shader: implement BUFFER_ATOMIC_FMAX

This commit is contained in:
nmzik
2026-08-05 02:33:26 +02:00
parent df44ad3fac
commit ac64d202af
15 changed files with 244 additions and 14 deletions
@@ -31,7 +31,8 @@ bool IsAtomic(Opcode op) {
case Opcode::AtomicAndU32:
case Opcode::AtomicOrU32:
case Opcode::AtomicXorU32:
case Opcode::AtomicFMinF32: return true;
case Opcode::AtomicFMinF32:
case Opcode::AtomicFMaxF32: return true;
default: return false;
}
}
+27 -1
View File
@@ -975,7 +975,12 @@ bool LowerControlInstruction(const Decoder::Instruction& decoded, BasicBlock& bl
return LowerControlMarker(decoded, block, Opcode::TtraceData, true, error);
case Decoder::Opcode::SInstPrefetch:
return LowerControlMarker(decoded, block, Opcode::InstPrefetch, true, error);
default: return false;
default:
if (error != nullptr) {
*error = fmt::format("control opcode has no specialized IR lowering: {}",
Decoder::OpcodeToString(decoded.opcode));
}
return false;
}
}
@@ -1064,6 +1069,18 @@ bool IsTerminatorOpcode(Decoder::Opcode opcode) {
bool LowerImplemented(const Decoder::Instruction& decoded, BasicBlock& block, std::string* error);
bool IsMemoryFamily(Decoder::Family family) {
switch (family) {
case Decoder::Family::SMEM:
case Decoder::Family::MUBUF:
case Decoder::Family::MTBUF:
case Decoder::Family::FLAT:
case Decoder::Family::DS:
case Decoder::Family::MIMG: return true;
default: return false;
}
}
bool LowerDecodedInstruction(const Decoder::Instruction& inst, BasicBlock& block,
std::string* error) {
switch (inst.opcode) {
@@ -1117,6 +1134,15 @@ bool LowerDecodedInstruction(const Decoder::Instruction& inst, BasicBlock& block
if (IsMemoryOpcode(inst.opcode)) {
return LowerMemoryInstruction(inst, block, error);
}
// Memory instructions require specialized lowering to populate MemoryInfo. Never let a
// newly decoded memory-family opcode fall through to generic register-only IR.
if (IsMemoryFamily(inst.family)) {
if (error != nullptr) {
*error = fmt::format("decoded memory-family opcode has no specialized IR lowering: {}",
Decoder::OpcodeToString(inst.opcode));
}
return false;
}
if (ScalarShiftLeftAddAmount(inst.opcode) != 0) {
return LowerScalarShiftLeftAdd(inst, block, error);
}
@@ -571,6 +571,8 @@ bool LowerMemoryInstruction(const Decoder::Instruction& decoded, BasicBlock& blo
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicXorU32, error);
case Decoder::Opcode::BufferAtomicFMin:
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicFMinF32, error);
case Decoder::Opcode::BufferAtomicFMax:
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicFMaxF32, error);
case Decoder::Opcode::FlatLoadUbyte:
case Decoder::Opcode::FlatLoadSbyte:
case Decoder::Opcode::FlatLoadUshort:
@@ -675,7 +677,12 @@ bool LowerMemoryInstruction(const Decoder::Instruction& decoded, BasicBlock& blo
case Decoder::Opcode::ImageAtomicAnd:
case Decoder::Opcode::ImageAtomicOr:
case Decoder::Opcode::ImageAtomicXor: return LowerImageAtomicU32(decoded, block, error);
default: return false;
default:
if (error != nullptr) {
*error = fmt::format("memory opcode has no specialized IR lowering: {}",
Decoder::OpcodeToString(decoded.opcode));
}
return false;
}
}
@@ -732,6 +739,7 @@ bool IsMemoryOpcode(Decoder::Opcode opcode) {
case Decoder::Opcode::BufferAtomicOr:
case Decoder::Opcode::BufferAtomicXor:
case Decoder::Opcode::BufferAtomicFMin:
case Decoder::Opcode::BufferAtomicFMax:
case Decoder::Opcode::FlatLoadUbyte:
case Decoder::Opcode::FlatLoadSbyte:
case Decoder::Opcode::FlatLoadUshort:
@@ -353,6 +353,7 @@ constexpr LowerMap LOWER_OPS[] = {
{Decoder::Opcode::BufferAtomicOr, Opcode::AtomicOrU32},
{Decoder::Opcode::BufferAtomicXor, Opcode::AtomicXorU32},
{Decoder::Opcode::BufferAtomicFMin, Opcode::AtomicFMinF32},
{Decoder::Opcode::BufferAtomicFMax, Opcode::AtomicFMaxF32},
{Decoder::Opcode::FlatLoadUbyte, Opcode::FlatLoadUbyte},
{Decoder::Opcode::FlatLoadSbyte, Opcode::FlatLoadSbyte},
{Decoder::Opcode::FlatLoadSshort, Opcode::FlatLoadSshort},
@@ -308,6 +308,7 @@ IR_OPCODE(AtomicAndU32, General)
IR_OPCODE(AtomicOrU32, General)
IR_OPCODE(AtomicXorU32, General)
IR_OPCODE(AtomicFMinF32, General)
IR_OPCODE(AtomicFMaxF32, General)
IR_OPCODE(FlatLoadUbyte, General)
IR_OPCODE(FlatLoadSbyte, General)
IR_OPCODE(FlatLoadUshort, General)