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

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
@@ -116,7 +116,8 @@ constexpr MemoryOpcodeInfo DS_OPCODE_LIST[] = {
{0x3fu, Opcode::DS_ORDERED_COUNT, 1, 32},
{0x40u, Opcode::DS_ADD_U64, 2, 32}, {0x4au, Opcode::DS_OR_B64, 2, 32},
{0x4du, Opcode::DS_WRITE_B64, 2, 32}, {0x4eu, Opcode::DS_WRITE2_B64, 4, 32},
{0x4fu, Opcode::DS_WRITE2ST64_B64, 4, 32}, {0x76u, Opcode::DS_READ_B64, 2, 32},
{0x4fu, Opcode::DS_WRITE2ST64_B64, 4, 32}, {0x68u, Opcode::DS_MAX_RTN_U64, 2, 32},
{0x70u, Opcode::DS_CMPST_RTN_B64, 2, 32}, {0x76u, Opcode::DS_READ_B64, 2, 32},
{0x77u, Opcode::DS_READ2_B64, 4, 32}, {0x78u, Opcode::DS_READ2ST64_B64, 4, 32},
{0xa0u, Opcode::DS_WRITE_B8_D16_HI, 1, 8},
{0xa1u, Opcode::DS_WRITE_B16_D16_HI, 1, 16},
@@ -192,6 +193,8 @@ bool IsDsAtomicOpcode(Opcode opcode) {
case Opcode::DS_MIN_RTN_U32:
case Opcode::DS_MAX_U32:
case Opcode::DS_MAX_RTN_U32:
case Opcode::DS_MAX_RTN_U64:
case Opcode::DS_CMPST_RTN_B64:
case Opcode::DS_AND_B32:
case Opcode::DS_AND_RTN_B32:
case Opcode::DS_OR_B32:
@@ -211,6 +214,7 @@ uint32_t DsSourceCount(Opcode opcode) {
case Opcode::DS_WRITE2ST64_B32:
case Opcode::DS_WRITE2_B64:
case Opcode::DS_WRITE2ST64_B64:
case Opcode::DS_CMPST_RTN_B64:
case Opcode::DS_MSKOR_B32: return 3u;
case Opcode::DS_MIN_F32:
case Opcode::DS_MAX_F32:
@@ -708,6 +708,8 @@ std::string InstructionToString(const Instruction& inst) {
case Opcode::DS_MIN_RTN_U32:
case Opcode::DS_MAX_U32:
case Opcode::DS_MAX_RTN_U32:
case Opcode::DS_MAX_RTN_U64:
case Opcode::DS_CMPST_RTN_B64:
case Opcode::DS_AND_B32:
case Opcode::DS_AND_RTN_B32:
case Opcode::DS_OR_B32:
@@ -624,6 +624,8 @@ enum class Opcode {
DS_MIN_RTN_U32,
DS_MAX_U32,
DS_MAX_RTN_U32,
DS_MAX_RTN_U64,
DS_CMPST_RTN_B64,
DS_AND_B32,
DS_AND_RTN_B32,
DS_OR_B32,
@@ -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: