shader: support S_QUADMASK_B32

This commit is contained in:
nmzik
2026-10-28 22:31:16 +01:00
parent 963aaeb3b6
commit 138d5b86c4
8 changed files with 69 additions and 9 deletions
@@ -70,6 +70,7 @@ constexpr OpcodeMap SOP1_OPCODE_LIST[] = {
{0x21u, Opcode::S_SWAPPC_B64},
{0x24u, Opcode::S_AND_SAVEEXEC_B64},
{0x28u, Opcode::S_ORN2_SAVEEXEC_B64},
{0x2cu, Opcode::S_QUADMASK_B32},
{0x2du, Opcode::S_QUADMASK_B64},
{0x34u, Opcode::S_ABS_I32},
{0x37u, Opcode::S_ANDN1_SAVEEXEC_B64},
@@ -517,6 +517,7 @@ std::string InstructionToString(const Instruction& inst) {
case Opcode::S_NOT_B64:
case Opcode::S_WQM_B32:
case Opcode::S_WQM_B64:
case Opcode::S_QUADMASK_B32:
case Opcode::S_QUADMASK_B64:
case Opcode::S_AND_SAVEEXEC_B32:
case Opcode::S_ORN2_SAVEEXEC_B32:
@@ -63,6 +63,7 @@ enum class Opcode {
S_FLBIT_I32,
S_FLBIT_I32_I64,
S_BITREPLICATE_B64_B32,
S_QUADMASK_B32,
S_QUADMASK_B64,
S_GETPC_B64,
S_SETPC_B64,
@@ -375,7 +375,7 @@ void Translator::S_BITREPLICATE_B64_B32(const Decoder::Instruction& inst) {
WriteOperand(DestinationOperand(inst), ir.ConstructU64(replicate(low), replicate(high)));
}
void Translator::S_QUADMASK_B64(const Decoder::Instruction& inst) {
void Translator::S_QUADMASK(const Decoder::Instruction& inst, bool is64) {
const auto compact = [&](IR::U32 value) {
auto bits = ir.BitwiseOr(value, ir.ShiftRightLogical(value, IR::U32(IR::Value(1u))));
bits = ir.BitwiseOr(bits, ir.ShiftRightLogical(bits, IR::U32(IR::Value(2u))));
@@ -388,12 +388,17 @@ void Translator::S_QUADMASK_B64(const Decoder::Instruction& inst) {
ir.BitwiseOr(bits, ir.ShiftRightLogical(bits, IR::U32(IR::Value(12u)))),
IR::U32(IR::Value(0xffu)));
};
const auto source = ReadU32Pair(inst.src0);
const auto quads = ir.BitwiseOr(
compact(source[0]), ir.ShiftLeftLogical(compact(source[1]), IR::U32(IR::Value(8u))));
const auto result = ir.ConstructU64(quads, IR::U32(IR::Value(0u)));
WriteOperand(DestinationOperand(inst), result);
ir.SetScc(IR::U1(ir.Emit(IR::ValueOpcode::INotEqual64, {result, IR::Value(uint64_t {0})})));
IR::U32 quads;
if (is64) {
const auto source = ReadU32Pair(inst.src0);
quads = ir.BitwiseOr(
compact(source[0]), ir.ShiftLeftLogical(compact(source[1]), IR::U32(IR::Value(8u))));
WriteOperand(DestinationOperand(inst), ir.ConstructU64(quads, IR::U32(IR::Value(0u))));
} else {
quads = compact(ReadU32(inst.src0));
WriteOperand(DestinationOperand(inst), quads);
}
ir.SetScc(ir.INotEqual(quads, IR::U32(IR::Value(0u))));
}
void Translator::BFM_B32(const Decoder::Instruction& inst) {
@@ -248,7 +248,8 @@ void Translator::EmitScalar(const Decoder::Instruction& inst) {
case O::S_BITSET0_B64: return S_BITSET_B64(inst, false);
case O::S_BITSET1_B64: return S_BITSET_B64(inst, true);
case O::S_BITREPLICATE_B64_B32: return S_BITREPLICATE_B64_B32(inst);
case O::S_QUADMASK_B64: return S_QUADMASK_B64(inst);
case O::S_QUADMASK_B32: return S_QUADMASK(inst, false);
case O::S_QUADMASK_B64: return S_QUADMASK(inst, true);
case O::S_BFM_B32: return BFM_B32(inst);
case O::S_BFM_B64: return S_BFM_B64(inst);
case O::S_BFE_U32: return S_BFE_U32(inst, false);
@@ -196,7 +196,7 @@ private:
void V_BCNT_U32_B32(const Decoder::Instruction& inst);
void V_MBCNT_U32_B32(const Decoder::Instruction& inst, bool low);
void S_BITREPLICATE_B64_B32(const Decoder::Instruction& inst);
void S_QUADMASK_B64(const Decoder::Instruction& inst);
void S_QUADMASK(const Decoder::Instruction& inst, bool is64);
void BFM_B32(const Decoder::Instruction& inst);
IR::U32 RightMask32(IR::U32 count);
void S_BFM_B64(const Decoder::Instruction& inst);