raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-23 04:16:57 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: encode SNORM16 formatted stores for skinning buffers

Browse Source
This commit is contained in:
nmzik
2026-09-17 17:40:05 +04:00
parent 4b77686566
commit c66d977c1d
2 changed files with 93 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp
+22 -25
View File
@@ -514,31 +514,6 @@ void StoreWord(ValueEmitContext& ctx, const IR::Inst& inst, IR::MemoryInfo mem)
});
}
void FormattedStorePrepared(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
uint32_t component, const MemoryResourceAccess& resource,
uint32_t data) {
const auto info = Format::GetFormatInfo(BufferFormat(ctx, mem));
if (info.type == Format::ComponentType::Unknown) {
StoreWordPrepared(ctx, inst, RebaseRawComponent(mem, component), resource, data);
return;
}
if (component >= info.component_count) return;
const auto bits = info.component_bits[component];
const auto component_mem = RebaseFormattedComponent(mem, info, component);
if (bits == 8u || bits == 16u) {
StoreSubwordPrepared(ctx, inst, component_mem, resource, bits, data);
} else {
StoreWordPrepared(ctx, inst, component_mem, resource, data);
}
}
void FormattedStore(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
EmitIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
const auto resource = PrepareMemoryResourceAccess(ctx.state, mem);
FormattedStorePrepared(ctx, inst, mem, 0u, resource, ctx.Arg(inst, inst.NumArgs() - 2));
});
}
spv::Op SpirvAtomicOpcode(IR::ValueOpcode opcode) {
switch (opcode) {
case IR::ValueOpcode::BufferAtomicCmpSwap32: return spv::OpAtomicCompareExchange;
@@ -729,6 +704,11 @@ void StoreFormattedInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
uint32_t data) {
if (component >= plan.info.component_count) return;
const auto bits = plan.info.component_bits[component];
if (plan.info.type == Format::ComponentType::Snorm && bits == 16u) {
const auto value = EmitBitCastF32U32(ctx.state, data);
data = EmitPackSnorm2x16(
ctx.state, EmitCompositeConstructF32x2(ctx.state, value, ConstantF32Value(ctx.state, 0.0f)));
}
if (bits == 8u || bits == 16u) {
StoreSubwordInBounds(ctx, mem, plan.resource, plan.addresses[component],
plan.indices[component], bits, data);
@@ -737,6 +717,23 @@ void StoreFormattedInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
}
}
void FormattedStore(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
EmitIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
const auto resource = PrepareMemoryResourceAccess(ctx.state, mem);
const auto data = ctx.Arg(inst, inst.NumArgs() - 2);
const auto info = Format::GetFormatInfo(BufferFormat(ctx, mem));
if (info.type == Format::ComponentType::Unknown) {
StoreWordPrepared(ctx, inst, RebaseRawComponent(mem, 0u), resource, data);
return;
}
const auto plan = PrepareFormattedMemory(ctx, inst, mem, resource, info, 1u,
FormattedAccess::Store);
EmitIfCondition(ctx.state, plan.in_bounds, [&]() {
StoreFormattedInBounds(ctx, mem, plan, 0u, data);
});
});
}
uint32_t LoadWideBuffer(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t components) {
auto& state = ctx.state;
return EmitValueOrDefaultIfCondition(
tests/ShaderRecompilerComputeTests.cpp
+71
View File
@@ -22382,6 +22382,63 @@ TestCase BufferStoreFormatXResource16UintPreservesAdjacentLanes() {
return test;
}
TestCase BufferStoreFormatXyzwSnorm16CapturedSkinningVectors() {
using O = ShaderOpcode;
// Normal and tangent from the character skinning dispatch (eadfd178c07feebd).
constexpr std::array<u32, 8> values = {
0x3d98f7c5u, 0xbedae712u, 0xbf66a19eu, 0x3f800000u,
0x3f7e6baeu, 0xbd339a7du, 0x3dd0a2e2u, 0xbf800000u};
std::vector<u32> code;
for (u32 record = 0; record < 2; record++) {
for (u32 component = 0; component < 4; component++) {
AppendVMovLiteral(&code, component, values[record * 4 + component]);
}
AppendVMovU32(&code, 20, record);
code.push_back(EncodeMubuf0(0x07u, 0, true, false));
code.push_back(EncodeMubuf1(0, 0, 20));
}
AppendEnd(&code);
TestCase test;
test.name = "BufferStoreFormatXyzwSnorm16CapturedSkinningVectors";
test.code = std::move(code);
test.initial = std::vector<u32>(4, 0xdeadbeefu);
test.expected = {0xc947098fu, 0x7fff8cb0u, 0xfa637f35u, 0x80010d0au};
test.user_data = MakeStructuredStorageBufferData(
8, 2, false, BufferFormat(Prospero::BufferFormat::k16_16_16_16SNorm));
test.has_user_data = true;
test.opcodes = {O::V_MOV_B32, O::BUFFER_STORE_FORMAT_XYZW, O::S_ENDPGM};
return test;
}
TestCase BufferStoreFormatXSnorm16ClampsRoundsAndPreservesHalfwords() {
using O = ShaderOpcode;
constexpr std::array<float, 9> values = {
-2.0f, -1.0f, -0.75f, -0.25f, 0.0f, 0.25f, 0.75f, 1.0f, 2.0f};
std::vector<u32> code;
for (u32 i = 0; i < values.size(); i++) {
AppendVMovLiteral(&code, 0, std::bit_cast<u32>(values[i]));
AppendVMovU32(&code, 20, i * 2 + 2);
code.push_back(EncodeMubuf0(0x04u));
code.push_back(EncodeMubuf1(0, 0, 20));
}
AppendEnd(&code);
TestCase test;
test.name = "BufferStoreFormatXSnorm16ClampsRoundsAndPreservesHalfwords";
test.code = std::move(code);
test.initial = std::vector<u32>(6, 0xdeadbeefu);
test.expected = {0x8001beefu, 0xa0018001u, 0x0000e000u,
0x5fff2000u, 0x7fff7fffu, 0xdeadbeefu};
test.user_data = MakeStructuredStorageBufferData(
0, 24, false, BufferFormat(Prospero::BufferFormat::k16SNorm));
test.has_user_data = true;
test.opcodes = {O::V_MOV_B32, O::BUFFER_STORE_FORMAT_X, O::S_ENDPGM};
return test;
}
TestCase BufferLoadFormatXyResource88UintExtractsBytes() {
using O = ShaderOpcode;
@@ -26901,6 +26958,8 @@ std::vector<TestCase> MakeCases() {
AddCase(BufferFormatStoreVariants);
AddCase(BufferStoreFormatXResource16UintWritesHalfword);
AddCase(BufferStoreFormatXResource16UintPreservesAdjacentLanes);
AddCase(BufferStoreFormatXyzwSnorm16CapturedSkinningVectors);
AddCase(BufferStoreFormatXSnorm16ClampsRoundsAndPreservesHalfwords);
AddCase(BufferLoadFormatXResource8UintZeroExtendsByte);
AddCase(BufferLoadFormatXyResource88UintExtractsBytes);
AddCase(BufferLoadFormatXyResource8888UnormConvertsFirstTwoComponents);
@@ -31512,6 +31571,18 @@ int main(int argc, char **argv) {
std::setvbuf(stdout, nullptr, _IONBF, 0);
EnsureConfigInitialized();
CheckLeastRecentlyUsedCacheOrdering();
if (argc == 2 && std::strcmp(argv[1], "--buffer-snorm-store-only") == 0) {
VulkanHarness vulkan;
RunCase(&vulkan, BufferStoreFormatXyzwSnorm16CapturedSkinningVectors());
RunCase(&vulkan, BufferStoreFormatXSnorm16ClampsRoundsAndPreservesHalfwords());
RunCase(&vulkan, BufferStoreFormatXResource16UintWritesHalfword());
RunCase(&vulkan, BufferStoreFormatXResource16UintPreservesAdjacentLanes());
RunCase(&vulkan, BufferStoreFormatXyResource88UintWritesBytes());
RunCase(&vulkan, BufferStoreFormatXyzwDropsPartialRecord());
RunCase(&vulkan, BufferStoreFormatXChecksOnlyTransferredComponent());
RunCase(&vulkan, BufferFormatStoreVariants());
return 0;
}
if (argc == 2 && std::strcmp(argv[1], "--s-ashr-i64-only") == 0) {
VulkanHarness vulkan;
RunCase(&vulkan, ScalarAshrI64Edges(false));