Implement RDNA2 64-bit image atomic maximum

Co-authored-by: shadowbeat070 <70843316+shadowbeat070@users.noreply.github.com>
This commit is contained in:
BlackCAT
2026-10-03 06:08:05 +04:00
committed by nmzik
parent e6cb880adb
commit ef4c5baf3a
19 changed files with 244 additions and 57 deletions
@@ -360,9 +360,19 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
SetUnsupported(inst, Family::MIMG, opcode,
"MIMG image gather requires exactly one dmask bit");
}
if (inst.opcode == Opcode::IMAGE_ATOMIC_CMPSWAP && inst.dmask != 0x3u) {
SetUnsupported(inst, Family::MIMG, opcode,
"MIMG image compare-and-swap requires 32-bit DMASK 0x3");
if (atomic != nullptr) {
const bool compare_swap = inst.opcode == Opcode::IMAGE_ATOMIC_CMPSWAP;
const auto mask32 = compare_swap ? 0x3u : 0x1u;
const auto mask64 = compare_swap ? 0xfu : 0x3u;
if (inst.dmask != mask32 && inst.dmask != mask64) {
SetUnsupported(inst, Family::MIMG, opcode, "MIMG image atomic has invalid DMASK");
} else if (inst.dmask == mask64) {
inst.data_bits = 64u;
if (inst.opcode != Opcode::IMAGE_ATOMIC_UMAX) {
SetUnsupported(inst, Family::MIMG, opcode,
"MIMG 64-bit image atomic opcode is not implemented");
}
}
}
const bool supports_d16 = sample != nullptr || gather != nullptr || opcode == 0x00u ||
opcode == 0x01u || opcode == 0x08u || opcode == 0x09u;
@@ -568,7 +568,9 @@ void Translator::IMAGE_ATOMIC(const Decoder::Instruction& inst, IR::ValueOpcode
ir.GetExec()},
flags);
} else {
result = ir.Emit(opcode, {resource, address, ReadU32(data), ir.GetExec()}, flags);
const IR::Value value =
memory.data_bits == 64u ? IR::Value(ReadU64(data)) : IR::Value(ReadU32(data));
result = ir.Emit(opcode, {resource, address, value, ir.GetExec()}, flags);
}
if (inst.glc) {
WriteOperand(inst.dst, result);
@@ -1052,7 +1054,8 @@ void Translator::EmitMemory(const Decoder::Instruction& inst) {
case Decoder::Opcode::IMAGE_ATOMIC_SMAX:
return IMAGE_ATOMIC(inst, IR::ValueOpcode::ImageAtomicSMax32);
case Decoder::Opcode::IMAGE_ATOMIC_UMAX:
return IMAGE_ATOMIC(inst, IR::ValueOpcode::ImageAtomicUMax32);
return IMAGE_ATOMIC(inst, inst.data_bits == 64u ? IR::ValueOpcode::ImageAtomicUMax64
: IR::ValueOpcode::ImageAtomicUMax32);
case Decoder::Opcode::IMAGE_ATOMIC_AND:
return IMAGE_ATOMIC(inst, IR::ValueOpcode::ImageAtomicAnd32);
case Decoder::Opcode::IMAGE_ATOMIC_OR: