raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 13:56:59 +01:00
Code Issues Packages Projects Releases Wiki Activity

shader: unpack buffer offsets from one load per packed word

Browse Source
This commit is contained in:
nmzik
2026-10-09 02:46:18 +02:00
parent b4853a0e08
commit 13e3ca314c
2 changed files with 29 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemoryHelpers.cpp
+5 -5
View File
@@ -104,13 +104,13 @@ Prospero::BufferFormat StorageBufferFormat(const EmitterState& state, const IR::
void EmitMemoryOffsets(EmitterState& state) { void EmitMemoryOffsets(EmitterState& state) {
const auto count = state.program.bindings.descriptor_counts[static_cast<size_t>(IR::DescriptorBindingKind::Buffers)]; const auto count = state.program.bindings.descriptor_counts[static_cast<size_t>(IR::DescriptorBindingKind::Buffers)];
for (uint32_t i = 0; i < count; i++) { for (uint32_t i = 0; i < count; i += 4u) {
const auto word = const auto word =
EmitShaderDataDwordLoad(state, state.program.bindings.memory_offset_dword + i / 4u); EmitShaderDataDwordLoad(state, state.program.bindings.memory_offset_dword + i / 4u);
const auto shift = ConstantU32(state, (i % 4u) * 8u); for (uint32_t lane = 0; lane < std::min(4u, count - i); ++lane) {
state.memory_byte_offsets[i] = EmitBinaryU32( state.memory_byte_offsets[i + lane] = EmitNative<spv::OpBitFieldUExtract, IR::Type::U32>(
state, spv::OpBitwiseAnd, EmitBinaryU32(state, spv::OpShiftRightLogical, word, shift), state, word, ConstantU32(state, lane * 8u), ConstantU32(state, 8u));
ConstantU32(state, 0xffu)); }
} }
} }
tests/ShaderRecompilerComputeTests.cpp
+24 -25
View File
@@ -28545,32 +28545,29 @@ TestCase BufferStoreDwordAppliesHostOffset() {
return test; return test;
} }
TestCase BufferOffsetsUsePackedLaneAndStorageFallback() { TestCase BufferOffsetsUsePackedWords(bool storage) {
using O = ShaderOpcode; using O = ShaderOpcode;
constexpr std::array offsets{0u, 12u, 20u, 28u, 36u};
std::vector<u32> code;
for (u32 base : {0u, 4u}) {
AppendSMovLiteral(&code, base, 0x1000u + base * 0x1000u);
AppendSMovLiteral(&code, base + 1u, 4u << 16u);
AppendSMovLiteral(&code, base + 2u, 1u);
AppendSMovLiteral(&code, base + 3u, 1u << 24u);
}
AppendVMovLiteral(&code, 0, 0x12345678u);
code.push_back(EncodeMubuf0(0x1cu, 0, false, false));
code.push_back(EncodeMubuf1(0, 0, 20));
AppendVMovLiteral(&code, 1, 0xabcdef01u);
code.push_back(EncodeMubuf0(0x1cu, 0, false, false));
code.push_back(EncodeMubuf1(1, 1, 20));
AppendEnd(&code);
TestCase test; TestCase test;
test.name = "BufferOffsetsUsePackedLaneAndStorageFallback"; test.name = storage ? "BufferOffsetsPackedWordsStorage" : "BufferOffsetsPackedWordsPush";
test.code = std::move(code); test.initial.resize(10);
test.initial = {0, 0, 0, 0}; test.expected = test.initial;
test.expected = {0x12345678u, 0, 0, 0xabcdef01u}; for (u32 i = 0; i < offsets.size(); ++i) {
const auto base = i * 4u;
AppendSMovLiteral(&test.code, base, 0x1000u + base * 0x1000u);
AppendSMovLiteral(&test.code, base + 1u, 4u << 16u);
AppendSMovLiteral(&test.code, base + 2u, 1u);
AppendSMovLiteral(&test.code, base + 3u, 1u << 24u);
const auto value = 0x12345678u + i;
AppendVMovLiteral(&test.code, i, value);
test.code.push_back(EncodeMubuf0(0x1cu, 0, false, false));
test.code.push_back(EncodeMubuf1(i, i, 20));
test.expected[offsets[i] / 4u] = value;
}
AppendEnd(&test.code);
test.storage_buffer_range_bytes = 4; test.storage_buffer_range_bytes = 4;
test.storage_buffer_offsets = {0, 12}; test.storage_buffer_offsets.assign(offsets.begin(), offsets.end());
test.expand_shader_data_storage = true; test.expand_shader_data_storage = storage;
test.opcodes = {O::S_MOV_B32, O::V_MOV_B32, O::BUFFER_STORE_DWORD, test.opcodes = {O::S_MOV_B32, O::V_MOV_B32, O::BUFFER_STORE_DWORD,
O::S_ENDPGM}; O::S_ENDPGM};
return test; return test;
@@ -36576,7 +36573,8 @@ std::vector<TestCase> MakeCases() {
AddCase(BufferLoadDwordIdxenUsesDescriptorStride); AddCase(BufferLoadDwordIdxenUsesDescriptorStride);
AddCase(BufferStoreDwordIdxenUsesDescriptorStride); AddCase(BufferStoreDwordIdxenUsesDescriptorStride);
AddCase(BufferStoreDwordAppliesHostOffset); AddCase(BufferStoreDwordAppliesHostOffset);
AddCase(BufferOffsetsUsePackedLaneAndStorageFallback); cases.push_back(BufferOffsetsUsePackedWords(false));
cases.push_back(BufferOffsetsUsePackedWords(true));
AddCase(BufferLoadVariants); AddCase(BufferLoadVariants);
for (const u32 component : {4u, 2u, 0u}) { for (const u32 component : {4u, 2u, 0u}) {
cases.push_back(BufferSubwordLoadsAtHostOffset(2, component, false)); cases.push_back(BufferSubwordLoadsAtHostOffset(2, component, false));
@@ -42370,7 +42368,8 @@ int main(int argc, char **argv) {
if (argc == 2 && std::strcmp(argv[1], "--shader-data-storage-only") == 0) { if (argc == 2 && std::strcmp(argv[1], "--shader-data-storage-only") == 0) {
CheckResourcePlanHandoff(); CheckResourcePlanHandoff();
VulkanHarness vulkan; VulkanHarness vulkan;
RunCase(&vulkan, BufferOffsetsUsePackedLaneAndStorageFallback()); RunCase(&vulkan, BufferOffsetsUsePackedWords(false));
RunCase(&vulkan, BufferOffsetsUsePackedWords(true));
return 0; return 0;
} }
if (argc == 2 && std::strcmp(argv[1], "--buffer-publication-only") == 0) { if (argc == 2 && std::strcmp(argv[1], "--buffer-publication-only") == 0) {