raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 05:46:56 +01:00
Code Issues Packages Projects Releases Wiki Activity

shader: support returning 64-bit LDS max and compare-exchange

Browse Source
This commit is contained in:
nmzik
2026-10-13 14:48:38 +02:00
parent e998f21e71
commit 9e41d8320a
12 changed files with 188 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterInstructions.h
+2
View File
@@ -322,6 +322,8 @@ inline constexpr auto EmitSharedAtomicDec32 = EmitSharedIncDec;
inline constexpr auto EmitSharedAtomicSMin32 = EmitAtomic32;
inline constexpr auto EmitSharedAtomicUMin32 = EmitAtomic32;
inline constexpr auto EmitSharedAtomicSMax32 = EmitAtomic32;
inline constexpr auto EmitSharedAtomicUMax64 = EmitSharedAtomic64;
inline constexpr auto EmitSharedAtomicCmpSwap64 = EmitSharedAtomic64;
inline constexpr auto EmitSharedAtomicUMax32 = EmitAtomic32;
inline constexpr auto EmitSharedAtomicAnd32 = EmitAtomic32;
inline constexpr auto EmitSharedAtomicOr32 = EmitAtomic32;
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp
+41 -25
View File
@@ -1240,32 +1240,48 @@ void EmitSharedAtomic64(ValueEmitContext& ctx, const IR::Inst& inst) {
auto& state = ctx.state;
const auto& mem = ctx.Memory(inst);
EnsureLdsStorage(state);
EmitIfCondition(state, ctx.Arg(inst, 2), [&]() {
const auto address = Binary(state, spv::OpBitwiseAnd, TypeU32(state),
ByteAddress(ctx, inst, mem), ConstantU32(state, 0xfff8u));
const auto index = Binary(state, spv::OpShiftRightLogical, TypeU32(state), address,
ConstantU32(state, 3u));
const auto in_bounds = Binary(state, spv::OpULessThan, TypeBool(state), index,
ConstantU32(state, LdsDwordCount(state) / 2u));
EmitIfCondition(state, in_bounds, [&]() {
auto native_index = index;
auto semantics = spv::MemorySemanticsWorkgroupMemoryMask;
if (state.lds_storage_class == spv::StorageClassStorageBuffer) {
native_index = EmitAddU32(state, index, EmitBinaryU32(
state, spv::OpShiftRightLogical, state.lds_base_dwords, ConstantU32(state, 1)));
semantics = spv::MemorySemanticsUniformMemoryMask;
}
const auto pointer = state.builder.AllocateId();
const auto address = Binary(state, spv::OpBitwiseAnd, TypeU32(state),
ByteAddress(ctx, inst, mem), ConstantU32(state, 0xfff8u));
const auto index = Binary(state, spv::OpShiftRightLogical, TypeU32(state), address,
ConstantU32(state, 3u));
const auto in_bounds = Binary(state, spv::OpULessThan, TypeBool(state), index,
ConstantU32(state, LdsDwordCount(state) / 2u));
const auto active = EmitLogicalAndBool(state, ctx.Arg(inst, inst.NumArgs() - 1), in_bounds);
const auto emit = [&]() {
auto native_index = index;
auto semantics = spv::MemorySemanticsWorkgroupMemoryMask;
if (state.lds_storage_class == spv::StorageClassStorageBuffer) {
native_index = EmitAddU32(state, index, EmitBinaryU32(
state, spv::OpShiftRightLogical, state.lds_base_dwords, ConstantU32(state, 1)));
semantics = spv::MemorySemanticsUniformMemoryMask;
}
const auto pointer = state.builder.AllocateId();
state.builder.AddFunction(
spv::OpAccessChain, TypePointer(state, state.lds_storage_class, TypeU64(state)),
pointer, state.lds_u64_variable, ConstantU32(state, 0), native_index);
const auto old = state.builder.AllocateId();
const auto opcode = SpirvAtomicOpcode(inst.GetOpcode());
const auto scope = ConstantU32(state, spv::ScopeWorkgroup);
const auto acquire_release =
ConstantU32(state, spv::MemorySemanticsAcquireReleaseMask | semantics);
if (opcode == spv::OpAtomicCompareExchange) {
// RDNA2 DS compares DATA and stores DATA2, opposite to BUFFER atomics.
state.builder.AddFunction(
spv::OpAccessChain, TypePointer(state, state.lds_storage_class, TypeU64(state)),
pointer, state.lds_u64_variable, ConstantU32(state, 0), native_index);
const auto value = ctx.Arg(inst, 1);
state.builder.AddFunction(
SpirvAtomicOpcode(inst.GetOpcode()), TypeU64(state), state.builder.AllocateId(),
pointer, ConstantU32(state, spv::ScopeWorkgroup),
ConstantU32(state, spv::MemorySemanticsAcquireReleaseMask | semantics), value);
});
});
opcode, TypeU64(state), old, pointer, scope, acquire_release,
ConstantU32(state, spv::MemorySemanticsAcquireMask | semantics),
ctx.Arg(inst, 2), ctx.Arg(inst, 1));
} else {
state.builder.AddFunction(opcode, TypeU64(state), old, pointer, scope,
acquire_release, ctx.Arg(inst, 1));
}
return old;
};
if (inst.HasUses()) {
ctx.Define(inst, EmitValueOrDefaultIfCondition(
state, active, TypeU64(state), ConstantU64(state, 0), emit));
} else {
EmitIfCondition(state, active, emit);
}
}
uint32_t EmitBufferFloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst) {
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemoryHelpers.cpp
+3 -1
View File
@@ -464,6 +464,7 @@ uint32_t EncodeFormattedStoreComponent(EmitterState& state, const Format::Buffer
spv::Op SpirvAtomicOpcode(IR::ValueOpcode opcode) {
switch (opcode) {
case IR::ValueOpcode::SharedAtomicCmpSwap64:
case IR::ValueOpcode::ImageAtomicCompareSwap32:
case IR::ValueOpcode::BufferAtomicCmpSwap32: return spv::OpAtomicCompareExchange;
case IR::ValueOpcode::ImageAtomicSwap32:
@@ -491,7 +492,8 @@ spv::Op SpirvAtomicOpcode(IR::ValueOpcode opcode) {
case IR::ValueOpcode::ImageAtomicUMax32:
case IR::ValueOpcode::ImageAtomicUMax64:
case IR::ValueOpcode::BufferAtomicUMax32:
case IR::ValueOpcode::SharedAtomicUMax32: return spv::OpAtomicUMax;
case IR::ValueOpcode::SharedAtomicUMax32:
case IR::ValueOpcode::SharedAtomicUMax64: return spv::OpAtomicUMax;
case IR::ValueOpcode::ImageAtomicAnd32:
case IR::ValueOpcode::ImageAtomicAnd64:
case IR::ValueOpcode::BufferAtomicAnd32: