raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: correct DS float min/max operands and special values

Browse Source 
Co-authored-by: shadowbeat070 <70843316+shadowbeat070@users.noreply.github.com>
This commit is contained in:
nmzik
2026-10-05 03:56:42 +04:00
parent d8be245c22
commit e85279ea7b
8 changed files with 179 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h
+3
View File
@@ -463,6 +463,9 @@ void EmitAtomicMemoryBarrier(EmitterState& state, IR::ResourceKind kind);
uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t source,
bool max_value);
uint32_t EmitDsFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t source,
bool max_value);
uint32_t EmitDsSwizzleTargetLane(EmitterState& state, uint32_t subid, uint32_t control);
inline constexpr auto EmitSelectValueU32 =
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp
+4 -28
View File
@@ -1227,34 +1227,10 @@ uint32_t EmitBufferFloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst) {
void EmitSharedFloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst) {
const auto& mem = ctx.Memory(inst);
const bool max_value = inst.GetOpcode() == IR::ValueOpcode::SharedAtomicFMax32;
EmitIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
const auto index = DwordIndex(ctx, inst, mem);
const auto resource = PrepareMemoryResourceAccess(ctx.state, mem);
EmitIfCondition(
ctx.state, EmitMemoryElementInBounds(ctx.state, resource, index), [&]() {
ctx.state.builder.AddFunction(spv::OpStore, ctx.scratch_u32_variable,
ctx.Arg(inst, 1));
const auto data = ctx.state.builder.AllocateId();
ctx.state.builder.AddFunction(spv::OpLoad, TypeU32(ctx.state), data,
ctx.scratch_u32_variable);
AtomicUpdate(
ctx.state, EmitMemoryElementPointer(ctx.state, resource, index),
mem.kind, [&](uint32_t old) {
const auto old_f =
Unary(ctx.state, spv::OpBitcast, TypeF32(ctx.state), old);
const auto compare_f =
Unary(ctx.state, spv::OpBitcast, TypeF32(ctx.state), ctx.Arg(inst, 2));
const auto data_f =
Unary(ctx.state, spv::OpBitcast, TypeF32(ctx.state), data);
const auto compare = Binary(
ctx.state, max_value ? spv::OpFOrdGreaterThan : spv::OpFOrdLessThan,
TypeBool(ctx.state), max_value ? old_f : compare_f,
max_value ? compare_f : old_f);
return Unary(ctx.state, spv::OpBitcast, TypeU32(ctx.state),
Select(ctx.state, TypeF32(ctx.state), compare, data_f, old_f));
});
});
});
EmitAtomicUpdate(ctx, inst, mem,
[max_value](EmitterState& state, uint32_t old, uint32_t value) {
return EmitDsFloatAtomicReplacement(state, old, value, max_value);
});
}
uint32_t EmitAppendConsume(ValueEmitContext& ctx, const IR::Inst& inst) {
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemoryHelpers.cpp
+54 -9
View File
@@ -438,6 +438,20 @@ void EmitAtomicMemoryBarrier(EmitterState& state, IR::ResourceKind kind) {
ConstantU32(state, spv::MemorySemanticsAcquireReleaseMask | memory));
}
namespace {
uint32_t FloatOrderKey(EmitterState& state, uint32_t bits) {
const auto negative = EmitCompareU32Constant(state, spv::OpINotEqual,
EmitAndConstant(state, bits, 0x80000000u), 0u);
const auto negative_key = state.builder.AllocateId();
state.builder.AddFunction(spv::OpNot, TypeU32(state), negative_key, bits);
const auto positive_key =
EmitBinaryU32(state, spv::OpBitwiseXor, bits, ConstantU32(state, 0x80000000u));
return EmitSelectValueU32(state, negative, negative_key, positive_key);
}
} // namespace
uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t source,
bool max_value) {
struct OrderedBits {
@@ -447,15 +461,7 @@ uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t
};
const auto classify = [&](uint32_t bits) {
const auto cls = EmitClassifyF32Bits(state, bits);
const auto negative = EmitCompareU32Constant(state, spv::OpINotEqual,
EmitAndConstant(state, bits, 0x80000000u), 0u);
const auto negative_key = state.builder.AllocateId();
state.builder.AddFunction(spv::OpNot, TypeU32(state), negative_key, bits);
const auto positive_key =
EmitBinaryU32(state, spv::OpBitwiseXor, bits, ConstantU32(state, 0x80000000u));
return OrderedBits {
cls.nan, cls.zero,
EmitSelectValueU32(state, negative, negative_key, positive_key)};
return OrderedBits {cls.nan, cls.zero, FloatOrderKey(state, bits)};
};
const auto source_class = classify(source);
const auto old_class = classify(old);
@@ -470,6 +476,45 @@ uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t
old);
}
uint32_t EmitDsFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t source,
bool max_value) {
struct OrderedBits {
uint32_t nan;
uint32_t signaling_nan;
uint32_t key;
};
const auto classify = [&](uint32_t bits) {
const auto abs = EmitAndConstant(state, bits, 0x7fffffffu);
const auto nan = EmitCompareU32Constant(state, spv::OpUGreaterThan, abs, 0x7f800000u);
const auto signaling_nan = EmitLogicalAndBool(
state, nan, EmitCompareU32Constant(state, spv::OpIEqual,
EmitAndConstant(state, bits, 0x00400000u), 0u));
const auto small = EmitCompareU32Constant(state, spv::OpULessThan, abs, 0x00800000u);
const auto sign = EmitAndConstant(state, bits, 0x80000000u);
const auto positive_denorm = EmitLogicalAndBool(
state, EmitCompareU32Constant(state, spv::OpIEqual, sign, 0u),
EmitCompareU32Constant(state, spv::OpINotEqual, abs, 0u));
// Negative denorms tie with -0; positive denorms tie only with each other.
const auto small_bits = EmitSelectValueU32(
state, positive_denorm, ConstantU32(state, 1u), sign);
const auto key_bits = EmitSelectValueU32(state, small, small_bits, bits);
return OrderedBits {nan, signaling_nan, FloatOrderKey(state, key_bits)};
};
const auto lhs = classify(old);
const auto rhs = classify(source);
const auto choose_source = state.builder.AllocateId();
state.builder.AddFunction(max_value ? spv::OpUGreaterThan : spv::OpULessThanEqual,
TypeBool(state), choose_source, rhs.key, lhs.key);
// Keep original bits: ordering denorms does not flush the selected operand.
auto result = EmitSelectValueU32(state, choose_source, source, old);
result = EmitSelectValueU32(state, lhs.nan, source, result);
result = EmitSelectValueU32(state, rhs.nan, old, result);
result = EmitSelectValueU32(state, rhs.signaling_nan,
EmitOrU32(state, source, ConstantU32(state, 0x00400000u)), result);
return EmitSelectValueU32(state, lhs.signaling_nan,
EmitOrU32(state, old, ConstantU32(state, 0x00400000u)), result);
}
uint32_t EmitDsSwizzleTargetLane(EmitterState& state, uint32_t subid, uint32_t control) {
if ((control & 0xc000u) == 0xc000u) {
const uint32_t mask = control & 0x1fu;
src/graphics/shader/recompiler/backend/spirv/spirvEmitterProgram.cpp
+1 -3
View File
@@ -689,9 +689,7 @@ void EmitProgram(EmitterState& state) {
DefineBvhIntersect(state);
for (const auto* block: program.blocks) {
if (std::ranges::any_of(*block, [](const IR::Inst& inst) {
return inst.GetOpcode() == IR::ValueOpcode::SwizzleU32 ||
inst.GetOpcode() == IR::ValueOpcode::SharedAtomicFMin32 ||
inst.GetOpcode() == IR::ValueOpcode::SharedAtomicFMax32;
return inst.GetOpcode() == IR::ValueOpcode::SwizzleU32;
})) {
ctx.scratch_u32_variable = state.builder.AllocateId();
if (state.lane_count == 2) {
src/graphics/shader/recompiler/frontend/decode/MemoryOps.cpp
+2 -2
View File
@@ -205,9 +205,9 @@ uint32_t DsSourceCount(Opcode opcode) {
case Opcode::DS_WRITE2ST64_B32:
case Opcode::DS_WRITE2_B64:
case Opcode::DS_WRITE2ST64_B64:
case Opcode::DS_MSKOR_B32:
case Opcode::DS_MSKOR_B32: return 3u;
case Opcode::DS_MIN_F32:
case Opcode::DS_MAX_F32: return 3u;
case Opcode::DS_MAX_F32:
case Opcode::DS_PERMUTE_B32:
case Opcode::DS_BPERMUTE_B32: return 2u;
case Opcode::DS_READ_ADDTID_B32:
src/graphics/shader/recompiler/ir/opcodes/ValueOpcodes.inc
+2 -2
View File
@@ -280,8 +280,8 @@ VALUE_OPCODE(WriteSharedU32, Void, U32, U32, U1)
VALUE_OPCODE(WriteSharedU32x2, Void, U32, U32, U32, U1)
VALUE_OPCODE(WriteSharedU32x3, Void, U32, U32, U32, U32, U1)
VALUE_OPCODE(WriteSharedU32x4, Void, U32, U32, U32, U32, U32, U1)
VALUE_OPCODE(SharedAtomicFMin32, Void, U32, U32, U32, U1)
VALUE_OPCODE(SharedAtomicFMax32, Void, U32, U32, U32, U1)
VALUE_OPCODE(SharedAtomicFMin32, Void, U32, U32, U1)
VALUE_OPCODE(SharedAtomicFMax32, Void, U32, U32, U1)
VALUE_OPCODE(SharedAtomicSwap32, U32, U32, U32, U1)
VALUE_OPCODE(SharedAtomicIAdd32, U32, U32, U32, U1)
VALUE_OPCODE(SharedAtomicIAdd64, Void, U32, U64, U1)