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

This commit is contained in:
nmzik committed 2026-10-13 21:33:50 +00:00
1 parent e998f21e71
commit 9e41d8320a
12 files changed
+188 -31

No files matched your search

@@ -578,11 +578,13 @@ void Translator::DS_ATOMIC(const Decoder::Instruction& inst, IR::ValueOpcode opc
bool returns_value) {
const auto memory = MemoryInfoFromDecoded(inst);
const auto address = ReadU32(inst.src0);
const IR::Value value =
memory.data_dwords == 2u ? IR::Value(ReadU64(inst.src1)) : IR::Value(ReadU32(inst.src1));
const auto read = [&](const Decoder::Operand& operand) -> IR::Value {
return memory.data_dwords == 2u ? IR::Value(ReadU64(operand)) : IR::Value(ReadU32(operand));
};
const auto value = read(inst.src1);
const auto flags = AddMemoryInfo(memory, inst.pc);
const auto result = inst.src_count == 3u
? ir.Emit(opcode, {address, value, ReadU32(inst.src2), ir.GetExec()}, flags)
? ir.Emit(opcode, {address, value, read(inst.src2), ir.GetExec()}, flags)
: ir.Emit(opcode, {address, value, ir.GetExec()}, flags);
if (returns_value) {
WriteOperand(inst.dst, result);
@@ -1073,6 +1075,10 @@ void Translator::EmitMemory(const Decoder::Instruction& inst) {
return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicUMax32, false);
case Decoder::Opcode::DS_MAX_RTN_U32:
return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicUMax32, true);
case Decoder::Opcode::DS_MAX_RTN_U64:
return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicUMax64, true);
case Decoder::Opcode::DS_CMPST_RTN_B64:
return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicCmpSwap64, true);
case Decoder::Opcode::DS_AND_B32:
return DS_ATOMIC(inst, IR::ValueOpcode::SharedAtomicAnd32, false);
case Decoder::Opcode::DS_AND_RTN_B32:
@@ -948,9 +948,14 @@ void IncludeInstructionVectorRegisters(const Decoder::Instruction& inst, uint32_
switch (inst.opcode) {
case Decoder::Opcode::DS_ADD_U64:
case Decoder::Opcode::DS_OR_B64:
case Decoder::Opcode::DS_MAX_RTN_U64:
case Decoder::Opcode::DS_WRITE_B64:
case Decoder::Opcode::DS_WRITE_B96:
case Decoder::Opcode::DS_WRITE_B128: include_vector(inst.src1, inst.data_dwords); break;
case Decoder::Opcode::DS_CMPST_RTN_B64:
include_vector(inst.src1, 2u);
include_vector(inst.src2, 2u);
break;
case Decoder::Opcode::DS_WRITE2_B32:
case Decoder::Opcode::DS_WRITE2ST64_B32:
case Decoder::Opcode::DS_WRITE2_B64: