Triforce v0.7.0 "Sage Of Courage"

The 0.6 crashes after warps and area changes fixed (no write-back into memory the capture leaves out), invisible objects fixed (occlusion gate), DLSS frame generation flicker fixed, achievements, the renderer built with a play profile, faster translated game code, helper threads judged on settled scenes, memory, shader preparation, audio effects, launcher, controls and Linux fixes.
This commit is contained in:
ruckus7382sfgsgdf
2026-11-22 11:59:06 -06:00
parent 70c210cee1
commit e6f2f248f2
529 changed files with 55958 additions and 1898 deletions

View File

@@ -3,6 +3,7 @@
#include <array>
#include <atomic>
#include <cstddef>
#include <functional>
#include <memory>
#include <string>
@@ -13,6 +14,8 @@
#include "shader_recompiler/shader_info.h"
#include "spv_store.h"
#include "compute_pin.h" // ((Triforce, 11-22, PIPE-52) which compute programs are held to 32 lanes: compute_pin_mode)
#include "fallback_size_policy.h" // ((Triforce, 11-22, PIPE-56) the mesh fallback's stages too big for two Mesa drivers)
#include "vk_internal.h"
#include "vk_state.h"
@@ -65,6 +68,16 @@ inline VkShaderModule stage_exact_key(const Stage& st) {
}
/// Whether stages are packed in this run (on unless RX_VK_SPV_PACK=0)
bool spv_pack_on();
/// ((Triforce, 11-22, the performance audit's PIPE-52) RX_VK_COMPUTE_PIN on this driver (compute_pin.h; read once, logged once
/// where it is not "all"), and whether a compute module was looked at and has no barrier of Subgroup scope (false for a
/// module nobody looked at, and always with "all", where no module is looked at)
compute_pin::Mode compute_pin_mode();
bool compute_lanes_free(VkShaderModule m);
/// ((Triforce, 11-22, the performance audit's PIPE-56) Which of the two drivers with a limit on a mesh-fallback stage's size
/// this is (fallback_size_policy.h; Other on every other driver, where no module is counted), and a stage's OpPhi and
/// OpFunctionCall counts as make_module took them (0 and 0 for a stage nobody counted). Any thread
fallback_size::Driver fallback_size_driver();
fallback_size::Size stage_size(const Stage& st);
/// (Triforce) A program stage's module let go (vkDestroyShaderModule), and forgotten by the memory line's count (mem_parts.h):
/// every place that lets a program's module go does it through this
void stage_module_destroy(VkShaderModule m);
@@ -77,6 +90,12 @@ struct Program {
uint64_t key = 0;
bool compute = false, failed = false;
std::atomic<bool> ready{false}; // translated (graphics programs are translated on a worker: vk_jobs.h)
// ((Triforce, 11-22, the performance audit's DESC-22, after Source A's; desc_counts.h) Its descriptor counts as
// build_descriptors sums them (buffers, images, texel views), packed: ~0u until its first build once it is ready.
// It sits in the four bytes of padding before stages, and THIS FIELD MUST NOT GROW THE STRUCT: sizeof(Program) is in
// the program cache's stamp (vk_shader.cpp disk_stamp), and a change of it throws every player's shader and pipeline
// caches away (it did on 11-18, 272 to 280: vk_draw.cpp). The static_assert under the struct holds it there)
mutable std::atomic<uint32_t> desc_n{~0u};
std::vector<Stage> stages;
std::array<uint32_t, 3> workgroup{1, 1, 1};
uint32_t shared_memory = 0;
@@ -121,6 +140,10 @@ struct Program {
std::vector<uint8_t> questions;
uint64_t codes[4]{}; // (hookshot) its stages' code keys: vertex (mesh), fragment, task, the header-only stage (the state list)
};
// ((Triforce, 11-22, DESC-22) desc_n in the padding between ready and stages, where it changes no size: see the field)
static_assert(sizeof(std::atomic<uint32_t>) == 4 && offsetof(Program, desc_n) + 4 <= offsetof(Program, stages) &&
offsetof(Program, stages) == 16,
"Program::desc_n must fit the padding before Program::stages: sizeof(Program) is in the program cache's stamp");
enum class ProgramKind { Graphics, Compute, Mesh };