Triforce v0.2.0

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
ruckus7382sfgsgdf
2026-11-10 19:32:48 -06:00
commit 34b3683441
1730 changed files with 554853 additions and 0 deletions

View File

@@ -0,0 +1,264 @@
# SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
# SPDX-License-Identifier: GPL-3.0-or-later
# SPDX-FileCopyrightText: 2018 yuzu Emulator Project
# SPDX-License-Identifier: GPL-2.0-or-later
add_library(shader_recompiler STATIC
backend/bindings.h
backend/glasm/emit_glasm.cpp
backend/glasm/emit_glasm.h
backend/glasm/emit_glasm_barriers.cpp
backend/glasm/emit_glasm_bitwise_conversion.cpp
backend/glasm/emit_glasm_composite.cpp
backend/glasm/emit_glasm_context_get_set.cpp
backend/glasm/emit_glasm_control_flow.cpp
backend/glasm/emit_glasm_convert.cpp
backend/glasm/emit_glasm_floating_point.cpp
backend/glasm/emit_glasm_image.cpp
backend/glasm/emit_glasm_instructions.h
backend/glasm/emit_glasm_integer.cpp
backend/glasm/emit_glasm_logical.cpp
backend/glasm/emit_glasm_memory.cpp
backend/glasm/emit_glasm_not_implemented.cpp
backend/glasm/emit_glasm_select.cpp
backend/glasm/emit_glasm_shared_memory.cpp
backend/glasm/emit_glasm_special.cpp
backend/glasm/emit_glasm_undefined.cpp
backend/glasm/emit_glasm_warp.cpp
backend/glasm/glasm_emit_context.cpp
backend/glasm/glasm_emit_context.h
backend/glasm/reg_alloc.cpp
backend/glasm/reg_alloc.h
backend/glsl/emit_glsl.cpp
backend/glsl/emit_glsl.h
backend/glsl/emit_glsl_atomic.cpp
backend/glsl/emit_glsl_barriers.cpp
backend/glsl/emit_glsl_bitwise_conversion.cpp
backend/glsl/emit_glsl_composite.cpp
backend/glsl/emit_glsl_context_get_set.cpp
backend/glsl/emit_glsl_control_flow.cpp
backend/glsl/emit_glsl_convert.cpp
backend/glsl/emit_glsl_floating_point.cpp
backend/glsl/emit_glsl_image.cpp
backend/glsl/emit_glsl_instructions.h
backend/glsl/emit_glsl_integer.cpp
backend/glsl/emit_glsl_logical.cpp
backend/glsl/emit_glsl_memory.cpp
backend/glsl/emit_glsl_not_implemented.cpp
backend/glsl/emit_glsl_select.cpp
backend/glsl/emit_glsl_shared_memory.cpp
backend/glsl/emit_glsl_special.cpp
backend/glsl/emit_glsl_undefined.cpp
backend/glsl/emit_glsl_warp.cpp
backend/glsl/glsl_emit_context.cpp
backend/glsl/glsl_emit_context.h
backend/glsl/var_alloc.cpp
backend/glsl/var_alloc.h
backend/spirv/emit_spirv.cpp
backend/spirv/emit_spirv.h
backend/spirv/emit_spirv_atomic.cpp
backend/spirv/emit_spirv_barriers.cpp
backend/spirv/emit_spirv_bitwise_conversion.cpp
backend/spirv/emit_spirv_composite.cpp
backend/spirv/emit_spirv_context_get_set.cpp
backend/spirv/emit_spirv_control_flow.cpp
backend/spirv/emit_spirv_convert.cpp
backend/spirv/emit_spirv_floating_point.cpp
backend/spirv/emit_spirv_image.cpp
backend/spirv/emit_spirv_image_atomic.cpp
backend/spirv/emit_spirv_instructions.h
backend/spirv/emit_spirv_integer.cpp
backend/spirv/emit_spirv_logical.cpp
backend/spirv/emit_spirv_memory.cpp
backend/spirv/emit_spirv_select.cpp
backend/spirv/emit_spirv_shared_memory.cpp
backend/spirv/emit_spirv_special.cpp
backend/spirv/emit_spirv_undefined.cpp
backend/spirv/emit_spirv_warp.cpp
backend/spirv/spirv_emit_context.cpp
backend/spirv/spirv_emit_context.h
environment.h
exception.h
frontend/ir/abstract_syntax_list.h
frontend/ir/attribute.h
frontend/ir/basic_block.cpp
frontend/ir/basic_block.h
frontend/ir/breadth_first_search.h
frontend/ir/condition.cpp
frontend/ir/condition.h
frontend/ir/flow_test.h
frontend/ir/ir_emitter.cpp
frontend/ir/ir_emitter.h
frontend/ir/microinstruction.cpp
frontend/ir/modifiers.h
frontend/ir/opcodes.cpp
frontend/ir/opcodes.h
frontend/ir/opcodes.inc
frontend/ir/patch.cpp
frontend/ir/patch.h
frontend/ir/post_order.cpp
frontend/ir/post_order.h
frontend/ir/pred.h
frontend/ir/program.cpp
frontend/ir/program.h
frontend/ir/reg.h
frontend/ir/type.cpp
frontend/ir/type.h
frontend/ir/value.cpp
frontend/ir/value.h
frontend/maxwell/control_flow.cpp
frontend/maxwell/control_flow.h
frontend/maxwell/decode.cpp
frontend/maxwell/decode.h
frontend/maxwell/indirect_branch_table_track.cpp
frontend/maxwell/indirect_branch_table_track.h
frontend/maxwell/instruction.h
frontend/maxwell/location.h
frontend/maxwell/maxwell.inc
frontend/maxwell/opcodes.cpp
frontend/maxwell/opcodes.h
frontend/maxwell/structured_control_flow.cpp
frontend/maxwell/structured_control_flow.h
frontend/maxwell/translate/impl/atomic_operations_global_memory.cpp
frontend/maxwell/translate/impl/atomic_operations_shared_memory.cpp
frontend/maxwell/translate/impl/attribute_memory_to_physical.cpp
frontend/maxwell/translate/impl/barrier_operations.cpp
frontend/maxwell/translate/impl/bitfield_extract.cpp
frontend/maxwell/translate/impl/bitfield_insert.cpp
frontend/maxwell/translate/impl/branch_indirect.cpp
frontend/maxwell/translate/impl/common_encoding.h
frontend/maxwell/translate/impl/common_funcs.cpp
frontend/maxwell/translate/impl/common_funcs.h
frontend/maxwell/translate/impl/condition_code_set.cpp
frontend/maxwell/translate/impl/double_add.cpp
frontend/maxwell/translate/impl/double_compare_and_set.cpp
frontend/maxwell/translate/impl/double_fused_multiply_add.cpp
frontend/maxwell/translate/impl/double_min_max.cpp
frontend/maxwell/translate/impl/double_multiply.cpp
frontend/maxwell/translate/impl/double_set_predicate.cpp
frontend/maxwell/translate/impl/exit_program.cpp
frontend/maxwell/translate/impl/find_leading_one.cpp
frontend/maxwell/translate/impl/floating_point_add.cpp
frontend/maxwell/translate/impl/floating_point_compare.cpp
frontend/maxwell/translate/impl/floating_point_compare_and_set.cpp
frontend/maxwell/translate/impl/floating_point_conversion_floating_point.cpp
frontend/maxwell/translate/impl/floating_point_conversion_integer.cpp
frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
frontend/maxwell/translate/impl/floating_point_min_max.cpp
frontend/maxwell/translate/impl/floating_point_multi_function.cpp
frontend/maxwell/translate/impl/floating_point_multiply.cpp
frontend/maxwell/translate/impl/floating_point_range_reduction.cpp
frontend/maxwell/translate/impl/floating_point_set_predicate.cpp
frontend/maxwell/translate/impl/floating_point_swizzled_add.cpp
frontend/maxwell/translate/impl/half_floating_point_add.cpp
frontend/maxwell/translate/impl/half_floating_point_fused_multiply_add.cpp
frontend/maxwell/translate/impl/half_floating_point_helper.cpp
frontend/maxwell/translate/impl/half_floating_point_helper.h
frontend/maxwell/translate/impl/half_floating_point_multiply.cpp
frontend/maxwell/translate/impl/half_floating_point_set.cpp
frontend/maxwell/translate/impl/half_floating_point_set_predicate.cpp
frontend/maxwell/translate/impl/impl.cpp
frontend/maxwell/translate/impl/impl.h
frontend/maxwell/translate/impl/integer_add.cpp
frontend/maxwell/translate/impl/integer_add_three_input.cpp
frontend/maxwell/translate/impl/integer_compare.cpp
frontend/maxwell/translate/impl/integer_compare_and_set.cpp
frontend/maxwell/translate/impl/integer_floating_point_conversion.cpp
frontend/maxwell/translate/impl/integer_funnel_shift.cpp
frontend/maxwell/translate/impl/integer_minimum_maximum.cpp
frontend/maxwell/translate/impl/integer_popcount.cpp
frontend/maxwell/translate/impl/integer_scaled_add.cpp
frontend/maxwell/translate/impl/integer_set_predicate.cpp
frontend/maxwell/translate/impl/integer_shift_left.cpp
frontend/maxwell/translate/impl/integer_shift_right.cpp
frontend/maxwell/translate/impl/integer_short_multiply_add.cpp
frontend/maxwell/translate/impl/integer_to_integer_conversion.cpp
frontend/maxwell/translate/impl/internal_stage_buffer_entry_read.cpp
frontend/maxwell/translate/impl/load_constant.cpp
frontend/maxwell/translate/impl/load_constant.h
frontend/maxwell/translate/impl/load_effective_address.cpp
frontend/maxwell/translate/impl/load_store_attribute.cpp
frontend/maxwell/translate/impl/load_store_local_shared.cpp
frontend/maxwell/translate/impl/load_store_memory.cpp
frontend/maxwell/translate/impl/logic_operation.cpp
frontend/maxwell/translate/impl/logic_operation_three_input.cpp
frontend/maxwell/translate/impl/move_predicate_to_register.cpp
frontend/maxwell/translate/impl/move_register.cpp
frontend/maxwell/translate/impl/move_register_to_predicate.cpp
frontend/maxwell/translate/impl/move_special_register.cpp
frontend/maxwell/translate/impl/not_implemented.cpp
frontend/maxwell/translate/impl/output_geometry.cpp
frontend/maxwell/translate/impl/pixel_load.cpp
frontend/maxwell/translate/impl/predicate_set_predicate.cpp
frontend/maxwell/translate/impl/predicate_set_register.cpp
frontend/maxwell/translate/impl/select_source_with_predicate.cpp
frontend/maxwell/translate/impl/surface_atomic_operations.cpp
frontend/maxwell/translate/impl/surface_load_store.cpp
frontend/maxwell/translate/impl/texture_fetch.cpp
frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp
frontend/maxwell/translate/impl/texture_gather.cpp
frontend/maxwell/translate/impl/texture_gather_swizzled.cpp
frontend/maxwell/translate/impl/texture_gradient.cpp
frontend/maxwell/translate/impl/texture_load.cpp
frontend/maxwell/translate/impl/texture_load_swizzled.cpp
frontend/maxwell/translate/impl/texture_mipmap_level.cpp
frontend/maxwell/translate/impl/texture_query.cpp
frontend/maxwell/translate/impl/video_helper.cpp
frontend/maxwell/translate/impl/video_helper.h
frontend/maxwell/translate/impl/video_minimum_maximum.cpp
frontend/maxwell/translate/impl/video_multiply_add.cpp
frontend/maxwell/translate/impl/video_set_predicate.cpp
frontend/maxwell/translate/impl/vote.cpp
frontend/maxwell/translate/impl/warp_shuffle.cpp
frontend/maxwell/translate/translate.cpp
frontend/maxwell/translate/translate.h
frontend/maxwell/translate_program.cpp
frontend/maxwell/translate_program.h
host_translate_info.h
ir_opt/collect_shader_info_pass.cpp
ir_opt/conditional_barrier_pass.cpp
ir_opt/constant_propagation_pass.cpp
ir_opt/dead_code_elimination_pass.cpp
ir_opt/dual_vertex_pass.cpp
ir_opt/global_memory_to_storage_buffer_pass.cpp
ir_opt/identity_removal_pass.cpp
ir_opt/layer_pass.cpp
ir_opt/lower_fp16_to_fp32.cpp
ir_opt/lower_fp64_to_fp32.cpp
ir_opt/lower_int64_to_int32.cpp
ir_opt/passes.h
ir_opt/position_pass.cpp
ir_opt/rescaling_pass.cpp
ir_opt/ssa_rewrite_pass.cpp
ir_opt/texture_pass.cpp
ir_opt/vendor_workaround_pass.cpp
ir_opt/verification_pass.cpp
object_pool.h
profile.h
program_header.h
runtime_info.h
shader_info.h
varying_state.h
)
target_link_libraries(shader_recompiler PUBLIC common fmt::fmt sirit::sirit)
if (MSVC)
target_compile_options(shader_recompiler PRIVATE
/we4242 # 'identifier': conversion from 'type1' to 'type2', possible loss of data
/we4254 # 'operator': conversion from 'type1:field_bits' to 'type2:field_bits', possible loss of data
/we4800 # Implicit conversion from 'type' to bool. Possible information loss
)
else()
target_compile_options(shader_recompiler PRIVATE
$<$<COMPILE_LANGUAGE:C,CXX>:-Werror=conversion>
# Bracket depth determines maximum size of a fold expression in Clang since 9c9974c3ccb6.
# And this in turns limits the size of a std::array.
$<$<CXX_COMPILER_ID:Clang>:-fbracket-depth=1024>
$<$<CXX_COMPILER_ID:AppleClang>:-fbracket-depth=1024>
)
endif()
create_target_directory_groups(shader_recompiler)

View File

@@ -0,0 +1,20 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "common/common_types.h"
namespace Shader::Backend {
struct Bindings {
u32 unified{};
u32 uniform_buffer{};
u32 storage_buffer{};
u32 texture{};
u32 image{};
u32 texture_scaling_index{};
u32 image_scaling_index{};
};
} // namespace Shader::Backend

View File

@@ -0,0 +1,252 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <string>
#include <tuple>
#include <type_traits>
#include "common/div_ceil.h"
#include "common/settings.h"
#include "shader_recompiler/backend/glsl/emit_glsl.h"
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/ir_emitter.h"
namespace Shader::Backend::GLSL {
namespace {
template <class Func>
struct FuncTraits {};
template <class ReturnType_, class... Args>
struct FuncTraits<ReturnType_ (*)(Args...)> {
using ReturnType = ReturnType_;
static constexpr size_t NUM_ARGS = sizeof...(Args);
template <size_t I>
using ArgType = std::tuple_element_t<I, std::tuple<Args...>>;
};
template <auto func, typename... Args>
void SetDefinition(EmitContext& ctx, IR::Inst* inst, Args... args) {
inst->SetDefinition<Id>(func(ctx, std::forward<Args>(args)...));
}
template <typename ArgType>
auto Arg(EmitContext& ctx, const IR::Value& arg) {
if constexpr (std::is_same_v<ArgType, std::string_view>) {
return ctx.var_alloc.Consume(arg);
} else if constexpr (std::is_same_v<ArgType, const IR::Value&>) {
return arg;
} else if constexpr (std::is_same_v<ArgType, u32>) {
return arg.U32();
} else if constexpr (std::is_same_v<ArgType, IR::Attribute>) {
return arg.Attribute();
} else if constexpr (std::is_same_v<ArgType, IR::Patch>) {
return arg.Patch();
} else if constexpr (std::is_same_v<ArgType, IR::Reg>) {
return arg.Reg();
}
}
template <auto func, bool is_first_arg_inst, size_t... I>
void Invoke(EmitContext& ctx, IR::Inst* inst, std::index_sequence<I...>) {
using Traits = FuncTraits<decltype(func)>;
if constexpr (std::is_same_v<typename Traits::ReturnType, Id>) {
if constexpr (is_first_arg_inst) {
SetDefinition<func>(
ctx, inst, *inst,
Arg<typename Traits::template ArgType<I + 2>>(ctx, inst->Arg(I))...);
} else {
SetDefinition<func>(
ctx, inst, Arg<typename Traits::template ArgType<I + 1>>(ctx, inst->Arg(I))...);
}
} else {
if constexpr (is_first_arg_inst) {
func(ctx, *inst, Arg<typename Traits::template ArgType<I + 2>>(ctx, inst->Arg(I))...);
} else {
func(ctx, Arg<typename Traits::template ArgType<I + 1>>(ctx, inst->Arg(I))...);
}
}
}
template <auto func>
void Invoke(EmitContext& ctx, IR::Inst* inst) {
using Traits = FuncTraits<decltype(func)>;
static_assert(Traits::NUM_ARGS >= 1, "Insufficient arguments");
if constexpr (Traits::NUM_ARGS == 1) {
Invoke<func, false>(ctx, inst, std::make_index_sequence<0>{});
} else {
using FirstArgType = typename Traits::template ArgType<1>;
static constexpr bool is_first_arg_inst = std::is_same_v<FirstArgType, IR::Inst&>;
using Indices = std::make_index_sequence<Traits::NUM_ARGS - (is_first_arg_inst ? 2 : 1)>;
Invoke<func, is_first_arg_inst>(ctx, inst, Indices{});
}
}
void EmitInst(EmitContext& ctx, IR::Inst* inst) {
switch (inst->GetOpcode()) {
#define OPCODE(name, result_type, ...) \
case IR::Opcode::name: \
return Invoke<&Emit##name>(ctx, inst);
#include "shader_recompiler/frontend/ir/opcodes.inc"
#undef OPCODE
}
throw LogicError("Invalid opcode {}", inst->GetOpcode());
}
bool IsReference(IR::Inst& inst) {
return inst.GetOpcode() == IR::Opcode::Reference;
}
void PrecolorInst(IR::Inst& phi) {
// Insert phi moves before references to avoid overwriting other phis
const size_t num_args{phi.NumArgs()};
for (size_t i = 0; i < num_args; ++i) {
IR::Block& phi_block{*phi.PhiBlock(i)};
auto it{std::find_if_not(phi_block.rbegin(), phi_block.rend(), IsReference).base()};
IR::IREmitter ir{phi_block, it};
const IR::Value arg{phi.Arg(i)};
if (arg.IsImmediate()) {
ir.PhiMove(phi, arg);
} else {
ir.PhiMove(phi, IR::Value{arg.InstRecursive()});
}
}
for (size_t i = 0; i < num_args; ++i) {
IR::IREmitter{*phi.PhiBlock(i)}.Reference(IR::Value{&phi});
}
}
void Precolor(const IR::Program& program) {
for (IR::Block* const block : program.blocks) {
for (IR::Inst& phi : block->Instructions()) {
if (!IR::IsPhi(phi)) {
break;
}
PrecolorInst(phi);
}
}
}
void EmitCode(EmitContext& ctx, const IR::Program& program) {
for (const IR::AbstractSyntaxNode& node : program.syntax_list) {
switch (node.type) {
case IR::AbstractSyntaxNode::Type::Block:
for (IR::Inst& inst : node.data.block->Instructions()) {
EmitInst(ctx, &inst);
}
break;
case IR::AbstractSyntaxNode::Type::If:
ctx.Add("if({}){{", ctx.var_alloc.Consume(node.data.if_node.cond));
break;
case IR::AbstractSyntaxNode::Type::EndIf:
ctx.Add("}}");
break;
case IR::AbstractSyntaxNode::Type::Break:
if (node.data.break_node.cond.IsImmediate()) {
if (node.data.break_node.cond.U1()) {
ctx.Add("break;");
}
} else {
ctx.Add("if({}){{break;}}", ctx.var_alloc.Consume(node.data.break_node.cond));
}
break;
case IR::AbstractSyntaxNode::Type::Return:
case IR::AbstractSyntaxNode::Type::Unreachable:
ctx.Add("return;");
break;
case IR::AbstractSyntaxNode::Type::Loop:
ctx.Add("for(;;){{");
break;
case IR::AbstractSyntaxNode::Type::Repeat:
if (Settings::values.disable_shader_loop_safety_checks) {
ctx.Add("if(!{}){{break;}}}}", ctx.var_alloc.Consume(node.data.repeat.cond));
} else {
ctx.Add("if(--loop{}<0 || !{}){{break;}}}}", ctx.num_safety_loop_vars++,
ctx.var_alloc.Consume(node.data.repeat.cond));
}
break;
default:
throw NotImplementedException("AbstractSyntaxNode Type {}", node.type);
}
}
}
std::string GlslVersionSpecifier(const EmitContext& ctx) {
if (ctx.uses_y_direction) {
return " compatibility";
}
return "";
}
bool IsPreciseType(GlslVarType type) {
switch (type) {
case GlslVarType::PrecF32:
case GlslVarType::PrecF64:
return true;
default:
return false;
}
}
void DefineVariables(const EmitContext& ctx, std::string& header) {
for (u32 i = 0; i < static_cast<u32>(GlslVarType::Void); ++i) {
const auto type{static_cast<GlslVarType>(i)};
const auto& tracker{ctx.var_alloc.GetUseTracker(type)};
const auto type_name{ctx.var_alloc.GetGlslType(type)};
const bool has_precise_bug{ctx.stage == Stage::Fragment && ctx.profile.has_gl_precise_bug};
const auto precise{!has_precise_bug && IsPreciseType(type) ? "precise " : ""};
// Temps/return types that are never used are stored at index 0
if (tracker.uses_temp) {
header += fmt::format("{}{} t{}={}(0);", precise, type_name,
ctx.var_alloc.Representation(0, type), type_name);
}
for (u32 index = 0; index < tracker.num_used; ++index) {
header += fmt::format("{}{} {}={}(0);", precise, type_name,
ctx.var_alloc.Representation(index, type), type_name);
}
}
for (u32 i = 0; i < ctx.num_safety_loop_vars; ++i) {
header += fmt::format("int loop{}=0x2000;", i);
}
}
} // Anonymous namespace
std::string EmitGLSL(const Profile& profile, const RuntimeInfo& runtime_info, IR::Program& program,
Bindings& bindings) {
EmitContext ctx{program, bindings, profile, runtime_info};
Precolor(program);
EmitCode(ctx, program);
const std::string version{fmt::format("#version 460{}\n", GlslVersionSpecifier(ctx))};
ctx.header.insert(0, version);
if (program.shared_memory_size > 0) {
const auto requested_size{program.shared_memory_size};
const auto max_size{profile.gl_max_compute_smem_size};
const bool needs_clamp{requested_size > max_size};
if (needs_clamp) {
LOG_WARNING(Shader_GLSL, "Requested shared memory size ({}) exceeds device limit ({})",
requested_size, max_size);
}
const auto smem_size{needs_clamp ? max_size : requested_size};
ctx.header += fmt::format("shared uint smem[{}];", Common::DivCeil(smem_size, 4U));
}
ctx.header += "void main(){\n";
if (program.local_memory_size > 0) {
ctx.header += fmt::format("uint lmem[{}];", Common::DivCeil(program.local_memory_size, 4U));
}
DefineVariables(ctx, ctx.header);
if (ctx.uses_cc_carry) {
ctx.header += "uint carry;";
}
if (program.info.uses_subgroup_shuffles) {
ctx.header += "bool shfl_in_bounds;";
ctx.header += "uint shfl_result;";
}
ctx.code.insert(0, ctx.header);
ctx.code += '}';
return ctx.code;
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,23 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <string>
#include "shader_recompiler/backend/bindings.h"
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/profile.h"
#include "shader_recompiler/runtime_info.h"
namespace Shader::Backend::GLSL {
[[nodiscard]] std::string EmitGLSL(const Profile& profile, const RuntimeInfo& runtime_info,
IR::Program& program, Bindings& bindings);
[[nodiscard]] inline std::string EmitGLSL(const Profile& profile, IR::Program& program) {
Bindings binding;
return EmitGLSL(profile, {}, program, binding);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,559 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
namespace {
constexpr char cas_loop[]{
"for (;;){{uint old={};{}=atomicCompSwap({},old,{}({},{}));if({}==old){{break;}}}}"};
void SharedCasFunction(EmitContext& ctx, IR::Inst& inst, std::string_view offset,
std::string_view value, std::string_view function) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32)};
const std::string smem{fmt::format("smem[{}>>2]", offset)};
ctx.Add(cas_loop, smem, ret, smem, function, smem, value, ret);
}
void SsboCasFunction(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value, std::string_view function) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32)};
const std::string ssbo{fmt::format("{}_ssbo{}[{}>>2]", ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset))};
ctx.Add(cas_loop, ssbo, ret, ssbo, function, ssbo, value, ret);
}
void SsboCasFunctionF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value,
std::string_view function) {
const std::string ssbo{fmt::format("{}_ssbo{}[{}>>2]", ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset))};
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32)};
ctx.Add(cas_loop, ssbo, ret, ssbo, function, ssbo, value, ret);
ctx.AddF32("{}=utof({});", inst, ret);
}
} // Anonymous namespace
void EmitSharedAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
ctx.AddU32("{}=atomicAdd(smem[{}>>2],{});", inst, pointer_offset, value);
}
void EmitSharedAtomicSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
const std::string u32_value{fmt::format("uint({})", value)};
SharedCasFunction(ctx, inst, pointer_offset, u32_value, "CasMinS32");
}
void EmitSharedAtomicUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
ctx.AddU32("{}=atomicMin(smem[{}>>2],{});", inst, pointer_offset, value);
}
void EmitSharedAtomicSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
const std::string u32_value{fmt::format("uint({})", value)};
SharedCasFunction(ctx, inst, pointer_offset, u32_value, "CasMaxS32");
}
void EmitSharedAtomicUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
ctx.AddU32("{}=atomicMax(smem[{}>>2],{});", inst, pointer_offset, value);
}
void EmitSharedAtomicInc32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
SharedCasFunction(ctx, inst, pointer_offset, value, "CasIncrement");
}
void EmitSharedAtomicDec32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
SharedCasFunction(ctx, inst, pointer_offset, value, "CasDecrement");
}
void EmitSharedAtomicAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
ctx.AddU32("{}=atomicAnd(smem[{}>>2],{});", inst, pointer_offset, value);
}
void EmitSharedAtomicOr32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
ctx.AddU32("{}=atomicOr(smem[{}>>2],{});", inst, pointer_offset, value);
}
void EmitSharedAtomicXor32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
ctx.AddU32("{}=atomicXor(smem[{}>>2],{});", inst, pointer_offset, value);
}
void EmitSharedAtomicExchange32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
ctx.AddU32("{}=atomicExchange(smem[{}>>2],{});", inst, pointer_offset, value);
}
void EmitSharedAtomicExchange64(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU64("{}=packUint2x32(uvec2(smem[{}>>2],smem[({}+4)>>2]));", inst, pointer_offset,
pointer_offset);
ctx.Add("smem[{}>>2]=unpackUint2x32({}).x;smem[({}+4)>>2]=unpackUint2x32({}).y;",
pointer_offset, value, pointer_offset, value);
}
void EmitSharedAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=uvec2(smem[{}>>2],smem[({}+4)>>2]);", inst, pointer_offset, pointer_offset);
ctx.Add("smem[{}>>2]={}.x;smem[({}+4)>>2]={}.y;", pointer_offset, value, pointer_offset, value);
}
void EmitStorageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU32("{}=atomicAdd({}_ssbo{}[{}>>2],{});", inst, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicSMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
const std::string u32_value{fmt::format("uint({})", value)};
SsboCasFunction(ctx, inst, binding, offset, u32_value, "CasMinS32");
}
void EmitStorageAtomicUMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU32("{}=atomicMin({}_ssbo{}[{}>>2],{});", inst, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicSMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
const std::string u32_value{fmt::format("uint({})", value)};
SsboCasFunction(ctx, inst, binding, offset, u32_value, "CasMaxS32");
}
void EmitStorageAtomicUMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU32("{}=atomicMax({}_ssbo{}[{}>>2],{});", inst, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicInc32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunction(ctx, inst, binding, offset, value, "CasIncrement");
}
void EmitStorageAtomicDec32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunction(ctx, inst, binding, offset, value, "CasDecrement");
}
void EmitStorageAtomicAnd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU32("{}=atomicAnd({}_ssbo{}[{}>>2],{});", inst, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicOr32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU32("{}=atomicOr({}_ssbo{}[{}>>2],{});", inst, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicXor32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU32("{}=atomicXor({}_ssbo{}[{}>>2],{});", inst, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicExchange32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU32("{}=atomicExchange({}_ssbo{}[{}>>2],{});", inst, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicIAdd64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU64("{}=packUint2x32(uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]));", inst,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset));
ctx.Add("{}_ssbo{}[{}>>2]+=unpackUint2x32({}).x;{}_ssbo{}[({}>>2)+1]+=unpackUint2x32({}).y;",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicSMin64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU64("{}=packInt2x32(ivec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]));", inst,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset));
ctx.Add("for(int i=0;i<2;++i){{ "
"{}_ssbo{}[({}>>2)+i]=uint(min(int({}_ssbo{}[({}>>2)+i]),unpackInt2x32(int64_t({}))[i])"
");}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicUMin64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU64("{}=packUint2x32(uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]));", inst,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset));
ctx.Add("for(int i=0;i<2;++i){{ "
"{}_ssbo{}[({}>>2)+i]=min({}_ssbo{}[({}>>2)+i],unpackUint2x32(uint64_t({}))[i]);}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicSMax64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU64("{}=packInt2x32(ivec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]));", inst,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset));
ctx.Add("for(int i=0;i<2;++i){{ "
"{}_ssbo{}[({}>>2)+i]=uint(max(int({}_ssbo{}[({}>>2)+i]),unpackInt2x32(int64_t({}))[i])"
");}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicUMax64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU64("{}=packUint2x32(uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]));", inst,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset));
ctx.Add("for(int "
"i=0;i<2;++i){{{}_ssbo{}[({}>>2)+i]=max({}_ssbo{}[({}>>2)+i],unpackUint2x32(uint64_t({}"
"))[i]);}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicAnd64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU64(
"{}=packUint2x32(uvec2(atomicAnd({}_ssbo{}[{}>>2],unpackUint2x32({}).x),atomicAnd({}_"
"ssbo{}[({}>>2)+1],unpackUint2x32({}).y)));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicOr64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU64("{}=packUint2x32(uvec2(atomicOr({}_ssbo{}[{}>>2],unpackUint2x32({}).x),atomicOr({}_"
"ssbo{}[({}>>2)+1],unpackUint2x32({}).y)));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicXor64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU64(
"{}=packUint2x32(uvec2(atomicXor({}_ssbo{}[{}>>2],unpackUint2x32({}).x),atomicXor({}_"
"ssbo{}[({}>>2)+1],unpackUint2x32({}).y)));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicExchange64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
ctx.AddU64("{}=packUint2x32(uvec2(atomicExchange({}_ssbo{}[{}>>2],unpackUint2x32({}).x),"
"atomicExchange({}_ssbo{}[({}>>2)+1],unpackUint2x32({}).y)));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicIAdd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("{}_ssbo{}[{}>>2]+={}.x;{}_ssbo{}[({}>>2)+1]+={}.y;", ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value, ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicSMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=ivec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("for(int "
"i=0;i<2;++i){{{}_ssbo{}[({}>>2)+i]=uint(min(int({}_ssbo{}[({}>>2)+i]),int({}[i])));}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicUMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("for(int i=0;i<2;++i){{ "
"{}_ssbo{}[({}>>2)+i]=min({}_ssbo{}[({}>>2)+i],{}[i]);}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicSMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=ivec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("for(int "
"i=0;i<2;++i){{{}_ssbo{}[({}>>2)+i]=uint(max(int({}_ssbo{}[({}>>2)+i]),int({}[i])));}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicUMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to non-atomic");
ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}>>2)+1]);", inst, ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name, binding.U32(),
ctx.var_alloc.Consume(offset));
ctx.Add("for(int i=0;i<2;++i){{{}_ssbo{}[({}>>2)+i]=max({}_ssbo{}[({}>>2)+i],{}[i]);}}",
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), ctx.stage_name,
binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicAnd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to 32x2");
ctx.AddU32x2("{}=uvec2(atomicAnd({}_ssbo{}[{}>>2],{}.x),atomicAnd({}_ssbo{}[({}>>2)+1],{}.y));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicOr32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to 32x2");
ctx.AddU32x2("{}=uvec2(atomicOr({}_ssbo{}[{}>>2],{}.x),atomicOr({}_ssbo{}[({}>>2)+1],{}.y));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicXor32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to 32x2");
ctx.AddU32x2("{}=uvec2(atomicXor({}_ssbo{}[{}>>2],{}.x),atomicXor({}_ssbo{}[({}>>2)+1],{}.y));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
LOG_WARNING(Shader_GLSL, "Int64 atomics not supported, fallback to 32x2");
ctx.AddU32x2("{}=uvec2(atomicExchange({}_ssbo{}[{}>>2],{}.x),atomicExchange({}_ssbo{}[({}>>2)+"
"1],{}.y));",
inst, ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value,
ctx.stage_name, binding.U32(), ctx.var_alloc.Consume(offset), value);
}
void EmitStorageAtomicAddF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunctionF32(ctx, inst, binding, offset, value, "CasFloatAdd");
}
void EmitStorageAtomicAddF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunction(ctx, inst, binding, offset, value, "CasFloatAdd16x2");
}
void EmitStorageAtomicAddF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunction(ctx, inst, binding, offset, value, "CasFloatAdd32x2");
}
void EmitStorageAtomicMinF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunction(ctx, inst, binding, offset, value, "CasFloatMin16x2");
}
void EmitStorageAtomicMinF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunction(ctx, inst, binding, offset, value, "CasFloatMin32x2");
}
void EmitStorageAtomicMaxF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunction(ctx, inst, binding, offset, value, "CasFloatMax16x2");
}
void EmitStorageAtomicMaxF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
SsboCasFunction(ctx, inst, binding, offset, value, "CasFloatMax32x2");
}
void EmitGlobalAtomicIAdd32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicSMin32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicUMin32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicSMax32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicUMax32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicInc32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicDec32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAnd32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicOr32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicXor32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicExchange32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicIAdd64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicSMin64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicUMin64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicSMax64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicUMax64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicInc64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicDec64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAnd64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicOr64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicXor64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicExchange64(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicIAdd32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicSMin32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicUMin32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicSMax32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicUMax32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicInc32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicDec32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAnd32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicOr32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicXor32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicExchange32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAddF32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAddF16x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAddF32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicMinF16x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicMinF32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicMaxF16x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicMaxF32x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,19 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
namespace Shader::Backend::GLSL {
void EmitBarrier(EmitContext& ctx) {
ctx.Add("barrier();");
}
void EmitWorkgroupMemoryBarrier(EmitContext& ctx) {
ctx.Add("groupMemoryBarrier();");
}
void EmitDeviceMemoryBarrier(EmitContext& ctx) {
ctx.Add("memoryBarrier();");
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,95 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::GLSL {
namespace {
void Alias(IR::Inst& inst, const IR::Value& value) {
if (value.IsImmediate()) {
return;
}
IR::Inst& value_inst{*value.InstRecursive()};
value_inst.DestructiveAddUsage(inst.UseCount());
value_inst.DestructiveRemoveUsage();
inst.SetDefinition(value_inst.Definition<Id>());
}
} // Anonymous namespace
void EmitIdentity(EmitContext&, IR::Inst& inst, const IR::Value& value) {
Alias(inst, value);
}
void EmitConditionRef(EmitContext& ctx, IR::Inst& inst, const IR::Value& value) {
// Fake one usage to get a real variable out of the condition
inst.DestructiveAddUsage(1);
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U1)};
const auto input{ctx.var_alloc.Consume(value)};
const auto suffix{ctx.profile.has_gl_bool_ref_bug ? "?true:false" : ""};
if (ret != input) {
ctx.Add("{}={}{};", ret, input, suffix);
}
}
void EmitBitCastU16F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst) {
NotImplemented();
}
void EmitBitCastU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=ftou({});", inst, value);
}
void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=doubleBitsToUint64({});", inst, value);
}
void EmitBitCastF16U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst) {
NotImplemented();
}
void EmitBitCastF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=utof({});", inst, value);
}
void EmitBitCastF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=uint64BitsToDouble({});", inst, value);
}
void EmitPackUint2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=packUint2x32({});", inst, value);
}
void EmitUnpackUint2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32x2("{}=unpackUint2x32({});", inst, value);
}
void EmitPackFloat2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=packFloat2x16({});", inst, value);
}
void EmitUnpackFloat2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF16x2("{}=unpackFloat2x16({});", inst, value);
}
void EmitPackHalf2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=packHalf2x16({});", inst, value);
}
void EmitUnpackHalf2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32x2("{}=unpackHalf2x16({});", inst, value);
}
void EmitPackDouble2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=packDouble2x32({});", inst, value);
}
void EmitUnpackDouble2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32x2("{}=unpackDouble2x32({});", inst, value);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,218 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
namespace {
constexpr std::string_view SWIZZLE{"xyzw"};
void CompositeInsert(EmitContext& ctx, std::string_view result, std::string_view composite,
std::string_view object, u32 index) {
if (result == composite) {
// The result is aliased with the composite
ctx.Add("{}.{}={};", composite, SWIZZLE[index], object);
} else {
ctx.Add("{}={};{}.{}={};", result, composite, result, SWIZZLE[index], object);
}
}
} // Anonymous namespace
void EmitCompositeConstructU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2) {
ctx.AddU32x2("{}=uvec2({},{});", inst, e1, e2);
}
void EmitCompositeConstructU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3) {
ctx.AddU32x3("{}=uvec3({},{},{});", inst, e1, e2, e3);
}
void EmitCompositeConstructU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3, std::string_view e4) {
ctx.AddU32x4("{}=uvec4({},{},{},{});", inst, e1, e2, e3, e4);
}
void EmitCompositeExtractU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddU32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeExtractU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddU32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeExtractU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddU32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeInsertU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32x2)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeInsertU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32x3)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeInsertU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32x4)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeConstructF16x2([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view e1,
[[maybe_unused]] std::string_view e2) {
NotImplemented();
}
void EmitCompositeConstructF16x3([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view e1,
[[maybe_unused]] std::string_view e2,
[[maybe_unused]] std::string_view e3) {
NotImplemented();
}
void EmitCompositeConstructF16x4([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view e1,
[[maybe_unused]] std::string_view e2,
[[maybe_unused]] std::string_view e3,
[[maybe_unused]] std::string_view e4) {
NotImplemented();
}
void EmitCompositeExtractF16x2([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeExtractF16x3([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeExtractF16x4([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeInsertF16x2([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] std::string_view object,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeInsertF16x3([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] std::string_view object,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeInsertF16x4([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] std::string_view object,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeConstructF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2) {
ctx.AddF32x2("{}=vec2({},{});", inst, e1, e2);
}
void EmitCompositeConstructF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3) {
ctx.AddF32x3("{}=vec3({},{},{});", inst, e1, e2, e3);
}
void EmitCompositeConstructF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3, std::string_view e4) {
ctx.AddF32x4("{}=vec4({},{},{},{});", inst, e1, e2, e3, e4);
}
void EmitCompositeExtractF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddF32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeExtractF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddF32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeExtractF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddF32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeInsertF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::F32x2)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeInsertF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::F32x3)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeInsertF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::F32x4)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeConstructF64x2([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeConstructF64x3([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeConstructF64x4([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeExtractF64x2([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeExtractF64x3([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeExtractF64x4([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeInsertF64x2(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index) {
ctx.Add("{}.{}={};", composite, SWIZZLE[index], object);
}
void EmitCompositeInsertF64x3(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index) {
ctx.Add("{}.{}={};", composite, SWIZZLE[index], object);
}
void EmitCompositeInsertF64x4(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index) {
ctx.Add("{}.{}={};", composite, SWIZZLE[index], object);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,476 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
#include "shader_recompiler/runtime_info.h"
namespace Shader::Backend::GLSL {
namespace {
constexpr char SWIZZLE[]{"xyzw"};
u32 CbufIndex(u32 offset) {
return (offset / 4) % 4;
}
char OffsetSwizzle(u32 offset) {
return SWIZZLE[CbufIndex(offset)];
}
bool IsInputArray(Stage stage) {
return stage == Stage::Geometry || stage == Stage::TessellationControl ||
stage == Stage::TessellationEval;
}
std::string InputVertexIndex(EmitContext& ctx, std::string_view vertex) {
return IsInputArray(ctx.stage) ? fmt::format("[{}]", vertex) : "";
}
std::string_view OutputVertexIndex(EmitContext& ctx) {
return ctx.stage == Stage::TessellationControl ? "[gl_InvocationID]" : "";
}
std::string ChooseCbuf(EmitContext& ctx, const IR::Value& binding, std::string_view index) {
if (binding.IsImmediate()) {
return fmt::format("{}_cbuf{}[{}]", ctx.stage_name, binding.U32(), index);
} else {
const auto binding_var{ctx.var_alloc.Consume(binding)};
return fmt::format("GetCbufIndirect({},{})", binding_var, index);
}
}
void GetCbuf(EmitContext& ctx, std::string_view ret, const IR::Value& binding,
const IR::Value& offset, u32 num_bits, std::string_view cast = {},
std::string_view bit_offset = {}) {
const bool is_immediate{offset.IsImmediate()};
const bool component_indexing_bug{!is_immediate && ctx.profile.has_gl_component_indexing_bug};
if (is_immediate) {
const s32 signed_offset{static_cast<s32>(offset.U32())};
static constexpr u32 cbuf_size{0x10000};
if (signed_offset < 0 || offset.U32() > cbuf_size) {
LOG_WARNING(Shader_GLSL, "Immediate constant buffer offset is out of bounds");
ctx.Add("{}=0u;", ret);
return;
}
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto index{is_immediate ? fmt::format("{}", offset.U32() / 16)
: fmt::format("{}>>4", offset_var)};
const auto swizzle{is_immediate ? fmt::format(".{}", OffsetSwizzle(offset.U32()))
: fmt::format("[({}>>2)%4]", offset_var)};
const auto cbuf{ChooseCbuf(ctx, binding, index)};
const auto cbuf_cast{fmt::format("{}({}{{}})", cast, cbuf)};
const auto extraction{num_bits == 32 ? cbuf_cast
: fmt::format("bitfieldExtract({},int({}),{})", cbuf_cast,
bit_offset, num_bits)};
if (!component_indexing_bug) {
const auto result{fmt::format(fmt::runtime(extraction), swizzle)};
ctx.Add("{}={};", ret, result);
return;
}
const auto cbuf_offset{fmt::format("{}>>2", offset_var)};
for (u32 i = 0; i < 4; ++i) {
const auto swizzle_string{fmt::format(".{}", "xyzw"[i])};
const auto result{fmt::format(fmt::runtime(extraction), swizzle_string)};
ctx.Add("if(({}&3)=={}){}={};", cbuf_offset, i, ret, result);
}
}
void GetCbuf8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, const IR::Value& offset,
std::string_view cast) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32)};
if (offset.IsImmediate()) {
const auto bit_offset{fmt::format("{}", (offset.U32() % 4) * 8)};
GetCbuf(ctx, ret, binding, offset, 8, cast, bit_offset);
} else {
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("({}%4)*8", offset_var)};
GetCbuf(ctx, ret, binding, offset, 8, cast, bit_offset);
}
}
void GetCbuf16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, const IR::Value& offset,
std::string_view cast) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32)};
if (offset.IsImmediate()) {
const auto bit_offset{fmt::format("{}", ((offset.U32() / 2) % 2) * 16)};
GetCbuf(ctx, ret, binding, offset, 16, cast, bit_offset);
} else {
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("(({}>>1)%2)*16", offset_var)};
GetCbuf(ctx, ret, binding, offset, 16, cast, bit_offset);
}
}
} // Anonymous namespace
void EmitGetCbufU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{ctx.profile.has_gl_cbuf_ftou_bug ? "" : "ftou"};
GetCbuf8(ctx, inst, binding, offset, cast);
}
void EmitGetCbufS8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{ctx.profile.has_gl_cbuf_ftou_bug ? "int" : "ftoi"};
GetCbuf8(ctx, inst, binding, offset, cast);
}
void EmitGetCbufU16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{ctx.profile.has_gl_cbuf_ftou_bug ? "" : "ftou"};
GetCbuf16(ctx, inst, binding, offset, cast);
}
void EmitGetCbufS16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{ctx.profile.has_gl_cbuf_ftou_bug ? "int" : "ftoi"};
GetCbuf16(ctx, inst, binding, offset, cast);
}
void EmitGetCbufU32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32)};
const auto cast{ctx.profile.has_gl_cbuf_ftou_bug ? "" : "ftou"};
GetCbuf(ctx, ret, binding, offset, 32, cast);
}
void EmitGetCbufF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::F32)};
const auto cast{ctx.profile.has_gl_cbuf_ftou_bug ? "utof" : ""};
GetCbuf(ctx, ret, binding, offset, 32, cast);
}
void EmitGetCbufU32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{ctx.profile.has_gl_cbuf_ftou_bug ? "" : "ftou"};
if (offset.IsImmediate()) {
const auto cbuf{fmt::format("{}_cbuf{}", ctx.stage_name, binding.U32())};
static constexpr u32 cbuf_size{0x10000};
const u32 u32_offset{offset.U32()};
const s32 signed_offset{static_cast<s32>(offset.U32())};
if (signed_offset < 0 || u32_offset > cbuf_size) {
LOG_WARNING(Shader_GLSL, "Immediate constant buffer offset is out of bounds");
ctx.AddU32x2("{}=uvec2(0u);", inst);
return;
}
if (u32_offset % 2 == 0) {
ctx.AddU32x2("{}={}({}[{}].{}{});", inst, cast, cbuf, u32_offset / 16,
OffsetSwizzle(u32_offset), OffsetSwizzle(u32_offset + 4));
} else {
ctx.AddU32x2("{}=uvec2({}({}[{}].{}),{}({}[{}].{}));", inst, cast, cbuf,
u32_offset / 16, OffsetSwizzle(u32_offset), cast, cbuf,
(u32_offset + 4) / 16, OffsetSwizzle(u32_offset + 4));
}
return;
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto cbuf{ChooseCbuf(ctx, binding, fmt::format("{}>>4", offset_var))};
if (!ctx.profile.has_gl_component_indexing_bug) {
ctx.AddU32x2("{}=uvec2({}({}[({}>>2)%4]),{}({}[(({}+4)>>2)%4]));", inst, cast, cbuf,
offset_var, cast, cbuf, offset_var);
return;
}
const auto ret{ctx.var_alloc.Define(inst, GlslVarType::U32x2)};
const auto cbuf_offset{fmt::format("{}>>2", offset_var)};
for (u32 swizzle = 0; swizzle < 4; ++swizzle) {
ctx.Add("if(({}&3)=={}){}=uvec2({}({}.{}),{}({}.{}));", cbuf_offset, swizzle, ret, cast,
cbuf, "xyzw"[swizzle], cast, cbuf, "xyzw"[(swizzle + 1) % 4]);
}
}
void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr,
std::string_view vertex) {
const u32 element{static_cast<u32>(attr) % 4};
const char swizzle{"xyzw"[element]};
if (IR::IsGeneric(attr)) {
const u32 index{IR::GenericAttributeIndex(attr)};
if (!ctx.runtime_info.previous_stage_stores.Generic(index, element)) {
if (element == 3) {
ctx.AddF32("{}=1.f;", inst, attr);
} else {
ctx.AddF32("{}=0.f;", inst, attr);
}
return;
}
ctx.AddF32("{}=in_attr{}{}.{};", inst, index, InputVertexIndex(ctx, vertex), swizzle);
return;
}
switch (attr) {
case IR::Attribute::PrimitiveId:
ctx.AddF32("{}=itof(gl_PrimitiveID);", inst);
break;
case IR::Attribute::Layer:
ctx.AddF32("{}=itof(gl_Layer);", inst);
break;
case IR::Attribute::PositionX:
case IR::Attribute::PositionY:
case IR::Attribute::PositionZ:
case IR::Attribute::PositionW: {
const bool is_array{IsInputArray(ctx.stage)};
const auto input_decorator{is_array ? fmt::format("gl_in[{}].", vertex) : ""};
ctx.AddF32("{}={}{}.{};", inst, input_decorator, ctx.position_name, swizzle);
break;
}
case IR::Attribute::PointSpriteS:
case IR::Attribute::PointSpriteT:
ctx.AddF32("{}=gl_PointCoord.{};", inst, swizzle);
break;
case IR::Attribute::TessellationEvaluationPointU:
case IR::Attribute::TessellationEvaluationPointV:
ctx.AddF32("{}=gl_TessCoord.{};", inst, swizzle);
break;
case IR::Attribute::InstanceId:
ctx.AddF32("{}=itof(gl_InstanceID);", inst);
break;
case IR::Attribute::VertexId:
ctx.AddF32("{}=itof(gl_VertexID);", inst);
break;
case IR::Attribute::FrontFace:
ctx.AddF32("{}=itof(gl_FrontFacing?-1:0);", inst);
break;
case IR::Attribute::BaseInstance:
ctx.AddF32("{}=itof(gl_BaseInstance);", inst);
break;
case IR::Attribute::BaseVertex:
ctx.AddF32("{}=itof(gl_BaseVertex);", inst);
break;
case IR::Attribute::DrawID:
ctx.AddF32("{}=itof(gl_DrawID);", inst);
break;
default:
throw NotImplementedException("Get attribute {}", attr);
}
}
void EmitGetAttributeU32(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, std::string_view) {
switch (attr) {
case IR::Attribute::PrimitiveId:
ctx.AddU32("{}=uint(gl_PrimitiveID);", inst);
break;
case IR::Attribute::InstanceId:
ctx.AddU32("{}=uint(gl_InstanceID);", inst);
break;
case IR::Attribute::VertexId:
ctx.AddU32("{}=uint(gl_VertexID);", inst);
break;
case IR::Attribute::BaseInstance:
ctx.AddU32("{}=uint(gl_BaseInstance);", inst);
break;
case IR::Attribute::BaseVertex:
ctx.AddU32("{}=uint(gl_BaseVertex);", inst);
break;
case IR::Attribute::DrawID:
ctx.AddU32("{}=uint(gl_DrawID);", inst);
break;
default:
throw NotImplementedException("Get U32 attribute {}", attr);
}
}
void EmitSetAttribute(EmitContext& ctx, IR::Attribute attr, std::string_view value,
[[maybe_unused]] std::string_view vertex) {
if (IR::IsGeneric(attr)) {
const u32 index{IR::GenericAttributeIndex(attr)};
const u32 attr_element{IR::GenericAttributeElement(attr)};
const GenericElementInfo& info{ctx.output_generics.at(index).at(attr_element)};
const auto output_decorator{OutputVertexIndex(ctx)};
if (info.num_components == 1) {
ctx.Add("{}{}={};", info.name, output_decorator, value);
} else {
const u32 index_element{attr_element - info.first_element};
ctx.Add("{}{}.{}={};", info.name, output_decorator, "xyzw"[index_element], value);
}
return;
}
const u32 element{static_cast<u32>(attr) % 4};
const char swizzle{"xyzw"[element]};
switch (attr) {
case IR::Attribute::Layer:
if (ctx.stage != Stage::Geometry &&
!ctx.profile.support_viewport_index_layer_non_geometry) {
LOG_WARNING(Shader_GLSL, "Shader stores viewport layer but device does not support "
"viewport layer extension");
break;
}
ctx.Add("gl_Layer=ftoi({});", value);
break;
case IR::Attribute::ViewportIndex:
if (ctx.stage != Stage::Geometry &&
!ctx.profile.support_viewport_index_layer_non_geometry) {
LOG_WARNING(Shader_GLSL, "Shader stores viewport index but device does not support "
"viewport layer extension");
break;
}
ctx.Add("gl_ViewportIndex=ftoi({});", value);
break;
case IR::Attribute::ViewportMask:
if (ctx.stage != Stage::Geometry && !ctx.profile.support_viewport_mask) {
LOG_WARNING(
Shader_GLSL,
"Shader stores viewport mask but device does not support viewport mask extension");
break;
}
ctx.Add("gl_ViewportMask[0]=ftoi({});", value);
break;
case IR::Attribute::PointSize:
ctx.Add("gl_PointSize={};", value);
break;
case IR::Attribute::PositionX:
case IR::Attribute::PositionY:
case IR::Attribute::PositionZ:
case IR::Attribute::PositionW:
ctx.Add("gl_Position.{}={};", swizzle, value);
break;
case IR::Attribute::ClipDistance0:
case IR::Attribute::ClipDistance1:
case IR::Attribute::ClipDistance2:
case IR::Attribute::ClipDistance3:
case IR::Attribute::ClipDistance4:
case IR::Attribute::ClipDistance5:
case IR::Attribute::ClipDistance6:
case IR::Attribute::ClipDistance7: {
const u32 index{static_cast<u32>(attr) - static_cast<u32>(IR::Attribute::ClipDistance0)};
ctx.Add("gl_ClipDistance[{}]={};", index, value);
break;
}
default:
throw NotImplementedException("Set attribute {}", attr);
}
}
void EmitGetAttributeIndexed(EmitContext& ctx, IR::Inst& inst, std::string_view offset,
std::string_view vertex) {
const bool is_array{ctx.stage == Stage::Geometry};
const auto vertex_arg{is_array ? fmt::format(",{}", vertex) : ""};
ctx.AddF32("{}=IndexedAttrLoad(int({}){});", inst, offset, vertex_arg);
}
void EmitSetAttributeIndexed([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view offset,
[[maybe_unused]] std::string_view value,
[[maybe_unused]] std::string_view vertex) {
NotImplemented();
}
void EmitGetPatch(EmitContext& ctx, IR::Inst& inst, IR::Patch patch) {
if (!IR::IsGeneric(patch)) {
throw NotImplementedException("Non-generic patch load");
}
const u32 index{IR::GenericPatchIndex(patch)};
const u32 element{IR::GenericPatchElement(patch)};
const char swizzle{"xyzw"[element]};
ctx.AddF32("{}=patch{}.{};", inst, index, swizzle);
}
void EmitSetPatch(EmitContext& ctx, IR::Patch patch, std::string_view value) {
if (IR::IsGeneric(patch)) {
const u32 index{IR::GenericPatchIndex(patch)};
const u32 element{IR::GenericPatchElement(patch)};
ctx.Add("patch{}.{}={};", index, "xyzw"[element], value);
return;
}
switch (patch) {
case IR::Patch::TessellationLodLeft:
case IR::Patch::TessellationLodRight:
case IR::Patch::TessellationLodTop:
case IR::Patch::TessellationLodBottom: {
const u32 index{static_cast<u32>(patch) - u32(IR::Patch::TessellationLodLeft)};
ctx.Add("gl_TessLevelOuter[{}]={};", index, value);
break;
}
case IR::Patch::TessellationLodInteriorU:
ctx.Add("gl_TessLevelInner[0]={};", value);
break;
case IR::Patch::TessellationLodInteriorV:
ctx.Add("gl_TessLevelInner[1]={};", value);
break;
default:
throw NotImplementedException("Patch {}", patch);
}
}
void EmitSetFragColor(EmitContext& ctx, u32 index, u32 component, std::string_view value) {
const char swizzle{"xyzw"[component]};
ctx.Add("frag_color{}.{}={};", index, swizzle, value);
}
void EmitSetSampleMask(EmitContext& ctx, std::string_view value) {
ctx.Add("gl_SampleMask[0]=int({});", value);
}
void EmitSetFragDepth(EmitContext& ctx, std::string_view value) {
ctx.Add("gl_FragDepth={};", value);
}
void EmitLocalInvocationId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32x3("{}=gl_LocalInvocationID;", inst);
}
void EmitWorkgroupId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32x3("{}=gl_WorkGroupID;", inst);
}
void EmitInvocationId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=uint(gl_InvocationID);", inst);
}
void EmitInvocationInfo(EmitContext& ctx, IR::Inst& inst) {
switch (ctx.stage) {
case Stage::TessellationControl:
case Stage::TessellationEval:
ctx.AddU32("{}=uint(gl_PatchVerticesIn)<<16;", inst);
break;
case Stage::Geometry:
ctx.AddU32("{}=uint({}<<16);", inst,
InputTopologyVertices::vertices(ctx.runtime_info.input_topology));
break;
default:
LOG_WARNING(Shader, "(STUBBED) called");
ctx.AddU32("{}=uint(0x00ff0000);", inst);
}
}
void EmitSampleId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=uint(gl_SampleID);", inst);
}
void EmitIsHelperInvocation(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU1("{}=gl_HelperInvocation;", inst);
}
void EmitSR_WScaleFactorXY(EmitContext& ctx, IR::Inst& inst) {
LOG_WARNING(Shader, "(STUBBED) called");
}
void EmitSR_WScaleFactorZ(EmitContext& ctx, IR::Inst& inst) {
LOG_WARNING(Shader, "(STUBBED) called");
}
void EmitYDirection(EmitContext& ctx, IR::Inst& inst) {
ctx.uses_y_direction = true;
ctx.AddF32("{}=gl_FrontMaterial.ambient.a;", inst);
}
void EmitResolutionDownFactor(EmitContext& ctx, IR::Inst& inst) {
ctx.AddF32("{}=scaling.z;", inst);
}
void EmitRenderArea(EmitContext& ctx, IR::Inst& inst) {
ctx.AddF32x4("{}=render_area;", inst);
}
void EmitLoadLocal(EmitContext& ctx, IR::Inst& inst, std::string_view word_offset) {
ctx.AddU32("{}=lmem[{}];", inst, word_offset);
}
void EmitWriteLocal(EmitContext& ctx, std::string_view word_offset, std::string_view value) {
ctx.Add("lmem[{}]={};", word_offset, value);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,18 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/exception.h"
namespace Shader::Backend::GLSL {
void EmitJoin(EmitContext&) {
throw NotImplementedException("Join shouldn't be emitted");
}
void EmitDemoteToHelperInvocation(EmitContext& ctx) {
ctx.Add("discard;");
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,259 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
void EmitConvertS16F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS16F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=(int({})&0xffff)|(bitfieldExtract(int({}),31,1)<<15);", inst, value, value);
}
void EmitConvertS16F64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS32F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=int({});", inst, value);
}
void EmitConvertS32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=int({});", inst, value);
}
void EmitConvertS64F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=int64_t({});", inst, value);
}
void EmitConvertS64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=int64_t({});", inst, value);
}
void EmitConvertU16F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU16F32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU16F64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU32F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU16U32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU32U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU8U32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU32U8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS32S8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS32S16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=uint({});", inst, value);
}
void EmitConvertU32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=uint({});", inst, value);
}
void EmitConvertU64F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=uint64_t({});", inst, value);
}
void EmitConvertU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=uint64_t({});", inst, value);
}
void EmitConvertU64U32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=uint64_t({});", inst, value);
}
void EmitConvertU32U64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=uint({});", inst, value);
}
void EmitConvertF16F32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=float({});", inst, value);
}
void EmitConvertF64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=double({});", inst, value);
}
void EmitConvertF16S8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16S16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16S32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16S64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16U8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16U32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16U64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32S8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32S16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32S32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=float(int({}));", inst, value);
}
void EmitConvertF32S64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=float(int64_t({}));", inst, value);
}
void EmitConvertF32U8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32U16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=float({}&0xffff);", inst, value);
}
void EmitConvertF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=float({});", inst, value);
}
void EmitConvertF32U64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=float({});", inst, value);
}
void EmitConvertF64S8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF64S16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF64S32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=double(int({}));", inst, value);
}
void EmitConvertF64S64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=double(int64_t({}));", inst, value);
}
void EmitConvertF64U8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF64U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF64U32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=double({});", inst, value);
}
void EmitConvertF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=double({});", inst, value);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,455 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/modifiers.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
namespace {
void Compare(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs,
std::string_view op, bool ordered) {
const auto nan_op{ordered ? "&&!" : "||"};
ctx.AddU1("{}={}{}{}"
"{}isnan({}){}isnan({});",
inst, lhs, op, rhs, nan_op, lhs, nan_op, rhs);
}
bool IsPrecise(const IR::Inst& inst) {
return inst.Flags<IR::FpControl>().no_contraction;
}
} // Anonymous namespace
void EmitFPAbs16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=abs({});", inst, value);
}
void EmitFPAbs64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=abs({});", inst, value);
}
void EmitFPAdd16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view a, [[maybe_unused]] std::string_view b) {
NotImplemented();
}
void EmitFPAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
if (IsPrecise(inst)) {
ctx.AddPrecF32("{}={}+{};", inst, a, b);
} else {
ctx.AddF32("{}={}+{};", inst, a, b);
}
}
void EmitFPAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
if (IsPrecise(inst)) {
ctx.AddPrecF64("{}={}+{};", inst, a, b);
} else {
ctx.AddF64("{}={}+{};", inst, a, b);
}
}
void EmitFPFma16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view a, [[maybe_unused]] std::string_view b,
[[maybe_unused]] std::string_view c) {
NotImplemented();
}
void EmitFPFma32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c) {
if (IsPrecise(inst)) {
ctx.AddPrecF32("{}=fma({},{},{});", inst, a, b, c);
} else {
ctx.AddF32("{}=fma({},{},{});", inst, a, b, c);
}
}
void EmitFPFma64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c) {
if (IsPrecise(inst)) {
ctx.AddPrecF64("{}=fma({},{},{});", inst, a, b, c);
} else {
ctx.AddF64("{}=fma({},{},{});", inst, a, b, c);
}
}
void EmitFPMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddF32("{}=max({},{});", inst, a, b);
}
void EmitFPMax64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddF64("{}=max({},{});", inst, a, b);
}
void EmitFPMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddF32("{}=min({},{});", inst, a, b);
}
void EmitFPMin64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddF64("{}=min({},{});", inst, a, b);
}
void EmitFPMul16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view a, [[maybe_unused]] std::string_view b) {
NotImplemented();
}
void EmitFPMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
if (IsPrecise(inst)) {
ctx.AddPrecF32("{}={}*{};", inst, a, b);
} else {
ctx.AddF32("{}={}*{};", inst, a, b);
}
}
void EmitFPMul64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
if (IsPrecise(inst)) {
ctx.AddPrecF64("{}={}*{};", inst, a, b);
} else {
ctx.AddF64("{}={}*{};", inst, a, b);
}
}
void EmitFPNeg16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPNeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=0.f-({});", inst, value);
}
void EmitFPNeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=double(0.)-({});", inst, value);
}
void EmitFPSin(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=sin({});", inst, value);
}
void EmitFPCos(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=cos({});", inst, value);
}
void EmitFPExp2(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=exp2({});", inst, value);
}
void EmitFPLog2(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=log2({});", inst, value);
}
void EmitFPRecip32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=(1.0f)/{};", inst, value);
}
void EmitFPRecip64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=1.0/{};", inst, value);
}
void EmitFPRecipSqrt32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
ctx.AddF32("{}=inversesqrt({});", inst, value);
}
void EmitFPRecipSqrt64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPSqrt(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=sqrt({});", inst, value);
}
void EmitFPSaturate16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPSaturate32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=min(max({},0.0),1.0);", inst, value);
}
void EmitFPSaturate64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=min(max({},0.0),1.0);", inst, value);
}
void EmitFPClamp16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value,
[[maybe_unused]] std::string_view min_value,
[[maybe_unused]] std::string_view max_value) {
NotImplemented();
}
void EmitFPClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value) {
// GLSL's clamp does not produce desirable results
ctx.AddF32("{}=min(max({},float({})),float({}));", inst, value, min_value, max_value);
}
void EmitFPClamp64(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value) {
// GLSL's clamp does not produce desirable results
ctx.AddF64("{}=min(max({},double({})),double({}));", inst, value, min_value, max_value);
}
void EmitFPRoundEven16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPRoundEven32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=roundEven({});", inst, value);
}
void EmitFPRoundEven64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=roundEven({});", inst, value);
}
void EmitFPFloor16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPFloor32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=floor({});", inst, value);
}
void EmitFPFloor64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=floor({});", inst, value);
}
void EmitFPCeil16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPCeil32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=ceil({});", inst, value);
}
void EmitFPCeil64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=ceil({});", inst, value);
}
void EmitFPTrunc16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPTrunc32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=trunc({});", inst, value);
}
void EmitFPTrunc64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=trunc({});", inst, value);
}
void EmitFPOrdEqual16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "==", true);
}
void EmitFPOrdEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "==", true);
}
void EmitFPUnordEqual16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "==", false);
}
void EmitFPUnordEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "==", false);
}
void EmitFPOrdNotEqual16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdNotEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "!=", true);
}
void EmitFPOrdNotEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "!=", true);
}
void EmitFPUnordNotEqual16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordNotEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "!=", false);
}
void EmitFPUnordNotEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "!=", false);
}
void EmitFPOrdLessThan16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdLessThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<", true);
}
void EmitFPOrdLessThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<", true);
}
void EmitFPUnordLessThan16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordLessThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<", false);
}
void EmitFPUnordLessThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<", false);
}
void EmitFPOrdGreaterThan16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdGreaterThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">", true);
}
void EmitFPOrdGreaterThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">", true);
}
void EmitFPUnordGreaterThan16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordGreaterThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">", false);
}
void EmitFPUnordGreaterThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">", false);
}
void EmitFPOrdLessThanEqual16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdLessThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<=", true);
}
void EmitFPOrdLessThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<=", true);
}
void EmitFPUnordLessThanEqual16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordLessThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<=", false);
}
void EmitFPUnordLessThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<=", false);
}
void EmitFPOrdGreaterThanEqual16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdGreaterThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">=", true);
}
void EmitFPOrdGreaterThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">=", true);
}
void EmitFPUnordGreaterThanEqual16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordGreaterThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">=", false);
}
void EmitFPUnordGreaterThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">=", false);
}
void EmitFPIsNan16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPIsNan32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU1("{}=isnan({});", inst, value);
}
void EmitFPIsNan64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU1("{}=isnan({});", inst, value);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,869 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/modifiers.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::GLSL {
namespace {
std::string Texture(EmitContext& ctx, const IR::TextureInstInfo& info, const IR::Value& index) {
const auto def{info.type == TextureType::Buffer ? ctx.texture_buffers.at(info.descriptor_index)
: ctx.textures.at(info.descriptor_index)};
const auto index_offset{def.count > 1 ? fmt::format("[{}]", ctx.var_alloc.Consume(index)) : ""};
return fmt::format("tex{}{}", def.binding, index_offset);
}
std::string Image(EmitContext& ctx, const IR::TextureInstInfo& info, const IR::Value& index) {
const auto def{info.type == TextureType::Buffer ? ctx.image_buffers.at(info.descriptor_index)
: ctx.images.at(info.descriptor_index)};
const auto index_offset{def.count > 1 ? fmt::format("[{}]", ctx.var_alloc.Consume(index)) : ""};
return fmt::format("img{}{}", def.binding, index_offset);
}
bool IsTextureMsaa(EmitContext& ctx, const IR::TextureInstInfo& info) {
if (info.type == TextureType::Buffer) {
return false;
}
return ctx.info.texture_descriptors.at(info.descriptor_index).is_multisample;
}
std::string CastToIntVec(std::string_view value, const IR::TextureInstInfo& info) {
switch (info.type) {
case TextureType::Color1D:
case TextureType::Buffer:
return fmt::format("int({})", value);
case TextureType::ColorArray1D:
case TextureType::Color2D:
case TextureType::ColorArray2D:
return fmt::format("ivec2({})", value);
case TextureType::Color3D:
case TextureType::ColorCube:
return fmt::format("ivec3({})", value);
case TextureType::ColorArrayCube:
return fmt::format("ivec4({})", value);
default:
throw NotImplementedException("Integer cast for TextureType {}", info.type.Value());
}
}
std::string CoordsCastToInt(std::string_view value, const IR::TextureInstInfo& info) {
switch (info.type) {
case TextureType::Color1D:
case TextureType::Buffer:
return fmt::format("int({})", value);
case TextureType::ColorArray1D:
case TextureType::Color2D:
return fmt::format("ivec2({})", value);
case TextureType::ColorArray2D:
case TextureType::Color3D:
case TextureType::ColorCube:
return fmt::format("ivec3({})", value);
case TextureType::ColorArrayCube:
return fmt::format("ivec4({})", value);
default:
throw NotImplementedException("TexelFetchCast type {}", info.type.Value());
}
}
bool NeedsShadowLodExt(TextureType type) {
switch (type) {
case TextureType::ColorArray2D:
case TextureType::ColorCube:
case TextureType::ColorArrayCube:
return true;
default:
return false;
}
}
std::string GetOffsetVec(EmitContext& ctx, const IR::Value& offset) {
if (offset.IsImmediate()) {
return fmt::format("int({})", offset.U32());
}
IR::Inst* const inst{offset.InstRecursive()};
if (inst->AreAllArgsImmediates()) {
switch (inst->GetOpcode()) {
case IR::Opcode::CompositeConstructU32x2:
return fmt::format("ivec2({},{})", inst->Arg(0).U32(), inst->Arg(1).U32());
case IR::Opcode::CompositeConstructU32x3:
return fmt::format("ivec3({},{},{})", inst->Arg(0).U32(), inst->Arg(1).U32(),
inst->Arg(2).U32());
case IR::Opcode::CompositeConstructU32x4:
return fmt::format("ivec4({},{},{},{})", inst->Arg(0).U32(), inst->Arg(1).U32(),
inst->Arg(2).U32(), inst->Arg(3).U32());
default:
break;
}
}
const bool has_var_aoffi{ctx.profile.support_gl_variable_aoffi};
if (!has_var_aoffi) {
LOG_WARNING(Shader_GLSL, "Device does not support variable texture offsets, STUBBING");
}
const auto offset_str{has_var_aoffi ? ctx.var_alloc.Consume(offset) : "0"};
switch (offset.Type()) {
case IR::Type::U32:
return fmt::format("int({})", offset_str);
case IR::Type::U32x2:
return fmt::format("ivec2({})", offset_str);
case IR::Type::U32x3:
return fmt::format("ivec3({})", offset_str);
case IR::Type::U32x4:
return fmt::format("ivec4({})", offset_str);
default:
throw NotImplementedException("Offset type {}", offset.Type());
}
}
std::string PtpOffsets(const IR::Value& offset, const IR::Value& offset2) {
const std::array values{offset.InstRecursive(), offset2.InstRecursive()};
if (!values[0]->AreAllArgsImmediates() || !values[1]->AreAllArgsImmediates()) {
LOG_WARNING(Shader_GLSL, "Not all arguments in PTP are immediate, STUBBING");
return "ivec2[](ivec2(0), ivec2(1), ivec2(2), ivec2(3))";
}
const IR::Opcode opcode{values[0]->GetOpcode()};
if (opcode != values[1]->GetOpcode() || opcode != IR::Opcode::CompositeConstructU32x4) {
throw LogicError("Invalid PTP arguments");
}
auto read{[&](unsigned int a, unsigned int b) { return values[a]->Arg(b).U32(); }};
return fmt::format("ivec2[](ivec2({},{}),ivec2({},{}),ivec2({},{}),ivec2({},{}))", read(0, 0),
read(0, 1), read(0, 2), read(0, 3), read(1, 0), read(1, 1), read(1, 2),
read(1, 3));
}
IR::Inst* PrepareSparse(IR::Inst& inst) {
const auto sparse_inst{inst.GetAssociatedPseudoOperation(IR::Opcode::GetSparseFromOp)};
if (sparse_inst) {
sparse_inst->Invalidate();
}
return sparse_inst;
}
std::string ImageGatherSubpixelOffset(const IR::TextureInstInfo& info, std::string_view texture,
std::string_view coords) {
switch (info.type) {
case TextureType::Color2D:
case TextureType::Color2DRect:
return fmt::format("{}+vec2(0.001953125)/vec2(textureSize({}, 0))", coords, texture);
case TextureType::ColorArray2D:
case TextureType::ColorCube:
return fmt::format("vec3({0}.xy+vec2(0.001953125)/vec2(textureSize({1}, 0)),{0}.z)", coords,
texture);
default:
return std::string{coords};
}
}
} // Anonymous namespace
void EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view bias_lc,
const IR::Value& offset) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageSampleImplicitLod Lod clamp samples");
}
const auto texture{Texture(ctx, info, index)};
const auto bias{info.has_bias ? fmt::format(",{}", bias_lc) : ""};
const auto texel{ctx.var_alloc.Define(inst, GlslVarType::F32x4)};
const auto sparse_inst{PrepareSparse(inst)};
const bool supports_sparse{ctx.profile.support_gl_sparse_textures};
if (sparse_inst && !supports_sparse) {
LOG_WARNING(Shader_GLSL, "Device does not support sparse texture queries. STUBBING");
ctx.AddU1("{}=true;", *sparse_inst);
}
if (!sparse_inst || !supports_sparse) {
if (!offset.IsEmpty()) {
const auto offset_str{GetOffsetVec(ctx, offset)};
if (ctx.stage == Stage::Fragment) {
ctx.Add("{}=textureOffset({},{},{}{});", texel, texture, coords, offset_str, bias);
} else {
ctx.Add("{}=textureLodOffset({},{},0.0,{});", texel, texture, coords, offset_str);
}
} else {
if (ctx.stage == Stage::Fragment) {
ctx.Add("{}=texture({},{}{});", texel, texture, coords, bias);
} else {
ctx.Add("{}=textureLod({},{},0.0);", texel, texture, coords);
}
}
return;
}
if (!offset.IsEmpty()) {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureOffsetARB({},{},{},{}{}));",
*sparse_inst, texture, coords, GetOffsetVec(ctx, offset), texel, bias);
} else {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureARB({},{},{}{}));", *sparse_inst,
texture, coords, texel, bias);
}
}
void EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view lod_lc,
const IR::Value& offset) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_bias) {
throw NotImplementedException("EmitImageSampleExplicitLod Bias texture samples");
}
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageSampleExplicitLod Lod clamp samples");
}
const auto texture{Texture(ctx, info, index)};
const auto texel{ctx.var_alloc.Define(inst, GlslVarType::F32x4)};
const auto sparse_inst{PrepareSparse(inst)};
const bool supports_sparse{ctx.profile.support_gl_sparse_textures};
if (sparse_inst && !supports_sparse) {
LOG_WARNING(Shader_GLSL, "Device does not support sparse texture queries. STUBBING");
ctx.AddU1("{}=true;", *sparse_inst);
}
if (!sparse_inst || !supports_sparse) {
if (!offset.IsEmpty()) {
ctx.Add("{}=textureLodOffset({},{},{},{});", texel, texture, coords, lod_lc,
GetOffsetVec(ctx, offset));
} else {
ctx.Add("{}=textureLod({},{},{});", texel, texture, coords, lod_lc);
}
return;
}
if (!offset.IsEmpty()) {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTexelFetchOffsetARB({},{},int({}),{},{}));",
*sparse_inst, texture, CastToIntVec(coords, info), lod_lc,
GetOffsetVec(ctx, offset), texel);
} else {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureLodARB({},{},{},{}));", *sparse_inst,
texture, coords, lod_lc, texel);
}
}
void EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view dref,
std::string_view bias_lc, const IR::Value& offset) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto sparse_inst{PrepareSparse(inst)};
if (sparse_inst) {
throw NotImplementedException("EmitImageSampleDrefImplicitLod Sparse texture samples");
}
if (info.has_bias) {
throw NotImplementedException("EmitImageSampleDrefImplicitLod Bias texture samples");
}
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageSampleDrefImplicitLod Lod clamp samples");
}
const auto texture{Texture(ctx, info, index)};
const auto bias{info.has_bias ? fmt::format(",{}", bias_lc) : ""};
const bool needs_shadow_ext{NeedsShadowLodExt(info.type)};
const auto cast{needs_shadow_ext ? "vec4" : "vec3"};
const bool use_grad{!ctx.profile.support_gl_texture_shadow_lod &&
ctx.stage != Stage::Fragment && needs_shadow_ext};
if (use_grad) {
LOG_WARNING(Shader_GLSL,
"Device lacks GL_EXT_texture_shadow_lod. Using textureGrad fallback");
if (info.type == TextureType::ColorArrayCube) {
LOG_WARNING(Shader_GLSL, "textureGrad does not support ColorArrayCube. Stubbing");
ctx.AddF32("{}=0.0f;", inst);
return;
}
const auto d_cast{info.type == TextureType::ColorArray2D ? "vec2" : "vec3"};
ctx.AddF32("{}=textureGrad({},{}({},{}),{}(0),{}(0));", inst, texture, cast, coords, dref,
d_cast, d_cast);
return;
}
if (!offset.IsEmpty()) {
const auto offset_str{GetOffsetVec(ctx, offset)};
if (ctx.stage == Stage::Fragment) {
ctx.AddF32("{}=textureOffset({},{}({},{}),{}{});", inst, texture, cast, coords, dref,
offset_str, bias);
} else {
ctx.AddF32("{}=textureLodOffset({},{}({},{}),0.0,{});", inst, texture, cast, coords,
dref, offset_str);
}
} else {
if (ctx.stage == Stage::Fragment) {
if (info.type == TextureType::ColorArrayCube) {
ctx.AddF32("{}=texture({},vec4({}),{});", inst, texture, coords, dref);
} else {
ctx.AddF32("{}=texture({},{}({},{}){});", inst, texture, cast, coords, dref, bias);
}
} else {
ctx.AddF32("{}=textureLod({},{}({},{}),0.0);", inst, texture, cast, coords, dref);
}
}
}
void EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view dref,
std::string_view lod_lc, const IR::Value& offset) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto sparse_inst{PrepareSparse(inst)};
if (sparse_inst) {
throw NotImplementedException("EmitImageSampleDrefExplicitLod Sparse texture samples");
}
if (info.has_bias) {
throw NotImplementedException("EmitImageSampleDrefExplicitLod Bias texture samples");
}
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageSampleDrefExplicitLod Lod clamp samples");
}
const auto texture{Texture(ctx, info, index)};
const bool needs_shadow_ext{NeedsShadowLodExt(info.type)};
const bool use_grad{!ctx.profile.support_gl_texture_shadow_lod && needs_shadow_ext};
const auto cast{needs_shadow_ext ? "vec4" : "vec3"};
if (use_grad) {
LOG_WARNING(Shader_GLSL,
"Device lacks GL_EXT_texture_shadow_lod. Using textureGrad fallback");
if (info.type == TextureType::ColorArrayCube) {
LOG_WARNING(Shader_GLSL, "textureGrad does not support ColorArrayCube. Stubbing");
ctx.AddF32("{}=0.0f;", inst);
return;
}
const auto d_cast{info.type == TextureType::ColorArray2D ? "vec2" : "vec3"};
ctx.AddF32("{}=textureGrad({},{}({},{}),{}(0),{}(0));", inst, texture, cast, coords, dref,
d_cast, d_cast);
return;
}
if (!offset.IsEmpty()) {
const auto offset_str{GetOffsetVec(ctx, offset)};
if (info.type == TextureType::ColorArrayCube) {
ctx.AddF32("{}=textureLodOffset({},{},{},{},{});", inst, texture, coords, dref, lod_lc,
offset_str);
} else {
ctx.AddF32("{}=textureLodOffset({},{}({},{}),{},{});", inst, texture, cast, coords,
dref, lod_lc, offset_str);
}
} else {
if (info.type == TextureType::ColorArrayCube) {
ctx.AddF32("{}=textureLod({},{},{},{});", inst, texture, coords, dref, lod_lc);
} else {
ctx.AddF32("{}=textureLod({},{}({},{}),{});", inst, texture, cast, coords, dref,
lod_lc);
}
}
}
void EmitImageGather(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, const IR::Value& offset2) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto texture{Texture(ctx, info, index)};
const auto texel{ctx.var_alloc.Define(inst, GlslVarType::F32x4)};
const auto sparse_inst{PrepareSparse(inst)};
const bool supports_sparse{ctx.profile.support_gl_sparse_textures};
if (sparse_inst && !supports_sparse) {
LOG_WARNING(Shader_GLSL, "Device does not support sparse texture queries. STUBBING");
ctx.AddU1("{}=true;", *sparse_inst);
}
std::string coords_with_subpixel_offset;
if (ctx.profile.need_gather_subpixel_offset) {
// Apply a subpixel offset of 1/512 the texel size of the texture to ensure same rounding on
// AMD hardware as on Maxwell or other Nvidia architectures.
coords_with_subpixel_offset = ImageGatherSubpixelOffset(info, texture, coords);
coords = coords_with_subpixel_offset;
}
if (!sparse_inst || !supports_sparse) {
if (offset.IsEmpty()) {
ctx.Add("{}=textureGather({},{},int({}));", texel, texture, coords,
info.gather_component);
return;
}
if (offset2.IsEmpty()) {
ctx.Add("{}=textureGatherOffset({},{},{},int({}));", texel, texture, coords,
GetOffsetVec(ctx, offset), info.gather_component);
return;
}
// PTP
const auto offsets{PtpOffsets(offset, offset2)};
ctx.Add("{}=textureGatherOffsets({},{},{},int({}));", texel, texture, coords, offsets,
info.gather_component);
return;
}
if (offset.IsEmpty()) {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureGatherARB({},{},{},int({})));",
*sparse_inst, texture, coords, texel, info.gather_component);
return;
}
if (offset2.IsEmpty()) {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureGatherOffsetARB({},{},{},{},int({})));",
*sparse_inst, texture, CastToIntVec(coords, info), GetOffsetVec(ctx, offset),
texel, info.gather_component);
return;
}
// PTP
const auto offsets{PtpOffsets(offset, offset2)};
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureGatherOffsetARB({},{},{},{},int({})));",
*sparse_inst, texture, CastToIntVec(coords, info), offsets, texel,
info.gather_component);
}
void EmitImageGatherDref(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, const IR::Value& offset2,
std::string_view dref) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto texture{Texture(ctx, info, index)};
const auto texel{ctx.var_alloc.Define(inst, GlslVarType::F32x4)};
const auto sparse_inst{PrepareSparse(inst)};
const bool supports_sparse{ctx.profile.support_gl_sparse_textures};
if (sparse_inst && !supports_sparse) {
LOG_WARNING(Shader_GLSL, "Device does not support sparse texture queries. STUBBING");
ctx.AddU1("{}=true;", *sparse_inst);
}
std::string coords_with_subpixel_offset;
if (ctx.profile.need_gather_subpixel_offset) {
// Apply a subpixel offset of 1/512 the texel size of the texture to ensure same rounding on
// AMD hardware as on Maxwell or other Nvidia architectures.
coords_with_subpixel_offset = ImageGatherSubpixelOffset(info, texture, coords);
coords = coords_with_subpixel_offset;
}
if (!sparse_inst || !supports_sparse) {
if (offset.IsEmpty()) {
ctx.Add("{}=textureGather({},{},{});", texel, texture, coords, dref);
return;
}
if (offset2.IsEmpty()) {
ctx.Add("{}=textureGatherOffset({},{},{},{});", texel, texture, coords, dref,
GetOffsetVec(ctx, offset));
return;
}
// PTP
const auto offsets{PtpOffsets(offset, offset2)};
ctx.Add("{}=textureGatherOffsets({},{},{},{});", texel, texture, coords, dref, offsets);
return;
}
if (offset.IsEmpty()) {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureGatherARB({},{},{},{}));", *sparse_inst,
texture, coords, dref, texel);
return;
}
if (offset2.IsEmpty()) {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureGatherOffsetARB({},{},{},,{},{}));",
*sparse_inst, texture, CastToIntVec(coords, info), dref,
GetOffsetVec(ctx, offset), texel);
return;
}
// PTP
const auto offsets{PtpOffsets(offset, offset2)};
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTextureGatherOffsetARB({},{},{},,{},{}));",
*sparse_inst, texture, CastToIntVec(coords, info), dref, offsets, texel);
}
void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, std::string_view lod,
std::string_view ms) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_bias) {
throw NotImplementedException("EmitImageFetch Bias texture samples");
}
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageFetch Lod clamp samples");
}
const auto texture{Texture(ctx, info, index)};
const auto sparse_inst{PrepareSparse(inst)};
const auto texel{ctx.var_alloc.Define(inst, GlslVarType::F32x4)};
const bool supports_sparse{ctx.profile.support_gl_sparse_textures};
if (sparse_inst && !supports_sparse) {
LOG_WARNING(Shader_GLSL, "Device does not support sparse texture queries. STUBBING");
ctx.AddU1("{}=true;", *sparse_inst);
}
if (!sparse_inst || !supports_sparse) {
const auto int_coords{CoordsCastToInt(coords, info)};
if (!ms.empty()) {
ctx.Add("{}=texelFetch({},{},int({}));", texel, texture, int_coords, ms);
} else if (!offset.IsEmpty()) {
ctx.Add("{}=texelFetchOffset({},{},int({}),{});", texel, texture, int_coords, lod,
GetOffsetVec(ctx, offset));
} else {
if (info.type == TextureType::Buffer) {
ctx.Add("{}=texelFetch({},int({}));", texel, texture, coords);
} else {
ctx.Add("{}=texelFetch({},{},int({}));", texel, texture, int_coords, lod);
}
}
return;
}
if (!ms.empty()) {
throw NotImplementedException("EmitImageFetch Sparse MSAA samples");
}
if (!offset.IsEmpty()) {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTexelFetchOffsetARB({},{},int({}),{},{}));",
*sparse_inst, texture, CastToIntVec(coords, info), lod, GetOffsetVec(ctx, offset),
texel);
} else {
ctx.AddU1("{}=sparseTexelsResidentARB(sparseTexelFetchARB({},{},int({}),{}));",
*sparse_inst, texture, CastToIntVec(coords, info), lod, texel);
}
}
void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view lod, const IR::Value& skip_mips_val) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto texture{Texture(ctx, info, index)};
const bool is_msaa{IsTextureMsaa(ctx, info)};
const bool skip_mips{skip_mips_val.U1()};
const auto mips{skip_mips ? "0u" : fmt::format("uint(textureQueryLevels({}))", texture)};
if (is_msaa && !skip_mips) {
throw NotImplementedException("EmitImageQueryDimensions MSAA QueryLevels");
}
if (info.type == TextureType::Buffer && !skip_mips) {
throw NotImplementedException("EmitImageQueryDimensions TextureType::Buffer QueryLevels");
}
const bool uses_lod{!is_msaa && info.type != TextureType::Buffer};
const auto lod_str{uses_lod ? fmt::format(",int({})", lod) : ""};
switch (info.type) {
case TextureType::Color1D:
return ctx.AddU32x4("{}=uvec4(uint(textureSize({}{})),0u,0u,{});", inst, texture, lod_str,
mips);
case TextureType::ColorArray1D:
case TextureType::Color2D:
case TextureType::ColorCube:
case TextureType::Color2DRect:
return ctx.AddU32x4("{}=uvec4(uvec2(textureSize({}{})),0u,{});", inst, texture, lod_str,
mips);
case TextureType::ColorArray2D:
case TextureType::Color3D:
case TextureType::ColorArrayCube:
return ctx.AddU32x4("{}=uvec4(uvec3(textureSize({}{})),{});", inst, texture, lod_str, mips);
case TextureType::Buffer:
return ctx.AddU32x4("{}=uvec4(uint(textureSize({})),0u,0u,{});", inst, texture, mips);
}
throw LogicError("Unspecified image type {}", info.type.Value());
}
void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto texture{Texture(ctx, info, index)};
return ctx.AddF32x4("{}=vec4(textureQueryLod({},{}),0.0,0.0);", inst, texture, coords);
}
void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& derivatives,
const IR::Value& offset, [[maybe_unused]] const IR::Value& lod_clamp) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageGradient Lod clamp samples");
}
const auto sparse_inst{PrepareSparse(inst)};
if (sparse_inst) {
throw NotImplementedException("EmitImageGradient Sparse");
}
if (!offset.IsEmpty() && info.num_derivatives <= 2) {
throw NotImplementedException("EmitImageGradient offset");
}
const auto texture{Texture(ctx, info, index)};
const auto texel{ctx.var_alloc.Define(inst, GlslVarType::F32x4)};
const bool multi_component{info.num_derivatives > 1 || info.has_lod_clamp};
const auto derivatives_vec{ctx.var_alloc.Consume(derivatives)};
if (multi_component) {
if (info.num_derivatives >= 3) {
const auto offset_vec{ctx.var_alloc.Consume(offset)};
ctx.Add("{}=textureGrad({},{},vec3({}.xz, {}.x),vec3({}.yw, {}.y));", texel, texture,
coords, derivatives_vec, offset_vec, derivatives_vec, offset_vec);
return;
}
ctx.Add("{}=textureGrad({},{},vec2({}.xz),vec2({}.yz));", texel, texture, coords,
derivatives_vec, derivatives_vec);
} else {
ctx.Add("{}=textureGrad({},{},float({}.x),float({}.y));", texel, texture, coords,
derivatives_vec, derivatives_vec);
}
}
void EmitImageRead(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto sparse_inst{PrepareSparse(inst)};
if (sparse_inst) {
throw NotImplementedException("EmitImageRead Sparse");
}
const auto image{Image(ctx, info, index)};
ctx.AddU32x4("{}=uvec4(imageLoad({},{}));", inst, image, CoordsCastToInt(coords, info));
}
void EmitImageWrite(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view color) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.Add("imageStore({},{},{});", image, CoordsCastToInt(coords, info), color);
}
void EmitImageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicAdd({},{},{});", inst, image, CoordsCastToInt(coords, info), value);
}
void EmitImageAtomicSMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicMin({},{},int({}));", inst, image, CoordsCastToInt(coords, info),
value);
}
void EmitImageAtomicUMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicMin({},{},uint({}));", inst, image, CoordsCastToInt(coords, info),
value);
}
void EmitImageAtomicSMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicMax({},{},int({}));", inst, image, CoordsCastToInt(coords, info),
value);
}
void EmitImageAtomicUMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicMax({},{},uint({}));", inst, image, CoordsCastToInt(coords, info),
value);
}
void EmitImageAtomicInc32(EmitContext&, IR::Inst&, const IR::Value&, std::string_view,
std::string_view) {
NotImplemented();
}
void EmitImageAtomicDec32(EmitContext&, IR::Inst&, const IR::Value&, std::string_view,
std::string_view) {
NotImplemented();
}
void EmitImageAtomicAnd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicAnd({},{},{});", inst, image, CoordsCastToInt(coords, info), value);
}
void EmitImageAtomicOr32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicOr({},{},{});", inst, image, CoordsCastToInt(coords, info), value);
}
void EmitImageAtomicXor32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicXor({},{},{});", inst, image, CoordsCastToInt(coords, info), value);
}
void EmitImageAtomicExchange32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto image{Image(ctx, info, index)};
ctx.AddU32("{}=imageAtomicExchange({},{},{});", inst, image, CoordsCastToInt(coords, info),
value);
}
void EmitIsTextureScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index) {
if (!index.IsImmediate()) {
throw NotImplementedException("Non-constant texture rescaling");
}
const u32 image_index{index.U32()};
ctx.AddU1("{}=(ftou(scaling.x)&{})!=0;", inst, 1u << image_index);
}
void EmitIsImageScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index) {
if (!index.IsImmediate()) {
throw NotImplementedException("Non-constant texture rescaling");
}
const u32 image_index{index.U32()};
ctx.AddU1("{}=(ftou(scaling.y)&{})!=0;", inst, 1u << image_index);
}
void EmitBindlessImageSampleImplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageSampleExplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageSampleDrefImplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageSampleDrefExplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageGather(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageGatherDref(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageFetch(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageQueryDimensions(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageQueryLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageGradient(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageRead(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageWrite(EmitContext&) {
NotImplemented();
}
void EmitBoundImageSampleImplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageSampleExplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageSampleDrefImplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageSampleDrefExplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageGather(EmitContext&) {
NotImplemented();
}
void EmitBoundImageGatherDref(EmitContext&) {
NotImplemented();
}
void EmitBoundImageFetch(EmitContext&) {
NotImplemented();
}
void EmitBoundImageQueryDimensions(EmitContext&) {
NotImplemented();
}
void EmitBoundImageQueryLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageGradient(EmitContext&) {
NotImplemented();
}
void EmitBoundImageRead(EmitContext&) {
NotImplemented();
}
void EmitBoundImageWrite(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicIAdd32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicSMin32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicUMin32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicSMax32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicUMax32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicInc32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicDec32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicAnd32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicOr32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicXor32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicExchange32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicIAdd32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicSMin32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicUMin32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicSMax32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicUMax32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicInc32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicDec32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicAnd32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicOr32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicXor32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicExchange32(EmitContext&) {
NotImplemented();
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,753 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <string_view>
#include "common/common_types.h"
namespace Shader::IR {
enum class Attribute : u64;
enum class Patch : u64;
class Inst;
class Value;
} // namespace Shader::IR
namespace Shader::Backend::GLSL {
class EmitContext;
#define NotImplemented() throw NotImplementedException("GLSL instruction {}", __func__)
// Microinstruction emitters
void EmitPhi(EmitContext& ctx, IR::Inst& inst);
void EmitVoid(EmitContext& ctx);
void EmitIdentity(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
void EmitConditionRef(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
void EmitReference(EmitContext& ctx, const IR::Value& value);
void EmitPhiMove(EmitContext& ctx, const IR::Value& phi, const IR::Value& value);
void EmitJoin(EmitContext& ctx);
void EmitDemoteToHelperInvocation(EmitContext& ctx);
void EmitBarrier(EmitContext& ctx);
void EmitWorkgroupMemoryBarrier(EmitContext& ctx);
void EmitDeviceMemoryBarrier(EmitContext& ctx);
void EmitPrologue(EmitContext& ctx);
void EmitEpilogue(EmitContext& ctx);
void EmitEmitVertex(EmitContext& ctx, const IR::Value& stream);
void EmitEndPrimitive(EmitContext& ctx, const IR::Value& stream);
void EmitGetRegister(EmitContext& ctx);
void EmitSetRegister(EmitContext& ctx);
void EmitGetPred(EmitContext& ctx);
void EmitSetPred(EmitContext& ctx);
void EmitSetGotoVariable(EmitContext& ctx);
void EmitGetGotoVariable(EmitContext& ctx);
void EmitSetIndirectBranchVariable(EmitContext& ctx);
void EmitGetIndirectBranchVariable(EmitContext& ctx);
void EmitGetCbufU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufS8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufU16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufS16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufU32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufU32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr,
std::string_view vertex);
void EmitGetAttributeU32(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr,
std::string_view vertex);
void EmitSetAttribute(EmitContext& ctx, IR::Attribute attr, std::string_view value,
std::string_view vertex);
void EmitGetAttributeIndexed(EmitContext& ctx, IR::Inst& inst, std::string_view offset,
std::string_view vertex);
void EmitSetAttributeIndexed(EmitContext& ctx, std::string_view offset, std::string_view value,
std::string_view vertex);
void EmitGetPatch(EmitContext& ctx, IR::Inst& inst, IR::Patch patch);
void EmitSetPatch(EmitContext& ctx, IR::Patch patch, std::string_view value);
void EmitSetFragColor(EmitContext& ctx, u32 index, u32 component, std::string_view value);
void EmitSetSampleMask(EmitContext& ctx, std::string_view value);
void EmitSetFragDepth(EmitContext& ctx, std::string_view value);
void EmitGetZFlag(EmitContext& ctx);
void EmitGetSFlag(EmitContext& ctx);
void EmitGetCFlag(EmitContext& ctx);
void EmitGetOFlag(EmitContext& ctx);
void EmitSetZFlag(EmitContext& ctx);
void EmitSetSFlag(EmitContext& ctx);
void EmitSetCFlag(EmitContext& ctx);
void EmitSetOFlag(EmitContext& ctx);
void EmitWorkgroupId(EmitContext& ctx, IR::Inst& inst);
void EmitLocalInvocationId(EmitContext& ctx, IR::Inst& inst);
void EmitInvocationId(EmitContext& ctx, IR::Inst& inst);
void EmitInvocationInfo(EmitContext& ctx, IR::Inst& inst);
void EmitSampleId(EmitContext& ctx, IR::Inst& inst);
void EmitIsHelperInvocation(EmitContext& ctx, IR::Inst& inst);
void EmitSR_WScaleFactorXY(EmitContext& ctx, IR::Inst& inst);
void EmitSR_WScaleFactorZ(EmitContext& ctx, IR::Inst& inst);
void EmitYDirection(EmitContext& ctx, IR::Inst& inst);
void EmitResolutionDownFactor(EmitContext& ctx, IR::Inst& inst);
void EmitRenderArea(EmitContext& ctx, IR::Inst& inst);
void EmitLoadLocal(EmitContext& ctx, IR::Inst& inst, std::string_view word_offset);
void EmitWriteLocal(EmitContext& ctx, std::string_view word_offset, std::string_view value);
void EmitUndefU1(EmitContext& ctx, IR::Inst& inst);
void EmitUndefU8(EmitContext& ctx, IR::Inst& inst);
void EmitUndefU16(EmitContext& ctx, IR::Inst& inst);
void EmitUndefU32(EmitContext& ctx, IR::Inst& inst);
void EmitUndefU64(EmitContext& ctx, IR::Inst& inst);
void EmitLoadGlobalU8(EmitContext& ctx);
void EmitLoadGlobalS8(EmitContext& ctx);
void EmitLoadGlobalU16(EmitContext& ctx);
void EmitLoadGlobalS16(EmitContext& ctx);
void EmitLoadGlobal32(EmitContext& ctx, IR::Inst& inst, std::string_view address);
void EmitLoadGlobal64(EmitContext& ctx, IR::Inst& inst, std::string_view address);
void EmitLoadGlobal128(EmitContext& ctx, IR::Inst& inst, std::string_view address);
void EmitWriteGlobalU8(EmitContext& ctx);
void EmitWriteGlobalS8(EmitContext& ctx);
void EmitWriteGlobalU16(EmitContext& ctx);
void EmitWriteGlobalS16(EmitContext& ctx);
void EmitWriteGlobal32(EmitContext& ctx, std::string_view address, std::string_view value);
void EmitWriteGlobal64(EmitContext& ctx, std::string_view address, std::string_view value);
void EmitWriteGlobal128(EmitContext& ctx, std::string_view address, std::string_view value);
void EmitLoadStorageU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorageS8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorageU16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorageS16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorage32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorage64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorage128(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitWriteStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorageS8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorageU16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorageS16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorage32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorage64(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorage128(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitLoadSharedU8(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedS8(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedU16(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedS16(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedU32(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedU64(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedU128(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitWriteSharedU8(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitWriteSharedU16(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitWriteSharedU32(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitWriteSharedU64(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitWriteSharedU128(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitCompositeConstructU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2);
void EmitCompositeConstructU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3);
void EmitCompositeConstructU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3, std::string_view e4);
void EmitCompositeExtractU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeExtractU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeExtractU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeInsertU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeInsertU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeInsertU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeConstructF16x2(EmitContext& ctx, std::string_view e1, std::string_view e2);
void EmitCompositeConstructF16x3(EmitContext& ctx, std::string_view e1, std::string_view e2,
std::string_view e3);
void EmitCompositeConstructF16x4(EmitContext& ctx, std::string_view e1, std::string_view e2,
std::string_view e3, std::string_view e4);
void EmitCompositeExtractF16x2(EmitContext& ctx, std::string_view composite, u32 index);
void EmitCompositeExtractF16x3(EmitContext& ctx, std::string_view composite, u32 index);
void EmitCompositeExtractF16x4(EmitContext& ctx, std::string_view composite, u32 index);
void EmitCompositeInsertF16x2(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeInsertF16x3(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeInsertF16x4(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeConstructF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2);
void EmitCompositeConstructF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3);
void EmitCompositeConstructF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3, std::string_view e4);
void EmitCompositeExtractF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeExtractF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeExtractF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeInsertF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeInsertF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeInsertF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeConstructF64x2(EmitContext& ctx);
void EmitCompositeConstructF64x3(EmitContext& ctx);
void EmitCompositeConstructF64x4(EmitContext& ctx);
void EmitCompositeExtractF64x2(EmitContext& ctx);
void EmitCompositeExtractF64x3(EmitContext& ctx);
void EmitCompositeExtractF64x4(EmitContext& ctx);
void EmitCompositeInsertF64x2(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeInsertF64x3(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeInsertF64x4(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitSelectU1(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitSelectU8(EmitContext& ctx, std::string_view cond, std::string_view true_value,
std::string_view false_value);
void EmitSelectU16(EmitContext& ctx, std::string_view cond, std::string_view true_value,
std::string_view false_value);
void EmitSelectU32(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitSelectU64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitSelectF16(EmitContext& ctx, std::string_view cond, std::string_view true_value,
std::string_view false_value);
void EmitSelectF32(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitSelectF64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitBitCastU16F16(EmitContext& ctx, IR::Inst& inst);
void EmitBitCastU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitCastF16U16(EmitContext& ctx, IR::Inst& inst);
void EmitBitCastF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitCastF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitPackUint2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitUnpackUint2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitPackFloat2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitUnpackFloat2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitPackHalf2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitUnpackHalf2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitPackDouble2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitUnpackDouble2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitGetZeroFromOp(EmitContext& ctx);
void EmitGetSignFromOp(EmitContext& ctx);
void EmitGetCarryFromOp(EmitContext& ctx);
void EmitGetOverflowFromOp(EmitContext& ctx);
void EmitGetSparseFromOp(EmitContext& ctx);
void EmitGetInBoundsFromOp(EmitContext& ctx);
void EmitFPAbs16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPAbs64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPAdd16(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPFma16(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c);
void EmitFPFma32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c);
void EmitFPFma64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c);
void EmitFPMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMax64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMin64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMul16(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMul64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPNeg16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPNeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPNeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSin(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPCos(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPExp2(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPLog2(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRecip32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRecip64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRecipSqrt32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRecipSqrt64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSqrt(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSaturate16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSaturate32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSaturate64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPClamp16(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value);
void EmitFPClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value);
void EmitFPClamp64(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value);
void EmitFPRoundEven16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRoundEven32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRoundEven64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPFloor16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPFloor32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPFloor64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPCeil16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPCeil32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPCeil64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPTrunc16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPTrunc32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPTrunc64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPOrdEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitFPOrdEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitFPUnordEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdNotEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdNotEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdNotEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordNotEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordNotEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordNotEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdLessThan16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdLessThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdLessThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordLessThan16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordLessThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordLessThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdGreaterThan16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdGreaterThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdGreaterThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordGreaterThan16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordGreaterThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordGreaterThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdLessThanEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdLessThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdLessThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordLessThanEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordLessThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordLessThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdGreaterThanEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdGreaterThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdGreaterThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordGreaterThanEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordGreaterThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordGreaterThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPIsNan16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPIsNan32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPIsNan64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitIAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitISub32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitISub64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitIMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitSDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitUDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitINeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitINeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitIAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitIAbs64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitShiftLeftLogical32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftLeftLogical64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftRightLogical32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftRightLogical64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftRightArithmetic32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftRightArithmetic64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitBitwiseAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitBitwiseOr32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitBitwiseXor32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitBitFieldInsert(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view insert, std::string_view offset, std::string_view count);
void EmitBitFieldSExtract(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view offset, std::string_view count);
void EmitBitFieldUExtract(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view offset, std::string_view count);
void EmitBitReverse32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitCount32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitwiseNot32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFindSMsb32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFindUMsb32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitSClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view min,
std::string_view max);
void EmitUClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view min,
std::string_view max);
void EmitSLessThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitULessThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitIEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitSLessThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitULessThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitSGreaterThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitUGreaterThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitINotEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitSGreaterThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitUGreaterThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitSharedAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicInc32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicDec32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicOr32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicXor32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicExchange32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicExchange64(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitStorageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicInc32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicDec32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAnd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicOr32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicXor32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicExchange32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicIAdd64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMin64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMin64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMax64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMax64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAnd64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicOr64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicXor64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicExchange64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicIAdd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAnd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicOr32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicXor32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAddF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAddF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAddF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicMinF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicMinF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicMaxF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicMaxF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitGlobalAtomicIAdd32(EmitContext& ctx);
void EmitGlobalAtomicSMin32(EmitContext& ctx);
void EmitGlobalAtomicUMin32(EmitContext& ctx);
void EmitGlobalAtomicSMax32(EmitContext& ctx);
void EmitGlobalAtomicUMax32(EmitContext& ctx);
void EmitGlobalAtomicInc32(EmitContext& ctx);
void EmitGlobalAtomicDec32(EmitContext& ctx);
void EmitGlobalAtomicAnd32(EmitContext& ctx);
void EmitGlobalAtomicOr32(EmitContext& ctx);
void EmitGlobalAtomicXor32(EmitContext& ctx);
void EmitGlobalAtomicExchange32(EmitContext& ctx);
void EmitGlobalAtomicIAdd64(EmitContext& ctx);
void EmitGlobalAtomicSMin64(EmitContext& ctx);
void EmitGlobalAtomicUMin64(EmitContext& ctx);
void EmitGlobalAtomicSMax64(EmitContext& ctx);
void EmitGlobalAtomicUMax64(EmitContext& ctx);
void EmitGlobalAtomicInc64(EmitContext& ctx);
void EmitGlobalAtomicDec64(EmitContext& ctx);
void EmitGlobalAtomicAnd64(EmitContext& ctx);
void EmitGlobalAtomicOr64(EmitContext& ctx);
void EmitGlobalAtomicXor64(EmitContext& ctx);
void EmitGlobalAtomicExchange64(EmitContext& ctx);
void EmitGlobalAtomicIAdd32x2(EmitContext& ctx);
void EmitGlobalAtomicSMin32x2(EmitContext& ctx);
void EmitGlobalAtomicUMin32x2(EmitContext& ctx);
void EmitGlobalAtomicSMax32x2(EmitContext& ctx);
void EmitGlobalAtomicUMax32x2(EmitContext& ctx);
void EmitGlobalAtomicInc32x2(EmitContext& ctx);
void EmitGlobalAtomicDec32x2(EmitContext& ctx);
void EmitGlobalAtomicAnd32x2(EmitContext& ctx);
void EmitGlobalAtomicOr32x2(EmitContext& ctx);
void EmitGlobalAtomicXor32x2(EmitContext& ctx);
void EmitGlobalAtomicExchange32x2(EmitContext& ctx);
void EmitGlobalAtomicAddF32(EmitContext& ctx);
void EmitGlobalAtomicAddF16x2(EmitContext& ctx);
void EmitGlobalAtomicAddF32x2(EmitContext& ctx);
void EmitGlobalAtomicMinF16x2(EmitContext& ctx);
void EmitGlobalAtomicMinF32x2(EmitContext& ctx);
void EmitGlobalAtomicMaxF16x2(EmitContext& ctx);
void EmitGlobalAtomicMaxF32x2(EmitContext& ctx);
void EmitLogicalOr(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitLogicalAnd(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitLogicalXor(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitLogicalNot(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS16F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS16F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS16F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS64F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU16F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU16F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU16F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU64F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU64U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16S8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16S16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16S32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16S64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16U8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16U16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32S8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32S16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32S32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32S64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32U8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32U16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64S8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64S16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64S32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64S64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64U8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64U16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU16U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32U16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU8U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32U8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32S8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32S16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBindlessImageSampleImplicitLod(EmitContext&);
void EmitBindlessImageSampleExplicitLod(EmitContext&);
void EmitBindlessImageSampleDrefImplicitLod(EmitContext&);
void EmitBindlessImageSampleDrefExplicitLod(EmitContext&);
void EmitBindlessImageGather(EmitContext&);
void EmitBindlessImageGatherDref(EmitContext&);
void EmitBindlessImageFetch(EmitContext&);
void EmitBindlessImageQueryDimensions(EmitContext&);
void EmitBindlessImageQueryLod(EmitContext&);
void EmitBindlessImageGradient(EmitContext&);
void EmitBindlessImageRead(EmitContext&);
void EmitBindlessImageWrite(EmitContext&);
void EmitBoundImageSampleImplicitLod(EmitContext&);
void EmitBoundImageSampleExplicitLod(EmitContext&);
void EmitBoundImageSampleDrefImplicitLod(EmitContext&);
void EmitBoundImageSampleDrefExplicitLod(EmitContext&);
void EmitBoundImageGather(EmitContext&);
void EmitBoundImageGatherDref(EmitContext&);
void EmitBoundImageFetch(EmitContext&);
void EmitBoundImageQueryDimensions(EmitContext&);
void EmitBoundImageQueryLod(EmitContext&);
void EmitBoundImageGradient(EmitContext&);
void EmitBoundImageRead(EmitContext&);
void EmitBoundImageWrite(EmitContext&);
void EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view bias_lc,
const IR::Value& offset);
void EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view lod_lc,
const IR::Value& offset);
void EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view dref,
std::string_view bias_lc, const IR::Value& offset);
void EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view dref,
std::string_view lod_lc, const IR::Value& offset);
void EmitImageGather(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, const IR::Value& offset2);
void EmitImageGatherDref(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, const IR::Value& offset2,
std::string_view dref);
void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, std::string_view lod,
std::string_view ms);
void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view lod, const IR::Value& skip_mips);
void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords);
void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& derivatives,
const IR::Value& offset, const IR::Value& lod_clamp);
void EmitImageRead(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords);
void EmitImageWrite(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view color);
void EmitIsTextureScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index);
void EmitIsImageScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index);
void EmitBindlessImageAtomicIAdd32(EmitContext&);
void EmitBindlessImageAtomicSMin32(EmitContext&);
void EmitBindlessImageAtomicUMin32(EmitContext&);
void EmitBindlessImageAtomicSMax32(EmitContext&);
void EmitBindlessImageAtomicUMax32(EmitContext&);
void EmitBindlessImageAtomicInc32(EmitContext&);
void EmitBindlessImageAtomicDec32(EmitContext&);
void EmitBindlessImageAtomicAnd32(EmitContext&);
void EmitBindlessImageAtomicOr32(EmitContext&);
void EmitBindlessImageAtomicXor32(EmitContext&);
void EmitBindlessImageAtomicExchange32(EmitContext&);
void EmitBoundImageAtomicIAdd32(EmitContext&);
void EmitBoundImageAtomicSMin32(EmitContext&);
void EmitBoundImageAtomicUMin32(EmitContext&);
void EmitBoundImageAtomicSMax32(EmitContext&);
void EmitBoundImageAtomicUMax32(EmitContext&);
void EmitBoundImageAtomicInc32(EmitContext&);
void EmitBoundImageAtomicDec32(EmitContext&);
void EmitBoundImageAtomicAnd32(EmitContext&);
void EmitBoundImageAtomicOr32(EmitContext&);
void EmitBoundImageAtomicXor32(EmitContext&);
void EmitBoundImageAtomicExchange32(EmitContext&);
void EmitImageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicSMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicUMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicSMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicUMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicInc32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicDec32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicAnd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicOr32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicXor32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicExchange32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitLaneId(EmitContext& ctx, IR::Inst& inst);
void EmitVoteAll(EmitContext& ctx, IR::Inst& inst, std::string_view pred);
void EmitVoteAny(EmitContext& ctx, IR::Inst& inst, std::string_view pred);
void EmitVoteEqual(EmitContext& ctx, IR::Inst& inst, std::string_view pred);
void EmitSubgroupBallot(EmitContext& ctx, IR::Inst& inst, std::string_view pred);
void EmitSubgroupEqMask(EmitContext& ctx, IR::Inst& inst);
void EmitSubgroupLtMask(EmitContext& ctx, IR::Inst& inst);
void EmitSubgroupLeMask(EmitContext& ctx, IR::Inst& inst);
void EmitSubgroupGtMask(EmitContext& ctx, IR::Inst& inst);
void EmitSubgroupGeMask(EmitContext& ctx, IR::Inst& inst);
void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask);
void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index,
std::string_view clamp, std::string_view segmentation_mask);
void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask);
void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask);
void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
std::string_view swizzle);
void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);
void EmitDPdyFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);
void EmitDPdxCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);
void EmitDPdyCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,266 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
namespace {
void SetZeroFlag(EmitContext& ctx, IR::Inst& inst, std::string_view result) {
IR::Inst* const zero{inst.GetAssociatedPseudoOperation(IR::Opcode::GetZeroFromOp)};
if (!zero) {
return;
}
ctx.AddU1("{}={}==0;", *zero, result);
zero->Invalidate();
}
void SetSignFlag(EmitContext& ctx, IR::Inst& inst, std::string_view result) {
IR::Inst* const sign{inst.GetAssociatedPseudoOperation(IR::Opcode::GetSignFromOp)};
if (!sign) {
return;
}
ctx.AddU1("{}=int({})<0;", *sign, result);
sign->Invalidate();
}
void BitwiseLogicalOp(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
char lop) {
const auto result{ctx.var_alloc.Define(inst, GlslVarType::U32)};
ctx.Add("{}={}{}{};", result, a, lop, b);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
} // Anonymous namespace
void EmitIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
// Compute the overflow CC first as it requires the original operand values,
// which may be overwritten by the result of the addition
if (IR::Inst * overflow{inst.GetAssociatedPseudoOperation(IR::Opcode::GetOverflowFromOp)}) {
// https://stackoverflow.com/questions/55468823/how-to-detect-integer-overflow-in-c
constexpr u32 s32_max{static_cast<u32>((std::numeric_limits<s32>::max)())};
const auto sub_a{fmt::format("{}u-{}", s32_max, a)};
const auto positive_result{fmt::format("int({})>int({})", b, sub_a)};
const auto negative_result{fmt::format("int({})<int({})", b, sub_a)};
ctx.AddU1("{}=int({})>=0?{}:{};", *overflow, a, positive_result, negative_result);
overflow->Invalidate();
}
const auto result{ctx.var_alloc.Define(inst, GlslVarType::U32)};
if (IR::Inst* const carry{inst.GetAssociatedPseudoOperation(IR::Opcode::GetCarryFromOp)}) {
ctx.uses_cc_carry = true;
ctx.Add("{}=uaddCarry({},{},carry);", result, a, b);
ctx.AddU1("{}=carry!=0;", *carry);
carry->Invalidate();
} else {
ctx.Add("{}={}+{};", result, a, b);
}
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitIAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU64("{}={}+{};", inst, a, b);
}
void EmitISub32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}={}-{};", inst, a, b);
}
void EmitISub64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU64("{}={}-{};", inst, a, b);
}
void EmitIMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=uint({}*{});", inst, a, b);
}
void EmitSDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=uint(int({})/int({}));", inst, a, b);
}
void EmitUDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}={}/{};", inst, a, b);
}
void EmitINeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=uint(int(0)-int({}));", inst, value);
}
void EmitINeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=uint64_t(int64_t(0)-int64_t({}));", inst, value);
}
void EmitIAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=abs(int({}));", inst, value);
}
void EmitIAbs64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
// TODO: Uhm, are you sure? This may crash on some drivers!
ctx.AddU32("{}=abs(int64_t({}));", inst, value);
}
void EmitShiftLeftLogical32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU32("{}={}<<{};", inst, base, shift);
}
void EmitShiftLeftLogical64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU64("{}={}<<{};", inst, base, shift);
}
void EmitShiftRightLogical32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU32("{}={}>>{};", inst, base, shift);
}
void EmitShiftRightLogical64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU64("{}={}>>{};", inst, base, shift);
}
void EmitShiftRightArithmetic32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU32("{}=int({})>>{};", inst, base, shift);
}
void EmitShiftRightArithmetic64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU64("{}=int64_t({})>>{};", inst, base, shift);
}
void EmitBitwiseAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
BitwiseLogicalOp(ctx, inst, a, b, '&');
}
void EmitBitwiseOr32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
BitwiseLogicalOp(ctx, inst, a, b, '|');
}
void EmitBitwiseXor32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
BitwiseLogicalOp(ctx, inst, a, b, '^');
}
void EmitBitFieldInsert(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view insert, std::string_view offset, std::string_view count) {
ctx.AddU32("{}=bitfieldInsert({},{},int({}),int({}));", inst, base, insert, offset, count);
}
void EmitBitFieldSExtract(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view offset, std::string_view count) {
const auto result{ctx.var_alloc.Define(inst, GlslVarType::U32)};
ctx.Add("{}=uint(bitfieldExtract(int({}),int({}),int({})));", result, base, offset, count);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitBitFieldUExtract(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view offset, std::string_view count) {
const auto result{ctx.var_alloc.Define(inst, GlslVarType::U32)};
ctx.Add("{}=uint(bitfieldExtract(uint({}),int({}),int({})));", result, base, offset, count);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitBitReverse32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=bitfieldReverse({});", inst, value);
}
void EmitBitCount32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=bitCount({});", inst, value);
}
void EmitBitwiseNot32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=~{};", inst, value);
}
void EmitFindSMsb32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=findMSB(int({}));", inst, value);
}
void EmitFindUMsb32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=findMSB(uint({}));", inst, value);
}
void EmitSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=min(int({}),int({}));", inst, a, b);
}
void EmitUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=min(uint({}),uint({}));", inst, a, b);
}
void EmitSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=max(int({}),int({}));", inst, a, b);
}
void EmitUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=max(uint({}),uint({}));", inst, a, b);
}
void EmitSClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view min,
std::string_view max) {
const auto result{ctx.var_alloc.Define(inst, GlslVarType::U32)};
ctx.Add("{}=clamp(int({}),int({}),int({}));", result, value, min, max);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitUClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view min,
std::string_view max) {
const auto result{ctx.var_alloc.Define(inst, GlslVarType::U32)};
ctx.Add("{}=clamp(uint({}),uint({}),uint({}));", result, value, min, max);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitSLessThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs) {
ctx.AddU1("{}=int({})<int({});", inst, lhs, rhs);
}
void EmitULessThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs) {
ctx.AddU1("{}=uint({})<uint({});", inst, lhs, rhs);
}
void EmitIEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs) {
ctx.AddU1("{}={}=={};", inst, lhs, rhs);
}
void EmitSLessThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=int({})<=int({});", inst, lhs, rhs);
}
void EmitULessThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=uint({})<=uint({});", inst, lhs, rhs);
}
void EmitSGreaterThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=int({})>int({});", inst, lhs, rhs);
}
void EmitUGreaterThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=uint({})>uint({});", inst, lhs, rhs);
}
void EmitINotEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs) {
ctx.AddU1("{}={}!={};", inst, lhs, rhs);
}
void EmitSGreaterThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=int({})>=int({});", inst, lhs, rhs);
}
void EmitUGreaterThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=uint({})>=uint({});", inst, lhs, rhs);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,25 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
void EmitLogicalOr(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU1("{}={}||{};", inst, a, b);
}
void EmitLogicalAnd(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU1("{}={}&&{};", inst, a, b);
}
void EmitLogicalXor(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU1("{}={}^^{};", inst, a, b);
}
void EmitLogicalNot(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU1("{}=!{};", inst, value);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,201 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::GLSL {
namespace {
constexpr char cas_loop[]{"for(;;){{uint old_value={};uint "
"cas_result=atomicCompSwap({},old_value,bitfieldInsert({},{},{},{}));"
"if(cas_result==old_value){{break;}}}}"};
void SsboWriteCas(EmitContext& ctx, const IR::Value& binding, std::string_view offset_var,
std::string_view value, std::string_view bit_offset, u32 num_bits) {
const auto ssbo{fmt::format("{}_ssbo{}[{}>>2]", ctx.stage_name, binding.U32(), offset_var)};
ctx.Add(cas_loop, ssbo, ssbo, ssbo, value, bit_offset, num_bits);
}
} // Anonymous namespace
void EmitLoadGlobalU8(EmitContext&) {
NotImplemented();
}
void EmitLoadGlobalS8(EmitContext&) {
NotImplemented();
}
void EmitLoadGlobalU16(EmitContext&) {
NotImplemented();
}
void EmitLoadGlobalS16(EmitContext&) {
NotImplemented();
}
void EmitLoadGlobal32(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
if (ctx.profile.support_int64) {
return ctx.AddU32("{}=LoadGlobal32({});", inst, address);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32("{}=0u;", inst);
}
void EmitLoadGlobal64(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
if (ctx.profile.support_int64) {
return ctx.AddU32x2("{}=LoadGlobal64({});", inst, address);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32x2("{}=uvec2(0);", inst);
}
void EmitLoadGlobal128(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
if (ctx.profile.support_int64) {
return ctx.AddU32x4("{}=LoadGlobal128({});", inst, address);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32x4("{}=uvec4(0);", inst);
}
void EmitWriteGlobalU8(EmitContext&) {
NotImplemented();
}
void EmitWriteGlobalS8(EmitContext&) {
NotImplemented();
}
void EmitWriteGlobalU16(EmitContext&) {
NotImplemented();
}
void EmitWriteGlobalS16(EmitContext&) {
NotImplemented();
}
void EmitWriteGlobal32(EmitContext& ctx, std::string_view address, std::string_view value) {
if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal32({},{});", address, value);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
}
void EmitWriteGlobal64(EmitContext& ctx, std::string_view address, std::string_view value) {
if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal64({},{});", address, value);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
}
void EmitWriteGlobal128(EmitContext& ctx, std::string_view address, std::string_view value) {
if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal128({},{});", address, value);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
}
void EmitLoadStorageU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}=bitfieldExtract({}_ssbo{}[{}>>2],int({}%4)*8,8);", inst, ctx.stage_name,
binding.U32(), offset_var, offset_var);
}
void EmitLoadStorageS8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}=bitfieldExtract(int({}_ssbo{}[{}>>2]),int({}%4)*8,8);", inst, ctx.stage_name,
binding.U32(), offset_var, offset_var);
}
void EmitLoadStorageU16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}=bitfieldExtract({}_ssbo{}[{}>>2],int(({}>>1)%2)*16,16);", inst, ctx.stage_name,
binding.U32(), offset_var, offset_var);
}
void EmitLoadStorageS16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}=bitfieldExtract(int({}_ssbo{}[{}>>2]),int(({}>>1)%2)*16,16);", inst,
ctx.stage_name, binding.U32(), offset_var, offset_var);
}
void EmitLoadStorage32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}={}_ssbo{}[{}>>2];", inst, ctx.stage_name, binding.U32(), offset_var);
}
void EmitLoadStorage64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32x2("{}=uvec2({}_ssbo{}[{}>>2],{}_ssbo{}[({}+4)>>2]);", inst, ctx.stage_name,
binding.U32(), offset_var, ctx.stage_name, binding.U32(), offset_var);
}
void EmitLoadStorage128(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32x4("{}=uvec4({}_ssbo{}[{}>>2],{}_ssbo{}[({}+4)>>2],{}_ssbo{}[({}+8)>>2],{}_ssbo{}[({}"
"+12)>>2]);",
inst, ctx.stage_name, binding.U32(), offset_var, ctx.stage_name, binding.U32(),
offset_var, ctx.stage_name, binding.U32(), offset_var, ctx.stage_name,
binding.U32(), offset_var);
}
void EmitWriteStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("int({}%4)*8", offset_var)};
SsboWriteCas(ctx, binding, offset_var, value, bit_offset, 8);
}
void EmitWriteStorageS8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("int({}%4)*8", offset_var)};
SsboWriteCas(ctx, binding, offset_var, value, bit_offset, 8);
}
void EmitWriteStorageU16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("int(({}>>1)%2)*16", offset_var)};
SsboWriteCas(ctx, binding, offset_var, value, bit_offset, 16);
}
void EmitWriteStorageS16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("int(({}>>1)%2)*16", offset_var)};
SsboWriteCas(ctx, binding, offset_var, value, bit_offset, 16);
}
void EmitWriteStorage32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.Add("{}_ssbo{}[{}>>2]={};", ctx.stage_name, binding.U32(), offset_var, value);
}
void EmitWriteStorage64(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.Add("{}_ssbo{}[{}>>2]={}.x;", ctx.stage_name, binding.U32(), offset_var, value);
ctx.Add("{}_ssbo{}[({}+4)>>2]={}.y;", ctx.stage_name, binding.U32(), offset_var, value);
}
void EmitWriteStorage128(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.Add("{}_ssbo{}[{}>>2]={}.x;", ctx.stage_name, binding.U32(), offset_var, value);
ctx.Add("{}_ssbo{}[({}+4)>>2]={}.y;", ctx.stage_name, binding.U32(), offset_var, value);
ctx.Add("{}_ssbo{}[({}+8)>>2]={}.z;", ctx.stage_name, binding.U32(), offset_var, value);
ctx.Add("{}_ssbo{}[({}+12)>>2]={}.w;", ctx.stage_name, binding.U32(), offset_var, value);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,100 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
void EmitGetRegister(EmitContext& ctx) {
NotImplemented();
}
void EmitSetRegister(EmitContext& ctx) {
NotImplemented();
}
void EmitGetPred(EmitContext& ctx) {
NotImplemented();
}
void EmitSetPred(EmitContext& ctx) {
NotImplemented();
}
void EmitSetGotoVariable(EmitContext& ctx) {
NotImplemented();
}
void EmitGetGotoVariable(EmitContext& ctx) {
NotImplemented();
}
void EmitSetIndirectBranchVariable(EmitContext& ctx) {
NotImplemented();
}
void EmitGetIndirectBranchVariable(EmitContext& ctx) {
NotImplemented();
}
void EmitGetZFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitGetSFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitGetCFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitGetOFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitSetZFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitSetSFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitSetCFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitSetOFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitGetZeroFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetSignFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetCarryFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetOverflowFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetSparseFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetInBoundsFromOp(EmitContext& ctx) {
NotImplemented();
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,54 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
void EmitSelectU1(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddU1("{}={}?{}:{};", inst, cond, true_value, false_value);
}
void EmitSelectU8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view cond,
[[maybe_unused]] std::string_view true_value,
[[maybe_unused]] std::string_view false_value) {
NotImplemented();
}
void EmitSelectU16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view cond,
[[maybe_unused]] std::string_view true_value,
[[maybe_unused]] std::string_view false_value) {
NotImplemented();
}
void EmitSelectU32(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddU32("{}={}?{}:{};", inst, cond, true_value, false_value);
}
void EmitSelectU64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddU64("{}={}?{}:{};", inst, cond, true_value, false_value);
}
void EmitSelectF16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view cond,
[[maybe_unused]] std::string_view true_value,
[[maybe_unused]] std::string_view false_value) {
NotImplemented();
}
void EmitSelectF32(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddF32("{}={}?{}:{};", inst, cond, true_value, false_value);
}
void EmitSelectF64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddF64("{}={}?{}:{};", inst, cond, true_value, false_value);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,76 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
namespace {
constexpr char cas_loop[]{"for(;;){{uint old_value={};uint "
"cas_result=atomicCompSwap({},old_value,bitfieldInsert({},{},{},{}));"
"if(cas_result==old_value){{break;}}}}"};
void SharedWriteCas(EmitContext& ctx, std::string_view offset, std::string_view value,
std::string_view bit_offset, u32 num_bits) {
const auto smem{fmt::format("smem[{}>>2]", offset)};
ctx.Add(cas_loop, smem, smem, smem, value, bit_offset, num_bits);
}
} // Anonymous namespace
void EmitLoadSharedU8(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=bitfieldExtract(smem[{}>>2],int({}%4)*8,8);", inst, offset, offset);
}
void EmitLoadSharedS8(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=bitfieldExtract(int(smem[{}>>2]),int({}%4)*8,8);", inst, offset, offset);
}
void EmitLoadSharedU16(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=bitfieldExtract(smem[{}>>2],int(({}>>1)%2)*16,16);", inst, offset, offset);
}
void EmitLoadSharedS16(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=bitfieldExtract(int(smem[{}>>2]),int(({}>>1)%2)*16,16);", inst, offset, offset);
}
void EmitLoadSharedU32(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=smem[{}>>2];", inst, offset);
}
void EmitLoadSharedU64(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32x2("{}=uvec2(smem[{}>>2],smem[({}+4)>>2]);", inst, offset, offset);
}
void EmitLoadSharedU128(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32x4("{}=uvec4(smem[{}>>2],smem[({}+4)>>2],smem[({}+8)>>2],smem[({}+12)>>2]);", inst,
offset, offset, offset, offset);
}
void EmitWriteSharedU8(EmitContext& ctx, std::string_view offset, std::string_view value) {
const auto bit_offset{fmt::format("int({}%4)*8", offset)};
SharedWriteCas(ctx, offset, value, bit_offset, 8);
}
void EmitWriteSharedU16(EmitContext& ctx, std::string_view offset, std::string_view value) {
const auto bit_offset{fmt::format("int(({}>>1)%2)*16", offset)};
SharedWriteCas(ctx, offset, value, bit_offset, 16);
}
void EmitWriteSharedU32(EmitContext& ctx, std::string_view offset, std::string_view value) {
ctx.Add("smem[{}>>2]={};", offset, value);
}
void EmitWriteSharedU64(EmitContext& ctx, std::string_view offset, std::string_view value) {
ctx.Add("smem[{}>>2]={}.x;", offset, value);
ctx.Add("smem[({}+4)>>2]={}.y;", offset, value);
}
void EmitWriteSharedU128(EmitContext& ctx, std::string_view offset, std::string_view value) {
ctx.Add("smem[{}>>2]={}.x;", offset, value);
ctx.Add("smem[({}+4)>>2]={}.y;", offset, value);
ctx.Add("smem[({}+8)>>2]={}.z;", offset, value);
ctx.Add("smem[({}+12)>>2]={}.w;", offset, value);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,110 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::GLSL {
namespace {
std::string_view OutputVertexIndex(EmitContext& ctx) {
return ctx.stage == Stage::TessellationControl ? "[gl_InvocationID]" : "";
}
void InitializeOutputVaryings(EmitContext& ctx) {
if (ctx.uses_geometry_passthrough) {
return;
}
if (ctx.stage == Stage::VertexB || ctx.stage == Stage::Geometry) {
ctx.Add("gl_Position=vec4(0,0,0,1);");
}
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (!ctx.info.stores.Generic(index)) {
continue;
}
const auto& info_array{ctx.output_generics.at(index)};
const auto output_decorator{OutputVertexIndex(ctx)};
size_t element{};
while (element < info_array.size()) {
const auto& info{info_array.at(element)};
const auto varying_name{fmt::format("{}{}", info.name, output_decorator)};
switch (info.num_components) {
case 1: {
const char value{element == 3 ? '1' : '0'};
ctx.Add("{}={}.f;", varying_name, value);
break;
}
case 2:
case 3:
if (element + info.num_components < 4) {
ctx.Add("{}=vec{}(0);", varying_name, info.num_components);
} else {
// last element is the w component, must be initialized to 1
const auto zeros{info.num_components == 3 ? "0,0," : "0,"};
ctx.Add("{}=vec{}({}1);", varying_name, info.num_components, zeros);
}
break;
case 4:
ctx.Add("{}=vec4(0,0,0,1);", varying_name);
break;
default:
break;
}
element += info.num_components;
}
}
}
} // Anonymous namespace
void EmitPhi(EmitContext& ctx, IR::Inst& phi) {
const size_t num_args{phi.NumArgs()};
for (size_t i = 0; i < num_args; ++i) {
ctx.var_alloc.Consume(phi.Arg(i));
}
if (!phi.Definition<Id>().is_valid) {
// The phi node wasn't forward defined
ctx.var_alloc.PhiDefine(phi, phi.Type());
}
}
void EmitVoid(EmitContext&) {}
void EmitReference(EmitContext& ctx, const IR::Value& value) {
ctx.var_alloc.Consume(value);
}
void EmitPhiMove(EmitContext& ctx, const IR::Value& phi_value, const IR::Value& value) {
IR::Inst& phi{*phi_value.InstRecursive()};
const auto phi_type{phi.Type()};
if (!phi.Definition<Id>().is_valid) {
// The phi node wasn't forward defined
ctx.var_alloc.PhiDefine(phi, phi_type);
}
const auto phi_reg{ctx.var_alloc.Consume(IR::Value{&phi})};
const auto val_reg{ctx.var_alloc.Consume(value)};
if (phi_reg == val_reg) {
return;
}
const bool needs_workaround{ctx.profile.has_gl_bool_ref_bug && phi_type == IR::Type::U1};
const auto suffix{needs_workaround ? "?true:false" : ""};
ctx.Add("{}={}{};", phi_reg, val_reg, suffix);
}
void EmitPrologue(EmitContext& ctx) {
InitializeOutputVaryings(ctx);
}
void EmitEpilogue(EmitContext&) {}
void EmitEmitVertex(EmitContext& ctx, const IR::Value& stream) {
ctx.Add("EmitStreamVertex(int({}));", ctx.var_alloc.Consume(stream));
InitializeOutputVaryings(ctx);
}
void EmitEndPrimitive(EmitContext& ctx, const IR::Value& stream) {
ctx.Add("EndStreamPrimitive(int({}));", ctx.var_alloc.Consume(stream));
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,29 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
namespace Shader::Backend::GLSL {
void EmitUndefU1(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU1("{}=false;", inst);
}
void EmitUndefU8(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=0u;", inst);
}
void EmitUndefU16(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=0u;", inst);
}
void EmitUndefU32(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=0u;", inst);
}
void EmitUndefU64(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU64("{}=0u;", inst);
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,246 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::GLSL {
namespace {
constexpr char THREAD_ID[]{"gl_SubGroupInvocationARB"};
void SetInBoundsFlag(EmitContext& ctx, IR::Inst& inst) {
IR::Inst* const in_bounds{inst.GetAssociatedPseudoOperation(IR::Opcode::GetInBoundsFromOp)};
if (!in_bounds) {
return;
}
ctx.AddU1("{}=shfl_in_bounds;", *in_bounds);
in_bounds->Invalidate();
}
std::string ComputeMinThreadId(std::string_view thread_id, std::string_view segmentation_mask) {
return fmt::format("({}&{})", thread_id, segmentation_mask);
}
std::string ComputeMaxThreadId(std::string_view min_thread_id, std::string_view clamp,
std::string_view not_seg_mask) {
return fmt::format("({})|({}&{})", min_thread_id, clamp, not_seg_mask);
}
std::string GetMaxThreadId(std::string_view thread_id, std::string_view clamp,
std::string_view segmentation_mask) {
const auto not_seg_mask{fmt::format("(~{})", segmentation_mask)};
const auto min_thread_id{ComputeMinThreadId(thread_id, segmentation_mask)};
return ComputeMaxThreadId(min_thread_id, clamp, not_seg_mask);
}
void UseShuffleNv(EmitContext& ctx, IR::Inst& inst, std::string_view shfl_op,
std::string_view value, std::string_view index,
[[maybe_unused]] std::string_view clamp, std::string_view segmentation_mask) {
const auto width{fmt::format("32u>>(bitCount({}&31u))", segmentation_mask)};
ctx.AddU32("{}={}({},{},{},shfl_in_bounds);", inst, shfl_op, value, index, width);
SetInBoundsFlag(ctx, inst);
}
std::string_view BallotIndex(EmitContext& ctx) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
return ".x";
}
return "[gl_SubGroupInvocationARB>>5]";
}
std::string GetMask(EmitContext& ctx, std::string_view mask) {
const auto ballot_index{BallotIndex(ctx)};
return fmt::format("uint(uvec2({}){})", mask, ballot_index);
}
} // Anonymous namespace
void EmitLaneId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}={}&31u;", inst, THREAD_ID);
}
void EmitVoteAll(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
ctx.AddU1("{}=allInvocationsEqualARB({});", inst, pred);
return;
}
const auto ballot_index{BallotIndex(ctx)};
const auto active_mask{fmt::format("uvec2(ballotARB(true)){}", ballot_index)};
const auto ballot{fmt::format("uvec2(ballotARB({})){}", pred, ballot_index)};
ctx.AddU1("{}=({}&{})=={};", inst, ballot, active_mask, active_mask);
}
void EmitVoteAny(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
ctx.AddU1("{}=anyInvocationARB({});", inst, pred);
return;
}
const auto ballot_index{BallotIndex(ctx)};
const auto active_mask{fmt::format("uvec2(ballotARB(true)){}", ballot_index)};
const auto ballot{fmt::format("uvec2(ballotARB({})){}", pred, ballot_index)};
ctx.AddU1("{}=({}&{})!=0u;", inst, ballot, active_mask, active_mask);
}
void EmitVoteEqual(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
ctx.AddU1("{}=allInvocationsEqualARB({});", inst, pred);
return;
}
const auto ballot_index{BallotIndex(ctx)};
const auto active_mask{fmt::format("uvec2(ballotARB(true)){}", ballot_index)};
const auto ballot{fmt::format("uvec2(ballotARB({})){}", pred, ballot_index)};
const auto value{fmt::format("({}^{})", ballot, active_mask)};
ctx.AddU1("{}=({}==0)||({}=={});", inst, value, value, active_mask);
}
void EmitSubgroupBallot(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
const auto ballot_index{BallotIndex(ctx)};
ctx.AddU32("{}=uvec2(ballotARB({})){};", inst, pred, ballot_index);
}
void EmitSubgroupEqMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}={};", inst, GetMask(ctx, "gl_SubGroupEqMaskARB"));
}
void EmitSubgroupLtMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}={};", inst, GetMask(ctx, "gl_SubGroupLtMaskARB"));
}
void EmitSubgroupLeMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}={};", inst, GetMask(ctx, "gl_SubGroupLeMaskARB"));
}
void EmitSubgroupGtMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}={};", inst, GetMask(ctx, "gl_SubGroupGtMaskARB"));
}
void EmitSubgroupGeMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}={};", inst, GetMask(ctx, "gl_SubGroupGeMaskARB"));
}
void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp, std::string_view seg_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleNV", value, index, clamp, seg_mask);
return;
}
const bool big_warp{ctx.profile.warp_size_potentially_larger_than_guest};
const auto is_upper_partition{"int(gl_SubGroupInvocationARB)>=32"};
const auto upper_index{fmt::format("{}?{}+32:{}", is_upper_partition, index, index)};
const auto upper_clamp{fmt::format("{}?{}+32:{}", is_upper_partition, clamp, clamp)};
const auto not_seg_mask{fmt::format("(~{})", seg_mask)};
const auto min_thread_id{ComputeMinThreadId(THREAD_ID, seg_mask)};
const auto max_thread_id{
ComputeMaxThreadId(min_thread_id, big_warp ? upper_clamp : clamp, not_seg_mask)};
const auto lhs{fmt::format("({}&{})", big_warp ? upper_index : index, not_seg_mask)};
const auto src_thread_id{fmt::format("({})|({})", lhs, min_thread_id)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id);
ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value);
}
void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index,
std::string_view clamp, std::string_view seg_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleUpNV", value, index, clamp, seg_mask);
return;
}
const bool big_warp{ctx.profile.warp_size_potentially_larger_than_guest};
const auto is_upper_partition{"int(gl_SubGroupInvocationARB)>=32"};
const auto upper_clamp{fmt::format("{}?{}+32:{}", is_upper_partition, clamp, clamp)};
const auto max_thread_id{GetMaxThreadId(THREAD_ID, big_warp ? upper_clamp : clamp, seg_mask)};
const auto src_thread_id{fmt::format("({}-{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=int({})>=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id);
ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value);
}
void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp, std::string_view seg_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleDownNV", value, index, clamp, seg_mask);
return;
}
const bool big_warp{ctx.profile.warp_size_potentially_larger_than_guest};
const auto is_upper_partition{"int(gl_SubGroupInvocationARB)>=32"};
const auto upper_clamp{fmt::format("{}?{}+32:{}", is_upper_partition, clamp, clamp)};
const auto max_thread_id{GetMaxThreadId(THREAD_ID, big_warp ? upper_clamp : clamp, seg_mask)};
const auto src_thread_id{fmt::format("({}+{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id);
ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value);
}
void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view seg_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleXorNV", value, index, clamp, seg_mask);
return;
}
const bool big_warp{ctx.profile.warp_size_potentially_larger_than_guest};
const auto is_upper_partition{"int(gl_SubGroupInvocationARB)>=32"};
const auto upper_clamp{fmt::format("{}?{}+32:{}", is_upper_partition, clamp, clamp)};
const auto max_thread_id{GetMaxThreadId(THREAD_ID, big_warp ? upper_clamp : clamp, seg_mask)};
const auto src_thread_id{fmt::format("({}^{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.Add("shfl_result=readInvocationARB({},{});", value, src_thread_id);
ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value);
}
void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
std::string_view swizzle) {
const auto mask{fmt::format("({}>>((gl_SubGroupInvocationARB&3)<<1))&3", swizzle)};
const std::string modifier_a = fmt::format("FSWZ_A[{}]", mask);
const std::string modifier_b = fmt::format("FSWZ_B[{}]", mask);
ctx.AddF32("{}=({}*{})+({}*{});", inst, op_a, modifier_a, op_b, modifier_b);
}
void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
if (ctx.profile.support_gl_derivative_control) {
ctx.AddF32("{}=dFdxFine({});", inst, op_a);
} else {
LOG_WARNING(Shader_GLSL, "Device does not support dFdxFine, fallback to dFdx");
ctx.AddF32("{}=dFdx({});", inst, op_a);
}
}
void EmitDPdyFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
if (ctx.profile.support_gl_derivative_control) {
ctx.AddF32("{}=dFdyFine({});", inst, op_a);
} else {
LOG_WARNING(Shader_GLSL, "Device does not support dFdyFine, fallback to dFdy");
ctx.AddF32("{}=dFdy({});", inst, op_a);
}
}
void EmitDPdxCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
if (ctx.profile.support_gl_derivative_control) {
ctx.AddF32("{}=dFdxCoarse({});", inst, op_a);
} else {
LOG_WARNING(Shader_GLSL, "Device does not support dFdxCoarse, fallback to dFdx");
ctx.AddF32("{}=dFdx({});", inst, op_a);
}
}
void EmitDPdyCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
if (ctx.profile.support_gl_derivative_control) {
ctx.AddF32("{}=dFdyCoarse({});", inst, op_a);
} else {
LOG_WARNING(Shader_GLSL, "Device does not support dFdyCoarse, fallback to dFdy");
ctx.AddF32("{}=dFdy({});", inst, op_a);
}
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,714 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "common/div_ceil.h"
#include "shader_recompiler/backend/bindings.h"
#include "shader_recompiler/backend/glsl/glsl_emit_context.h"
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/profile.h"
#include "shader_recompiler/runtime_info.h"
namespace Shader::Backend::GLSL {
namespace {
u32 CbufIndex(size_t offset) {
return (offset / 4) % 4;
}
char Swizzle(size_t offset) {
return "xyzw"[CbufIndex(offset)];
}
std::string_view InterpDecorator(Interpolation interp) {
switch (interp) {
case Interpolation::Smooth:
return "";
case Interpolation::Flat:
return "flat ";
case Interpolation::NoPerspective:
return "noperspective ";
}
throw InvalidArgument("Invalid interpolation {}", interp);
}
std::string_view InputArrayDecorator(Stage stage) {
switch (stage) {
case Stage::Geometry:
case Stage::TessellationControl:
case Stage::TessellationEval:
return "[]";
default:
return "";
}
}
bool StoresPerVertexAttributes(Stage stage) {
switch (stage) {
case Stage::VertexA:
case Stage::VertexB:
case Stage::Geometry:
case Stage::TessellationEval:
return true;
default:
return false;
}
}
std::string OutputDecorator(Stage stage, u32 size) {
switch (stage) {
case Stage::TessellationControl:
return fmt::format("[{}]", size);
default:
return "";
}
}
std::string_view DepthSamplerType(TextureType type) {
switch (type) {
case TextureType::Color1D:
return "sampler1DShadow";
case TextureType::ColorArray1D:
return "sampler1DArrayShadow";
case TextureType::Color2D:
return "sampler2DShadow";
case TextureType::ColorArray2D:
return "sampler2DArrayShadow";
case TextureType::ColorCube:
return "samplerCubeShadow";
case TextureType::ColorArrayCube:
return "samplerCubeArrayShadow";
default:
throw NotImplementedException("Texture type: {}", type);
}
}
std::string_view ColorSamplerType(TextureType type, bool is_multisample = false) {
if (is_multisample) {
ASSERT(type == TextureType::Color2D || type == TextureType::ColorArray2D);
}
switch (type) {
case TextureType::Color1D:
return "sampler1D";
case TextureType::ColorArray1D:
return "sampler1DArray";
case TextureType::Color2D:
case TextureType::Color2DRect:
return is_multisample ? "sampler2DMS" : "sampler2D";
case TextureType::ColorArray2D:
return is_multisample ? "sampler2DMSArray" : "sampler2DArray";
case TextureType::Color3D:
return "sampler3D";
case TextureType::ColorCube:
return "samplerCube";
case TextureType::ColorArrayCube:
return "samplerCubeArray";
case TextureType::Buffer:
return "samplerBuffer";
default:
throw NotImplementedException("Texture type: {}", type);
}
}
std::string_view ImageType(TextureType type) {
switch (type) {
case TextureType::Color1D:
return "uimage1D";
case TextureType::ColorArray1D:
return "uimage1DArray";
case TextureType::Color2D:
return "uimage2D";
case TextureType::ColorArray2D:
return "uimage2DArray";
case TextureType::Color3D:
return "uimage3D";
case TextureType::ColorCube:
return "uimageCube";
case TextureType::ColorArrayCube:
return "uimageCubeArray";
case TextureType::Buffer:
return "uimageBuffer";
default:
throw NotImplementedException("Image type: {}", type);
}
}
std::string_view ImageFormatString(ImageFormat format) {
switch (format) {
case ImageFormat::Typeless:
return "";
case ImageFormat::R8_UINT:
return ",r8ui";
case ImageFormat::R8_SINT:
return ",r8i";
case ImageFormat::R16_UINT:
return ",r16ui";
case ImageFormat::R16_SINT:
return ",r16i";
case ImageFormat::R32_UINT:
return ",r32ui";
case ImageFormat::R32G32_UINT:
return ",rg32ui";
case ImageFormat::R32G32B32A32_UINT:
return ",rgba32ui";
default:
throw NotImplementedException("Image format: {}", format);
}
}
std::string_view ImageAccessQualifier(bool is_written, bool is_read) {
if (is_written && !is_read) {
return "writeonly ";
}
if (is_read && !is_written) {
return "readonly ";
}
return "";
}
std::string_view GetTessMode(TessPrimitive primitive) {
switch (primitive) {
case TessPrimitive::Triangles:
return "triangles";
case TessPrimitive::Quads:
return "quads";
case TessPrimitive::Isolines:
return "isolines";
}
throw InvalidArgument("Invalid tessellation primitive {}", primitive);
}
std::string_view GetTessSpacing(TessSpacing spacing) {
switch (spacing) {
case TessSpacing::Equal:
return "equal_spacing";
case TessSpacing::FractionalOdd:
return "fractional_odd_spacing";
case TessSpacing::FractionalEven:
return "fractional_even_spacing";
}
throw InvalidArgument("Invalid tessellation spacing {}", spacing);
}
std::string_view InputPrimitive(InputTopology topology) {
switch (topology) {
case InputTopology::Points:
return "points";
case InputTopology::Lines:
return "lines";
case InputTopology::LinesAdjacency:
return "lines_adjacency";
case InputTopology::Triangles:
return "triangles";
case InputTopology::TrianglesAdjacency:
return "triangles_adjacency";
}
throw InvalidArgument("Invalid input topology {}", topology);
}
std::string_view OutputPrimitive(OutputTopology topology) {
switch (topology) {
case OutputTopology::PointList:
return "points";
case OutputTopology::LineStrip:
return "line_strip";
case OutputTopology::TriangleStrip:
return "triangle_strip";
}
throw InvalidArgument("Invalid output topology {}", topology);
}
void SetupOutPerVertex(EmitContext& ctx, std::string& header) {
if (!StoresPerVertexAttributes(ctx.stage)) {
return;
}
if (ctx.uses_geometry_passthrough) {
return;
}
header += "out gl_PerVertex{vec4 gl_Position;";
if (ctx.info.stores[IR::Attribute::PointSize]) {
header += "float gl_PointSize;";
}
if (ctx.info.stores.ClipDistances()) {
header += "float gl_ClipDistance[];";
}
if (ctx.info.stores[IR::Attribute::ViewportIndex] &&
ctx.profile.support_viewport_index_layer_non_geometry && ctx.stage != Stage::Geometry) {
header += "int gl_ViewportIndex;";
}
header += "};";
if (ctx.info.stores[IR::Attribute::ViewportIndex] && ctx.stage == Stage::Geometry) {
header += "out int gl_ViewportIndex;";
}
}
void SetupInPerVertex(EmitContext& ctx, std::string& header) {
// Currently only required for TessellationControl to adhere to
// ARB_separate_shader_objects requirements
if (ctx.stage != Stage::TessellationControl) {
return;
}
const bool loads_position{ctx.info.loads.AnyComponent(IR::Attribute::PositionX)};
const bool loads_point_size{ctx.info.loads[IR::Attribute::PointSize]};
const bool loads_clip_distance{ctx.info.loads.ClipDistances()};
const bool loads_per_vertex{loads_position || loads_point_size || loads_clip_distance};
if (!loads_per_vertex) {
return;
}
header += "in gl_PerVertex{";
if (loads_position) {
header += "vec4 gl_Position;";
}
if (loads_point_size) {
header += "float gl_PointSize;";
}
if (loads_clip_distance) {
header += "float gl_ClipDistance[];";
}
header += "}gl_in[gl_MaxPatchVertices];";
}
} // Anonymous namespace
EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile& profile_,
const RuntimeInfo& runtime_info_)
: info{program.info}, profile{profile_}, runtime_info{runtime_info_}, stage{program.stage},
uses_geometry_passthrough{program.is_geometry_passthrough &&
profile.support_geometry_shader_passthrough} {
if (profile.need_fastmath_off) {
header += "#pragma optionNV(fastmath off)\n";
}
SetupExtensions();
switch (program.stage) {
case Stage::VertexA:
case Stage::VertexB:
stage_name = "vs";
break;
case Stage::TessellationControl:
stage_name = "tcs";
header += fmt::format("layout(vertices={})out;", program.invocations);
break;
case Stage::TessellationEval:
stage_name = "tes";
header += fmt::format("layout({},{},{})in;", GetTessMode(runtime_info.tess_primitive),
GetTessSpacing(runtime_info.tess_spacing),
runtime_info.tess_clockwise ? "cw" : "ccw");
break;
case Stage::Geometry:
stage_name = "gs";
header += fmt::format("layout({})in;", InputPrimitive(runtime_info.input_topology));
if (uses_geometry_passthrough) {
// Passthru REQUIRES the layout to be defined with a corresponding name, for our sanity
// we will just use `gl_in[]`, if you don't the driver will complain with:
// 0(56) : error C7593: Builtin block member gl_Position not found in redeclaration of in gl_PerVertex
header += "layout(passthrough)in gl_PerVertex{vec4 gl_Position;}gl_in[];";
break;
} else if (program.is_geometry_passthrough && !profile.support_geometry_shader_passthrough) {
LOG_WARNING(Shader_GLSL, "Passthrough geometry program used but not supported");
}
header += fmt::format(
"layout({},max_vertices={})out;in gl_PerVertex{{vec4 gl_Position;}}gl_in[];",
OutputPrimitive(program.output_topology), program.output_vertices);
break;
case Stage::Fragment:
stage_name = "fs";
position_name = "gl_FragCoord";
if (runtime_info.force_early_z) {
header += "layout(early_fragment_tests)in;";
}
break;
case Stage::Compute:
stage_name = "cs";
const u32 local_x{(std::max)(program.workgroup_size[0], 1u)};
const u32 local_y{(std::max)(program.workgroup_size[1], 1u)};
const u32 local_z{(std::max)(program.workgroup_size[2], 1u)};
header += fmt::format("layout(local_size_x={},local_size_y={},local_size_z={}) in;",
local_x, local_y, local_z);
break;
}
SetupOutPerVertex(*this, header);
SetupInPerVertex(*this, header);
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (!info.loads.Generic(index) || !runtime_info.previous_stage_stores.Generic(index)) {
continue;
}
const auto qualifier{uses_geometry_passthrough ? "passthrough"
: fmt::format("location={}", index)};
header += fmt::format("layout({}){}in vec4 in_attr{}{};", qualifier,
InterpDecorator(info.interpolation[index]), index,
InputArrayDecorator(stage));
}
for (size_t index = 0; index < info.uses_patches.size(); ++index) {
if (!info.uses_patches[index]) {
continue;
}
const auto qualifier{stage == Stage::TessellationControl ? "out" : "in"};
header += fmt::format("layout(location={})patch {} vec4 patch{};", index, qualifier, index);
}
if (stage == Stage::Fragment) {
for (size_t index = 0; index < info.stores_frag_color.size(); ++index) {
if (!info.stores_frag_color[index] && !profile.need_declared_frag_colors) {
continue;
}
header += fmt::format("layout(location={})out vec4 frag_color{};", index, index);
}
}
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (info.stores.Generic(index)) {
DefineGenericOutput(index, program.invocations);
}
}
if (info.uses_rescaling_uniform) {
header += "layout(location=0) uniform vec4 scaling;";
}
if (info.uses_render_area) {
header += "layout(location=1) uniform vec4 render_area;";
}
DefineConstantBuffers(bindings);
DefineConstantBufferIndirect();
DefineStorageBuffers(bindings);
SetupImages(bindings);
SetupTextures(bindings);
DefineHelperFunctions();
DefineConstants();
}
void EmitContext::SetupExtensions() {
header += "#extension GL_ARB_separate_shader_objects : enable\n";
if (info.uses_shadow_lod && profile.support_gl_texture_shadow_lod) {
header += "#extension GL_EXT_texture_shadow_lod : enable\n";
}
if (info.uses_int64 && profile.support_int64) {
header += "#extension GL_ARB_gpu_shader_int64 : enable\n";
}
if (info.uses_int64_bit_atomics) {
header += "#extension GL_NV_shader_atomic_int64 : enable\n";
}
if (info.uses_atomic_f32_add) {
header += "#extension GL_NV_shader_atomic_float : enable\n";
}
if (info.uses_atomic_f16x2_add || info.uses_atomic_f16x2_min || info.uses_atomic_f16x2_max) {
header += "#extension GL_NV_shader_atomic_fp16_vector : enable\n";
}
if (info.uses_fp16) {
if (profile.support_gl_nv_gpu_shader_5) {
header += "#extension GL_NV_gpu_shader5 : enable\n";
}
if (profile.support_gl_amd_gpu_shader_half_float) {
header += "#extension GL_AMD_gpu_shader_half_float : enable\n";
}
}
if (info.uses_subgroup_invocation_id || info.uses_subgroup_mask || info.uses_subgroup_vote ||
info.uses_subgroup_shuffles || info.uses_fswzadd) {
header += "#extension GL_ARB_shader_ballot : enable\n"
"#extension GL_ARB_shader_group_vote : enable\n";
if (!info.uses_int64 && profile.support_int64) {
header += "#extension GL_ARB_gpu_shader_int64 : enable\n";
}
if (profile.support_gl_warp_intrinsics) {
header += "#extension GL_NV_shader_thread_shuffle : enable\n";
}
}
if ((info.stores[IR::Attribute::ViewportIndex] || info.stores[IR::Attribute::Layer]) &&
profile.support_viewport_index_layer_non_geometry && stage != Stage::Geometry) {
header += "#extension GL_ARB_shader_viewport_layer_array : enable\n";
}
if (info.uses_sparse_residency && profile.support_gl_sparse_textures) {
header += "#extension GL_ARB_sparse_texture2 : enable\n";
}
if (info.stores[IR::Attribute::ViewportMask] && profile.support_viewport_mask) {
header += "#extension GL_NV_viewport_array2 : enable\n";
}
if (info.uses_typeless_image_reads) {
header += "#extension GL_EXT_shader_image_load_formatted : enable\n";
}
if (info.uses_derivatives && profile.support_gl_derivative_control) {
header += "#extension GL_ARB_derivative_control : enable\n";
}
if (uses_geometry_passthrough) {
header += "#extension GL_NV_geometry_shader_passthrough : enable\n";
}
}
void EmitContext::DefineConstantBuffers(Bindings& bindings) {
if (info.constant_buffer_descriptors.empty()) {
return;
}
for (const auto& desc : info.constant_buffer_descriptors) {
const auto cbuf_type{profile.has_gl_cbuf_ftou_bug ? "uvec4" : "vec4"};
const u32 cbuf_used_size{Common::DivCeil(info.constant_buffer_used_sizes[desc.index], 16U)};
const u32 cbuf_binding_size{info.uses_global_memory ? 0x1000U : cbuf_used_size};
header += fmt::format("layout(std140,binding={}) uniform {}_cbuf_{}{{{} {}_cbuf{}[{}];}};",
bindings.uniform_buffer, stage_name, desc.index, cbuf_type,
stage_name, desc.index, cbuf_binding_size);
bindings.uniform_buffer += desc.count;
}
}
void EmitContext::DefineConstantBufferIndirect() {
if (!info.uses_cbuf_indirect) {
return;
}
header += profile.has_gl_cbuf_ftou_bug ? "uvec4 " : "vec4 ";
header += "GetCbufIndirect(uint binding, uint offset){"
"switch(binding){"
"default:";
for (const auto& desc : info.constant_buffer_descriptors) {
header +=
fmt::format("case {}:return {}_cbuf{}[offset];", desc.index, stage_name, desc.index);
}
header += "}}";
}
void EmitContext::DefineStorageBuffers(Bindings& bindings) {
if (info.storage_buffers_descriptors.empty()) {
return;
}
u32 index{};
for (const auto& desc : info.storage_buffers_descriptors) {
header += fmt::format("layout(std430,binding={}) buffer {}_ssbo_{}{{uint {}_ssbo{}[];}};",
bindings.storage_buffer, stage_name, bindings.storage_buffer,
stage_name, index);
bindings.storage_buffer += desc.count;
index += desc.count;
}
}
void EmitContext::DefineGenericOutput(size_t index, u32 invocations) {
static constexpr std::string_view swizzle{"xyzw"};
const size_t base_index{static_cast<size_t>(IR::Attribute::Generic0X) + index * 4};
u32 element{0};
while (element < 4) {
std::string definition{fmt::format("layout(location={}", index)};
const u32 remainder{4 - element};
const TransformFeedbackVarying* xfb_varying{};
const size_t xfb_varying_index{base_index + element};
if (xfb_varying_index < runtime_info.xfb_count) {
xfb_varying = &runtime_info.xfb_varyings[xfb_varying_index];
xfb_varying = xfb_varying->components > 0 ? xfb_varying : nullptr;
}
const u32 num_components{xfb_varying ? xfb_varying->components : remainder};
if (element > 0) {
definition += fmt::format(",component={}", element);
}
if (xfb_varying) {
definition +=
fmt::format(",xfb_buffer={},xfb_stride={},xfb_offset={}", xfb_varying->buffer,
xfb_varying->stride, xfb_varying->offset);
}
std::string name{fmt::format("out_attr{}", index)};
if (num_components < 4 || element > 0) {
name += fmt::format("_{}", swizzle.substr(element, num_components));
}
const auto type{num_components == 1 ? "float" : fmt::format("vec{}", num_components)};
definition += fmt::format(")out {} {}{};", type, name, OutputDecorator(stage, invocations));
header += definition;
const GenericElementInfo element_info{
.name = name,
.first_element = element,
.num_components = num_components,
};
std::fill_n(output_generics[index].begin() + element, num_components, element_info);
element += num_components;
}
}
void EmitContext::DefineHelperFunctions() {
header += "\n#define ftoi floatBitsToInt\n#define ftou floatBitsToUint\n"
"#define itof intBitsToFloat\n#define utof uintBitsToFloat\n";
if (info.uses_global_increment || info.uses_shared_increment) {
header += "uint CasIncrement(uint op_a,uint op_b){return op_a>=op_b?0u:(op_a+1u);}";
}
if (info.uses_global_decrement || info.uses_shared_decrement) {
header += "uint CasDecrement(uint op_a,uint op_b){"
"return op_a==0||op_a>op_b?op_b:(op_a-1u);}";
}
if (info.uses_atomic_f32_add) {
header += "uint CasFloatAdd(uint op_a,float op_b){"
"return ftou(utof(op_a)+op_b);}";
}
if (info.uses_atomic_f32x2_add) {
header += "uint CasFloatAdd32x2(uint op_a,vec2 op_b){"
"return packHalf2x16(unpackHalf2x16(op_a)+op_b);}";
}
if (info.uses_atomic_f32x2_min) {
header += "uint CasFloatMin32x2(uint op_a,vec2 op_b){return "
"packHalf2x16(min(unpackHalf2x16(op_a),op_b));}";
}
if (info.uses_atomic_f32x2_max) {
header += "uint CasFloatMax32x2(uint op_a,vec2 op_b){return "
"packHalf2x16(max(unpackHalf2x16(op_a),op_b));}";
}
if (info.uses_atomic_f16x2_add) {
header += "uint CasFloatAdd16x2(uint op_a,f16vec2 op_b){return "
"packFloat2x16(unpackFloat2x16(op_a)+op_b);}";
}
if (info.uses_atomic_f16x2_min) {
header += "uint CasFloatMin16x2(uint op_a,f16vec2 op_b){return "
"packFloat2x16(min(unpackFloat2x16(op_a),op_b));}";
}
if (info.uses_atomic_f16x2_max) {
header += "uint CasFloatMax16x2(uint op_a,f16vec2 op_b){return "
"packFloat2x16(max(unpackFloat2x16(op_a),op_b));}";
}
if (info.uses_atomic_s32_min) {
header += "uint CasMinS32(uint op_a,uint op_b){return uint(min(int(op_a),int(op_b)));}";
}
if (info.uses_atomic_s32_max) {
header += "uint CasMaxS32(uint op_a,uint op_b){return uint(max(int(op_a),int(op_b)));}";
}
if (info.uses_global_memory && profile.support_int64) {
header += DefineGlobalMemoryFunctions();
}
if (info.loads_indexed_attributes) {
const bool is_array{stage == Stage::Geometry};
const auto vertex_arg{is_array ? ",uint vertex" : ""};
std::string func{
fmt::format("float IndexedAttrLoad(int offset{}){{int base_index=offset>>2;uint "
"masked_index=uint(base_index)&3u;switch(base_index>>2){{",
vertex_arg)};
if (info.loads.AnyComponent(IR::Attribute::PositionX)) {
const auto position_idx{is_array ? "gl_in[vertex]." : ""};
func += fmt::format("case {}:return {}{}[masked_index];",
static_cast<u32>(IR::Attribute::PositionX) >> 2, position_idx,
position_name);
}
const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2;
for (u32 index = 0; index < IR::NUM_GENERICS; ++index) {
if (!info.loads.Generic(index)) {
continue;
}
const auto vertex_idx{is_array ? "[vertex]" : ""};
func += fmt::format("case {}:return in_attr{}{}[masked_index];",
base_attribute_value + index, index, vertex_idx);
}
func += "default: return 0.0;}}";
header += func;
}
if (info.stores_indexed_attributes) {
// TODO
}
}
std::string EmitContext::DefineGlobalMemoryFunctions() {
const auto define_body{[&](std::string& func, size_t index, std::string_view return_statement) {
const auto& ssbo{info.storage_buffers_descriptors[index]};
const u32 size_cbuf_offset{ssbo.cbuf_offset + 8};
const auto ssbo_addr{fmt::format("ssbo_addr{}", index)};
const auto cbuf{fmt::format("{}_cbuf{}", stage_name, ssbo.cbuf_index)};
std::array<std::string, 2> addr_xy;
std::array<std::string, 2> size_xy;
for (size_t i = 0; i < addr_xy.size(); ++i) {
const auto addr_loc{ssbo.cbuf_offset + 4 * i};
const auto size_loc{size_cbuf_offset + 4 * i};
addr_xy[i] = fmt::format("ftou({}[{}].{})", cbuf, addr_loc / 16, Swizzle(addr_loc));
size_xy[i] = fmt::format("ftou({}[{}].{})", cbuf, size_loc / 16, Swizzle(size_loc));
}
const u32 ssbo_align_mask{~(static_cast<u32>(profile.min_ssbo_alignment) - 1U)};
const auto aligned_low_addr{fmt::format("{}&{}", addr_xy[0], ssbo_align_mask)};
const auto aligned_addr{fmt::format("uvec2({},{})", aligned_low_addr, addr_xy[1])};
const auto addr_pack{fmt::format("packUint2x32({})", aligned_addr)};
const auto addr_statement{fmt::format("uint64_t {}={};", ssbo_addr, addr_pack)};
func += addr_statement;
const auto size_vec{fmt::format("uvec2({},{})", size_xy[0], size_xy[1])};
const auto comp_lhs{fmt::format("(addr>={})", ssbo_addr)};
const auto comp_rhs{fmt::format("(addr<({}+uint64_t({})))", ssbo_addr, size_vec)};
const auto comparison{fmt::format("if({}&&{}){{", comp_lhs, comp_rhs)};
func += comparison;
const auto ssbo_name{fmt::format("{}_ssbo{}", stage_name, index)};
func += fmt::format(fmt::runtime(return_statement), ssbo_name, ssbo_addr);
}};
std::string write_func{"void WriteGlobal32(uint64_t addr,uint data){"};
std::string write_func_64{"void WriteGlobal64(uint64_t addr,uvec2 data){"};
std::string write_func_128{"void WriteGlobal128(uint64_t addr,uvec4 data){"};
std::string load_func{"uint LoadGlobal32(uint64_t addr){"};
std::string load_func_64{"uvec2 LoadGlobal64(uint64_t addr){"};
std::string load_func_128{"uvec4 LoadGlobal128(uint64_t addr){"};
const size_t num_buffers{info.storage_buffers_descriptors.size()};
for (size_t index = 0; index < num_buffers; ++index) {
if (!info.nvn_buffer_used[index]) {
continue;
}
define_body(write_func, index, "{0}[uint(addr-{1})>>2]=data;return;}}");
define_body(write_func_64, index,
"{0}[uint(addr-{1})>>2]=data.x;{0}[uint(addr-{1}+4)>>2]=data.y;return;}}");
define_body(write_func_128, index,
"{0}[uint(addr-{1})>>2]=data.x;{0}[uint(addr-{1}+4)>>2]=data.y;{0}[uint("
"addr-{1}+8)>>2]=data.z;{0}[uint(addr-{1}+12)>>2]=data.w;return;}}");
define_body(load_func, index, "return {0}[uint(addr-{1})>>2];}}");
define_body(load_func_64, index,
"return uvec2({0}[uint(addr-{1})>>2],{0}[uint(addr-{1}+4)>>2]);}}");
define_body(load_func_128, index,
"return uvec4({0}[uint(addr-{1})>>2],{0}[uint(addr-{1}+4)>>2],{0}["
"uint(addr-{1}+8)>>2],{0}[uint(addr-{1}+12)>>2]);}}");
}
write_func += '}';
write_func_64 += '}';
write_func_128 += '}';
load_func += "return 0u;}";
load_func_64 += "return uvec2(0);}";
load_func_128 += "return uvec4(0);}";
return write_func + write_func_64 + write_func_128 + load_func + load_func_64 + load_func_128;
}
void EmitContext::SetupImages(Bindings& bindings) {
image_buffers.reserve(info.image_buffer_descriptors.size());
for (const auto& desc : info.image_buffer_descriptors) {
image_buffers.push_back({bindings.image, desc.count});
const auto format{ImageFormatString(desc.format)};
const auto qualifier{ImageAccessQualifier(desc.is_written, desc.is_read)};
const auto array_decorator{desc.count > 1 ? fmt::format("[{}]", desc.count) : ""};
header += fmt::format("layout(binding={}{}) uniform {}uimageBuffer img{}{};",
bindings.image, format, qualifier, bindings.image, array_decorator);
bindings.image += desc.count;
}
images.reserve(info.image_descriptors.size());
for (const auto& desc : info.image_descriptors) {
images.push_back({bindings.image, desc.count});
const auto format{ImageFormatString(desc.format)};
const auto image_type{ImageType(desc.type)};
const auto qualifier{ImageAccessQualifier(desc.is_written, desc.is_read)};
const auto array_decorator{desc.count > 1 ? fmt::format("[{}]", desc.count) : ""};
header += fmt::format("layout(binding={}{})uniform {}{} img{}{};", bindings.image, format,
qualifier, image_type, bindings.image, array_decorator);
bindings.image += desc.count;
}
}
void EmitContext::SetupTextures(Bindings& bindings) {
texture_buffers.reserve(info.texture_buffer_descriptors.size());
for (const auto& desc : info.texture_buffer_descriptors) {
texture_buffers.push_back({bindings.texture, desc.count});
const auto sampler_type{ColorSamplerType(TextureType::Buffer)};
const auto array_decorator{desc.count > 1 ? fmt::format("[{}]", desc.count) : ""};
header += fmt::format("layout(binding={}) uniform {} tex{}{};", bindings.texture,
sampler_type, bindings.texture, array_decorator);
bindings.texture += desc.count;
}
textures.reserve(info.texture_descriptors.size());
for (const auto& desc : info.texture_descriptors) {
textures.push_back({bindings.texture, desc.count});
const auto sampler_type{desc.is_depth ? DepthSamplerType(desc.type)
: ColorSamplerType(desc.type, desc.is_multisample)};
const auto array_decorator{desc.count > 1 ? fmt::format("[{}]", desc.count) : ""};
header += fmt::format("layout(binding={}) uniform {} tex{}{};", bindings.texture,
sampler_type, bindings.texture, array_decorator);
bindings.texture += desc.count;
}
}
void EmitContext::DefineConstants() {
if (info.uses_fswzadd) {
header += "const float FSWZ_A[]=float[4](-1.f,1.f,-1.f,0.f);"
"const float FSWZ_B[]=float[4](-1.f,-1.f,1.f,-1.f);";
}
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,174 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <string>
#include <utility>
#include <vector>
#include <fmt/ranges.h>
#include "shader_recompiler/backend/glsl/var_alloc.h"
#include "shader_recompiler/stage.h"
namespace Shader {
struct Info;
struct Profile;
struct RuntimeInfo;
} // namespace Shader
namespace Shader::Backend {
struct Bindings;
}
namespace Shader::IR {
class Inst;
struct Program;
} // namespace Shader::IR
namespace Shader::Backend::GLSL {
struct GenericElementInfo {
std::string name;
u32 first_element{};
u32 num_components{};
};
struct TextureImageDefinition {
u32 binding;
u32 count;
};
class EmitContext {
public:
explicit EmitContext(IR::Program& program, Bindings& bindings, const Profile& profile_,
const RuntimeInfo& runtime_info_);
template <GlslVarType type, typename... Args>
void Add(const char* format_str, IR::Inst& inst, Args&&... args) {
const auto var_def{var_alloc.AddDefine(inst, type)};
if (var_def.empty()) {
// skip assignment.
code += fmt::format(fmt::runtime(format_str + 3), std::forward<Args>(args)...);
} else {
code += fmt::format(fmt::runtime(format_str), var_def, std::forward<Args>(args)...);
}
// TODO: Remove this
code += '\n';
}
template <typename... Args>
void AddU1(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::U1>(format_str, inst, args...);
}
template <typename... Args>
void AddF16x2(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::F16x2>(format_str, inst, args...);
}
template <typename... Args>
void AddU32(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::U32>(format_str, inst, args...);
}
template <typename... Args>
void AddF32(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::F32>(format_str, inst, args...);
}
template <typename... Args>
void AddU64(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::U64>(format_str, inst, args...);
}
template <typename... Args>
void AddF64(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::F64>(format_str, inst, args...);
}
template <typename... Args>
void AddU32x2(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::U32x2>(format_str, inst, args...);
}
template <typename... Args>
void AddF32x2(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::F32x2>(format_str, inst, args...);
}
template <typename... Args>
void AddU32x3(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::U32x3>(format_str, inst, args...);
}
template <typename... Args>
void AddF32x3(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::F32x3>(format_str, inst, args...);
}
template <typename... Args>
void AddU32x4(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::U32x4>(format_str, inst, args...);
}
template <typename... Args>
void AddF32x4(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::F32x4>(format_str, inst, args...);
}
template <typename... Args>
void AddPrecF32(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::PrecF32>(format_str, inst, args...);
}
template <typename... Args>
void AddPrecF64(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<GlslVarType::PrecF64>(format_str, inst, args...);
}
template <typename... Args>
void Add(const char* format_str, Args&&... args) {
code += fmt::format(fmt::runtime(format_str), std::forward<Args>(args)...);
// TODO: Remove this
code += '\n';
}
std::string header;
std::string code;
VarAlloc var_alloc;
const Info& info;
const Profile& profile;
const RuntimeInfo& runtime_info;
Stage stage{};
std::string_view stage_name = "invalid";
std::string_view position_name = "gl_Position";
std::vector<TextureImageDefinition> texture_buffers;
std::vector<TextureImageDefinition> image_buffers;
std::vector<TextureImageDefinition> textures;
std::vector<TextureImageDefinition> images;
std::array<std::array<GenericElementInfo, 4>, 32> output_generics{};
u32 num_safety_loop_vars{};
bool uses_y_direction{};
bool uses_cc_carry{};
bool uses_geometry_passthrough{};
private:
void SetupExtensions();
void DefineConstantBuffers(Bindings& bindings);
void DefineConstantBufferIndirect();
void DefineStorageBuffers(Bindings& bindings);
void DefineGenericOutput(size_t index, u32 invocations);
void DefineHelperFunctions();
void DefineConstants();
std::string DefineGlobalMemoryFunctions();
void SetupImages(Bindings& bindings);
void SetupTextures(Bindings& bindings);
};
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,309 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string>
#include <string_view>
#include <fmt/ranges.h>
#include "shader_recompiler/backend/glsl/var_alloc.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::GLSL {
namespace {
std::string TypePrefix(GlslVarType type) {
switch (type) {
case GlslVarType::U1:
return "b_";
case GlslVarType::F16x2:
return "f16x2_";
case GlslVarType::U32:
return "u_";
case GlslVarType::F32:
return "f_";
case GlslVarType::U64:
return "u64_";
case GlslVarType::F64:
return "d_";
case GlslVarType::U32x2:
return "u2_";
case GlslVarType::F32x2:
return "f2_";
case GlslVarType::U32x3:
return "u3_";
case GlslVarType::F32x3:
return "f3_";
case GlslVarType::U32x4:
return "u4_";
case GlslVarType::F32x4:
return "f4_";
case GlslVarType::PrecF32:
return "pf_";
case GlslVarType::PrecF64:
return "pd_";
case GlslVarType::Void:
return "";
default:
throw NotImplementedException("Type {}", type);
}
}
std::string FormatFloat(std::string_view value, IR::Type type) {
// TODO: Confirm FP64 nan/inf
if (type == IR::Type::F32) {
if (value == "nan") {
return "utof(0x7fc00000)";
}
if (value == "inf") {
return "utof(0x7f800000)";
}
if (value == "-inf") {
return "utof(0xff800000)";
}
}
if (value.find_first_of('e') != std::string_view::npos) {
// scientific notation
const auto cast{type == IR::Type::F32 ? "float" : "double"};
return fmt::format("{}({})", cast, value);
}
const bool needs_dot{value.find_first_of('.') == std::string_view::npos};
const bool needs_suffix{!value.ends_with('f')};
const auto suffix{type == IR::Type::F32 ? "f" : "lf"};
return fmt::format("{}{}{}", value, needs_dot ? "." : "", needs_suffix ? suffix : "");
}
std::string MakeImm(const IR::Value& value) {
switch (value.Type()) {
case IR::Type::U1:
return fmt::format("{}", value.U1() ? "true" : "false");
case IR::Type::U32:
return fmt::format("{}u", value.U32());
case IR::Type::F32:
return FormatFloat(fmt::format("{}", value.F32()), IR::Type::F32);
case IR::Type::U64:
return fmt::format("{}ul", value.U64());
case IR::Type::F64:
return FormatFloat(fmt::format("{}", value.F64()), IR::Type::F64);
case IR::Type::Void:
return "";
default:
throw NotImplementedException("Immediate type {}", value.Type());
}
}
} // Anonymous namespace
std::string VarAlloc::Representation(u32 index, GlslVarType type) const {
const auto prefix{TypePrefix(type)};
return fmt::format("{}{}", prefix, index);
}
std::string VarAlloc::Representation(Id id) const {
return Representation(id.index, id.type);
}
std::string VarAlloc::Define(IR::Inst& inst, GlslVarType type) {
if (inst.HasUses()) {
inst.SetDefinition<Id>(Alloc(type));
return Representation(inst.Definition<Id>());
} else {
Id id{};
id.type.Assign(type);
GetUseTracker(type).uses_temp = true;
inst.SetDefinition<Id>(id);
return 't' + Representation(inst.Definition<Id>());
}
}
std::string VarAlloc::Define(IR::Inst& inst, IR::Type type) {
return Define(inst, RegType(type));
}
std::string VarAlloc::PhiDefine(IR::Inst& inst, IR::Type type) {
return AddDefine(inst, RegType(type));
}
std::string VarAlloc::AddDefine(IR::Inst& inst, GlslVarType type) {
if (inst.HasUses()) {
inst.SetDefinition<Id>(Alloc(type));
return Representation(inst.Definition<Id>());
} else {
return "";
}
}
std::string VarAlloc::Consume(const IR::Value& value) {
return value.IsImmediate() ? MakeImm(value) : ConsumeInst(*value.InstRecursive());
}
std::string VarAlloc::ConsumeInst(IR::Inst& inst) {
inst.DestructiveRemoveUsage();
if (!inst.HasUses()) {
Free(inst.Definition<Id>());
}
return Representation(inst.Definition<Id>());
}
std::string VarAlloc::GetGlslType(IR::Type type) const {
return GetGlslType(RegType(type));
}
Id VarAlloc::Alloc(GlslVarType type) {
auto& use_tracker{GetUseTracker(type)};
const auto num_vars{use_tracker.var_use.size()};
for (size_t var = 0; var < num_vars; ++var) {
if (use_tracker.var_use[var]) {
continue;
}
use_tracker.num_used = (std::max)(use_tracker.num_used, var + 1);
use_tracker.var_use[var] = true;
Id ret{};
ret.is_valid.Assign(1);
ret.type.Assign(type);
ret.index.Assign(static_cast<u32>(var));
return ret;
}
// Allocate a new variable
use_tracker.var_use.push_back(true);
Id ret{};
ret.is_valid.Assign(1);
ret.type.Assign(type);
ret.index.Assign(static_cast<u32>(use_tracker.num_used));
++use_tracker.num_used;
return ret;
}
void VarAlloc::Free(Id id) {
if (id.is_valid == 0) {
throw LogicError("Freeing invalid variable");
}
auto& use_tracker{GetUseTracker(id.type)};
use_tracker.var_use[id.index] = false;
}
GlslVarType VarAlloc::RegType(IR::Type type) const {
switch (type) {
case IR::Type::U1:
return GlslVarType::U1;
case IR::Type::U32:
return GlslVarType::U32;
case IR::Type::F32:
return GlslVarType::F32;
case IR::Type::U64:
return GlslVarType::U64;
case IR::Type::F64:
return GlslVarType::F64;
default:
throw NotImplementedException("IR type {}", type);
}
}
std::string VarAlloc::GetGlslType(GlslVarType type) const {
switch (type) {
case GlslVarType::U1:
return "bool";
case GlslVarType::F16x2:
return "f16vec2";
case GlslVarType::U32:
return "uint";
case GlslVarType::F32:
case GlslVarType::PrecF32:
return "float";
case GlslVarType::U64:
return "uint64_t";
case GlslVarType::F64:
case GlslVarType::PrecF64:
return "double";
case GlslVarType::U32x2:
return "uvec2";
case GlslVarType::F32x2:
return "vec2";
case GlslVarType::U32x3:
return "uvec3";
case GlslVarType::F32x3:
return "vec3";
case GlslVarType::U32x4:
return "uvec4";
case GlslVarType::F32x4:
return "vec4";
case GlslVarType::Void:
return "";
default:
throw NotImplementedException("Type {}", type);
}
}
VarAlloc::UseTracker& VarAlloc::GetUseTracker(GlslVarType type) {
switch (type) {
case GlslVarType::U1:
return var_bool;
case GlslVarType::F16x2:
return var_f16x2;
case GlslVarType::U32:
return var_u32;
case GlslVarType::F32:
return var_f32;
case GlslVarType::U64:
return var_u64;
case GlslVarType::F64:
return var_f64;
case GlslVarType::U32x2:
return var_u32x2;
case GlslVarType::F32x2:
return var_f32x2;
case GlslVarType::U32x3:
return var_u32x3;
case GlslVarType::F32x3:
return var_f32x3;
case GlslVarType::U32x4:
return var_u32x4;
case GlslVarType::F32x4:
return var_f32x4;
case GlslVarType::PrecF32:
return var_precf32;
case GlslVarType::PrecF64:
return var_precf64;
default:
throw NotImplementedException("Type {}", type);
}
}
const VarAlloc::UseTracker& VarAlloc::GetUseTracker(GlslVarType type) const {
switch (type) {
case GlslVarType::U1:
return var_bool;
case GlslVarType::F16x2:
return var_f16x2;
case GlslVarType::U32:
return var_u32;
case GlslVarType::F32:
return var_f32;
case GlslVarType::U64:
return var_u64;
case GlslVarType::F64:
return var_f64;
case GlslVarType::U32x2:
return var_u32x2;
case GlslVarType::F32x2:
return var_f32x2;
case GlslVarType::U32x3:
return var_u32x3;
case GlslVarType::F32x3:
return var_f32x3;
case GlslVarType::U32x4:
return var_u32x4;
case GlslVarType::F32x4:
return var_f32x4;
case GlslVarType::PrecF32:
return var_precf32;
case GlslVarType::PrecF64:
return var_precf64;
default:
throw NotImplementedException("Type {}", type);
}
}
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,104 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <bitset>
#include <string>
#include <vector>
#include "common/bit_field.h"
#include "common/common_types.h"
namespace Shader::IR {
class Inst;
class Value;
enum class Type;
} // namespace Shader::IR
namespace Shader::Backend::GLSL {
enum class GlslVarType : u32 {
U1,
F16x2,
U32,
F32,
U64,
F64,
U32x2,
F32x2,
U32x3,
F32x3,
U32x4,
F32x4,
PrecF32,
PrecF64,
Void,
};
struct Id {
union {
u32 raw;
BitField<0, 1, u32> is_valid;
BitField<1, 4, GlslVarType> type;
BitField<6, 26, u32> index;
};
bool operator==(Id rhs) const noexcept {
return raw == rhs.raw;
}
bool operator!=(Id rhs) const noexcept {
return !operator==(rhs);
}
};
static_assert(sizeof(Id) == sizeof(u32));
class VarAlloc {
public:
struct UseTracker {
bool uses_temp{};
size_t num_used{};
std::vector<bool> var_use;
};
/// Used for explicit usages of variables, may revert to temporaries
std::string Define(IR::Inst& inst, GlslVarType type);
std::string Define(IR::Inst& inst, IR::Type type);
/// Used to assign variables used by the IR. May return a blank string if
/// the instruction's result is unused in the IR.
std::string AddDefine(IR::Inst& inst, GlslVarType type);
std::string PhiDefine(IR::Inst& inst, IR::Type type);
std::string Consume(const IR::Value& value);
std::string ConsumeInst(IR::Inst& inst);
std::string GetGlslType(GlslVarType type) const;
std::string GetGlslType(IR::Type type) const;
const UseTracker& GetUseTracker(GlslVarType type) const;
std::string Representation(u32 index, GlslVarType type) const;
private:
GlslVarType RegType(IR::Type type) const;
Id Alloc(GlslVarType type);
void Free(Id id);
UseTracker& GetUseTracker(GlslVarType type);
std::string Representation(Id id) const;
UseTracker var_bool{};
UseTracker var_f16x2{};
UseTracker var_u32{};
UseTracker var_u32x2{};
UseTracker var_u32x3{};
UseTracker var_u32x4{};
UseTracker var_f32{};
UseTracker var_f32x2{};
UseTracker var_f32x3{};
UseTracker var_f32x4{};
UseTracker var_u64{};
UseTracker var_f64{};
UseTracker var_precf32{};
UseTracker var_precf64{};
};
} // namespace Shader::Backend::GLSL

View File

@@ -0,0 +1,565 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <cstdlib>
#include <string>
#include <tuple>
#include <type_traits>
#include "common/div_ceil.h"
#include "common/settings.h"
#include "shader_recompiler/backend/wgsl/emit_wgsl.h"
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/ir_emitter.h"
namespace Shader::Backend::WGSL {
namespace {
template <class Func>
struct FuncTraits {};
template <class ReturnType_, class... Args>
struct FuncTraits<ReturnType_ (*)(Args...)> {
using ReturnType = ReturnType_;
static constexpr size_t NUM_ARGS = sizeof...(Args);
template <size_t I>
using ArgType = std::tuple_element_t<I, std::tuple<Args...>>;
};
template <auto func, typename... Args>
void SetDefinition(EmitContext& ctx, IR::Inst* inst, Args... args) {
inst->SetDefinition<Id>(func(ctx, std::forward<Args>(args)...));
}
template <typename ArgType>
auto Arg(EmitContext& ctx, const IR::Value& arg) {
if constexpr (std::is_same_v<ArgType, std::string_view>) {
return ctx.var_alloc.Consume(arg);
} else if constexpr (std::is_same_v<ArgType, const IR::Value&>) {
return arg;
} else if constexpr (std::is_same_v<ArgType, u32>) {
return arg.U32();
} else if constexpr (std::is_same_v<ArgType, IR::Attribute>) {
return arg.Attribute();
} else if constexpr (std::is_same_v<ArgType, IR::Patch>) {
return arg.Patch();
} else if constexpr (std::is_same_v<ArgType, IR::Reg>) {
return arg.Reg();
}
}
template <auto func, bool is_first_arg_inst, size_t... I>
void Invoke(EmitContext& ctx, IR::Inst* inst, std::index_sequence<I...>) {
using Traits = FuncTraits<decltype(func)>;
constexpr size_t off = is_first_arg_inst ? 2 : 1;
constexpr size_t N = sizeof...(I);
// The operands are consumed last to first, in a defined order: Consume has side effects (a value's last use
// releases its variable for reuse) and this backend was written and checked where function arguments are
// evaluated right to left (the MSVC ABI). Compiled for wasm, which evaluates them left to right, comparisons
// came out as "x > x" and constants changed. (A braced list is evaluated in order.)
std::tuple<decltype(Arg<typename Traits::template ArgType<N - 1 - I + off>>(ctx, inst->Arg(N - 1 - I)))...> rev{
Arg<typename Traits::template ArgType<N - 1 - I + off>>(ctx, inst->Arg(N - 1 - I))...};
(void)rev;
if constexpr (std::is_same_v<typename Traits::ReturnType, Id>) {
if constexpr (is_first_arg_inst) {
SetDefinition<func>(ctx, inst, *inst, std::get<N - 1 - I>(rev)...);
} else {
SetDefinition<func>(ctx, inst, std::get<N - 1 - I>(rev)...);
}
} else {
if constexpr (is_first_arg_inst) {
func(ctx, *inst, std::get<N - 1 - I>(rev)...);
} else {
func(ctx, std::get<N - 1 - I>(rev)...);
}
}
}
template <auto func>
void Invoke(EmitContext& ctx, IR::Inst* inst) {
using Traits = FuncTraits<decltype(func)>;
static_assert(Traits::NUM_ARGS >= 1, "Insufficient arguments");
if constexpr (Traits::NUM_ARGS == 1) {
Invoke<func, false>(ctx, inst, std::make_index_sequence<0>{});
} else {
using FirstArgType = typename Traits::template ArgType<1>;
static constexpr bool is_first_arg_inst = std::is_same_v<FirstArgType, IR::Inst&>;
using Indices = std::make_index_sequence<Traits::NUM_ARGS - (is_first_arg_inst ? 2 : 1)>;
Invoke<func, is_first_arg_inst>(ctx, inst, Indices{});
}
}
void EmitInst(EmitContext& ctx, IR::Inst* inst) {
switch (inst->GetOpcode()) {
#define OPCODE(name, result_type, ...) \
case IR::Opcode::name: \
return Invoke<&Emit##name>(ctx, inst);
#include "shader_recompiler/frontend/ir/opcodes.inc"
#undef OPCODE
}
throw LogicError("Invalid opcode {}", inst->GetOpcode());
}
bool IsReference(IR::Inst& inst) {
return inst.GetOpcode() == IR::Opcode::Reference;
}
void PrecolorInst(IR::Inst& phi) {
// Insert phi moves before references to avoid overwriting other phis
const size_t num_args{phi.NumArgs()};
for (size_t i = 0; i < num_args; ++i) {
IR::Block& phi_block{*phi.PhiBlock(i)};
auto it{std::find_if_not(phi_block.rbegin(), phi_block.rend(), IsReference).base()};
IR::IREmitter ir{phi_block, it};
const IR::Value arg{phi.Arg(i)};
if (arg.IsImmediate()) {
ir.PhiMove(phi, arg);
} else {
ir.PhiMove(phi, IR::Value{arg.InstRecursive()});
}
}
for (size_t i = 0; i < num_args; ++i) {
IR::IREmitter{*phi.PhiBlock(i)}.Reference(IR::Value{&phi});
}
}
void Precolor(const IR::Program& program) {
for (IR::Block* const block : program.blocks) {
for (IR::Inst& phi : block->Instructions()) {
if (!IR::IsPhi(phi)) {
break;
}
PrecolorInst(phi);
}
}
}
void EmitCode(EmitContext& ctx, const IR::Program& program) {
for (const IR::AbstractSyntaxNode& node : program.syntax_list) {
switch (node.type) {
case IR::AbstractSyntaxNode::Type::Block:
for (IR::Inst& inst : node.data.block->Instructions()) {
EmitInst(ctx, &inst);
}
break;
case IR::AbstractSyntaxNode::Type::If:
ctx.Add("if ({}) {{", ctx.var_alloc.Consume(node.data.if_node.cond));
break;
case IR::AbstractSyntaxNode::Type::EndIf:
ctx.Add("}}");
break;
case IR::AbstractSyntaxNode::Type::Break:
if (node.data.break_node.cond.IsImmediate()) {
if (node.data.break_node.cond.U1()) {
ctx.Add("break;");
}
} else {
ctx.Add("if ({}) {{ break; }}", ctx.var_alloc.Consume(node.data.break_node.cond));
}
break;
case IR::AbstractSyntaxNode::Type::Return:
case IR::AbstractSyntaxNode::Type::Unreachable:
ctx.Add("return;");
break;
case IR::AbstractSyntaxNode::Type::Loop:
ctx.Add("loop {{");
break;
case IR::AbstractSyntaxNode::Type::Repeat:
if (Settings::values.disable_shader_loop_safety_checks) {
ctx.Add("if (!{}) {{ break; }} }}", ctx.var_alloc.Consume(node.data.repeat.cond));
} else {
const u32 n{ctx.num_safety_loop_vars++};
ctx.Add("loop{} = loop{} - 1; if (loop{} < 0 || !{}) {{ break; }} }}", n, n, n,
ctx.var_alloc.Consume(node.data.repeat.cond));
}
break;
default:
throw NotImplementedException("AbstractSyntaxNode Type {}", node.type);
}
}
}
void DefineVariables(const EmitContext& ctx, std::string& header) {
for (u32 i = 0; i < static_cast<u32>(WgslVarType::Void); ++i) {
const auto type{static_cast<WgslVarType>(i)};
const auto& tracker{ctx.var_alloc.GetUseTracker(type)};
if (!tracker.uses_temp && tracker.num_used == 0) {
continue;
}
const auto type_name{ctx.var_alloc.GetWgslType(type)};
// Temps/return types that are never used are stored at index 0
if (tracker.uses_temp) {
header += fmt::format("var t{}: {};\n", ctx.var_alloc.Representation(0, type), type_name);
}
for (u32 index = 0; index < tracker.num_used; ++index) {
header += fmt::format("var {}: {};\n", ctx.var_alloc.Representation(index, type), type_name);
}
}
for (u32 i = 0; i < ctx.num_safety_loop_vars; ++i) {
header += fmt::format("var loop{}: i32 = 0x2000;\n", i);
}
}
std::string_view AttributeTypeName(AttributeType type) {
switch (type) {
case AttributeType::SignedInt:
case AttributeType::SignedScaled:
return "vec4<i32>";
case AttributeType::UnsignedInt:
case AttributeType::UnsignedScaled:
return "vec4<u32>";
default:
return "vec4<f32>";
}
}
bool IsIntegerAttribute(AttributeType type) {
return type == AttributeType::SignedInt || type == AttributeType::UnsignedInt;
}
std::string_view InterpolationAttribute(Interpolation interp) {
switch (interp) {
case Interpolation::Flat:
return "@interpolate(flat) ";
case Interpolation::NoPerspective:
return "@interpolate(linear) ";
default:
return "";
}
}
/// A pulled vertex attribute (Options::VertexPull): its bytes read from the stream's storage binding and converted
/// as the vertex fetch would (Maxwell's attribute format word: offset 7-20, size 21-26, type 27-29, bgra 31).
std::string PulledAttribute(const EmitContext& ctx, size_t i, const Options::VertexPull& p) {
const u32 a{p.attribute}, offset{(a >> 7) & 0x3fff}, size{(a >> 21) & 0x3f}, type{(a >> 27) & 7};
const bool bgra{(a >> 31) != 0};
int bits[4]{};
switch (size) {
case 0x01: bits[0] = bits[1] = bits[2] = bits[3] = 32; break;
case 0x02: bits[0] = bits[1] = bits[2] = 32; break;
case 0x03: bits[0] = bits[1] = bits[2] = bits[3] = 16; break;
case 0x04: bits[0] = bits[1] = 32; break;
case 0x05: bits[0] = bits[1] = bits[2] = 16; break;
case 0x0A: case 0x33: bits[0] = bits[1] = bits[2] = bits[3] = 8; break;
case 0x0F: bits[0] = bits[1] = 16; break;
case 0x12: bits[0] = 32; break;
case 0x13: bits[0] = bits[1] = bits[2] = 8; break;
case 0x18: case 0x32: bits[0] = bits[1] = 8; break;
case 0x1B: bits[0] = 16; break;
case 0x1D: case 0x34: bits[0] = 8; break;
default: break; // packed (0x30, 0x31): below
}
const char k{"xyz"[p.stream % 3]};
const auto element{p.instanced ? std::string{"(input.instance_index + vs_pull.w)"}
: std::string{"u32(i32(input.vertex_index) + bitcast<i32>(rescale_(3u).z))"}};
const auto base{fmt::format("(vs_pull.{} + {} * {}u + {}u)", k, element, p.stride, offset)};
const bool integer{type == 3 || type == 4}, scaled{type == 5 || type == 6};
const bool is_signed{type == 1 || type == 3 || type == 6};
std::string out{fmt::format(" {{\n let o = {};\n", base)};
std::string comps[4];
if (size == 0x30 || size == 0x31) {
out += fmt::format(" let w = vb{}_32(o);\n", p.stream);
if (size == 0x31) {
out += fmt::format(" in_attr{} = vec4<f32>(uf11_(w & 0x7ffu), uf11_((w >> 11u) & 0x7ffu), uf10_(w >> 22u), 1.0);\n }}\n", i);
return out;
}
const char* u{"vec4<u32>(w & 0x3ffu, (w >> 10u) & 0x3ffu, (w >> 20u) & 0x3ffu, w >> 30u)"};
const char* s{"vec4<i32>(i32(w << 22u) >> 22u, i32(w << 12u) >> 22u, i32(w << 2u) >> 22u, i32(w) >> 30u)"};
std::string v;
switch (type) {
case 1: v = fmt::format("max(vec4<f32>({}) / vec4<f32>(511.0, 511.0, 511.0, 1.0), vec4<f32>(-1.0))", s); break;
case 2: v = fmt::format("vec4<f32>({}) / vec4<f32>(1023.0, 1023.0, 1023.0, 3.0)", u); break;
case 3: v = fmt::format("bitcast<vec4<f32>>({})", s); break;
case 4: v = fmt::format("bitcast<vec4<f32>>({})", u); break;
case 5: v = fmt::format("vec4<f32>({})", u); break;
case 6: v = fmt::format("vec4<f32>({})", s); break;
default: v = "vec4<f32>(0.0, 0.0, 0.0, 1.0)"; break;
}
out += fmt::format(" in_attr{} = {};\n }}\n", i, v);
return out;
}
u32 at{0};
for (int c = 0; c < 4; ++c) {
if (!bits[c]) {
comps[c] = integer ? (c == 3 ? "1u" : "0u") : (c == 3 ? "1.0" : "0.0");
continue;
}
const int n{bits[c]};
const auto raw{fmt::format("vb{}_{}(o + {}u)", p.stream, n, at)};
at += n / 8;
const auto sext{n == 32 ? fmt::format("i32({})", raw)
: fmt::format("(i32({} << {}u) >> {}u)", raw, 32 - n, 32 - n)};
if (integer) {
comps[c] = is_signed ? fmt::format("u32({})", sext) : raw;
} else if (scaled) {
comps[c] = is_signed ? fmt::format("f32({})", sext) : fmt::format("f32({})", raw);
} else if (type == 7) { // float
comps[c] = n == 32 ? fmt::format("bitcast<f32>({})", raw)
: n == 16 ? fmt::format("unpack2x16float({}).x", raw)
: fmt::format("f32({})", raw);
} else if (type == 2) { // unorm
comps[c] = fmt::format("(f32({}) / {}.0)", raw, n == 32 ? 4294967295.0 : (double)((1u << n) - 1));
} else { // snorm
comps[c] = fmt::format("max(f32({}) / {}.0, -1.0)", sext, (1u << (n - 1)) - 1);
}
}
if (bgra) std::swap(comps[0], comps[2]);
if (size == 0x32) std::swap(comps[0], comps[1]);
out += integer ? fmt::format(" in_attr{} = bitcast<vec4<f32>>(vec4<u32>({}, {}, {}, {}));\n }}\n", i, comps[0],
comps[1], comps[2], comps[3])
: fmt::format(" in_attr{} = vec4<f32>({}, {}, {}, {});\n }}\n", i, comps[0], comps[1], comps[2],
comps[3]);
return out;
}
/// The WGSL entry point: copies the stage's inputs into the private variables the body reads, runs it, and returns
/// the private variables it wrote as the stage's outputs.
std::string EntryPoint(const EmitContext& ctx, const IR::Program& program, const Options& options) {
const Info& info{ctx.info};
std::string out;
switch (ctx.stage) {
case Stage::VertexA:
case Stage::VertexB: {
out += "struct VsIn {\n @builtin(vertex_index) vertex_index: u32,\n"
" @builtin(instance_index) instance_index: u32,\n";
for (size_t i = 0; i < IR::NUM_GENERICS; ++i) {
if (info.loads.Generic(i) && !options.vertex_pull[i].enabled) {
out += fmt::format(" @location({}) a{}: {},\n", i, i,
options.vertex_unpack[i] != VertexUnpack::None
? std::string_view{"u32"}
: AttributeTypeName(ctx.runtime_info.generic_input_types[i]));
}
}
out += "};\nstruct VsOut {\n @builtin(position) position: vec4<f32>,\n";
for (size_t i = 0; i < IR::NUM_GENERICS; ++i) {
if (info.stores.Generic(i)) {
out += fmt::format(" @location({}) {}o{}: vec4<f32>,\n", i,
InterpolationAttribute(options.output_interpolation[i]), i);
}
}
// unsigned 11/10-bit floats (5-bit exponent, no sign)
out += "};\nfn uf11_(v: u32) -> f32 { let e = v >> 6u; let m = f32(v & 63u);"
" if (e == 0u) { return m * exp2(-20.0); } if (e == 31u) { return 65504.0; }"
" return (1.0 + m / 64.0) * exp2(f32(e) - 15.0); }\n"
"fn uf10_(v: u32) -> f32 { let e = v >> 5u; let m = f32(v & 31u);"
" if (e == 0u) { return m * exp2(-19.0); } if (e == 31u) { return 65504.0; }"
" return (1.0 + m / 32.0) * exp2(f32(e) - 15.0); }\n"
"@vertex\nfn main(input: VsIn) -> VsOut {\n"
" gl_VertexID = i32(input.vertex_index) + bitcast<i32>(rescale_(3u).z);\n"
" gl_InstanceID = i32(input.instance_index) + bitcast<i32>(rescale_(3u).w);\n";
for (size_t i = 0; i < IR::NUM_GENERICS; ++i) {
if (info.loads.Generic(i) && options.vertex_pull[i].enabled) {
out += PulledAttribute(ctx, i, options.vertex_pull[i]);
} else if (info.loads.Generic(i) && options.vertex_unpack[i] != VertexUnpack::None) {
// fields of a packed word: x = bits 0-9, y 10-19, z 20-29, w 30-31 (10/11-bit floats: 0-10,
// 11-21, 22-31)
const auto v{fmt::format("input.a{}", i)};
const auto u{fmt::format("vec4<u32>({0}&0x3ffu,({0}>>10u)&0x3ffu,({0}>>20u)&0x3ffu,{0}>>30u)", v)};
const auto s{fmt::format("vec4<i32>(i32({0}<<22u)>>22u,i32({0}<<12u)>>22u,i32({0}<<2u)>>22u,"
"i32({0})>>30u)", v)};
std::string value;
switch (options.vertex_unpack[i]) {
case VertexUnpack::A2B10G10R10_SNORM:
value = fmt::format("max(vec4<f32>({})/vec4<f32>(511.0,511.0,511.0,1.0),vec4<f32>(-1.0))", s);
break;
case VertexUnpack::A2B10G10R10_UINT:
value = fmt::format("bitcast<vec4<f32>>({})", u);
break;
case VertexUnpack::A2B10G10R10_SINT:
value = fmt::format("bitcast<vec4<f32>>({})", s);
break;
case VertexUnpack::A2B10G10R10_USCALED:
value = fmt::format("vec4<f32>({})", u);
break;
case VertexUnpack::A2B10G10R10_SSCALED:
value = fmt::format("vec4<f32>({})", s);
break;
case VertexUnpack::B10G11R11_FLOAT:
value = fmt::format("vec4<f32>(uf11_({0}&0x7ffu),uf11_(({0}>>11u)&0x7ffu),uf10_({0}>>22u),1.0)", v);
break;
default:
break;
}
out += fmt::format(" in_attr{} = {};\n", i, value);
} else if (info.loads.Generic(i)) {
const AttributeType type{ctx.runtime_info.generic_input_types[i]};
out += IsIntegerAttribute(type) ? fmt::format(" in_attr{} = bitcast<vec4<f32>>(input.a{});\n", i, i)
: type == AttributeType::SignedScaled || type == AttributeType::UnsignedScaled
? fmt::format(" in_attr{} = vec4<f32>(input.a{});\n", i, i)
: fmt::format(" in_attr{} = input.a{};\n", i, i);
}
}
out += " gl_Position = vec4<f32>(0.0f, 0.0f, 0.0f, 1.0f);\n shader_main();\n var output: VsOut;\n"
" let p = gl_Position;\n let t = rescale_(2u);\n let z = rescale_(3u);\n"
" output.position = vec4<f32>(t.x * p.x + t.y * p.w, t.z * p.y + t.w * p.w, z.x * p.z + z.y * p.w,"
" p.w);\n"
// a 16-bit depth target: the depth on its step, as pixel shaders write it (d16q_); vertices outside
// the depth range keep theirs (the visible part of the primitive keeps its slope)
" if (render_area().z != 0.0f && p.w > 0.0f) {\n"
" let d = output.position.z / p.w;\n"
" if (d >= 0.0f && d <= 1.0f) { output.position.z = d16q_(d) * p.w; }\n"
" }\n";
// RX_DEBUG_VS_NONFINITE=1 (debugging): a position with an infinite or NaN component goes to the screen's
// centre; =2: all of it NaN (a bit test: compilers may fold x != x)
if (static const char* nonfinite = std::getenv("RX_DEBUG_VS_NONFINITE"); nonfinite)
out += fmt::format(" if (any((bitcast<vec4<u32>>(p) & vec4<u32>(0x7f800000u)) == vec4<u32>(0x7f800000u))) {{"
" output.position = {}; }}\n",
*nonfinite == '2' ? "bitcast<vec4<f32>>(vec4<u32>(0x7fc00000u))"
: "vec4<f32>(0.0f, 0.0f, 0.5f, 1.0f)");
for (size_t i = 0; i < IR::NUM_GENERICS; ++i) {
if (info.stores.Generic(i)) {
out += fmt::format(" output.o{} = out_attr{};\n", i, i);
}
}
out += " return output;\n}\n";
if (ctx.vs_compute) { // (vs_as_compute_next: the same body as a compute entry, outputs dropped)
const auto swap = [&out](std::string_view from, std::string_view to) {
const size_t at{out.find(from)};
if (at == std::string::npos) throw NotImplementedException("vertex program as compute: {}", from);
out.replace(at, from.size(), to);
};
if (out.find("@location(", 0) < out.find("};\nstruct VsOut")) {
throw NotImplementedException("vertex program as compute: attributes not pulled");
}
swap("struct VsIn {\n @builtin(vertex_index) vertex_index: u32,\n"
" @builtin(instance_index) instance_index: u32,\n",
"struct VsIn {\n vertex_index: u32,\n instance_index: u32,\n");
swap("@vertex\nfn main(input: VsIn) -> VsOut {\n",
"@compute @workgroup_size(1)\nfn main(@builtin(global_invocation_id) gid_: vec3<u32>) {\n"
" let input = VsIn(gid_.x, gid_.y);\n");
swap(" return output;\n}\n", "}\n");
}
break;
}
case Stage::Fragment: {
out += "struct FsIn {\n @builtin(position) frag_coord: vec4<f32>,\n"
" @builtin(front_facing) front_facing: bool,\n";
if (info.uses_sample_id) {
out += " @builtin(sample_index) sample_index: u32,\n";
}
for (size_t i = 0; i < IR::NUM_GENERICS; ++i) {
if (info.loads.Generic(i) && ctx.runtime_info.previous_stage_stores.Generic(i)) {
out += fmt::format(" @location({}) {}a{}: vec4<f32>,\n", i,
InterpolationAttribute(info.interpolation[i]), i);
}
}
out += "};\n";
bool any_output{info.stores_frag_depth};
std::string outs{"struct FsOut {\n"};
for (size_t i = 0; i < info.stores_frag_color.size(); ++i) {
if (info.stores_frag_color[i]) {
outs += fmt::format(" @location({}) c{}: {},\n", i, i,
AttributeTypeName(ctx.runtime_info.color_output_types[i]));
any_output = true;
}
}
if (info.stores_frag_depth) {
outs += " @builtin(frag_depth) depth: f32,\n";
}
outs += "};\n";
if (any_output) {
out += outs;
}
// (force_early_z: depth and stencil tests before the shader runs, so the fragments they reject make no stores:
// the GPU's early-Z for a program that asks for it)
const char* early = "";
if (ctx.runtime_info.force_early_z && !info.stores_frag_depth) {
static const char* attr = getenv("RX_EARLY_Z_ATTR") ? getenv("RX_EARLY_Z_ATTR") : "@early_depth_test(force)";
early = attr;
}
out += fmt::format("@fragment {}\nfn main(input: FsIn){} {{\n", early, any_output ? " -> FsOut" : "");
out += " gl_FragCoord = input.frag_coord;\n gl_FrontFacing = input.front_facing;\n";
if (info.uses_sample_id) {
out += " gl_SampleID = i32(input.sample_index);\n";
}
for (size_t i = 0; i < IR::NUM_GENERICS; ++i) {
if (info.loads.Generic(i) && ctx.runtime_info.previous_stage_stores.Generic(i)) {
out += fmt::format(" in_attr{} = input.a{};\n", i, i);
}
}
out += " gl_FragDepth = input.frag_coord.z;\n shader_main();\n";
if (any_output) {
out += " var output: FsOut;\n";
// render_area().w: bit i = colour target i has a UNORM format kept as float (16-bit UNORM targets
// cannot be render attachments in WebGPU), bit 8 + i the same for SNORM: clamped as the hardware's
// store would
out += " let ntc_ = u32(render_area().w);\n";
for (size_t i = 0; i < info.stores_frag_color.size(); ++i) {
if (info.stores_frag_color[i]) {
const auto type{ctx.runtime_info.color_output_types[i]};
out += IsIntegerAttribute(type)
? fmt::format(" output.c{} = bitcast<{}>(frag_color{});\n", i, AttributeTypeName(type), i)
: fmt::format(" output.c{0} = frag_color{0};\n"
" if ((ntc_ & {1}u) != 0u) {{ output.c{0} = clamp(output.c{0}, vec4<f32>(0.0), vec4<f32>(1.0)); }}\n"
" if ((ntc_ & {2}u) != 0u) {{ output.c{0} = clamp(output.c{0}, vec4<f32>(-1.0), vec4<f32>(1.0)); }}\n",
i, 1u << i, 0x100u << i);
}
}
if (info.stores_frag_depth) {
out += " output.depth = select(gl_FragDepth, d16q_(gl_FragDepth), render_area().z != 0.0f);\n";
}
out += " return output;\n";
}
out += "}\n";
break;
}
case Stage::Compute:
out += fmt::format("@compute @workgroup_size({}, {}, {})\n"
"fn main(@builtin(local_invocation_id) lid: vec3<u32>, "
"@builtin(workgroup_id) wid: vec3<u32>) {{\n"
" gl_LocalInvocationID = lid;\n gl_WorkGroupID = wid;\n shader_main();\n}}\n",
(std::max)(program.workgroup_size[0], 1u), (std::max)(program.workgroup_size[1], 1u),
(std::max)(program.workgroup_size[2], 1u));
break;
default:
throw NotImplementedException("WebGPU has no {} shaders", ctx.stage);
}
return out;
}
} // Anonymous namespace
std::string EmitWGSL(const Profile& profile, const RuntimeInfo& runtime_info, IR::Program& program,
Bindings& bindings, const Options& options) {
EmitContext ctx{program, bindings, profile, runtime_info};
u32 pulled_streams{0};
for (const auto& p : options.vertex_pull) {
if (p.enabled) {
pulled_streams = (std::max)(pulled_streams, static_cast<u32>(p.stream) + 1);
}
}
if (pulled_streams > 0 && (program.stage == Stage::VertexA || program.stage == Stage::VertexB)) {
ctx.header += fmt::format("@group({}) @binding({}) var<uniform> vs_pull: vec4<u32>;\n", ctx.group,
bindings.unified++);
for (u32 k = 0; k < pulled_streams; ++k) {
ctx.header += fmt::format(
"@group({0}) @binding({1}) var<storage, read> vs_vb{2}: array<u32>;\n"
"fn vb{2}_8(o: u32) -> u32 {{ return (vs_vb{2}[o >> 2u] >> ((o & 3u) * 8u)) & 0xffu; }}\n"
"fn vb{2}_16(o: u32) -> u32 {{ return vb{2}_8(o) | (vb{2}_8(o + 1u) << 8u); }}\n"
"fn vb{2}_32(o: u32) -> u32 {{ return vb{2}_16(o) | (vb{2}_16(o + 2u) << 16u); }}\n",
ctx.group, bindings.unified++, k);
}
}
Precolor(program);
EmitCode(ctx, program);
if (program.shared_memory_size > 0) {
ctx.header += fmt::format("var<workgroup> smem: array<{}, {}>;\n", ctx.atomic_smem ? "atomic<u32>" : "u32",
Common::DivCeil(program.shared_memory_size, 4U));
}
std::string body{"fn shader_main() {\n"};
if (program.local_memory_size > 0) {
body += fmt::format("var lmem: array<u32, {}>;\n", Common::DivCeil(program.local_memory_size, 4U));
}
DefineVariables(ctx, body);
if (ctx.uses_cc_carry) {
body += "var carry: u32;\n";
}
if (program.info.uses_subgroup_shuffles) {
body += "var shfl_in_bounds: bool;\nvar shfl_result: u32;\n";
}
return ctx.header + body + ctx.code + "}\n" + EntryPoint(ctx, program, options);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,76 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
//
// WGSL backend (Rexcomp): one stage of a WebGPU pipeline as a WGSL module with a `main` entry point.
#pragma once
#include <array>
#include <string>
#include "shader_recompiler/backend/bindings.h"
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/profile.h"
#include "shader_recompiler/runtime_info.h"
#include "shader_recompiler/shader_info.h"
namespace Shader::Backend::WGSL {
/// Vertex attributes in formats WebGPU's vertex fetch lacks: read as one u32 and unpacked in the vertex stage.
enum class VertexUnpack : u8 {
None,
A2B10G10R10_SNORM,
A2B10G10R10_UINT, ///< (the generic is an unsigned integer)
A2B10G10R10_SINT, ///< (signed integer)
A2B10G10R10_USCALED, ///< (floats of the unsigned values)
A2B10G10R10_SSCALED, ///< (floats of the signed values)
B10G11R11_FLOAT,
};
/// What one stage needs to know about its neighbours in the pipeline.
struct Options {
/// Vertex stages: the interpolation the fragment stage declares for each generic (WebGPU requires both sides
/// of a location to agree).
std::array<Interpolation, 32> output_interpolation{};
/// Vertex stages: attributes fetched packed (see VertexUnpack).
std::array<VertexUnpack, 32> vertex_unpack{};
/// Vertex stages: attributes the vertex stage reads from memory itself ("vertex pulling": layouts WebGPU's vertex
/// fetch cannot describe, like strides that are not multiples of 4). Each names one of up to 3 pulled streams,
/// bound after every other resource as `var<storage, read> vs_vb<k>: array<u32>`, preceded by
/// `var<uniform> vs_pull: vec4<u32>` (x, y, z: where stream 0, 1, 2 starts in its binding, in bytes; w: added to
/// the instance index of instanced streams).
struct VertexPull {
bool enabled{};
u8 stream{};
bool instanced{};
u32 attribute{}; ///< the Maxwell attribute format word (offset, size, type, bgra)
u32 stride{};
};
std::array<VertexPull, 32> vertex_pull{};
};
/// Storage buffers bound to one same range (the browser forbids two writable bindings of one range): for the next
/// EmitWGSL on this thread, ssbo_alias_next[j] = i + 1 makes descriptor j use descriptor i's binding (j is not
/// declared). Reset by the emission.
extern thread_local std::array<u8, 64> ssbo_alias_next;
/// ...and where descriptor j's data starts in descriptor i's binding, in words (overlapping ranges merged into one)
extern thread_local std::array<u32, 64> ssbo_alias_words_next;
/// For the next EmitWGSL of a vertex program on this thread: emit it as a compute program instead (one invocation per
/// vertex: x the vertex index, y the instance), with its storage writes, which WebGPU's vertex stage cannot make
/// (TOTK's Ultrahand reads the object under the reticle in a one-point draw whose vertex program stores it). Its outputs
/// are dropped. Only for vertex programs whose attributes are all pulled (vertex_pull) or none. Reset by the emission.
extern thread_local bool vs_as_compute_next;
/// (mkw-recomp) The storage texture format of each image descriptor (WGSL's name: "rgba16float"...) for the next
/// EmitWGSL on this thread: a formatted image store or load takes its format from the bound image, which WGSL must
/// declare. Empty: the descriptor's own (typeless: r32uint). Reset by the emission.
extern thread_local std::array<std::string, 16> image_formats_next;
[[nodiscard]] std::string EmitWGSL(const Profile& profile, const RuntimeInfo& runtime_info,
IR::Program& program, Bindings& bindings, const Options& options = {});
[[nodiscard]] inline std::string EmitWGSL(const Profile& profile, IR::Program& program) {
Bindings binding;
return EmitWGSL(profile, {}, program, binding);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,449 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
namespace {
// WGSL atomics on atomic<u32> memory (EmitContext::atomic_ssbo / atomic_smem). The result is assigned after the
// operation, as it may share a variable with an operand.
void Native(EmitContext& ctx, IR::Inst& inst, std::string_view ptr, std::string_view function,
std::string_view value) {
const std::string op{fmt::format("{}({},{})", function, ptr, value)};
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32)};
ctx.Add("{}={};", ret, op);
}
/// An operation WGSL has no atomic for: a compare-exchange loop over function(old, value) -> new.
void CasLoop(EmitContext& ctx, IR::Inst& inst, std::string_view ptr, std::string_view function,
std::string_view value) {
const std::string p{ptr}, v{value};
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32)};
ctx.Add("loop {{let old_=atomicLoad({});let r_=atomicCompareExchangeWeak({},old_,{}(old_,{}));"
"if (r_.exchanged) {{{}=old_;break;}}}}",
p, p, function, v, ret);
}
std::string Smem(EmitContext&, std::string_view offset) {
return fmt::format("&smem[{}>>2]", offset);
}
std::string Ssbo(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) {
return ctx.SsboPtr(binding.U32(), fmt::format("{}>>2", ctx.var_alloc.Consume(offset)));
}
} // Anonymous namespace
void EmitSharedAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
Native(ctx, inst, Smem(ctx, pointer_offset), "atomicAdd", value);
}
void EmitSharedAtomicSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
CasLoop(ctx, inst, Smem(ctx, pointer_offset), "CasMinS32", fmt::format("u32({})", value));
}
void EmitSharedAtomicUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
Native(ctx, inst, Smem(ctx, pointer_offset), "atomicMin", value);
}
void EmitSharedAtomicSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
CasLoop(ctx, inst, Smem(ctx, pointer_offset), "CasMaxS32", fmt::format("u32({})", value));
}
void EmitSharedAtomicUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
Native(ctx, inst, Smem(ctx, pointer_offset), "atomicMax", value);
}
void EmitSharedAtomicInc32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
CasLoop(ctx, inst, Smem(ctx, pointer_offset), "CasIncrement", value);
}
void EmitSharedAtomicDec32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
CasLoop(ctx, inst, Smem(ctx, pointer_offset), "CasDecrement", value);
}
void EmitSharedAtomicAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
Native(ctx, inst, Smem(ctx, pointer_offset), "atomicAnd", value);
}
void EmitSharedAtomicOr32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
Native(ctx, inst, Smem(ctx, pointer_offset), "atomicOr", value);
}
void EmitSharedAtomicXor32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
Native(ctx, inst, Smem(ctx, pointer_offset), "atomicXor", value);
}
void EmitSharedAtomicExchange32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
Native(ctx, inst, Smem(ctx, pointer_offset), "atomicExchange", value);
}
void EmitSharedAtomicExchange64(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
NotImplemented();
}
void EmitSharedAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
Native(ctx, inst, Ssbo(ctx, binding, offset), "atomicAdd", value);
}
void EmitStorageAtomicSMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
CasLoop(ctx, inst, Ssbo(ctx, binding, offset), "CasMinS32", fmt::format("u32({})", value));
}
void EmitStorageAtomicUMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
Native(ctx, inst, Ssbo(ctx, binding, offset), "atomicMin", value);
}
void EmitStorageAtomicSMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
CasLoop(ctx, inst, Ssbo(ctx, binding, offset), "CasMaxS32", fmt::format("u32({})", value));
}
void EmitStorageAtomicUMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
Native(ctx, inst, Ssbo(ctx, binding, offset), "atomicMax", value);
}
void EmitStorageAtomicInc32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
CasLoop(ctx, inst, Ssbo(ctx, binding, offset), "CasIncrement", value);
}
void EmitStorageAtomicDec32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
CasLoop(ctx, inst, Ssbo(ctx, binding, offset), "CasDecrement", value);
}
void EmitStorageAtomicAnd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
Native(ctx, inst, Ssbo(ctx, binding, offset), "atomicAnd", value);
}
void EmitStorageAtomicOr32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
Native(ctx, inst, Ssbo(ctx, binding, offset), "atomicOr", value);
}
void EmitStorageAtomicXor32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
Native(ctx, inst, Ssbo(ctx, binding, offset), "atomicXor", value);
}
void EmitStorageAtomicExchange32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
Native(ctx, inst, Ssbo(ctx, binding, offset), "atomicExchange", value);
}
void EmitStorageAtomicIAdd64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicSMin64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicUMin64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicSMax64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicUMax64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicAnd64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicOr64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicXor64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicExchange64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicIAdd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicSMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicUMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicSMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicUMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicAnd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicOr32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicXor32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicAddF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
const std::string ptr{Ssbo(ctx, binding, offset)}, v{value};
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32)};
ctx.Add("loop {{let old_=atomicLoad({});let r_=atomicCompareExchangeWeak({},old_,CasFloatAdd(old_,{}));"
"if (r_.exchanged) {{{}=old_;break;}}}}",
ptr, ptr, v, ret);
ctx.AddF32("{}=bitcast<f32>({});", inst, ret);
}
void EmitStorageAtomicAddF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
CasLoop(ctx, inst, Ssbo(ctx, binding, offset), "CasFloatAdd16x2", value);
}
void EmitStorageAtomicAddF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicMinF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
CasLoop(ctx, inst, Ssbo(ctx, binding, offset), "CasFloatMin16x2", value);
}
void EmitStorageAtomicMinF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitStorageAtomicMaxF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
CasLoop(ctx, inst, Ssbo(ctx, binding, offset), "CasFloatMax16x2", value);
}
void EmitStorageAtomicMaxF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value) {
NotImplemented();
}
void EmitGlobalAtomicIAdd32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicSMin32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicUMin32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicSMax32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicUMax32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicInc32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicDec32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAnd32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicOr32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicXor32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicExchange32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicIAdd64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicSMin64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicUMin64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicSMax64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicUMax64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicInc64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicDec64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicAnd64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicOr64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicXor64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicExchange64(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicIAdd32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicSMin32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicUMin32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicSMax32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicUMax32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicInc32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicDec32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicAnd32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicOr32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicXor32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicExchange32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicAddF32(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAddF16x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicAddF32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicMinF16x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicMinF32x2(EmitContext&) {
NotImplemented();
}
void EmitGlobalAtomicMaxF16x2(EmitContext&) {
throw NotImplementedException("GLSL Instruction");
}
void EmitGlobalAtomicMaxF32x2(EmitContext&) {
NotImplemented();
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,19 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
namespace Shader::Backend::WGSL {
void EmitBarrier(EmitContext& ctx) {
ctx.Add("workgroupBarrier();");
}
void EmitWorkgroupMemoryBarrier(EmitContext& ctx) {
ctx.Add("workgroupBarrier();");
}
void EmitDeviceMemoryBarrier(EmitContext& ctx) {
ctx.Add("storageBarrier();");
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,95 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::WGSL {
namespace {
void Alias(IR::Inst& inst, const IR::Value& value) {
if (value.IsImmediate()) {
return;
}
IR::Inst& value_inst{*value.InstRecursive()};
value_inst.DestructiveAddUsage(inst.UseCount());
value_inst.DestructiveRemoveUsage();
inst.SetDefinition(value_inst.Definition<Id>());
}
} // Anonymous namespace
void EmitIdentity(EmitContext&, IR::Inst& inst, const IR::Value& value) {
Alias(inst, value);
}
void EmitConditionRef(EmitContext& ctx, IR::Inst& inst, const IR::Value& value) {
// Fake one usage to get a real variable out of the condition
inst.DestructiveAddUsage(1);
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U1)};
const auto input{ctx.var_alloc.Consume(value)};
const auto suffix{ctx.profile.has_gl_bool_ref_bug ? "?true:false" : ""};
if (ret != input) {
ctx.Add("{}={}{};", ret, input, suffix);
}
}
void EmitBitCastU16F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst) {
NotImplemented();
}
void EmitBitCastU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=bitcast<u32>({});", inst, value);
}
void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
void EmitBitCastF16U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst) {
NotImplemented();
}
void EmitBitCastF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=bitcast<f32>({});", inst, value);
}
void EmitBitCastF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
void EmitPackUint2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}={};", inst, value); // u64 is vec2<u32>
}
void EmitUnpackUint2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32x2("{}={};", inst, value);
}
void EmitPackFloat2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=pack2x16float({});", inst, value); // f16x2 is vec2<f32>
}
void EmitUnpackFloat2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF16x2("{}=unpack2x16float({});", inst, value);
}
void EmitPackHalf2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=pack2x16float({});", inst, value);
}
void EmitUnpackHalf2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32x2("{}=unpack2x16float({});", inst, value);
}
void EmitPackDouble2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
void EmitUnpackDouble2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,218 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
namespace {
constexpr std::string_view SWIZZLE{"xyzw"};
void CompositeInsert(EmitContext& ctx, std::string_view result, std::string_view composite,
std::string_view object, u32 index) {
if (result == composite) {
// The result is aliased with the composite
ctx.Add("{}.{}={};", composite, SWIZZLE[index], object);
} else {
ctx.Add("{}={};{}.{}={};", result, composite, result, SWIZZLE[index], object);
}
}
} // Anonymous namespace
void EmitCompositeConstructU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2) {
ctx.AddU32x2("{}=vec2<u32>({},{});", inst, e1, e2);
}
void EmitCompositeConstructU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3) {
ctx.AddU32x3("{}=vec3<u32>({},{},{});", inst, e1, e2, e3);
}
void EmitCompositeConstructU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3, std::string_view e4) {
ctx.AddU32x4("{}=vec4<u32>({},{},{},{});", inst, e1, e2, e3, e4);
}
void EmitCompositeExtractU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddU32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeExtractU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddU32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeExtractU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddU32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeInsertU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32x2)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeInsertU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32x3)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeInsertU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32x4)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeConstructF16x2([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view e1,
[[maybe_unused]] std::string_view e2) {
NotImplemented();
}
void EmitCompositeConstructF16x3([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view e1,
[[maybe_unused]] std::string_view e2,
[[maybe_unused]] std::string_view e3) {
NotImplemented();
}
void EmitCompositeConstructF16x4([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view e1,
[[maybe_unused]] std::string_view e2,
[[maybe_unused]] std::string_view e3,
[[maybe_unused]] std::string_view e4) {
NotImplemented();
}
void EmitCompositeExtractF16x2([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeExtractF16x3([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeExtractF16x4([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeInsertF16x2([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] std::string_view object,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeInsertF16x3([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] std::string_view object,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeInsertF16x4([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view composite,
[[maybe_unused]] std::string_view object,
[[maybe_unused]] u32 index) {
NotImplemented();
}
void EmitCompositeConstructF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2) {
ctx.AddF32x2("{}=vec2<f32>({},{});", inst, e1, e2);
}
void EmitCompositeConstructF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3) {
ctx.AddF32x3("{}=vec3<f32>({},{},{});", inst, e1, e2, e3);
}
void EmitCompositeConstructF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3, std::string_view e4) {
ctx.AddF32x4("{}=vec4<f32>({},{},{},{});", inst, e1, e2, e3, e4);
}
void EmitCompositeExtractF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddF32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeExtractF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddF32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeExtractF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index) {
ctx.AddF32("{}={}.{};", inst, composite, SWIZZLE[index]);
}
void EmitCompositeInsertF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::F32x2)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeInsertF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::F32x3)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeInsertF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::F32x4)};
CompositeInsert(ctx, ret, composite, object, index);
}
void EmitCompositeConstructF64x2([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeConstructF64x3([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeConstructF64x4([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeExtractF64x2([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeExtractF64x3([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeExtractF64x4([[maybe_unused]] EmitContext& ctx) {
NotImplemented();
}
void EmitCompositeInsertF64x2(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index) {
ctx.Add("{}.{}={};", composite, SWIZZLE[index], object);
}
void EmitCompositeInsertF64x3(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index) {
ctx.Add("{}.{}={};", composite, SWIZZLE[index], object);
}
void EmitCompositeInsertF64x4(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index) {
ctx.Add("{}.{}={};", composite, SWIZZLE[index], object);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,479 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
#include "shader_recompiler/runtime_info.h"
namespace Shader::Backend::WGSL {
namespace {
constexpr char SWIZZLE[]{"xyzw"};
u32 CbufIndex(u32 offset) {
return (offset / 4) % 4;
}
char OffsetSwizzle(u32 offset) {
return SWIZZLE[CbufIndex(offset)];
}
bool IsInputArray(Stage stage) {
return stage == Stage::Geometry || stage == Stage::TessellationControl ||
stage == Stage::TessellationEval;
}
std::string InputVertexIndex(EmitContext& ctx, std::string_view vertex) {
return IsInputArray(ctx.stage) ? fmt::format("[{}]", vertex) : "";
}
std::string_view OutputVertexIndex(EmitContext& ctx) {
return ctx.stage == Stage::TessellationControl ? "[gl_InvocationID]" : "";
}
/// dynamic: the index is computed at run time (a storage-bound buffer then reads 0 past its bound size: CbufAt)
std::string ChooseCbuf(EmitContext& ctx, const IR::Value& binding, std::string_view index, bool dynamic = false) {
if (binding.IsImmediate()) {
if (ctx.cbuf_in_ub[binding.U32() % 18]) {
return fmt::format("{}[{}u+({})]", ctx.ub_name, ctx.ub_cbuf_base[binding.U32() % 18], index);
}
if (dynamic && !CbufClamp()) return fmt::format("{}_cbuf{}_at(u32({}))", ctx.stage_name, binding.U32(), index);
return fmt::format("{}_cbuf{}[{}]", ctx.stage_name, binding.U32(), index);
} else {
const auto binding_var{ctx.var_alloc.Consume(binding)};
return fmt::format("GetCbufIndirect({},{})", binding_var, index);
}
}
void GetCbuf(EmitContext& ctx, std::string_view ret, const IR::Value& binding,
const IR::Value& offset, u32 num_bits, std::string_view cast = {},
std::string_view bit_offset = {}) {
const bool is_immediate{offset.IsImmediate()};
const bool component_indexing_bug{!is_immediate && ctx.profile.has_gl_component_indexing_bug};
if (is_immediate) {
const s32 signed_offset{static_cast<s32>(offset.U32())};
static constexpr u32 cbuf_size{0x10000};
if (signed_offset < 0 || offset.U32() > cbuf_size) {
LOG_WARNING(Shader_WGSL, "Immediate constant buffer offset is out of bounds");
ctx.Add("{}={};", ret, cast == "bitcast<f32>" ? "0.0f" : "0u");
return;
}
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto index{is_immediate ? fmt::format("{}", offset.U32() / 16)
: fmt::format("{}>>4", offset_var)};
const auto swizzle{is_immediate ? fmt::format(".{}", OffsetSwizzle(offset.U32()))
: fmt::format("[({}>>2)%4]", offset_var)};
const auto cbuf{ChooseCbuf(ctx, binding, index, !is_immediate)};
const auto cbuf_cast{fmt::format("{}({}{{}})", cast, cbuf)};
const auto extraction{num_bits == 32 ? cbuf_cast
: fmt::format("u32(extractBits({},u32({}),{}u))", cbuf_cast,
bit_offset, num_bits)};
if (!component_indexing_bug) {
const auto result{fmt::format(fmt::runtime(extraction), swizzle)};
ctx.Add("{}={};", ret, result);
return;
}
const auto cbuf_offset{fmt::format("{}>>2", offset_var)};
for (u32 i = 0; i < 4; ++i) {
const auto swizzle_string{fmt::format(".{}", "xyzw"[i])};
const auto result{fmt::format(fmt::runtime(extraction), swizzle_string)};
ctx.Add("if(({}&3)=={}){}={};", cbuf_offset, i, ret, result);
}
}
void GetCbuf8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, const IR::Value& offset,
std::string_view cast) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32)};
if (offset.IsImmediate()) {
const auto bit_offset{fmt::format("{}", (offset.U32() % 4) * 8)};
GetCbuf(ctx, ret, binding, offset, 8, cast, bit_offset);
} else {
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("({}%4)*8", offset_var)};
GetCbuf(ctx, ret, binding, offset, 8, cast, bit_offset);
}
}
void GetCbuf16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, const IR::Value& offset,
std::string_view cast) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32)};
if (offset.IsImmediate()) {
const auto bit_offset{fmt::format("{}", ((offset.U32() / 2) % 2) * 16)};
GetCbuf(ctx, ret, binding, offset, 16, cast, bit_offset);
} else {
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("(({}>>1)%2)*16", offset_var)};
GetCbuf(ctx, ret, binding, offset, 16, cast, bit_offset);
}
}
} // Anonymous namespace
void EmitGetCbufU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{""};
GetCbuf8(ctx, inst, binding, offset, cast);
}
void EmitGetCbufS8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{"bitcast<i32>"};
GetCbuf8(ctx, inst, binding, offset, cast);
}
void EmitGetCbufU16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{""};
GetCbuf16(ctx, inst, binding, offset, cast);
}
void EmitGetCbufS16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{"bitcast<i32>"};
GetCbuf16(ctx, inst, binding, offset, cast);
}
void EmitGetCbufU32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32)};
GetCbuf(ctx, ret, binding, offset, 32, "");
}
void EmitGetCbufF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::F32)};
GetCbuf(ctx, ret, binding, offset, 32, "bitcast<f32>");
}
void EmitGetCbufU32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto cast{""};
if (offset.IsImmediate()) {
static constexpr u32 cbuf_size{0x10000};
const u32 u32_offset{offset.U32()};
const s32 signed_offset{static_cast<s32>(offset.U32())};
if (signed_offset < 0 || u32_offset > cbuf_size) {
LOG_WARNING(Shader_WGSL, "Immediate constant buffer offset is out of bounds");
ctx.AddU32x2("{}=vec2<u32>(0u);", inst);
return;
}
const auto first{ChooseCbuf(ctx, binding, fmt::format("{}", u32_offset / 16))};
if (u32_offset % 16 <= 8 && u32_offset % 4 == 0) {
ctx.AddU32x2("{}={}({}.{}{});", inst, cast, first, OffsetSwizzle(u32_offset),
OffsetSwizzle(u32_offset + 4));
} else {
const auto second{ChooseCbuf(ctx, binding, fmt::format("{}", (u32_offset + 4) / 16))};
ctx.AddU32x2("{}=vec2<u32>({}({}.{}),{}({}.{}));", inst, cast, first, OffsetSwizzle(u32_offset), cast,
second, OffsetSwizzle(u32_offset + 4));
}
return;
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto cbuf{ChooseCbuf(ctx, binding, fmt::format("{}>>4", offset_var), true)};
if (!ctx.profile.has_gl_component_indexing_bug) {
ctx.AddU32x2("{}=vec2<u32>({}({}[({}>>2)%4]),{}({}[(({}+4)>>2)%4]));", inst, cast, cbuf,
offset_var, cast, cbuf, offset_var);
return;
}
const auto ret{ctx.var_alloc.Define(inst, WgslVarType::U32x2)};
const auto cbuf_offset{fmt::format("{}>>2", offset_var)};
for (u32 swizzle = 0; swizzle < 4; ++swizzle) {
ctx.Add("if(({}&3)=={}){}=vec2<u32>({}({}.{}),{}({}.{}));", cbuf_offset, swizzle, ret, cast,
cbuf, "xyzw"[swizzle], cast, cbuf, "xyzw"[(swizzle + 1) % 4]);
}
}
void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr,
std::string_view vertex) {
const u32 element{static_cast<u32>(attr) % 4};
const char swizzle{"xyzw"[element]};
if (IR::IsGeneric(attr)) {
const u32 index{IR::GenericAttributeIndex(attr)};
if (!ctx.runtime_info.previous_stage_stores.Generic(index, element)) {
if (element == 3) {
ctx.AddF32("{}=1.f;", inst, attr);
} else {
ctx.AddF32("{}=0.f;", inst, attr);
}
return;
}
ctx.AddF32("{}=in_attr{}{}.{};", inst, index, InputVertexIndex(ctx, vertex), swizzle);
return;
}
switch (attr) {
case IR::Attribute::PrimitiveId:
ctx.AddF32("{}=bitcast<f32>(gl_PrimitiveID);", inst);
break;
case IR::Attribute::Layer:
ctx.AddF32("{}=bitcast<f32>(gl_Layer);", inst);
break;
case IR::Attribute::PositionX:
case IR::Attribute::PositionY:
case IR::Attribute::PositionZ:
case IR::Attribute::PositionW: {
const bool is_array{IsInputArray(ctx.stage)};
const auto input_decorator{is_array ? fmt::format("gl_in[{}].", vertex) : ""};
ctx.AddF32("{}={}{}.{};", inst, input_decorator, ctx.position_name, swizzle);
break;
}
case IR::Attribute::PointSpriteS:
case IR::Attribute::PointSpriteT:
ctx.AddF32("{}=gl_PointCoord.{};", inst, swizzle);
break;
case IR::Attribute::TessellationEvaluationPointU:
case IR::Attribute::TessellationEvaluationPointV:
ctx.AddF32("{}=gl_TessCoord.{};", inst, swizzle);
break;
case IR::Attribute::InstanceId:
ctx.AddF32("{}=bitcast<f32>(gl_InstanceID);", inst);
break;
case IR::Attribute::VertexId:
ctx.AddF32("{}=bitcast<f32>(gl_VertexID);", inst);
break;
case IR::Attribute::FrontFace:
ctx.AddF32("{}=bitcast<f32>(select(0i,-1i,gl_FrontFacing));", inst);
break;
case IR::Attribute::BaseInstance:
ctx.AddF32("{}=bitcast<f32>(gl_BaseInstance);", inst);
break;
case IR::Attribute::BaseVertex:
ctx.AddF32("{}=bitcast<f32>(gl_BaseVertex);", inst);
break;
case IR::Attribute::DrawID:
ctx.AddF32("{}=bitcast<f32>(gl_DrawID);", inst);
break;
default:
throw NotImplementedException("Get attribute {}", attr);
}
}
void EmitGetAttributeU32(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, std::string_view) {
switch (attr) {
case IR::Attribute::PrimitiveId:
ctx.AddU32("{}=u32(gl_PrimitiveID);", inst);
break;
case IR::Attribute::InstanceId:
ctx.AddU32("{}=u32(gl_InstanceID);", inst);
break;
case IR::Attribute::VertexId:
ctx.AddU32("{}=u32(gl_VertexID);", inst);
break;
case IR::Attribute::BaseInstance:
ctx.AddU32("{}=u32(gl_BaseInstance);", inst);
break;
case IR::Attribute::BaseVertex:
ctx.AddU32("{}=u32(gl_BaseVertex);", inst);
break;
case IR::Attribute::DrawID:
ctx.AddU32("{}=u32(gl_DrawID);", inst);
break;
default:
throw NotImplementedException("Get U32 attribute {}", attr);
}
}
void EmitSetAttribute(EmitContext& ctx, IR::Attribute attr, std::string_view value,
[[maybe_unused]] std::string_view vertex) {
if (IR::IsGeneric(attr)) {
const u32 index{IR::GenericAttributeIndex(attr)};
const u32 attr_element{IR::GenericAttributeElement(attr)};
const GenericElementInfo& info{ctx.output_generics.at(index).at(attr_element)};
const auto output_decorator{OutputVertexIndex(ctx)};
if (info.num_components == 1) {
ctx.Add("{}{}={};", info.name, output_decorator, value);
} else {
const u32 index_element{attr_element - info.first_element};
ctx.Add("{}{}.{}={};", info.name, output_decorator, "xyzw"[index_element], value);
}
return;
}
const u32 element{static_cast<u32>(attr) % 4};
const char swizzle{"xyzw"[element]};
switch (attr) {
case IR::Attribute::Layer:
if (ctx.stage != Stage::Geometry &&
!ctx.profile.support_viewport_index_layer_non_geometry) {
LOG_WARNING(Shader_WGSL, "Shader stores viewport layer but device does not support "
"viewport layer extension");
break;
}
ctx.Add("gl_Layer=bitcast<i32>({});", value);
break;
case IR::Attribute::ViewportIndex:
if (ctx.stage != Stage::Geometry &&
!ctx.profile.support_viewport_index_layer_non_geometry) {
LOG_WARNING(Shader_WGSL, "Shader stores viewport index but device does not support "
"viewport layer extension");
break;
}
ctx.Add("gl_ViewportIndex=bitcast<i32>({});", value);
break;
case IR::Attribute::ViewportMask:
if (ctx.stage != Stage::Geometry && !ctx.profile.support_viewport_mask) {
LOG_WARNING(
Shader_WGSL,
"Shader stores viewport mask but device does not support viewport mask extension");
break;
}
ctx.Add("gl_ViewportMask[0]=bitcast<i32>({});", value);
break;
case IR::Attribute::PointSize:
ctx.Add("gl_PointSize={};", value);
break;
case IR::Attribute::PositionX:
case IR::Attribute::PositionY:
case IR::Attribute::PositionZ:
case IR::Attribute::PositionW:
ctx.Add("gl_Position.{}={};", swizzle, value);
break;
case IR::Attribute::ClipDistance0:
case IR::Attribute::ClipDistance1:
case IR::Attribute::ClipDistance2:
case IR::Attribute::ClipDistance3:
case IR::Attribute::ClipDistance4:
case IR::Attribute::ClipDistance5:
case IR::Attribute::ClipDistance6:
case IR::Attribute::ClipDistance7: {
const u32 index{static_cast<u32>(attr) - static_cast<u32>(IR::Attribute::ClipDistance0)};
ctx.Add("gl_ClipDistance[{}]={};", index, value);
break;
}
default:
throw NotImplementedException("Set attribute {}", attr);
}
}
void EmitGetAttributeIndexed(EmitContext& ctx, IR::Inst& inst, std::string_view offset,
std::string_view vertex) {
const bool is_array{ctx.stage == Stage::Geometry};
const auto vertex_arg{is_array ? fmt::format(",{}", vertex) : ""};
ctx.AddF32("{}=IndexedAttrLoad(i32({}){});", inst, offset, vertex_arg);
}
void EmitSetAttributeIndexed([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view offset,
[[maybe_unused]] std::string_view value,
[[maybe_unused]] std::string_view vertex) {
NotImplemented();
}
void EmitGetPatch(EmitContext& ctx, IR::Inst& inst, IR::Patch patch) {
if (!IR::IsGeneric(patch)) {
throw NotImplementedException("Non-generic patch load");
}
const u32 index{IR::GenericPatchIndex(patch)};
const u32 element{IR::GenericPatchElement(patch)};
const char swizzle{"xyzw"[element]};
ctx.AddF32("{}=patch{}.{};", inst, index, swizzle);
}
void EmitSetPatch(EmitContext& ctx, IR::Patch patch, std::string_view value) {
if (IR::IsGeneric(patch)) {
const u32 index{IR::GenericPatchIndex(patch)};
const u32 element{IR::GenericPatchElement(patch)};
ctx.Add("patch{}.{}={};", index, "xyzw"[element], value);
return;
}
switch (patch) {
case IR::Patch::TessellationLodLeft:
case IR::Patch::TessellationLodRight:
case IR::Patch::TessellationLodTop:
case IR::Patch::TessellationLodBottom: {
const u32 index{static_cast<u32>(patch) - u32(IR::Patch::TessellationLodLeft)};
ctx.Add("gl_TessLevelOuter[{}]={};", index, value);
break;
}
case IR::Patch::TessellationLodInteriorU:
ctx.Add("gl_TessLevelInner[0]={};", value);
break;
case IR::Patch::TessellationLodInteriorV:
ctx.Add("gl_TessLevelInner[1]={};", value);
break;
default:
throw NotImplementedException("Patch {}", patch);
}
}
void EmitSetFragColor(EmitContext& ctx, u32 index, u32 component, std::string_view value) {
const char swizzle{"xyzw"[component]};
ctx.Add("frag_color{}.{}={};", index, swizzle, value);
}
void EmitSetSampleMask(EmitContext& ctx, std::string_view value) {
ctx.Add("gl_SampleMask[0]=i32({});", value);
}
void EmitSetFragDepth(EmitContext& ctx, std::string_view value) {
ctx.Add("gl_FragDepth={};", value);
}
void EmitLocalInvocationId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32x3("{}=gl_LocalInvocationID;", inst);
}
void EmitWorkgroupId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32x3("{}=gl_WorkGroupID;", inst);
}
void EmitInvocationId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=u32(gl_InvocationID);", inst);
}
void EmitInvocationInfo(EmitContext& ctx, IR::Inst& inst) {
switch (ctx.stage) {
case Stage::TessellationControl:
case Stage::TessellationEval:
ctx.AddU32("{}=u32(gl_PatchVerticesIn)<<16;", inst);
break;
case Stage::Geometry:
ctx.AddU32("{}=u32({}<<16);", inst,
InputTopologyVertices::vertices(ctx.runtime_info.input_topology));
break;
default:
LOG_WARNING(Shader, "(STUBBED) called");
ctx.AddU32("{}=u32(0x00ff0000);", inst);
}
}
void EmitSampleId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=u32(gl_SampleID);", inst);
}
void EmitIsHelperInvocation(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU1("{}=gl_HelperInvocation;", inst);
}
void EmitSR_WScaleFactorXY(EmitContext& ctx, IR::Inst& inst) {
LOG_WARNING(Shader, "(STUBBED) called");
}
void EmitSR_WScaleFactorZ(EmitContext& ctx, IR::Inst& inst) {
LOG_WARNING(Shader, "(STUBBED) called");
}
void EmitYDirection(EmitContext& ctx, IR::Inst& inst) {
ctx.uses_y_direction = true;
ctx.AddF32("{}=1.0f;", inst); // the renderer flips Y in the viewport
}
void EmitResolutionDownFactor(EmitContext& ctx, IR::Inst& inst) {
ctx.AddF32("{}=scaling().z;", inst);
}
void EmitRenderArea(EmitContext& ctx, IR::Inst& inst) {
ctx.AddF32x4("{}=render_area();", inst);
}
void EmitLoadLocal(EmitContext& ctx, IR::Inst& inst, std::string_view word_offset) {
ctx.AddU32("{}=lmem[{}];", inst, word_offset);
}
void EmitWriteLocal(EmitContext& ctx, std::string_view word_offset, std::string_view value) {
ctx.Add("lmem[{}]={};", word_offset, value);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,18 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/exception.h"
namespace Shader::Backend::WGSL {
void EmitJoin(EmitContext&) {
throw NotImplementedException("Join shouldn't be emitted");
}
void EmitDemoteToHelperInvocation(EmitContext& ctx) {
ctx.Add("discard;");
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,259 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
void EmitConvertS16F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS16F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=u32((i32({})&0xffff)|(extractBits(i32({}),31u,1u)<<15u));", inst, value, value);
}
void EmitConvertS16F64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS32F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=u32(i32({}));", inst, value);
}
void EmitConvertS32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=u32(i32({}));", inst, value);
}
void EmitConvertS64F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=vec2<u32>(u32(i32({})),select(0u,0xffffffffu,{}<0.0f));", inst, value, value);
}
void EmitConvertS64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=vec2<u32>(u32(i32({})),select(0u,0xffffffffu,{}<0.0f));", inst, value, value);
}
void EmitConvertU16F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU16F32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU16F64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU32F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU16U32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU32U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU8U32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU32U8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS32S8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertS32S16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=u32({});", inst, value);
}
void EmitConvertU32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=u32({});", inst, value);
}
void EmitConvertU64F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertU64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=vec2<u32>(u32({}),0u);", inst, value);
}
void EmitConvertU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=vec2<u32>(u32({}),0u);", inst, value);
}
void EmitConvertU64U32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=vec2<u32>(u32({}),0u);", inst, value);
}
void EmitConvertU32U64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}={}.x;", inst, value);
}
void EmitConvertF16F32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32F16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=f32({});", inst, value);
}
void EmitConvertF64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
void EmitConvertF16S8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16S16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16S32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16S64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16U8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16U32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF16U64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32S8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32S16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32S32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=f32(i32({}));", inst, value);
}
void EmitConvertF32S64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=f32(i32({}.x));", inst, value);
}
void EmitConvertF32U8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF32U16(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=f32({}&0xffffu);", inst, value);
}
void EmitConvertF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=f32({});", inst, value);
}
void EmitConvertF32U64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=f32({}.x);", inst, value);
}
void EmitConvertF64S8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF64S16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF64S32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
void EmitConvertF64S64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
void EmitConvertF64U8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF64U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitConvertF64U32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
void EmitConvertF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
NotImplemented();
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,489 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <cctype>
#include <cmath>
#include <string>
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/modifiers.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
namespace {
/// A literal operand wrapped in rt_(): WGSL would fold the call at shader creation and reject a domain error.
std::string Rt(std::string_view v) {
const bool literal{!v.empty() && (std::isdigit(static_cast<unsigned char>(v[0])) || v[0] == '-' || v[0] == '.')};
return literal ? fmt::format("rt_({})", v) : std::string{v};
}
void Compare(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs,
std::string_view op, bool ordered) {
const auto nan_op{ordered ? "&&!" : "||"};
ctx.AddU1("{}={}{}{}"
"{}isnan_({}){}isnan_({});",
inst, lhs, op, rhs, nan_op, lhs, nan_op, rhs);
}
bool IsPrecise(const IR::Inst& inst) {
return inst.Flags<IR::FpControl>().no_contraction;
}
/// A float literal operand of magnitude below 1e-6 (an epsilon guarding a division or a logarithm).
bool IsEpsilon(std::string_view v) {
if (v.size() < 2 || v.back() != 'f') return false;
const char c{v[0]};
if (!std::isdigit(static_cast<unsigned char>(c)) && c != '-' && c != '.') return false;
try {
const float f{std::stof(std::string{v.substr(0, v.size() - 1)})};
return f != 0.0f && std::fabs(f) < 1e-6f;
} catch (...) {
return false;
}
}
} // Anonymous namespace
void EmitFPAbs16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=abs({});", inst, value);
}
void EmitFPAbs64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=abs({});", inst, value);
}
void EmitFPAdd16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view a, [[maybe_unused]] std::string_view b) {
NotImplemented();
}
void EmitFPAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
// x + epsilon: x is made opaque, so that the host compiler cannot fold the epsilon into an earlier constant
if (IsEpsilon(b) && !IsEpsilon(a)) {
ctx.AddF32("{}=opq_({})+{};", inst, a, b);
return;
}
if (IsEpsilon(a) && !IsEpsilon(b)) {
ctx.AddF32("{}={}+opq_({});", inst, a, b);
return;
}
if (IsPrecise(inst)) {
ctx.AddPrecF32("{}={}+{};", inst, a, b);
} else {
ctx.AddF32("{}={}+{};", inst, a, b);
}
}
void EmitFPAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
if (IsPrecise(inst)) {
ctx.AddPrecF64("{}={}+{};", inst, a, b);
} else {
ctx.AddF64("{}={}+{};", inst, a, b);
}
}
void EmitFPFma16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view a, [[maybe_unused]] std::string_view b,
[[maybe_unused]] std::string_view c) {
NotImplemented();
}
void EmitFPFma32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c) {
if (IsPrecise(inst)) {
ctx.AddPrecF32("{}=fma({},{},{});", inst, a, b, c);
} else {
ctx.AddF32("{}=fma({},{},{});", inst, a, b, c);
}
}
void EmitFPFma64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c) {
if (IsPrecise(inst)) {
ctx.AddPrecF64("{}=fma({},{},{});", inst, a, b, c);
} else {
ctx.AddF64("{}=fma({},{},{});", inst, a, b, c);
}
}
void EmitFPMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddF32("{}=max({},{});", inst, a, b);
}
void EmitFPMax64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddF64("{}=max({},{});", inst, a, b);
}
void EmitFPMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddF32("{}=min({},{});", inst, a, b);
}
void EmitFPMin64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddF64("{}=min({},{});", inst, a, b);
}
void EmitFPMul16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view a, [[maybe_unused]] std::string_view b) {
NotImplemented();
}
void EmitFPMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
if (IsPrecise(inst)) {
ctx.AddPrecF32("{}={}*{};", inst, a, b);
} else {
ctx.AddF32("{}={}*{};", inst, a, b);
}
}
void EmitFPMul64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
if (IsPrecise(inst)) {
ctx.AddPrecF64("{}={}*{};", inst, a, b);
} else {
ctx.AddF64("{}={}*{};", inst, a, b);
}
}
void EmitFPNeg16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPNeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=0.f-({});", inst, value);
}
void EmitFPNeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=double(0.)-({});", inst, value);
}
void EmitFPSin(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=sin({});", inst, value);
}
void EmitFPCos(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=cos({});", inst, value);
}
void EmitFPExp2(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
// Maxwell's EX2 takes its input through RRO (a fixed-point range reduction) where NaN becomes 0: pow(0, 0)
// compiled as exp2(0 * log2(0)) is 1 there, NaN on PC GPUs (TOTK sky islands: white blobs in the lighting)
ctx.AddF32("{}=exp2(lodok_({}));", inst, Rt(value));
}
void EmitFPLog2(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=log2({});", inst, Rt(value));
}
void EmitFPRecip32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=(1.0f)/{};", inst, Rt(value));
}
void EmitFPRecip64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=1.0/{};", inst, value);
}
void EmitFPRecipSqrt32([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
ctx.AddF32("{}=inverseSqrt({});", inst, Rt(value));
}
void EmitFPRecipSqrt64([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPSqrt(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=sqrt({});", inst, Rt(value));
}
void EmitFPSaturate16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPSaturate32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
// Maxwell's .SAT makes NaN 0 (the frontend's FMZ emulation relies on it); WGSL's max(NaN, 0) may be NaN
ctx.AddF32("{0}=select(min(max({1},0.0),1.0),0.0,isnan_({1}));", inst, value);
}
void EmitFPSaturate64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=min(max({},0.0),1.0);", inst, value);
}
void EmitFPClamp16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value,
[[maybe_unused]] std::string_view min_value,
[[maybe_unused]] std::string_view max_value) {
NotImplemented();
}
void EmitFPClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value) {
// GLSL's clamp does not produce desirable results
ctx.AddF32("{}=min(max({},f32({})),f32({}));", inst, value, min_value, max_value);
}
void EmitFPClamp64(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value) {
// GLSL's clamp does not produce desirable results
ctx.AddF64("{}=min(max({},double({})),double({}));", inst, value, min_value, max_value);
}
void EmitFPRoundEven16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPRoundEven32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=round({});", inst, value);
}
void EmitFPRoundEven64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=round({});", inst, value);
}
void EmitFPFloor16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPFloor32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=floor({});", inst, value);
}
void EmitFPFloor64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=floor({});", inst, value);
}
void EmitFPCeil16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPCeil32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=ceil({});", inst, value);
}
void EmitFPCeil64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=ceil({});", inst, value);
}
void EmitFPTrunc16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPTrunc32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF32("{}=trunc({});", inst, value);
}
void EmitFPTrunc64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddF64("{}=trunc({});", inst, value);
}
void EmitFPOrdEqual16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "==", true);
}
void EmitFPOrdEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "==", true);
}
void EmitFPUnordEqual16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "==", false);
}
void EmitFPUnordEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "==", false);
}
void EmitFPOrdNotEqual16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdNotEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "!=", true);
}
void EmitFPOrdNotEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "!=", true);
}
void EmitFPUnordNotEqual16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordNotEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "!=", false);
}
void EmitFPUnordNotEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "!=", false);
}
void EmitFPOrdLessThan16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdLessThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<", true);
}
void EmitFPOrdLessThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<", true);
}
void EmitFPUnordLessThan16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordLessThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<", false);
}
void EmitFPUnordLessThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<", false);
}
void EmitFPOrdGreaterThan16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdGreaterThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">", true);
}
void EmitFPOrdGreaterThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">", true);
}
void EmitFPUnordGreaterThan16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordGreaterThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">", false);
}
void EmitFPUnordGreaterThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">", false);
}
void EmitFPOrdLessThanEqual16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdLessThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<=", true);
}
void EmitFPOrdLessThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<=", true);
}
void EmitFPUnordLessThanEqual16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordLessThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<=", false);
}
void EmitFPUnordLessThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, "<=", false);
}
void EmitFPOrdGreaterThanEqual16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPOrdGreaterThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">=", true);
}
void EmitFPOrdGreaterThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">=", true);
}
void EmitFPUnordGreaterThanEqual16([[maybe_unused]] EmitContext& ctx,
[[maybe_unused]] std::string_view lhs,
[[maybe_unused]] std::string_view rhs) {
NotImplemented();
}
void EmitFPUnordGreaterThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">=", false);
}
void EmitFPUnordGreaterThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
Compare(ctx, inst, lhs, rhs, ">=", false);
}
void EmitFPIsNan16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst,
[[maybe_unused]] std::string_view value) {
NotImplemented();
}
void EmitFPIsNan32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU1("{}=isnan_({});", inst, value);
}
void EmitFPIsNan64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU1("{}=isnan_({});", inst, value);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,644 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
//
// WGSL backend (Rexcomp): texture and image instructions, rewritten for WGSL's separate texture/sampler model.
#include <cstdlib>
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/modifiers.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::WGSL {
namespace {
const TextureImageDefinition& TextureDef(EmitContext& ctx, const IR::TextureInstInfo& info) {
if (info.type == TextureType::Buffer) {
throw NotImplementedException("WGSL texture buffers");
}
return ctx.textures.at(info.descriptor_index);
}
/// "texN,sampN": the texture and its sampler
std::string TexSamp(EmitContext& ctx, const IR::TextureInstInfo& info) {
const auto& def{TextureDef(ctx, info)};
return fmt::format("tex{},samp{}", def.binding, def.textures_sampler_binding());
}
std::string Tex(EmitContext& ctx, const IR::TextureInstInfo& info) {
return fmt::format("tex{}", TextureDef(ctx, info).binding);
}
std::string Swizzled(EmitContext& ctx, const IR::TextureInstInfo& info, const std::string& expression) {
return fmt::format("swizzle{}({})", TextureDef(ctx, info).binding, expression);
}
/// The normalized coordinates a border test looks at (xyz; axes the texture type has not are inside), or "" when
/// the type has no border (cubes, unnormalized coordinates).
std::string BorderCoords(const IR::TextureInstInfo& info, std::string_view c) {
switch (info.type) {
case TextureType::Color1D: return fmt::format("vec3<f32>(f32({}),0.5,0.5)", c);
case TextureType::Color2D:
case TextureType::ColorArray2D: return fmt::format("vec3<f32>({}.xy,0.5)", c);
case TextureType::Color3D: return fmt::format("vec3<f32>({}.xyz)", c);
default: return "";
}
}
/// A sampled color with the sampler's border color outside [0, 1] on its border axes.
std::string Bordered(EmitContext& ctx, const IR::TextureInstInfo& info, std::string_view coords,
const std::string& expression) {
const auto bc{BorderCoords(info, coords)};
if (bc.empty()) {
return expression;
}
return fmt::format("border{}({},{})", TextureDef(ctx, info).binding, bc, expression);
}
/// A depth comparison's reference, clamped to [0, 1] as the console compares it. WebGPU (Vulkan) clamps it only for a
/// UNORM depth texture: against a float one a point just past a shadow cascade's far plane compared 1.0x <= 1.0 and
/// fell into shadow. TOTK's moonlight cascades end within view: dark bands across the ground at night, moving with the
/// camera (gone with the clamp, 2026-10-28). RX_NO_DREF_CLAMP=1: the reference as computed (debugging).
/// Against an array texture (TOTK's shadow cascades: D16 arrays) the reference is also rounded to 16 bits, as Maxwell
/// compares at the texture's precision: a driver that compares the full float (AMD's) left a lit face failing its own
/// compare by a fraction of a step, shadowed (totk-pc-recomp 6029719 / MacNativeGames 6283599, MIT).
/// RX_NO_D16_DREF=1: not rounded.
std::string Dref(std::string_view dref, bool array = false) {
static const bool no_clamp{std::getenv("RX_NO_DREF_CLAMP") != nullptr};
static const bool no_d16{std::getenv("RX_NO_D16_DREF") != nullptr};
if (no_clamp) return std::string{dref};
if (array && !no_d16) return fmt::format("(floor(clamp({},0.0,1.0)*65535.0+0.5)/65535.0)", dref);
return fmt::format("clamp({},0.0,1.0)", dref);
}
/// The same for a depth comparison: outside, the reference is compared with the border depth.
std::string BorderedCompare(EmitContext& ctx, const IR::TextureInstInfo& info, std::string_view coords,
std::string_view dref, const std::string& expression) {
const auto bc{BorderCoords(info, coords)};
if (bc.empty()) {
return expression;
}
return fmt::format("bordercmp{}({},{},{})", TextureDef(ctx, info).binding, bc, dref, expression);
}
std::string Image(EmitContext& ctx, const IR::TextureInstInfo& info) {
if (info.type == TextureType::Buffer) {
throw NotImplementedException("WGSL image buffers");
}
return fmt::format("img{}", ctx.images.at(info.descriptor_index).binding);
}
bool IsArray(const IR::TextureInstInfo& info) {
return info.type == TextureType::ColorArray2D || info.type == TextureType::ColorArrayCube;
}
/// Sampling coordinates, with the array layer as WGSL's separate argument.
std::string SampleCoords(const IR::TextureInstInfo& info, std::string_view c) {
switch (info.type) {
case TextureType::Color1D:
case TextureType::Color2D:
case TextureType::Color2DRect:
case TextureType::Color3D:
case TextureType::ColorCube:
return std::string{c};
case TextureType::ColorArray2D:
return fmt::format("{}.xy,i32({}.z)", c, c);
case TextureType::ColorArrayCube:
return fmt::format("{}.xyz,i32({}.w)", c, c);
default:
throw NotImplementedException("WGSL sampling of texture type {}", info.type.Value());
}
}
/// Integer coordinates for textureLoad/textureStore (u32 inputs), with the array layer separate; `offset` (an
/// i32 vector expression or empty) is added to the texel coordinates.
std::string LoadCoords(const IR::TextureInstInfo& info, std::string_view c, std::string_view offset = {}) {
const auto add{offset.empty() ? std::string{} : fmt::format("+{}", offset)};
switch (info.type) {
case TextureType::Color1D:
return fmt::format("i32({}){}", c, add);
case TextureType::Color2D:
case TextureType::Color2DRect:
return fmt::format("vec2<i32>({}){}", c, add);
case TextureType::ColorArray2D:
return fmt::format("vec2<i32>({}.xy){},i32({}.z)", c, add, c);
case TextureType::Color3D:
return fmt::format("vec3<i32>({}){}", c, add);
default:
throw NotImplementedException("WGSL texel loads of texture type {}", info.type.Value());
}
}
bool OffsetAllowed(const IR::TextureInstInfo& info) {
switch (info.type) {
case TextureType::Color2D:
case TextureType::Color2DRect:
case TextureType::ColorArray2D:
case TextureType::Color3D:
return true;
default:
return false;
}
}
/// A texel offset as a WGSL constant expression (WGSL requires one), or "" when there is none or it is not constant.
std::string ConstOffset(const IR::Value& offset) {
if (offset.IsEmpty()) {
return "";
}
if (offset.IsImmediate()) {
return fmt::format("{}", static_cast<s32>(offset.U32()));
}
IR::Inst* const inst{offset.InstRecursive()};
if (inst->AreAllArgsImmediates()) {
switch (inst->GetOpcode()) {
case IR::Opcode::CompositeConstructU32x2:
return fmt::format("vec2<i32>({},{})", static_cast<s32>(inst->Arg(0).U32()),
static_cast<s32>(inst->Arg(1).U32()));
case IR::Opcode::CompositeConstructU32x3:
return fmt::format("vec3<i32>({},{},{})", static_cast<s32>(inst->Arg(0).U32()),
static_cast<s32>(inst->Arg(1).U32()), static_cast<s32>(inst->Arg(2).U32()));
default:
break;
}
}
LOG_WARNING(Shader_WGSL, "Variable texture offset dropped (WGSL offsets are constants)");
return "";
}
/// ",offset" for a sampling call, when the texture type takes one
std::string SampleOffsetArg(const IR::TextureInstInfo& info, const IR::Value& offset) {
const auto o{ConstOffset(offset)};
if (o.empty() && !offset.IsEmpty()) return "/*VAROFFSET*/";
return o.empty() || !OffsetAllowed(info) ? "" : "," + o;
}
/// A texel offset as an i32 vector expression for textureLoad (variable offsets are fine here).
std::string LoadOffset(EmitContext& ctx, const IR::TextureInstInfo& info, const IR::Value& offset) {
if (offset.IsEmpty()) {
return "";
}
const auto o{ConstOffset(offset)};
if (!o.empty()) {
return o;
}
const auto v{ctx.var_alloc.Consume(offset)};
switch (info.type) {
case TextureType::Color1D:
return fmt::format("i32({})", v);
case TextureType::Color3D:
return fmt::format("vec3<i32>({})", v);
default:
return fmt::format("vec2<i32>({})", v);
}
}
IR::Inst* PrepareSparse(EmitContext& ctx, IR::Inst& inst) {
IR::Inst* const sparse_inst{inst.GetAssociatedPseudoOperation(IR::Opcode::GetSparseFromOp)};
if (sparse_inst) {
// WebGPU has no sparse residency: every texel is resident
ctx.AddU1("{}=true;", *sparse_inst);
sparse_inst->Invalidate();
}
return sparse_inst;
}
} // Anonymous namespace
void EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view bias_lc,
const IR::Value& offset) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageSampleImplicitLod Lod clamp samples");
}
(void)index;
PrepareSparse(ctx, inst);
const auto ts{TexSamp(ctx, info)};
const auto c{SampleCoords(info, coords)};
const auto o{SampleOffsetArg(info, offset)};
// the sampler's LOD bias (lodbiasN, from the uniform block) is added: WebGPU samplers have none
const auto lb{fmt::format("lodbias{}()", TextureDef(ctx, info).binding)};
if (ctx.stage != Stage::Fragment) {
ctx.AddF32x4("{}={};", inst, Bordered(ctx, info, coords, Swizzled(ctx, info, fmt::format("textureSampleLevel({},{},{}{})", ts, c, lb, o))));
} else if (info.has_bias) {
ctx.AddF32x4("{}={};", inst, Bordered(ctx, info, coords, Swizzled(ctx, info, fmt::format("textureSampleBias({},{},lodok_({}+{}){})", ts, c, bias_lc, lb, o))));
} else if (info.type == TextureType::Color1D) {
ctx.AddF32x4("{}={};", inst, Bordered(ctx, info, coords, Swizzled(ctx, info, fmt::format("textureSample({},{}{})", ts, c, o))));
} else {
ctx.AddF32x4("{}={};", inst, Bordered(ctx, info, coords, Swizzled(ctx, info, fmt::format("textureSampleBias({},{},{}{})", ts, c, lb, o))));
}
}
void EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view lod_lc,
const IR::Value& offset) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_bias) {
throw NotImplementedException("EmitImageSampleExplicitLod Bias texture samples");
}
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageSampleExplicitLod Lod clamp samples");
}
(void)index;
PrepareSparse(ctx, inst);
ctx.AddF32x4("{}={};", inst, Bordered(ctx, info, coords, Swizzled(ctx, info, fmt::format("textureSampleLevel({},{},lodok_({}+lodbias{}()){})",
TexSamp(ctx, info), SampleCoords(info, coords), lod_lc, TextureDef(ctx, info).binding,
SampleOffsetArg(info, offset)))));
}
void EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view dref,
std::string_view bias_lc, const IR::Value& offset) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageSampleDrefImplicitLod Lod clamp samples");
}
(void)index;
(void)bias_lc;
PrepareSparse(ctx, inst);
const auto fn{ctx.stage == Stage::Fragment ? "textureSampleCompare" : "textureSampleCompareLevel"};
const auto d{Dref(dref, IsArray(info))};
ctx.AddF32("{}={};", inst, BorderedCompare(ctx, info, coords, d, fmt::format("{}({},{},{}{})", fn,
TexSamp(ctx, info), SampleCoords(info, coords), d, SampleOffsetArg(info, offset))));
}
void EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view dref,
std::string_view lod_lc, const IR::Value& offset) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageSampleDrefExplicitLod Lod clamp samples");
}
(void)index;
(void)lod_lc; // WGSL compares at level 0 only
PrepareSparse(ctx, inst);
const auto d{Dref(dref, IsArray(info))};
ctx.AddF32("{}={};", inst, BorderedCompare(ctx, info, coords, d, fmt::format("textureSampleCompareLevel({},{},{}{})",
TexSamp(ctx, info), SampleCoords(info, coords), d, SampleOffsetArg(info, offset))));
}
void EmitImageGather(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, const IR::Value& offset2) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
(void)index;
if (!offset2.IsEmpty()) ctx.Add("/*PTP*/"); // per-texel offsets (PTP): WGSL has none; the first offset is used
PrepareSparse(ctx, inst);
const bool depth{ctx.info.texture_descriptors.at(info.descriptor_index).is_depth};
const auto component{depth ? std::string{} : fmt::format("{},", info.gather_component)};
// A gather at a texel's center sits on the edge between two footprints: the game uses .w of such a gather as a
// point fetch (TOTK's half-resolution lighting upsample), so 1/256 of a texel decides the footprint the way
// Maxwell's rounding does instead of the interpolation error of the coordinate
std::string c{coords};
static const bool no_nudge{std::getenv("RX_NO_GATHER_NUDGE") != nullptr}; // (debugging)
if (no_nudge) {
} else if (info.type == TextureType::Color2D || info.type == TextureType::Color2DRect) {
c = fmt::format("({}+vec2<f32>(0.00390625)/vec2<f32>(textureDimensions({},0)))", coords, Tex(ctx, info));
} else if (info.type == TextureType::ColorArray2D) {
c = fmt::format("vec3<f32>({0}.xy+vec2<f32>(0.00390625)/vec2<f32>(textureDimensions({1},0)),{0}.z)", coords,
Tex(ctx, info));
}
ctx.AddF32x4("{}=textureGather({}{},{}{});", inst, component, TexSamp(ctx, info), SampleCoords(info, c),
SampleOffsetArg(info, offset));
}
void EmitImageGatherDref(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, const IR::Value& offset2,
std::string_view dref) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
(void)index;
(void)offset2;
PrepareSparse(ctx, inst);
ctx.AddF32x4("{}=textureGatherCompare({},{},{}{});", inst, TexSamp(ctx, info), SampleCoords(info, coords),
Dref(dref, IsArray(info)), SampleOffsetArg(info, offset));
}
void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, std::string_view lod,
std::string_view ms) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_bias) {
throw NotImplementedException("EmitImageFetch Bias texture samples");
}
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageFetch Lod clamp samples");
}
(void)index;
PrepareSparse(ctx, inst);
const auto o{LoadOffset(ctx, info, offset)};
const auto c{LoadCoords(info, coords, o)};
const bool depth{ctx.info.texture_descriptors.at(info.descriptor_index).is_depth};
const auto level{!ms.empty() ? fmt::format("i32({})", ms) : fmt::format("i32({})", lod)};
if (depth) {
ctx.AddF32x4("{}={};", inst, Swizzled(ctx, info, fmt::format("vec4<f32>(textureLoad({},{},{}),0.0f,0.0f,1.0f)", Tex(ctx, info), c, level)));
} else {
ctx.AddF32x4("{}={};", inst, Swizzled(ctx, info, fmt::format("textureLoad({},{},{})", Tex(ctx, info), c, level)));
}
}
void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view lod, const IR::Value& skip_mips_val) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
(void)index;
const auto t{Tex(ctx, info)};
const bool is_msaa{ctx.info.texture_descriptors.at(info.descriptor_index).is_multisample};
const auto mips{skip_mips_val.U1() || is_msaa ? std::string{"0u"} : fmt::format("textureNumLevels({})", t)};
const auto dims{is_msaa ? fmt::format("textureDimensions({})", t)
: fmt::format("textureDimensions({},u32({}))", t, lod)};
switch (info.type) {
case TextureType::Color1D:
return ctx.AddU32x4("{}=vec4<u32>({},0u,0u,{});", inst, dims, mips);
case TextureType::Color2D:
case TextureType::Color2DRect:
case TextureType::ColorCube:
return ctx.AddU32x4("{}=vec4<u32>({},0u,{});", inst, dims, mips);
case TextureType::ColorArray2D:
case TextureType::ColorArrayCube:
return ctx.AddU32x4("{}=vec4<u32>({},textureNumLayers({}),{});", inst, dims, t, mips);
case TextureType::Color3D:
return ctx.AddU32x4("{}=vec4<u32>({},{});", inst, dims, mips);
default:
throw NotImplementedException("WGSL dimension query of texture type {}", info.type.Value());
}
}
void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
(void)index;
if (ctx.stage != Stage::Fragment || (info.type != TextureType::Color2D && info.type != TextureType::ColorArray2D)) {
ctx.AddF32x4("{}=vec4<f32>(0.0f);", inst);
return;
}
// WGSL has no textureQueryLod: the level from the screen-space derivatives of the texel coordinates
const auto t{Tex(ctx, info)};
ctx.AddF32x4("{}=vec4<f32>(vec2<f32>(max(0.5f*log2(max(dot(dpdx({}.xy*vec2<f32>(textureDimensions({},0u))),"
"dpdx({}.xy*vec2<f32>(textureDimensions({},0u)))),dot(dpdy({}.xy*vec2<f32>(textureDimensions({},0u))),"
"dpdy({}.xy*vec2<f32>(textureDimensions({},0u)))))),0.0f)),0.0f,0.0f);",
inst, coords, t, coords, t, coords, t, coords, t);
}
void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& derivatives,
const IR::Value& offset, [[maybe_unused]] const IR::Value& lod_clamp) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
if (info.has_lod_clamp) {
throw NotImplementedException("EmitImageGradient Lod clamp samples");
}
(void)index;
PrepareSparse(ctx, inst);
const auto d{ctx.var_alloc.Consume(derivatives)};
const auto ts{TexSamp(ctx, info)};
const auto c{SampleCoords(info, coords)};
if (info.num_derivatives >= 3) {
const auto o{ctx.var_alloc.Consume(offset)};
ctx.AddF32x4("{}={};", inst, Swizzled(ctx, info, fmt::format("textureSampleGrad({},{},vec3<f32>({}.xz,{}.x),vec3<f32>({}.yw,{}.y))", ts, c, d, o, d, o)));
return;
}
if (info.num_derivatives == 2) {
// (the LOD bias as scaled derivatives)
ctx.AddF32x4("{}={};", inst, Bordered(ctx, info, coords, Swizzled(ctx, info, fmt::format("textureSampleGrad({},{},{}.xz*exp2(lodbias{}()),{}.yw*exp2(lodbias{}()){})", ts, c, d,
TextureDef(ctx, info).binding, d, TextureDef(ctx, info).binding, SampleOffsetArg(info, offset)))));
return;
}
throw NotImplementedException("WGSL 1D gradient samples");
}
void EmitImageRead(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
(void)index;
PrepareSparse(ctx, inst);
// (mkw-recomp) a float or signed image's texels as the program's 32-bit words
const u32 kind = ctx.images.at(info.descriptor_index).kind;
if (kind == 0) ctx.AddU32x4("{}=textureLoad({},{});", inst, Image(ctx, info), LoadCoords(info, coords));
else ctx.AddU32x4("{}=bitcast<vec4<u32>>(textureLoad({},{}));", inst, Image(ctx, info), LoadCoords(info, coords));
}
void EmitImageWrite(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view color) {
const auto info{inst.Flags<IR::TextureInstInfo>()};
(void)index;
const u32 kind = ctx.images.at(info.descriptor_index).kind;
ctx.Add("textureStore({},{},{});", Image(ctx, info), LoadCoords(info, coords),
kind == 0 ? std::string(color) : fmt::format("bitcast<vec4<{}>>({})", kind == 1 ? "f32" : "i32", color));
}
#define IMAGE_ATOMIC(name) \
void EmitImageAtomic##name(EmitContext&, IR::Inst&, const IR::Value&, std::string_view, \
std::string_view) { \
throw NotImplementedException("WGSL image atomics"); \
}
IMAGE_ATOMIC(IAdd32)
IMAGE_ATOMIC(SMin32)
IMAGE_ATOMIC(UMin32)
IMAGE_ATOMIC(SMax32)
IMAGE_ATOMIC(UMax32)
IMAGE_ATOMIC(Inc32)
IMAGE_ATOMIC(Dec32)
IMAGE_ATOMIC(And32)
IMAGE_ATOMIC(Or32)
IMAGE_ATOMIC(Xor32)
IMAGE_ATOMIC(Exchange32)
#undef IMAGE_ATOMIC
void EmitIsTextureScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index) {
if (!index.IsImmediate()) {
throw NotImplementedException("Non-constant texture rescaling");
}
ctx.AddU1("{}=(bitcast<u32>(scaling().x)&{}u)!=0u;", inst, 1u << index.U32());
}
void EmitIsImageScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index) {
if (!index.IsImmediate()) {
throw NotImplementedException("Non-constant texture rescaling");
}
ctx.AddU1("{}=(bitcast<u32>(scaling().y)&{}u)!=0u;", inst, 1u << index.U32());
}
void EmitBindlessImageSampleImplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageSampleExplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageSampleDrefImplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageSampleDrefExplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageGather(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageGatherDref(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageFetch(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageQueryDimensions(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageQueryLod(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageGradient(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageRead(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageWrite(EmitContext&) {
NotImplemented();
}
void EmitBoundImageSampleImplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageSampleExplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageSampleDrefImplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageSampleDrefExplicitLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageGather(EmitContext&) {
NotImplemented();
}
void EmitBoundImageGatherDref(EmitContext&) {
NotImplemented();
}
void EmitBoundImageFetch(EmitContext&) {
NotImplemented();
}
void EmitBoundImageQueryDimensions(EmitContext&) {
NotImplemented();
}
void EmitBoundImageQueryLod(EmitContext&) {
NotImplemented();
}
void EmitBoundImageGradient(EmitContext&) {
NotImplemented();
}
void EmitBoundImageRead(EmitContext&) {
NotImplemented();
}
void EmitBoundImageWrite(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicIAdd32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicSMin32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicUMin32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicSMax32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicUMax32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicInc32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicDec32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicAnd32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicOr32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicXor32(EmitContext&) {
NotImplemented();
}
void EmitBindlessImageAtomicExchange32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicIAdd32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicSMin32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicUMin32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicSMax32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicUMax32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicInc32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicDec32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicAnd32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicOr32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicXor32(EmitContext&) {
NotImplemented();
}
void EmitBoundImageAtomicExchange32(EmitContext&) {
NotImplemented();
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,753 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <string_view>
#include "common/common_types.h"
namespace Shader::IR {
enum class Attribute : u64;
enum class Patch : u64;
class Inst;
class Value;
} // namespace Shader::IR
namespace Shader::Backend::WGSL {
class EmitContext;
#define NotImplemented() throw NotImplementedException("GLSL instruction {}", __func__)
// Microinstruction emitters
void EmitPhi(EmitContext& ctx, IR::Inst& inst);
void EmitVoid(EmitContext& ctx);
void EmitIdentity(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
void EmitConditionRef(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
void EmitReference(EmitContext& ctx, const IR::Value& value);
void EmitPhiMove(EmitContext& ctx, const IR::Value& phi, const IR::Value& value);
void EmitJoin(EmitContext& ctx);
void EmitDemoteToHelperInvocation(EmitContext& ctx);
void EmitBarrier(EmitContext& ctx);
void EmitWorkgroupMemoryBarrier(EmitContext& ctx);
void EmitDeviceMemoryBarrier(EmitContext& ctx);
void EmitPrologue(EmitContext& ctx);
void EmitEpilogue(EmitContext& ctx);
void EmitEmitVertex(EmitContext& ctx, const IR::Value& stream);
void EmitEndPrimitive(EmitContext& ctx, const IR::Value& stream);
void EmitGetRegister(EmitContext& ctx);
void EmitSetRegister(EmitContext& ctx);
void EmitGetPred(EmitContext& ctx);
void EmitSetPred(EmitContext& ctx);
void EmitSetGotoVariable(EmitContext& ctx);
void EmitGetGotoVariable(EmitContext& ctx);
void EmitSetIndirectBranchVariable(EmitContext& ctx);
void EmitGetIndirectBranchVariable(EmitContext& ctx);
void EmitGetCbufU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufS8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufU16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufS16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufU32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetCbufU32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr,
std::string_view vertex);
void EmitGetAttributeU32(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr,
std::string_view vertex);
void EmitSetAttribute(EmitContext& ctx, IR::Attribute attr, std::string_view value,
std::string_view vertex);
void EmitGetAttributeIndexed(EmitContext& ctx, IR::Inst& inst, std::string_view offset,
std::string_view vertex);
void EmitSetAttributeIndexed(EmitContext& ctx, std::string_view offset, std::string_view value,
std::string_view vertex);
void EmitGetPatch(EmitContext& ctx, IR::Inst& inst, IR::Patch patch);
void EmitSetPatch(EmitContext& ctx, IR::Patch patch, std::string_view value);
void EmitSetFragColor(EmitContext& ctx, u32 index, u32 component, std::string_view value);
void EmitSetSampleMask(EmitContext& ctx, std::string_view value);
void EmitSetFragDepth(EmitContext& ctx, std::string_view value);
void EmitGetZFlag(EmitContext& ctx);
void EmitGetSFlag(EmitContext& ctx);
void EmitGetCFlag(EmitContext& ctx);
void EmitGetOFlag(EmitContext& ctx);
void EmitSetZFlag(EmitContext& ctx);
void EmitSetSFlag(EmitContext& ctx);
void EmitSetCFlag(EmitContext& ctx);
void EmitSetOFlag(EmitContext& ctx);
void EmitWorkgroupId(EmitContext& ctx, IR::Inst& inst);
void EmitLocalInvocationId(EmitContext& ctx, IR::Inst& inst);
void EmitInvocationId(EmitContext& ctx, IR::Inst& inst);
void EmitInvocationInfo(EmitContext& ctx, IR::Inst& inst);
void EmitSampleId(EmitContext& ctx, IR::Inst& inst);
void EmitIsHelperInvocation(EmitContext& ctx, IR::Inst& inst);
void EmitSR_WScaleFactorXY(EmitContext& ctx, IR::Inst& inst);
void EmitSR_WScaleFactorZ(EmitContext& ctx, IR::Inst& inst);
void EmitYDirection(EmitContext& ctx, IR::Inst& inst);
void EmitResolutionDownFactor(EmitContext& ctx, IR::Inst& inst);
void EmitRenderArea(EmitContext& ctx, IR::Inst& inst);
void EmitLoadLocal(EmitContext& ctx, IR::Inst& inst, std::string_view word_offset);
void EmitWriteLocal(EmitContext& ctx, std::string_view word_offset, std::string_view value);
void EmitUndefU1(EmitContext& ctx, IR::Inst& inst);
void EmitUndefU8(EmitContext& ctx, IR::Inst& inst);
void EmitUndefU16(EmitContext& ctx, IR::Inst& inst);
void EmitUndefU32(EmitContext& ctx, IR::Inst& inst);
void EmitUndefU64(EmitContext& ctx, IR::Inst& inst);
void EmitLoadGlobalU8(EmitContext& ctx);
void EmitLoadGlobalS8(EmitContext& ctx);
void EmitLoadGlobalU16(EmitContext& ctx);
void EmitLoadGlobalS16(EmitContext& ctx);
void EmitLoadGlobal32(EmitContext& ctx, IR::Inst& inst, std::string_view address);
void EmitLoadGlobal64(EmitContext& ctx, IR::Inst& inst, std::string_view address);
void EmitLoadGlobal128(EmitContext& ctx, IR::Inst& inst, std::string_view address);
void EmitWriteGlobalU8(EmitContext& ctx);
void EmitWriteGlobalS8(EmitContext& ctx);
void EmitWriteGlobalU16(EmitContext& ctx);
void EmitWriteGlobalS16(EmitContext& ctx);
void EmitWriteGlobal32(EmitContext& ctx, std::string_view address, std::string_view value);
void EmitWriteGlobal64(EmitContext& ctx, std::string_view address, std::string_view value);
void EmitWriteGlobal128(EmitContext& ctx, std::string_view address, std::string_view value);
void EmitLoadStorageU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorageS8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorageU16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorageS16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorage32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorage64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitLoadStorage128(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset);
void EmitWriteStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorageS8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorageU16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorageS16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorage32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorage64(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitWriteStorage128(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value);
void EmitLoadSharedU8(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedS8(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedU16(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedS16(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedU32(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedU64(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitLoadSharedU128(EmitContext& ctx, IR::Inst& inst, std::string_view offset);
void EmitWriteSharedU8(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitWriteSharedU16(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitWriteSharedU32(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitWriteSharedU64(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitWriteSharedU128(EmitContext& ctx, std::string_view offset, std::string_view value);
void EmitCompositeConstructU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2);
void EmitCompositeConstructU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3);
void EmitCompositeConstructU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3, std::string_view e4);
void EmitCompositeExtractU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeExtractU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeExtractU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeInsertU32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeInsertU32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeInsertU32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeConstructF16x2(EmitContext& ctx, std::string_view e1, std::string_view e2);
void EmitCompositeConstructF16x3(EmitContext& ctx, std::string_view e1, std::string_view e2,
std::string_view e3);
void EmitCompositeConstructF16x4(EmitContext& ctx, std::string_view e1, std::string_view e2,
std::string_view e3, std::string_view e4);
void EmitCompositeExtractF16x2(EmitContext& ctx, std::string_view composite, u32 index);
void EmitCompositeExtractF16x3(EmitContext& ctx, std::string_view composite, u32 index);
void EmitCompositeExtractF16x4(EmitContext& ctx, std::string_view composite, u32 index);
void EmitCompositeInsertF16x2(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeInsertF16x3(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeInsertF16x4(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeConstructF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2);
void EmitCompositeConstructF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3);
void EmitCompositeConstructF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view e1,
std::string_view e2, std::string_view e3, std::string_view e4);
void EmitCompositeExtractF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeExtractF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeExtractF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
u32 index);
void EmitCompositeInsertF32x2(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeInsertF32x3(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeInsertF32x4(EmitContext& ctx, IR::Inst& inst, std::string_view composite,
std::string_view object, u32 index);
void EmitCompositeConstructF64x2(EmitContext& ctx);
void EmitCompositeConstructF64x3(EmitContext& ctx);
void EmitCompositeConstructF64x4(EmitContext& ctx);
void EmitCompositeExtractF64x2(EmitContext& ctx);
void EmitCompositeExtractF64x3(EmitContext& ctx);
void EmitCompositeExtractF64x4(EmitContext& ctx);
void EmitCompositeInsertF64x2(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeInsertF64x3(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitCompositeInsertF64x4(EmitContext& ctx, std::string_view composite, std::string_view object,
u32 index);
void EmitSelectU1(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitSelectU8(EmitContext& ctx, std::string_view cond, std::string_view true_value,
std::string_view false_value);
void EmitSelectU16(EmitContext& ctx, std::string_view cond, std::string_view true_value,
std::string_view false_value);
void EmitSelectU32(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitSelectU64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitSelectF16(EmitContext& ctx, std::string_view cond, std::string_view true_value,
std::string_view false_value);
void EmitSelectF32(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitSelectF64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value);
void EmitBitCastU16F16(EmitContext& ctx, IR::Inst& inst);
void EmitBitCastU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitCastF16U16(EmitContext& ctx, IR::Inst& inst);
void EmitBitCastF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitCastF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitPackUint2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitUnpackUint2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitPackFloat2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitUnpackFloat2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitPackHalf2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitUnpackHalf2x16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitPackDouble2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitUnpackDouble2x32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitGetZeroFromOp(EmitContext& ctx);
void EmitGetSignFromOp(EmitContext& ctx);
void EmitGetCarryFromOp(EmitContext& ctx);
void EmitGetOverflowFromOp(EmitContext& ctx);
void EmitGetSparseFromOp(EmitContext& ctx);
void EmitGetInBoundsFromOp(EmitContext& ctx);
void EmitFPAbs16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPAbs64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPAdd16(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPFma16(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c);
void EmitFPFma32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c);
void EmitFPFma64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
std::string_view c);
void EmitFPMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMax64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMin64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMul16(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPMul64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitFPNeg16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPNeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPNeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSin(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPCos(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPExp2(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPLog2(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRecip32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRecip64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRecipSqrt32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRecipSqrt64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSqrt(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSaturate16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSaturate32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPSaturate64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPClamp16(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value);
void EmitFPClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value);
void EmitFPClamp64(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view min_value, std::string_view max_value);
void EmitFPRoundEven16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRoundEven32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPRoundEven64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPFloor16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPFloor32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPFloor64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPCeil16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPCeil32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPCeil64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPTrunc16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPTrunc32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPTrunc64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPOrdEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitFPOrdEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitFPUnordEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdNotEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdNotEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdNotEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordNotEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordNotEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordNotEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdLessThan16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdLessThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdLessThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordLessThan16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordLessThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordLessThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdGreaterThan16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdGreaterThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdGreaterThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordGreaterThan16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordGreaterThan32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordGreaterThan64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdLessThanEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdLessThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdLessThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordLessThanEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordLessThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordLessThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdGreaterThanEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPOrdGreaterThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPOrdGreaterThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordGreaterThanEqual16(EmitContext& ctx, std::string_view lhs, std::string_view rhs);
void EmitFPUnordGreaterThanEqual32(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPUnordGreaterThanEqual64(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitFPIsNan16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPIsNan32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFPIsNan64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitIAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitISub32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitISub64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitIMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitSDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitUDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitINeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitINeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitIAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitIAbs64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitShiftLeftLogical32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftLeftLogical64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftRightLogical32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftRightLogical64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftRightArithmetic32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitShiftRightArithmetic64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift);
void EmitBitwiseAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitBitwiseOr32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitBitwiseXor32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitBitFieldInsert(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view insert, std::string_view offset, std::string_view count);
void EmitBitFieldSExtract(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view offset, std::string_view count);
void EmitBitFieldUExtract(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view offset, std::string_view count);
void EmitBitReverse32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitCount32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBitwiseNot32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFindSMsb32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitFindUMsb32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitSClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view min,
std::string_view max);
void EmitUClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view min,
std::string_view max);
void EmitSLessThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitULessThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitIEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitSLessThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitULessThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitSGreaterThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitUGreaterThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitINotEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs);
void EmitSGreaterThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitUGreaterThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs);
void EmitSharedAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicInc32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicDec32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicOr32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicXor32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicExchange32(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicExchange64(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitSharedAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, std::string_view pointer_offset,
std::string_view value);
void EmitStorageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicInc32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicDec32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAnd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicOr32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicXor32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicExchange32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicIAdd64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMin64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMin64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMax64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMax64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAnd64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicOr64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicXor64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicExchange64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicIAdd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMin32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicSMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicUMax32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAnd32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicOr32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicXor32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicExchange32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAddF32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAddF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicAddF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicMinF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicMinF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicMaxF16x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitStorageAtomicMaxF32x2(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset, std::string_view value);
void EmitGlobalAtomicIAdd32(EmitContext& ctx);
void EmitGlobalAtomicSMin32(EmitContext& ctx);
void EmitGlobalAtomicUMin32(EmitContext& ctx);
void EmitGlobalAtomicSMax32(EmitContext& ctx);
void EmitGlobalAtomicUMax32(EmitContext& ctx);
void EmitGlobalAtomicInc32(EmitContext& ctx);
void EmitGlobalAtomicDec32(EmitContext& ctx);
void EmitGlobalAtomicAnd32(EmitContext& ctx);
void EmitGlobalAtomicOr32(EmitContext& ctx);
void EmitGlobalAtomicXor32(EmitContext& ctx);
void EmitGlobalAtomicExchange32(EmitContext& ctx);
void EmitGlobalAtomicIAdd64(EmitContext& ctx);
void EmitGlobalAtomicSMin64(EmitContext& ctx);
void EmitGlobalAtomicUMin64(EmitContext& ctx);
void EmitGlobalAtomicSMax64(EmitContext& ctx);
void EmitGlobalAtomicUMax64(EmitContext& ctx);
void EmitGlobalAtomicInc64(EmitContext& ctx);
void EmitGlobalAtomicDec64(EmitContext& ctx);
void EmitGlobalAtomicAnd64(EmitContext& ctx);
void EmitGlobalAtomicOr64(EmitContext& ctx);
void EmitGlobalAtomicXor64(EmitContext& ctx);
void EmitGlobalAtomicExchange64(EmitContext& ctx);
void EmitGlobalAtomicIAdd32x2(EmitContext& ctx);
void EmitGlobalAtomicSMin32x2(EmitContext& ctx);
void EmitGlobalAtomicUMin32x2(EmitContext& ctx);
void EmitGlobalAtomicSMax32x2(EmitContext& ctx);
void EmitGlobalAtomicUMax32x2(EmitContext& ctx);
void EmitGlobalAtomicInc32x2(EmitContext& ctx);
void EmitGlobalAtomicDec32x2(EmitContext& ctx);
void EmitGlobalAtomicAnd32x2(EmitContext& ctx);
void EmitGlobalAtomicOr32x2(EmitContext& ctx);
void EmitGlobalAtomicXor32x2(EmitContext& ctx);
void EmitGlobalAtomicExchange32x2(EmitContext& ctx);
void EmitGlobalAtomicAddF32(EmitContext& ctx);
void EmitGlobalAtomicAddF16x2(EmitContext& ctx);
void EmitGlobalAtomicAddF32x2(EmitContext& ctx);
void EmitGlobalAtomicMinF16x2(EmitContext& ctx);
void EmitGlobalAtomicMinF32x2(EmitContext& ctx);
void EmitGlobalAtomicMaxF16x2(EmitContext& ctx);
void EmitGlobalAtomicMaxF32x2(EmitContext& ctx);
void EmitLogicalOr(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitLogicalAnd(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitLogicalXor(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b);
void EmitLogicalNot(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS16F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS16F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS16F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS64F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU16F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU16F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU16F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU64F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU64U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32F16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16S8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16S16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16S32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16S64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16U8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16U16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF16U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32S8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32S16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32S32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32S64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32U8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32U16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF32U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64S8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64S16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64S32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64S64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64U8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64U16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU16U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32U16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU8U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertU32U8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32S8(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitConvertS32S16(EmitContext& ctx, IR::Inst& inst, std::string_view value);
void EmitBindlessImageSampleImplicitLod(EmitContext&);
void EmitBindlessImageSampleExplicitLod(EmitContext&);
void EmitBindlessImageSampleDrefImplicitLod(EmitContext&);
void EmitBindlessImageSampleDrefExplicitLod(EmitContext&);
void EmitBindlessImageGather(EmitContext&);
void EmitBindlessImageGatherDref(EmitContext&);
void EmitBindlessImageFetch(EmitContext&);
void EmitBindlessImageQueryDimensions(EmitContext&);
void EmitBindlessImageQueryLod(EmitContext&);
void EmitBindlessImageGradient(EmitContext&);
void EmitBindlessImageRead(EmitContext&);
void EmitBindlessImageWrite(EmitContext&);
void EmitBoundImageSampleImplicitLod(EmitContext&);
void EmitBoundImageSampleExplicitLod(EmitContext&);
void EmitBoundImageSampleDrefImplicitLod(EmitContext&);
void EmitBoundImageSampleDrefExplicitLod(EmitContext&);
void EmitBoundImageGather(EmitContext&);
void EmitBoundImageGatherDref(EmitContext&);
void EmitBoundImageFetch(EmitContext&);
void EmitBoundImageQueryDimensions(EmitContext&);
void EmitBoundImageQueryLod(EmitContext&);
void EmitBoundImageGradient(EmitContext&);
void EmitBoundImageRead(EmitContext&);
void EmitBoundImageWrite(EmitContext&);
void EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view bias_lc,
const IR::Value& offset);
void EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view lod_lc,
const IR::Value& offset);
void EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view dref,
std::string_view bias_lc, const IR::Value& offset);
void EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view dref,
std::string_view lod_lc, const IR::Value& offset);
void EmitImageGather(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, const IR::Value& offset2);
void EmitImageGatherDref(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, const IR::Value& offset2,
std::string_view dref);
void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& offset, std::string_view lod,
std::string_view ms);
void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view lod, const IR::Value& skip_mips);
void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords);
void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, const IR::Value& derivatives,
const IR::Value& offset, const IR::Value& lod_clamp);
void EmitImageRead(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords);
void EmitImageWrite(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view color);
void EmitIsTextureScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index);
void EmitIsImageScaled(EmitContext& ctx, IR::Inst& inst, const IR::Value& index);
void EmitBindlessImageAtomicIAdd32(EmitContext&);
void EmitBindlessImageAtomicSMin32(EmitContext&);
void EmitBindlessImageAtomicUMin32(EmitContext&);
void EmitBindlessImageAtomicSMax32(EmitContext&);
void EmitBindlessImageAtomicUMax32(EmitContext&);
void EmitBindlessImageAtomicInc32(EmitContext&);
void EmitBindlessImageAtomicDec32(EmitContext&);
void EmitBindlessImageAtomicAnd32(EmitContext&);
void EmitBindlessImageAtomicOr32(EmitContext&);
void EmitBindlessImageAtomicXor32(EmitContext&);
void EmitBindlessImageAtomicExchange32(EmitContext&);
void EmitBoundImageAtomicIAdd32(EmitContext&);
void EmitBoundImageAtomicSMin32(EmitContext&);
void EmitBoundImageAtomicUMin32(EmitContext&);
void EmitBoundImageAtomicSMax32(EmitContext&);
void EmitBoundImageAtomicUMax32(EmitContext&);
void EmitBoundImageAtomicInc32(EmitContext&);
void EmitBoundImageAtomicDec32(EmitContext&);
void EmitBoundImageAtomicAnd32(EmitContext&);
void EmitBoundImageAtomicOr32(EmitContext&);
void EmitBoundImageAtomicXor32(EmitContext&);
void EmitBoundImageAtomicExchange32(EmitContext&);
void EmitImageAtomicIAdd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicSMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicUMin32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicSMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicUMax32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicInc32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicDec32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicAnd32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicOr32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicXor32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitImageAtomicExchange32(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view value);
void EmitLaneId(EmitContext& ctx, IR::Inst& inst);
void EmitVoteAll(EmitContext& ctx, IR::Inst& inst, std::string_view pred);
void EmitVoteAny(EmitContext& ctx, IR::Inst& inst, std::string_view pred);
void EmitVoteEqual(EmitContext& ctx, IR::Inst& inst, std::string_view pred);
void EmitSubgroupBallot(EmitContext& ctx, IR::Inst& inst, std::string_view pred);
void EmitSubgroupEqMask(EmitContext& ctx, IR::Inst& inst);
void EmitSubgroupLtMask(EmitContext& ctx, IR::Inst& inst);
void EmitSubgroupLeMask(EmitContext& ctx, IR::Inst& inst);
void EmitSubgroupGtMask(EmitContext& ctx, IR::Inst& inst);
void EmitSubgroupGeMask(EmitContext& ctx, IR::Inst& inst);
void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask);
void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index,
std::string_view clamp, std::string_view segmentation_mask);
void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask);
void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask);
void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
std::string_view swizzle);
void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);
void EmitDPdyFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);
void EmitDPdxCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);
void EmitDPdyCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,268 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
namespace {
void SetZeroFlag(EmitContext& ctx, IR::Inst& inst, std::string_view result) {
IR::Inst* const zero{inst.GetAssociatedPseudoOperation(IR::Opcode::GetZeroFromOp)};
if (!zero) {
return;
}
ctx.AddU1("{}={}==0;", *zero, result);
zero->Invalidate();
}
void SetSignFlag(EmitContext& ctx, IR::Inst& inst, std::string_view result) {
IR::Inst* const sign{inst.GetAssociatedPseudoOperation(IR::Opcode::GetSignFromOp)};
if (!sign) {
return;
}
ctx.AddU1("{}=i32({})<0;", *sign, result);
sign->Invalidate();
}
void BitwiseLogicalOp(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b,
char lop) {
const auto result{ctx.var_alloc.Define(inst, WgslVarType::U32)};
ctx.Add("{}={}{}{};", result, a, lop, b);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
} // Anonymous namespace
void EmitIAdd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
// Compute the overflow CC first as it requires the original operand values,
// which may be overwritten by the result of the addition
if (IR::Inst * overflow{inst.GetAssociatedPseudoOperation(IR::Opcode::GetOverflowFromOp)}) {
// https://stackoverflow.com/questions/55468823/how-to-detect-integer-overflow-in-c
constexpr u32 s32_max{static_cast<u32>((std::numeric_limits<s32>::max)())};
const auto sub_a{fmt::format("{}u-{}", s32_max, a)};
const auto positive_result{fmt::format("i32({})>i32({})", b, sub_a)};
const auto negative_result{fmt::format("i32({})<i32({})", b, sub_a)};
ctx.AddU1("{}=select({},{},i32({})>=0);", *overflow, negative_result, positive_result, a);
overflow->Invalidate();
}
const auto result{ctx.var_alloc.Define(inst, WgslVarType::U32)};
if (IR::Inst* const carry{inst.GetAssociatedPseudoOperation(IR::Opcode::GetCarryFromOp)}) {
ctx.uses_cc_carry = true;
// the carry first: the result may share a variable with an operand
ctx.Add("carry=select(0u,1u,({}+{})<{});", a, b, a);
ctx.Add("{}={}+{};", result, a, b);
ctx.AddU1("{}=carry!=0;", *carry);
carry->Invalidate();
} else {
ctx.Add("{}={}+{};", result, a, b);
}
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitIAdd64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU64("{}=add64({},{});", inst, a, b);
}
void EmitISub32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}={}-{};", inst, a, b);
}
void EmitISub64(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU64("{}=sub64({},{});", inst, a, b);
}
void EmitIMul32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=u32({}*{});", inst, a, b);
}
void EmitSDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=u32(i32({})/i32({}));", inst, a, b);
}
void EmitUDiv32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}={}/{};", inst, a, b);
}
void EmitINeg32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=u32(i32(0)-i32({}));", inst, value);
}
void EmitINeg64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU64("{}=sub64(vec2<u32>(0u),{});", inst, value);
}
void EmitIAbs32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=u32(abs(i32({})));", inst, value);
}
void EmitIAbs64(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
// TODO: Uhm, are you sure? This may crash on some drivers!
NotImplemented();
}
void EmitShiftLeftLogical32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU32("{}={}<<{};", inst, base, shift);
}
void EmitShiftLeftLogical64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU64("{}=shl64({},{});", inst, base, shift);
}
void EmitShiftRightLogical32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU32("{}={}>>{};", inst, base, shift);
}
void EmitShiftRightLogical64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU64("{}=shr64({},{});", inst, base, shift);
}
void EmitShiftRightArithmetic32(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU32("{}=u32(i32({})>>{});", inst, base, shift);
}
void EmitShiftRightArithmetic64(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view shift) {
ctx.AddU64("{}=sar64({},{});", inst, base, shift);
}
void EmitBitwiseAnd32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
BitwiseLogicalOp(ctx, inst, a, b, '&');
}
void EmitBitwiseOr32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
BitwiseLogicalOp(ctx, inst, a, b, '|');
}
void EmitBitwiseXor32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
BitwiseLogicalOp(ctx, inst, a, b, '^');
}
void EmitBitFieldInsert(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view insert, std::string_view offset, std::string_view count) {
ctx.AddU32("{}=insertBits({},{},u32({}),u32({}));", inst, base, insert, offset, count);
}
void EmitBitFieldSExtract(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view offset, std::string_view count) {
const auto result{ctx.var_alloc.Define(inst, WgslVarType::U32)};
ctx.Add("{}=u32(extractBits(i32({}),u32({}),u32({})));", result, base, offset, count);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitBitFieldUExtract(EmitContext& ctx, IR::Inst& inst, std::string_view base,
std::string_view offset, std::string_view count) {
const auto result{ctx.var_alloc.Define(inst, WgslVarType::U32)};
ctx.Add("{}=extractBits(u32({}),u32({}),u32({}));", result, base, offset, count);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitBitReverse32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=reverseBits({});", inst, value);
}
void EmitBitCount32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=countOneBits({});", inst, value);
}
void EmitBitwiseNot32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=~{};", inst, value);
}
void EmitFindSMsb32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=u32(firstLeadingBit(i32({})));", inst, value);
}
void EmitFindUMsb32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU32("{}=firstLeadingBit(u32({}));", inst, value);
}
void EmitSMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=u32(min(i32({}),i32({})));", inst, a, b);
}
void EmitUMin32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=min(u32({}),u32({}));", inst, a, b);
}
void EmitSMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=u32(max(i32({}),i32({})));", inst, a, b);
}
void EmitUMax32(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU32("{}=max(u32({}),u32({}));", inst, a, b);
}
void EmitSClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view min,
std::string_view max) {
const auto result{ctx.var_alloc.Define(inst, WgslVarType::U32)};
ctx.Add("{}=u32(clamp(i32({}),i32({}),i32({})));", result, value, min, max);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitUClamp32(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view min,
std::string_view max) {
const auto result{ctx.var_alloc.Define(inst, WgslVarType::U32)};
ctx.Add("{}=clamp(u32({}),u32({}),u32({}));", result, value, min, max);
SetZeroFlag(ctx, inst, result);
SetSignFlag(ctx, inst, result);
}
void EmitSLessThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs) {
ctx.AddU1("{}=i32({})<i32({});", inst, lhs, rhs);
}
void EmitULessThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs) {
ctx.AddU1("{}=u32({})<u32({});", inst, lhs, rhs);
}
void EmitIEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs) {
ctx.AddU1("{}={}=={};", inst, lhs, rhs);
}
void EmitSLessThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=i32({})<=i32({});", inst, lhs, rhs);
}
void EmitULessThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=u32({})<=u32({});", inst, lhs, rhs);
}
void EmitSGreaterThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=i32({})>i32({});", inst, lhs, rhs);
}
void EmitUGreaterThan(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=u32({})>u32({});", inst, lhs, rhs);
}
void EmitINotEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs, std::string_view rhs) {
ctx.AddU1("{}={}!={};", inst, lhs, rhs);
}
void EmitSGreaterThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=i32({})>=i32({});", inst, lhs, rhs);
}
void EmitUGreaterThanEqual(EmitContext& ctx, IR::Inst& inst, std::string_view lhs,
std::string_view rhs) {
ctx.AddU1("{}=u32({})>=u32({});", inst, lhs, rhs);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,25 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
void EmitLogicalOr(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU1("{}={}||{};", inst, a, b);
}
void EmitLogicalAnd(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU1("{}={}&&{};", inst, a, b);
}
void EmitLogicalXor(EmitContext& ctx, IR::Inst& inst, std::string_view a, std::string_view b) {
ctx.AddU1("{}={}!={};", inst, a, b);
}
void EmitLogicalNot(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
ctx.AddU1("{}=!{};", inst, value);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,223 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::WGSL {
namespace {
// a byte or a short: read-modify-write of its word
void SsboWriteCas(EmitContext& ctx, const IR::Value& binding, std::string_view offset_var,
std::string_view value, std::string_view bit_offset, u32 num_bits) {
const auto word{fmt::format("{}>>2", offset_var)};
const auto merged{fmt::format("insertBits({},{},u32({}),{}u)", ctx.SsboLoad(binding.U32(), word), value,
bit_offset, num_bits)};
ctx.Add("{}", ctx.SsboStore(binding.U32(), word, merged));
}
std::string Word(std::string_view offset_var, u32 plus = 0) {
return plus ? fmt::format("({}+{})>>2", offset_var, plus) : fmt::format("{}>>2", offset_var);
}
} // Anonymous namespace
void EmitLoadGlobalU8(EmitContext&) {
NotImplemented();
}
void EmitLoadGlobalS8(EmitContext&) {
NotImplemented();
}
void EmitLoadGlobalU16(EmitContext&) {
NotImplemented();
}
void EmitLoadGlobalS16(EmitContext&) {
NotImplemented();
}
void EmitLoadGlobal32(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
if (ctx.profile.support_int64) {
return ctx.AddU32("{}=LoadGlobal32({});", inst, address);
}
LOG_WARNING(Shader_WGSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32("{}=0u;", inst);
}
void EmitLoadGlobal64(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
if (ctx.profile.support_int64) {
return ctx.AddU32x2("{}=LoadGlobal64({});", inst, address);
}
LOG_WARNING(Shader_WGSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32x2("{}=vec2<u32>(0);", inst);
}
void EmitLoadGlobal128(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
if (ctx.profile.support_int64) {
return ctx.AddU32x4("{}=LoadGlobal128({});", inst, address);
}
LOG_WARNING(Shader_WGSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32x4("{}=vec4<u32>(0);", inst);
}
void EmitWriteGlobalU8(EmitContext&) {
NotImplemented();
}
void EmitWriteGlobalS8(EmitContext&) {
NotImplemented();
}
void EmitWriteGlobalU16(EmitContext&) {
NotImplemented();
}
void EmitWriteGlobalS16(EmitContext&) {
NotImplemented();
}
void EmitWriteGlobal32(EmitContext& ctx, std::string_view address, std::string_view value) {
if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal32({},{});", address, value);
}
LOG_WARNING(Shader_WGSL, "Int64 not supported, ignoring memory operation");
}
void EmitWriteGlobal64(EmitContext& ctx, std::string_view address, std::string_view value) {
if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal64({},{});", address, value);
}
LOG_WARNING(Shader_WGSL, "Int64 not supported, ignoring memory operation");
}
void EmitWriteGlobal128(EmitContext& ctx, std::string_view address, std::string_view value) {
if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal128({},{});", address, value);
}
LOG_WARNING(Shader_WGSL, "Int64 not supported, ignoring memory operation");
}
void EmitLoadStorageU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}=extractBits({},u32({}%4)*8u,8u);", inst, ctx.SsboLoad(binding.U32(), Word(offset_var)),
offset_var);
}
void EmitLoadStorageS8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}=u32(extractBits(i32({}),u32({}%4)*8u,8u));", inst, ctx.SsboLoad(binding.U32(), Word(offset_var)),
offset_var);
}
void EmitLoadStorageU16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}=extractBits({},u32(({}>>1)%2)*16u,16u);", inst, ctx.SsboLoad(binding.U32(), Word(offset_var)),
offset_var);
}
void EmitLoadStorageS16(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}=u32(extractBits(i32({}),u32(({}>>1)%2)*16u,16u));", inst,
ctx.SsboLoad(binding.U32(), Word(offset_var)), offset_var);
}
void EmitLoadStorage32(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32("{}={};", inst, ctx.SsboLoad(binding.U32(), Word(offset_var)));
}
void EmitLoadStorage64(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32x2("{}=vec2<u32>({},{});", inst, ctx.SsboLoad(binding.U32(), Word(offset_var)),
ctx.SsboLoad(binding.U32(), Word(offset_var, 4)));
}
void EmitLoadStorage128(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,
const IR::Value& offset) {
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.AddU32x4("{}=vec4<u32>({},{},{},{});", inst, ctx.SsboLoad(binding.U32(), Word(offset_var)),
ctx.SsboLoad(binding.U32(), Word(offset_var, 4)), ctx.SsboLoad(binding.U32(), Word(offset_var, 8)),
ctx.SsboLoad(binding.U32(), Word(offset_var, 12)));
}
void EmitWriteStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
if ((ctx.stage == Stage::VertexA || ctx.stage == Stage::VertexB) && !ctx.vs_compute) {
return; // WebGPU has no writable storage in vertex shaders
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("i32({}%4)*8", offset_var)};
SsboWriteCas(ctx, binding, offset_var, value, bit_offset, 8);
}
void EmitWriteStorageS8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
if ((ctx.stage == Stage::VertexA || ctx.stage == Stage::VertexB) && !ctx.vs_compute) {
return; // WebGPU has no writable storage in vertex shaders
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("i32({}%4)*8", offset_var)};
SsboWriteCas(ctx, binding, offset_var, value, bit_offset, 8);
}
void EmitWriteStorageU16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
if ((ctx.stage == Stage::VertexA || ctx.stage == Stage::VertexB) && !ctx.vs_compute) {
return; // WebGPU has no writable storage in vertex shaders
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("i32(({}>>1)%2)*16", offset_var)};
SsboWriteCas(ctx, binding, offset_var, value, bit_offset, 16);
}
void EmitWriteStorageS16(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
if ((ctx.stage == Stage::VertexA || ctx.stage == Stage::VertexB) && !ctx.vs_compute) {
return; // WebGPU has no writable storage in vertex shaders
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
const auto bit_offset{fmt::format("i32(({}>>1)%2)*16", offset_var)};
SsboWriteCas(ctx, binding, offset_var, value, bit_offset, 16);
}
void EmitWriteStorage32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
if ((ctx.stage == Stage::VertexA || ctx.stage == Stage::VertexB) && !ctx.vs_compute) {
return; // WebGPU has no writable storage in vertex shaders
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.Add("{}", ctx.SsboStore(binding.U32(), Word(offset_var), value));
}
void EmitWriteStorage64(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
if ((ctx.stage == Stage::VertexA || ctx.stage == Stage::VertexB) && !ctx.vs_compute) {
return; // WebGPU has no writable storage in vertex shaders
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.Add("{}", ctx.SsboStore(binding.U32(), Word(offset_var), fmt::format("{}.x", value)));
ctx.Add("{}", ctx.SsboStore(binding.U32(), Word(offset_var, 4), fmt::format("{}.y", value)));
}
void EmitWriteStorage128(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
std::string_view value) {
if ((ctx.stage == Stage::VertexA || ctx.stage == Stage::VertexB) && !ctx.vs_compute) {
return; // WebGPU has no writable storage in vertex shaders
}
const auto offset_var{ctx.var_alloc.Consume(offset)};
ctx.Add("{}", ctx.SsboStore(binding.U32(), Word(offset_var), fmt::format("{}.x", value)));
ctx.Add("{}", ctx.SsboStore(binding.U32(), Word(offset_var, 4), fmt::format("{}.y", value)));
ctx.Add("{}", ctx.SsboStore(binding.U32(), Word(offset_var, 8), fmt::format("{}.z", value)));
ctx.Add("{}", ctx.SsboStore(binding.U32(), Word(offset_var, 12), fmt::format("{}.w", value)));
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,100 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
void EmitGetRegister(EmitContext& ctx) {
NotImplemented();
}
void EmitSetRegister(EmitContext& ctx) {
NotImplemented();
}
void EmitGetPred(EmitContext& ctx) {
NotImplemented();
}
void EmitSetPred(EmitContext& ctx) {
NotImplemented();
}
void EmitSetGotoVariable(EmitContext& ctx) {
NotImplemented();
}
void EmitGetGotoVariable(EmitContext& ctx) {
NotImplemented();
}
void EmitSetIndirectBranchVariable(EmitContext& ctx) {
NotImplemented();
}
void EmitGetIndirectBranchVariable(EmitContext& ctx) {
NotImplemented();
}
void EmitGetZFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitGetSFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitGetCFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitGetOFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitSetZFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitSetSFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitSetCFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitSetOFlag(EmitContext& ctx) {
NotImplemented();
}
void EmitGetZeroFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetSignFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetCarryFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetOverflowFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetSparseFromOp(EmitContext& ctx) {
NotImplemented();
}
void EmitGetInBoundsFromOp(EmitContext& ctx) {
NotImplemented();
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,54 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
void EmitSelectU1(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddU1("{}=select({},{},{});", inst, false_value, true_value, cond);
}
void EmitSelectU8([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view cond,
[[maybe_unused]] std::string_view true_value,
[[maybe_unused]] std::string_view false_value) {
NotImplemented();
}
void EmitSelectU16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view cond,
[[maybe_unused]] std::string_view true_value,
[[maybe_unused]] std::string_view false_value) {
NotImplemented();
}
void EmitSelectU32(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddU32("{}=select({},{},{});", inst, false_value, true_value, cond);
}
void EmitSelectU64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddU64("{}=select({},{},{});", inst, false_value, true_value, cond);
}
void EmitSelectF16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] std::string_view cond,
[[maybe_unused]] std::string_view true_value,
[[maybe_unused]] std::string_view false_value) {
NotImplemented();
}
void EmitSelectF32(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddF32("{}=select({},{},{});", inst, false_value, true_value, cond);
}
void EmitSelectF64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
std::string_view true_value, std::string_view false_value) {
ctx.AddF64("{}=select({},{},{});", inst, false_value, true_value, cond);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,78 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
namespace {
// a byte or a short: read-modify-write of its word
void SharedWriteCas(EmitContext& ctx, std::string_view offset, std::string_view value,
std::string_view bit_offset, u32 num_bits) {
const auto word{fmt::format("{}>>2", offset)};
ctx.Add("{}", ctx.SmemStore(word, fmt::format("insertBits({},{},u32({}),{}u)", ctx.SmemLoad(word), value,
bit_offset, num_bits)));
}
std::string Word(std::string_view offset, u32 plus = 0) {
return plus ? fmt::format("({}+{})>>2", offset, plus) : fmt::format("{}>>2", offset);
}
} // Anonymous namespace
void EmitLoadSharedU8(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=extractBits({},u32({}%4)*8u,8u);", inst, ctx.SmemLoad(Word(offset)), offset);
}
void EmitLoadSharedS8(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=u32(extractBits(i32({}),u32({}%4)*8u,8u));", inst, ctx.SmemLoad(Word(offset)), offset);
}
void EmitLoadSharedU16(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=extractBits({},u32(({}>>1)%2)*16u,16u);", inst, ctx.SmemLoad(Word(offset)), offset);
}
void EmitLoadSharedS16(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}=u32(extractBits(i32({}),u32(({}>>1)%2)*16u,16u));", inst, ctx.SmemLoad(Word(offset)), offset);
}
void EmitLoadSharedU32(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32("{}={};", inst, ctx.SmemLoad(Word(offset)));
}
void EmitLoadSharedU64(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32x2("{}=vec2<u32>({},{});", inst, ctx.SmemLoad(Word(offset)), ctx.SmemLoad(Word(offset, 4)));
}
void EmitLoadSharedU128(EmitContext& ctx, IR::Inst& inst, std::string_view offset) {
ctx.AddU32x4("{}=vec4<u32>({},{},{},{});", inst, ctx.SmemLoad(Word(offset)), ctx.SmemLoad(Word(offset, 4)),
ctx.SmemLoad(Word(offset, 8)), ctx.SmemLoad(Word(offset, 12)));
}
void EmitWriteSharedU8(EmitContext& ctx, std::string_view offset, std::string_view value) {
const auto bit_offset{fmt::format("i32({}%4)*8", offset)};
SharedWriteCas(ctx, offset, value, bit_offset, 8);
}
void EmitWriteSharedU16(EmitContext& ctx, std::string_view offset, std::string_view value) {
const auto bit_offset{fmt::format("i32(({}>>1)%2)*16", offset)};
SharedWriteCas(ctx, offset, value, bit_offset, 16);
}
void EmitWriteSharedU32(EmitContext& ctx, std::string_view offset, std::string_view value) {
ctx.Add("{}", ctx.SmemStore(Word(offset), value));
}
void EmitWriteSharedU64(EmitContext& ctx, std::string_view offset, std::string_view value) {
ctx.Add("{}", ctx.SmemStore(Word(offset), fmt::format("{}.x", value)));
ctx.Add("{}", ctx.SmemStore(Word(offset, 4), fmt::format("{}.y", value)));
}
void EmitWriteSharedU128(EmitContext& ctx, std::string_view offset, std::string_view value) {
ctx.Add("{}", ctx.SmemStore(Word(offset), fmt::format("{}.x", value)));
ctx.Add("{}", ctx.SmemStore(Word(offset, 4), fmt::format("{}.y", value)));
ctx.Add("{}", ctx.SmemStore(Word(offset, 8), fmt::format("{}.z", value)));
ctx.Add("{}", ctx.SmemStore(Word(offset, 12), fmt::format("{}.w", value)));
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,109 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::WGSL {
namespace {
std::string_view OutputVertexIndex(EmitContext& ctx) {
return ctx.stage == Stage::TessellationControl ? "[gl_InvocationID]" : "";
}
void InitializeOutputVaryings(EmitContext& ctx) {
if (ctx.uses_geometry_passthrough) {
return;
}
if (ctx.stage == Stage::VertexB || ctx.stage == Stage::Geometry) {
ctx.Add("gl_Position=vec4<f32>(0,0,0,1);");
}
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (!ctx.info.stores.Generic(index)) {
continue;
}
const auto& info_array{ctx.output_generics.at(index)};
const auto output_decorator{OutputVertexIndex(ctx)};
size_t element{};
while (element < info_array.size()) {
const auto& info{info_array.at(element)};
const auto varying_name{fmt::format("{}{}", info.name, output_decorator)};
switch (info.num_components) {
case 1: {
ctx.Add("{}={}f;", varying_name, element == 3 ? "1.0" : "0.0");
break;
}
case 2:
case 3:
if (element + info.num_components < 4) {
ctx.Add("{}=vec{}<f32>(0);", varying_name, info.num_components);
} else {
// last element is the w component, must be initialized to 1
const auto zeros{info.num_components == 3 ? "0,0," : "0,"};
ctx.Add("{}=vec{}<f32>({}1);", varying_name, info.num_components, zeros);
}
break;
case 4:
ctx.Add("{}=vec4<f32>(0,0,0,1);", varying_name);
break;
default:
break;
}
element += info.num_components;
}
}
}
} // Anonymous namespace
void EmitPhi(EmitContext& ctx, IR::Inst& phi) {
const size_t num_args{phi.NumArgs()};
for (size_t i = 0; i < num_args; ++i) {
ctx.var_alloc.Consume(phi.Arg(i));
}
if (!phi.Definition<Id>().is_valid) {
// The phi node wasn't forward defined
ctx.var_alloc.PhiDefine(phi, phi.Type());
}
}
void EmitVoid(EmitContext&) {}
void EmitReference(EmitContext& ctx, const IR::Value& value) {
ctx.var_alloc.Consume(value);
}
void EmitPhiMove(EmitContext& ctx, const IR::Value& phi_value, const IR::Value& value) {
IR::Inst& phi{*phi_value.InstRecursive()};
const auto phi_type{phi.Type()};
if (!phi.Definition<Id>().is_valid) {
// The phi node wasn't forward defined
ctx.var_alloc.PhiDefine(phi, phi_type);
}
const auto phi_reg{ctx.var_alloc.Consume(IR::Value{&phi})};
const auto val_reg{ctx.var_alloc.Consume(value)};
if (phi_reg == val_reg) {
return;
}
const bool needs_workaround{ctx.profile.has_gl_bool_ref_bug && phi_type == IR::Type::U1};
const auto suffix{needs_workaround ? "?true:false" : ""};
ctx.Add("{}={}{};", phi_reg, val_reg, suffix);
}
void EmitPrologue(EmitContext& ctx) {
InitializeOutputVaryings(ctx);
}
void EmitEpilogue(EmitContext&) {}
void EmitEmitVertex(EmitContext& ctx, const IR::Value& stream) {
ctx.Add("EmitStreamVertex(i32({}));", ctx.var_alloc.Consume(stream));
InitializeOutputVaryings(ctx);
}
void EmitEndPrimitive(EmitContext& ctx, const IR::Value& stream) {
ctx.Add("EndStreamPrimitive(i32({}));", ctx.var_alloc.Consume(stream));
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,29 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
namespace Shader::Backend::WGSL {
void EmitUndefU1(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU1("{}=false;", inst);
}
void EmitUndefU8(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=0u;", inst);
}
void EmitUndefU16(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=0u;", inst);
}
void EmitUndefU32(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=0u;", inst);
}
void EmitUndefU64(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU64("{}=vec2<u32>(0u);", inst);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,203 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string_view>
#include "shader_recompiler/backend/wgsl/emit_wgsl_instructions.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::WGSL {
namespace {
constexpr char THREAD_ID[]{"gl_SubGroupInvocationARB"};
void SetInBoundsFlag(EmitContext& ctx, IR::Inst& inst) {
IR::Inst* const in_bounds{inst.GetAssociatedPseudoOperation(IR::Opcode::GetInBoundsFromOp)};
if (!in_bounds) {
return;
}
ctx.AddU1("{}=shfl_in_bounds;", *in_bounds);
in_bounds->Invalidate();
}
std::string ComputeMinThreadId(std::string_view thread_id, std::string_view segmentation_mask) {
return fmt::format("({}&{})", thread_id, segmentation_mask);
}
std::string ComputeMaxThreadId(std::string_view min_thread_id, std::string_view clamp,
std::string_view not_seg_mask) {
return fmt::format("({})|({}&{})", min_thread_id, clamp, not_seg_mask);
}
std::string GetMaxThreadId(std::string_view thread_id, std::string_view clamp,
std::string_view segmentation_mask) {
const auto not_seg_mask{fmt::format("(~{})", segmentation_mask)};
const auto min_thread_id{ComputeMinThreadId(thread_id, segmentation_mask)};
return ComputeMaxThreadId(min_thread_id, clamp, not_seg_mask);
}
void UseShuffleNv(EmitContext& ctx, IR::Inst& inst, std::string_view shfl_op,
std::string_view value, std::string_view index,
[[maybe_unused]] std::string_view clamp, std::string_view segmentation_mask) {
const auto width{fmt::format("32u>>(countOneBits({}&31u))", segmentation_mask)};
ctx.AddU32("{}={}({},{},{},shfl_in_bounds);", inst, shfl_op, value, index, width);
SetInBoundsFlag(ctx, inst);
}
std::string_view BallotIndex(EmitContext& ctx) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
return ".x";
}
return "[gl_SubGroupInvocationARB>>5]";
}
std::string GetMask(EmitContext& ctx, std::string_view mask) {
const auto ballot_index{BallotIndex(ctx)};
return fmt::format("u32(vec2<u32>({}){})", mask, ballot_index);
}
} // Anonymous namespace
void EmitLaneId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}={}&31u;", inst, THREAD_ID);
}
void EmitVoteAll(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
ctx.AddU1("{}={};", inst, pred);
}
void EmitVoteAny(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
ctx.AddU1("{}={};", inst, pred);
}
void EmitVoteEqual(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
ctx.AddU1("{}=true;", inst);
}
void EmitSubgroupBallot(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
ctx.AddU32("{}=select(0u,1u,{});", inst, pred);
}
void EmitSubgroupEqMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=1u;", inst);
}
void EmitSubgroupLtMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=0u;", inst);
}
void EmitSubgroupLeMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=1u;", inst);
}
void EmitSubgroupGtMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=0u;", inst);
}
void EmitSubgroupGeMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=1u;", inst);
}
void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp, std::string_view seg_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleNV", value, index, clamp, seg_mask);
return;
}
const bool big_warp{ctx.profile.warp_size_potentially_larger_than_guest};
const auto is_upper_partition{"i32(gl_SubGroupInvocationARB)>=32"};
const auto upper_index{fmt::format("select({},{}+32u,{})", index, index, is_upper_partition)};
const auto upper_clamp{fmt::format("select({},{}+32u,{})", clamp, clamp, is_upper_partition)};
const auto not_seg_mask{fmt::format("(~{})", seg_mask)};
const auto min_thread_id{ComputeMinThreadId(THREAD_ID, seg_mask)};
const auto max_thread_id{
ComputeMaxThreadId(min_thread_id, big_warp ? upper_clamp : clamp, not_seg_mask)};
const auto lhs{fmt::format("({}&{})", big_warp ? upper_index : index, not_seg_mask)};
const auto src_thread_id{fmt::format("({})|({})", lhs, min_thread_id)};
ctx.Add("shfl_in_bounds=i32({})<=i32({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.Add("shfl_result={};", value);
ctx.AddU32("{}=select({},shfl_result,shfl_in_bounds);", inst, value);
}
void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index,
std::string_view clamp, std::string_view seg_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleUpNV", value, index, clamp, seg_mask);
return;
}
const bool big_warp{ctx.profile.warp_size_potentially_larger_than_guest};
const auto is_upper_partition{"i32(gl_SubGroupInvocationARB)>=32"};
const auto upper_clamp{fmt::format("select({},{}+32u,{})", clamp, clamp, is_upper_partition)};
const auto max_thread_id{GetMaxThreadId(THREAD_ID, big_warp ? upper_clamp : clamp, seg_mask)};
const auto src_thread_id{fmt::format("({}-{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=i32({})>=i32({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.Add("shfl_result={};", value);
ctx.AddU32("{}=select({},shfl_result,shfl_in_bounds);", inst, value);
}
void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp, std::string_view seg_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleDownNV", value, index, clamp, seg_mask);
return;
}
const bool big_warp{ctx.profile.warp_size_potentially_larger_than_guest};
const auto is_upper_partition{"i32(gl_SubGroupInvocationARB)>=32"};
const auto upper_clamp{fmt::format("select({},{}+32u,{})", clamp, clamp, is_upper_partition)};
const auto max_thread_id{GetMaxThreadId(THREAD_ID, big_warp ? upper_clamp : clamp, seg_mask)};
const auto src_thread_id{fmt::format("({}+{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=i32({})<=i32({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.Add("shfl_result={};", value);
ctx.AddU32("{}=select({},shfl_result,shfl_in_bounds);", inst, value);
}
void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view seg_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleXorNV", value, index, clamp, seg_mask);
return;
}
const bool big_warp{ctx.profile.warp_size_potentially_larger_than_guest};
const auto is_upper_partition{"i32(gl_SubGroupInvocationARB)>=32"};
const auto upper_clamp{fmt::format("select({},{}+32u,{})", clamp, clamp, is_upper_partition)};
const auto max_thread_id{GetMaxThreadId(THREAD_ID, big_warp ? upper_clamp : clamp, seg_mask)};
const auto src_thread_id{fmt::format("({}^{})", THREAD_ID, index)};
ctx.Add("shfl_in_bounds=i32({})<=i32({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.Add("shfl_result={};", value);
ctx.AddU32("{}=select({},shfl_result,shfl_in_bounds);", inst, value);
}
void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
std::string_view swizzle) {
const auto mask{fmt::format("({}>>((gl_SubGroupInvocationARB&3)<<1))&3", swizzle)};
const std::string modifier_a = fmt::format("FSWZ_A[{}]", mask);
const std::string modifier_b = fmt::format("FSWZ_B[{}]", mask);
ctx.AddF32("{}=({}*{})+({}*{});", inst, op_a, modifier_a, op_b, modifier_b);
}
void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
ctx.AddF32("{}=dpdxFine({});", inst, op_a);
}
void EmitDPdyFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
ctx.AddF32("{}=dpdyFine({});", inst, op_a);
}
void EmitDPdxCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
ctx.AddF32("{}=dpdxCoarse({});", inst, op_a);
}
void EmitDPdyCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
ctx.AddF32("{}=dpdyCoarse({});", inst, op_a);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,301 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string>
#include <string_view>
#include <fmt/ranges.h>
#include "shader_recompiler/backend/wgsl/var_alloc.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::Backend::WGSL {
namespace {
std::string TypePrefix(WgslVarType type) {
switch (type) {
case WgslVarType::U1:
return "b_";
case WgslVarType::F16x2:
return "f16x2_";
case WgslVarType::U32:
return "u_";
case WgslVarType::F32:
return "f_";
case WgslVarType::U64:
return "u64_";
case WgslVarType::F64:
return "d_";
case WgslVarType::U32x2:
return "u2_";
case WgslVarType::F32x2:
return "f2_";
case WgslVarType::U32x3:
return "u3_";
case WgslVarType::F32x3:
return "f3_";
case WgslVarType::U32x4:
return "u4_";
case WgslVarType::F32x4:
return "f4_";
case WgslVarType::PrecF32:
return "pf_";
case WgslVarType::PrecF64:
return "pd_";
case WgslVarType::Void:
return "";
default:
throw NotImplementedException("Type {}", type);
}
}
std::string FormatFloat(std::string_view value, IR::Type type) {
if (type != IR::Type::F32) {
throw NotImplementedException("WGSL has no 64-bit floats");
}
// WGSL const-evaluates bitcasts of literals and rejects non-finite results: read them from private variables
if (value == "nan") {
return "bitcast<f32>(F32_NAN_)";
}
if (value == "inf") {
return "bitcast<f32>(F32_INF_)";
}
if (value == "-inf") {
return "bitcast<f32>(F32_NINF_)";
}
const bool scientific{value.find_first_of('e') != std::string_view::npos};
const bool needs_dot{!scientific && value.find_first_of('.') == std::string_view::npos};
return fmt::format("{}{}f", value, needs_dot ? ".0" : "");
}
std::string MakeImm(const IR::Value& value) {
switch (value.Type()) {
case IR::Type::U1:
return value.U1() ? "true" : "false";
case IR::Type::U32:
return fmt::format("{}u", value.U32());
case IR::Type::F32:
return FormatFloat(fmt::format("{}", value.F32()), IR::Type::F32);
case IR::Type::U64:
return fmt::format("vec2<u32>({}u, {}u)", static_cast<u32>(value.U64()),
static_cast<u32>(value.U64() >> 32));
case IR::Type::Void:
return "";
default:
throw NotImplementedException("Immediate type {}", value.Type());
}
}
} // Anonymous namespace
std::string VarAlloc::Representation(u32 index, WgslVarType type) const {
const auto prefix{TypePrefix(type)};
return fmt::format("{}{}", prefix, index);
}
std::string VarAlloc::Representation(Id id) const {
return Representation(id.index, id.type);
}
std::string VarAlloc::Define(IR::Inst& inst, WgslVarType type) {
if (inst.HasUses()) {
inst.SetDefinition<Id>(Alloc(type));
return Representation(inst.Definition<Id>());
} else {
Id id{};
id.type.Assign(type);
GetUseTracker(type).uses_temp = true;
inst.SetDefinition<Id>(id);
return 't' + Representation(inst.Definition<Id>());
}
}
std::string VarAlloc::Define(IR::Inst& inst, IR::Type type) {
return Define(inst, RegType(type));
}
std::string VarAlloc::PhiDefine(IR::Inst& inst, IR::Type type) {
return AddDefine(inst, RegType(type));
}
std::string VarAlloc::AddDefine(IR::Inst& inst, WgslVarType type) {
if (inst.HasUses()) {
inst.SetDefinition<Id>(Alloc(type));
return Representation(inst.Definition<Id>());
} else {
return "";
}
}
std::string VarAlloc::Consume(const IR::Value& value) {
return value.IsImmediate() ? MakeImm(value) : ConsumeInst(*value.InstRecursive());
}
std::string VarAlloc::ConsumeInst(IR::Inst& inst) {
inst.DestructiveRemoveUsage();
if (!inst.HasUses()) {
Free(inst.Definition<Id>());
}
return Representation(inst.Definition<Id>());
}
std::string VarAlloc::GetWgslType(IR::Type type) const {
return GetWgslType(RegType(type));
}
Id VarAlloc::Alloc(WgslVarType type) {
auto& use_tracker{GetUseTracker(type)};
const auto num_vars{use_tracker.var_use.size()};
for (size_t var = 0; var < num_vars; ++var) {
if (use_tracker.var_use[var]) {
continue;
}
use_tracker.num_used = (std::max)(use_tracker.num_used, var + 1);
use_tracker.var_use[var] = true;
Id ret{};
ret.is_valid.Assign(1);
ret.type.Assign(type);
ret.index.Assign(static_cast<u32>(var));
return ret;
}
// Allocate a new variable
use_tracker.var_use.push_back(true);
Id ret{};
ret.is_valid.Assign(1);
ret.type.Assign(type);
ret.index.Assign(static_cast<u32>(use_tracker.num_used));
++use_tracker.num_used;
return ret;
}
void VarAlloc::Free(Id id) {
if (id.is_valid == 0) {
throw LogicError("Freeing invalid variable");
}
auto& use_tracker{GetUseTracker(id.type)};
use_tracker.var_use[id.index] = false;
}
WgslVarType VarAlloc::RegType(IR::Type type) const {
switch (type) {
case IR::Type::U1:
return WgslVarType::U1;
case IR::Type::U32:
return WgslVarType::U32;
case IR::Type::F32:
return WgslVarType::F32;
case IR::Type::U64:
return WgslVarType::U64;
case IR::Type::F64:
return WgslVarType::F64;
default:
throw NotImplementedException("IR type {}", type);
}
}
std::string VarAlloc::GetWgslType(WgslVarType type) const {
switch (type) {
case WgslVarType::U1:
return "bool";
case WgslVarType::F16x2: // no f16 in core WebGPU: half pairs are carried as f32 pairs
case WgslVarType::F32x2:
return "vec2<f32>";
case WgslVarType::U32:
return "u32";
case WgslVarType::F32:
case WgslVarType::PrecF32:
return "f32";
case WgslVarType::U64: // no 64-bit integers either: low and high words
case WgslVarType::U32x2:
return "vec2<u32>";
case WgslVarType::U32x3:
return "vec3<u32>";
case WgslVarType::F32x3:
return "vec3<f32>";
case WgslVarType::U32x4:
return "vec4<u32>";
case WgslVarType::F32x4:
return "vec4<f32>";
case WgslVarType::F64:
case WgslVarType::PrecF64:
throw NotImplementedException("WGSL has no 64-bit floats");
case WgslVarType::Void:
return "";
default:
throw NotImplementedException("Type {}", type);
}
}
VarAlloc::UseTracker& VarAlloc::GetUseTracker(WgslVarType type) {
switch (type) {
case WgslVarType::U1:
return var_bool;
case WgslVarType::F16x2:
return var_f16x2;
case WgslVarType::U32:
return var_u32;
case WgslVarType::F32:
return var_f32;
case WgslVarType::U64:
return var_u64;
case WgslVarType::F64:
return var_f64;
case WgslVarType::U32x2:
return var_u32x2;
case WgslVarType::F32x2:
return var_f32x2;
case WgslVarType::U32x3:
return var_u32x3;
case WgslVarType::F32x3:
return var_f32x3;
case WgslVarType::U32x4:
return var_u32x4;
case WgslVarType::F32x4:
return var_f32x4;
case WgslVarType::PrecF32:
return var_precf32;
case WgslVarType::PrecF64:
return var_precf64;
default:
throw NotImplementedException("Type {}", type);
}
}
const VarAlloc::UseTracker& VarAlloc::GetUseTracker(WgslVarType type) const {
switch (type) {
case WgslVarType::U1:
return var_bool;
case WgslVarType::F16x2:
return var_f16x2;
case WgslVarType::U32:
return var_u32;
case WgslVarType::F32:
return var_f32;
case WgslVarType::U64:
return var_u64;
case WgslVarType::F64:
return var_f64;
case WgslVarType::U32x2:
return var_u32x2;
case WgslVarType::F32x2:
return var_f32x2;
case WgslVarType::U32x3:
return var_u32x3;
case WgslVarType::F32x3:
return var_f32x3;
case WgslVarType::U32x4:
return var_u32x4;
case WgslVarType::F32x4:
return var_f32x4;
case WgslVarType::PrecF32:
return var_precf32;
case WgslVarType::PrecF64:
return var_precf64;
default:
throw NotImplementedException("Type {}", type);
}
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,104 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <bitset>
#include <string>
#include <vector>
#include "common/bit_field.h"
#include "common/common_types.h"
namespace Shader::IR {
class Inst;
class Value;
enum class Type;
} // namespace Shader::IR
namespace Shader::Backend::WGSL {
enum class WgslVarType : u32 {
U1,
F16x2,
U32,
F32,
U64,
F64,
U32x2,
F32x2,
U32x3,
F32x3,
U32x4,
F32x4,
PrecF32,
PrecF64,
Void,
};
struct Id {
union {
u32 raw;
BitField<0, 1, u32> is_valid;
BitField<1, 4, WgslVarType> type;
BitField<6, 26, u32> index;
};
bool operator==(Id rhs) const noexcept {
return raw == rhs.raw;
}
bool operator!=(Id rhs) const noexcept {
return !operator==(rhs);
}
};
static_assert(sizeof(Id) == sizeof(u32));
class VarAlloc {
public:
struct UseTracker {
bool uses_temp{};
size_t num_used{};
std::vector<bool> var_use;
};
/// Used for explicit usages of variables, may revert to temporaries
std::string Define(IR::Inst& inst, WgslVarType type);
std::string Define(IR::Inst& inst, IR::Type type);
/// Used to assign variables used by the IR. May return a blank string if
/// the instruction's result is unused in the IR.
std::string AddDefine(IR::Inst& inst, WgslVarType type);
std::string PhiDefine(IR::Inst& inst, IR::Type type);
std::string Consume(const IR::Value& value);
std::string ConsumeInst(IR::Inst& inst);
std::string GetWgslType(WgslVarType type) const;
std::string GetWgslType(IR::Type type) const;
const UseTracker& GetUseTracker(WgslVarType type) const;
std::string Representation(u32 index, WgslVarType type) const;
private:
WgslVarType RegType(IR::Type type) const;
Id Alloc(WgslVarType type);
void Free(Id id);
UseTracker& GetUseTracker(WgslVarType type);
std::string Representation(Id id) const;
UseTracker var_bool{};
UseTracker var_f16x2{};
UseTracker var_u32{};
UseTracker var_u32x2{};
UseTracker var_u32x3{};
UseTracker var_u32x4{};
UseTracker var_f32{};
UseTracker var_f32x2{};
UseTracker var_f32x3{};
UseTracker var_f32x4{};
UseTracker var_u64{};
UseTracker var_f64{};
UseTracker var_precf32{};
UseTracker var_precf64{};
};
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,482 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
//
// WGSL backend (Rexcomp): derived from suyu's GLSL backend context. Module-scope declarations for one shader stage
// of a WebGPU pipeline.
//
// Layout:
// - every resource of a stage is in one bind group: vertex 0, fragment 1, compute 0; binding numbers come from
// Bindings::unified in declaration order (the uniform block, constant buffers indexed at run time, storage
// buffers, textures (each a texture then its sampler), images)
// - binding 0, the stage's uniform block <stage>_ub: array<vec4<u32>, N>, which the renderer binds at a dynamic
// offset per draw: [0] resolution scaling, [1] render area, [2] (ax, bx, ay, by), [3] (az, bz, vertex id
// offset, instance id offset): the viewport transform the vertex stage applies (WebGPU has no negative or
// out-of-target viewports), [4 + t] the component swizzle of texture t (WebGPU views have none), then the
// constant buffers (raw words; floats are bitcast), each at its base in declaration order
// - constant buffers indexed at run time: var<storage, read> <stage>_cbuf<i>: array<vec4<u32>> of the bound size
// - storage buffers: var<storage, read|read_write> <stage>_ssbo<i>: array<u32>
// - GLSL's built-ins (gl_Position, gl_FragCoord, ...) and the stage's inputs/outputs (in_attr<i>, out_attr<i>,
// frag_color<i>) are module-scope private variables with the GLSL names, so the instruction emitters read
// like the GLSL backend's; the entry point (emit_wgsl.cpp) copies them from/to the WGSL built-ins and locations.
#include "common/div_ceil.h"
#include "shader_recompiler/backend/bindings.h"
#include "shader_recompiler/backend/wgsl/wgsl_emit_context.h"
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/profile.h"
#include "shader_recompiler/runtime_info.h"
namespace Shader::Backend::WGSL {
namespace {
std::string_view ColorTextureType(TextureType type, bool is_multisample) {
switch (type) {
case TextureType::Color1D:
return "texture_1d<f32>";
case TextureType::Color2D:
case TextureType::Color2DRect:
return is_multisample ? "texture_multisampled_2d<f32>" : "texture_2d<f32>";
case TextureType::ColorArray2D:
return "texture_2d_array<f32>";
case TextureType::Color3D:
return "texture_3d<f32>";
case TextureType::ColorCube:
return "texture_cube<f32>";
case TextureType::ColorArrayCube:
return "texture_cube_array<f32>";
default:
throw NotImplementedException("WGSL texture type {}", type);
}
}
std::string_view DepthTextureType(TextureType type) {
switch (type) {
case TextureType::Color2D:
case TextureType::Color2DRect:
return "texture_depth_2d";
case TextureType::ColorArray2D:
return "texture_depth_2d_array";
case TextureType::ColorCube:
return "texture_depth_cube";
case TextureType::ColorArrayCube:
return "texture_depth_cube_array";
default:
throw NotImplementedException("WGSL depth texture type {}", type);
}
}
std::string_view StorageTextureType(TextureType type) {
switch (type) {
case TextureType::Color1D:
return "texture_storage_1d";
case TextureType::Color2D:
return "texture_storage_2d";
case TextureType::ColorArray2D:
return "texture_storage_2d_array";
case TextureType::Color3D:
return "texture_storage_3d";
default:
throw NotImplementedException("WGSL storage texture type {}", type);
}
}
std::string_view StorageFormat(ImageFormat format) {
switch (format) {
case ImageFormat::Typeless:
case ImageFormat::R32_UINT:
return "r32uint";
case ImageFormat::R32G32_UINT:
return "rg32uint";
case ImageFormat::R32G32B32A32_UINT:
return "rgba32uint";
default:
throw NotImplementedException("WGSL storage texture format {}", format);
}
}
} // Anonymous namespace
thread_local std::array<u8, 64> ssbo_alias_next{};
thread_local std::array<u32, 64> ssbo_alias_words_next{};
thread_local bool vs_as_compute_next{};
thread_local std::array<std::string, 16> image_formats_next{};
EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile& profile_,
const RuntimeInfo& runtime_info_)
: info{program.info}, profile{profile_}, runtime_info{runtime_info_}, stage{program.stage} {
ssbo_alias = ssbo_alias_next;
ssbo_alias_next = {};
ssbo_alias_words = ssbo_alias_words_next;
ssbo_alias_words_next = {};
vs_compute = vs_as_compute_next && (stage == Stage::VertexA || stage == Stage::VertexB);
vs_as_compute_next = false;
image_formats = image_formats_next;
image_formats_next = {};
for (IR::Block* const block : program.blocks) {
for (IR::Inst& inst : block->Instructions()) {
const std::string_view name{IR::NameOf(inst.GetOpcode())};
if (name.starts_with("StorageAtomic") && inst.Arg(0).IsImmediate() && inst.Arg(0).U32() < 64) {
atomic_ssbo.set(MapSsbo(inst.Arg(0).U32()));
} else if (name.starts_with("SharedAtomic")) {
atomic_smem = true;
}
}
}
switch (program.stage) {
case Stage::VertexA:
case Stage::VertexB:
stage_name = "vs";
group = 0;
break;
case Stage::Fragment:
stage_name = "fs";
position_name = "gl_FragCoord";
group = 1;
break;
case Stage::Compute:
stage_name = "cs";
group = 0;
break;
default:
throw NotImplementedException("WebGPU has no {} shaders", program.stage);
}
header += "diagnostic(off, derivative_uniformity);\n";
DefineBuiltins();
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (info.loads.Generic(index)) {
header += fmt::format("var<private> in_attr{}: vec4<f32>;\n", index);
}
if (info.stores.Generic(index)) {
header += fmt::format("var<private> out_attr{}: vec4<f32>;\n", index);
const GenericElementInfo element_info{
.name = fmt::format("out_attr{}", index),
.first_element = 0,
.num_components = 4,
};
output_generics[index].fill(element_info);
}
}
if (stage == Stage::Fragment) {
for (size_t index = 0; index < info.stores_frag_color.size(); ++index) {
if (info.stores_frag_color[index]) {
header += fmt::format("var<private> frag_color{}: vec4<f32>;\n", index);
}
}
}
// binding 0: the uniform block (see the layout above)
ub_name = fmt::format("{}_ub", stage_name);
ub_size = 4 + 2 * static_cast<u32>(info.texture_descriptors.size());
if (!(info.uses_global_memory || info.uses_cbuf_indirect)) {
// in declaration order, while they fit (the renderer lays the block out the same way)
for (const auto& desc : info.constant_buffer_descriptors) {
const u32 size{(std::max)(Common::DivCeil(info.constant_buffer_used_sizes[desc.index], 16U), 1U)};
if (runtime_info.cbuf_ub_limit && desc.index >= 3 &&
info.constant_buffer_used_sizes[desc.index] > runtime_info.cbuf_ub_limit) {
continue; // (bound in place: copying it for every draw costs more)
}
if (ub_size + size > 0x1000) {
continue;
}
cbuf_in_ub[desc.index] = true;
ub_cbuf_base[desc.index] = ub_size;
ub_size += size;
}
}
header += fmt::format("@group({}) @binding({}) var<uniform> {}: array<vec4<u32>, {}>;\n", group,
bindings.unified++, ub_name, ub_size);
header += fmt::format("fn rescale_(i: u32) -> vec4<f32> {{ return bitcast<vec4<f32>>({}[i]); }}\n", ub_name);
// Depth for a 16-bit target, on its step: the nearest step (halves up) plus a quarter, so that rounding and
// truncating conversions agree and a value one draw writes passes another draw's EQUAL test (TOTK keeps
// material numbers in a D16 target as 0.5 + id/510: even numbers fall halfway between two steps). On when
// render_area().z is not 0 (the renderer sets it for draws into a 16-bit depth target).
header += "fn d16q_(d: f32) -> f32 { return min((floor(clamp(d, 0.0, 1.0) * 65535.0 + 0.5) + 0.25) / 65535.0, 1.0); }\n";
header += "fn scaling() -> vec4<f32> { return rescale_(0u); }\n"
"fn render_area() -> vec4<f32> { return rescale_(1u); }\n";
// opq_(x) == x, but the host compiler cannot see it (the uniform block's lane [0].w is always 0): it keeps an
// addition of a tiny epsilon from being reassociated away, e.g. (1 - a) + 1e-10 into (1 + 1e-10) - a = 1 - a,
// which then divides by zero where the guest's compiler kept the epsilon
header += fmt::format("fn opq_(x: f32) -> f32 {{ return bitcast<f32>(bitcast<u32>(x) | {}[0u].w); }}\n", ub_name);
DefineConstantBuffers(bindings);
DefineConstantBufferIndirect();
DefineStorageBuffers(bindings);
SetupTextures(bindings);
SetupImages(bindings);
DefineHelperFunctions();
DefineConstants();
}
void EmitContext::DefineBuiltins() {
header += "var<private> gl_Position: vec4<f32>;\n"
"var<private> gl_FragCoord: vec4<f32>;\n"
"var<private> gl_VertexID: i32;\n"
"var<private> gl_InstanceID: i32;\n"
"var<private> gl_BaseVertex: i32;\n"
"var<private> gl_BaseInstance: i32;\n"
"var<private> gl_DrawID: i32;\n"
"var<private> gl_FrontFacing: bool;\n"
"var<private> gl_FragDepth: f32;\n"
"var<private> gl_PointSize: f32;\n"
"var<private> gl_PointCoord: vec2<f32>;\n"
"var<private> gl_ClipDistance: array<f32, 8>;\n"
"var<private> gl_Layer: i32;\n"
"var<private> gl_ViewportIndex: i32;\n"
"var<private> gl_PrimitiveID: i32;\n"
"var<private> gl_SampleID: i32;\n"
"var<private> gl_SampleMask: array<i32, 1>;\n"
"var<private> gl_HelperInvocation: bool;\n"
"var<private> gl_LocalInvocationID: vec3<u32>;\n"
"var<private> gl_WorkGroupID: vec3<u32>;\n"
"var<private> gl_SubGroupInvocationARB: u32;\n"
"var<private> discarded_: bool;\n"
"var<private> F32_INF_: u32 = 0x7f800000u;\n"
"var<private> F32_NINF_: u32 = 0xff800000u;\n"
"var<private> F32_NAN_: u32 = 0x7fc00000u;\n"
// a call is never a const-expression: literal arguments of log2, sqrt, ... are not folded (and so
// not rejected) at shader creation
"fn rt_(x: f32) -> f32 { return x; }\n";
}
void EmitContext::DefineConstantBuffers(Bindings& bindings) {
for (const auto& desc : info.constant_buffer_descriptors) {
if (cbuf_in_ub[desc.index]) {
continue;
}
// not in the uniform block (indexed at run time, or no room): a read-only storage buffer of the bound size
header += fmt::format("@group({}) @binding({}) var<storage, read> {}_cbuf{}: array<vec4<u32>>;\n", group,
bindings.unified, stage_name, desc.index);
// reads at a run-time offset (CbufAt): past the bound size the hardware reads 0, where WebGPU's bounds checks
// would give the buffer's last vec4 (TOTK's particle programs index their emitter's tables so: long spikes)
if (!CbufClamp()) {
header += fmt::format("fn {0}_cbuf{1}_at(i: u32) -> vec4<u32> {{ if (i < arrayLength(&{0}_cbuf{1})) {{ "
"return {0}_cbuf{1}[i]; }} return vec4<u32>(0u); }}\n",
stage_name, desc.index);
}
cbuf_bindings.push_back({bindings.unified, desc.index});
bindings.unified += desc.count;
}
}
void EmitContext::DefineConstantBufferIndirect() {
if (!info.uses_cbuf_indirect) {
return;
}
header += "fn GetCbufIndirect(binding: u32, offset: u32) -> vec4<u32> {\n switch (binding) {\n";
for (const auto& desc : info.constant_buffer_descriptors) {
header += fmt::format(fmt::runtime(CbufClamp() ? " case {0}u: {{ return {1}_cbuf{0}[offset]; }}\n"
: " case {0}u: {{ return {1}_cbuf{0}_at(offset); }}\n"),
desc.index, stage_name);
}
header += " default: { return vec4<u32>(0u); }\n }\n}\n";
}
void EmitContext::DefineStorageBuffers(Bindings& bindings) {
u32 index{};
for (const auto& desc : info.storage_buffers_descriptors) {
if (index < 64 && ssbo_alias[index]) { // (uses another descriptor's binding)
bindings.unified += desc.count;
index += desc.count;
continue;
}
// WebGPU forbids writable storage buffers in vertex shaders
// compute: every storage buffer read_write (they may share one GPU buffer, and WebGPU forbids a buffer in
// one dispatch as both read-only and writable storage)
const bool writable{stage == Stage::Compute ||
(desc.is_written && ((stage != Stage::VertexA && stage != Stage::VertexB) || vs_compute))};
const bool atomic{index < 64 && atomic_ssbo[index]};
header += fmt::format("@group({}) @binding({}) var<storage, {}> {}_ssbo{}: array<{}>;\n", group,
bindings.unified, writable || atomic ? "read_write" : "read", stage_name, index,
atomic ? "atomic<u32>" : "u32");
ssbo_bindings.push_back({bindings.unified, index});
bindings.unified += desc.count;
index += desc.count;
}
}
void EmitContext::SetupTextures(Bindings& bindings) {
if (!info.texture_buffer_descriptors.empty()) {
throw NotImplementedException("WGSL texture buffers");
}
textures.reserve(info.texture_descriptors.size());
// Browsers allow 16 samplers a stage (wgpu more): from the 17th texture on, a texture samples with the sampler of
// the last earlier texture of its kind that has one (gpu/render/rx_draw.cpp binds the same way)
u32 last_own[2] = {0, 0}; // the last texture with a sampler of its own, by is_depth (binding + 1; 0: none)
for (const auto& desc : info.texture_descriptors) {
if (desc.count > 1) {
throw NotImplementedException("WGSL texture arrays");
}
const u32 binding{bindings.unified};
u32 shared{0};
#ifdef __EMSCRIPTEN__
if (textures.size() >= 16) shared = last_own[desc.is_depth ? 1 : 0]; // (binding + 1)
#endif
textures.push_back({binding, desc.count, shared});
if (!shared) last_own[desc.is_depth ? 1 : 0] = binding + 1;
const auto type{desc.is_depth ? DepthTextureType(desc.type)
: ColorTextureType(desc.type, desc.is_multisample)};
header += fmt::format("@group({}) @binding({}) var tex{}: {};\n", group, binding, binding, type);
if (!shared)
header += fmt::format("@group({}) @binding({}) var samp{}: {};\n", group, binding + 1, binding,
desc.is_depth ? "sampler_comparison" : "sampler");
// WebGPU texture views have no component swizzle and samplers no border color or LOD bias: per texture,
// two slots of the uniform block: [0] the TIC's component mapping (0-7 each; x bits 3-5: the axes whose
// wrap mode is border, y bits 3-5: the sampler's depth compare function, z bits 8-31: the sampler's LOD bias
// in 1/256), [1] the border color
const u32 swizzle_index{4 + 2 * static_cast<u32>(textures.size() - 1)};
header += fmt::format("fn swizzle{}(v: vec4<f32>) -> vec4<f32> {{\n"
" let s = {}[{}u] & vec4<u32>(7u);\n"
" let c = array<f32,8>(0.0,0.0,v.x,v.y,v.z,v.w,1.0,1.0);\n"
" return vec4<f32>(c[s.x],c[s.y],c[s.z],c[s.w]);\n}}\n",
binding, ub_name, swizzle_index);
header += fmt::format("fn lodbias{}() -> f32 {{ return f32(bitcast<i32>({}[{}u].z) >> 8u) / 256.0; }}\n",
binding, ub_name, swizzle_index);
header += fmt::format(
"fn outside{0}(c: vec3<f32>) -> bool {{\n"
" let m = ({1}[{2}u].x >> 3u) & 7u;\n"
" return ((m & 1u) != 0u && (c.x < 0.0 || c.x > 1.0)) || ((m & 2u) != 0u && (c.y < 0.0 || c.y > 1.0)) ||"
" ((m & 4u) != 0u && (c.z < 0.0 || c.z > 1.0));\n}}\n"
"fn border{0}(c: vec3<f32>, v: vec4<f32>) -> vec4<f32> {{\n"
" return select(v, bitcast<vec4<f32>>({1}[{3}u]), outside{0}(c));\n}}\n"
"fn bordercmp{0}(c: vec3<f32>, dref: f32, v: f32) -> f32 {{\n"
" if (!outside{0}(c)) {{ return v; }}\n"
" let b = bitcast<f32>({1}[{3}u].x);\n"
" switch (({1}[{2}u].y >> 3u) & 7u) {{\n"
" case 0u: {{ return 0.0; }}\n"
" case 1u: {{ return select(0.0, 1.0, dref < b); }}\n"
" case 2u: {{ return select(0.0, 1.0, dref == b); }}\n"
" case 3u: {{ return select(0.0, 1.0, dref <= b); }}\n"
" case 4u: {{ return select(0.0, 1.0, dref > b); }}\n"
" case 5u: {{ return select(0.0, 1.0, dref != b); }}\n"
" case 6u: {{ return select(0.0, 1.0, dref >= b); }}\n"
" default: {{ return 1.0; }}\n"
" }}\n}}\n",
binding, ub_name, swizzle_index, swizzle_index + 1);
texture_bindings.push_back({binding, static_cast<u32>(textures.size() - 1)});
bindings.unified += 2;
}
}
void EmitContext::SetupImages(Bindings& bindings) {
if (!info.image_buffer_descriptors.empty()) {
throw NotImplementedException("WGSL image buffers");
}
images.reserve(info.image_descriptors.size());
for (const auto& desc : info.image_descriptors) {
if (desc.count > 1) {
throw NotImplementedException("WGSL image arrays");
}
// (mkw-recomp) the bound image's format when the renderer gives it; the access the program needs
const size_t index = images.size();
const std::string format = index < image_formats.size() && !image_formats[index].empty()
? image_formats[index] : std::string(StorageFormat(desc.format));
const u32 kind = format.ends_with("uint") ? 0U : format.ends_with("sint") ? 2U : 1U;
images.push_back({bindings.unified, desc.count, 0, kind});
header += fmt::format("@group({}) @binding({}) var img{}: {}<{}, {}>;\n", group, bindings.unified,
bindings.unified, StorageTextureType(desc.type), format,
desc.is_written ? (desc.is_read ? "read_write" : "write") : "read");
bindings.unified += 1;
}
}
void EmitContext::DefineHelperFunctions() {
header += "fn add64(a: vec2<u32>, b: vec2<u32>) -> vec2<u32> {"
" let lo = a.x + b.x; return vec2<u32>(lo, a.y + b.y + select(0u, 1u, lo < a.x)); }\n"
"fn sub64(a: vec2<u32>, b: vec2<u32>) -> vec2<u32> {"
" return vec2<u32>(a.x - b.x, a.y - b.y - select(0u, 1u, a.x < b.x)); }\n"
"fn shl64(a: vec2<u32>, s: u32) -> vec2<u32> {"
" if (s == 0u) { return a; } if (s >= 32u) { return vec2<u32>(0u, a.x << (s - 32u)); }"
" return vec2<u32>(a.x << s, (a.y << s) | (a.x >> (32u - s))); }\n"
"fn shr64(a: vec2<u32>, s: u32) -> vec2<u32> {"
" if (s == 0u) { return a; } if (s >= 32u) { return vec2<u32>(a.y >> (s - 32u), 0u); }"
" return vec2<u32>((a.x >> s) | (a.y << (32u - s)), a.y >> s); }\n"
"fn sar64(a: vec2<u32>, s: u32) -> vec2<u32> {"
" let hi = i32(a.y); if (s == 0u) { return a; }"
" if (s >= 32u) { return vec2<u32>(u32(hi >> (s - 32u)), u32(hi >> 31u)); }"
" return vec2<u32>((a.x >> s) | (a.y << (32u - s)), u32(hi >> s)); }\n";
header += "fn isnan_(x: f32) -> bool { return (bitcast<u32>(x) & 0x7fffffffu) > 0x7f800000u; }\n"
"fn isinf_(x: f32) -> bool { return (bitcast<u32>(x) & 0x7fffffffu) == 0x7f800000u; }\n"
"fn isnan2_(x: vec2<f32>) -> vec2<bool> { return vec2<bool>(isnan_(x.x), isnan_(x.y)); }\n"
// a NaN level of detail or bias picks level 0 on Maxwell (its fixed-point conversion); on PC GPUs the
// sample itself becomes NaN (TOTK: white blobs from an environment cube sampled at a NaN roughness LOD)
"fn lodok_(x: f32) -> f32 { return select(x, 0.0, isnan_(x)); }\n";
if (info.uses_global_increment || info.uses_shared_increment) {
header += "fn CasIncrement(op_a: u32, op_b: u32) -> u32 { return select(op_a + 1u, 0u, op_a >= op_b); }\n";
}
if (info.uses_global_decrement || info.uses_shared_decrement) {
header += "fn CasDecrement(op_a: u32, op_b: u32) -> u32 {"
" return select(op_a - 1u, op_b, op_a == 0u || op_a > op_b); }\n";
}
if (info.uses_atomic_f32_add) {
header += "fn CasFloatAdd(op_a: u32, op_b: f32) -> u32 { return bitcast<u32>(bitcast<f32>(op_a) + op_b); }\n";
}
if (info.uses_atomic_s32_min) {
header += "fn CasMinS32(op_a: u32, op_b: u32) -> u32 { return u32(min(i32(op_a), i32(op_b))); }\n";
}
if (info.uses_atomic_s32_max) {
header += "fn CasMaxS32(op_a: u32, op_b: u32) -> u32 { return u32(max(i32(op_a), i32(op_b))); }\n";
}
if (info.uses_global_memory) {
throw NotImplementedException("WGSL global memory (64-bit addresses)");
}
if (info.loads_indexed_attributes) {
std::string func{"fn IndexedAttrLoad(offset: i32) -> f32 {\n"
" let base_index = offset >> 2u;\n let masked_index = u32(base_index) & 3u;\n"
" switch (base_index >> 2u) {\n"};
if (info.loads.AnyComponent(IR::Attribute::PositionX)) {
func += fmt::format(" case {}: {{ return {}[masked_index]; }}\n",
static_cast<u32>(IR::Attribute::PositionX) >> 2, position_name);
}
const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2;
for (u32 index = 0; index < IR::NUM_GENERICS; ++index) {
if (info.loads.Generic(index)) {
func += fmt::format(" case {}: {{ return in_attr{}[masked_index]; }}\n",
base_attribute_value + index, index);
}
}
func += " default: { return 0.0f; }\n }\n}\n";
header += func;
}
}
void EmitContext::DefineConstants() {
if (info.uses_fswzadd) {
header += "var<private> FSWZ_A: array<f32, 4> = array<f32, 4>(-1.0f, 1.0f, -1.0f, 0.0f);\n"
"var<private> FSWZ_B: array<f32, 4> = array<f32, 4>(-1.0f, -1.0f, 1.0f, -1.0f);\n";
}
}
} // namespace Shader::Backend::WGSL
namespace Shader::Backend::WGSL {
std::string EmitContext::SsboPtr(u32 binding, std::string_view word) const {
const std::string w{SsboWord(binding, word)};
word = w;
binding = MapSsbo(binding);
return fmt::format("&{}_ssbo{}[{}]", stage_name, binding, word);
}
std::string EmitContext::SsboLoad(u32 binding, std::string_view word) const {
const std::string w{SsboWord(binding, word)};
word = w;
binding = MapSsbo(binding);
if (binding < 64 && atomic_ssbo[binding]) {
return fmt::format("atomicLoad(&{}_ssbo{}[{}])", stage_name, binding, word);
}
return fmt::format("{}_ssbo{}[{}]", stage_name, binding, word);
}
std::string EmitContext::SsboStore(u32 binding, std::string_view word, std::string_view value) const {
const std::string w{SsboWord(binding, word)};
word = w;
binding = MapSsbo(binding);
if (binding < 64 && atomic_ssbo[binding]) {
return fmt::format("atomicStore(&{}_ssbo{}[{}],{});", stage_name, binding, word, value);
}
return fmt::format("{}_ssbo{}[{}]={};", stage_name, binding, word, value);
}
std::string EmitContext::SmemLoad(std::string_view word) const {
return atomic_smem ? fmt::format("atomicLoad(&smem[{}])", word) : fmt::format("smem[{}]", word);
}
std::string EmitContext::SmemStore(std::string_view word, std::string_view value) const {
return atomic_smem ? fmt::format("atomicStore(&smem[{}],{});", word, value) : fmt::format("smem[{}]={};", word, value);
}
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,223 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <bitset>
#include <cstdlib>
#include <string>
#include <utility>
#include <vector>
#include <fmt/ranges.h>
#include "shader_recompiler/backend/wgsl/var_alloc.h"
#include "shader_recompiler/stage.h"
namespace Shader {
struct Info;
struct Profile;
struct RuntimeInfo;
} // namespace Shader
namespace Shader::Backend {
struct Bindings;
}
namespace Shader::IR {
class Inst;
struct Program;
} // namespace Shader::IR
namespace Shader::Backend::WGSL {
/// RX_CBUF_CLAMP=1: constant buffer reads at run-time offsets past the bound size get WebGPU's clamped read (the old
/// translation) instead of the hardware's 0 (<stage>_cbuf<n>_at).
inline bool CbufClamp() {
static const bool clamp = std::getenv("RX_CBUF_CLAMP") != nullptr;
return clamp;
}
struct GenericElementInfo {
std::string name;
u32 first_element{};
u32 num_components{};
};
struct TextureImageDefinition {
u32 binding;
u32 count;
u32 sampler_plus1 = 0; // the texture whose sampler it samples with, binding + 1 (0: its own; shared: web)
u32 kind = 0; // (mkw-recomp, images: the declared format's texel type: 0 u32, 1 f32, 2 i32)
u32 textures_sampler_binding() const { return sampler_plus1 ? sampler_plus1 - 1 : binding; }
};
/// A resource of the stage's bind group: its binding number and the descriptor index it serves.
struct ResourceBinding {
u32 binding;
u32 index;
};
class EmitContext {
public:
explicit EmitContext(IR::Program& program, Bindings& bindings, const Profile& profile_,
const RuntimeInfo& runtime_info_);
template <WgslVarType type, typename... Args>
void Add(const char* format_str, IR::Inst& inst, Args&&... args) {
const auto var_def{var_alloc.AddDefine(inst, type)};
if (var_def.empty()) {
// the result is unused: WGSL needs a phony assignment to evaluate an expression for its effects
code += "_=";
code += fmt::format(fmt::runtime(format_str + 3), std::forward<Args>(args)...);
} else {
code += fmt::format(fmt::runtime(format_str), var_def, std::forward<Args>(args)...);
}
// TODO: Remove this
code += '\n';
}
template <typename... Args>
void AddU1(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::U1>(format_str, inst, args...);
}
template <typename... Args>
void AddF16x2(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::F16x2>(format_str, inst, args...);
}
template <typename... Args>
void AddU32(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::U32>(format_str, inst, args...);
}
template <typename... Args>
void AddF32(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::F32>(format_str, inst, args...);
}
template <typename... Args>
void AddU64(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::U64>(format_str, inst, args...);
}
template <typename... Args>
void AddF64(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::F64>(format_str, inst, args...);
}
template <typename... Args>
void AddU32x2(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::U32x2>(format_str, inst, args...);
}
template <typename... Args>
void AddF32x2(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::F32x2>(format_str, inst, args...);
}
template <typename... Args>
void AddU32x3(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::U32x3>(format_str, inst, args...);
}
template <typename... Args>
void AddF32x3(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::F32x3>(format_str, inst, args...);
}
template <typename... Args>
void AddU32x4(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::U32x4>(format_str, inst, args...);
}
template <typename... Args>
void AddF32x4(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::F32x4>(format_str, inst, args...);
}
template <typename... Args>
void AddPrecF32(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::PrecF32>(format_str, inst, args...);
}
template <typename... Args>
void AddPrecF64(const char* format_str, IR::Inst& inst, Args&&... args) {
Add<WgslVarType::PrecF64>(format_str, inst, args...);
}
template <typename... Args>
void Add(const char* format_str, Args&&... args) {
code += fmt::format(fmt::runtime(format_str), std::forward<Args>(args)...);
// TODO: Remove this
code += '\n';
}
std::string header;
std::string code;
VarAlloc var_alloc;
const Info& info;
const Profile& profile;
const RuntimeInfo& runtime_info;
Stage stage{};
std::string_view stage_name = "invalid";
std::string_view position_name = "gl_Position";
u32 group{}; ///< the bind group of this stage's resources
/// The stage's uniform block (binding 0, bound at a dynamic offset per draw): its name, size in vec4s, and
/// which constant buffers are in it and where (the others are storage buffers: indexed at run time, or no
/// room left in the 64 KB a uniform binding may have).
std::string ub_name;
u32 ub_size{};
std::array<bool, 18> cbuf_in_ub{};
std::array<u32, 18> ub_cbuf_base{};
std::vector<ResourceBinding> cbuf_bindings;
std::vector<ResourceBinding> ssbo_bindings;
// Storage buffers (by index) and shared memory an atomic operation touches: declared array<atomic<u32>> and
// read and written through atomicLoad / atomicStore (WGSL atomics need atomic-typed memory)
std::bitset<64> atomic_ssbo;
std::array<u8, 64> ssbo_alias{}; // (see ssbo_alias_next)
bool vs_compute{}; // (see vs_as_compute_next)
std::array<std::string, 16> image_formats{}; // (see image_formats_next)
std::array<u32, 64> ssbo_alias_words{};
std::string SsboWord(u32 binding, std::string_view word) const {
return binding < 64 && ssbo_alias[binding] && ssbo_alias_words[binding]
? fmt::format("(({}) + {}u)", word, ssbo_alias_words[binding]) : std::string(word);
}
u32 MapSsbo(u32 binding) const { return binding < 64 && ssbo_alias[binding] ? ssbo_alias[binding] - 1u : binding; }
bool atomic_smem{};
std::string SsboLoad(u32 binding, std::string_view word) const;
std::string SsboStore(u32 binding, std::string_view word, std::string_view value) const;
std::string SsboPtr(u32 binding, std::string_view word) const;
std::string SmemLoad(std::string_view word) const;
std::string SmemStore(std::string_view word, std::string_view value) const;
std::vector<ResourceBinding> texture_bindings;
std::vector<TextureImageDefinition> texture_buffers;
std::vector<TextureImageDefinition> image_buffers;
std::vector<TextureImageDefinition> textures;
std::vector<TextureImageDefinition> images;
std::array<std::array<GenericElementInfo, 4>, 32> output_generics{};
u32 num_safety_loop_vars{};
bool uses_y_direction{};
bool uses_cc_carry{};
bool uses_geometry_passthrough{};
private:
void DefineBuiltins();
void DefineConstantBuffers(Bindings& bindings);
void DefineConstantBufferIndirect();
void DefineStorageBuffers(Bindings& bindings);
void DefineHelperFunctions();
void DefineConstants();
void SetupImages(Bindings& bindings);
void SetupTextures(Bindings& bindings);
};
} // namespace Shader::Backend::WGSL

View File

@@ -0,0 +1,138 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <ankerl/unordered_dense.h>
#include "common/common_types.h"
#include "shader_recompiler/program_header.h"
#include "shader_recompiler/shader_info.h"
#include "shader_recompiler/stage.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
class Inst;
}
namespace Shader {
struct CbufWordKey {
u32 index;
u32 offset;
constexpr bool operator==(const CbufWordKey& o) const noexcept {
return index == o.index && offset == o.offset;
}
};
struct CbufWordKeyHash {
constexpr size_t operator()(const CbufWordKey& k) const noexcept {
return (size_t(k.index) << 32) ^ k.offset;
}
};
struct HandleKey {
u32 index, offset, shift_left;
u32 sec_index, sec_offset, sec_shift_left;
u32 count;
bool has_secondary;
constexpr bool operator==(const HandleKey& o) const noexcept {
return std::tie(index, offset, shift_left, sec_index, sec_offset, sec_shift_left, count, has_secondary)
== std::tie(o.index, o.offset, o.shift_left, o.sec_index, o.sec_offset, o.sec_shift_left, o.count, o.has_secondary);
}
};
struct HandleKeyHash {
constexpr size_t operator()(const HandleKey& k) const noexcept {
size_t h = (size_t(k.index) << 32) ^ k.offset;
h ^= (size_t(k.shift_left) << 1);
h ^= (size_t(k.sec_index) << 33) ^ (size_t(k.sec_offset) << 2);
h ^= (size_t(k.sec_shift_left) << 3);
h ^= (size_t(k.count) << 7);
return h ^ (k.has_secondary ? 0x9e3779b97f4a7c15ULL : 0ULL);
}
};
struct ConstBufferAddr {
u32 index;
u32 offset;
u32 shift_left;
u32 secondary_index;
u32 secondary_offset;
u32 secondary_shift_left;
IR::U32 dynamic_offset;
u32 count;
bool has_secondary;
};
class Environment {
public:
virtual ~Environment() = default;
[[nodiscard]] virtual u64 ReadInstruction(u32 address) = 0;
/// (mkw-recomp, from Eglantine) an indirect branch's target, from the native program's address to the
/// translated one's: Ampere programs are lowered to Maxwell (frontend/ampere) before the Maxwell frontend
[[nodiscard]] virtual u32 TranslateProgramAddress(u32 address) const { return address; }
[[nodiscard]] virtual u32 ReadCbufValue(u32 cbuf_index, u32 cbuf_offset) = 0;
[[nodiscard]] virtual TextureType ReadTextureType(u32 raw_handle) = 0;
[[nodiscard]] virtual TexturePixelFormat ReadTexturePixelFormat(u32 raw_handle) = 0;
[[nodiscard]] virtual bool IsTexturePixelFormatInteger(u32 raw_handle) = 0;
[[nodiscard]] virtual u32 ReadViewportTransformState() = 0;
[[nodiscard]] virtual u32 TextureBoundBuffer() const = 0;
[[nodiscard]] virtual u32 LocalMemorySize() const = 0;
[[nodiscard]] virtual u32 SharedMemorySize() const = 0;
[[nodiscard]] virtual std::array<u32, 3> WorkgroupSize() const = 0;
[[nodiscard]] virtual bool HasHLEMacroState() const = 0;
[[nodiscard]] virtual std::optional<ReplaceConstant> GetReplaceConstBuffer(u32 bank,
u32 offset) = 0;
virtual void Dump(u64 pipeline_hash, u64 shader_hash) = 0;
[[nodiscard]] const ProgramHeader& SPH() const noexcept {
return sph;
}
[[nodiscard]] const std::array<u32, 8>& GpPassthroughMask() const noexcept {
return gp_passthrough_mask;
}
[[nodiscard]] Stage ShaderStage() const noexcept {
return stage;
}
[[nodiscard]] u32 StartAddress() const noexcept {
return start_address;
}
[[nodiscard]] bool IsProprietaryDriver() const noexcept {
return is_proprietary_driver;
}
protected:
ProgramHeader sph{};
std::array<u32, 8> gp_passthrough_mask{};
Stage stage{};
u32 start_address{};
bool is_proprietary_driver{};
public:
ankerl::unordered_dense::map<CbufWordKey, u32, CbufWordKeyHash> cbuf_word_cache;
ankerl::unordered_dense::map<HandleKey, u32, HandleKeyHash> handle_cache;
ankerl::unordered_dense::map<const IR::Inst*, ConstBufferAddr> track_cache;
};
} // namespace Shader

View File

@@ -0,0 +1,67 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <exception>
#include <string>
#include <utility>
#include "common/logging.h"
namespace Shader {
class Exception : public std::exception {
public:
explicit Exception(std::string message) noexcept : err_message{std::move(message)} {}
[[nodiscard]] const char* what() const noexcept override {
return err_message.c_str();
}
void Prepend(std::string_view prepend) {
err_message.insert(0, prepend);
}
void Append(std::string_view append) {
err_message += append;
}
private:
std::string err_message;
};
class LogicError : public Exception {
public:
template <typename... Args>
explicit LogicError(const char* message, Args&&... args)
: Exception{fmt::format(fmt::runtime(message), std::forward<Args>(args)...)} {}
};
class RuntimeError : public Exception {
public:
template <typename... Args>
explicit RuntimeError(const char* message, Args&&... args)
: Exception{fmt::format(fmt::runtime(message), std::forward<Args>(args)...)} {}
};
class NotImplementedException : public Exception {
public:
template <typename... Args>
explicit NotImplementedException(const char* message, Args&&... args)
: Exception{fmt::format(fmt::runtime(message), std::forward<Args>(args)...)} {
Append(" is not implemented");
}
};
class InvalidArgument : public Exception {
public:
template <typename... Args>
explicit InvalidArgument(const char* message, Args&&... args)
: Exception{fmt::format(fmt::runtime(message), std::forward<Args>(args)...)} {}
};
} // namespace Shader

File diff suppressed because it is too large Load Diff

View File

@@ -0,0 +1,29 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include <optional>
#include <span>
#include <vector>
#include "common/common_types.h"
namespace Shader::Ampere {
constexpr u64 NativeAddressMapMagic = 0x50414d4e4544454eULL;
inline bool HasNativeAddressMap(std::span<const u64> code) {
return code.size() >= 2 && code.back() == NativeAddressMapMagic;
}
inline void AppendNativeAddressMap(std::vector<u64>& code, std::span<const u32> offsets) {
if (offsets.empty()) return;
for (u32 index : offsets) code.push_back(0x50U + index * 8 + ((index + 1) / 3) * 8);
code.push_back(offsets.size());
code.push_back(NativeAddressMapMagic);
}
inline std::optional<u32> NativeProgramAddress(std::span<const u64> code, u32 address) {
if (!HasNativeAddressMap(code) || address % 16) return {};
const u64 count = code[code.size() - 2];
if (!count || count > 65536 || count + 2 > code.size() || address / 16 >= count) return {};
const size_t start = code.size() - static_cast<size_t>(count) - 2;
const u64 target = code[start + address / 16];
if (target < 0x50 || target >= start * 8 || target % 8 || target % 32 == 0) return {};
return static_cast<u32>(target);
}
} // namespace Shader::Ampere

View File

@@ -0,0 +1,54 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include "shader_recompiler/frontend/ampere/lower_to_maxwell.h"
namespace Shader::Ampere {
class ProgramCaptureState {
public:
enum class Result { Continue, Complete, InvalidTarget };
static constexpr u32 MaximumInstructions = 65536;
Result Observe(const Instruction& instruction, u32 index) {
const u32 opcode = instruction.Bits(0, 12);
if (opcode == 0x944 || opcode == 0x947 || opcode == 0x945) {
const u64 raw = instruction.words[1] | (u64{instruction.words[2] & 0x3ffff} << 32);
const s64 relative = opcode == 0x945 ? static_cast<s32>(instruction.words[1])
: static_cast<s64>(raw << 14) >> 14;
const s64 target = s64{index} * 16 + 16 + relative;
if (target < 0 || target % 16 || target >= MaximumInstructions * 16)
return Result::InvalidTarget;
// CALL.REL.NOINC (bit 86) pushes no return address: the compiler uses it as a jump
// (the lowering treats it as one), so it never pairs with a RET.
const bool noinc_call = opcode == 0x944 && (instruction.words[2] & 0x400000) != 0;
if (opcode == 0x944 && !noinc_call) {
last_call_target = std::max(last_call_target, static_cast<u32>(target / 16));
} else {
last_forward_target = std::max(last_forward_target, static_cast<u32>(target / 16));
}
// The compiler ends every program with an unconditional branch to itself; with a jump
// table (whose targets live in a constant buffer) that is the only reliable end.
if (opcode == 0x947 && computed_branch && instruction.Bits(12, 4) == 7 &&
target == s64{index} * 16)
return Result::Complete;
}
if (opcode == 0x949 || opcode == 0x958) computed_branch = true;
if (computed_branch) return Result::Continue;
if (opcode == 0x94d && instruction.Bits(12, 4) == 7 && index >= last_forward_target) {
if (!last_call_target) return Result::Complete;
main_exited = true;
}
if (opcode == 0x950 && instruction.Bits(12, 4) == 7 && main_exited &&
index >= last_call_target && index >= last_forward_target)
return Result::Complete;
return Result::Continue;
}
private:
u32 last_call_target{};
u32 last_forward_target{};
bool main_exited{};
bool computed_branch{};
};
} // namespace Shader::Ampere

View File

@@ -0,0 +1,85 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include <cstring>
#include "shader_recompiler/frontend/ampere/lower_to_maxwell.h"
namespace Shader::Ampere {
inline u64 NativeProgramKey(std::span<const Instruction> source) {
u64 hash = 14695981039346656037ULL;
for (const auto& instruction : source) {
for (u32 word : instruction.words) {
for (u32 byte = 0; byte < 4; ++byte) hash = (hash ^ ((word >> (byte * 8)) & 255)) * 1099511628211ULL;
}
}
return hash;
}
// Private iteration format: version, lengths, texture bank, exact original
// instructions, then the newly generated canonical instructions. Source bytes
// are compared directly; the filename hash is only an index.
inline std::vector<u8> EncodeProgramOverride(std::span<const Instruction> source,
const LoweredProgram& program) {
if (!program || program.source_size != source.size_bytes() || program.source_instruction_offsets.size() != source.size()) return {};
const std::array<u32, 6> header{0x4e454445, 0x3230564f,
static_cast<u32>(source.size_bytes()), static_cast<u32>(program.instructions.size()),
program.required_texture_buffer, static_cast<u32>(program.source_instruction_offsets.size())};
std::vector<u8> bytes(sizeof(header) + source.size_bytes() + program.instructions.size() * 8 + program.source_instruction_offsets.size() * 4);
std::memcpy(bytes.data(), header.data(), sizeof(header));
std::memcpy(bytes.data() + sizeof(header), source.data(), source.size_bytes());
std::memcpy(bytes.data() + sizeof(header) + source.size_bytes(), program.instructions.data(), program.instructions.size() * 8);
std::memcpy(bytes.data() + sizeof(header) + source.size_bytes() + program.instructions.size() * 8,
program.source_instruction_offsets.data(), program.source_instruction_offsets.size() * 4);
return bytes;
}
inline u32 ProgramOverrideSourceSize(std::span<const u8> bytes) {
std::array<u32, 6> header{};
if (bytes.size() < sizeof(header)) return 0;
std::memcpy(header.data(), bytes.data(), sizeof(header));
const bool mapped = header[1] == 0x3230564f;
if (header[0] != 0x4e454445 || (!mapped && header[1] != 0x3130564f) ||
(mapped ? header[5] != header[2] / 16 : header[5] != 0) ||
header[2] == 0 || header[2] % sizeof(Instruction) || header[2] > 1048576 ||
header[3] == 0 || header[3] > 1048576 || (header[4] != ~0U && header[4] >= 18)) return 0;
if (bytes.size() != sizeof(header) + size_t{header[2]} + size_t{header[3]} * 8 + size_t{header[5]} * 4) return 0;
return header[2];
}
inline LoweredProgram DecodeProgramOverride(std::span<const u8> bytes,
std::span<const Instruction> source) {
LoweredProgram result;
const auto fail = [&result](std::string_view message) {
result.instructions.clear();
result.source_instruction_offsets.clear();
result.error = message;
return result;
};
std::array<u32, 6> header{};
if (!ProgramOverrideSourceSize(bytes)) return fail("invalid native override header or length");
std::memcpy(header.data(), bytes.data(), sizeof(header));
if (header[2] != source.size_bytes()) return fail("native override source length mismatch");
if (std::memcmp(bytes.data() + sizeof(header), source.data(), source.size_bytes()))
return fail("native override source mismatch");
result.instructions.resize(header[3]);
std::memcpy(result.instructions.data(), bytes.data() + sizeof(header) + source.size_bytes(), header[3] * 8);
if (header[5]) {
result.source_instruction_offsets.resize(header[5]);
std::memcpy(result.source_instruction_offsets.data(),
bytes.data() + sizeof(header) + source.size_bytes() + size_t{header[3]} * 8,
size_t{header[5]} * 4);
u32 previous = 0;
for (u32 offset : result.source_instruction_offsets) {
if (offset >= result.instructions.size() || offset < previous)
return fail("native override address map mismatch");
previous = offset;
}
}
result.source_size = header[2];
result.required_texture_buffer = header[4];
return result;
}
} // namespace Shader::Ampere

View File

@@ -0,0 +1,57 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <vector>
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
class Block;
struct AbstractSyntaxNode {
enum class Type {
Block,
If,
EndIf,
Loop,
Repeat,
Break,
Return,
Unreachable,
};
union Data {
Block* block;
struct {
U1 cond;
Block* body;
Block* merge;
} if_node;
struct {
Block* merge;
} end_if;
struct {
Block* body;
Block* continue_block;
Block* merge;
} loop;
struct {
U1 cond;
Block* loop_header;
Block* merge;
} repeat;
struct {
U1 cond;
Block* merge;
Block* skip;
} break_node;
};
Data data{};
Type type{};
};
using AbstractSyntaxList = std::vector<AbstractSyntaxNode>;
} // namespace Shader::IR

View File

@@ -0,0 +1,459 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <fmt/format.h>
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/attribute.h"
namespace Shader::IR {
bool IsGeneric(Attribute attribute) noexcept {
return attribute >= Attribute::Generic0X && attribute <= Attribute::Generic31X;
}
u32 GenericAttributeIndex(Attribute attribute) {
if (!IsGeneric(attribute)) {
throw InvalidArgument("Attribute is not generic {}", attribute);
}
return (static_cast<u32>(attribute) - static_cast<u32>(Attribute::Generic0X)) / 4u;
}
u32 GenericAttributeElement(Attribute attribute) {
if (!IsGeneric(attribute)) {
throw InvalidArgument("Attribute is not generic {}", attribute);
}
return static_cast<u32>(attribute) % 4;
}
std::string NameOf(Attribute attribute) {
switch (attribute) {
case Attribute::PrimitiveId:
return "PrimitiveId";
case Attribute::Layer:
return "Layer";
case Attribute::ViewportIndex:
return "ViewportIndex";
case Attribute::PointSize:
return "PointSize";
case Attribute::PositionX:
return "Position.X";
case Attribute::PositionY:
return "Position.Y";
case Attribute::PositionZ:
return "Position.Z";
case Attribute::PositionW:
return "Position.W";
case Attribute::Generic0X:
return "Generic[0].X";
case Attribute::Generic0Y:
return "Generic[0].Y";
case Attribute::Generic0Z:
return "Generic[0].Z";
case Attribute::Generic0W:
return "Generic[0].W";
case Attribute::Generic1X:
return "Generic[1].X";
case Attribute::Generic1Y:
return "Generic[1].Y";
case Attribute::Generic1Z:
return "Generic[1].Z";
case Attribute::Generic1W:
return "Generic[1].W";
case Attribute::Generic2X:
return "Generic[2].X";
case Attribute::Generic2Y:
return "Generic[2].Y";
case Attribute::Generic2Z:
return "Generic[2].Z";
case Attribute::Generic2W:
return "Generic[2].W";
case Attribute::Generic3X:
return "Generic[3].X";
case Attribute::Generic3Y:
return "Generic[3].Y";
case Attribute::Generic3Z:
return "Generic[3].Z";
case Attribute::Generic3W:
return "Generic[3].W";
case Attribute::Generic4X:
return "Generic[4].X";
case Attribute::Generic4Y:
return "Generic[4].Y";
case Attribute::Generic4Z:
return "Generic[4].Z";
case Attribute::Generic4W:
return "Generic[4].W";
case Attribute::Generic5X:
return "Generic[5].X";
case Attribute::Generic5Y:
return "Generic[5].Y";
case Attribute::Generic5Z:
return "Generic[5].Z";
case Attribute::Generic5W:
return "Generic[5].W";
case Attribute::Generic6X:
return "Generic[6].X";
case Attribute::Generic6Y:
return "Generic[6].Y";
case Attribute::Generic6Z:
return "Generic[6].Z";
case Attribute::Generic6W:
return "Generic[6].W";
case Attribute::Generic7X:
return "Generic[7].X";
case Attribute::Generic7Y:
return "Generic[7].Y";
case Attribute::Generic7Z:
return "Generic[7].Z";
case Attribute::Generic7W:
return "Generic[7].W";
case Attribute::Generic8X:
return "Generic[8].X";
case Attribute::Generic8Y:
return "Generic[8].Y";
case Attribute::Generic8Z:
return "Generic[8].Z";
case Attribute::Generic8W:
return "Generic[8].W";
case Attribute::Generic9X:
return "Generic[9].X";
case Attribute::Generic9Y:
return "Generic[9].Y";
case Attribute::Generic9Z:
return "Generic[9].Z";
case Attribute::Generic9W:
return "Generic[9].W";
case Attribute::Generic10X:
return "Generic[10].X";
case Attribute::Generic10Y:
return "Generic[10].Y";
case Attribute::Generic10Z:
return "Generic[10].Z";
case Attribute::Generic10W:
return "Generic[10].W";
case Attribute::Generic11X:
return "Generic[11].X";
case Attribute::Generic11Y:
return "Generic[11].Y";
case Attribute::Generic11Z:
return "Generic[11].Z";
case Attribute::Generic11W:
return "Generic[11].W";
case Attribute::Generic12X:
return "Generic[12].X";
case Attribute::Generic12Y:
return "Generic[12].Y";
case Attribute::Generic12Z:
return "Generic[12].Z";
case Attribute::Generic12W:
return "Generic[12].W";
case Attribute::Generic13X:
return "Generic[13].X";
case Attribute::Generic13Y:
return "Generic[13].Y";
case Attribute::Generic13Z:
return "Generic[13].Z";
case Attribute::Generic13W:
return "Generic[13].W";
case Attribute::Generic14X:
return "Generic[14].X";
case Attribute::Generic14Y:
return "Generic[14].Y";
case Attribute::Generic14Z:
return "Generic[14].Z";
case Attribute::Generic14W:
return "Generic[14].W";
case Attribute::Generic15X:
return "Generic[15].X";
case Attribute::Generic15Y:
return "Generic[15].Y";
case Attribute::Generic15Z:
return "Generic[15].Z";
case Attribute::Generic15W:
return "Generic[15].W";
case Attribute::Generic16X:
return "Generic[16].X";
case Attribute::Generic16Y:
return "Generic[16].Y";
case Attribute::Generic16Z:
return "Generic[16].Z";
case Attribute::Generic16W:
return "Generic[16].W";
case Attribute::Generic17X:
return "Generic[17].X";
case Attribute::Generic17Y:
return "Generic[17].Y";
case Attribute::Generic17Z:
return "Generic[17].Z";
case Attribute::Generic17W:
return "Generic[17].W";
case Attribute::Generic18X:
return "Generic[18].X";
case Attribute::Generic18Y:
return "Generic[18].Y";
case Attribute::Generic18Z:
return "Generic[18].Z";
case Attribute::Generic18W:
return "Generic[18].W";
case Attribute::Generic19X:
return "Generic[19].X";
case Attribute::Generic19Y:
return "Generic[19].Y";
case Attribute::Generic19Z:
return "Generic[19].Z";
case Attribute::Generic19W:
return "Generic[19].W";
case Attribute::Generic20X:
return "Generic[20].X";
case Attribute::Generic20Y:
return "Generic[20].Y";
case Attribute::Generic20Z:
return "Generic[20].Z";
case Attribute::Generic20W:
return "Generic[20].W";
case Attribute::Generic21X:
return "Generic[21].X";
case Attribute::Generic21Y:
return "Generic[21].Y";
case Attribute::Generic21Z:
return "Generic[21].Z";
case Attribute::Generic21W:
return "Generic[21].W";
case Attribute::Generic22X:
return "Generic[22].X";
case Attribute::Generic22Y:
return "Generic[22].Y";
case Attribute::Generic22Z:
return "Generic[22].Z";
case Attribute::Generic22W:
return "Generic[22].W";
case Attribute::Generic23X:
return "Generic[23].X";
case Attribute::Generic23Y:
return "Generic[23].Y";
case Attribute::Generic23Z:
return "Generic[23].Z";
case Attribute::Generic23W:
return "Generic[23].W";
case Attribute::Generic24X:
return "Generic[24].X";
case Attribute::Generic24Y:
return "Generic[24].Y";
case Attribute::Generic24Z:
return "Generic[24].Z";
case Attribute::Generic24W:
return "Generic[24].W";
case Attribute::Generic25X:
return "Generic[25].X";
case Attribute::Generic25Y:
return "Generic[25].Y";
case Attribute::Generic25Z:
return "Generic[25].Z";
case Attribute::Generic25W:
return "Generic[25].W";
case Attribute::Generic26X:
return "Generic[26].X";
case Attribute::Generic26Y:
return "Generic[26].Y";
case Attribute::Generic26Z:
return "Generic[26].Z";
case Attribute::Generic26W:
return "Generic[26].W";
case Attribute::Generic27X:
return "Generic[27].X";
case Attribute::Generic27Y:
return "Generic[27].Y";
case Attribute::Generic27Z:
return "Generic[27].Z";
case Attribute::Generic27W:
return "Generic[27].W";
case Attribute::Generic28X:
return "Generic[28].X";
case Attribute::Generic28Y:
return "Generic[28].Y";
case Attribute::Generic28Z:
return "Generic[28].Z";
case Attribute::Generic28W:
return "Generic[28].W";
case Attribute::Generic29X:
return "Generic[29].X";
case Attribute::Generic29Y:
return "Generic[29].Y";
case Attribute::Generic29Z:
return "Generic[29].Z";
case Attribute::Generic29W:
return "Generic[29].W";
case Attribute::Generic30X:
return "Generic[30].X";
case Attribute::Generic30Y:
return "Generic[30].Y";
case Attribute::Generic30Z:
return "Generic[30].Z";
case Attribute::Generic30W:
return "Generic[30].W";
case Attribute::Generic31X:
return "Generic[31].X";
case Attribute::Generic31Y:
return "Generic[31].Y";
case Attribute::Generic31Z:
return "Generic[31].Z";
case Attribute::Generic31W:
return "Generic[31].W";
case Attribute::ColorFrontDiffuseR:
return "ColorFrontDiffuse.R";
case Attribute::ColorFrontDiffuseG:
return "ColorFrontDiffuse.G";
case Attribute::ColorFrontDiffuseB:
return "ColorFrontDiffuse.B";
case Attribute::ColorFrontDiffuseA:
return "ColorFrontDiffuse.A";
case Attribute::ColorFrontSpecularR:
return "ColorFrontSpecular.R";
case Attribute::ColorFrontSpecularG:
return "ColorFrontSpecular.G";
case Attribute::ColorFrontSpecularB:
return "ColorFrontSpecular.B";
case Attribute::ColorFrontSpecularA:
return "ColorFrontSpecular.A";
case Attribute::ColorBackDiffuseR:
return "ColorBackDiffuse.R";
case Attribute::ColorBackDiffuseG:
return "ColorBackDiffuse.G";
case Attribute::ColorBackDiffuseB:
return "ColorBackDiffuse.B";
case Attribute::ColorBackDiffuseA:
return "ColorBackDiffuse.A";
case Attribute::ColorBackSpecularR:
return "ColorBackSpecular.R";
case Attribute::ColorBackSpecularG:
return "ColorBackSpecular.G";
case Attribute::ColorBackSpecularB:
return "ColorBackSpecular.B";
case Attribute::ColorBackSpecularA:
return "ColorBackSpecular.A";
case Attribute::ClipDistance0:
return "ClipDistance[0]";
case Attribute::ClipDistance1:
return "ClipDistance[1]";
case Attribute::ClipDistance2:
return "ClipDistance[2]";
case Attribute::ClipDistance3:
return "ClipDistance[3]";
case Attribute::ClipDistance4:
return "ClipDistance[4]";
case Attribute::ClipDistance5:
return "ClipDistance[5]";
case Attribute::ClipDistance6:
return "ClipDistance[6]";
case Attribute::ClipDistance7:
return "ClipDistance[7]";
case Attribute::PointSpriteS:
return "PointSprite.S";
case Attribute::PointSpriteT:
return "PointSprite.T";
case Attribute::FogCoordinate:
return "FogCoordinate";
case Attribute::TessellationEvaluationPointU:
return "TessellationEvaluationPoint.U";
case Attribute::TessellationEvaluationPointV:
return "TessellationEvaluationPoint.V";
case Attribute::InstanceId:
return "InstanceId";
case Attribute::VertexId:
return "VertexId";
case Attribute::FixedFncTexture0S:
return "FixedFncTexture[0].S";
case Attribute::FixedFncTexture0T:
return "FixedFncTexture[0].T";
case Attribute::FixedFncTexture0R:
return "FixedFncTexture[0].R";
case Attribute::FixedFncTexture0Q:
return "FixedFncTexture[0].Q";
case Attribute::FixedFncTexture1S:
return "FixedFncTexture[1].S";
case Attribute::FixedFncTexture1T:
return "FixedFncTexture[1].T";
case Attribute::FixedFncTexture1R:
return "FixedFncTexture[1].R";
case Attribute::FixedFncTexture1Q:
return "FixedFncTexture[1].Q";
case Attribute::FixedFncTexture2S:
return "FixedFncTexture[2].S";
case Attribute::FixedFncTexture2T:
return "FixedFncTexture[2].T";
case Attribute::FixedFncTexture2R:
return "FixedFncTexture[2].R";
case Attribute::FixedFncTexture2Q:
return "FixedFncTexture[2].Q";
case Attribute::FixedFncTexture3S:
return "FixedFncTexture[3].S";
case Attribute::FixedFncTexture3T:
return "FixedFncTexture[3].T";
case Attribute::FixedFncTexture3R:
return "FixedFncTexture[3].R";
case Attribute::FixedFncTexture3Q:
return "FixedFncTexture[3].Q";
case Attribute::FixedFncTexture4S:
return "FixedFncTexture[4].S";
case Attribute::FixedFncTexture4T:
return "FixedFncTexture[4].T";
case Attribute::FixedFncTexture4R:
return "FixedFncTexture[4].R";
case Attribute::FixedFncTexture4Q:
return "FixedFncTexture[4].Q";
case Attribute::FixedFncTexture5S:
return "FixedFncTexture[5].S";
case Attribute::FixedFncTexture5T:
return "FixedFncTexture[5].T";
case Attribute::FixedFncTexture5R:
return "FixedFncTexture[5].R";
case Attribute::FixedFncTexture5Q:
return "FixedFncTexture[5].Q";
case Attribute::FixedFncTexture6S:
return "FixedFncTexture[6].S";
case Attribute::FixedFncTexture6T:
return "FixedFncTexture[6].T";
case Attribute::FixedFncTexture6R:
return "FixedFncTexture[6].R";
case Attribute::FixedFncTexture6Q:
return "FixedFncTexture[6].Q";
case Attribute::FixedFncTexture7S:
return "FixedFncTexture[7].S";
case Attribute::FixedFncTexture7T:
return "FixedFncTexture[7].T";
case Attribute::FixedFncTexture7R:
return "FixedFncTexture[7].R";
case Attribute::FixedFncTexture7Q:
return "FixedFncTexture[7].Q";
case Attribute::FixedFncTexture8S:
return "FixedFncTexture[8].S";
case Attribute::FixedFncTexture8T:
return "FixedFncTexture[8].T";
case Attribute::FixedFncTexture8R:
return "FixedFncTexture[8].R";
case Attribute::FixedFncTexture8Q:
return "FixedFncTexture[8].Q";
case Attribute::FixedFncTexture9S:
return "FixedFncTexture[9].S";
case Attribute::FixedFncTexture9T:
return "FixedFncTexture[9].T";
case Attribute::FixedFncTexture9R:
return "FixedFncTexture[9].R";
case Attribute::FixedFncTexture9Q:
return "FixedFncTexture[9].Q";
case Attribute::ViewportMask:
return "ViewportMask";
case Attribute::FrontFace:
return "FrontFace";
case Attribute::BaseInstance:
return "BaseInstance";
case Attribute::BaseVertex:
return "BaseVertex";
case Attribute::DrawID:
return "DrawID";
}
return fmt::format("<reserved attribute {}>", static_cast<int>(attribute));
}
} // namespace Shader::IR

View File

@@ -0,0 +1,288 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <fmt/ranges.h>
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
namespace Shader::IR {
enum class Attribute : u64 {
#define SRIR_ATTRIBUTE_LIST \
SRIR_ATTRIBUTE_ELEM(PrimitiveId, 24) \
SRIR_ATTRIBUTE_ELEM(Layer, 25) \
SRIR_ATTRIBUTE_ELEM(ViewportIndex, 26) \
SRIR_ATTRIBUTE_ELEM(PointSize, 27) \
SRIR_ATTRIBUTE_ELEM(PositionX, 28) \
SRIR_ATTRIBUTE_ELEM(PositionY, 29) \
SRIR_ATTRIBUTE_ELEM(PositionZ, 30) \
SRIR_ATTRIBUTE_ELEM(PositionW, 31) \
SRIR_ATTRIBUTE_ELEM(Generic0X, 32) \
SRIR_ATTRIBUTE_ELEM(Generic0Y, 33) \
SRIR_ATTRIBUTE_ELEM(Generic0Z, 34) \
SRIR_ATTRIBUTE_ELEM(Generic0W, 35) \
SRIR_ATTRIBUTE_ELEM(Generic1X, 36) \
SRIR_ATTRIBUTE_ELEM(Generic1Y, 37) \
SRIR_ATTRIBUTE_ELEM(Generic1Z, 38) \
SRIR_ATTRIBUTE_ELEM(Generic1W, 39) \
SRIR_ATTRIBUTE_ELEM(Generic2X, 40) \
SRIR_ATTRIBUTE_ELEM(Generic2Y, 41) \
SRIR_ATTRIBUTE_ELEM(Generic2Z, 42) \
SRIR_ATTRIBUTE_ELEM(Generic2W, 43) \
SRIR_ATTRIBUTE_ELEM(Generic3X, 44) \
SRIR_ATTRIBUTE_ELEM(Generic3Y, 45) \
SRIR_ATTRIBUTE_ELEM(Generic3Z, 46) \
SRIR_ATTRIBUTE_ELEM(Generic3W, 47) \
SRIR_ATTRIBUTE_ELEM(Generic4X, 48) \
SRIR_ATTRIBUTE_ELEM(Generic4Y, 49) \
SRIR_ATTRIBUTE_ELEM(Generic4Z, 50) \
SRIR_ATTRIBUTE_ELEM(Generic4W, 51) \
SRIR_ATTRIBUTE_ELEM(Generic5X, 52) \
SRIR_ATTRIBUTE_ELEM(Generic5Y, 53) \
SRIR_ATTRIBUTE_ELEM(Generic5Z, 54) \
SRIR_ATTRIBUTE_ELEM(Generic5W, 55) \
SRIR_ATTRIBUTE_ELEM(Generic6X, 56) \
SRIR_ATTRIBUTE_ELEM(Generic6Y, 57) \
SRIR_ATTRIBUTE_ELEM(Generic6Z, 58) \
SRIR_ATTRIBUTE_ELEM(Generic6W, 59) \
SRIR_ATTRIBUTE_ELEM(Generic7X, 60) \
SRIR_ATTRIBUTE_ELEM(Generic7Y, 61) \
SRIR_ATTRIBUTE_ELEM(Generic7Z, 62) \
SRIR_ATTRIBUTE_ELEM(Generic7W, 63) \
SRIR_ATTRIBUTE_ELEM(Generic8X, 64) \
SRIR_ATTRIBUTE_ELEM(Generic8Y, 65) \
SRIR_ATTRIBUTE_ELEM(Generic8Z, 66) \
SRIR_ATTRIBUTE_ELEM(Generic8W, 67) \
SRIR_ATTRIBUTE_ELEM(Generic9X, 68) \
SRIR_ATTRIBUTE_ELEM(Generic9Y, 69) \
SRIR_ATTRIBUTE_ELEM(Generic9Z, 70) \
SRIR_ATTRIBUTE_ELEM(Generic9W, 71) \
SRIR_ATTRIBUTE_ELEM(Generic10X, 72) \
SRIR_ATTRIBUTE_ELEM(Generic10Y, 73) \
SRIR_ATTRIBUTE_ELEM(Generic10Z, 74) \
SRIR_ATTRIBUTE_ELEM(Generic10W, 75) \
SRIR_ATTRIBUTE_ELEM(Generic11X, 76) \
SRIR_ATTRIBUTE_ELEM(Generic11Y, 77) \
SRIR_ATTRIBUTE_ELEM(Generic11Z, 78) \
SRIR_ATTRIBUTE_ELEM(Generic11W, 79) \
SRIR_ATTRIBUTE_ELEM(Generic12X, 80) \
SRIR_ATTRIBUTE_ELEM(Generic12Y, 81) \
SRIR_ATTRIBUTE_ELEM(Generic12Z, 82) \
SRIR_ATTRIBUTE_ELEM(Generic12W, 83) \
SRIR_ATTRIBUTE_ELEM(Generic13X, 84) \
SRIR_ATTRIBUTE_ELEM(Generic13Y, 85) \
SRIR_ATTRIBUTE_ELEM(Generic13Z, 86) \
SRIR_ATTRIBUTE_ELEM(Generic13W, 87) \
SRIR_ATTRIBUTE_ELEM(Generic14X, 88) \
SRIR_ATTRIBUTE_ELEM(Generic14Y, 89) \
SRIR_ATTRIBUTE_ELEM(Generic14Z, 90) \
SRIR_ATTRIBUTE_ELEM(Generic14W, 91) \
SRIR_ATTRIBUTE_ELEM(Generic15X, 92) \
SRIR_ATTRIBUTE_ELEM(Generic15Y, 93) \
SRIR_ATTRIBUTE_ELEM(Generic15Z, 94) \
SRIR_ATTRIBUTE_ELEM(Generic15W, 95) \
SRIR_ATTRIBUTE_ELEM(Generic16X, 96) \
SRIR_ATTRIBUTE_ELEM(Generic16Y, 97) \
SRIR_ATTRIBUTE_ELEM(Generic16Z, 98) \
SRIR_ATTRIBUTE_ELEM(Generic16W, 99) \
SRIR_ATTRIBUTE_ELEM(Generic17X, 100) \
SRIR_ATTRIBUTE_ELEM(Generic17Y, 101) \
SRIR_ATTRIBUTE_ELEM(Generic17Z, 102) \
SRIR_ATTRIBUTE_ELEM(Generic17W, 103) \
SRIR_ATTRIBUTE_ELEM(Generic18X, 104) \
SRIR_ATTRIBUTE_ELEM(Generic18Y, 105) \
SRIR_ATTRIBUTE_ELEM(Generic18Z, 106) \
SRIR_ATTRIBUTE_ELEM(Generic18W, 107) \
SRIR_ATTRIBUTE_ELEM(Generic19X, 108) \
SRIR_ATTRIBUTE_ELEM(Generic19Y, 109) \
SRIR_ATTRIBUTE_ELEM(Generic19Z, 110) \
SRIR_ATTRIBUTE_ELEM(Generic19W, 111) \
SRIR_ATTRIBUTE_ELEM(Generic20X, 112) \
SRIR_ATTRIBUTE_ELEM(Generic20Y, 113) \
SRIR_ATTRIBUTE_ELEM(Generic20Z, 114) \
SRIR_ATTRIBUTE_ELEM(Generic20W, 115) \
SRIR_ATTRIBUTE_ELEM(Generic21X, 116) \
SRIR_ATTRIBUTE_ELEM(Generic21Y, 117) \
SRIR_ATTRIBUTE_ELEM(Generic21Z, 118) \
SRIR_ATTRIBUTE_ELEM(Generic21W, 119) \
SRIR_ATTRIBUTE_ELEM(Generic22X, 120) \
SRIR_ATTRIBUTE_ELEM(Generic22Y, 121) \
SRIR_ATTRIBUTE_ELEM(Generic22Z, 122) \
SRIR_ATTRIBUTE_ELEM(Generic22W, 123) \
SRIR_ATTRIBUTE_ELEM(Generic23X, 124) \
SRIR_ATTRIBUTE_ELEM(Generic23Y, 125) \
SRIR_ATTRIBUTE_ELEM(Generic23Z, 126) \
SRIR_ATTRIBUTE_ELEM(Generic23W, 127) \
SRIR_ATTRIBUTE_ELEM(Generic24X, 128) \
SRIR_ATTRIBUTE_ELEM(Generic24Y, 129) \
SRIR_ATTRIBUTE_ELEM(Generic24Z, 130) \
SRIR_ATTRIBUTE_ELEM(Generic24W, 131) \
SRIR_ATTRIBUTE_ELEM(Generic25X, 132) \
SRIR_ATTRIBUTE_ELEM(Generic25Y, 133) \
SRIR_ATTRIBUTE_ELEM(Generic25Z, 134) \
SRIR_ATTRIBUTE_ELEM(Generic25W, 135) \
SRIR_ATTRIBUTE_ELEM(Generic26X, 136) \
SRIR_ATTRIBUTE_ELEM(Generic26Y, 137) \
SRIR_ATTRIBUTE_ELEM(Generic26Z, 138) \
SRIR_ATTRIBUTE_ELEM(Generic26W, 139) \
SRIR_ATTRIBUTE_ELEM(Generic27X, 140) \
SRIR_ATTRIBUTE_ELEM(Generic27Y, 141) \
SRIR_ATTRIBUTE_ELEM(Generic27Z, 142) \
SRIR_ATTRIBUTE_ELEM(Generic27W, 143) \
SRIR_ATTRIBUTE_ELEM(Generic28X, 144) \
SRIR_ATTRIBUTE_ELEM(Generic28Y, 145) \
SRIR_ATTRIBUTE_ELEM(Generic28Z, 146) \
SRIR_ATTRIBUTE_ELEM(Generic28W, 147) \
SRIR_ATTRIBUTE_ELEM(Generic29X, 148) \
SRIR_ATTRIBUTE_ELEM(Generic29Y, 149) \
SRIR_ATTRIBUTE_ELEM(Generic29Z, 150) \
SRIR_ATTRIBUTE_ELEM(Generic29W, 151) \
SRIR_ATTRIBUTE_ELEM(Generic30X, 152) \
SRIR_ATTRIBUTE_ELEM(Generic30Y, 153) \
SRIR_ATTRIBUTE_ELEM(Generic30Z, 154) \
SRIR_ATTRIBUTE_ELEM(Generic30W, 155) \
SRIR_ATTRIBUTE_ELEM(Generic31X, 156) \
SRIR_ATTRIBUTE_ELEM(Generic31Y, 157) \
SRIR_ATTRIBUTE_ELEM(Generic31Z, 158) \
SRIR_ATTRIBUTE_ELEM(Generic31W, 159) \
SRIR_ATTRIBUTE_ELEM(ColorFrontDiffuseR, 160) \
SRIR_ATTRIBUTE_ELEM(ColorFrontDiffuseG, 161) \
SRIR_ATTRIBUTE_ELEM(ColorFrontDiffuseB, 162) \
SRIR_ATTRIBUTE_ELEM(ColorFrontDiffuseA, 163) \
SRIR_ATTRIBUTE_ELEM(ColorFrontSpecularR, 164) \
SRIR_ATTRIBUTE_ELEM(ColorFrontSpecularG, 165) \
SRIR_ATTRIBUTE_ELEM(ColorFrontSpecularB, 166) \
SRIR_ATTRIBUTE_ELEM(ColorFrontSpecularA, 167) \
SRIR_ATTRIBUTE_ELEM(ColorBackDiffuseR, 168) \
SRIR_ATTRIBUTE_ELEM(ColorBackDiffuseG, 169) \
SRIR_ATTRIBUTE_ELEM(ColorBackDiffuseB, 170) \
SRIR_ATTRIBUTE_ELEM(ColorBackDiffuseA, 171) \
SRIR_ATTRIBUTE_ELEM(ColorBackSpecularR, 172) \
SRIR_ATTRIBUTE_ELEM(ColorBackSpecularG, 173) \
SRIR_ATTRIBUTE_ELEM(ColorBackSpecularB, 174) \
SRIR_ATTRIBUTE_ELEM(ColorBackSpecularA, 175) \
SRIR_ATTRIBUTE_ELEM(ClipDistance0, 176) \
SRIR_ATTRIBUTE_ELEM(ClipDistance1, 177) \
SRIR_ATTRIBUTE_ELEM(ClipDistance2, 178) \
SRIR_ATTRIBUTE_ELEM(ClipDistance3, 179) \
SRIR_ATTRIBUTE_ELEM(ClipDistance4, 180) \
SRIR_ATTRIBUTE_ELEM(ClipDistance5, 181) \
SRIR_ATTRIBUTE_ELEM(ClipDistance6, 182) \
SRIR_ATTRIBUTE_ELEM(ClipDistance7, 183) \
SRIR_ATTRIBUTE_ELEM(PointSpriteS, 184) \
SRIR_ATTRIBUTE_ELEM(PointSpriteT, 185) \
SRIR_ATTRIBUTE_ELEM(FogCoordinate, 186) \
SRIR_ATTRIBUTE_ELEM(TessellationEvaluationPointU, 188) \
SRIR_ATTRIBUTE_ELEM(TessellationEvaluationPointV, 189) \
SRIR_ATTRIBUTE_ELEM(InstanceId, 190) \
SRIR_ATTRIBUTE_ELEM(VertexId, 191) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture0S, 192) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture0T, 193) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture0R, 194) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture0Q, 195) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture1S, 196) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture1T, 197) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture1R, 198) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture1Q, 199) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture2S, 200) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture2T, 201) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture2R, 202) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture2Q, 203) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture3S, 204) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture3T, 205) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture3R, 206) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture3Q, 207) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture4S, 208) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture4T, 209) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture4R, 210) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture4Q, 211) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture5S, 212) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture5T, 213) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture5R, 214) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture5Q, 215) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture6S, 216) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture6T, 217) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture6R, 218) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture6Q, 219) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture7S, 220) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture7T, 221) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture7R, 222) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture7Q, 223) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture8S, 224) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture8T, 225) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture8R, 226) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture8Q, 227) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture9S, 228) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture9T, 229) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture9R, 230) \
SRIR_ATTRIBUTE_ELEM(FixedFncTexture9Q, 231) \
SRIR_ATTRIBUTE_ELEM(ViewportMask, 232) \
SRIR_ATTRIBUTE_ELEM(FrontFace, 255) \
/* Implementation attributes */ \
SRIR_ATTRIBUTE_ELEM(BaseInstance, 256) \
SRIR_ATTRIBUTE_ELEM(BaseVertex, 257) \
SRIR_ATTRIBUTE_ELEM(DrawID, 258)
#define SRIR_ATTRIBUTE_ELEM(n, v) n = v,
SRIR_ATTRIBUTE_LIST
#undef SRIR_ATTRIBUTE_ELEM
};
constexpr size_t NUM_GENERICS = 32;
constexpr size_t NUM_FIXEDFNCTEXTURE = 10;
[[nodiscard]] inline bool IsGeneric(Attribute attribute) noexcept {
return attribute >= Attribute::Generic0X && attribute <= Attribute::Generic31X;
}
[[nodiscard]] inline bool IsLegacyAttribute(Attribute attribute) noexcept {
return (attribute >= Attribute::ColorFrontDiffuseR &&
attribute <= Attribute::ColorBackSpecularA) ||
attribute == Attribute::FogCoordinate ||
(attribute >= Attribute::FixedFncTexture0S &&
attribute <= Attribute::FixedFncTexture9Q);
}
[[nodiscard]] inline u32 GenericAttributeIndex(Attribute attribute) {
if (!IsGeneric(attribute))
throw InvalidArgument("Attribute is not generic {}", attribute);
return (u32(attribute) - u32(Attribute::Generic0X)) / 4u;
}
[[nodiscard]] inline u32 GenericAttributeElement(Attribute attribute) {
if (!IsGeneric(attribute))
throw InvalidArgument("Attribute is not generic {}", attribute);
return u32(attribute) % 4;
}
[[nodiscard]] inline std::string NameOf(Attribute attribute) {
switch (attribute) {
#define SRIR_ATTRIBUTE_ELEM(n, v) case Attribute::n: return #n;
SRIR_ATTRIBUTE_LIST
#undef SRIR_ATTRIBUTE_ELEM
default:
return fmt::format("<reserved attribute {}>", int(attribute));
}
}
[[nodiscard]] constexpr IR::Attribute operator+(IR::Attribute attribute, size_t value) noexcept {
return IR::Attribute(size_t(attribute) + value);
}
} // namespace Shader::IR
template <>
struct fmt::formatter<Shader::IR::Attribute> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::IR::Attribute& attribute, FormatContext& ctx) const {
return fmt::format_to(ctx.out(), "{}", Shader::IR::NameOf(attribute));
}
};

View File

@@ -0,0 +1,154 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <initializer_list>
#include <map>
#include "common/common_types.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
Block::Block(ObjectPool<Inst>& inst_pool_) : inst_pool{&inst_pool_} {}
Block::~Block() = default;
void Block::AppendNewInst(Opcode op, std::initializer_list<Value> args) {
PrependNewInst(instructions.end(), op, args);
}
Block::iterator Block::PrependNewInst(iterator insertion_point, const Inst& base_inst) {
Inst* const inst{inst_pool->Create(base_inst)};
return instructions.insert(insertion_point, *inst);
}
Block::iterator Block::PrependNewInst(iterator insertion_point, Opcode op,
std::initializer_list<Value> args, u32 flags) {
Inst* const inst{inst_pool->Create(op, flags)};
const auto result_it{instructions.insert(insertion_point, *inst)};
if (inst->NumArgs() != args.size()) {
throw InvalidArgument("Invalid number of arguments {} in {}", args.size(), op);
}
std::ranges::for_each(args, [inst, index = size_t{0}](const Value& arg) mutable {
inst->SetArg(index, arg);
++index;
});
return result_it;
}
void Block::AddBranch(Block* block) {
if (std::ranges::find(imm_successors, block) != imm_successors.end()) {
throw LogicError("Successor already inserted");
}
if (std::ranges::find(block->imm_predecessors, this) != block->imm_predecessors.end()) {
throw LogicError("Predecessor already inserted");
}
imm_successors.push_back(block);
block->imm_predecessors.push_back(this);
}
static std::string BlockToIndex(const std::map<const Block*, size_t>& block_to_index,
Block* block) {
if (const auto it{block_to_index.find(block)}; it != block_to_index.end()) {
return fmt::format("{{Block ${}}}", it->second);
}
return fmt::format("$<unknown block {:016x}>", reinterpret_cast<u64>(block));
}
static size_t InstIndex(std::map<const Inst*, size_t>& inst_to_index, size_t& inst_index,
const Inst* inst) {
const auto [it, is_inserted]{inst_to_index.emplace(inst, inst_index + 1)};
if (is_inserted) {
++inst_index;
}
return it->second;
}
static std::string ArgToIndex(std::map<const Inst*, size_t>& inst_to_index, size_t& inst_index,
const Value& arg) {
if (arg.IsEmpty()) {
return "<null>";
}
if (!arg.IsImmediate() || arg.IsIdentity()) {
return fmt::format("%{}", InstIndex(inst_to_index, inst_index, arg.Inst()));
}
switch (arg.Type()) {
case Type::U1:
return fmt::format("#{}", arg.U1() ? "true" : "false");
case Type::U8:
return fmt::format("#{}", arg.U8());
case Type::U16:
return fmt::format("#{}", arg.U16());
case Type::U32:
return fmt::format("#{}", arg.U32());
case Type::U64:
return fmt::format("#{}", arg.U64());
case Type::F32:
return fmt::format("#{}", arg.F32());
case Type::Reg:
return fmt::format("{}", arg.Reg());
case Type::Pred:
return fmt::format("{}", arg.Pred());
case Type::Attribute:
return fmt::format("{}", arg.Attribute());
default:
return "<unknown immediate type>";
}
}
std::string DumpBlock(const Block& block) {
size_t inst_index{0};
std::map<const Inst*, size_t> inst_to_index;
return DumpBlock(block, {}, inst_to_index, inst_index);
}
std::string DumpBlock(const Block& block, const std::map<const Block*, size_t>& block_to_index,
std::map<const Inst*, size_t>& inst_to_index, size_t& inst_index) {
std::string ret{"Block"};
if (const auto it{block_to_index.find(&block)}; it != block_to_index.end()) {
ret += fmt::format(" ${}", it->second);
}
ret += '\n';
for (const Inst& inst : block) {
const Opcode op{inst.GetOpcode()};
ret += fmt::format("[{:016x}] ", reinterpret_cast<u64>(&inst));
if (TypeOf(op) != Type::Void) {
ret += fmt::format("%{:<5} = {}", InstIndex(inst_to_index, inst_index, &inst), op);
} else {
ret += fmt::format(" {}", op); // '%00000 = ' -> 1 + 5 + 3 = 9 spaces
}
const size_t arg_count{inst.NumArgs()};
for (size_t arg_index = 0; arg_index < arg_count; ++arg_index) {
const Value arg{inst.Arg(arg_index)};
const std::string arg_str{ArgToIndex(inst_to_index, inst_index, arg)};
ret += arg_index != 0 ? ", " : " ";
if (op == Opcode::Phi) {
ret += fmt::format("[ {}, {} ]", arg_str,
BlockToIndex(block_to_index, inst.PhiBlock(arg_index)));
} else {
ret += arg_str;
}
if (op != Opcode::Phi) {
const Type actual_type{arg.Type()};
const Type expected_type{ArgTypeOf(op, arg_index)};
if (!AreTypesCompatible(actual_type, expected_type)) {
ret += fmt::format("<type error: {} != {}>", actual_type, expected_type);
}
}
}
if (TypeOf(op) != Type::Void) {
ret += fmt::format(" (uses: {})\n", inst.UseCount());
} else {
ret += '\n';
}
}
return ret;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,204 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <initializer_list>
#include <map>
#include <span>
#include <vector>
#include <bit>
#include <numeric>
#include <boost/intrusive/list.hpp>
#include "common/common_types.h"
#include "shader_recompiler/frontend/ir/condition.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/object_pool.h"
namespace Shader::IR {
class Block {
public:
using InstructionList = boost::intrusive::list<Inst>;
using size_type = InstructionList::size_type;
using iterator = InstructionList::iterator;
using const_iterator = InstructionList::const_iterator;
using reverse_iterator = InstructionList::reverse_iterator;
using const_reverse_iterator = InstructionList::const_reverse_iterator;
explicit Block(ObjectPool<Inst>& inst_pool_);
~Block();
Block(const Block&) = delete;
Block& operator=(const Block&) = delete;
Block(Block&&) = default;
Block& operator=(Block&&) = default;
/// Appends a new instruction to the end of this basic block.
void AppendNewInst(Opcode op, std::initializer_list<Value> args);
/// Prepends a copy of an instruction to this basic block before the insertion point.
iterator PrependNewInst(iterator insertion_point, const Inst& base_inst);
/// Prepends a new instruction to this basic block before the insertion point.
iterator PrependNewInst(iterator insertion_point, Opcode op,
std::initializer_list<Value> args = {}, u32 flags = 0);
/// Adds a new branch to this basic block.
void AddBranch(Block* block);
/// Gets a mutable reference to the instruction list for this basic block.
[[nodiscard]] InstructionList& Instructions() noexcept {
return instructions;
}
/// Gets an immutable reference to the instruction list for this basic block.
[[nodiscard]] const InstructionList& Instructions() const noexcept {
return instructions;
}
/// Gets an immutable span to the immediate predecessors.
[[nodiscard]] std::span<Block* const> ImmPredecessors() const noexcept {
return imm_predecessors;
}
/// Gets an immutable span to the immediate successors.
[[nodiscard]] std::span<Block* const> ImmSuccessors() const noexcept {
return imm_successors;
}
/// Intrusively store the host definition of this instruction.
template <typename DefinitionType>
void SetDefinition(DefinitionType def) {
definition = std::bit_cast<u32>(def);
}
/// Return the intrusively stored host definition of this instruction.
template <typename DefinitionType>
[[nodiscard]] DefinitionType Definition() const noexcept {
return std::bit_cast<DefinitionType>(definition);
}
void SetSsaRegValue(IR::Reg reg, const Value& value) noexcept {
ssa_reg_values[RegIndex(reg)] = value;
}
const Value& SsaRegValue(IR::Reg reg) const noexcept {
return ssa_reg_values[RegIndex(reg)];
}
void SsaSeal() noexcept {
is_ssa_sealed = true;
}
[[nodiscard]] bool IsSsaSealed() const noexcept {
return is_ssa_sealed;
}
[[nodiscard]] bool empty() const {
return instructions.empty();
}
[[nodiscard]] size_type size() const {
return instructions.size();
}
[[nodiscard]] Inst& front() {
return instructions.front();
}
[[nodiscard]] const Inst& front() const {
return instructions.front();
}
[[nodiscard]] Inst& back() {
return instructions.back();
}
[[nodiscard]] const Inst& back() const {
return instructions.back();
}
[[nodiscard]] iterator begin() {
return instructions.begin();
}
[[nodiscard]] const_iterator begin() const {
return instructions.begin();
}
[[nodiscard]] iterator end() {
return instructions.end();
}
[[nodiscard]] const_iterator end() const {
return instructions.end();
}
[[nodiscard]] reverse_iterator rbegin() {
return instructions.rbegin();
}
[[nodiscard]] const_reverse_iterator rbegin() const {
return instructions.rbegin();
}
[[nodiscard]] reverse_iterator rend() {
return instructions.rend();
}
[[nodiscard]] const_reverse_iterator rend() const {
return instructions.rend();
}
[[nodiscard]] const_iterator cbegin() const {
return instructions.cbegin();
}
[[nodiscard]] const_iterator cend() const {
return instructions.cend();
}
[[nodiscard]] const_reverse_iterator crbegin() const {
return instructions.crbegin();
}
[[nodiscard]] const_reverse_iterator crend() const {
return instructions.crend();
}
// Set the order of the block, it can be set pre order, the user decides
void SetOrder(u32 new_order) {
order = new_order;
}
// Get the order of the block.
// The higher, the closer is the block to the end.
[[nodiscard]] u32 GetOrder() const {
return order;
}
private:
/// Memory pool for instruction list
ObjectPool<Inst>* inst_pool;
/// List of instructions in this block
InstructionList instructions;
/// Block immediate predecessors
std::vector<Block*> imm_predecessors;
/// Block immediate successors
std::vector<Block*> imm_successors;
/// Intrusively store the value of a register in the block.
std::array<Value, NUM_REGS> ssa_reg_values;
/// Intrusively store if the block is sealed in the SSA pass.
bool is_ssa_sealed{false};
/// Intrusively stored host definition of this block.
u32 definition{};
/// Order of the block.
u32 order{};
};
using BlockList = std::vector<Block*>;
[[nodiscard]] std::string DumpBlock(const Block& block);
[[nodiscard]] std::string DumpBlock(const Block& block,
const std::map<const Block*, size_t>& block_to_index,
std::map<const Inst*, size_t>& inst_to_index,
size_t& inst_index);
} // namespace Shader::IR

View File

@@ -0,0 +1,55 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <optional>
#include <type_traits>
#include <queue>
#include <boost/container/small_vector.hpp>
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
template <typename Pred>
auto BreadthFirstSearch(const Value& value, Pred&& pred)
-> std::invoke_result_t<Pred, const Inst*> {
if (value.IsImmediate()) {
// Nothing to do with immediates
return std::nullopt;
}
// Breadth-first search visiting the right most arguments first
// Small vector has been determined from shaders in Super Smash Bros. Ultimate
boost::container::small_vector<const Inst*, 2> visited;
std::queue<const Inst*> queue;
queue.push(value.InstRecursive());
while (!queue.empty()) {
// Pop one instruction from the queue
const Inst* const inst{queue.front()};
queue.pop();
if (const std::optional result = pred(inst)) {
// This is the instruction we were looking for
return result;
}
// Visit the right most arguments first
for (size_t arg = inst->NumArgs(); arg--;) {
const Value arg_value{inst->Arg(arg)};
if (arg_value.IsImmediate()) {
continue;
}
// Queue instruction if it hasn't been visited
const Inst* const arg_inst{arg_value.InstRecursive()};
if (std::ranges::find(visited, arg_inst) == visited.end()) {
visited.push_back(arg_inst);
queue.push(arg_inst);
}
}
}
// SSA tree has been traversed and the result hasn't been found
return std::nullopt;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,28 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string>
#include <fmt/ranges.h>
#include "shader_recompiler/frontend/ir/condition.h"
namespace Shader::IR {
std::string NameOf(Condition condition) {
std::string ret;
if (condition.GetFlowTest() != FlowTest::T) {
ret = fmt::to_string(condition.GetFlowTest());
}
const auto [pred, negated]{condition.GetPred()};
if (!ret.empty()) {
ret += '&';
}
if (negated) {
ret += '!';
}
ret += fmt::to_string(pred);
return ret;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,58 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <string>
#include <fmt/ranges.h>
#include "common/common_types.h"
#include "shader_recompiler/frontend/ir/flow_test.h"
#include "shader_recompiler/frontend/ir/pred.h"
namespace Shader::IR {
class Condition {
public:
Condition() noexcept = default;
explicit Condition(FlowTest flow_test_, Pred pred_, bool pred_negated_ = false) noexcept
: flow_test{static_cast<u16>(flow_test_)}, pred{static_cast<u8>(pred_)},
pred_negated{pred_negated_ ? u8{1} : u8{0}} {}
explicit Condition(Pred pred_, bool pred_negated_ = false) noexcept
: Condition(FlowTest::T, pred_, pred_negated_) {}
explicit Condition(bool value) : Condition(Pred::PT, !value) {}
auto operator<=>(const Condition&) const noexcept = default;
[[nodiscard]] IR::FlowTest GetFlowTest() const noexcept {
return static_cast<IR::FlowTest>(flow_test);
}
[[nodiscard]] std::pair<IR::Pred, bool> GetPred() const noexcept {
return {static_cast<IR::Pred>(pred), pred_negated != 0};
}
private:
u16 flow_test;
u8 pred;
u8 pred_negated;
};
std::string NameOf(Condition condition);
} // namespace Shader::IR
template <>
struct fmt::formatter<Shader::IR::Condition> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::IR::Condition& cond, FormatContext& ctx) const {
return fmt::format_to(ctx.out(), "{}", Shader::IR::NameOf(cond));
}
};

View File

@@ -0,0 +1,82 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <string>
#include <fmt/format.h>
#include "shader_recompiler/frontend/ir/flow_test.h"
namespace Shader::IR {
std::string NameOf(FlowTest flow_test) {
switch (flow_test) {
case FlowTest::F:
return "F";
case FlowTest::LT:
return "LT";
case FlowTest::EQ:
return "EQ";
case FlowTest::LE:
return "LE";
case FlowTest::GT:
return "GT";
case FlowTest::NE:
return "NE";
case FlowTest::GE:
return "GE";
case FlowTest::NUM:
return "NUM";
case FlowTest::NaN:
return "NAN";
case FlowTest::LTU:
return "LTU";
case FlowTest::EQU:
return "EQU";
case FlowTest::LEU:
return "LEU";
case FlowTest::GTU:
return "GTU";
case FlowTest::NEU:
return "NEU";
case FlowTest::GEU:
return "GEU";
case FlowTest::T:
return "T";
case FlowTest::OFF:
return "OFF";
case FlowTest::LO:
return "LO";
case FlowTest::SFF:
return "SFF";
case FlowTest::LS:
return "LS";
case FlowTest::HI:
return "HI";
case FlowTest::SFT:
return "SFT";
case FlowTest::HS:
return "HS";
case FlowTest::OFT:
return "OFT";
case FlowTest::CSM_TA:
return "CSM_TA";
case FlowTest::CSM_TR:
return "CSM_TR";
case FlowTest::CSM_MX:
return "CSM_MX";
case FlowTest::FCSM_TA:
return "FCSM_TA";
case FlowTest::FCSM_TR:
return "FCSM_TR";
case FlowTest::FCSM_MX:
return "FCSM_MX";
case FlowTest::RLE:
return "RLE";
case FlowTest::RGT:
return "RGT";
}
return fmt::format("<invalid flow test {}>", static_cast<int>(flow_test));
}
} // namespace Shader::IR

View File

@@ -0,0 +1,77 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <string>
#include <fmt/ranges.h>
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
namespace Shader::IR {
enum class FlowTest : u64 {
#define SRIR_FLOW_TEST_LIST \
SRIR_FLOW_TEST_ELEM(F) \
SRIR_FLOW_TEST_ELEM(LT) \
SRIR_FLOW_TEST_ELEM(EQ) \
SRIR_FLOW_TEST_ELEM(LE) \
SRIR_FLOW_TEST_ELEM(GT) \
SRIR_FLOW_TEST_ELEM(NE) \
SRIR_FLOW_TEST_ELEM(GE) \
SRIR_FLOW_TEST_ELEM(NUM) \
SRIR_FLOW_TEST_ELEM(NaN) \
SRIR_FLOW_TEST_ELEM(LTU) \
SRIR_FLOW_TEST_ELEM(EQU) \
SRIR_FLOW_TEST_ELEM(LEU) \
SRIR_FLOW_TEST_ELEM(GTU) \
SRIR_FLOW_TEST_ELEM(NEU) \
SRIR_FLOW_TEST_ELEM(GEU) \
SRIR_FLOW_TEST_ELEM(T) \
SRIR_FLOW_TEST_ELEM(OFF) \
SRIR_FLOW_TEST_ELEM(LO) \
SRIR_FLOW_TEST_ELEM(SFF) \
SRIR_FLOW_TEST_ELEM(LS) \
SRIR_FLOW_TEST_ELEM(HI) \
SRIR_FLOW_TEST_ELEM(SFT) \
SRIR_FLOW_TEST_ELEM(HS) \
SRIR_FLOW_TEST_ELEM(OFT) \
SRIR_FLOW_TEST_ELEM(CSM_TA) \
SRIR_FLOW_TEST_ELEM(CSM_TR) \
SRIR_FLOW_TEST_ELEM(CSM_MX) \
SRIR_FLOW_TEST_ELEM(FCSM_TA) \
SRIR_FLOW_TEST_ELEM(FCSM_TR) \
SRIR_FLOW_TEST_ELEM(FCSM_MX) \
SRIR_FLOW_TEST_ELEM(RLE) \
SRIR_FLOW_TEST_ELEM(RGT)
#define SRIR_FLOW_TEST_ELEM(n) n,
SRIR_FLOW_TEST_LIST
#undef SRIR_FLOW_TEST_ELEM
};
[[nodiscard]] inline std::string NameOf(FlowTest flow_test) {
switch (flow_test) {
#define SRIR_FLOW_TEST_ELEM(n) case FlowTest::n: return #n;
SRIR_FLOW_TEST_LIST
#undef SRIR_FLOW_TEST_ELEM
default:
return fmt::format("<invalid flow test {}>", int(flow_test));
}
}
} // namespace Shader::IR
template <>
struct fmt::formatter<Shader::IR::FlowTest> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::IR::FlowTest& flow_test, FormatContext& ctx) const {
return fmt::format_to(ctx.out(), "{}", Shader::IR::NameOf(flow_test));
}
};

File diff suppressed because it is too large Load Diff

View File

@@ -0,0 +1,435 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <cstring>
#include <type_traits>
#include "shader_recompiler/frontend/ir/attribute.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/modifiers.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
class IREmitter {
public:
explicit IREmitter(Block& block_) : block{&block_}, insertion_point{block->end()} {}
explicit IREmitter(Block& block_, Block::iterator insertion_point_)
: block{&block_}, insertion_point{insertion_point_} {}
Block* block;
[[nodiscard]] U1 Imm1(bool value) const;
[[nodiscard]] U8 Imm8(u8 value) const;
[[nodiscard]] U16 Imm16(u16 value) const;
[[nodiscard]] U32 Imm32(u32 value) const;
[[nodiscard]] U32 Imm32(s32 value) const;
[[nodiscard]] F32 Imm32(f32 value) const;
[[nodiscard]] U64 Imm64(u64 value) const;
[[nodiscard]] U64 Imm64(s64 value) const;
[[nodiscard]] F64 Imm64(f64 value) const;
U1 ConditionRef(const U1& value);
void Reference(const Value& value);
void PhiMove(IR::Inst& phi, const Value& value);
void Prologue();
void Epilogue();
void DemoteToHelperInvocation();
void EmitVertex(const U32& stream);
void EndPrimitive(const U32& stream);
[[nodiscard]] U32 GetReg(IR::Reg reg);
void SetReg(IR::Reg reg, const U32& value);
[[nodiscard]] U1 GetPred(IR::Pred pred, bool is_negated = false);
void SetPred(IR::Pred pred, const U1& value);
[[nodiscard]] U1 GetGotoVariable(u32 id);
void SetGotoVariable(u32 id, const U1& value);
[[nodiscard]] U32 GetIndirectBranchVariable();
void SetIndirectBranchVariable(const U32& value);
[[nodiscard]] U32 GetCbuf(const U32& binding, const U32& byte_offset);
[[nodiscard]] Value GetCbuf(const U32& binding, const U32& byte_offset, size_t bitsize,
bool is_signed);
[[nodiscard]] F32 GetFloatCbuf(const U32& binding, const U32& byte_offset);
[[nodiscard]] U1 GetZFlag();
[[nodiscard]] U1 GetSFlag();
[[nodiscard]] U1 GetCFlag();
[[nodiscard]] U1 GetOFlag();
void SetZFlag(const U1& value);
void SetSFlag(const U1& value);
void SetCFlag(const U1& value);
void SetOFlag(const U1& value);
[[nodiscard]] U1 Condition(IR::Condition cond);
[[nodiscard]] U1 GetFlowTestResult(FlowTest test);
[[nodiscard]] F32 GetAttribute(IR::Attribute attribute);
[[nodiscard]] F32 GetAttribute(IR::Attribute attribute, const U32& vertex);
[[nodiscard]] U32 GetAttributeU32(IR::Attribute attribute);
[[nodiscard]] U32 GetAttributeU32(IR::Attribute attribute, const U32& vertex);
void SetAttribute(IR::Attribute attribute, const F32& value, const U32& vertex);
[[nodiscard]] F32 GetAttributeIndexed(const U32& phys_address);
[[nodiscard]] F32 GetAttributeIndexed(const U32& phys_address, const U32& vertex);
void SetAttributeIndexed(const U32& phys_address, const F32& value, const U32& vertex);
[[nodiscard]] F32 GetPatch(Patch patch);
void SetPatch(Patch patch, const F32& value);
void SetFragColor(u32 index, u32 component, const F32& value);
void SetSampleMask(const U32& value);
void SetFragDepth(const F32& value);
[[nodiscard]] U32 WorkgroupIdX();
[[nodiscard]] U32 WorkgroupIdY();
[[nodiscard]] U32 WorkgroupIdZ();
[[nodiscard]] Value LocalInvocationId();
[[nodiscard]] U32 LocalInvocationIdX();
[[nodiscard]] U32 LocalInvocationIdY();
[[nodiscard]] U32 LocalInvocationIdZ();
[[nodiscard]] U32 InvocationId();
[[nodiscard]] U32 InvocationInfo();
[[nodiscard]] U32 SampleId();
[[nodiscard]] U1 IsHelperInvocation();
[[nodiscard]] U32 SR_WScaleFactorXY();
[[nodiscard]] U32 SR_WScaleFactorZ();
[[nodiscard]] F32 YDirection();
[[nodiscard]] F32 ResolutionDownFactor();
[[nodiscard]] F32 RenderAreaWidth();
[[nodiscard]] F32 RenderAreaHeight();
[[nodiscard]] U32 LaneId();
[[nodiscard]] U32 LoadGlobalU8(const U64& address);
[[nodiscard]] U32 LoadGlobalS8(const U64& address);
[[nodiscard]] U32 LoadGlobalU16(const U64& address);
[[nodiscard]] U32 LoadGlobalS16(const U64& address);
[[nodiscard]] U32 LoadGlobal32(const U64& address);
[[nodiscard]] Value LoadGlobal64(const U64& address);
[[nodiscard]] Value LoadGlobal128(const U64& address);
void WriteGlobalU8(const U64& address, const U32& value);
void WriteGlobalS8(const U64& address, const U32& value);
void WriteGlobalU16(const U64& address, const U32& value);
void WriteGlobalS16(const U64& address, const U32& value);
void WriteGlobal32(const U64& address, const U32& value);
void WriteGlobal64(const U64& address, const IR::Value& vector);
void WriteGlobal128(const U64& address, const IR::Value& vector);
[[nodiscard]] U32 LoadLocal(const U32& word_offset);
void WriteLocal(const U32& word_offset, const U32& value);
[[nodiscard]] Value LoadShared(int bit_size, bool is_signed, const U32& offset);
void WriteShared(int bit_size, const U32& offset, const Value& value);
[[nodiscard]] U1 GetZeroFromOp(const Value& op);
[[nodiscard]] U1 GetSignFromOp(const Value& op);
[[nodiscard]] U1 GetCarryFromOp(const Value& op);
[[nodiscard]] U1 GetOverflowFromOp(const Value& op);
[[nodiscard]] U1 GetSparseFromOp(const Value& op);
[[nodiscard]] U1 GetInBoundsFromOp(const Value& op);
[[nodiscard]] Value CompositeConstruct(const Value& e1, const Value& e2);
[[nodiscard]] Value CompositeConstruct(const Value& e1, const Value& e2, const Value& e3);
[[nodiscard]] Value CompositeConstruct(const Value& e1, const Value& e2, const Value& e3,
const Value& e4);
[[nodiscard]] Value CompositeExtract(const Value& vector, size_t element);
[[nodiscard]] Value CompositeInsert(const Value& vector, const Value& object, size_t element);
[[nodiscard]] Value Select(const U1& condition, const Value& true_value,
const Value& false_value);
void Barrier();
void WorkgroupMemoryBarrier();
void DeviceMemoryBarrier();
template <typename Dest, typename Source>
[[nodiscard]] Dest BitCast(const Source& value);
[[nodiscard]] U64 PackUint2x32(const Value& vector);
[[nodiscard]] Value UnpackUint2x32(const U64& value);
[[nodiscard]] U32 PackFloat2x16(const Value& vector);
[[nodiscard]] Value UnpackFloat2x16(const U32& value);
[[nodiscard]] U32 PackHalf2x16(const Value& vector);
[[nodiscard]] Value UnpackHalf2x16(const U32& value);
[[nodiscard]] F64 PackDouble2x32(const Value& vector);
[[nodiscard]] Value UnpackDouble2x32(const F64& value);
[[nodiscard]] F16F32F64 FPAdd(const F16F32F64& a, const F16F32F64& b, FpControl control = {});
[[nodiscard]] F16F32F64 FPMul(const F16F32F64& a, const F16F32F64& b, FpControl control = {});
[[nodiscard]] F16F32F64 FPFma(const F16F32F64& a, const F16F32F64& b, const F16F32F64& c,
FpControl control = {});
[[nodiscard]] F16F32F64 FPAbs(const F16F32F64& value);
[[nodiscard]] F16F32F64 FPNeg(const F16F32F64& value);
[[nodiscard]] F16F32F64 FPAbsNeg(const F16F32F64& value, bool abs, bool neg);
[[nodiscard]] F32 FPCos(const F32& value);
[[nodiscard]] F32 FPSin(const F32& value);
[[nodiscard]] F32 FPExp2(const F32& value);
[[nodiscard]] F32 FPLog2(const F32& value);
[[nodiscard]] F32F64 FPRecip(const F32F64& value);
[[nodiscard]] F32F64 FPRecipSqrt(const F32F64& value);
[[nodiscard]] F32 FPSqrt(const F32& value);
[[nodiscard]] F16F32F64 FPSaturate(const F16F32F64& value);
[[nodiscard]] F16F32F64 FPClamp(const F16F32F64& value, const F16F32F64& min_value,
const F16F32F64& max_value);
[[nodiscard]] F16F32F64 FPRoundEven(const F16F32F64& value, FpControl control = {});
[[nodiscard]] F16F32F64 FPFloor(const F16F32F64& value, FpControl control = {});
[[nodiscard]] F16F32F64 FPCeil(const F16F32F64& value, FpControl control = {});
[[nodiscard]] F16F32F64 FPTrunc(const F16F32F64& value, FpControl control = {});
[[nodiscard]] U1 FPEqual(const F16F32F64& lhs, const F16F32F64& rhs, FpControl control = {},
bool ordered = true);
[[nodiscard]] U1 FPNotEqual(const F16F32F64& lhs, const F16F32F64& rhs, FpControl control = {},
bool ordered = true);
[[nodiscard]] U1 FPLessThan(const F16F32F64& lhs, const F16F32F64& rhs, FpControl control = {},
bool ordered = true);
[[nodiscard]] U1 FPGreaterThan(const F16F32F64& lhs, const F16F32F64& rhs,
FpControl control = {}, bool ordered = true);
[[nodiscard]] U1 FPLessThanEqual(const F16F32F64& lhs, const F16F32F64& rhs,
FpControl control = {}, bool ordered = true);
[[nodiscard]] U1 FPGreaterThanEqual(const F16F32F64& lhs, const F16F32F64& rhs,
FpControl control = {}, bool ordered = true);
[[nodiscard]] U1 FPIsNan(const F16F32F64& value);
[[nodiscard]] U1 FPOrdered(const F16F32F64& lhs, const F16F32F64& rhs);
[[nodiscard]] U1 FPUnordered(const F16F32F64& lhs, const F16F32F64& rhs);
[[nodiscard]] F32F64 FPMax(const F32F64& lhs, const F32F64& rhs, FpControl control = {});
[[nodiscard]] F32F64 FPMin(const F32F64& lhs, const F32F64& rhs, FpControl control = {});
[[nodiscard]] U32U64 IAdd(const U32U64& a, const U32U64& b);
[[nodiscard]] U32U64 ISub(const U32U64& a, const U32U64& b);
[[nodiscard]] U32 IMul(const U32& a, const U32& b);
[[nodiscard]] U32 IDiv(const U32& a, const U32& b, bool is_signed = false);
[[nodiscard]] U32U64 INeg(const U32U64& value);
[[nodiscard]] U32U64 IAbs(const U32U64& value);
[[nodiscard]] U32U64 ShiftLeftLogical(const U32U64& base, const U32& shift);
[[nodiscard]] U32U64 ShiftRightLogical(const U32U64& base, const U32& shift);
[[nodiscard]] U32U64 ShiftRightArithmetic(const U32U64& base, const U32& shift);
[[nodiscard]] U32 BitwiseAnd(const U32& a, const U32& b);
[[nodiscard]] U32 BitwiseOr(const U32& a, const U32& b);
[[nodiscard]] U32 BitwiseXor(const U32& a, const U32& b);
[[nodiscard]] U32 BitFieldInsert(const U32& base, const U32& insert, const U32& offset,
const U32& count);
[[nodiscard]] U32 BitFieldExtract(const U32& base, const U32& offset, const U32& count,
bool is_signed = false);
[[nodiscard]] U32 BitReverse(const U32& value);
[[nodiscard]] U32 BitCount(const U32& value);
[[nodiscard]] U32 BitwiseNot(const U32& value);
[[nodiscard]] U32 FindSMsb(const U32& value);
[[nodiscard]] U32 FindUMsb(const U32& value);
[[nodiscard]] U32 SMin(const U32& a, const U32& b);
[[nodiscard]] U32 UMin(const U32& a, const U32& b);
[[nodiscard]] U32 IMin(const U32& a, const U32& b, bool is_signed);
[[nodiscard]] U32 SMax(const U32& a, const U32& b);
[[nodiscard]] U32 UMax(const U32& a, const U32& b);
[[nodiscard]] U32 IMax(const U32& a, const U32& b, bool is_signed);
[[nodiscard]] U32 SClamp(const U32& value, const U32& min, const U32& max);
[[nodiscard]] U32 UClamp(const U32& value, const U32& min, const U32& max);
[[nodiscard]] U1 ILessThan(const U32& lhs, const U32& rhs, bool is_signed);
[[nodiscard]] U1 IEqual(const U32U64& lhs, const U32U64& rhs);
[[nodiscard]] U1 ILessThanEqual(const U32& lhs, const U32& rhs, bool is_signed);
[[nodiscard]] U1 IGreaterThan(const U32& lhs, const U32& rhs, bool is_signed);
[[nodiscard]] U1 INotEqual(const U32& lhs, const U32& rhs);
[[nodiscard]] U1 IGreaterThanEqual(const U32& lhs, const U32& rhs, bool is_signed);
[[nodiscard]] U32 SharedAtomicIAdd(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicSMin(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicUMin(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicIMin(const U32& pointer_offset, const U32& value, bool is_signed);
[[nodiscard]] U32 SharedAtomicSMax(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicUMax(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicIMax(const U32& pointer_offset, const U32& value, bool is_signed);
[[nodiscard]] U32 SharedAtomicInc(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicDec(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicAnd(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicOr(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32 SharedAtomicXor(const U32& pointer_offset, const U32& value);
[[nodiscard]] U32U64 SharedAtomicExchange(const U32& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicIAdd(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicSMin(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicUMin(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicIMin(const U64& pointer_offset, const U32U64& value,
bool is_signed);
[[nodiscard]] U32U64 GlobalAtomicSMax(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicUMax(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicIMax(const U64& pointer_offset, const U32U64& value,
bool is_signed);
[[nodiscard]] U32 GlobalAtomicInc(const U64& pointer_offset, const U32& value);
[[nodiscard]] U32 GlobalAtomicDec(const U64& pointer_offset, const U32& value);
[[nodiscard]] U32U64 GlobalAtomicAnd(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicOr(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicXor(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] U32U64 GlobalAtomicExchange(const U64& pointer_offset, const U32U64& value);
[[nodiscard]] F32 GlobalAtomicF32Add(const U64& pointer_offset, const Value& value,
const FpControl control = {});
[[nodiscard]] Value GlobalAtomicF16x2Add(const U64& pointer_offset, const Value& value,
const FpControl control = {});
[[nodiscard]] Value GlobalAtomicF16x2Min(const U64& pointer_offset, const Value& value,
const FpControl control = {});
[[nodiscard]] Value GlobalAtomicF16x2Max(const U64& pointer_offset, const Value& value,
const FpControl control = {});
[[nodiscard]] U1 LogicalOr(const U1& a, const U1& b);
[[nodiscard]] U1 LogicalAnd(const U1& a, const U1& b);
[[nodiscard]] U1 LogicalXor(const U1& a, const U1& b);
[[nodiscard]] U1 LogicalNot(const U1& value);
[[nodiscard]] U32U64 ConvertFToS(size_t bitsize, const F16F32F64& value);
[[nodiscard]] U32U64 ConvertFToU(size_t bitsize, const F16F32F64& value);
[[nodiscard]] U32U64 ConvertFToI(size_t bitsize, bool is_signed, const F16F32F64& value);
[[nodiscard]] F16F32F64 ConvertSToF(size_t dest_bitsize, size_t src_bitsize, const Value& value,
FpControl control = {});
[[nodiscard]] F16F32F64 ConvertUToF(size_t dest_bitsize, size_t src_bitsize, const Value& value,
FpControl control = {});
[[nodiscard]] F16F32F64 ConvertIToF(size_t dest_bitsize, size_t src_bitsize, bool is_signed,
const Value& value, FpControl control = {});
[[nodiscard]] U8U16U32U64 UConvert(size_t result_bitsize, const U8U16U32U64& value);
[[nodiscard]] U8U16U32U64 SConvert(size_t result_bitsize, const U8U16U32U64& value);
[[nodiscard]] F16F32F64 FPConvert(size_t result_bitsize, const F16F32F64& value,
FpControl control = {});
[[nodiscard]] Value ImageSampleImplicitLod(const Value& handle, const Value& coords,
const F32& bias, const Value& offset,
const F32& lod_clamp, TextureInstInfo info);
[[nodiscard]] Value ImageSampleExplicitLod(const Value& handle, const Value& coords,
const F32& lod, const Value& offset,
TextureInstInfo info);
[[nodiscard]] F32 ImageSampleDrefImplicitLod(const Value& handle, const Value& coords,
const F32& dref, const F32& bias,
const Value& offset, const F32& lod_clamp,
TextureInstInfo info);
[[nodiscard]] F32 ImageSampleDrefExplicitLod(const Value& handle, const Value& coords,
const F32& dref, const F32& lod,
const Value& offset, TextureInstInfo info);
[[nodiscard]] Value ImageQueryDimension(const Value& handle, const IR::U32& lod,
const IR::U1& skip_mips);
[[nodiscard]] Value ImageQueryDimension(const Value& handle, const IR::U32& lod,
const IR::U1& skip_mips, TextureInstInfo info);
[[nodiscard]] Value ImageQueryLod(const Value& handle, const Value& coords,
TextureInstInfo info);
[[nodiscard]] Value ImageGather(const Value& handle, const Value& coords, const Value& offset,
const Value& offset2, TextureInstInfo info);
[[nodiscard]] Value ImageGatherDref(const Value& handle, const Value& coords,
const Value& offset, const Value& offset2, const F32& dref,
TextureInstInfo info);
[[nodiscard]] Value ImageFetch(const Value& handle, const Value& coords, const Value& offset,
const U32& lod, const U32& multisampling, TextureInstInfo info);
[[nodiscard]] Value ImageGradient(const Value& handle, const Value& coords,
const Value& derivatives, const Value& offset,
const F32& lod_clamp, TextureInstInfo info);
[[nodiscard]] Value ImageRead(const Value& handle, const Value& coords, TextureInstInfo info);
void ImageWrite(const Value& handle, const Value& coords, const Value& color,
TextureInstInfo info);
[[nodiscard]] Value ImageAtomicIAdd(const Value& handle, const Value& coords,
const Value& value, TextureInstInfo info);
[[nodiscard]] Value ImageAtomicSMin(const Value& handle, const Value& coords,
const Value& value, TextureInstInfo info);
[[nodiscard]] Value ImageAtomicUMin(const Value& handle, const Value& coords,
const Value& value, TextureInstInfo info);
[[nodiscard]] Value ImageAtomicIMin(const Value& handle, const Value& coords,
const Value& value, bool is_signed, TextureInstInfo info);
[[nodiscard]] Value ImageAtomicSMax(const Value& handle, const Value& coords,
const Value& value, TextureInstInfo info);
[[nodiscard]] Value ImageAtomicUMax(const Value& handle, const Value& coords,
const Value& value, TextureInstInfo info);
[[nodiscard]] Value ImageAtomicIMax(const Value& handle, const Value& coords,
const Value& value, bool is_signed, TextureInstInfo info);
[[nodiscard]] Value ImageAtomicInc(const Value& handle, const Value& coords, const Value& value,
TextureInstInfo info);
[[nodiscard]] Value ImageAtomicDec(const Value& handle, const Value& coords, const Value& value,
TextureInstInfo info);
[[nodiscard]] Value ImageAtomicAnd(const Value& handle, const Value& coords, const Value& value,
TextureInstInfo info);
[[nodiscard]] Value ImageAtomicOr(const Value& handle, const Value& coords, const Value& value,
TextureInstInfo info);
[[nodiscard]] Value ImageAtomicXor(const Value& handle, const Value& coords, const Value& value,
TextureInstInfo info);
[[nodiscard]] Value ImageAtomicExchange(const Value& handle, const Value& coords,
const Value& value, TextureInstInfo info);
[[nodiscard]] U1 IsTextureScaled(const U32& index);
[[nodiscard]] U1 IsImageScaled(const U32& index);
[[nodiscard]] U1 VoteAll(const U1& value);
[[nodiscard]] U1 VoteAny(const U1& value);
[[nodiscard]] U1 VoteEqual(const U1& value);
[[nodiscard]] U32 SubgroupBallot(const U1& value);
[[nodiscard]] U32 SubgroupEqMask();
[[nodiscard]] U32 SubgroupLtMask();
[[nodiscard]] U32 SubgroupLeMask();
[[nodiscard]] U32 SubgroupGtMask();
[[nodiscard]] U32 SubgroupGeMask();
[[nodiscard]] U32 ShuffleIndex(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
const IR::U32& seg_mask);
[[nodiscard]] U32 ShuffleUp(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
const IR::U32& seg_mask);
[[nodiscard]] U32 ShuffleDown(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
const IR::U32& seg_mask);
[[nodiscard]] U32 ShuffleButterfly(const IR::U32& value, const IR::U32& index,
const IR::U32& clamp, const IR::U32& seg_mask);
[[nodiscard]] F32 FSwizzleAdd(const F32& a, const F32& b, const U32& swizzle,
FpControl control = {});
[[nodiscard]] F32 DPdxFine(const F32& a);
[[nodiscard]] F32 DPdyFine(const F32& a);
[[nodiscard]] F32 DPdxCoarse(const F32& a);
[[nodiscard]] F32 DPdyCoarse(const F32& a);
private:
IR::Block::iterator insertion_point;
template <typename T = Value, typename... Args>
T Inst(Opcode op, Args... args) {
auto it{block->PrependNewInst(insertion_point, op, {Value{args}...})};
return T{Value{&*it}};
}
template <typename T>
requires(sizeof(T) <= sizeof(u32) && std::is_trivially_copyable_v<T>)
struct Flags {
Flags() = default;
Flags(T proxy_) : proxy{proxy_} {}
T proxy;
};
template <typename T = Value, typename FlagType, typename... Args>
T Inst(Opcode op, Flags<FlagType> flags, Args... args) {
u32 raw_flags{};
std::memcpy(&raw_flags, &flags.proxy, sizeof(flags.proxy));
auto it{block->PrependNewInst(insertion_point, op, {Value{args}...}, raw_flags)};
return T{Value{&*it}};
}
};
} // namespace Shader::IR

View File

@@ -0,0 +1,455 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <memory>
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/type.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
namespace {
void CheckPseudoInstruction(IR::Inst* inst, IR::Opcode opcode) {
if (inst && inst->GetOpcode() != opcode) {
throw LogicError("Invalid pseudo-instruction");
}
}
void SetPseudoInstruction(IR::Inst*& dest_inst, IR::Inst* pseudo_inst) {
if (dest_inst) {
throw LogicError("Only one of each type of pseudo-op allowed");
}
dest_inst = pseudo_inst;
}
void RemovePseudoInstruction(IR::Inst*& inst, IR::Opcode expected_opcode) {
if (inst->GetOpcode() != expected_opcode) {
throw LogicError("Undoing use of invalid pseudo-op");
}
inst = nullptr;
}
void AllocAssociatedInsts(std::unique_ptr<AssociatedInsts>& associated_insts) {
if (!associated_insts) {
associated_insts = std::make_unique<AssociatedInsts>();
}
}
} // Anonymous namespace
Inst::Inst(IR::Opcode op_, u32 flags_) noexcept : op{op_}, flags{flags_} {
if (op == Opcode::Phi) {
std::construct_at(&phi_args);
} else {
std::construct_at(&args);
}
}
Inst::Inst(const Inst& base) : op{base.op}, flags{base.flags} {
if (base.op == Opcode::Phi) {
throw NotImplementedException("Copying phi node");
}
std::construct_at(&args);
const size_t num_args{base.NumArgs()};
for (size_t index = 0; index < num_args; ++index) {
SetArg(index, base.Arg(index));
}
}
Inst::~Inst() {
if (op == Opcode::Phi) {
std::destroy_at(&phi_args);
} else {
std::destroy_at(&args);
}
}
bool Inst::MayHaveSideEffects() const noexcept {
switch (op) {
case Opcode::ConditionRef:
case Opcode::Reference:
case Opcode::PhiMove:
case Opcode::Prologue:
case Opcode::Epilogue:
case Opcode::Join:
case Opcode::DemoteToHelperInvocation:
case Opcode::Barrier:
case Opcode::WorkgroupMemoryBarrier:
case Opcode::DeviceMemoryBarrier:
case Opcode::EmitVertex:
case Opcode::EndPrimitive:
case Opcode::SetAttribute:
case Opcode::SetAttributeIndexed:
case Opcode::SetPatch:
case Opcode::SetFragColor:
case Opcode::SetSampleMask:
case Opcode::SetFragDepth:
case Opcode::WriteGlobalU8:
case Opcode::WriteGlobalS8:
case Opcode::WriteGlobalU16:
case Opcode::WriteGlobalS16:
case Opcode::WriteGlobal32:
case Opcode::WriteGlobal64:
case Opcode::WriteGlobal128:
case Opcode::WriteStorageU8:
case Opcode::WriteStorageS8:
case Opcode::WriteStorageU16:
case Opcode::WriteStorageS16:
case Opcode::WriteStorage32:
case Opcode::WriteStorage64:
case Opcode::WriteStorage128:
case Opcode::WriteLocal:
case Opcode::WriteSharedU8:
case Opcode::WriteSharedU16:
case Opcode::WriteSharedU32:
case Opcode::WriteSharedU64:
case Opcode::WriteSharedU128:
case Opcode::SharedAtomicIAdd32:
case Opcode::SharedAtomicSMin32:
case Opcode::SharedAtomicUMin32:
case Opcode::SharedAtomicSMax32:
case Opcode::SharedAtomicUMax32:
case Opcode::SharedAtomicInc32:
case Opcode::SharedAtomicDec32:
case Opcode::SharedAtomicAnd32:
case Opcode::SharedAtomicOr32:
case Opcode::SharedAtomicXor32:
case Opcode::SharedAtomicExchange32:
case Opcode::SharedAtomicExchange64:
case Opcode::SharedAtomicExchange32x2:
case Opcode::GlobalAtomicIAdd32:
case Opcode::GlobalAtomicSMin32:
case Opcode::GlobalAtomicUMin32:
case Opcode::GlobalAtomicSMax32:
case Opcode::GlobalAtomicUMax32:
case Opcode::GlobalAtomicInc32:
case Opcode::GlobalAtomicDec32:
case Opcode::GlobalAtomicAnd32:
case Opcode::GlobalAtomicOr32:
case Opcode::GlobalAtomicXor32:
case Opcode::GlobalAtomicExchange32:
case Opcode::GlobalAtomicIAdd64:
case Opcode::GlobalAtomicSMin64:
case Opcode::GlobalAtomicUMin64:
case Opcode::GlobalAtomicSMax64:
case Opcode::GlobalAtomicUMax64:
case Opcode::GlobalAtomicAnd64:
case Opcode::GlobalAtomicOr64:
case Opcode::GlobalAtomicXor64:
case Opcode::GlobalAtomicExchange64:
case Opcode::GlobalAtomicIAdd32x2:
case Opcode::GlobalAtomicSMin32x2:
case Opcode::GlobalAtomicUMin32x2:
case Opcode::GlobalAtomicSMax32x2:
case Opcode::GlobalAtomicUMax32x2:
case Opcode::GlobalAtomicAnd32x2:
case Opcode::GlobalAtomicOr32x2:
case Opcode::GlobalAtomicXor32x2:
case Opcode::GlobalAtomicExchange32x2:
case Opcode::GlobalAtomicAddF32:
case Opcode::GlobalAtomicAddF16x2:
case Opcode::GlobalAtomicAddF32x2:
case Opcode::GlobalAtomicMinF16x2:
case Opcode::GlobalAtomicMinF32x2:
case Opcode::GlobalAtomicMaxF16x2:
case Opcode::GlobalAtomicMaxF32x2:
case Opcode::StorageAtomicIAdd32:
case Opcode::StorageAtomicSMin32:
case Opcode::StorageAtomicUMin32:
case Opcode::StorageAtomicSMax32:
case Opcode::StorageAtomicUMax32:
case Opcode::StorageAtomicInc32:
case Opcode::StorageAtomicDec32:
case Opcode::StorageAtomicAnd32:
case Opcode::StorageAtomicOr32:
case Opcode::StorageAtomicXor32:
case Opcode::StorageAtomicExchange32:
case Opcode::StorageAtomicIAdd64:
case Opcode::StorageAtomicSMin64:
case Opcode::StorageAtomicUMin64:
case Opcode::StorageAtomicSMax64:
case Opcode::StorageAtomicUMax64:
case Opcode::StorageAtomicAnd64:
case Opcode::StorageAtomicOr64:
case Opcode::StorageAtomicXor64:
case Opcode::StorageAtomicExchange64:
case Opcode::StorageAtomicIAdd32x2:
case Opcode::StorageAtomicSMin32x2:
case Opcode::StorageAtomicUMin32x2:
case Opcode::StorageAtomicSMax32x2:
case Opcode::StorageAtomicUMax32x2:
case Opcode::StorageAtomicAnd32x2:
case Opcode::StorageAtomicOr32x2:
case Opcode::StorageAtomicXor32x2:
case Opcode::StorageAtomicExchange32x2:
case Opcode::StorageAtomicAddF32:
case Opcode::StorageAtomicAddF16x2:
case Opcode::StorageAtomicAddF32x2:
case Opcode::StorageAtomicMinF16x2:
case Opcode::StorageAtomicMinF32x2:
case Opcode::StorageAtomicMaxF16x2:
case Opcode::StorageAtomicMaxF32x2:
case Opcode::BindlessImageWrite:
case Opcode::BoundImageWrite:
case Opcode::ImageWrite:
case IR::Opcode::BindlessImageAtomicIAdd32:
case IR::Opcode::BindlessImageAtomicSMin32:
case IR::Opcode::BindlessImageAtomicUMin32:
case IR::Opcode::BindlessImageAtomicSMax32:
case IR::Opcode::BindlessImageAtomicUMax32:
case IR::Opcode::BindlessImageAtomicInc32:
case IR::Opcode::BindlessImageAtomicDec32:
case IR::Opcode::BindlessImageAtomicAnd32:
case IR::Opcode::BindlessImageAtomicOr32:
case IR::Opcode::BindlessImageAtomicXor32:
case IR::Opcode::BindlessImageAtomicExchange32:
case IR::Opcode::BoundImageAtomicIAdd32:
case IR::Opcode::BoundImageAtomicSMin32:
case IR::Opcode::BoundImageAtomicUMin32:
case IR::Opcode::BoundImageAtomicSMax32:
case IR::Opcode::BoundImageAtomicUMax32:
case IR::Opcode::BoundImageAtomicInc32:
case IR::Opcode::BoundImageAtomicDec32:
case IR::Opcode::BoundImageAtomicAnd32:
case IR::Opcode::BoundImageAtomicOr32:
case IR::Opcode::BoundImageAtomicXor32:
case IR::Opcode::BoundImageAtomicExchange32:
case IR::Opcode::ImageAtomicIAdd32:
case IR::Opcode::ImageAtomicSMin32:
case IR::Opcode::ImageAtomicUMin32:
case IR::Opcode::ImageAtomicSMax32:
case IR::Opcode::ImageAtomicUMax32:
case IR::Opcode::ImageAtomicInc32:
case IR::Opcode::ImageAtomicDec32:
case IR::Opcode::ImageAtomicAnd32:
case IR::Opcode::ImageAtomicOr32:
case IR::Opcode::ImageAtomicXor32:
case IR::Opcode::ImageAtomicExchange32:
return true;
default:
return false;
}
}
bool Inst::IsPseudoInstruction() const noexcept {
switch (op) {
case Opcode::GetZeroFromOp:
case Opcode::GetSignFromOp:
case Opcode::GetCarryFromOp:
case Opcode::GetOverflowFromOp:
case Opcode::GetSparseFromOp:
case Opcode::GetInBoundsFromOp:
return true;
default:
return false;
}
}
bool Inst::AreAllArgsImmediates() const {
if (op == Opcode::Phi) {
throw LogicError("Testing for all arguments are immediates on phi instruction");
}
return std::all_of(args.begin(), args.begin() + NumArgs(),
[](const IR::Value& value) { return value.IsImmediate(); });
}
Inst* Inst::GetAssociatedPseudoOperation(IR::Opcode opcode) {
if (!associated_insts) {
return nullptr;
}
switch (opcode) {
case Opcode::GetZeroFromOp:
CheckPseudoInstruction(associated_insts->zero_inst, Opcode::GetZeroFromOp);
return associated_insts->zero_inst;
case Opcode::GetSignFromOp:
CheckPseudoInstruction(associated_insts->sign_inst, Opcode::GetSignFromOp);
return associated_insts->sign_inst;
case Opcode::GetCarryFromOp:
CheckPseudoInstruction(associated_insts->carry_inst, Opcode::GetCarryFromOp);
return associated_insts->carry_inst;
case Opcode::GetOverflowFromOp:
CheckPseudoInstruction(associated_insts->overflow_inst, Opcode::GetOverflowFromOp);
return associated_insts->overflow_inst;
case Opcode::GetSparseFromOp:
CheckPseudoInstruction(associated_insts->sparse_inst, Opcode::GetSparseFromOp);
return associated_insts->sparse_inst;
case Opcode::GetInBoundsFromOp:
CheckPseudoInstruction(associated_insts->in_bounds_inst, Opcode::GetInBoundsFromOp);
return associated_insts->in_bounds_inst;
default:
throw InvalidArgument("{} is not a pseudo-instruction", opcode);
}
}
IR::Type Inst::Type() const {
if (op == IR::Opcode::Phi) {
// The type of a phi node is stored in its flags
return Flags<IR::Type>();
}
return TypeOf(op);
}
void Inst::SetArg(size_t index, Value value) {
if (index >= NumArgs()) {
throw InvalidArgument("Out of bounds argument index {} in opcode {}", index, op);
}
const IR::Value arg{Arg(index)};
if (!arg.IsImmediate()) {
UndoUse(arg);
}
if (!value.IsImmediate()) {
Use(value);
}
if (op == Opcode::Phi) {
phi_args[index].second = value;
} else {
args[index] = value;
}
}
Block* Inst::PhiBlock(size_t index) const {
if (op != Opcode::Phi) {
throw LogicError("{} is not a Phi instruction", op);
}
if (index >= phi_args.size()) {
throw InvalidArgument("Out of bounds argument index {} in phi instruction");
}
return phi_args[index].first;
}
void Inst::AddPhiOperand(Block* predecessor, const Value& value) {
if (!value.IsImmediate()) {
Use(value);
}
phi_args.emplace_back(predecessor, value);
}
void Inst::OrderPhiArgs() {
if (op != Opcode::Phi) {
throw LogicError("{} is not a Phi instruction", op);
}
std::sort(phi_args.begin(), phi_args.end(),
[](const std::pair<Block*, Value>& a, const std::pair<Block*, Value>& b) {
return a.first->GetOrder() < b.first->GetOrder();
});
}
void Inst::Invalidate() {
ClearArgs();
ReplaceOpcode(Opcode::Void);
}
void Inst::ClearArgs() {
if (op == Opcode::Phi) {
for (auto& pair : phi_args) {
IR::Value& value{pair.second};
if (!value.IsImmediate()) {
UndoUse(value);
}
}
phi_args.clear();
} else {
for (auto& value : args) {
if (!value.IsImmediate()) {
UndoUse(value);
}
}
// Reset arguments to null
// std::memset was measured to be faster on MSVC than std::ranges:fill
std::memset(reinterpret_cast<char*>(&args), 0, sizeof(args));
}
}
void Inst::ReplaceUsesWith(Value replacement) {
Invalidate();
ReplaceOpcode(Opcode::Identity);
if (!replacement.IsImmediate()) {
Use(replacement);
}
args[0] = replacement;
}
void Inst::ReplaceOpcode(IR::Opcode opcode) {
if (opcode == IR::Opcode::Phi) {
throw LogicError("Cannot transition into Phi");
}
if (op == Opcode::Phi) {
// Transition out of phi arguments into non-phi
std::destroy_at(&phi_args);
std::construct_at(&args);
}
op = opcode;
}
void Inst::Use(const Value& value) {
Inst* const inst{value.Inst()};
++inst->use_count;
std::unique_ptr<AssociatedInsts>& assoc_inst{inst->associated_insts};
switch (op) {
case Opcode::GetZeroFromOp:
AllocAssociatedInsts(assoc_inst);
SetPseudoInstruction(assoc_inst->zero_inst, this);
break;
case Opcode::GetSignFromOp:
AllocAssociatedInsts(assoc_inst);
SetPseudoInstruction(assoc_inst->sign_inst, this);
break;
case Opcode::GetCarryFromOp:
AllocAssociatedInsts(assoc_inst);
SetPseudoInstruction(assoc_inst->carry_inst, this);
break;
case Opcode::GetOverflowFromOp:
AllocAssociatedInsts(assoc_inst);
SetPseudoInstruction(assoc_inst->overflow_inst, this);
break;
case Opcode::GetSparseFromOp:
AllocAssociatedInsts(assoc_inst);
SetPseudoInstruction(assoc_inst->sparse_inst, this);
break;
case Opcode::GetInBoundsFromOp:
AllocAssociatedInsts(assoc_inst);
SetPseudoInstruction(assoc_inst->in_bounds_inst, this);
break;
default:
break;
}
}
void Inst::UndoUse(const Value& value) {
Inst* const inst{value.Inst()};
--inst->use_count;
std::unique_ptr<AssociatedInsts>& assoc_inst{inst->associated_insts};
switch (op) {
case Opcode::GetZeroFromOp:
AllocAssociatedInsts(assoc_inst);
RemovePseudoInstruction(assoc_inst->zero_inst, Opcode::GetZeroFromOp);
break;
case Opcode::GetSignFromOp:
AllocAssociatedInsts(assoc_inst);
RemovePseudoInstruction(assoc_inst->sign_inst, Opcode::GetSignFromOp);
break;
case Opcode::GetCarryFromOp:
AllocAssociatedInsts(assoc_inst);
RemovePseudoInstruction(assoc_inst->carry_inst, Opcode::GetCarryFromOp);
break;
case Opcode::GetOverflowFromOp:
AllocAssociatedInsts(assoc_inst);
RemovePseudoInstruction(assoc_inst->overflow_inst, Opcode::GetOverflowFromOp);
break;
case Opcode::GetSparseFromOp:
AllocAssociatedInsts(assoc_inst);
RemovePseudoInstruction(assoc_inst->sparse_inst, Opcode::GetSparseFromOp);
break;
case Opcode::GetInBoundsFromOp:
AllocAssociatedInsts(assoc_inst);
RemovePseudoInstruction(assoc_inst->in_bounds_inst, Opcode::GetInBoundsFromOp);
break;
default:
break;
}
}
} // namespace Shader::IR

View File

@@ -0,0 +1,56 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "common/bit_field.h"
#include "common/common_types.h"
#include "shader_recompiler/shader_info.h"
namespace Shader::IR {
enum class FmzMode : u8 {
DontCare, // Not specified for this instruction
FTZ, // Flush denorms to zero, NAN is propagated (D3D11, NVN, GL, VK)
FMZ, // Flush denorms to zero, x * 0 == 0 (D3D9)
None, // Denorms are not flushed, NAN is propagated (nouveau)
};
enum class FpRounding : u8 {
DontCare, // Not specified for this instruction
RN, // Round to nearest even,
RM, // Round towards negative infinity
RP, // Round towards positive infinity
RZ, // Round towards zero
};
struct FpControl {
bool no_contraction{false};
FpRounding rounding{FpRounding::DontCare};
FmzMode fmz_mode{FmzMode::DontCare};
};
static_assert(sizeof(FpControl) <= sizeof(u32));
union TextureInstInfo {
u32 raw;
BitField<0, 16, u32> descriptor_index;
BitField<16, 3, TextureType> type;
BitField<19, 1, u32> is_depth;
BitField<20, 1, u32> has_bias;
BitField<21, 1, u32> has_lod_clamp;
BitField<22, 1, u32> relaxed_precision;
BitField<23, 2, u32> gather_component;
BitField<25, 2, u32> num_derivatives;
BitField<27, 3, ImageFormat> image_format;
BitField<30, 1, u32> ndv_is_active;
/// Only meaningful for the sampled-texture family (ImageSampleImplicitLod, ImageFetch,
/// ImageGather, etc.); unused/zero for storage-image opcodes, which carry their own
/// is_integer via ImageDescriptor/ImageBufferDescriptor instead.
BitField<31, 1, u32> is_integer;
};
static_assert(sizeof(TextureInstInfo) <= sizeof(u32));
} // namespace Shader::IR

View File

@@ -0,0 +1,36 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/frontend/ir/opcodes.h"
namespace Shader::IR {
namespace Detail {
OpcodeMeta META_TABLE[532] = {
#define OPCODE(name_token, type_token, ...) \
{ \
.name{#name_token}, \
.type = type_token, \
.arg_types{__VA_ARGS__}, \
},
#include "opcodes.inc"
#undef OPCODE
};
u8 NUM_ARGS[532] = {
#define OPCODE(name_token, type_token, ...) u8(CalculateNumArgsOf(Opcode::name_token)),
#include "opcodes.inc"
#undef OPCODE
};
}
std::string_view NameOf(Opcode op) {
return Detail::META_TABLE[static_cast<size_t>(op)].name;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,97 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <algorithm>
#include <array>
#include <fmt/ranges.h>
#include <ranges>
#include "shader_recompiler/frontend/ir/type.h"
namespace Shader::IR {
enum class Opcode {
#define OPCODE(name, ...) name,
#include "opcodes.inc"
#undef OPCODE
};
namespace Detail {
struct OpcodeMeta {
std::string_view name;
Type type;
std::array<Type, 5> arg_types;
};
// using enum Type;
static constexpr Type Void{Type::Void};
static constexpr Type Opaque{Type::Opaque};
static constexpr Type Reg{Type::Reg};
static constexpr Type Pred{Type::Pred};
static constexpr Type Attribute{Type::Attribute};
static constexpr Type Patch{Type::Patch};
static constexpr Type U1{Type::U1};
static constexpr Type U8{Type::U8};
static constexpr Type U16{Type::U16};
static constexpr Type U32{Type::U32};
static constexpr Type U64{Type::U64};
static constexpr Type F16{Type::F16};
static constexpr Type F32{Type::F32};
static constexpr Type F64{Type::F64};
static constexpr Type U32x2{Type::U32x2};
static constexpr Type U32x3{Type::U32x3};
static constexpr Type U32x4{Type::U32x4};
static constexpr Type F16x2{Type::F16x2};
static constexpr Type F16x3{Type::F16x3};
static constexpr Type F16x4{Type::F16x4};
static constexpr Type F32x2{Type::F32x2};
static constexpr Type F32x3{Type::F32x3};
static constexpr Type F32x4{Type::F32x4};
static constexpr Type F64x2{Type::F64x2};
static constexpr Type F64x3{Type::F64x3};
static constexpr Type F64x4{Type::F64x4};
extern OpcodeMeta META_TABLE[532];
constexpr size_t CalculateNumArgsOf(Opcode op) noexcept {
const auto& arg_types = META_TABLE[size_t(op)].arg_types;
return size_t(std::distance(arg_types.begin(), std::ranges::find(arg_types, Type::Void)));
}
extern u8 NUM_ARGS[532];
} // namespace Detail
/// Get return type of an opcode
[[nodiscard]] inline Type TypeOf(Opcode op) noexcept {
return Detail::META_TABLE[size_t(op)].type;
}
/// Get the number of arguments an opcode accepts
[[nodiscard]] inline size_t NumArgsOf(Opcode op) noexcept {
return size_t(Detail::NUM_ARGS[size_t(op)]);
}
/// Get the required type of an argument of an opcode
[[nodiscard]] inline Type ArgTypeOf(Opcode op, size_t arg_index) noexcept {
return Detail::META_TABLE[size_t(op)].arg_types[arg_index];
}
/// Get the name of an opcode
[[nodiscard]] std::string_view NameOf(Opcode op);
} // namespace Shader::IR
template <>
struct fmt::formatter<Shader::IR::Opcode> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::IR::Opcode& op, FormatContext& ctx) const {
return fmt::format_to(ctx.out(), "{}", Shader::IR::NameOf(op));
}
};

View File

@@ -0,0 +1,586 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
// opcode name, return type, arg1 type, arg2 type, arg3 type, arg4 type, arg4 type, ...
OPCODE(Phi, Opaque, )
OPCODE(Identity, Opaque, Opaque, )
OPCODE(Void, Void, )
OPCODE(ConditionRef, U1, U1, )
OPCODE(Reference, Void, Opaque, )
OPCODE(PhiMove, Void, Opaque, Opaque, )
// Special operations
OPCODE(Prologue, Void, )
OPCODE(Epilogue, Void, )
OPCODE(Join, Void, )
OPCODE(DemoteToHelperInvocation, Void, )
OPCODE(EmitVertex, Void, U32, )
OPCODE(EndPrimitive, Void, U32, )
// Barriers
OPCODE(Barrier, Void, )
OPCODE(WorkgroupMemoryBarrier, Void, )
OPCODE(DeviceMemoryBarrier, Void, )
// Context getters/setters
OPCODE(GetRegister, U32, Reg, )
OPCODE(SetRegister, Void, Reg, U32, )
OPCODE(GetPred, U1, Pred, )
OPCODE(SetPred, Void, Pred, U1, )
OPCODE(GetGotoVariable, U1, U32, )
OPCODE(SetGotoVariable, Void, U32, U1, )
OPCODE(GetIndirectBranchVariable, U32, )
OPCODE(SetIndirectBranchVariable, Void, U32, )
OPCODE(GetCbufU8, U32, U32, U32, )
OPCODE(GetCbufS8, U32, U32, U32, )
OPCODE(GetCbufU16, U32, U32, U32, )
OPCODE(GetCbufS16, U32, U32, U32, )
OPCODE(GetCbufU32, U32, U32, U32, )
OPCODE(GetCbufF32, F32, U32, U32, )
OPCODE(GetCbufU32x2, U32x2, U32, U32, )
OPCODE(GetAttribute, F32, Attribute, U32, )
OPCODE(GetAttributeU32, U32, Attribute, U32, )
OPCODE(SetAttribute, Void, Attribute, F32, U32, )
OPCODE(GetAttributeIndexed, F32, U32, U32, )
OPCODE(SetAttributeIndexed, Void, U32, F32, U32, )
OPCODE(GetPatch, F32, Patch, )
OPCODE(SetPatch, Void, Patch, F32, )
OPCODE(SetFragColor, Void, U32, U32, F32, )
OPCODE(SetSampleMask, Void, U32, )
OPCODE(SetFragDepth, Void, F32, )
OPCODE(GetZFlag, U1, Void, )
OPCODE(GetSFlag, U1, Void, )
OPCODE(GetCFlag, U1, Void, )
OPCODE(GetOFlag, U1, Void, )
OPCODE(SetZFlag, Void, U1, )
OPCODE(SetSFlag, Void, U1, )
OPCODE(SetCFlag, Void, U1, )
OPCODE(SetOFlag, Void, U1, )
OPCODE(WorkgroupId, U32x3, )
OPCODE(LocalInvocationId, U32x3, )
OPCODE(InvocationId, U32, )
OPCODE(InvocationInfo, U32, )
OPCODE(SampleId, U32, )
OPCODE(IsHelperInvocation, U1, )
OPCODE(SR_WScaleFactorXY, U32, )
OPCODE(SR_WScaleFactorZ, U32, )
OPCODE(YDirection, F32, )
OPCODE(ResolutionDownFactor, F32, )
OPCODE(RenderArea, F32x4, )
// Undefined
OPCODE(UndefU1, U1, )
OPCODE(UndefU8, U8, )
OPCODE(UndefU16, U16, )
OPCODE(UndefU32, U32, )
OPCODE(UndefU64, U64, )
// Memory operations
OPCODE(LoadGlobalU8, U32, Opaque, )
OPCODE(LoadGlobalS8, U32, Opaque, )
OPCODE(LoadGlobalU16, U32, Opaque, )
OPCODE(LoadGlobalS16, U32, Opaque, )
OPCODE(LoadGlobal32, U32, Opaque, )
OPCODE(LoadGlobal64, U32x2, Opaque, )
OPCODE(LoadGlobal128, U32x4, Opaque, )
OPCODE(WriteGlobalU8, Void, Opaque, U32, )
OPCODE(WriteGlobalS8, Void, Opaque, U32, )
OPCODE(WriteGlobalU16, Void, Opaque, U32, )
OPCODE(WriteGlobalS16, Void, Opaque, U32, )
OPCODE(WriteGlobal32, Void, Opaque, U32, )
OPCODE(WriteGlobal64, Void, Opaque, U32x2, )
OPCODE(WriteGlobal128, Void, Opaque, U32x4, )
// Storage buffer operations
OPCODE(LoadStorageU8, U32, U32, U32, )
OPCODE(LoadStorageS8, U32, U32, U32, )
OPCODE(LoadStorageU16, U32, U32, U32, )
OPCODE(LoadStorageS16, U32, U32, U32, )
OPCODE(LoadStorage32, U32, U32, U32, )
OPCODE(LoadStorage64, U32x2, U32, U32, )
OPCODE(LoadStorage128, U32x4, U32, U32, )
OPCODE(WriteStorageU8, Void, U32, U32, U32, )
OPCODE(WriteStorageS8, Void, U32, U32, U32, )
OPCODE(WriteStorageU16, Void, U32, U32, U32, )
OPCODE(WriteStorageS16, Void, U32, U32, U32, )
OPCODE(WriteStorage32, Void, U32, U32, U32, )
OPCODE(WriteStorage64, Void, U32, U32, U32x2, )
OPCODE(WriteStorage128, Void, U32, U32, U32x4, )
// Local memory operations
OPCODE(LoadLocal, U32, U32, )
OPCODE(WriteLocal, Void, U32, U32, )
// Shared memory operations
OPCODE(LoadSharedU8, U32, U32, )
OPCODE(LoadSharedS8, U32, U32, )
OPCODE(LoadSharedU16, U32, U32, )
OPCODE(LoadSharedS16, U32, U32, )
OPCODE(LoadSharedU32, U32, U32, )
OPCODE(LoadSharedU64, U32x2, U32, )
OPCODE(LoadSharedU128, U32x4, U32, )
OPCODE(WriteSharedU8, Void, U32, U32, )
OPCODE(WriteSharedU16, Void, U32, U32, )
OPCODE(WriteSharedU32, Void, U32, U32, )
OPCODE(WriteSharedU64, Void, U32, U32x2, )
OPCODE(WriteSharedU128, Void, U32, U32x4, )
// Vector utility
OPCODE(CompositeConstructU32x2, U32x2, U32, U32, )
OPCODE(CompositeConstructU32x3, U32x3, U32, U32, U32, )
OPCODE(CompositeConstructU32x4, U32x4, U32, U32, U32, U32, )
OPCODE(CompositeExtractU32x2, U32, U32x2, U32, )
OPCODE(CompositeExtractU32x3, U32, U32x3, U32, )
OPCODE(CompositeExtractU32x4, U32, U32x4, U32, )
OPCODE(CompositeInsertU32x2, U32x2, U32x2, U32, U32, )
OPCODE(CompositeInsertU32x3, U32x3, U32x3, U32, U32, )
OPCODE(CompositeInsertU32x4, U32x4, U32x4, U32, U32, )
OPCODE(CompositeConstructF16x2, F16x2, F16, F16, )
OPCODE(CompositeConstructF16x3, F16x3, F16, F16, F16, )
OPCODE(CompositeConstructF16x4, F16x4, F16, F16, F16, F16, )
OPCODE(CompositeExtractF16x2, F16, F16x2, U32, )
OPCODE(CompositeExtractF16x3, F16, F16x3, U32, )
OPCODE(CompositeExtractF16x4, F16, F16x4, U32, )
OPCODE(CompositeInsertF16x2, F16x2, F16x2, F16, U32, )
OPCODE(CompositeInsertF16x3, F16x3, F16x3, F16, U32, )
OPCODE(CompositeInsertF16x4, F16x4, F16x4, F16, U32, )
OPCODE(CompositeConstructF32x2, F32x2, F32, F32, )
OPCODE(CompositeConstructF32x3, F32x3, F32, F32, F32, )
OPCODE(CompositeConstructF32x4, F32x4, F32, F32, F32, F32, )
OPCODE(CompositeExtractF32x2, F32, F32x2, U32, )
OPCODE(CompositeExtractF32x3, F32, F32x3, U32, )
OPCODE(CompositeExtractF32x4, F32, F32x4, U32, )
OPCODE(CompositeInsertF32x2, F32x2, F32x2, F32, U32, )
OPCODE(CompositeInsertF32x3, F32x3, F32x3, F32, U32, )
OPCODE(CompositeInsertF32x4, F32x4, F32x4, F32, U32, )
OPCODE(CompositeConstructF64x2, F64x2, F64, F64, )
OPCODE(CompositeConstructF64x3, F64x3, F64, F64, F64, )
OPCODE(CompositeConstructF64x4, F64x4, F64, F64, F64, F64, )
OPCODE(CompositeExtractF64x2, F64, F64x2, U32, )
OPCODE(CompositeExtractF64x3, F64, F64x3, U32, )
OPCODE(CompositeExtractF64x4, F64, F64x4, U32, )
OPCODE(CompositeInsertF64x2, F64x2, F64x2, F64, U32, )
OPCODE(CompositeInsertF64x3, F64x3, F64x3, F64, U32, )
OPCODE(CompositeInsertF64x4, F64x4, F64x4, F64, U32, )
// Select operations
OPCODE(SelectU1, U1, U1, U1, U1, )
OPCODE(SelectU8, U8, U1, U8, U8, )
OPCODE(SelectU16, U16, U1, U16, U16, )
OPCODE(SelectU32, U32, U1, U32, U32, )
OPCODE(SelectU64, U64, U1, U64, U64, )
OPCODE(SelectF16, F16, U1, F16, F16, )
OPCODE(SelectF32, F32, U1, F32, F32, )
OPCODE(SelectF64, F64, U1, F64, F64, )
// Bitwise conversions
OPCODE(BitCastU16F16, U16, F16, )
OPCODE(BitCastU32F32, U32, F32, )
OPCODE(BitCastU64F64, U64, F64, )
OPCODE(BitCastF16U16, F16, U16, )
OPCODE(BitCastF32U32, F32, U32, )
OPCODE(BitCastF64U64, F64, U64, )
OPCODE(PackUint2x32, U64, U32x2, )
OPCODE(UnpackUint2x32, U32x2, U64, )
OPCODE(PackFloat2x16, U32, F16x2, )
OPCODE(UnpackFloat2x16, F16x2, U32, )
OPCODE(PackHalf2x16, U32, F32x2, )
OPCODE(UnpackHalf2x16, F32x2, U32, )
OPCODE(PackDouble2x32, F64, U32x2, )
OPCODE(UnpackDouble2x32, U32x2, F64, )
// Pseudo-operation, handled specially at final emit
OPCODE(GetZeroFromOp, U1, Opaque, )
OPCODE(GetSignFromOp, U1, Opaque, )
OPCODE(GetCarryFromOp, U1, Opaque, )
OPCODE(GetOverflowFromOp, U1, Opaque, )
OPCODE(GetSparseFromOp, U1, Opaque, )
OPCODE(GetInBoundsFromOp, U1, Opaque, )
// Floating-point operations
OPCODE(FPAbs16, F16, F16, )
OPCODE(FPAbs32, F32, F32, )
OPCODE(FPAbs64, F64, F64, )
OPCODE(FPAdd16, F16, F16, F16, )
OPCODE(FPAdd32, F32, F32, F32, )
OPCODE(FPAdd64, F64, F64, F64, )
OPCODE(FPFma16, F16, F16, F16, F16, )
OPCODE(FPFma32, F32, F32, F32, F32, )
OPCODE(FPFma64, F64, F64, F64, F64, )
OPCODE(FPMax32, F32, F32, F32, )
OPCODE(FPMax64, F64, F64, F64, )
OPCODE(FPMin32, F32, F32, F32, )
OPCODE(FPMin64, F64, F64, F64, )
OPCODE(FPMul16, F16, F16, F16, )
OPCODE(FPMul32, F32, F32, F32, )
OPCODE(FPMul64, F64, F64, F64, )
OPCODE(FPNeg16, F16, F16, )
OPCODE(FPNeg32, F32, F32, )
OPCODE(FPNeg64, F64, F64, )
OPCODE(FPRecip32, F32, F32, )
OPCODE(FPRecip64, F64, F64, )
OPCODE(FPRecipSqrt32, F32, F32, )
OPCODE(FPRecipSqrt64, F64, F64, )
OPCODE(FPSqrt, F32, F32, )
OPCODE(FPSin, F32, F32, )
OPCODE(FPExp2, F32, F32, )
OPCODE(FPCos, F32, F32, )
OPCODE(FPLog2, F32, F32, )
OPCODE(FPSaturate16, F16, F16, )
OPCODE(FPSaturate32, F32, F32, )
OPCODE(FPSaturate64, F64, F64, )
OPCODE(FPClamp16, F16, F16, F16, F16, )
OPCODE(FPClamp32, F32, F32, F32, F32, )
OPCODE(FPClamp64, F64, F64, F64, F64, )
OPCODE(FPRoundEven16, F16, F16, )
OPCODE(FPRoundEven32, F32, F32, )
OPCODE(FPRoundEven64, F64, F64, )
OPCODE(FPFloor16, F16, F16, )
OPCODE(FPFloor32, F32, F32, )
OPCODE(FPFloor64, F64, F64, )
OPCODE(FPCeil16, F16, F16, )
OPCODE(FPCeil32, F32, F32, )
OPCODE(FPCeil64, F64, F64, )
OPCODE(FPTrunc16, F16, F16, )
OPCODE(FPTrunc32, F32, F32, )
OPCODE(FPTrunc64, F64, F64, )
OPCODE(FPOrdEqual16, U1, F16, F16, )
OPCODE(FPOrdEqual32, U1, F32, F32, )
OPCODE(FPOrdEqual64, U1, F64, F64, )
OPCODE(FPUnordEqual16, U1, F16, F16, )
OPCODE(FPUnordEqual32, U1, F32, F32, )
OPCODE(FPUnordEqual64, U1, F64, F64, )
OPCODE(FPOrdNotEqual16, U1, F16, F16, )
OPCODE(FPOrdNotEqual32, U1, F32, F32, )
OPCODE(FPOrdNotEqual64, U1, F64, F64, )
OPCODE(FPUnordNotEqual16, U1, F16, F16, )
OPCODE(FPUnordNotEqual32, U1, F32, F32, )
OPCODE(FPUnordNotEqual64, U1, F64, F64, )
OPCODE(FPOrdLessThan16, U1, F16, F16, )
OPCODE(FPOrdLessThan32, U1, F32, F32, )
OPCODE(FPOrdLessThan64, U1, F64, F64, )
OPCODE(FPUnordLessThan16, U1, F16, F16, )
OPCODE(FPUnordLessThan32, U1, F32, F32, )
OPCODE(FPUnordLessThan64, U1, F64, F64, )
OPCODE(FPOrdGreaterThan16, U1, F16, F16, )
OPCODE(FPOrdGreaterThan32, U1, F32, F32, )
OPCODE(FPOrdGreaterThan64, U1, F64, F64, )
OPCODE(FPUnordGreaterThan16, U1, F16, F16, )
OPCODE(FPUnordGreaterThan32, U1, F32, F32, )
OPCODE(FPUnordGreaterThan64, U1, F64, F64, )
OPCODE(FPOrdLessThanEqual16, U1, F16, F16, )
OPCODE(FPOrdLessThanEqual32, U1, F32, F32, )
OPCODE(FPOrdLessThanEqual64, U1, F64, F64, )
OPCODE(FPUnordLessThanEqual16, U1, F16, F16, )
OPCODE(FPUnordLessThanEqual32, U1, F32, F32, )
OPCODE(FPUnordLessThanEqual64, U1, F64, F64, )
OPCODE(FPOrdGreaterThanEqual16, U1, F16, F16, )
OPCODE(FPOrdGreaterThanEqual32, U1, F32, F32, )
OPCODE(FPOrdGreaterThanEqual64, U1, F64, F64, )
OPCODE(FPUnordGreaterThanEqual16, U1, F16, F16, )
OPCODE(FPUnordGreaterThanEqual32, U1, F32, F32, )
OPCODE(FPUnordGreaterThanEqual64, U1, F64, F64, )
OPCODE(FPIsNan16, U1, F16, )
OPCODE(FPIsNan32, U1, F32, )
OPCODE(FPIsNan64, U1, F64, )
// Integer operations
OPCODE(IAdd32, U32, U32, U32, )
OPCODE(IAdd64, U64, U64, U64, )
OPCODE(ISub32, U32, U32, U32, )
OPCODE(ISub64, U64, U64, U64, )
OPCODE(IMul32, U32, U32, U32, )
OPCODE(SDiv32, U32, U32, U32, )
OPCODE(UDiv32, U32, U32, U32, )
OPCODE(INeg32, U32, U32, )
OPCODE(INeg64, U64, U64, )
OPCODE(IAbs32, U32, U32, )
OPCODE(IAbs64, U64, U64, )
OPCODE(ShiftLeftLogical32, U32, U32, U32, )
OPCODE(ShiftLeftLogical64, U64, U64, U32, )
OPCODE(ShiftRightLogical32, U32, U32, U32, )
OPCODE(ShiftRightLogical64, U64, U64, U32, )
OPCODE(ShiftRightArithmetic32, U32, U32, U32, )
OPCODE(ShiftRightArithmetic64, U64, U64, U32, )
OPCODE(BitwiseAnd32, U32, U32, U32, )
OPCODE(BitwiseOr32, U32, U32, U32, )
OPCODE(BitwiseXor32, U32, U32, U32, )
OPCODE(BitFieldInsert, U32, U32, U32, U32, U32, )
OPCODE(BitFieldSExtract, U32, U32, U32, U32, )
OPCODE(BitFieldUExtract, U32, U32, U32, U32, )
OPCODE(BitReverse32, U32, U32, )
OPCODE(BitCount32, U32, U32, )
OPCODE(BitwiseNot32, U32, U32, )
OPCODE(FindSMsb32, U32, U32, )
OPCODE(FindUMsb32, U32, U32, )
OPCODE(SMin32, U32, U32, U32, )
OPCODE(UMin32, U32, U32, U32, )
OPCODE(SMax32, U32, U32, U32, )
OPCODE(UMax32, U32, U32, U32, )
OPCODE(SClamp32, U32, U32, U32, U32, )
OPCODE(UClamp32, U32, U32, U32, U32, )
OPCODE(SLessThan, U1, U32, U32, )
OPCODE(ULessThan, U1, U32, U32, )
OPCODE(IEqual, U1, U32, U32, )
OPCODE(SLessThanEqual, U1, U32, U32, )
OPCODE(ULessThanEqual, U1, U32, U32, )
OPCODE(SGreaterThan, U1, U32, U32, )
OPCODE(UGreaterThan, U1, U32, U32, )
OPCODE(INotEqual, U1, U32, U32, )
OPCODE(SGreaterThanEqual, U1, U32, U32, )
OPCODE(UGreaterThanEqual, U1, U32, U32, )
// Atomic operations
OPCODE(SharedAtomicIAdd32, U32, U32, U32, )
OPCODE(SharedAtomicSMin32, U32, U32, U32, )
OPCODE(SharedAtomicUMin32, U32, U32, U32, )
OPCODE(SharedAtomicSMax32, U32, U32, U32, )
OPCODE(SharedAtomicUMax32, U32, U32, U32, )
OPCODE(SharedAtomicInc32, U32, U32, U32, )
OPCODE(SharedAtomicDec32, U32, U32, U32, )
OPCODE(SharedAtomicAnd32, U32, U32, U32, )
OPCODE(SharedAtomicOr32, U32, U32, U32, )
OPCODE(SharedAtomicXor32, U32, U32, U32, )
OPCODE(SharedAtomicExchange32, U32, U32, U32, )
OPCODE(SharedAtomicExchange64, U64, U32, U64, )
OPCODE(SharedAtomicExchange32x2, U32x2, U32, U32x2, )
OPCODE(GlobalAtomicIAdd32, U32, U64, U32, )
OPCODE(GlobalAtomicSMin32, U32, U64, U32, )
OPCODE(GlobalAtomicUMin32, U32, U64, U32, )
OPCODE(GlobalAtomicSMax32, U32, U64, U32, )
OPCODE(GlobalAtomicUMax32, U32, U64, U32, )
OPCODE(GlobalAtomicInc32, U32, U64, U32, )
OPCODE(GlobalAtomicDec32, U32, U64, U32, )
OPCODE(GlobalAtomicAnd32, U32, U64, U32, )
OPCODE(GlobalAtomicOr32, U32, U64, U32, )
OPCODE(GlobalAtomicXor32, U32, U64, U32, )
OPCODE(GlobalAtomicExchange32, U32, U64, U32, )
OPCODE(GlobalAtomicIAdd64, U64, U64, U64, )
OPCODE(GlobalAtomicSMin64, U64, U64, U64, )
OPCODE(GlobalAtomicUMin64, U64, U64, U64, )
OPCODE(GlobalAtomicSMax64, U64, U64, U64, )
OPCODE(GlobalAtomicUMax64, U64, U64, U64, )
OPCODE(GlobalAtomicAnd64, U64, U64, U64, )
OPCODE(GlobalAtomicOr64, U64, U64, U64, )
OPCODE(GlobalAtomicXor64, U64, U64, U64, )
OPCODE(GlobalAtomicExchange64, U64, U64, U64, )
OPCODE(GlobalAtomicIAdd32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicSMin32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicUMin32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicSMax32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicUMax32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicAnd32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicOr32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicXor32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicExchange32x2, U32x2, U32x2, U32x2, )
OPCODE(GlobalAtomicAddF32, F32, U64, F32, )
OPCODE(GlobalAtomicAddF16x2, U32, U64, F16x2, )
OPCODE(GlobalAtomicAddF32x2, U32, U64, F32x2, )
OPCODE(GlobalAtomicMinF16x2, U32, U64, F16x2, )
OPCODE(GlobalAtomicMinF32x2, U32, U64, F32x2, )
OPCODE(GlobalAtomicMaxF16x2, U32, U64, F16x2, )
OPCODE(GlobalAtomicMaxF32x2, U32, U64, F32x2, )
OPCODE(StorageAtomicIAdd32, U32, U32, U32, U32, )
OPCODE(StorageAtomicSMin32, U32, U32, U32, U32, )
OPCODE(StorageAtomicUMin32, U32, U32, U32, U32, )
OPCODE(StorageAtomicSMax32, U32, U32, U32, U32, )
OPCODE(StorageAtomicUMax32, U32, U32, U32, U32, )
OPCODE(StorageAtomicInc32, U32, U32, U32, U32, )
OPCODE(StorageAtomicDec32, U32, U32, U32, U32, )
OPCODE(StorageAtomicAnd32, U32, U32, U32, U32, )
OPCODE(StorageAtomicOr32, U32, U32, U32, U32, )
OPCODE(StorageAtomicXor32, U32, U32, U32, U32, )
OPCODE(StorageAtomicExchange32, U32, U32, U32, U32, )
OPCODE(StorageAtomicIAdd64, U64, U32, U32, U64, )
OPCODE(StorageAtomicSMin64, U64, U32, U32, U64, )
OPCODE(StorageAtomicUMin64, U64, U32, U32, U64, )
OPCODE(StorageAtomicSMax64, U64, U32, U32, U64, )
OPCODE(StorageAtomicUMax64, U64, U32, U32, U64, )
OPCODE(StorageAtomicAnd64, U64, U32, U32, U64, )
OPCODE(StorageAtomicOr64, U64, U32, U32, U64, )
OPCODE(StorageAtomicXor64, U64, U32, U32, U64, )
OPCODE(StorageAtomicExchange64, U64, U32, U32, U64, )
OPCODE(StorageAtomicIAdd32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicSMin32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicUMin32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicSMax32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicUMax32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicAnd32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicOr32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicXor32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicExchange32x2, U32x2, U32, U32, U32x2, )
OPCODE(StorageAtomicAddF32, F32, U32, U32, F32, )
OPCODE(StorageAtomicAddF16x2, U32, U32, U32, F16x2, )
OPCODE(StorageAtomicAddF32x2, U32, U32, U32, F32x2, )
OPCODE(StorageAtomicMinF16x2, U32, U32, U32, F16x2, )
OPCODE(StorageAtomicMinF32x2, U32, U32, U32, F32x2, )
OPCODE(StorageAtomicMaxF16x2, U32, U32, U32, F16x2, )
OPCODE(StorageAtomicMaxF32x2, U32, U32, U32, F32x2, )
// Logical operations
OPCODE(LogicalOr, U1, U1, U1, )
OPCODE(LogicalAnd, U1, U1, U1, )
OPCODE(LogicalXor, U1, U1, U1, )
OPCODE(LogicalNot, U1, U1, )
// Conversion operations
OPCODE(ConvertS16F16, U32, F16, )
OPCODE(ConvertS16F32, U32, F32, )
OPCODE(ConvertS16F64, U32, F64, )
OPCODE(ConvertS32F16, U32, F16, )
OPCODE(ConvertS32F32, U32, F32, )
OPCODE(ConvertS32F64, U32, F64, )
OPCODE(ConvertS64F16, U64, F16, )
OPCODE(ConvertS64F32, U64, F32, )
OPCODE(ConvertS64F64, U64, F64, )
OPCODE(ConvertU16F16, U32, F16, )
OPCODE(ConvertU16F32, U32, F32, )
OPCODE(ConvertU16F64, U32, F64, )
OPCODE(ConvertU32F16, U32, F16, )
OPCODE(ConvertU32F32, U32, F32, )
OPCODE(ConvertU32F64, U32, F64, )
OPCODE(ConvertU64F16, U64, F16, )
OPCODE(ConvertU64F32, U64, F32, )
OPCODE(ConvertU64F64, U64, F64, )
OPCODE(ConvertU64U32, U64, U32, )
OPCODE(ConvertU32U64, U32, U64, )
OPCODE(ConvertF16F32, F16, F32, )
OPCODE(ConvertF32F16, F32, F16, )
OPCODE(ConvertF32F64, F32, F64, )
OPCODE(ConvertF64F32, F64, F32, )
OPCODE(ConvertF16S8, F16, U32, )
OPCODE(ConvertF16S16, F16, U32, )
OPCODE(ConvertF16S32, F16, U32, )
OPCODE(ConvertF16S64, F16, U64, )
OPCODE(ConvertF16U8, F16, U32, )
OPCODE(ConvertF16U16, F16, U32, )
OPCODE(ConvertF16U32, F16, U32, )
OPCODE(ConvertF16U64, F16, U64, )
OPCODE(ConvertF32S8, F32, U32, )
OPCODE(ConvertF32S16, F32, U32, )
OPCODE(ConvertF32S32, F32, U32, )
OPCODE(ConvertF32S64, F32, U64, )
OPCODE(ConvertF32U8, F32, U32, )
OPCODE(ConvertF32U16, F32, U32, )
OPCODE(ConvertF32U32, F32, U32, )
OPCODE(ConvertF32U64, F32, U64, )
OPCODE(ConvertF64S8, F64, U32, )
OPCODE(ConvertF64S16, F64, U32, )
OPCODE(ConvertF64S32, F64, U32, )
OPCODE(ConvertF64S64, F64, U64, )
OPCODE(ConvertF64U8, F64, U32, )
OPCODE(ConvertF64U16, F64, U32, )
OPCODE(ConvertF64U32, F64, U32, )
OPCODE(ConvertF64U64, F64, U64, )
OPCODE(ConvertU16U32, U16, U32, )
OPCODE(ConvertU32U16, U32, U16, )
OPCODE(ConvertU8U32, U8, U32, )
OPCODE(ConvertU32U8, U32, U8, )
OPCODE(ConvertS32S8, U32, U8, )
OPCODE(ConvertS32S16, U32, U16, )
// Image operations
OPCODE(BindlessImageSampleImplicitLod, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BindlessImageSampleExplicitLod, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BindlessImageSampleDrefImplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, )
OPCODE(BindlessImageSampleDrefExplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, )
OPCODE(BindlessImageGather, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BindlessImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, )
OPCODE(BindlessImageFetch, F32x4, U32, Opaque, Opaque, U32, Opaque, )
OPCODE(BindlessImageQueryDimensions, U32x4, U32, U32, U1, )
OPCODE(BindlessImageQueryLod, F32x4, U32, Opaque, )
OPCODE(BindlessImageGradient, F32x4, U32, Opaque, Opaque, Opaque, Opaque, )
OPCODE(BindlessImageRead, U32x4, U32, Opaque, )
OPCODE(BindlessImageWrite, Void, U32, Opaque, U32x4, )
OPCODE(BoundImageSampleImplicitLod, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BoundImageSampleExplicitLod, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BoundImageSampleDrefImplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, )
OPCODE(BoundImageSampleDrefExplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, )
OPCODE(BoundImageGather, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BoundImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, )
OPCODE(BoundImageFetch, F32x4, U32, Opaque, Opaque, U32, Opaque, )
OPCODE(BoundImageQueryDimensions, U32x4, U32, U32, U1, )
OPCODE(BoundImageQueryLod, F32x4, U32, Opaque, )
OPCODE(BoundImageGradient, F32x4, U32, Opaque, Opaque, Opaque, Opaque, )
OPCODE(BoundImageRead, U32x4, U32, Opaque, )
OPCODE(BoundImageWrite, Void, U32, Opaque, U32x4, )
OPCODE(ImageSampleImplicitLod, F32x4, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageSampleExplicitLod, F32x4, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageSampleDrefImplicitLod, F32, Opaque, Opaque, F32, Opaque, Opaque, )
OPCODE(ImageSampleDrefExplicitLod, F32, Opaque, Opaque, F32, Opaque, Opaque, )
OPCODE(ImageGather, F32x4, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageGatherDref, F32x4, Opaque, Opaque, Opaque, Opaque, F32, )
OPCODE(ImageFetch, F32x4, Opaque, Opaque, Opaque, U32, Opaque, )
OPCODE(ImageQueryDimensions, U32x4, Opaque, U32, U1, )
OPCODE(ImageQueryLod, F32x4, Opaque, Opaque, )
OPCODE(ImageGradient, F32x4, Opaque, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageRead, U32x4, Opaque, Opaque, )
OPCODE(ImageWrite, Void, Opaque, Opaque, U32x4, )
OPCODE(IsTextureScaled, U1, U32, )
OPCODE(IsImageScaled, U1, U32, )
// Atomic Image operations
OPCODE(BindlessImageAtomicIAdd32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicSMin32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicUMin32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicSMax32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicUMax32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicInc32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicDec32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicAnd32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicOr32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicXor32, U32, U32, Opaque, U32, )
OPCODE(BindlessImageAtomicExchange32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicIAdd32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicSMin32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicUMin32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicSMax32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicUMax32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicInc32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicDec32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicAnd32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicOr32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicXor32, U32, U32, Opaque, U32, )
OPCODE(BoundImageAtomicExchange32, U32, U32, Opaque, U32, )
OPCODE(ImageAtomicIAdd32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicSMin32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicUMin32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicSMax32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicUMax32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicInc32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicDec32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicAnd32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicOr32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicXor32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicExchange32, U32, Opaque, Opaque, U32, )
// Warp operations
OPCODE(LaneId, U32, )
OPCODE(VoteAll, U1, U1, )
OPCODE(VoteAny, U1, U1, )
OPCODE(VoteEqual, U1, U1, )
OPCODE(SubgroupBallot, U32, U1, )
OPCODE(SubgroupEqMask, U32, )
OPCODE(SubgroupLtMask, U32, )
OPCODE(SubgroupLeMask, U32, )
OPCODE(SubgroupGtMask, U32, )
OPCODE(SubgroupGeMask, U32, )
OPCODE(ShuffleIndex, U32, U32, U32, U32, U32, )
OPCODE(ShuffleUp, U32, U32, U32, U32, U32, )
OPCODE(ShuffleDown, U32, U32, U32, U32, U32, )
OPCODE(ShuffleButterfly, U32, U32, U32, U32, U32, )
OPCODE(FSwizzleAdd, F32, F32, F32, U32, )
OPCODE(DPdxFine, F32, F32, )
OPCODE(DPdyFine, F32, F32, )
OPCODE(DPdxCoarse, F32, F32, )
OPCODE(DPdyCoarse, F32, F32, )

View File

@@ -0,0 +1,27 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/patch.h"
namespace Shader::IR {
bool IsGeneric(Patch patch) noexcept {
return patch >= Patch::Component0 && patch <= Patch::Component119;
}
u32 GenericPatchIndex(Patch patch) {
if (!IsGeneric(patch)) {
throw InvalidArgument("Patch {} is not generic", patch);
}
return (static_cast<u32>(patch) - static_cast<u32>(Patch::Component0)) / 4;
}
u32 GenericPatchElement(Patch patch) {
if (!IsGeneric(patch)) {
throw InvalidArgument("Patch {} is not generic", patch);
}
return (static_cast<u32>(patch) - static_cast<u32>(Patch::Component0)) % 4;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,146 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "common/common_types.h"
namespace Shader::IR {
enum class Patch : u64 {
TessellationLodLeft,
TessellationLodTop,
TessellationLodRight,
TessellationLodBottom,
TessellationLodInteriorU,
TessellationLodInteriorV,
Component0,
Component1,
Component2,
Component3,
Component4,
Component5,
Component6,
Component7,
Component8,
Component9,
Component10,
Component11,
Component12,
Component13,
Component14,
Component15,
Component16,
Component17,
Component18,
Component19,
Component20,
Component21,
Component22,
Component23,
Component24,
Component25,
Component26,
Component27,
Component28,
Component29,
Component30,
Component31,
Component32,
Component33,
Component34,
Component35,
Component36,
Component37,
Component38,
Component39,
Component40,
Component41,
Component42,
Component43,
Component44,
Component45,
Component46,
Component47,
Component48,
Component49,
Component50,
Component51,
Component52,
Component53,
Component54,
Component55,
Component56,
Component57,
Component58,
Component59,
Component60,
Component61,
Component62,
Component63,
Component64,
Component65,
Component66,
Component67,
Component68,
Component69,
Component70,
Component71,
Component72,
Component73,
Component74,
Component75,
Component76,
Component77,
Component78,
Component79,
Component80,
Component81,
Component82,
Component83,
Component84,
Component85,
Component86,
Component87,
Component88,
Component89,
Component90,
Component91,
Component92,
Component93,
Component94,
Component95,
Component96,
Component97,
Component98,
Component99,
Component100,
Component101,
Component102,
Component103,
Component104,
Component105,
Component106,
Component107,
Component108,
Component109,
Component110,
Component111,
Component112,
Component113,
Component114,
Component115,
Component116,
Component117,
Component118,
Component119,
};
static_assert(static_cast<u64>(Patch::Component119) == 125);
[[nodiscard]] bool IsGeneric(Patch patch) noexcept;
[[nodiscard]] u32 GenericPatchIndex(Patch patch);
[[nodiscard]] u32 GenericPatchElement(Patch patch);
} // namespace Shader::IR

View File

@@ -0,0 +1,45 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <boost/container/flat_set.hpp>
#include <boost/container/small_vector.hpp>
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/post_order.h"
namespace Shader::IR {
BlockList PostOrder(const AbstractSyntaxNode& root) {
boost::container::small_vector<Block*, 16> block_stack;
boost::container::flat_set<Block*> visited;
BlockList post_order_blocks;
if (root.type != AbstractSyntaxNode::Type::Block) {
throw LogicError("First node in abstract syntax list root is not a block");
}
Block* const first_block{root.data.block};
visited.insert(first_block);
block_stack.push_back(first_block);
while (!block_stack.empty()) {
Block* const block{block_stack.back()};
const auto visit{[&](Block* branch) {
if (!visited.insert(branch).second) {
return false;
}
// Calling push_back twice is faster than insert on MSVC
block_stack.push_back(block);
block_stack.push_back(branch);
return true;
}};
block_stack.pop_back();
if (std::ranges::none_of(block->ImmSuccessors(), visit)) {
post_order_blocks.push_back(block);
}
}
return post_order_blocks;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,13 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "shader_recompiler/frontend/ir/abstract_syntax_list.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
namespace Shader::IR {
BlockList PostOrder(const AbstractSyntaxNode& root);
} // namespace Shader::IR

View File

@@ -0,0 +1,43 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <fmt/ranges.h>
namespace Shader::IR {
enum class Pred : u64 {
P0,
P1,
P2,
P3,
P4,
P5,
P6,
PT,
};
constexpr size_t NUM_USER_PREDS = 7;
constexpr size_t NUM_PREDS = 8;
[[nodiscard]] constexpr size_t PredIndex(Pred pred) noexcept {
return static_cast<size_t>(pred);
}
} // namespace Shader::IR
template <>
struct fmt::formatter<Shader::IR::Pred> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::IR::Pred& pred, FormatContext& ctx) const {
if (pred == Shader::IR::Pred::PT) {
return fmt::format_to(ctx.out(), "PT");
} else {
return fmt::format_to(ctx.out(), "P{}", static_cast<int>(pred));
}
}
};

View File

@@ -0,0 +1,31 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <map>
#include <string>
#include <fmt/ranges.h>
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/program.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
std::string DumpProgram(const Program& program) {
size_t index{0};
std::map<const IR::Inst*, size_t> inst_to_index;
std::map<const IR::Block*, size_t> block_to_index;
for (const IR::Block* const block : program.blocks) {
block_to_index.emplace(block, index);
++index;
}
std::string ret;
for (const auto& block : program.blocks) {
ret += IR::DumpBlock(*block, block_to_index, inst_to_index, index) + '\n';
}
return ret;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,34 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <string>
#include "shader_recompiler/frontend/ir/abstract_syntax_list.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/program_header.h"
#include "shader_recompiler/shader_info.h"
#include "shader_recompiler/stage.h"
namespace Shader::IR {
struct Program {
AbstractSyntaxList syntax_list;
BlockList blocks;
BlockList post_order_blocks;
Info info;
Stage stage{};
std::array<u32, 3> workgroup_size{};
OutputTopology output_topology{};
u32 output_vertices{};
u32 invocations{};
u32 local_memory_size{};
u32 shared_memory_size{};
bool is_geometry_passthrough{};
};
[[nodiscard]] std::string DumpProgram(const Program& program);
} // namespace Shader::IR

View File

@@ -0,0 +1,331 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <fmt/ranges.h>
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
namespace Shader::IR {
enum class Reg : u64 {
R0,
R1,
R2,
R3,
R4,
R5,
R6,
R7,
R8,
R9,
R10,
R11,
R12,
R13,
R14,
R15,
R16,
R17,
R18,
R19,
R20,
R21,
R22,
R23,
R24,
R25,
R26,
R27,
R28,
R29,
R30,
R31,
R32,
R33,
R34,
R35,
R36,
R37,
R38,
R39,
R40,
R41,
R42,
R43,
R44,
R45,
R46,
R47,
R48,
R49,
R50,
R51,
R52,
R53,
R54,
R55,
R56,
R57,
R58,
R59,
R60,
R61,
R62,
R63,
R64,
R65,
R66,
R67,
R68,
R69,
R70,
R71,
R72,
R73,
R74,
R75,
R76,
R77,
R78,
R79,
R80,
R81,
R82,
R83,
R84,
R85,
R86,
R87,
R88,
R89,
R90,
R91,
R92,
R93,
R94,
R95,
R96,
R97,
R98,
R99,
R100,
R101,
R102,
R103,
R104,
R105,
R106,
R107,
R108,
R109,
R110,
R111,
R112,
R113,
R114,
R115,
R116,
R117,
R118,
R119,
R120,
R121,
R122,
R123,
R124,
R125,
R126,
R127,
R128,
R129,
R130,
R131,
R132,
R133,
R134,
R135,
R136,
R137,
R138,
R139,
R140,
R141,
R142,
R143,
R144,
R145,
R146,
R147,
R148,
R149,
R150,
R151,
R152,
R153,
R154,
R155,
R156,
R157,
R158,
R159,
R160,
R161,
R162,
R163,
R164,
R165,
R166,
R167,
R168,
R169,
R170,
R171,
R172,
R173,
R174,
R175,
R176,
R177,
R178,
R179,
R180,
R181,
R182,
R183,
R184,
R185,
R186,
R187,
R188,
R189,
R190,
R191,
R192,
R193,
R194,
R195,
R196,
R197,
R198,
R199,
R200,
R201,
R202,
R203,
R204,
R205,
R206,
R207,
R208,
R209,
R210,
R211,
R212,
R213,
R214,
R215,
R216,
R217,
R218,
R219,
R220,
R221,
R222,
R223,
R224,
R225,
R226,
R227,
R228,
R229,
R230,
R231,
R232,
R233,
R234,
R235,
R236,
R237,
R238,
R239,
R240,
R241,
R242,
R243,
R244,
R245,
R246,
R247,
R248,
R249,
R250,
R251,
R252,
R253,
R254,
RZ,
};
static_assert(static_cast<int>(Reg::RZ) == 255);
constexpr size_t NUM_USER_REGS = 255;
constexpr size_t NUM_REGS = 256;
[[nodiscard]] constexpr Reg operator+(Reg reg, int num) {
if (reg == Reg::RZ) {
// Adding or subtracting registers from RZ yields RZ
return Reg::RZ;
}
const int result{static_cast<int>(reg) + num};
if (result >= static_cast<int>(Reg::RZ)) {
throw LogicError("Overflow on register arithmetic");
}
if (result < 0) {
throw LogicError("Underflow on register arithmetic");
}
return static_cast<Reg>(result);
}
[[nodiscard]] constexpr Reg operator-(Reg reg, int num) {
return reg + (-num);
}
constexpr Reg operator++(Reg& reg) {
reg = reg + 1;
return reg;
}
constexpr Reg operator++(Reg& reg, int) {
const Reg copy{reg};
reg = reg + 1;
return copy;
}
[[nodiscard]] constexpr size_t RegIndex(Reg reg) noexcept {
return static_cast<size_t>(reg);
}
[[nodiscard]] constexpr bool IsAligned(Reg reg, size_t align) {
return RegIndex(reg) % align == 0 || reg == Reg::RZ;
}
} // namespace Shader::IR
template <>
struct fmt::formatter<Shader::IR::Reg> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::IR::Reg& reg, FormatContext& ctx) const {
if (reg == Shader::IR::Reg::RZ) {
return fmt::format_to(ctx.out(), "RZ");
} else if (static_cast<int>(reg) >= 0 && static_cast<int>(reg) < 255) {
return fmt::format_to(ctx.out(), "R{}", static_cast<int>(reg));
} else {
throw Shader::LogicError("Invalid register with raw value {}", static_cast<int>(reg));
}
}
};

View File

@@ -0,0 +1,37 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <array>
#include <string>
#include "shader_recompiler/frontend/ir/type.h"
namespace Shader::IR {
std::string NameOf(Type type) {
static constexpr std::array names{
"Opaque", "Label", "Reg", "Pred", "Attribute", "U1", "U8", "U16", "U32",
"U64", "F16", "F32", "F64", "U32x2", "U32x3", "U32x4", "F16x2", "F16x3",
"F16x4", "F32x2", "F32x3", "F32x4", "F64x2", "F64x3", "F64x4",
};
const size_t bits{static_cast<size_t>(type)};
if (bits == 0) {
return "Void";
}
std::string result;
for (size_t i = 0; i < names.size(); i++) {
if ((bits & (size_t{1} << i)) != 0) {
if (!result.empty()) {
result += '|';
}
result += names[i];
}
}
return result;
}
bool AreTypesCompatible(Type lhs, Type rhs) noexcept {
return lhs == rhs || lhs == Type::Opaque || rhs == Type::Opaque;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,60 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <string>
#include <fmt/ranges.h>
#include "common/common_funcs.h"
#include "shader_recompiler/exception.h"
namespace Shader::IR {
enum class Type {
Void = 0,
Opaque = 1 << 0,
Reg = 1 << 1,
Pred = 1 << 2,
Attribute = 1 << 3,
Patch = 1 << 4,
U1 = 1 << 5,
U8 = 1 << 6,
U16 = 1 << 7,
U32 = 1 << 8,
U64 = 1 << 9,
F16 = 1 << 10,
F32 = 1 << 11,
F64 = 1 << 12,
U32x2 = 1 << 13,
U32x3 = 1 << 14,
U32x4 = 1 << 15,
F16x2 = 1 << 16,
F16x3 = 1 << 17,
F16x4 = 1 << 18,
F32x2 = 1 << 19,
F32x3 = 1 << 20,
F32x4 = 1 << 21,
F64x2 = 1 << 22,
F64x3 = 1 << 23,
F64x4 = 1 << 24,
};
DECLARE_ENUM_FLAG_OPERATORS(Type)
[[nodiscard]] std::string NameOf(Type type);
[[nodiscard]] bool AreTypesCompatible(Type lhs, Type rhs) noexcept;
} // namespace Shader::IR
template <>
struct fmt::formatter<Shader::IR::Type> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::IR::Type& type, FormatContext& ctx) const {
return fmt::format_to(ctx.out(), "{}", NameOf(type));
}
};

View File

@@ -0,0 +1,97 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
Value::Value(IR::Inst* value) noexcept : type{Type::Opaque}, inst{value} {}
Value::Value(IR::Reg value) noexcept : type{Type::Reg}, reg{value} {}
Value::Value(IR::Pred value) noexcept : type{Type::Pred}, pred{value} {}
Value::Value(IR::Attribute value) noexcept : type{Type::Attribute}, attribute{value} {}
Value::Value(IR::Patch value) noexcept : type{Type::Patch}, patch{value} {}
Value::Value(bool value) noexcept : type{Type::U1}, imm_u1{value} {}
Value::Value(u8 value) noexcept : type{Type::U8}, imm_u8{value} {}
Value::Value(u16 value) noexcept : type{Type::U16}, imm_u16{value} {}
Value::Value(u32 value) noexcept : type{Type::U32}, imm_u32{value} {}
Value::Value(f32 value) noexcept : type{Type::F32}, imm_f32{value} {}
Value::Value(u64 value) noexcept : type{Type::U64}, imm_u64{value} {}
Value::Value(f64 value) noexcept : type{Type::F64}, imm_f64{value} {}
IR::Type Value::Type() const noexcept {
if (IsPhi()) {
// The type of a phi node is stored in its flags
return inst->Flags<IR::Type>();
}
if (IsIdentity()) {
return inst->Arg(0).Type();
}
if (type == Type::Opaque) {
return inst->Type();
}
return type;
}
bool Value::operator==(const Value& other) const {
if (type != other.type) {
return false;
}
switch (type) {
case Type::Void:
return true;
case Type::Opaque:
return inst == other.inst;
case Type::Reg:
return reg == other.reg;
case Type::Pred:
return pred == other.pred;
case Type::Attribute:
return attribute == other.attribute;
case Type::Patch:
return patch == other.patch;
case Type::U1:
return imm_u1 == other.imm_u1;
case Type::U8:
return imm_u8 == other.imm_u8;
case Type::U16:
case Type::F16:
return imm_u16 == other.imm_u16;
case Type::U32:
case Type::F32:
return imm_u32 == other.imm_u32;
case Type::U64:
case Type::F64:
return imm_u64 == other.imm_u64;
case Type::U32x2:
case Type::U32x3:
case Type::U32x4:
case Type::F16x2:
case Type::F16x3:
case Type::F16x4:
case Type::F32x2:
case Type::F32x3:
case Type::F32x4:
case Type::F64x2:
case Type::F64x3:
case Type::F64x4:
break;
}
throw LogicError("Invalid type {}", type);
}
bool Value::operator!=(const Value& other) const {
return !operator==(other);
}
} // namespace Shader::IR

View File

@@ -0,0 +1,410 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <cstring>
#include <memory>
#include <type_traits>
#include <utility>
#include <vector>
#include <boost/container/small_vector.hpp>
#include <boost/intrusive/list.hpp>
#include "common/assert.h"
#include <numeric>
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/attribute.h"
#include "shader_recompiler/frontend/ir/opcodes.h"
#include "shader_recompiler/frontend/ir/patch.h"
#include "shader_recompiler/frontend/ir/pred.h"
#include "shader_recompiler/frontend/ir/reg.h"
#include "shader_recompiler/frontend/ir/type.h"
namespace Shader::IR {
class Block;
class Inst;
struct AssociatedInsts;
class Value {
public:
Value() noexcept = default;
explicit Value(IR::Inst* value) noexcept;
explicit Value(IR::Reg value) noexcept;
explicit Value(IR::Pred value) noexcept;
explicit Value(IR::Attribute value) noexcept;
explicit Value(IR::Patch value) noexcept;
explicit Value(bool value) noexcept;
explicit Value(u8 value) noexcept;
explicit Value(u16 value) noexcept;
explicit Value(u32 value) noexcept;
explicit Value(f32 value) noexcept;
explicit Value(u64 value) noexcept;
explicit Value(f64 value) noexcept;
[[nodiscard]] bool IsIdentity() const noexcept;
[[nodiscard]] bool IsPhi() const noexcept;
[[nodiscard]] bool IsEmpty() const noexcept;
[[nodiscard]] bool IsImmediate() const noexcept;
[[nodiscard]] IR::Type Type() const noexcept;
[[nodiscard]] IR::Inst* Inst() const;
[[nodiscard]] IR::Inst* InstRecursive() const;
[[nodiscard]] IR::Inst* TryInstRecursive() const;
[[nodiscard]] IR::Value Resolve() const;
[[nodiscard]] IR::Reg Reg() const;
[[nodiscard]] IR::Pred Pred() const;
[[nodiscard]] IR::Attribute Attribute() const;
[[nodiscard]] IR::Patch Patch() const;
[[nodiscard]] bool U1() const;
[[nodiscard]] u8 U8() const;
[[nodiscard]] u16 U16() const;
[[nodiscard]] u32 U32() const;
[[nodiscard]] f32 F32() const;
[[nodiscard]] u64 U64() const;
[[nodiscard]] f64 F64() const;
[[nodiscard]] bool operator==(const Value& other) const;
[[nodiscard]] bool operator!=(const Value& other) const;
private:
IR::Type type{};
union {
IR::Inst* inst{};
IR::Reg reg;
IR::Pred pred;
IR::Attribute attribute;
IR::Patch patch;
bool imm_u1;
u8 imm_u8;
u16 imm_u16;
u32 imm_u32;
f32 imm_f32;
u64 imm_u64;
f64 imm_f64;
};
};
static_assert(static_cast<u32>(IR::Type::Void) == 0, "memset relies on IR::Type being zero");
static_assert(std::is_trivially_copyable_v<Value>);
template <IR::Type type_>
class TypedValue : public Value {
public:
TypedValue() = default;
template <IR::Type other_type>
requires((other_type & type_) != IR::Type::Void)
explicit(false) TypedValue(const TypedValue<other_type>& value) : Value(value) {}
explicit TypedValue(const Value& value) : Value(value) {
if ((value.Type() & type_) == IR::Type::Void) {
throw InvalidArgument("Incompatible types {} and {}", type_, value.Type());
}
}
explicit TypedValue(IR::Inst* inst_) : TypedValue(Value(inst_)) {}
};
class Inst : public boost::intrusive::list_base_hook<> {
public:
explicit Inst(IR::Opcode op_, u32 flags_) noexcept;
explicit Inst(const Inst& base);
~Inst();
Inst& operator=(const Inst&) = delete;
Inst& operator=(Inst&&) = delete;
Inst(Inst&&) = delete;
/// Get the number of uses this instruction has.
[[nodiscard]] int UseCount() const noexcept {
return use_count;
}
/// Determines whether this instruction has uses or not.
[[nodiscard]] bool HasUses() const noexcept {
return use_count > 0;
}
/// Get the opcode this microinstruction represents.
[[nodiscard]] IR::Opcode GetOpcode() const noexcept {
return op;
}
/// Determines if there is a pseudo-operation associated with this instruction.
[[nodiscard]] bool HasAssociatedPseudoOperation() const noexcept {
return associated_insts != nullptr;
}
/// Determines whether or not this instruction may have side effects.
[[nodiscard]] bool MayHaveSideEffects() const noexcept;
/// Determines whether or not this instruction is a pseudo-instruction.
/// Pseudo-instructions depend on their parent instructions for their semantics.
[[nodiscard]] bool IsPseudoInstruction() const noexcept;
/// Determines if all arguments of this instruction are immediates.
[[nodiscard]] bool AreAllArgsImmediates() const;
/// Gets a pseudo-operation associated with this instruction
[[nodiscard]] Inst* GetAssociatedPseudoOperation(IR::Opcode opcode);
/// Get the type this instruction returns.
[[nodiscard]] IR::Type Type() const;
/// Get the number of arguments this instruction has.
[[nodiscard]] size_t NumArgs() const {
return op == IR::Opcode::Phi ? phi_args.size() : NumArgsOf(op);
}
/// Get the value of a given argument index.
[[nodiscard]] Value Arg(size_t index) const noexcept {
if (op == IR::Opcode::Phi) {
return phi_args[index].second;
} else {
return args[index];
}
}
/// Set the value of a given argument index.
void SetArg(size_t index, Value value);
/// Get a pointer to the block of a phi argument.
[[nodiscard]] Block* PhiBlock(size_t index) const;
/// Add phi operand to a phi instruction.
void AddPhiOperand(Block* predecessor, const Value& value);
/// Orders the Phi arguments from farthest away to nearest.
void OrderPhiArgs();
void Invalidate();
void ClearArgs();
void ReplaceUsesWith(Value replacement);
void ReplaceOpcode(IR::Opcode opcode);
template <typename FlagsType>
requires(sizeof(FlagsType) <= sizeof(u32) && std::is_trivially_copyable_v<FlagsType>)
[[nodiscard]] FlagsType Flags() const noexcept {
FlagsType ret;
std::memcpy(reinterpret_cast<char*>(&ret), &flags, sizeof(ret));
return ret;
}
template <typename FlagsType>
requires(sizeof(FlagsType) <= sizeof(u32) && std::is_trivially_copyable_v<FlagsType>)
void SetFlags(FlagsType value) noexcept {
std::memcpy(&flags, &value, sizeof(value));
}
/// Intrusively store the host definition of this instruction.
template <typename DefinitionType>
void SetDefinition(DefinitionType def) {
definition = std::bit_cast<u32>(def);
}
/// Return the intrusively stored host definition of this instruction.
template <typename DefinitionType>
[[nodiscard]] DefinitionType Definition() const noexcept {
return std::bit_cast<DefinitionType>(definition);
}
/// Destructively remove one reference count from the instruction
/// Useful for register allocation
void DestructiveRemoveUsage() {
--use_count;
}
/// Destructively add usages to the instruction
/// Useful for register allocation
void DestructiveAddUsage(int count) {
use_count += count;
}
private:
struct NonTriviallyDummy {
NonTriviallyDummy() noexcept {}
};
void Use(const Value& value);
void UndoUse(const Value& value);
IR::Opcode op{};
int use_count{};
u32 flags{};
u32 definition{};
union {
NonTriviallyDummy dummy{};
boost::container::small_vector<std::pair<Block*, Value>, 2> phi_args;
std::array<Value, 5> args;
};
std::unique_ptr<AssociatedInsts> associated_insts;
};
static_assert(sizeof(Inst) <= 128, "Inst size unintentionally increased");
struct AssociatedInsts {
union {
Inst* in_bounds_inst;
Inst* sparse_inst;
Inst* zero_inst{};
};
Inst* sign_inst{};
Inst* carry_inst{};
Inst* overflow_inst{};
};
using U1 = TypedValue<Type::U1>;
using U8 = TypedValue<Type::U8>;
using U16 = TypedValue<Type::U16>;
using U32 = TypedValue<Type::U32>;
using U64 = TypedValue<Type::U64>;
using F16 = TypedValue<Type::F16>;
using F32 = TypedValue<Type::F32>;
using F64 = TypedValue<Type::F64>;
using U32U64 = TypedValue<Type::U32 | Type::U64>;
using F32F64 = TypedValue<Type::F32 | Type::F64>;
using U8U16U32U64 = TypedValue<Type::U8 | Type::U16 | Type::U32 | Type::U64>;
using U16U32U64 = TypedValue<Type::U16 | Type::U32 | Type::U64>;
using F16F32F64 = TypedValue<Type::F16 | Type::F32 | Type::F64>;
using UAny = TypedValue<Type::U8 | Type::U16 | Type::U32 | Type::U64>;
inline bool Value::IsIdentity() const noexcept {
return type == Type::Opaque && inst->GetOpcode() == Opcode::Identity;
}
inline bool Value::IsPhi() const noexcept {
return type == Type::Opaque && inst->GetOpcode() == Opcode::Phi;
}
inline bool Value::IsEmpty() const noexcept {
return type == Type::Void;
}
inline bool Value::IsImmediate() const noexcept {
IR::Type current_type{type};
const IR::Inst* current_inst{inst};
while (current_type == Type::Opaque && current_inst->GetOpcode() == Opcode::Identity) {
const Value& arg{current_inst->Arg(0)};
current_type = arg.type;
current_inst = arg.inst;
}
return current_type != Type::Opaque;
}
inline IR::Inst* Value::Inst() const {
DEBUG_ASSERT(type == Type::Opaque);
return inst;
}
inline IR::Inst* Value::InstRecursive() const {
DEBUG_ASSERT(type == Type::Opaque);
if (IsIdentity()) {
return inst->Arg(0).InstRecursive();
}
return inst;
}
inline IR::Inst* Value::TryInstRecursive() const {
if (IsIdentity()) {
return inst->Arg(0).TryInstRecursive();
}
return type == Type::Opaque ? inst : nullptr;
}
inline IR::Value Value::Resolve() const {
if (IsIdentity()) {
return inst->Arg(0).Resolve();
}
return *this;
}
inline IR::Reg Value::Reg() const {
DEBUG_ASSERT(type == Type::Reg);
return reg;
}
inline IR::Pred Value::Pred() const {
DEBUG_ASSERT(type == Type::Pred);
return pred;
}
inline IR::Attribute Value::Attribute() const {
DEBUG_ASSERT(type == Type::Attribute);
return attribute;
}
inline IR::Patch Value::Patch() const {
DEBUG_ASSERT(type == Type::Patch);
return patch;
}
inline bool Value::U1() const {
if (IsIdentity()) {
return inst->Arg(0).U1();
}
DEBUG_ASSERT(type == Type::U1);
return imm_u1;
}
inline u8 Value::U8() const {
if (IsIdentity()) {
return inst->Arg(0).U8();
}
DEBUG_ASSERT(type == Type::U8);
return imm_u8;
}
inline u16 Value::U16() const {
if (IsIdentity()) {
return inst->Arg(0).U16();
}
DEBUG_ASSERT(type == Type::U16);
return imm_u16;
}
inline u32 Value::U32() const {
if (IsIdentity()) {
return inst->Arg(0).U32();
}
DEBUG_ASSERT(type == Type::U32);
return imm_u32;
}
inline f32 Value::F32() const {
if (IsIdentity()) {
return inst->Arg(0).F32();
}
DEBUG_ASSERT(type == Type::F32);
return imm_f32;
}
inline u64 Value::U64() const {
if (IsIdentity()) {
return inst->Arg(0).U64();
}
DEBUG_ASSERT(type == Type::U64);
return imm_u64;
}
inline f64 Value::F64() const {
if (IsIdentity()) {
return inst->Arg(0).F64();
}
DEBUG_ASSERT(type == Type::F64);
return imm_f64;
}
[[nodiscard]] inline bool IsPhi(const Inst& inst) {
return inst.GetOpcode() == Opcode::Phi;
}
} // namespace Shader::IR

View File

@@ -0,0 +1,645 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <array>
#include <optional>
#include <string>
#include <utility>
#include <fmt/ranges.h>
#include <ranges>
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/frontend/maxwell/decode.h"
#include "shader_recompiler/frontend/maxwell/indirect_branch_table_track.h"
#include "shader_recompiler/frontend/maxwell/location.h"
namespace Shader::Maxwell::Flow {
namespace {
struct Compare {
bool operator()(const Block& lhs, Location rhs) const noexcept {
return lhs.begin < rhs;
}
bool operator()(Location lhs, const Block& rhs) const noexcept {
return lhs < rhs.begin;
}
bool operator()(const Block& lhs, const Block& rhs) const noexcept {
return lhs.begin < rhs.begin;
}
};
u32 BranchOffset(Location pc, Instruction inst) {
return pc.Offset() + static_cast<u32>(inst.branch.Offset()) + 8u;
}
void Split(Block* old_block, Block* new_block, Location pc) {
if (pc <= old_block->begin || pc >= old_block->end) {
throw InvalidArgument("Invalid address to split={}", pc);
}
*new_block = Block{};
new_block->begin = pc;
new_block->end = old_block->end;
new_block->end_class = old_block->end_class;
new_block->cond = old_block->cond;
new_block->stack = old_block->stack;
new_block->branch_true = old_block->branch_true;
new_block->branch_false = old_block->branch_false;
new_block->function_call = old_block->function_call;
new_block->return_block = old_block->return_block;
new_block->branch_reg = old_block->branch_reg;
new_block->branch_offset = old_block->branch_offset;
new_block->indirect_branches = std::move(old_block->indirect_branches);
const Location old_begin{old_block->begin};
Stack old_stack{std::move(old_block->stack)};
*old_block = Block{};
old_block->begin = old_begin;
old_block->end = pc;
old_block->end_class = EndClass::Branch;
old_block->cond = IR::Condition(true);
old_block->stack = old_stack;
old_block->branch_true = new_block;
old_block->branch_false = nullptr;
}
Token OpcodeToken(Opcode opcode) {
switch (opcode) {
case Opcode::PBK:
case Opcode::BRK:
return Token::PBK;
case Opcode::PCNT:
case Opcode::CONT:
return Token::PCNT;
case Opcode::PEXIT:
case Opcode::EXIT:
return Token::PEXIT;
case Opcode::PLONGJMP:
case Opcode::LONGJMP:
return Token::PLONGJMP;
case Opcode::PRET:
case Opcode::RET:
case Opcode::CAL:
return Token::PRET;
case Opcode::SSY:
case Opcode::SYNC:
return Token::SSY;
default:
throw InvalidArgument("{}", opcode);
}
}
bool IsAbsoluteJump(Opcode opcode) {
switch (opcode) {
case Opcode::JCAL:
case Opcode::JMP:
case Opcode::JMX:
return true;
default:
return false;
}
}
bool HasFlowTest(Opcode opcode) {
switch (opcode) {
case Opcode::BRA:
case Opcode::BRX:
case Opcode::EXIT:
case Opcode::JMP:
case Opcode::JMX:
case Opcode::KIL:
case Opcode::BRK:
case Opcode::CONT:
case Opcode::LONGJMP:
case Opcode::RET:
case Opcode::SYNC:
return true;
case Opcode::CAL:
case Opcode::JCAL:
return false;
default:
throw InvalidArgument("Invalid branch {}", opcode);
}
}
std::string NameOf(const Block& block) {
if (block.begin.IsVirtual()) {
return fmt::format("\"Virtual {}\"", block.begin);
} else {
return fmt::format("\"{}\"", block.begin);
}
}
} // Anonymous namespace
void Stack::Push(Token token, Location target) {
entries.push_back({
.token = token,
.target{target},
});
}
std::pair<Location, Stack> Stack::Pop(Token token) const {
const std::optional<Location> pc{Peek(token)};
if (!pc) {
throw LogicError("Token could not be found");
}
return {*pc, Remove(token)};
}
std::optional<Location> Stack::Peek(Token token) const {
const auto it{std::find_if(entries.rbegin(), entries.rend(),
[token](const auto& entry) { return entry.token == token; })};
if (it == entries.rend()) {
return std::nullopt;
}
return it->target;
}
Stack Stack::Remove(Token token) const {
const auto it{std::find_if(entries.rbegin(), entries.rend(),
[token](const auto& entry) { return entry.token == token; })};
const auto pos{std::distance(entries.rbegin(), it)};
Stack result;
result.entries.insert(result.entries.end(), entries.begin(), entries.end() - pos - 1);
return result;
}
bool Block::Contains(Location pc) const noexcept {
return pc >= begin && pc < end;
}
Function::Function(ObjectPool<Block>& block_pool, Location start_address)
: entrypoint{start_address} {
Label& label{labels.emplace_back()};
label.address = start_address;
label.block = block_pool.Create(Block{});
label.block->begin = start_address;
label.block->end = start_address;
label.block->end_class = EndClass::Branch;
label.block->cond = IR::Condition(true);
label.block->branch_true = nullptr;
label.block->branch_false = nullptr;
}
CFG::CFG(Environment& env_, ObjectPool<Block>& block_pool_, Location start_address,
bool exits_to_dispatcher_)
: env{env_}, block_pool{block_pool_}, program_start{start_address}, exits_to_dispatcher{
exits_to_dispatcher_} {
if (exits_to_dispatcher) {
dispatch_block = block_pool.Create(Block{});
dispatch_block->begin = {};
dispatch_block->end = {};
dispatch_block->end_class = EndClass::Exit;
dispatch_block->cond = IR::Condition(true);
dispatch_block->stack = {};
dispatch_block->branch_true = nullptr;
dispatch_block->branch_false = nullptr;
}
functions.emplace_back(block_pool, start_address);
for (FunctionId function_id = 0; function_id < functions.size(); ++function_id) {
while (!functions[function_id].labels.empty()) {
Function& function{functions[function_id]};
Label label{function.labels.back()};
function.labels.pop_back();
AnalyzeLabel(function_id, label);
}
}
if (exits_to_dispatcher) {
const auto last_block{functions[0].blocks.rbegin()};
dispatch_block->begin = last_block->end + 1;
dispatch_block->end = last_block->end + 1;
functions[0].blocks.insert(*dispatch_block);
}
}
void CFG::AnalyzeLabel(FunctionId function_id, Label& label) {
if (InspectVisitedBlocks(function_id, label)) {
// Label address has been visited
return;
}
// Try to find the next block
Function* const function{&functions[function_id]};
Location pc{label.address};
const auto next_it{function->blocks.upper_bound(pc, Compare{})};
const bool is_last{next_it == function->blocks.end()};
Block* const next{is_last ? nullptr : &*next_it};
// Insert before the next block
Block* const block{label.block};
// Analyze instructions until it reaches an already visited block or there's a branch
bool is_branch{false};
while (!next || pc < next->begin) {
is_branch = AnalyzeInst(block, function_id, pc) == AnalysisState::Branch;
if (is_branch) {
break;
}
++pc;
}
if (!is_branch) {
// If the block finished without a branch,
// it means that the next instruction is already visited, jump to it
block->end = pc;
block->cond = IR::Condition{true};
block->branch_true = next;
block->branch_false = nullptr;
}
// Function's pointer might be invalid, resolve it again
// Insert the new block
functions[function_id].blocks.insert(*block);
}
bool CFG::InspectVisitedBlocks(FunctionId function_id, const Label& label) {
const Location pc{label.address};
Function& function{functions[function_id]};
const auto it{
std::ranges::find_if(function.blocks, [pc](auto& block) { return block.Contains(pc); })};
if (it == function.blocks.end()) {
// Address has not been visited
return false;
}
Block* const visited_block{&*it};
if (visited_block->begin == pc) {
throw LogicError("Dangling block");
}
Block* const new_block{label.block};
Split(visited_block, new_block, pc);
function.blocks.insert(it, *new_block);
return true;
}
CFG::AnalysisState CFG::AnalyzeInst(Block* block, FunctionId function_id, Location pc) {
const Instruction inst{env.ReadInstruction(pc.Offset())};
const Opcode opcode{Decode(inst.raw)};
switch (opcode) {
case Opcode::BRA:
case Opcode::JMP:
case Opcode::RET:
if (!AnalyzeBranch(block, function_id, pc, inst, opcode)) {
return AnalysisState::Continue;
}
switch (opcode) {
case Opcode::BRA:
case Opcode::JMP:
AnalyzeBRA(block, function_id, pc, inst, IsAbsoluteJump(opcode));
break;
case Opcode::RET:
block->end_class = EndClass::Return;
break;
default:
break;
}
block->end = pc;
return AnalysisState::Branch;
case Opcode::BRK:
case Opcode::CONT:
case Opcode::LONGJMP:
case Opcode::SYNC: {
if (!AnalyzeBranch(block, function_id, pc, inst, opcode)) {
return AnalysisState::Continue;
}
const auto [stack_pc, new_stack]{block->stack.Pop(OpcodeToken(opcode))};
block->branch_true = AddLabel(block, new_stack, stack_pc, function_id);
block->end = pc;
return AnalysisState::Branch;
}
case Opcode::KIL: {
const Predicate pred{inst.Pred()};
const auto ir_pred{static_cast<IR::Pred>(pred.index)};
const IR::Condition cond{inst.branch.flow_test, ir_pred, pred.negated};
AnalyzeCondInst(block, function_id, pc, EndClass::Kill, cond);
return AnalysisState::Branch;
}
case Opcode::PBK:
case Opcode::PCNT:
case Opcode::PEXIT:
case Opcode::PLONGJMP:
case Opcode::SSY:
block->stack.Push(OpcodeToken(opcode), BranchOffset(pc, inst));
return AnalysisState::Continue;
case Opcode::BRX:
case Opcode::JMX:
return AnalyzeBRX(block, pc, inst, IsAbsoluteJump(opcode), function_id);
case Opcode::EXIT:
return AnalyzeEXIT(block, function_id, pc, inst);
case Opcode::PRET:
throw NotImplementedException("PRET flow analysis");
case Opcode::CAL:
case Opcode::JCAL: {
const bool is_absolute{IsAbsoluteJump(opcode)};
const Location cal_pc{is_absolute ? inst.branch.Absolute() : BranchOffset(pc, inst)};
// Technically CAL pushes into PRET, but that's implicit in the function call for us
// Insert the function into the list if it doesn't exist
const auto it{std::ranges::find(functions, cal_pc, &Function::entrypoint)};
const bool exists{it != functions.end()};
const FunctionId call_id{exists ? static_cast<size_t>(std::distance(functions.begin(), it))
: functions.size()};
if (!exists) {
functions.emplace_back(block_pool, cal_pc);
}
block->end_class = EndClass::Call;
block->function_call = call_id;
block->return_block = AddLabel(block, block->stack, pc + 1, function_id);
block->end = pc;
return AnalysisState::Branch;
}
default:
break;
}
const Predicate pred{inst.Pred()};
if (pred == Predicate{true} || pred == Predicate{false}) {
return AnalysisState::Continue;
}
const IR::Condition cond{static_cast<IR::Pred>(pred.index), pred.negated};
AnalyzeCondInst(block, function_id, pc, EndClass::Branch, cond);
return AnalysisState::Branch;
}
void CFG::AnalyzeCondInst(Block* block, FunctionId function_id, Location pc,
EndClass insn_end_class, IR::Condition cond) {
if (block->begin != pc) {
// If the block doesn't start in the conditional instruction
// mark it as a label to visit it later
block->end = pc;
block->cond = IR::Condition{true};
block->branch_true = AddLabel(block, block->stack, pc, function_id);
block->branch_false = nullptr;
return;
}
// Create a virtual block and a conditional block
Block* const conditional_block{block_pool.Create()};
Block virtual_block{};
virtual_block.begin = block->begin.Virtual();
virtual_block.end = block->begin.Virtual();
virtual_block.end_class = EndClass::Branch;
virtual_block.stack = block->stack;
virtual_block.cond = cond;
virtual_block.branch_true = conditional_block;
virtual_block.branch_false = nullptr;
// Save the contents of the visited block in the conditional block
*conditional_block = std::move(*block);
// Impersonate the visited block with a virtual block
*block = std::move(virtual_block);
// Set the end properties of the conditional instruction
conditional_block->end = pc + 1;
conditional_block->end_class = insn_end_class;
// Add a label to the instruction after the conditional instruction
Block* const endif_block{AddLabel(conditional_block, block->stack, pc + 1, function_id)};
// Branch to the next instruction from the virtual block
block->branch_false = endif_block;
// And branch to it from the conditional instruction if it is a branch or a kill instruction
// Kill instructions are considered a branch because they demote to a helper invocation and
// execution may continue.
if (insn_end_class == EndClass::Branch || insn_end_class == EndClass::Kill) {
conditional_block->cond = IR::Condition{true};
conditional_block->branch_true = endif_block;
conditional_block->branch_false = nullptr;
}
// Finally insert the condition block into the list of blocks
functions[function_id].blocks.insert(*conditional_block);
}
bool CFG::AnalyzeBranch(Block* block, FunctionId function_id, Location pc, Instruction inst,
Opcode opcode) {
if (inst.branch.is_cbuf) {
throw NotImplementedException("Branch with constant buffer offset");
}
const Predicate pred{inst.Pred()};
if (pred == Predicate{false}) {
return false;
}
const bool has_flow_test{HasFlowTest(opcode)};
const IR::FlowTest flow_test{has_flow_test ? inst.branch.flow_test.Value() : IR::FlowTest::T};
if (pred != Predicate{true} || flow_test != IR::FlowTest::T) {
block->cond = IR::Condition(flow_test, static_cast<IR::Pred>(pred.index), pred.negated);
block->branch_false = AddLabel(block, block->stack, pc + 1, function_id);
} else {
block->cond = IR::Condition{true};
}
return true;
}
void CFG::AnalyzeBRA(Block* block, FunctionId function_id, Location pc, Instruction inst,
bool is_absolute) {
const Location bra_pc{is_absolute ? inst.branch.Absolute() : BranchOffset(pc, inst)};
block->branch_true = AddLabel(block, block->stack, bra_pc, function_id);
}
CFG::AnalysisState CFG::AnalyzeBRX(Block* block, Location pc, Instruction inst, bool is_absolute,
FunctionId function_id) {
const std::optional brx_table{TrackIndirectBranchTable(env, pc, program_start)};
if (!brx_table) {
TrackIndirectBranchTable(env, pc, program_start);
throw NotImplementedException("Failed to track indirect branch");
}
const IR::FlowTest flow_test{inst.branch.flow_test};
const Predicate pred{inst.Pred()};
if (flow_test != IR::FlowTest::T || pred != Predicate{true}) {
throw NotImplementedException("Conditional indirect branch");
}
std::vector<u32> targets;
targets.reserve(brx_table->num_entries);
for (u32 i = 0; i < brx_table->num_entries; ++i) {
u32 target{env.ReadCbufValue(brx_table->cbuf_index, brx_table->cbuf_offset + i * 4)};
if (!is_absolute) {
target += pc.Offset();
}
target += static_cast<u32>(brx_table->branch_offset);
target += 8;
targets.push_back(target);
}
std::ranges::sort(targets);
targets.erase(std::unique(targets.begin(), targets.end()), targets.end());
block->indirect_branches.reserve(targets.size());
for (const u32 target : targets) {
Block* const branch{AddLabel(block, block->stack, env.TranslateProgramAddress(target), function_id)};
block->indirect_branches.push_back({
.block = branch,
.address = target,
});
}
block->cond = IR::Condition{true};
block->end = pc + 1;
block->end_class = EndClass::IndirectBranch;
block->branch_reg = brx_table->branch_reg;
block->branch_offset = brx_table->branch_offset + 8;
if (!is_absolute) {
block->branch_offset += pc.Offset();
}
return AnalysisState::Branch;
}
CFG::AnalysisState CFG::AnalyzeEXIT(Block* block, FunctionId function_id, Location pc,
Instruction inst) {
const IR::FlowTest flow_test{inst.branch.flow_test};
const Predicate pred{inst.Pred()};
if (pred == Predicate{false} || flow_test == IR::FlowTest::F) {
// EXIT will never be taken
return AnalysisState::Continue;
}
if (exits_to_dispatcher && function_id != 0) {
throw NotImplementedException("Dispatch EXIT on external function");
}
if (pred != Predicate{true} || flow_test != IR::FlowTest::T) {
if (block->stack.Peek(Token::PEXIT).has_value()) {
throw NotImplementedException("Conditional EXIT with PEXIT token");
}
const IR::Condition cond{flow_test, static_cast<IR::Pred>(pred.index), pred.negated};
if (exits_to_dispatcher) {
block->end = pc;
block->end_class = EndClass::Branch;
block->cond = cond;
block->branch_true = dispatch_block;
block->branch_false = AddLabel(block, block->stack, pc + 1, function_id);
return AnalysisState::Branch;
}
AnalyzeCondInst(block, function_id, pc, EndClass::Exit, cond);
return AnalysisState::Branch;
}
if (const std::optional<Location> exit_pc{block->stack.Peek(Token::PEXIT)}) {
const Stack popped_stack{block->stack.Remove(Token::PEXIT)};
block->cond = IR::Condition{true};
block->branch_true = AddLabel(block, popped_stack, *exit_pc, function_id);
block->branch_false = nullptr;
return AnalysisState::Branch;
}
if (exits_to_dispatcher) {
block->cond = IR::Condition{true};
block->end = pc;
block->end_class = EndClass::Branch;
block->branch_true = dispatch_block;
block->branch_false = nullptr;
return AnalysisState::Branch;
}
block->end = pc + 1;
block->end_class = EndClass::Exit;
return AnalysisState::Branch;
}
Block* CFG::AddLabel(Block* block, Stack stack, Location pc, FunctionId function_id) {
Function& function{functions[function_id]};
if (block->begin == pc) {
// Jumps to itself
return block;
}
if (const auto it{function.blocks.find(pc, Compare{})}; it != function.blocks.end()) {
// Block already exists and it has been visited
if (function.blocks.begin() != it) {
// Check if the previous node is the virtual variant of the label
// This won't exist if a virtual node is not needed or it hasn't been visited
// If it hasn't been visited and a virtual node is needed, this will still behave as
// expected because the node impersonated with its virtual node.
const auto prev{std::prev(it)};
if (it->begin.Virtual() == prev->begin) {
return &*prev;
}
}
return &*it;
}
// Make sure we don't insert the same layer twice
const auto label_it{std::ranges::find(function.labels, pc, &Label::address)};
if (label_it != function.labels.end()) {
return label_it->block;
}
Block* const new_block{block_pool.Create()};
new_block->begin = pc;
new_block->end = pc;
new_block->end_class = EndClass::Branch;
new_block->cond = IR::Condition(true);
new_block->stack = stack;
new_block->branch_true = nullptr;
new_block->branch_false = nullptr;
function.labels.push_back(Label{
.address{pc},
.block = new_block,
.stack{std::move(stack)},
});
return new_block;
}
std::string CFG::Dot() const {
int node_uid{0};
std::string dot{"digraph shader {\n"};
for (const Function& function : functions) {
dot += fmt::format("\tsubgraph cluster_{} {{\n", function.entrypoint);
dot += fmt::format("\t\tnode [style=filled];\n");
for (const Block& block : function.blocks) {
const std::string name{NameOf(block)};
const auto add_branch = [&](Block* branch, bool add_label) {
dot += fmt::format("\t\t{}->{}", name, NameOf(*branch));
if (add_label && block.cond != IR::Condition{true} &&
block.cond != IR::Condition{false}) {
dot += fmt::format(" [label=\"{}\"]", block.cond);
}
dot += '\n';
};
dot += fmt::format("\t\t{};\n", name);
switch (block.end_class) {
case EndClass::Branch:
if (block.cond != IR::Condition{false}) {
add_branch(block.branch_true, true);
}
if (block.cond != IR::Condition{true}) {
add_branch(block.branch_false, false);
}
break;
case EndClass::IndirectBranch:
for (const IndirectBranch& branch : block.indirect_branches) {
add_branch(branch.block, false);
}
break;
case EndClass::Call:
dot += fmt::format("\t\t{}->N{};\n", name, node_uid);
dot += fmt::format("\t\tN{}->{};\n", node_uid, NameOf(*block.return_block));
dot += fmt::format("\t\tN{} [label=\"Call {}\"][shape=square][style=stripped];\n",
node_uid, block.function_call);
dot += '\n';
++node_uid;
break;
case EndClass::Exit:
dot += fmt::format("\t\t{}->N{};\n", name, node_uid);
dot += fmt::format("\t\tN{} [label=\"Exit\"][shape=square][style=stripped];\n",
node_uid);
++node_uid;
break;
case EndClass::Return:
dot += fmt::format("\t\t{}->N{};\n", name, node_uid);
dot += fmt::format("\t\tN{} [label=\"Return\"][shape=square][style=stripped];\n",
node_uid);
++node_uid;
break;
case EndClass::Kill:
dot += fmt::format("\t\t{}->N{};\n", name, node_uid);
dot += fmt::format("\t\tN{} [label=\"Kill\"][shape=square][style=stripped];\n",
node_uid);
++node_uid;
break;
}
}
if (function.entrypoint == 8) {
dot += fmt::format("\t\tlabel = \"main\";\n");
} else {
dot += fmt::format("\t\tlabel = \"Function {}\";\n", function.entrypoint);
}
dot += "\t}\n";
}
if (!functions.empty()) {
auto& function{functions.front()};
if (function.blocks.empty()) {
dot += "Start;\n";
} else {
dot += fmt::format("\tStart -> {};\n", NameOf(*function.blocks.begin()));
}
dot += fmt::format("\tStart [shape=diamond];\n");
}
dot += "}\n";
return dot;
}
} // namespace Shader::Maxwell::Flow

View File

@@ -0,0 +1,168 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <optional>
#include <span>
#include <string>
#include <vector>
#include <boost/container/small_vector.hpp>
#include <boost/intrusive/set.hpp>
#include "shader_recompiler/environment.h"
#include "shader_recompiler/frontend/ir/condition.h"
#include "shader_recompiler/frontend/ir/reg.h"
#include "shader_recompiler/frontend/maxwell/instruction.h"
#include "shader_recompiler/frontend/maxwell/location.h"
#include "shader_recompiler/frontend/maxwell/opcodes.h"
#include "shader_recompiler/object_pool.h"
namespace Shader::Maxwell::Flow {
struct Block;
using FunctionId = size_t;
enum class EndClass {
Branch,
IndirectBranch,
Call,
Exit,
Return,
Kill,
};
enum class Token {
SSY,
PBK,
PEXIT,
PRET,
PCNT,
PLONGJMP,
};
struct StackEntry {
auto operator<=>(const StackEntry&) const noexcept = default;
Token token;
Location target;
};
class Stack {
public:
void Push(Token token, Location target);
[[nodiscard]] std::pair<Location, Stack> Pop(Token token) const;
[[nodiscard]] std::optional<Location> Peek(Token token) const;
[[nodiscard]] Stack Remove(Token token) const;
private:
std::vector<StackEntry> entries;
};
struct IndirectBranch {
Block* block;
u32 address;
};
struct Block : boost::intrusive::set_base_hook<
// Normal link is ~2.5% faster compared to safe link
boost::intrusive::link_mode<boost::intrusive::normal_link>> {
[[nodiscard]] bool Contains(Location pc) const noexcept;
bool operator<(const Block& rhs) const noexcept {
return begin < rhs.begin;
}
Location begin;
Location end;
EndClass end_class{};
IR::Condition cond{};
Stack stack;
Block* branch_true{};
Block* branch_false{};
FunctionId function_call{};
Block* return_block{};
IR::Reg branch_reg{};
s32 branch_offset{};
std::vector<IndirectBranch> indirect_branches;
};
struct Label {
Location address;
Block* block;
Stack stack;
};
struct Function {
explicit Function(ObjectPool<Block>& block_pool, Location start_address);
Location entrypoint;
boost::container::small_vector<Label, 16> labels;
boost::intrusive::set<Block> blocks;
};
class CFG {
enum class AnalysisState {
Branch,
Continue,
};
public:
explicit CFG(Environment& env, ObjectPool<Block>& block_pool, Location start_address,
bool exits_to_dispatcher = false);
CFG& operator=(const CFG&) = delete;
CFG(const CFG&) = delete;
CFG& operator=(CFG&&) = delete;
CFG(CFG&&) = delete;
[[nodiscard]] std::string Dot() const;
[[nodiscard]] std::span<const Function> Functions() const noexcept {
return std::span(functions.data(), functions.size());
}
[[nodiscard]] std::span<Function> Functions() noexcept {
return std::span(functions.data(), functions.size());
}
[[nodiscard]] bool ExitsToDispatcher() const {
return exits_to_dispatcher;
}
private:
void AnalyzeLabel(FunctionId function_id, Label& label);
/// Inspect already visited blocks.
/// Return true when the block has already been visited
bool InspectVisitedBlocks(FunctionId function_id, const Label& label);
AnalysisState AnalyzeInst(Block* block, FunctionId function_id, Location pc);
void AnalyzeCondInst(Block* block, FunctionId function_id, Location pc, EndClass insn_end_class,
IR::Condition cond);
/// Return true when the branch instruction is confirmed to be a branch
bool AnalyzeBranch(Block* block, FunctionId function_id, Location pc, Instruction inst,
Opcode opcode);
void AnalyzeBRA(Block* block, FunctionId function_id, Location pc, Instruction inst,
bool is_absolute);
AnalysisState AnalyzeBRX(Block* block, Location pc, Instruction inst, bool is_absolute,
FunctionId function_id);
AnalysisState AnalyzeEXIT(Block* block, FunctionId function_id, Location pc, Instruction inst);
/// Return the branch target block id
Block* AddLabel(Block* block, Stack stack, Location pc, FunctionId function_id);
Environment& env;
ObjectPool<Block>& block_pool;
boost::container::small_vector<Function, 1> functions;
Location program_start;
bool exits_to_dispatcher{};
Block* dispatch_block{};
};
} // namespace Shader::Maxwell::Flow

View File

@@ -0,0 +1,53 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <array>
#include <bit>
#include <memory>
#include <ranges>
#include <string_view>
#include "common/assert.h"
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/maxwell/decode.h"
#include "shader_recompiler/frontend/maxwell/opcodes.h"
namespace Shader::Maxwell {
consteval std::pair<u64, u64> MaskValueFromEncoding(const char data[20]) noexcept {
u64 mask = 0, value = 0, bit = u64(1) << 63;
for (int i = 0; i < 20; ++i)
switch (data[i]) {
case '0':
mask |= bit;
bit >>= 1;
break;
case '1':
mask |= bit;
value |= bit;
bit >>= 1;
break;
case '-':
bit >>= 1;
break;
default:
break;
}
return { mask, value };
}
Opcode Decode(u64 insn) {
#define INST(name, cute, encode) \
if (auto const p = MaskValueFromEncoding(encode); (insn & p.first) == p.second) \
return Opcode::name;
#include "maxwell.inc"
#undef INST
ASSERT_MSG(false, "Invalid insn 0x{:016x}", insn);
return Opcode::NOP;
}
} // namespace Shader::Maxwell

View File

@@ -0,0 +1,13 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "common/common_types.h"
#include "shader_recompiler/frontend/maxwell/opcodes.h"
namespace Shader::Maxwell {
[[nodiscard]] Opcode Decode(u64 insn);
} // namespace Shader::Maxwell

View File

@@ -0,0 +1,107 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <optional>
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/maxwell/decode.h"
#include "shader_recompiler/frontend/maxwell/indirect_branch_table_track.h"
#include "shader_recompiler/frontend/maxwell/opcodes.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/load_constant.h"
namespace Shader::Maxwell {
namespace {
union Encoding {
u64 raw;
BitField<0, 8, IR::Reg> dest_reg;
BitField<8, 8, IR::Reg> src_reg;
BitField<20, 19, u64> immediate;
BitField<56, 1, u64> is_negative;
BitField<20, 24, s64> brx_offset;
};
template <typename Callable>
std::optional<u64> Track(Environment& env, Location block_begin, Location& pos, Callable&& func) {
while (pos >= block_begin) {
const u64 insn{env.ReadInstruction(pos.Offset())};
--pos;
if (func(insn, Decode(insn))) {
return insn;
}
}
return std::nullopt;
}
std::optional<u64> TrackLDC(Environment& env, Location block_begin, Location& pos,
IR::Reg brx_reg) {
return Track(env, block_begin, pos, [brx_reg](u64 insn, Opcode opcode) {
const LDC::Encoding ldc{insn};
return opcode == Opcode::LDC && ldc.dest_reg == brx_reg && ldc.size == LDC::Size::B32 &&
ldc.mode == LDC::Mode::Default;
});
}
std::optional<u64> TrackSHL(Environment& env, Location block_begin, Location& pos,
IR::Reg ldc_reg) {
return Track(env, block_begin, pos, [ldc_reg](u64 insn, Opcode opcode) {
const Encoding shl{insn};
return opcode == Opcode::SHL_imm && shl.dest_reg == ldc_reg;
});
}
std::optional<u64> TrackIMNMX(Environment& env, Location block_begin, Location& pos,
IR::Reg shl_reg) {
return Track(env, block_begin, pos, [shl_reg](u64 insn, Opcode opcode) {
const Encoding imnmx{insn};
return opcode == Opcode::IMNMX_imm && imnmx.dest_reg == shl_reg;
});
}
} // Anonymous namespace
std::optional<IndirectBranchTableInfo> TrackIndirectBranchTable(Environment& env, Location brx_pos,
Location block_begin) {
const u64 brx_insn{env.ReadInstruction(brx_pos.Offset())};
const Opcode brx_opcode{Decode(brx_insn)};
if (brx_opcode != Opcode::BRX && brx_opcode != Opcode::JMX) {
throw LogicError("Tracked instruction is not BRX or JMX");
}
const IR::Reg brx_reg{Encoding{brx_insn}.src_reg};
const s32 brx_offset{static_cast<s32>(Encoding{brx_insn}.brx_offset)};
Location pos{brx_pos};
const std::optional<u64> ldc_insn{TrackLDC(env, block_begin, pos, brx_reg)};
if (!ldc_insn) {
return std::nullopt;
}
const LDC::Encoding ldc{*ldc_insn};
const u32 cbuf_index{static_cast<u32>(ldc.index)};
const u32 cbuf_offset{static_cast<u32>(static_cast<s32>(ldc.offset.Value()))};
const IR::Reg ldc_reg{ldc.src_reg};
const std::optional<u64> shl_insn{TrackSHL(env, block_begin, pos, ldc_reg)};
if (!shl_insn) {
return std::nullopt;
}
const Encoding shl{*shl_insn};
const IR::Reg shl_reg{shl.src_reg};
const std::optional<u64> imnmx_insn{TrackIMNMX(env, block_begin, pos, shl_reg)};
if (!imnmx_insn) {
return std::nullopt;
}
const Encoding imnmx{*imnmx_insn};
if (imnmx.is_negative != 0) {
return std::nullopt;
}
const u32 imnmx_immediate{static_cast<u32>(imnmx.immediate.Value())};
return IndirectBranchTableInfo{
.cbuf_index = cbuf_index,
.cbuf_offset = cbuf_offset,
.num_entries = imnmx_immediate + 1,
.branch_offset = brx_offset,
.branch_reg = brx_reg,
};
}
} // namespace Shader::Maxwell

View File

@@ -0,0 +1,26 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <optional>
#include "common/common_types.h"
#include "shader_recompiler/environment.h"
#include "shader_recompiler/frontend/ir/reg.h"
#include "shader_recompiler/frontend/maxwell/location.h"
namespace Shader::Maxwell {
struct IndirectBranchTableInfo {
u32 cbuf_index{};
u32 cbuf_offset{};
u32 num_entries{};
s32 branch_offset{};
IR::Reg branch_reg{};
};
std::optional<IndirectBranchTableInfo> TrackIndirectBranchTable(Environment& env, Location brx_pos,
Location block_begin);
} // namespace Shader::Maxwell

View File

@@ -0,0 +1,61 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "common/bit_field.h"
#include "common/common_types.h"
#include "shader_recompiler/frontend/ir/flow_test.h"
namespace Shader::Maxwell {
struct Predicate {
Predicate() = default;
Predicate(unsigned index_, bool negated_ = false) : index{index_}, negated{negated_} {}
Predicate(bool value) : index{7}, negated{!value} {}
Predicate(u64 raw) : index{static_cast<unsigned>(raw & 7)}, negated{(raw & 8) != 0} {}
unsigned index;
bool negated;
};
inline bool operator==(const Predicate& lhs, const Predicate& rhs) noexcept {
return lhs.index == rhs.index && lhs.negated == rhs.negated;
}
inline bool operator!=(const Predicate& lhs, const Predicate& rhs) noexcept {
return !(lhs == rhs);
}
union Instruction {
Instruction(u64 raw_) : raw{raw_} {}
u64 raw;
union {
BitField<5, 1, u64> is_cbuf;
BitField<0, 5, IR::FlowTest> flow_test;
[[nodiscard]] u32 Absolute() const noexcept {
return static_cast<u32>(absolute);
}
[[nodiscard]] s32 Offset() const noexcept {
return static_cast<s32>(offset);
}
private:
BitField<20, 24, s64> offset;
BitField<20, 32, u64> absolute;
} branch;
[[nodiscard]] Predicate Pred() const noexcept {
return Predicate{pred};
}
private:
BitField<16, 4, u64> pred;
};
static_assert(std::is_trivially_copyable_v<Instruction>);
} // namespace Shader::Maxwell

View File

@@ -0,0 +1,108 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <fmt/ranges.h>
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
namespace Shader::Maxwell {
class Location {
static constexpr u32 VIRTUAL_BIAS{4};
public:
constexpr Location() = default;
constexpr Location(u32 initial_offset) : offset{initial_offset} {
if (initial_offset % 8 != 0) {
throw InvalidArgument("initial_offset={} is not a multiple of 8", initial_offset);
}
Align();
}
constexpr Location Virtual() const noexcept {
Location virtual_location;
virtual_location.offset = offset - VIRTUAL_BIAS;
return virtual_location;
}
[[nodiscard]] constexpr u32 Offset() const noexcept {
return offset;
}
[[nodiscard]] constexpr bool IsVirtual() const {
return offset % 8 == VIRTUAL_BIAS;
}
constexpr auto operator<=>(const Location&) const noexcept = default;
constexpr Location operator++() noexcept {
const Location copy{*this};
Step();
return copy;
}
constexpr Location operator++(int) noexcept {
Step();
return *this;
}
constexpr Location operator--() noexcept {
const Location copy{*this};
Back();
return copy;
}
constexpr Location operator--(int) noexcept {
Back();
return *this;
}
constexpr Location operator+(int number) const {
Location new_pc{*this};
while (number > 0) {
--number;
++new_pc;
}
while (number < 0) {
++number;
--new_pc;
}
return new_pc;
}
constexpr Location operator-(int number) const {
return operator+(-number);
}
private:
constexpr void Align() {
offset += offset % 32 == 0 ? 8 : 0;
}
constexpr void Step() {
offset += 8 + (offset % 32 == 24 ? 8 : 0);
}
constexpr void Back() {
offset -= 8 + (offset % 32 == 8 ? 8 : 0);
}
u32 offset{0xcccccccc};
};
} // namespace Shader::Maxwell
template <>
struct fmt::formatter<Shader::Maxwell::Location> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::Maxwell::Location& location, FormatContext& ctx) const {
return fmt::format_to(ctx.out(), "{:04x}", location.Offset());
}
};

View File

@@ -0,0 +1,283 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
INST(CCTLT, "CCTLT", "1110 1011 1111 0--0")
INST(AL2P, "AL2P", "1110 1111 1010 0---")
INST(ALD, "ALD", "1110 1111 1101 1---")
INST(AST, "AST", "1110 1111 1111 0---")
INST(B2R, "B2R", "1111 0000 1011 1---")
INST(BAR, "BAR", "1111 0000 1010 1---")
INST(BFE_cbuf, "BFE (cbuf)", "0100 1100 0000 0---")
INST(BFE_reg, "BFE (reg)", "0101 1100 0000 0---")
INST(BFI_cr, "BFI (cr)", "0100 1011 1111 0---")
INST(BFI_rc, "BFI (rc)", "0101 0011 1111 0---")
INST(BFI_reg, "BFI (reg)", "0101 1011 1111 0---")
INST(CS2R, "CS2R", "0101 0000 1100 1---")
INST(CSET, "CSET", "0101 0000 1001 1---")
INST(CSETP, "CSETP", "0101 0000 1010 0---")
INST(DADD_cbuf, "DADD (cbuf)", "0100 1100 0111 0---")
INST(DADD_reg, "DADD (reg)", "0101 1100 0111 0---")
INST(DEPBAR, "DEPBAR", "1111 0000 1111 0---")
INST(DMNMX_cbuf, "DMNMX (cbuf)", "0100 1100 0101 0---")
INST(DMNMX_reg, "DMNMX (reg)", "0101 1100 0101 0---")
INST(DMUL_cbuf, "DMUL (cbuf)", "0100 1100 1000 0---")
INST(DMUL_reg, "DMUL (reg)", "0101 1100 1000 0---")
INST(F2F_cbuf, "F2F (cbuf)", "0100 1100 1010 1---")
INST(F2F_reg, "F2F (reg)", "0101 1100 1010 1---")
INST(F2I_cbuf, "F2I (cbuf)", "0100 1100 1011 0---")
INST(F2I_reg, "F2I (reg)", "0101 1100 1011 0---")
INST(FADD_cbuf, "FADD (cbuf)", "0100 1100 0101 1---")
INST(FADD_reg, "FADD (reg)", "0101 1100 0101 1---")
INST(FCHK_cbuf, "FCHK (cbuf)", "0100 1100 1000 1---")
INST(FCHK_reg, "FCHK (reg)", "0101 1100 1000 1---")
INST(FLO_cbuf, "FLO (cbuf)", "0100 1100 0011 0---")
INST(FLO_reg, "FLO (reg)", "0101 1100 0011 0---")
INST(FMNMX_cbuf, "FMNMX (cbuf)", "0100 1100 0110 0---")
INST(FMNMX_reg, "FMNMX (reg)", "0101 1100 0110 0---")
INST(FMUL_cbuf, "FMUL (cbuf)", "0100 1100 0110 1---")
INST(FMUL_reg, "FMUL (reg)", "0101 1100 0110 1---")
INST(FSWZADD, "FSWZADD", "0101 0000 1111 1---")
INST(HADD2_reg, "HADD2 (reg)", "0101 1101 0001 0---")
INST(HFMA2_reg, "HFMA2 (reg)", "0101 1101 0000 0---")
INST(HMUL2_reg, "HMUL2 (reg)", "0101 1101 0000 1---")
INST(HSET2_reg, "HSET2 (reg)", "0101 1101 0001 1---")
INST(HSETP2_reg, "HSETP2 (reg)", "0101 1101 0010 0---")
INST(I2F_cbuf, "I2F (cbuf)", "0100 1100 1011 1---")
INST(I2F_reg, "I2F (reg)", "0101 1100 1011 1---")
INST(I2I_cbuf, "I2I (cbuf)", "0100 1100 1110 0---")
INST(I2I_reg, "I2I (reg)", "0101 1100 1110 0---")
INST(IADD_cbuf, "IADD (cbuf)", "0100 1100 0001 0---")
INST(IADD_reg, "IADD (reg)", "0101 1100 0001 0---")
INST(IDP_imm, "IDP (imm)", "0101 0011 1101 1---")
INST(IDP_reg, "IDP (reg)", "0101 0011 1111 1---")
INST(IMNMX_cbuf, "IMNMX (cbuf)", "0100 1100 0010 0---")
INST(IMNMX_reg, "IMNMX (reg)", "0101 1100 0010 0---")
INST(IMUL_cbuf, "IMUL (cbuf)", "0100 1100 0011 1---")
INST(IMUL_reg, "IMUL (reg)", "0101 1100 0011 1---")
INST(ISBERD, "ISBERD", "1110 1111 1101 0---")
INST(ISCADD_cbuf, "ISCADD (cbuf)", "0100 1100 0001 1---")
INST(ISCADD_reg, "ISCADD (reg)", "0101 1100 0001 1---")
INST(LDC, "LDC", "1110 1111 1001 0---")
INST(LDG, "LDG", "1110 1110 1101 0---")
INST(LDL, "LDL", "1110 1111 0100 0---")
INST(LDS, "LDS", "1110 1111 0100 1---")
INST(LEA_hi_reg, "LEA (hi reg)", "0101 1011 1101 1---")
INST(LEA_lo_reg, "LEA (lo reg)", "0101 1011 1101 0---")
INST(LEPC, "LEPC", "0101 0000 1101 0---")
INST(LOP3_reg, "LOP3 (reg)", "0101 1011 1110 0---")
INST(LOP_cbuf, "LOP (cbuf)", "0100 1100 0100 0---")
INST(LOP_reg, "LOP (reg)", "0101 1100 0100 0---")
INST(MEMBAR, "MEMBAR", "1110 1111 1001 1---")
INST(MOV_cbuf, "MOV (cbuf)", "0100 1100 1001 1---")
INST(MOV_reg, "MOV (reg)", "0101 1100 1001 1---")
INST(MUFU, "MUFU", "0101 0000 1000 0---")
INST(NOP, "NOP", "0101 0000 1011 0---")
INST(OUT_cbuf, "OUT (cbuf)", "1110 1011 1110 0---")
INST(OUT_reg, "OUT (reg)", "1111 1011 1110 0---")
INST(P2R_cbuf, "P2R (cbuf)", "0100 1100 1110 1---")
INST(P2R_imm, "P2R (imm)", "0011 1000 1110 1---")
INST(P2R_reg, "P2R (reg)", "0101 1100 1110 1---")
INST(PIXLD, "PIXLD", "1110 1111 1110 1---")
INST(POPC_cbuf, "POPC (cbuf)", "0100 1100 0000 1---")
INST(POPC_reg, "POPC (reg)", "0101 1100 0000 1---")
INST(PSET, "PSET", "0101 0000 1000 1---")
INST(PSETP, "PSETP", "0101 0000 1001 0---")
INST(R2B, "R2B", "1111 0000 1100 0---")
INST(R2P_cbuf, "R2P (cbuf)", "0100 1100 1111 0---")
INST(R2P_reg, "R2P (reg)", "0101 1100 1111 0---")
INST(RED, "RED", "1110 1011 1111 1---")
INST(RRO_cbuf, "RRO (cbuf)", "0100 1100 1001 0---")
INST(RRO_reg, "RRO (reg)", "0101 1100 1001 0---")
INST(S2R, "S2R", "1111 0000 1100 1---")
INST(SEL_cbuf, "SEL (cbuf)", "0100 1100 1010 0---")
INST(SEL_reg, "SEL (reg)", "0101 1100 1010 0---")
INST(SHFL, "SHFL", "1110 1111 0001 0---")
INST(SHF_l_reg, "SHF (l reg)", "0101 1011 1111 1---")
INST(SHF_r_reg, "SHF (r reg)", "0101 1100 1111 1---")
INST(SHL_cbuf, "SHL (cbuf)", "0100 1100 0100 1---")
INST(SHL_reg, "SHL (reg)", "0101 1100 0100 1---")
INST(SHR_cbuf, "SHR (cbuf)", "0100 1100 0010 1---")
INST(SHR_reg, "SHR (reg)", "0101 1100 0010 1---")
INST(STG, "STG", "1110 1110 1101 1---")
INST(STL, "STL", "1110 1111 0101 0---")
INST(STP, "STP", "1110 1110 1010 0---")
INST(STS, "STS", "1110 1111 0101 1---")
INST(SYNC, "SYNC", "1111 0000 1111 1---")
INST(TMML, "TMML", "1101 1111 0101 1---")
INST(TMML_b, "TMML (b)", "1101 1111 0110 0---")
INST(TXA, "TXA", "1101 1111 0100 0---")
INST(TXQ, "TXQ", "1101 1111 0100 1---")
INST(TXQ_b, "TXQ (b)", "1101 1111 0101 0---")
INST(VOTE, "VOTE", "0101 0000 1101 1---")
INST(VOTE_vtg, "VOTE (vtg)", "0101 0000 1110 0---")
INST(VSETP, "VSETP", "0101 0000 1111 0---")
INST(ATOM_cas, "ATOM (cas)", "1110 1110 1111 ----")
INST(BFE_imm, "BFE (imm)", "0011 100- 0000 0---")
INST(BFI_imm, "BFI (imm)", "0011 011- 1111 0---")
INST(BPT, "BPT", "1110 0011 1010 ----")
INST(BRA, "BRA", "1110 0010 0100 ----")
INST(BRK, "BRK", "1110 0011 0100 ----")
INST(BRX, "BRX", "1110 0010 0101 ----")
INST(CAL, "CAL", "1110 0010 0110 ----")
INST(CONT, "CONT", "1110 0011 0101 ----")
INST(DADD_imm, "DADD (imm)", "0011 100- 0111 0---")
INST(DFMA_cr, "DFMA (cr)", "0100 1011 0111 ----")
INST(DFMA_rc, "DFMA (rc)", "0101 0011 0111 ----")
INST(DFMA_reg, "DFMA (reg)", "0101 1011 0111 ----")
INST(DMNMX_imm, "DMNMX (imm)", "0011 100- 0101 0---")
INST(DMUL_imm, "DMUL (imm)", "0011 100- 1000 0---")
INST(DSETP_cbuf, "DSETP (cbuf)", "0100 1011 1000 ----")
INST(DSETP_reg, "DSETP (reg)", "0101 1011 1000 ----")
INST(EXIT, "EXIT", "1110 0011 0000 ----")
INST(F2F_imm, "F2F (imm)", "0011 100- 1010 1---")
INST(F2I_imm, "F2I (imm)", "0011 100- 1011 0---")
INST(FADD_imm, "FADD (imm)", "0011 100- 0101 1---")
INST(FCHK_imm, "FCHK (imm)", "0011 100- 1000 1---")
INST(FCMP_cr, "FCMP (cr)", "0100 1011 1010 ----")
INST(FCMP_rc, "FCMP (rc)", "0101 0011 1010 ----")
INST(FCMP_reg, "FCMP (reg)", "0101 1011 1010 ----")
INST(FLO_imm, "FLO (imm)", "0011 100- 0011 0---")
INST(FMNMX_imm, "FMNMX (imm)", "0011 100- 0110 0---")
INST(FMUL_imm, "FMUL (imm)", "0011 100- 0110 1---")
INST(FSETP_cbuf, "FSETP (cbuf)", "0100 1011 1011 ----")
INST(FSETP_reg, "FSETP (reg)", "0101 1011 1011 ----")
INST(GETCRSPTR, "GETCRSPTR", "1110 0010 1100 ----")
INST(GETLMEMBASE, "GETLMEMBASE", "1110 0010 1101 ----")
INST(I2F_imm, "I2F (imm)", "0011 100- 1011 1---")
INST(I2I_imm, "I2I (imm)", "0011 100- 1110 0---")
INST(IADD3_cbuf, "IADD3 (cbuf)", "0100 1100 1100 ----")
INST(IADD3_reg, "IADD3 (reg)", "0101 1100 1100 ----")
INST(IADD_imm, "IADD (imm)", "0011 100- 0001 0---")
INST(ICMP_cr, "ICMP (cr)", "0100 1011 0100 ----")
INST(ICMP_rc, "ICMP (rc)", "0101 0011 0100 ----")
INST(ICMP_reg, "ICMP (reg)", "0101 1011 0100 ----")
INST(IDE, "IDE", "1110 0011 1001 ----")
INST(IMNMX_imm, "IMNMX (imm)", "0011 100- 0010 0---")
INST(IMUL_imm, "IMUL (imm)", "0011 100- 0011 1---")
INST(ISCADD_imm, "ISCADD (imm)", "0011 100- 0001 1---")
INST(ISETP_cbuf, "ISETP (cbuf)", "0100 1011 0110 ----")
INST(ISETP_reg, "ISETP (reg)", "0101 1011 0110 ----")
INST(ISET_cbuf, "ISET (cbuf)", "0100 1011 0101 ----")
INST(ISET_reg, "ISET (reg)", "0101 1011 0101 ----")
INST(JCAL, "JCAL", "1110 0010 0010 ----")
INST(JMP, "JMP", "1110 0010 0001 ----")
INST(JMX, "JMX", "1110 0010 0000 ----")
INST(KIL, "KIL", "1110 0011 0011 ----")
INST(LEA_lo_cbuf, "LEA (lo cbuf)", "0100 1011 1101 ----")
INST(LEA_lo_imm, "LEA (lo imm)", "0011 011- 1101 0---")
INST(LONGJMP, "LONGJMP", "1110 0011 0001 ----")
INST(LOP_imm, "LOP (imm)", "0011 100- 0100 0---")
INST(MOV32I, "MOV32I", "0000 0001 0000 ----")
INST(MOV_imm, "MOV (imm)", "0011 100- 1001 1---")
INST(OUT_imm, "OUT (imm)", "1111 011- 1110 0---")
INST(PBK, "PBK", "1110 0010 1010 ----")
INST(PCNT, "PCNT", "1110 0010 1011 ----")
INST(PEXIT, "PEXIT", "1110 0010 0011 ----")
INST(PLONGJMP, "PLONGJMP", "1110 0010 1000 ----")
INST(POPC_imm, "POPC (imm)", "0011 100- 0000 1---")
INST(PRET, "PRET", "1110 0010 0111 ----")
INST(PRMT_cr, "PRMT (cr)", "0100 1011 1100 ----")
INST(PRMT_rc, "PRMT (rc)", "0101 0011 1100 ----")
INST(PRMT_reg, "PRMT (reg)", "0101 1011 1100 ----")
INST(R2P_imm, "R2P (imm)", "0011 100- 1111 0---")
INST(RAM, "RAM", "1110 0011 1000 ----")
INST(RET, "RET", "1110 0011 0010 ----")
INST(RRO_imm, "RRO (imm)", "0011 100- 1001 0---")
INST(RTT, "RTT", "1110 0011 0110 ----")
INST(SAM, "SAM", "1110 0011 0111 ----")
INST(SEL_imm, "SEL (imm)", "0011 100- 1010 0---")
INST(SETCRSPTR, "SETCRSPTR", "1110 0010 1110 ----")
INST(SETLMEMBASE, "SETLMEMBASE", "1110 0010 1111 ----")
INST(SHF_l_imm, "SHF (l imm)", "0011 011- 1111 1---")
INST(SHF_r_imm, "SHF (r imm)", "0011 100- 1111 1---")
INST(SHL_imm, "SHL (imm)", "0011 100- 0100 1---")
INST(SHR_imm, "SHR (imm)", "0011 100- 0010 1---")
INST(SSY, "SSY", "1110 0010 1001 ----")
INST(CCTL, "CCTL", "1110 1111 011- ----")
INST(CCTLL, "CCTLL", "1110 1111 100- ----")
INST(DFMA_imm, "DFMA (imm)", "0011 011- 0111 ----")
INST(DSETP_imm, "DSETP (imm)", "0011 011- 1000 ----")
INST(FCMP_imm, "FCMP (imm)", "0011 011- 1010 ----")
INST(FSETP_imm, "FSETP (imm)", "0011 011- 1011 ----")
INST(IADD3_imm, "IADD3 (imm)", "0011 100- 1100 ----")
INST(ICMP_imm, "ICMP (imm)", "0011 011- 0100 ----")
INST(ISETP_imm, "ISETP (imm)", "0011 011- 0110 ----")
INST(ISET_imm, "ISET (imm)", "0011 011- 0101 ----")
INST(PRMT_imm, "PRMT (imm)", "0011 011- 1100 ----")
INST(SULD, "SULD", "1110 1011 000- ----")
INST(SURED, "SURED", "1110 1011 010- ----")
INST(SUST, "SUST", "1110 1011 001- ----")
INST(TEX_b, "TEX (b)", "1101 1110 10-- ----")
INST(TLD4_b, "TLD4 (b)", "1101 1110 11-- ----")
INST(TXD, "TXD", "1101 1110 00-- ----")
INST(TXD_b, "TXD (b)", "1101 1110 01-- ----")
INST(XMAD_reg, "XMAD (reg)", "0101 1011 00-- ----")
INST(DSET_cbuf, "DSET (cbuf)", "0100 1001 0--- ----")
INST(DSET_reg, "DSET (reg)", "0101 1001 0--- ----")
INST(FFMA_cr, "FFMA (cr)", "0100 1001 1--- ----")
INST(FFMA_rc, "FFMA (rc)", "0101 0001 1--- ----")
INST(FFMA_reg, "FFMA (reg)", "0101 1001 1--- ----")
INST(IMADSP_cr, "IMADSP (cr)", "0100 1010 1--- ----")
INST(IMADSP_rc, "IMADSP (rc)", "0101 0010 1--- ----")
INST(IMADSP_reg, "IMADSP (reg)", "0101 1010 1--- ----")
INST(IMAD_cr, "IMAD (cr)", "0100 1010 0--- ----")
INST(IMAD_rc, "IMAD (rc)", "0101 0010 0--- ----")
INST(IMAD_reg, "IMAD (reg)", "0101 1010 0--- ----")
INST(SUATOM, "SUATOM", "1110 1010 0--- ----")
INST(SUATOM_cas, "SUATOM_cas", "1110 1010 1--- ----")
INST(TLD4S, "TLD4S", "1101 1111 -0-- ----")
INST(VABSDIFF4, "VABSDIFF4", "0101 0000 0--- ----")
INST(XMAD_imm, "XMAD (imm)", "0011 011- 00-- ----")
INST(XMAD_rc, "XMAD (rc)", "0101 0001 0--- ----")
INST(ATOM, "ATOM", "1110 1101 ---- ----")
INST(ATOMS, "ATOMS", "1110 1100 ---- ----")
INST(ATOMS_cas, "ATOMS (cas)", "1110 1110 ---- ----")
INST(DSET_imm, "DSET (imm)", "0011 001- 0--- ----")
INST(FFMA_imm, "FFMA (imm)", "0011 001- 1--- ----")
INST(FMUL32I, "FMUL32I", "0001 1110 ---- ----")
INST(FSET_cbuf, "FSET (cbuf)", "0100 1000 ---- ----")
INST(FSET_reg, "FSET (reg)", "0101 1000 ---- ----")
INST(HADD2_cbuf, "HADD2 (cbuf)", "0111 101- 1--- ----")
INST(HADD2_imm, "HADD2 (imm)", "0111 101- 0--- ----")
INST(HMUL2_cbuf, "HMUL2 (cbuf)", "0111 100- 1--- ----")
INST(HMUL2_imm, "HMUL2 (imm)", "0111 100- 0--- ----")
INST(HSET2_cbuf, "HSET2 (cbuf)", "0111 110- 1--- ----")
INST(HSET2_imm, "HSET2 (imm)", "0111 110- 0--- ----")
INST(HSETP2_cbuf, "HSETP2 (cbuf)", "0111 111- 1--- ----")
INST(HSETP2_imm, "HSETP2 (imm)", "0111 111- 0--- ----")
INST(IMADSP_imm, "IMADSP (imm)", "0011 010- 1--- ----")
INST(IMAD_imm, "IMAD (imm)", "0011 010- 0--- ----")
INST(IMUL32I, "IMUL32I", "0001 1111 ---- ----")
INST(IPA, "IPA", "1110 0000 ---- ----")
INST(TLD, "TLD", "1101 1100 ---- ----")
INST(TLD_b, "TLD (b)", "1101 1101 ---- ----")
INST(VABSDIFF, "VABSDIFF", "0101 0100 ---- ----")
INST(VMAD, "VMAD", "0101 1111 ---- ----")
INST(VSHL, "VSHL", "0101 0111 ---- ----")
INST(VSHR, "VSHR", "0101 0110 ---- ----")
INST(FSET_imm, "FSET (imm)", "0011 000- ---- ----")
INST(HADD2_32I, "HADD2_32I", "0010 110- ---- ----")
INST(HFMA2_32I, "HFMA2_32I", "0010 100- ---- ----")
INST(HMUL2_32I, "HMUL2_32I", "0010 101- ---- ----")
INST(IADD32I, "IADD32I", "0001 110- ---- ----")
INST(LOP3_cbuf, "LOP3 (cbuf)", "0000 001- ---- ----")
INST(VMNMX, "VMNMX", "0011 101- ---- ----")
INST(VSET, "VSET", "0100 000- ---- ----")
INST(XMAD_cr, "XMAD (cr)", "0100 111- ---- ----")
INST(FADD32I, "FADD32I", "0000 10-- ---- ----")
INST(FFMA32I, "FFMA32I", "0000 11-- ---- ----")
INST(HFMA2_cr, "HFMA2 (cr)", "0111 0--- 1--- ----")
INST(HFMA2_imm, "HFMA2 (imm)", "0111 0--- 0--- ----")
INST(HFMA2_rc, "HFMA2 (rc)", "0110 0--- 1--- ----")
INST(IMAD32I, "IMAD32I", "1000 00-- ---- ----")
INST(ISCADD32I, "ISCADD32I", "0001 01-- ---- ----")
INST(LEA_hi_cbuf, "LEA (hi cbuf)", "0001 10-- ---- ----")
INST(LOP32I, "LOP32I", "0000 01-- ---- ----")
INST(LOP3_imm, "LOP3 (imm)", "0011 11-- ---- ----")
INST(TEXS, "TEXS", "1101 -00- ---- ----")
INST(TLD4, "TLD4", "1100 10-- ---- ----")
INST(TLDS, "TLDS", "1101 -01- ---- ----")
INST(VADD, "VADD", "0010 00-- ---- ----")
INST(TEX, "TEX", "1100 0--- ---- ----")
INST(LD, "LD", "100- ---- ---- ----")
INST(ST, "ST", "101- ---- ---- ----")

View File

@@ -0,0 +1,25 @@
// SPDX-FileCopyrightText: Copyright 2025 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <array>
#include <string_view>
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/maxwell/opcodes.h"
namespace Shader::Maxwell {
const char* NameOf(Opcode opcode) {
constexpr const char* NAME_TABLE[] = {
#define INST(name, cute, encode) cute,
#include "maxwell.inc"
#undef INST
};
if (size_t(opcode) >= sizeof(NAME_TABLE) / sizeof(NAME_TABLE[0]))
throw InvalidArgument("Invalid opcode with raw value {}", int(opcode));
return NAME_TABLE[size_t(opcode)];
}
} // namespace Shader::Maxwell

View File

@@ -0,0 +1,29 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <fmt/ranges.h>
namespace Shader::Maxwell {
enum class Opcode {
#define INST(name, cute, encode) name,
#include "maxwell.inc"
#undef INST
};
const char* NameOf(Opcode opcode);
} // namespace Shader::Maxwell
template <>
struct fmt::formatter<Shader::Maxwell::Opcode> {
constexpr auto parse(format_parse_context& ctx) {
return ctx.begin();
}
template <typename FormatContext>
auto format(const Shader::Maxwell::Opcode& opcode, FormatContext& ctx) const {
return fmt::format_to(ctx.out(), "{}", NameOf(opcode));
}
};

View File

@@ -0,0 +1,998 @@
// SPDX-FileCopyrightText: Copyright 2026 suyu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <memory>
#include <string>
#include <ankerl/unordered_dense.h>
#include <utility>
#include <vector>
#include <fmt/ranges.h>
#include <boost/intrusive/list.hpp>
#include <ranges>
#include "shader_recompiler/environment.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/ir_emitter.h"
#include "shader_recompiler/frontend/maxwell/structured_control_flow.h"
#include "shader_recompiler/frontend/maxwell/translate/translate.h"
#include "shader_recompiler/host_translate_info.h"
#include "shader_recompiler/object_pool.h"
namespace Shader::Maxwell {
namespace {
struct Statement;
// Use normal_link because we are not guaranteed to destroy the tree in order
using ListBaseHook =
boost::intrusive::list_base_hook<boost::intrusive::link_mode<boost::intrusive::normal_link>>;
using Tree = boost::intrusive::list<Statement,
// Allow using Statement without a definition
boost::intrusive::base_hook<ListBaseHook>,
// Avoid linear complexity on splice, size is never called
boost::intrusive::constant_time_size<false>>;
using Node = Tree::iterator;
enum class StatementType {
Code,
Goto,
Label,
If,
Loop,
Break,
Return,
Kill,
Unreachable,
Function,
Identity,
Not,
Or,
SetVariable,
SetIndirectBranchVariable,
Variable,
IndirectBranchCond,
};
bool HasChildren(StatementType type) {
switch (type) {
case StatementType::If:
case StatementType::Loop:
case StatementType::Function:
return true;
default:
return false;
}
}
struct Goto {};
struct Label {};
struct If {};
struct Loop {};
struct Break {};
struct Return {};
struct Kill {};
struct Unreachable {};
struct FunctionTag {};
struct Identity {};
struct Not {};
struct Or {};
struct SetVariable {};
struct SetIndirectBranchVariable {};
struct Variable {};
struct IndirectBranchCond {};
#ifdef _MSC_VER
#pragma warning(push)
#pragma warning(disable : 26495) // Always initialize a member variable, expected in Statement
#endif
struct Statement : ListBaseHook {
Statement(const Flow::Block* block_, Statement* up_)
: block{block_}, up{up_}, type{StatementType::Code} {}
Statement(Goto, Statement* cond_, Node label_, Statement* up_)
: label{label_}, cond{cond_}, up{up_}, type{StatementType::Goto} {}
Statement(Label, u32 id_, Statement* up_) : id{id_}, up{up_}, type{StatementType::Label} {}
Statement(If, Statement* cond_, Tree&& children_, Statement* up_)
: children{std::move(children_)}, cond{cond_}, up{up_}, type{StatementType::If} {}
Statement(Loop, Statement* cond_, Tree&& children_, Statement* up_)
: children{std::move(children_)}, cond{cond_}, up{up_}, type{StatementType::Loop} {}
Statement(Break, Statement* cond_, Statement* up_)
: cond{cond_}, up{up_}, type{StatementType::Break} {}
Statement(Return, Statement* up_) : up{up_}, type{StatementType::Return} {}
Statement(Kill, Statement* up_) : up{up_}, type{StatementType::Kill} {}
Statement(Unreachable, Statement* up_) : up{up_}, type{StatementType::Unreachable} {}
Statement(FunctionTag) : children{}, type{StatementType::Function} {}
Statement(Identity, IR::Condition cond_, Statement* up_)
: guest_cond{cond_}, up{up_}, type{StatementType::Identity} {}
Statement(Not, Statement* op_, Statement* up_) : op{op_}, up{up_}, type{StatementType::Not} {}
Statement(Or, Statement* op_a_, Statement* op_b_, Statement* up_)
: op_a{op_a_}, op_b{op_b_}, up{up_}, type{StatementType::Or} {}
Statement(SetVariable, u32 id_, Statement* op_, Statement* up_)
: op{op_}, id{id_}, up{up_}, type{StatementType::SetVariable} {}
Statement(SetIndirectBranchVariable, IR::Reg branch_reg_, s32 branch_offset_, Statement* up_)
: branch_offset{branch_offset_},
branch_reg{branch_reg_}, up{up_}, type{StatementType::SetIndirectBranchVariable} {}
Statement(Variable, u32 id_, Statement* up_)
: id{id_}, up{up_}, type{StatementType::Variable} {}
Statement(IndirectBranchCond, u32 location_, Statement* up_)
: location{location_}, up{up_}, type{StatementType::IndirectBranchCond} {}
~Statement() {
if (HasChildren(type)) {
std::destroy_at(&children);
}
}
union {
const Flow::Block* block;
Node label;
Tree children;
IR::Condition guest_cond;
Statement* op;
Statement* op_a;
u32 location;
s32 branch_offset;
};
union {
Statement* cond;
Statement* op_b;
u32 id;
IR::Reg branch_reg;
};
Statement* up{};
StatementType type;
};
#ifdef _MSC_VER
#pragma warning(pop)
#endif
std::string DumpExpr(const Statement* stmt) {
switch (stmt->type) {
case StatementType::Identity:
return fmt::format("{}", stmt->guest_cond);
case StatementType::Not:
return fmt::format("!{}", DumpExpr(stmt->op));
case StatementType::Or:
return fmt::format("{} || {}", DumpExpr(stmt->op_a), DumpExpr(stmt->op_b));
case StatementType::Variable:
return fmt::format("goto_L{}", stmt->id);
case StatementType::IndirectBranchCond:
return fmt::format("(indirect_branch == {:x})", stmt->location);
default:
return "<invalid type>";
}
}
[[maybe_unused]] std::string DumpTree(const Tree& tree, u32 indentation = 0) {
std::string ret;
std::string indent(indentation, ' ');
for (auto stmt = tree.begin(); stmt != tree.end(); ++stmt) {
switch (stmt->type) {
case StatementType::Code:
ret += fmt::format("{} Block {:04x} -> {:04x} (0x{:016x});\n", indent,
stmt->block->begin.Offset(), stmt->block->end.Offset(),
reinterpret_cast<uintptr_t>(stmt->block));
break;
case StatementType::Goto:
ret += fmt::format("{} if ({}) goto L{};\n", indent, DumpExpr(stmt->cond),
stmt->label->id);
break;
case StatementType::Label:
ret += fmt::format("{}L{}:\n", indent, stmt->id);
break;
case StatementType::If:
ret += fmt::format("{} if ({}) {{\n", indent, DumpExpr(stmt->cond));
ret += DumpTree(stmt->children, indentation + 4);
ret += fmt::format("{} }}\n", indent);
break;
case StatementType::Loop:
ret += fmt::format("{} do {{\n", indent);
ret += DumpTree(stmt->children, indentation + 4);
ret += fmt::format("{} }} while ({});\n", indent, DumpExpr(stmt->cond));
break;
case StatementType::Break:
ret += fmt::format("{} if ({}) break;\n", indent, DumpExpr(stmt->cond));
break;
case StatementType::Return:
ret += fmt::format("{} return;\n", indent);
break;
case StatementType::Kill:
ret += fmt::format("{} kill;\n", indent);
break;
case StatementType::Unreachable:
ret += fmt::format("{} unreachable;\n", indent);
break;
case StatementType::SetVariable:
ret += fmt::format("{} goto_L{} = {};\n", indent, stmt->id, DumpExpr(stmt->op));
break;
case StatementType::SetIndirectBranchVariable:
ret += fmt::format("{} indirect_branch = {} + {};\n", indent, stmt->branch_reg,
stmt->branch_offset);
break;
case StatementType::Function:
case StatementType::Identity:
case StatementType::Not:
case StatementType::Or:
case StatementType::Variable:
case StatementType::IndirectBranchCond:
throw LogicError("Statement can't be printed");
}
}
return ret;
}
void SanitizeNoBreaks(const Tree& tree) {
if (std::find_if(tree.begin(), tree.end(), [](const Statement& s) { return s.type == StatementType::Break; }) != tree.end()) {
throw NotImplementedException("Capturing statement with break nodes");
}
}
size_t Level(Node stmt) {
size_t level{0};
Statement* node{stmt->up};
while (node) {
++level;
node = node->up;
}
return level;
}
bool IsDirectlyRelated(Node goto_stmt, Node label_stmt) {
const size_t goto_level{Level(goto_stmt)};
const size_t label_level{Level(label_stmt)};
size_t min_level;
size_t max_level;
Node min;
Node max;
if (label_level < goto_level) {
min_level = label_level;
max_level = goto_level;
min = label_stmt;
max = goto_stmt;
} else { // goto_level < label_level
min_level = goto_level;
max_level = label_level;
min = goto_stmt;
max = label_stmt;
}
while (max_level > min_level) {
--max_level;
max = max->up;
}
return min->up == max->up;
}
bool IsIndirectlyRelated(Node goto_stmt, Node label_stmt) {
return goto_stmt->up != label_stmt->up && !IsDirectlyRelated(goto_stmt, label_stmt);
}
[[maybe_unused]] bool AreSiblings(Node goto_stmt, Node label_stmt) noexcept {
Node it{goto_stmt};
do {
if (it == label_stmt) {
return true;
}
--it;
} while (it != goto_stmt->up->children.begin());
while (it != goto_stmt->up->children.end()) {
if (it == label_stmt) {
return true;
}
++it;
}
return false;
}
Node SiblingFromNephew(Node uncle, Node nephew) noexcept {
Statement* const parent{uncle->up};
Statement* it{&*nephew};
while (it->up != parent) {
it = it->up;
}
return Tree::s_iterator_to(*it);
}
bool AreOrdered(Node left_sibling, Node right_sibling) noexcept {
const Node end{right_sibling->up->children.end()};
for (auto it = right_sibling; it != end; ++it) {
if (it == left_sibling) {
return false;
}
}
return true;
}
bool NeedsLift(Node goto_stmt, Node label_stmt) noexcept {
const Node sibling{SiblingFromNephew(goto_stmt, label_stmt)};
return AreOrdered(sibling, goto_stmt);
}
class GotoPass {
public:
explicit GotoPass(Flow::CFG& cfg, ObjectPool<Statement>& stmt_pool) : pool{stmt_pool} {
std::vector gotos{BuildTree(cfg)};
const auto end{gotos.rend()};
for (auto goto_stmt = gotos.rbegin(); goto_stmt != end; ++goto_stmt) {
RemoveGoto(*goto_stmt);
}
}
Statement& RootStatement() noexcept {
return root_stmt;
}
private:
void RemoveGoto(Node goto_stmt) {
// Force goto_stmt and label_stmt to be directly related
const Node label_stmt{goto_stmt->label};
if (IsIndirectlyRelated(goto_stmt, label_stmt)) {
// Move goto_stmt out using outward-movement transformation until it becomes
// directly related to label_stmt
while (!IsDirectlyRelated(goto_stmt, label_stmt)) {
goto_stmt = MoveOutward(goto_stmt);
}
}
// Force goto_stmt and label_stmt to be siblings
if (IsDirectlyRelated(goto_stmt, label_stmt)) {
const size_t label_level{Level(label_stmt)};
size_t goto_level{Level(goto_stmt)};
if (goto_level > label_level) {
// Move goto_stmt out of its level using outward-movement transformations
while (goto_level > label_level) {
goto_stmt = MoveOutward(goto_stmt);
--goto_level;
}
} else { // Level(goto_stmt) < Level(label_stmt)
if (NeedsLift(goto_stmt, label_stmt)) {
// Lift goto_stmt to above stmt containing label_stmt using goto-lifting
// transformations
goto_stmt = Lift(goto_stmt);
}
// Move goto_stmt into label_stmt's level using inward-movement transformation
while (goto_level < label_level) {
goto_stmt = MoveInward(goto_stmt);
++goto_level;
}
}
}
// Expensive operation:
// if (!AreSiblings(goto_stmt, label_stmt)) {
// throw LogicError("Goto is not a sibling with the label");
// }
// goto_stmt and label_stmt are guaranteed to be siblings, eliminate
if (std::next(goto_stmt) == label_stmt) {
// Simply eliminate the goto if the label is next to it
goto_stmt->up->children.erase(goto_stmt);
} else if (AreOrdered(goto_stmt, label_stmt)) {
// Eliminate goto_stmt with a conditional
EliminateAsConditional(goto_stmt, label_stmt);
} else {
// Eliminate goto_stmt with a loop
EliminateAsLoop(goto_stmt, label_stmt);
}
}
std::vector<Node> BuildTree(Flow::CFG& cfg) {
u32 label_id{0};
std::vector<Node> gotos;
Flow::Function& first_function{cfg.Functions().front()};
BuildTree(cfg, first_function, label_id, gotos, root_stmt.children.end(), std::nullopt);
return gotos;
}
void BuildTree(Flow::CFG& cfg, Flow::Function& function, u32& label_id,
std::vector<Node>& gotos, Node function_insert_point,
std::optional<Node> return_label) {
Statement* const false_stmt{pool.Create(Identity{}, IR::Condition{false}, &root_stmt)};
Tree& root{root_stmt.children};
ankerl::unordered_dense::map<Flow::Block*, Node> local_labels;
local_labels.reserve(function.blocks.size());
for (Flow::Block& block : function.blocks) {
Statement* const label{pool.Create(Label{}, label_id, &root_stmt)};
const Node label_it{root.insert(function_insert_point, *label)};
local_labels.emplace(&block, label_it);
++label_id;
}
for (Flow::Block& block : function.blocks) {
const Node label{local_labels.at(&block)};
// Insertion point
const Node ip{std::next(label)};
// Reset goto variables before the first block and after its respective label
const auto make_reset_variable{[&]() -> Statement& {
return *pool.Create(SetVariable{}, label->id, false_stmt, &root_stmt);
}};
root.push_front(make_reset_variable());
root.insert(ip, make_reset_variable());
root.insert(ip, *pool.Create(&block, &root_stmt));
switch (block.end_class) {
case Flow::EndClass::Branch: {
Statement* const always_cond{
pool.Create(Identity{}, IR::Condition{true}, &root_stmt)};
if (block.cond == IR::Condition{true}) {
const Node true_label{local_labels.at(block.branch_true)};
gotos.push_back(
root.insert(ip, *pool.Create(Goto{}, always_cond, true_label, &root_stmt)));
} else if (block.cond == IR::Condition{false}) {
const Node false_label{local_labels.at(block.branch_false)};
gotos.push_back(root.insert(
ip, *pool.Create(Goto{}, always_cond, false_label, &root_stmt)));
} else {
const Node true_label{local_labels.at(block.branch_true)};
const Node false_label{local_labels.at(block.branch_false)};
Statement* const true_cond{pool.Create(Identity{}, block.cond, &root_stmt)};
gotos.push_back(
root.insert(ip, *pool.Create(Goto{}, true_cond, true_label, &root_stmt)));
gotos.push_back(root.insert(
ip, *pool.Create(Goto{}, always_cond, false_label, &root_stmt)));
}
break;
}
case Flow::EndClass::IndirectBranch:
root.insert(ip, *pool.Create(SetIndirectBranchVariable{}, block.branch_reg,
block.branch_offset, &root_stmt));
for (const Flow::IndirectBranch& indirect : block.indirect_branches) {
const Node indirect_label{local_labels.at(indirect.block)};
Statement* cond{
pool.Create(IndirectBranchCond{}, indirect.address, &root_stmt)};
Statement* goto_stmt{pool.Create(Goto{}, cond, indirect_label, &root_stmt)};
gotos.push_back(root.insert(ip, *goto_stmt));
}
root.insert(ip, *pool.Create(Unreachable{}, &root_stmt));
break;
case Flow::EndClass::Call: {
Flow::Function& call{cfg.Functions()[block.function_call]};
const Node call_return_label{local_labels.at(block.return_block)};
BuildTree(cfg, call, label_id, gotos, ip, call_return_label);
break;
}
case Flow::EndClass::Exit:
root.insert(ip, *pool.Create(Return{}, &root_stmt));
break;
case Flow::EndClass::Return: {
Statement* const always_cond{pool.Create(Identity{}, block.cond, &root_stmt)};
auto goto_stmt{pool.Create(Goto{}, always_cond, return_label.value(), &root_stmt)};
gotos.push_back(root.insert(ip, *goto_stmt));
break;
}
case Flow::EndClass::Kill:
root.insert(ip, *pool.Create(Kill{}, &root_stmt));
break;
}
}
}
void UpdateTreeUp(Statement* tree) {
for (Statement& stmt : tree->children) {
stmt.up = tree;
}
}
void EliminateAsConditional(Node goto_stmt, Node label_stmt) {
Tree& body{goto_stmt->up->children};
Tree if_body;
if_body.splice(if_body.begin(), body, std::next(goto_stmt), label_stmt);
Statement* const cond{pool.Create(Not{}, goto_stmt->cond, &root_stmt)};
Statement* const if_stmt{pool.Create(If{}, cond, std::move(if_body), goto_stmt->up)};
UpdateTreeUp(if_stmt);
body.insert(goto_stmt, *if_stmt);
body.erase(goto_stmt);
}
void EliminateAsLoop(Node goto_stmt, Node label_stmt) {
Tree& body{goto_stmt->up->children};
Tree loop_body;
loop_body.splice(loop_body.begin(), body, label_stmt, goto_stmt);
Statement* const cond{goto_stmt->cond};
Statement* const loop{pool.Create(Loop{}, cond, std::move(loop_body), goto_stmt->up)};
UpdateTreeUp(loop);
body.insert(goto_stmt, *loop);
body.erase(goto_stmt);
}
[[nodiscard]] Node MoveOutward(Node goto_stmt) {
switch (goto_stmt->up->type) {
case StatementType::If:
return MoveOutwardIf(goto_stmt);
case StatementType::Loop:
return MoveOutwardLoop(goto_stmt);
default:
throw LogicError("Invalid outward movement");
}
}
[[nodiscard]] Node MoveInward(Node goto_stmt) {
Statement* const parent{goto_stmt->up};
Tree& body{parent->children};
const Node label{goto_stmt->label};
const Node label_nested_stmt{SiblingFromNephew(goto_stmt, label)};
const u32 label_id{label->id};
Statement* const goto_cond{goto_stmt->cond};
Statement* const set_var{pool.Create(SetVariable{}, label_id, goto_cond, parent)};
body.insert(goto_stmt, *set_var);
Tree if_body;
if_body.splice(if_body.begin(), body, std::next(goto_stmt), label_nested_stmt);
Statement* const variable{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const neg_var{pool.Create(Not{}, variable, &root_stmt)};
if (!if_body.empty()) {
Statement* const if_stmt{pool.Create(If{}, neg_var, std::move(if_body), parent)};
UpdateTreeUp(if_stmt);
body.insert(goto_stmt, *if_stmt);
}
body.erase(goto_stmt);
switch (label_nested_stmt->type) {
case StatementType::If:
// Update nested if condition
label_nested_stmt->cond =
pool.Create(Or{}, variable, label_nested_stmt->cond, &root_stmt);
break;
case StatementType::Loop:
break;
default:
throw LogicError("Invalid inward movement");
}
Tree& nested_tree{label_nested_stmt->children};
Statement* const new_goto{pool.Create(Goto{}, variable, label, &*label_nested_stmt)};
return nested_tree.insert(nested_tree.begin(), *new_goto);
}
[[nodiscard]] Node Lift(Node goto_stmt) {
Statement* const parent{goto_stmt->up};
Tree& body{parent->children};
const Node label{goto_stmt->label};
const u32 label_id{label->id};
const Node label_nested_stmt{SiblingFromNephew(goto_stmt, label)};
Tree loop_body;
loop_body.splice(loop_body.begin(), body, label_nested_stmt, goto_stmt);
SanitizeNoBreaks(loop_body);
Statement* const variable{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const loop_stmt{pool.Create(Loop{}, variable, std::move(loop_body), parent)};
UpdateTreeUp(loop_stmt);
body.insert(goto_stmt, *loop_stmt);
Statement* const new_goto{pool.Create(Goto{}, variable, label, loop_stmt)};
loop_stmt->children.push_front(*new_goto);
const Node new_goto_node{loop_stmt->children.begin()};
Statement* const set_var{pool.Create(SetVariable{}, label_id, goto_stmt->cond, loop_stmt)};
loop_stmt->children.push_back(*set_var);
body.erase(goto_stmt);
return new_goto_node;
}
Node MoveOutwardIf(Node goto_stmt) {
const Node parent{Tree::s_iterator_to(*goto_stmt->up)};
Tree& body{parent->children};
const u32 label_id{goto_stmt->label->id};
Statement* const goto_cond{goto_stmt->cond};
Statement* const set_goto_var{pool.Create(SetVariable{}, label_id, goto_cond, &*parent)};
body.insert(goto_stmt, *set_goto_var);
Tree if_body;
if_body.splice(if_body.begin(), body, std::next(goto_stmt), body.end());
if_body.pop_front();
Statement* const cond{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const neg_cond{pool.Create(Not{}, cond, &root_stmt)};
Statement* const if_stmt{pool.Create(If{}, neg_cond, std::move(if_body), &*parent)};
UpdateTreeUp(if_stmt);
body.insert(goto_stmt, *if_stmt);
body.erase(goto_stmt);
Statement* const new_cond{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const new_goto{pool.Create(Goto{}, new_cond, goto_stmt->label, parent->up)};
Tree& parent_tree{parent->up->children};
return parent_tree.insert(std::next(parent), *new_goto);
}
Node MoveOutwardLoop(Node goto_stmt) {
Statement* const parent{goto_stmt->up};
Tree& body{parent->children};
const u32 label_id{goto_stmt->label->id};
Statement* const goto_cond{goto_stmt->cond};
Statement* const set_goto_var{pool.Create(SetVariable{}, label_id, goto_cond, parent)};
Statement* const cond{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const break_stmt{pool.Create(Break{}, cond, parent)};
body.insert(goto_stmt, *set_goto_var);
body.insert(goto_stmt, *break_stmt);
body.erase(goto_stmt);
const Node loop{Tree::s_iterator_to(*goto_stmt->up)};
Statement* const new_goto_cond{pool.Create(Variable{}, label_id, &root_stmt)};
Statement* const new_goto{pool.Create(Goto{}, new_goto_cond, goto_stmt->label, loop->up)};
Tree& parent_tree{loop->up->children};
return parent_tree.insert(std::next(loop), *new_goto);
}
ObjectPool<Statement>& pool;
Statement root_stmt{FunctionTag{}};
};
[[nodiscard]] Statement* TryFindForwardBlock(Statement& stmt) {
Tree& tree{stmt.up->children};
const Node end{tree.end()};
Node forward_node{std::next(Tree::s_iterator_to(stmt))};
while (forward_node != end && !HasChildren(forward_node->type)) {
if (forward_node->type == StatementType::Code) {
return &*forward_node;
}
++forward_node;
}
return nullptr;
}
[[nodiscard]] IR::U1 VisitExpr(IR::IREmitter& ir, const Statement& stmt) {
switch (stmt.type) {
case StatementType::Identity:
return ir.Condition(stmt.guest_cond);
case StatementType::Not:
return ir.LogicalNot(IR::U1{VisitExpr(ir, *stmt.op)});
case StatementType::Or:
return ir.LogicalOr(VisitExpr(ir, *stmt.op_a), VisitExpr(ir, *stmt.op_b));
case StatementType::Variable:
return ir.GetGotoVariable(stmt.id);
case StatementType::IndirectBranchCond:
return ir.IEqual(ir.GetIndirectBranchVariable(), ir.Imm32(stmt.location));
default:
throw NotImplementedException("Statement type {}", stmt.type);
}
}
class TranslatePass {
public:
TranslatePass(ObjectPool<IR::Inst>& inst_pool_, ObjectPool<IR::Block>& block_pool_,
ObjectPool<Statement>& stmt_pool_, Environment& env_, Statement& root_stmt,
IR::AbstractSyntaxList& syntax_list_, const HostTranslateInfo& host_info)
: stmt_pool{stmt_pool_}, inst_pool{inst_pool_}, block_pool{block_pool_}, env{env_},
syntax_list{syntax_list_} {
Visit(root_stmt, nullptr, nullptr);
IR::Block& first_block{*syntax_list.front().data.block};
IR::IREmitter ir(first_block, first_block.begin());
ir.Prologue();
if (uses_demote_to_helper && host_info.needs_demote_reorder) {
DemoteCombinationPass();
}
}
private:
void Visit(Statement& parent, IR::Block* break_block, IR::Block* fallthrough_block) {
IR::Block* current_block{};
const auto ensure_block{[&] {
if (current_block) {
return;
}
current_block = block_pool.Create(inst_pool);
auto& node{syntax_list.emplace_back()};
node.type = IR::AbstractSyntaxNode::Type::Block;
node.data.block = current_block;
}};
Tree& tree{parent.children};
for (auto it = tree.begin(); it != tree.end(); ++it) {
Statement& stmt{*it};
switch (stmt.type) {
case StatementType::Label:
// Labels can be ignored
break;
case StatementType::Code: {
ensure_block();
Translate(env, current_block, stmt.block->begin.Offset(), stmt.block->end.Offset());
break;
}
case StatementType::SetVariable: {
ensure_block();
IR::IREmitter ir{*current_block};
ir.SetGotoVariable(stmt.id, VisitExpr(ir, *stmt.op));
break;
}
case StatementType::SetIndirectBranchVariable: {
ensure_block();
IR::IREmitter ir{*current_block};
IR::U32 address{ir.IAdd(ir.GetReg(stmt.branch_reg), ir.Imm32(stmt.branch_offset))};
ir.SetIndirectBranchVariable(address);
break;
}
case StatementType::If: {
ensure_block();
IR::Block* const merge_block{MergeBlock(parent, stmt)};
// Implement if header block
IR::IREmitter ir{*current_block};
const IR::U1 cond{ir.ConditionRef(VisitExpr(ir, *stmt.cond))};
const size_t if_node_index{syntax_list.size()};
syntax_list.emplace_back();
// Visit children
const size_t then_block_index{syntax_list.size()};
Visit(stmt, break_block, merge_block);
IR::Block* const then_block{syntax_list.at(then_block_index).data.block};
current_block->AddBranch(then_block);
current_block->AddBranch(merge_block);
current_block = merge_block;
auto& if_node{syntax_list[if_node_index]};
if_node.type = IR::AbstractSyntaxNode::Type::If;
if_node.data.if_node.cond = cond;
if_node.data.if_node.body = then_block;
if_node.data.if_node.merge = merge_block;
auto& endif_node{syntax_list.emplace_back()};
endif_node.type = IR::AbstractSyntaxNode::Type::EndIf;
endif_node.data.end_if.merge = merge_block;
auto& merge{syntax_list.emplace_back()};
merge.type = IR::AbstractSyntaxNode::Type::Block;
merge.data.block = merge_block;
break;
}
case StatementType::Loop: {
IR::Block* const loop_header_block{block_pool.Create(inst_pool)};
if (current_block) {
current_block->AddBranch(loop_header_block);
}
auto& header_node{syntax_list.emplace_back()};
header_node.type = IR::AbstractSyntaxNode::Type::Block;
header_node.data.block = loop_header_block;
IR::Block* const continue_block{block_pool.Create(inst_pool)};
IR::Block* const merge_block{MergeBlock(parent, stmt)};
const size_t loop_node_index{syntax_list.size()};
syntax_list.emplace_back();
// Visit children
const size_t body_block_index{syntax_list.size()};
Visit(stmt, merge_block, continue_block);
// The continue block is located at the end of the loop
IR::IREmitter ir{*continue_block};
const IR::U1 cond{ir.ConditionRef(VisitExpr(ir, *stmt.cond))};
IR::Block* const body_block{syntax_list.at(body_block_index).data.block};
loop_header_block->AddBranch(body_block);
continue_block->AddBranch(loop_header_block);
continue_block->AddBranch(merge_block);
current_block = merge_block;
auto& loop{syntax_list[loop_node_index]};
loop.type = IR::AbstractSyntaxNode::Type::Loop;
loop.data.loop.body = body_block;
loop.data.loop.continue_block = continue_block;
loop.data.loop.merge = merge_block;
auto& continue_block_node{syntax_list.emplace_back()};
continue_block_node.type = IR::AbstractSyntaxNode::Type::Block;
continue_block_node.data.block = continue_block;
auto& repeat{syntax_list.emplace_back()};
repeat.type = IR::AbstractSyntaxNode::Type::Repeat;
repeat.data.repeat.cond = cond;
repeat.data.repeat.loop_header = loop_header_block;
repeat.data.repeat.merge = merge_block;
auto& merge{syntax_list.emplace_back()};
merge.type = IR::AbstractSyntaxNode::Type::Block;
merge.data.block = merge_block;
break;
}
case StatementType::Break: {
ensure_block();
IR::Block* const skip_block{MergeBlock(parent, stmt)};
IR::IREmitter ir{*current_block};
const IR::U1 cond{ir.ConditionRef(VisitExpr(ir, *stmt.cond))};
current_block->AddBranch(break_block);
current_block->AddBranch(skip_block);
current_block = skip_block;
auto& break_node{syntax_list.emplace_back()};
break_node.type = IR::AbstractSyntaxNode::Type::Break;
break_node.data.break_node.cond = cond;
break_node.data.break_node.merge = break_block;
break_node.data.break_node.skip = skip_block;
auto& merge{syntax_list.emplace_back()};
merge.type = IR::AbstractSyntaxNode::Type::Block;
merge.data.block = skip_block;
break;
}
case StatementType::Return: {
ensure_block();
IR::Block* return_block{block_pool.Create(inst_pool)};
IR::IREmitter{*return_block}.Epilogue();
current_block->AddBranch(return_block);
auto& merge{syntax_list.emplace_back()};
merge.type = IR::AbstractSyntaxNode::Type::Block;
merge.data.block = return_block;
current_block = nullptr;
syntax_list.emplace_back().type = IR::AbstractSyntaxNode::Type::Return;
break;
}
case StatementType::Kill: {
ensure_block();
IR::Block* demote_block{MergeBlock(parent, stmt)};
IR::IREmitter{*current_block}.DemoteToHelperInvocation();
current_block->AddBranch(demote_block);
current_block = demote_block;
auto& merge{syntax_list.emplace_back()};
merge.type = IR::AbstractSyntaxNode::Type::Block;
merge.data.block = demote_block;
uses_demote_to_helper = true;
break;
}
case StatementType::Unreachable: {
ensure_block();
current_block = nullptr;
syntax_list.emplace_back().type = IR::AbstractSyntaxNode::Type::Unreachable;
break;
}
default:
throw NotImplementedException("Statement type {}", stmt.type);
}
}
if (current_block) {
if (fallthrough_block) {
current_block->AddBranch(fallthrough_block);
} else {
syntax_list.emplace_back().type = IR::AbstractSyntaxNode::Type::Unreachable;
}
}
}
IR::Block* MergeBlock(Statement& parent, Statement& stmt) {
Statement* merge_stmt{TryFindForwardBlock(stmt)};
if (!merge_stmt) {
// Create a merge block we can visit later
merge_stmt = stmt_pool.Create(&dummy_flow_block, &parent);
parent.children.insert(std::next(Tree::s_iterator_to(stmt)), *merge_stmt);
}
return block_pool.Create(inst_pool);
}
void DemoteCombinationPass() {
using Type = IR::AbstractSyntaxNode::Type;
std::vector<IR::Block*> demote_blocks;
std::vector<IR::U1> demote_conds;
u32 num_epilogues{};
u32 branch_depth{};
for (const IR::AbstractSyntaxNode& node : syntax_list) {
if (node.type == Type::If) {
++branch_depth;
}
if (node.type == Type::EndIf) {
--branch_depth;
}
if (node.type != Type::Block) {
continue;
}
if (branch_depth > 1) {
// Skip reordering nested demote branches.
continue;
}
for (const IR::Inst& inst : node.data.block->Instructions()) {
const IR::Opcode op{inst.GetOpcode()};
if (op == IR::Opcode::DemoteToHelperInvocation) {
demote_blocks.push_back(node.data.block);
break;
}
if (op == IR::Opcode::Epilogue) {
++num_epilogues;
}
}
}
if (demote_blocks.size() == 0) {
return;
}
if (num_epilogues > 1) {
LOG_DEBUG(Shader, "Combining demotes with more than one return is not implemented.");
return;
}
s64 last_iterator_offset{};
auto& asl{syntax_list};
for (const IR::Block* demote_block : demote_blocks) {
const auto start_it{asl.begin() + last_iterator_offset};
auto asl_it{std::find_if(start_it, asl.end(), [&](const IR::AbstractSyntaxNode& asn) {
return asn.type == Type::If && asn.data.if_node.body == demote_block;
})};
if (asl_it == asl.end()) {
// Demote without a conditional branch.
// No need to proceed since all fragment instances will be demoted regardless.
return;
}
const IR::Block* const end_if = asl_it->data.if_node.merge;
demote_conds.push_back(asl_it->data.if_node.cond);
last_iterator_offset = std::distance(asl.begin(), asl_it);
asl_it = asl.erase(asl_it);
asl_it = std::find_if(asl_it, asl.end(), [&](const IR::AbstractSyntaxNode& asn) {
return asn.type == Type::Block && asn.data.block == demote_block;
});
asl_it = asl.erase(asl_it);
asl_it = std::find_if(asl_it, asl.end(), [&](const IR::AbstractSyntaxNode& asn) {
return asn.type == Type::EndIf && asn.data.end_if.merge == end_if;
});
asl_it = asl.erase(asl_it);
}
const auto epilogue_func{[](const IR::AbstractSyntaxNode& asn) {
if (asn.type != Type::Block) {
return false;
}
for (const auto& inst : asn.data.block->Instructions()) {
if (inst.GetOpcode() == IR::Opcode::Epilogue) {
return true;
}
}
return false;
}};
const auto reverse_it{std::find_if(asl.rbegin(), asl.rend(), epilogue_func)};
const auto return_block_it{(reverse_it + 1).base()};
IR::IREmitter ir{*(return_block_it - 1)->data.block};
IR::U1 cond(IR::Value(false));
for (const auto& demote_cond : demote_conds) {
cond = ir.LogicalOr(cond, demote_cond);
}
cond.Inst()->DestructiveAddUsage(1);
IR::AbstractSyntaxNode demote_if_node{};
demote_if_node.type = Type::If;
demote_if_node.data.if_node.cond = cond;
demote_if_node.data.if_node.body = demote_blocks[0];
demote_if_node.data.if_node.merge = return_block_it->data.block;
IR::AbstractSyntaxNode demote_node{};
demote_node.type = Type::Block;
demote_node.data.block = demote_blocks[0];
IR::AbstractSyntaxNode demote_endif_node{};
demote_endif_node.type = Type::EndIf;
demote_endif_node.data.end_if.merge = return_block_it->data.block;
const auto next_it_1 = asl.insert(return_block_it, demote_endif_node);
const auto next_it_2 = asl.insert(next_it_1, demote_node);
asl.insert(next_it_2, demote_if_node);
}
ObjectPool<Statement>& stmt_pool;
ObjectPool<IR::Inst>& inst_pool;
ObjectPool<IR::Block>& block_pool;
Environment& env;
IR::AbstractSyntaxList& syntax_list;
bool uses_demote_to_helper{};
const Flow::Block dummy_flow_block{};
};
} // Anonymous namespace
IR::AbstractSyntaxList BuildASL(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Block>& block_pool,
Environment& env, Flow::CFG& cfg,
const HostTranslateInfo& host_info) {
ObjectPool<Statement> stmt_pool{64};
GotoPass goto_pass{cfg, stmt_pool};
Statement& root{goto_pass.RootStatement()};
IR::AbstractSyntaxList syntax_list;
TranslatePass{inst_pool, block_pool, stmt_pool, env, root, syntax_list, host_info};
stmt_pool.ReleaseContents();
return syntax_list;
}
} // namespace Shader::Maxwell

View File

@@ -0,0 +1,22 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "shader_recompiler/environment.h"
#include "shader_recompiler/frontend/ir/abstract_syntax_list.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/frontend/maxwell/control_flow.h"
#include "shader_recompiler/object_pool.h"
namespace Shader {
struct HostTranslateInfo;
namespace Maxwell {
[[nodiscard]] IR::AbstractSyntaxList BuildASL(ObjectPool<IR::Inst>& inst_pool,
ObjectPool<IR::Block>& block_pool, Environment& env,
Flow::CFG& cfg, const HostTranslateInfo& host_info);
} // namespace Maxwell
} // namespace Shader

Some files were not shown because too many files have changed in this diff Show More