Decode the upscaler stand-in's picture with a square, emit SURED with its real opcode and keep the translator's warnings.

The post pass stores the square root of each channel, so the stand-in for Ocarina's untranslatable upscaler now
filters those values and squares them in its compute pass (shaders/fallback_tonemap.spvasm, vk_tonemap.c) instead of
decoding them with the sRGB curve, which darkened mid-tones and lifted the darkest values; the pass is compiled in again,
its round-24 device loss having been the push descriptor call outside the recorder. NX2NAVI_FALLBACK_EXPOSURE is gone.
AmpereLowering emitted SURED as 0xeb48, which matches no Maxwell instruction, so the translator skipped every image
reduction; with 0xeb58 the edge AA's blend-weight pass writes its image. The translator's own warnings go to res.json
notes, and shader dumps are named by program serial so they match the frame dumps' program numbers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
chrg6
2026-11-06 13:11:58 -05:00
parent 5dfa690d96
commit eed3e6b786
8 changed files with 92 additions and 123 deletions

View File

@@ -1,18 +1,20 @@
; Tonemap and scale for the untranslatable upscaler (runtime/vk_exec_draw.c untranslated_dispatch_fallback).
; Scale and decode for the untranslatable scene upscaler (runtime/vk_exec_draw.c untranslated_dispatch_fallback).
; Its input is the post pass's picture with each channel stored as the square root of its linear value (prog 102 packs
; sqrt(composite) into RGB9E5), and its output is linear (the HUD is drawn over it, the frame's last pass encodes to
; sRGB). So: filter in the stored (sqrt) values, as the game's upscaler does, then square. The pass replaced a
; decode with the sRGB curve, which darkened mid-tones and lifted the darkest values.
; Written in SPIR-V assembly (spirv-as; no GLSL compiler in the build). As GLSL:
; layout(local_size_x = 8, local_size_y = 8) in;
; layout(binding = 0) uniform sampler2D src; // the HDR scene (linear filtering)
; layout(binding = 0) uniform sampler2D src; // the stored picture (linear filtering)
; layout(binding = 1, rgba16f) uniform writeonly image2D dst; // the pass output (vk_tonemap.c), blitted into the target
; layout(push_constant) uniform P { vec2 inv; float exposure; float pad; ivec2 size; } pc;
; layout(push_constant) uniform P { vec2 inv; float unused; float pad; ivec2 size; } pc;
; void main() {
; ivec2 p = ivec2(gl_GlobalInvocationID.xy);
; if (p.x >= pc.size.x || p.y >= pc.size.y) return;
; vec3 c = textureLod(src, (vec2(p) + 0.5) * pc.inv, 0.0).rgb * pc.exposure;
; c = clamp(c * (2.51 * c + 0.03) / (c * (2.43 * c + 0.59) + 0.14), 0.0, 1.0); // Narkowicz's ACES fit
; imageStore(dst, p, vec4(c, 1.0));
; vec3 s = textureLod(src, (vec2(p) + 0.5) * pc.inv, 0.0).rgb;
; imageStore(dst, p, vec4(s * s, 1.0));
; }
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint GLCompute %main "main" %gid %src %dst %pc
OpExecutionMode %main LocalSize 8 8 1
@@ -52,27 +54,13 @@
%ptr_pc_P = OpTypePointer PushConstant %P
%pc = OpVariable %ptr_pc_P PushConstant
%ptr_pc_v2float = OpTypePointer PushConstant %v2float
%ptr_pc_float = OpTypePointer PushConstant %float
%ptr_pc_v2int = OpTypePointer PushConstant %v2int
%int_0 = OpConstant %int 0
%int_1 = OpConstant %int 1
%int_3 = OpConstant %int 3
%float_0 = OpConstant %float 0
%float_1 = OpConstant %float 1
%float_half = OpConstant %float 0.5
%float_2_51 = OpConstant %float 2.51
%float_0_03 = OpConstant %float 0.03
%float_2_43 = OpConstant %float 2.43
%float_0_59 = OpConstant %float 0.59
%float_0_14 = OpConstant %float 0.14
%v2half = OpConstantComposite %v2float %float_half %float_half
%v3_0 = OpConstantComposite %v3float %float_0 %float_0 %float_0
%v3_1 = OpConstantComposite %v3float %float_1 %float_1 %float_1
%v3_2_51 = OpConstantComposite %v3float %float_2_51 %float_2_51 %float_2_51
%v3_0_03 = OpConstantComposite %v3float %float_0_03 %float_0_03 %float_0_03
%v3_2_43 = OpConstantComposite %v3float %float_2_43 %float_2_43 %float_2_43
%v3_0_59 = OpConstantComposite %v3float %float_0_59 %float_0_59 %float_0_59
%v3_0_14 = OpConstantComposite %v3float %float_0_14 %float_0_14 %float_0_14
%v2bool = OpTypeVector %bool 2
%main = OpFunction %void None %fnvoid
%entry = OpLabel
@@ -96,18 +84,7 @@
%si = OpLoad %simg2d %src
%s4 = OpImageSampleExplicitLod %v4float %si %uv Lod %float_0
%s3 = OpVectorShuffle %v3float %s4 %s4 0 1 2
%pexp = OpAccessChain %ptr_pc_float %pc %int_1
%exp = OpLoad %float %pexp
%c = OpVectorTimesScalar %v3float %s3 %exp
%a1 = OpFMul %v3float %c %v3_2_51
%a2 = OpFAdd %v3float %a1 %v3_0_03
%num = OpFMul %v3float %c %a2
%b1 = OpFMul %v3float %c %v3_2_43
%b2 = OpFAdd %v3float %b1 %v3_0_59
%b3 = OpFMul %v3float %c %b2
%den = OpFAdd %v3float %b3 %v3_0_14
%t = OpFDiv %v3float %num %den
%tc = OpExtInst %v3float %1 FClamp %t %v3_0 %v3_1
%tc = OpFMul %v3float %s3 %s3
%o4 = OpCompositeConstruct %v4float %tc %float_1
%di = OpLoad %stimg %dst
OpImageWrite %di %p %o4