mirror of
https://github.com/KytyPS5/KytyPS5.git
synced 2026-10-26 21:06:53 +01:00
shader: use ISA-style decoder opcode names
This commit is contained in:
@@ -71,10 +71,10 @@ bool AppendScalarResultSccNonZero(const Decoder::Instruction& decoded, BasicBloc
|
||||
|
||||
uint32_t VectorByteConvertIndex(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::VCvtF32Ubyte0: return 0;
|
||||
case Decoder::Opcode::VCvtF32Ubyte1: return 1;
|
||||
case Decoder::Opcode::VCvtF32Ubyte2: return 2;
|
||||
case Decoder::Opcode::VCvtF32Ubyte3: return 3;
|
||||
case Decoder::Opcode::V_CVT_F32_UBYTE0: return 0;
|
||||
case Decoder::Opcode::V_CVT_F32_UBYTE1: return 1;
|
||||
case Decoder::Opcode::V_CVT_F32_UBYTE2: return 2;
|
||||
case Decoder::Opcode::V_CVT_F32_UBYTE3: return 3;
|
||||
default: return 0xffffffffu;
|
||||
}
|
||||
}
|
||||
@@ -228,8 +228,8 @@ MemoryInfo MemoryInfoFromDecoded(const Decoder::Instruction& decoded, ResourceKi
|
||||
mem.data_signed = decoded.data_signed;
|
||||
mem.typed = decoded.typed;
|
||||
mem.formatted = decoded.formatted;
|
||||
mem.image_has_mip = decoded.opcode == Decoder::Opcode::ImageLoadMip ||
|
||||
decoded.opcode == Decoder::Opcode::ImageStoreMip;
|
||||
mem.image_has_mip = decoded.opcode == Decoder::Opcode::IMAGE_LOAD_MIP ||
|
||||
decoded.opcode == Decoder::Opcode::IMAGE_STORE_MIP;
|
||||
mem.glc = decoded.glc;
|
||||
mem.slc = decoded.slc;
|
||||
mem.idxen = decoded.idxen;
|
||||
@@ -451,10 +451,10 @@ bool LowerScalarSelect64(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
|
||||
uint32_t ScalarShiftLeftAddAmount(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SLshl1AddU32: return 1;
|
||||
case Decoder::Opcode::SLshl2AddU32: return 2;
|
||||
case Decoder::Opcode::SLshl3AddU32: return 3;
|
||||
case Decoder::Opcode::SLshl4AddU32: return 4;
|
||||
case Decoder::Opcode::S_LSHL1_ADD_U32: return 1;
|
||||
case Decoder::Opcode::S_LSHL2_ADD_U32: return 2;
|
||||
case Decoder::Opcode::S_LSHL3_ADD_U32: return 3;
|
||||
case Decoder::Opcode::S_LSHL4_ADD_U32: return 4;
|
||||
default: return 0;
|
||||
}
|
||||
}
|
||||
@@ -484,20 +484,20 @@ bool LowerScalarShiftLeftAdd(const Decoder::Instruction& decoded, BasicBlock& bl
|
||||
|
||||
Opcode ScalarMinMaxSccCompareOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SMinI32: return Opcode::CompareLtI32;
|
||||
case Decoder::Opcode::SMaxI32: return Opcode::CompareGtI32;
|
||||
case Decoder::Opcode::SMinU32: return Opcode::CompareLtU32;
|
||||
case Decoder::Opcode::SMaxU32: return Opcode::CompareGtU32;
|
||||
case Decoder::Opcode::S_MIN_I32: return Opcode::CompareLtI32;
|
||||
case Decoder::Opcode::S_MAX_I32: return Opcode::CompareGtI32;
|
||||
case Decoder::Opcode::S_MIN_U32: return Opcode::CompareLtU32;
|
||||
case Decoder::Opcode::S_MAX_U32: return Opcode::CompareGtU32;
|
||||
default: return Opcode::CompareFalse;
|
||||
}
|
||||
}
|
||||
|
||||
bool IsScalarMinMaxOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SMinI32:
|
||||
case Decoder::Opcode::SMaxI32:
|
||||
case Decoder::Opcode::SMinU32:
|
||||
case Decoder::Opcode::SMaxU32: return true;
|
||||
case Decoder::Opcode::S_MIN_I32:
|
||||
case Decoder::Opcode::S_MAX_I32:
|
||||
case Decoder::Opcode::S_MIN_U32:
|
||||
case Decoder::Opcode::S_MAX_U32: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
@@ -507,7 +507,7 @@ bool ResolveIrOpcode(Decoder::Opcode decoded, Opcode& lowered, std::string* erro
|
||||
if (!opcode.has_value()) {
|
||||
if (error != nullptr) {
|
||||
*error = fmt::format("decoded opcode has no IR lowering: {}",
|
||||
Decoder::OpcodeToString(decoded));
|
||||
magic_enum::enum_name(decoded));
|
||||
}
|
||||
return false;
|
||||
}
|
||||
@@ -763,17 +763,19 @@ bool LowerVectorCarryOut(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
Instruction inst;
|
||||
inst.pc = decoded.pc;
|
||||
inst.op =
|
||||
decoded.opcode == Decoder::Opcode::VAddI32 ? Opcode::IAddCarryU32 : Opcode::ISubBorrowU32;
|
||||
inst.src_count = decoded.opcode == Decoder::Opcode::VAddI32 ? 3u : 2u;
|
||||
const auto& src0 = decoded.opcode == Decoder::Opcode::VSubrevI32 ? decoded.src1 : decoded.src0;
|
||||
const auto& src1 = decoded.opcode == Decoder::Opcode::VSubrevI32 ? decoded.src0 : decoded.src1;
|
||||
decoded.opcode == Decoder::Opcode::V_ADD_I32 ? Opcode::IAddCarryU32 : Opcode::ISubBorrowU32;
|
||||
inst.src_count = decoded.opcode == Decoder::Opcode::V_ADD_I32 ? 3u : 2u;
|
||||
const auto& src0 =
|
||||
decoded.opcode == Decoder::Opcode::V_SUBREV_I32 ? decoded.src1 : decoded.src0;
|
||||
const auto& src1 =
|
||||
decoded.opcode == Decoder::Opcode::V_SUBREV_I32 ? decoded.src0 : decoded.src1;
|
||||
if (!LowerRegisterOperand(decoded.dst, inst.dst, error) ||
|
||||
!LowerRegisterOperand(decoded.dst2, inst.dst2, error) ||
|
||||
!LowerSourceOperand(src0, inst.src[0], error) ||
|
||||
!LowerSourceOperand(src1, inst.src[1], error)) {
|
||||
return false;
|
||||
}
|
||||
if (decoded.opcode == Decoder::Opcode::VAddI32) {
|
||||
if (decoded.opcode == Decoder::Opcode::V_ADD_I32) {
|
||||
inst.src[2] = MakeImmediateU32(0);
|
||||
}
|
||||
block.instructions.push_back(inst);
|
||||
@@ -892,7 +894,7 @@ bool LowerVInterpP1F32(const Decoder::Instruction& decoded, BasicBlock& block) {
|
||||
|
||||
bool LowerVInterpLoadF32(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
std::string* error) {
|
||||
if (decoded.opcode == Decoder::Opcode::VInterpMovF32 && decoded.src0.value != 2u) {
|
||||
if (decoded.opcode == Decoder::Opcode::V_INTERP_MOV_F32 && decoded.src0.value != 2u) {
|
||||
if (error != nullptr) {
|
||||
*error = fmt::format("v_interp_mov_f32 mode {} is not implemented at pc 0x{:08x}",
|
||||
decoded.src0.value, decoded.pc);
|
||||
@@ -989,29 +991,29 @@ bool LowerControlMarker(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
bool LowerControlInstruction(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
std::string* error) {
|
||||
switch (decoded.opcode) {
|
||||
case Decoder::Opcode::SNop:
|
||||
case Decoder::Opcode::S_NOP:
|
||||
return LowerControlMarker(decoded, block, Opcode::ControlNop, true, error);
|
||||
case Decoder::Opcode::SWaitcnt:
|
||||
case Decoder::Opcode::SWaitcntDepctr:
|
||||
case Decoder::Opcode::S_WAITCNT:
|
||||
case Decoder::Opcode::S_WAITCNT_DEPCTR:
|
||||
return LowerControlMarker(decoded, block, Opcode::Waitcnt, true, error);
|
||||
case Decoder::Opcode::SBarrier:
|
||||
case Decoder::Opcode::S_BARRIER:
|
||||
return LowerControlMarker(decoded, block, Opcode::Barrier, false, error);
|
||||
case Decoder::Opcode::SSendmsg:
|
||||
case Decoder::Opcode::S_SENDMSG:
|
||||
return LowerControlMarker(decoded, block, Opcode::Sendmsg, true, error);
|
||||
case Decoder::Opcode::SSetregB32:
|
||||
case Decoder::Opcode::SSleep:
|
||||
case Decoder::Opcode::S_SETREG_B32:
|
||||
case Decoder::Opcode::S_SLEEP:
|
||||
return LowerControlMarker(decoded, block, Opcode::ControlNop, true, error);
|
||||
case Decoder::Opcode::STrap:
|
||||
case Decoder::Opcode::S_TRAP:
|
||||
// RDNA2 converts S_TRAP to a NOP when the unmodeled trap state is disabled.
|
||||
return LowerControlMarker(decoded, block, Opcode::ControlNop, true, error);
|
||||
case Decoder::Opcode::STtraceData:
|
||||
case Decoder::Opcode::S_TTRACEDATA:
|
||||
return LowerControlMarker(decoded, block, Opcode::TtraceData, true, error);
|
||||
case Decoder::Opcode::SInstPrefetch:
|
||||
case Decoder::Opcode::S_INST_PREFETCH:
|
||||
return LowerControlMarker(decoded, block, Opcode::InstPrefetch, true, error);
|
||||
default:
|
||||
if (error != nullptr) {
|
||||
*error = fmt::format("control opcode has no specialized IR lowering: {}",
|
||||
Decoder::OpcodeToString(decoded.opcode));
|
||||
magic_enum::enum_name(decoded.opcode));
|
||||
}
|
||||
return false;
|
||||
}
|
||||
@@ -1036,20 +1038,20 @@ bool LowerScalarGetpcB64(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
|
||||
bool IsScalarSaveexecOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SAndSaveexecB32:
|
||||
case Decoder::Opcode::SAndn1SaveexecB32:
|
||||
case Decoder::Opcode::SAndSaveexecB64:
|
||||
case Decoder::Opcode::SOrn2SaveexecB64:
|
||||
case Decoder::Opcode::SAndn1SaveexecB64: return true;
|
||||
case Decoder::Opcode::S_AND_SAVEEXEC_B32:
|
||||
case Decoder::Opcode::S_ANDN1_SAVEEXEC_B32:
|
||||
case Decoder::Opcode::S_AND_SAVEEXEC_B64:
|
||||
case Decoder::Opcode::S_ORN2_SAVEEXEC_B64:
|
||||
case Decoder::Opcode::S_ANDN1_SAVEEXEC_B64: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
|
||||
SaveexecMode ScalarSaveexecMode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SOrn2SaveexecB64: return SaveexecMode::Orn2;
|
||||
case Decoder::Opcode::SAndn1SaveexecB32:
|
||||
case Decoder::Opcode::SAndn1SaveexecB64: return SaveexecMode::Andn1;
|
||||
case Decoder::Opcode::S_ORN2_SAVEEXEC_B64: return SaveexecMode::Orn2;
|
||||
case Decoder::Opcode::S_ANDN1_SAVEEXEC_B32:
|
||||
case Decoder::Opcode::S_ANDN1_SAVEEXEC_B64: return SaveexecMode::Andn1;
|
||||
default: return SaveexecMode::And;
|
||||
}
|
||||
}
|
||||
@@ -1058,8 +1060,8 @@ bool LowerScalarSaveexec(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
std::string* error) {
|
||||
Instruction inst;
|
||||
inst.pc = decoded.pc;
|
||||
inst.op = (decoded.opcode == Decoder::Opcode::SAndSaveexecB32 ||
|
||||
decoded.opcode == Decoder::Opcode::SAndn1SaveexecB32)
|
||||
inst.op = (decoded.opcode == Decoder::Opcode::S_AND_SAVEEXEC_B32 ||
|
||||
decoded.opcode == Decoder::Opcode::S_ANDN1_SAVEEXEC_B32)
|
||||
? Opcode::SaveexecB32
|
||||
: Opcode::SaveexecB64;
|
||||
inst.saveexec_mode = ScalarSaveexecMode(decoded.opcode);
|
||||
@@ -1074,31 +1076,31 @@ bool LowerScalarSaveexec(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
|
||||
bool IsControlOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SNop:
|
||||
case Decoder::Opcode::SWaitcnt:
|
||||
case Decoder::Opcode::SWaitcntDepctr:
|
||||
case Decoder::Opcode::SBarrier:
|
||||
case Decoder::Opcode::SSendmsg:
|
||||
case Decoder::Opcode::SSetregB32:
|
||||
case Decoder::Opcode::SSleep:
|
||||
case Decoder::Opcode::STrap:
|
||||
case Decoder::Opcode::STtraceData:
|
||||
case Decoder::Opcode::SInstPrefetch: return true;
|
||||
case Decoder::Opcode::S_NOP:
|
||||
case Decoder::Opcode::S_WAITCNT:
|
||||
case Decoder::Opcode::S_WAITCNT_DEPCTR:
|
||||
case Decoder::Opcode::S_BARRIER:
|
||||
case Decoder::Opcode::S_SENDMSG:
|
||||
case Decoder::Opcode::S_SETREG_B32:
|
||||
case Decoder::Opcode::S_SLEEP:
|
||||
case Decoder::Opcode::S_TRAP:
|
||||
case Decoder::Opcode::S_TTRACEDATA:
|
||||
case Decoder::Opcode::S_INST_PREFETCH: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
|
||||
bool IsTerminatorOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SEndpgm:
|
||||
case Decoder::Opcode::SBranch:
|
||||
case Decoder::Opcode::SCbranchScc0:
|
||||
case Decoder::Opcode::SCbranchScc1:
|
||||
case Decoder::Opcode::SCbranchVccz:
|
||||
case Decoder::Opcode::SCbranchVccnz:
|
||||
case Decoder::Opcode::SCbranchExecz:
|
||||
case Decoder::Opcode::SCbranchExecnz:
|
||||
case Decoder::Opcode::SSetpcB64: return true;
|
||||
case Decoder::Opcode::S_ENDPGM:
|
||||
case Decoder::Opcode::S_BRANCH:
|
||||
case Decoder::Opcode::S_CBRANCH_SCC0:
|
||||
case Decoder::Opcode::S_CBRANCH_SCC1:
|
||||
case Decoder::Opcode::S_CBRANCH_VCCZ:
|
||||
case Decoder::Opcode::S_CBRANCH_VCCNZ:
|
||||
case Decoder::Opcode::S_CBRANCH_EXECZ:
|
||||
case Decoder::Opcode::S_CBRANCH_EXECNZ:
|
||||
case Decoder::Opcode::S_SETPC_B64: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
@@ -1120,43 +1122,44 @@ bool IsMemoryFamily(Decoder::Family family) {
|
||||
bool LowerDecodedInstruction(const Decoder::Instruction& inst, BasicBlock& block,
|
||||
std::string* error) {
|
||||
switch (inst.opcode) {
|
||||
case Decoder::Opcode::Unsupported:
|
||||
case Decoder::Opcode::UNSUPPORTED:
|
||||
if (error != nullptr) {
|
||||
*error = fmt::format("unsupported decoded instruction: {}",
|
||||
Decoder::InstructionToString(inst).c_str());
|
||||
}
|
||||
return false;
|
||||
case Decoder::Opcode::SGetpcB64: return LowerScalarGetpcB64(inst, block, error);
|
||||
case Decoder::Opcode::VNop: return true;
|
||||
case Decoder::Opcode::Exp: return LowerExportInstruction(inst, block, error);
|
||||
case Decoder::Opcode::SCselectB32: return LowerScalarSelect(inst, block, error);
|
||||
case Decoder::Opcode::SCselectB64: return LowerScalarSelect64(inst, block, error);
|
||||
case Decoder::Opcode::SAddU32: return LowerScalarAddCarry(inst, block, false, error);
|
||||
case Decoder::Opcode::SAddcU32: return LowerScalarAddCarry(inst, block, true, error);
|
||||
case Decoder::Opcode::SSubbU32: return LowerScalarSubBorrowCarry(inst, block, error);
|
||||
case Decoder::Opcode::SSubU32:
|
||||
case Decoder::Opcode::S_GETPC_B64: return LowerScalarGetpcB64(inst, block, error);
|
||||
case Decoder::Opcode::V_NOP: return true;
|
||||
case Decoder::Opcode::EXP: return LowerExportInstruction(inst, block, error);
|
||||
case Decoder::Opcode::S_CSELECT_B32: return LowerScalarSelect(inst, block, error);
|
||||
case Decoder::Opcode::S_CSELECT_B64: return LowerScalarSelect64(inst, block, error);
|
||||
case Decoder::Opcode::S_ADD_U32: return LowerScalarAddCarry(inst, block, false, error);
|
||||
case Decoder::Opcode::S_ADDC_U32: return LowerScalarAddCarry(inst, block, true, error);
|
||||
case Decoder::Opcode::S_SUBB_U32: return LowerScalarSubBorrowCarry(inst, block, error);
|
||||
case Decoder::Opcode::S_SUB_U32:
|
||||
return LowerScalarBinaryScc(inst, block, Opcode::ScalarSubBorrowU32, error);
|
||||
case Decoder::Opcode::SBitset0B32: return LowerScalarBitset0B32(inst, block, error);
|
||||
case Decoder::Opcode::SBitset1B32: return LowerScalarBitset1B32(inst, block, error);
|
||||
case Decoder::Opcode::SAddI32:
|
||||
case Decoder::Opcode::S_BITSET0_B32: return LowerScalarBitset0B32(inst, block, error);
|
||||
case Decoder::Opcode::S_BITSET1_B32: return LowerScalarBitset1B32(inst, block, error);
|
||||
case Decoder::Opcode::S_ADD_I32:
|
||||
return LowerScalarBinaryScc(inst, block, Opcode::ScalarSignedAddOverflowI32, error);
|
||||
case Decoder::Opcode::SSubI32:
|
||||
case Decoder::Opcode::S_SUB_I32:
|
||||
return LowerScalarBinaryScc(inst, block, Opcode::ScalarSignedSubOverflowI32, error);
|
||||
case Decoder::Opcode::VCndmaskB32: return LowerVectorCndmask(inst, block, error);
|
||||
case Decoder::Opcode::VMovB32: return LowerVectorMoveB32(inst, block, error);
|
||||
case Decoder::Opcode::VMovreldB32: return LowerVectorMoveRelDestination(inst, block, error);
|
||||
case Decoder::Opcode::VMovrelsB32: return LowerVectorMoveRelSource(inst, block, error);
|
||||
case Decoder::Opcode::VAddcU32: return LowerVectorAddCarry(inst, block, error);
|
||||
case Decoder::Opcode::VSubrevCoCiU32:
|
||||
case Decoder::Opcode::V_CNDMASK_B32: return LowerVectorCndmask(inst, block, error);
|
||||
case Decoder::Opcode::V_MOV_B32: return LowerVectorMoveB32(inst, block, error);
|
||||
case Decoder::Opcode::V_MOVRELD_B32:
|
||||
return LowerVectorMoveRelDestination(inst, block, error);
|
||||
case Decoder::Opcode::V_MOVRELS_B32: return LowerVectorMoveRelSource(inst, block, error);
|
||||
case Decoder::Opcode::V_ADDC_U32: return LowerVectorAddCarry(inst, block, error);
|
||||
case Decoder::Opcode::V_SUBREV_CO_CI_U32:
|
||||
return LowerVectorSubrevBorrowCarry(inst, block, error);
|
||||
case Decoder::Opcode::VMadU64U32: return LowerVectorMadU64U32(inst, block, error);
|
||||
case Decoder::Opcode::VMacF32: return LowerVectorMacF32(inst, block, error);
|
||||
case Decoder::Opcode::VPkFmacF16: return LowerVectorPkFmacF16(inst, block, error);
|
||||
case Decoder::Opcode::VFmacF16: return LowerVectorFmacF16(inst, block, error);
|
||||
case Decoder::Opcode::VDot2cF32F16: return LowerVectorDot2cF32F16(inst, block, error);
|
||||
case Decoder::Opcode::VInterpP1F32: return LowerVInterpP1F32(inst, block);
|
||||
case Decoder::Opcode::VInterpP2F32:
|
||||
case Decoder::Opcode::VInterpMovF32: return LowerVInterpLoadF32(inst, block, error);
|
||||
case Decoder::Opcode::V_MAD_U64_U32: return LowerVectorMadU64U32(inst, block, error);
|
||||
case Decoder::Opcode::V_MAC_F32: return LowerVectorMacF32(inst, block, error);
|
||||
case Decoder::Opcode::V_PK_FMAC_F16: return LowerVectorPkFmacF16(inst, block, error);
|
||||
case Decoder::Opcode::V_FMAC_F16: return LowerVectorFmacF16(inst, block, error);
|
||||
case Decoder::Opcode::V_DOT2C_F32_F16: return LowerVectorDot2cF32F16(inst, block, error);
|
||||
case Decoder::Opcode::V_INTERP_P1_F32: return LowerVInterpP1F32(inst, block);
|
||||
case Decoder::Opcode::V_INTERP_P2_F32:
|
||||
case Decoder::Opcode::V_INTERP_MOV_F32: return LowerVInterpLoadF32(inst, block, error);
|
||||
default: break;
|
||||
}
|
||||
|
||||
@@ -1177,7 +1180,7 @@ bool LowerDecodedInstruction(const Decoder::Instruction& inst, BasicBlock& block
|
||||
if (IsMemoryFamily(inst.family)) {
|
||||
if (error != nullptr) {
|
||||
*error = fmt::format("decoded memory-family opcode has no specialized IR lowering: {}",
|
||||
Decoder::OpcodeToString(inst.opcode));
|
||||
magic_enum::enum_name(inst.opcode));
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
@@ -60,17 +60,17 @@ Opcode DsWriteOpcode(uint32_t data_bits) {
|
||||
|
||||
Opcode LoweredImageOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::ImageGetResinfo: return Opcode::ImageGetResinfo;
|
||||
case Decoder::Opcode::ImageGetLod: return Opcode::ImageGetLod;
|
||||
case Decoder::Opcode::ImageLoad:
|
||||
case Decoder::Opcode::ImageLoadMip: return Opcode::ImageLoad;
|
||||
case Decoder::Opcode::ImageGather4Lz:
|
||||
case Decoder::Opcode::ImageGather4C:
|
||||
case Decoder::Opcode::ImageGather4CLz:
|
||||
case Decoder::Opcode::ImageGather4LzO:
|
||||
case Decoder::Opcode::ImageGather4CO:
|
||||
case Decoder::Opcode::ImageGather4CLzO:
|
||||
case Decoder::Opcode::ImageGather4H: return Opcode::ImageGather4;
|
||||
case Decoder::Opcode::IMAGE_GET_RESINFO: return Opcode::ImageGetResinfo;
|
||||
case Decoder::Opcode::IMAGE_GET_LOD: return Opcode::ImageGetLod;
|
||||
case Decoder::Opcode::IMAGE_LOAD:
|
||||
case Decoder::Opcode::IMAGE_LOAD_MIP: return Opcode::ImageLoad;
|
||||
case Decoder::Opcode::IMAGE_GATHER4_LZ:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_LZ:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_LZ_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_LZ_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4H: return Opcode::ImageGather4;
|
||||
default: return Opcode::ImageSample;
|
||||
}
|
||||
}
|
||||
@@ -334,12 +334,12 @@ bool LowerDsAppendConsume(const Decoder::Instruction& decoded, BasicBlock& block
|
||||
|
||||
Opcode ImageAtomicIrOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::ImageAtomicAdd: return Opcode::AtomicAddU32;
|
||||
case Decoder::Opcode::ImageAtomicUMin: return Opcode::AtomicUMinU32;
|
||||
case Decoder::Opcode::ImageAtomicUMax: return Opcode::AtomicUMaxU32;
|
||||
case Decoder::Opcode::ImageAtomicAnd: return Opcode::AtomicAndU32;
|
||||
case Decoder::Opcode::ImageAtomicOr: return Opcode::AtomicOrU32;
|
||||
case Decoder::Opcode::ImageAtomicXor: return Opcode::AtomicXorU32;
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_ADD: return Opcode::AtomicAddU32;
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_UMIN: return Opcode::AtomicUMinU32;
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_UMAX: return Opcode::AtomicUMaxU32;
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_AND: return Opcode::AtomicAndU32;
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_OR: return Opcode::AtomicOrU32;
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_XOR: return Opcode::AtomicXorU32;
|
||||
default: return Opcode::AtomicAddU32;
|
||||
}
|
||||
}
|
||||
@@ -477,8 +477,8 @@ bool LowerImageOperation(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
std::string* error) {
|
||||
Instruction inst;
|
||||
inst.pc = decoded.pc;
|
||||
if (decoded.opcode == Decoder::Opcode::ImageStore ||
|
||||
decoded.opcode == Decoder::Opcode::ImageStoreMip) {
|
||||
if (decoded.opcode == Decoder::Opcode::IMAGE_STORE ||
|
||||
decoded.opcode == Decoder::Opcode::IMAGE_STORE_MIP) {
|
||||
inst.op = Opcode::ImageStore;
|
||||
inst.src_count = 2;
|
||||
inst.memory = MemoryInfoFromDecoded(decoded, ResourceKind::StorageImage);
|
||||
@@ -508,181 +508,184 @@ bool LowerImageOperation(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
bool LowerMemoryInstruction(const Decoder::Instruction& decoded, BasicBlock& block,
|
||||
std::string* error) {
|
||||
switch (decoded.opcode) {
|
||||
case Decoder::Opcode::SLoadDword:
|
||||
case Decoder::Opcode::SLoadDwordx2:
|
||||
case Decoder::Opcode::SLoadDwordx4:
|
||||
case Decoder::Opcode::SLoadDwordx8:
|
||||
case Decoder::Opcode::SLoadDwordx16:
|
||||
case Decoder::Opcode::S_LOAD_DWORD:
|
||||
case Decoder::Opcode::S_LOAD_DWORDX2:
|
||||
case Decoder::Opcode::S_LOAD_DWORDX4:
|
||||
case Decoder::Opcode::S_LOAD_DWORDX8:
|
||||
case Decoder::Opcode::S_LOAD_DWORDX16:
|
||||
return LowerScalarMemoryLoadDword(decoded, block, Opcode::SLoadDword, error);
|
||||
case Decoder::Opcode::SBufferLoadDword:
|
||||
case Decoder::Opcode::SBufferLoadDwordx2:
|
||||
case Decoder::Opcode::SBufferLoadDwordx4:
|
||||
case Decoder::Opcode::SBufferLoadDwordx8:
|
||||
case Decoder::Opcode::SBufferLoadDwordx16:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORD:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORDX2:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORDX4:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORDX8:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORDX16:
|
||||
return LowerScalarMemoryLoadDword(decoded, block, Opcode::SBufferLoadDword, error);
|
||||
case Decoder::Opcode::BufferLoadUbyte:
|
||||
case Decoder::Opcode::BufferLoadSbyte:
|
||||
case Decoder::Opcode::BufferLoadUshort:
|
||||
case Decoder::Opcode::BufferLoadSshort:
|
||||
case Decoder::Opcode::BufferLoadDword:
|
||||
case Decoder::Opcode::BufferLoadDwordx2:
|
||||
case Decoder::Opcode::BufferLoadDwordx3:
|
||||
case Decoder::Opcode::BufferLoadDwordx4:
|
||||
case Decoder::Opcode::BufferLoadFormatX:
|
||||
case Decoder::Opcode::BufferLoadFormatXy:
|
||||
case Decoder::Opcode::BufferLoadFormatXyz:
|
||||
case Decoder::Opcode::BufferLoadFormatXyzw:
|
||||
case Decoder::Opcode::TBufferLoadFormatX:
|
||||
case Decoder::Opcode::TBufferLoadFormatXy:
|
||||
case Decoder::Opcode::TBufferLoadFormatXyz:
|
||||
case Decoder::Opcode::TBufferLoadFormatXyzw: return LowerBufferLoad(decoded, block, error);
|
||||
case Decoder::Opcode::BufferStoreDword:
|
||||
case Decoder::Opcode::BufferStoreDwordx2:
|
||||
case Decoder::Opcode::BufferStoreDwordx3:
|
||||
case Decoder::Opcode::BufferStoreDwordx4:
|
||||
case Decoder::Opcode::BufferStoreByte:
|
||||
case Decoder::Opcode::BufferStoreShort:
|
||||
case Decoder::Opcode::BufferStoreFormatX:
|
||||
case Decoder::Opcode::BufferStoreFormatXy:
|
||||
case Decoder::Opcode::BufferStoreFormatXyz:
|
||||
case Decoder::Opcode::BufferStoreFormatXyzw:
|
||||
case Decoder::Opcode::TBufferStoreFormatX:
|
||||
case Decoder::Opcode::TBufferStoreFormatXy:
|
||||
case Decoder::Opcode::TBufferStoreFormatXyz:
|
||||
case Decoder::Opcode::TBufferStoreFormatXyzw:
|
||||
case Decoder::Opcode::BUFFER_LOAD_UBYTE:
|
||||
case Decoder::Opcode::BUFFER_LOAD_SBYTE:
|
||||
case Decoder::Opcode::BUFFER_LOAD_USHORT:
|
||||
case Decoder::Opcode::BUFFER_LOAD_SSHORT:
|
||||
case Decoder::Opcode::BUFFER_LOAD_DWORD:
|
||||
case Decoder::Opcode::BUFFER_LOAD_DWORDX2:
|
||||
case Decoder::Opcode::BUFFER_LOAD_DWORDX3:
|
||||
case Decoder::Opcode::BUFFER_LOAD_DWORDX4:
|
||||
case Decoder::Opcode::BUFFER_LOAD_FORMAT_X:
|
||||
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XY:
|
||||
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XYZ:
|
||||
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XYZW:
|
||||
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_X:
|
||||
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_XY:
|
||||
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_XYZ:
|
||||
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_XYZW:
|
||||
return LowerBufferLoad(decoded, block, error);
|
||||
case Decoder::Opcode::BUFFER_STORE_DWORD:
|
||||
case Decoder::Opcode::BUFFER_STORE_DWORDX2:
|
||||
case Decoder::Opcode::BUFFER_STORE_DWORDX3:
|
||||
case Decoder::Opcode::BUFFER_STORE_DWORDX4:
|
||||
case Decoder::Opcode::BUFFER_STORE_BYTE:
|
||||
case Decoder::Opcode::BUFFER_STORE_SHORT:
|
||||
case Decoder::Opcode::BUFFER_STORE_FORMAT_X:
|
||||
case Decoder::Opcode::BUFFER_STORE_FORMAT_XY:
|
||||
case Decoder::Opcode::BUFFER_STORE_FORMAT_XYZ:
|
||||
case Decoder::Opcode::BUFFER_STORE_FORMAT_XYZW:
|
||||
case Decoder::Opcode::TBUFFER_STORE_FORMAT_X:
|
||||
case Decoder::Opcode::TBUFFER_STORE_FORMAT_XY:
|
||||
case Decoder::Opcode::TBUFFER_STORE_FORMAT_XYZ:
|
||||
case Decoder::Opcode::TBUFFER_STORE_FORMAT_XYZW:
|
||||
return LowerBufferStore(decoded, block, error);
|
||||
case Decoder::Opcode::BufferAtomicSwap:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_SWAP:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicSwapU32, error);
|
||||
case Decoder::Opcode::BufferAtomicAdd:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_ADD:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicAddU32, error);
|
||||
case Decoder::Opcode::BufferAtomicSub:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_SUB:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicSubU32, error);
|
||||
case Decoder::Opcode::BufferAtomicSMin:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_SMIN:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicSMinI32, error);
|
||||
case Decoder::Opcode::BufferAtomicUMin:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_UMIN:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicUMinU32, error);
|
||||
case Decoder::Opcode::BufferAtomicSMax:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_SMAX:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicSMaxI32, error);
|
||||
case Decoder::Opcode::BufferAtomicUMax:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_UMAX:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicUMaxU32, error);
|
||||
case Decoder::Opcode::BufferAtomicAnd:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_AND:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicAndU32, error);
|
||||
case Decoder::Opcode::BufferAtomicOr:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_OR:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicOrU32, error);
|
||||
case Decoder::Opcode::BufferAtomicXor:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_XOR:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicXorU32, error);
|
||||
case Decoder::Opcode::BufferAtomicFMin:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_FMIN:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicFMinF32, error);
|
||||
case Decoder::Opcode::BufferAtomicFMax:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_FMAX:
|
||||
return LowerBufferAtomicDword(decoded, block, Opcode::AtomicFMaxF32, error);
|
||||
case Decoder::Opcode::FlatLoadUbyte:
|
||||
case Decoder::Opcode::FlatLoadSbyte:
|
||||
case Decoder::Opcode::FlatLoadUshort:
|
||||
case Decoder::Opcode::FlatLoadSshort:
|
||||
case Decoder::Opcode::FlatLoadDword:
|
||||
case Decoder::Opcode::FlatLoadDwordx2:
|
||||
case Decoder::Opcode::FlatLoadDwordx3:
|
||||
case Decoder::Opcode::FlatLoadDwordx4: return LowerFlatLoad(decoded, block, error);
|
||||
case Decoder::Opcode::FlatStoreByte:
|
||||
case Decoder::Opcode::FlatStoreShort:
|
||||
case Decoder::Opcode::FlatStoreDword:
|
||||
case Decoder::Opcode::FlatStoreDwordx2:
|
||||
case Decoder::Opcode::FlatStoreDwordx3:
|
||||
case Decoder::Opcode::FlatStoreDwordx4: return LowerFlatStore(decoded, block, error);
|
||||
case Decoder::Opcode::DsAddU32:
|
||||
case Decoder::Opcode::FLAT_LOAD_UBYTE:
|
||||
case Decoder::Opcode::FLAT_LOAD_SBYTE:
|
||||
case Decoder::Opcode::FLAT_LOAD_USHORT:
|
||||
case Decoder::Opcode::FLAT_LOAD_SSHORT:
|
||||
case Decoder::Opcode::FLAT_LOAD_DWORD:
|
||||
case Decoder::Opcode::FLAT_LOAD_DWORDX2:
|
||||
case Decoder::Opcode::FLAT_LOAD_DWORDX3:
|
||||
case Decoder::Opcode::FLAT_LOAD_DWORDX4: return LowerFlatLoad(decoded, block, error);
|
||||
case Decoder::Opcode::FLAT_STORE_BYTE:
|
||||
case Decoder::Opcode::FLAT_STORE_SHORT:
|
||||
case Decoder::Opcode::FLAT_STORE_DWORD:
|
||||
case Decoder::Opcode::FLAT_STORE_DWORDX2:
|
||||
case Decoder::Opcode::FLAT_STORE_DWORDX3:
|
||||
case Decoder::Opcode::FLAT_STORE_DWORDX4: return LowerFlatStore(decoded, block, error);
|
||||
case Decoder::Opcode::DS_ADD_U32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicAddU32, false, error);
|
||||
case Decoder::Opcode::DsAddRtnU32:
|
||||
case Decoder::Opcode::DS_ADD_RTN_U32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicAddU32, true, error);
|
||||
case Decoder::Opcode::DsSubU32:
|
||||
case Decoder::Opcode::DS_SUB_U32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicSubU32, false, error);
|
||||
case Decoder::Opcode::DsSubRtnU32:
|
||||
case Decoder::Opcode::DS_SUB_RTN_U32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicSubU32, true, error);
|
||||
case Decoder::Opcode::DsMinI32:
|
||||
case Decoder::Opcode::DS_MIN_I32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicSMinI32, false, error);
|
||||
case Decoder::Opcode::DsMinRtnI32:
|
||||
case Decoder::Opcode::DS_MIN_RTN_I32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicSMinI32, true, error);
|
||||
case Decoder::Opcode::DsMaxI32:
|
||||
case Decoder::Opcode::DS_MAX_I32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicSMaxI32, false, error);
|
||||
case Decoder::Opcode::DsMaxRtnI32:
|
||||
case Decoder::Opcode::DS_MAX_RTN_I32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicSMaxI32, true, error);
|
||||
case Decoder::Opcode::DsMinU32:
|
||||
case Decoder::Opcode::DS_MIN_U32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicUMinU32, false, error);
|
||||
case Decoder::Opcode::DsMinRtnU32:
|
||||
case Decoder::Opcode::DS_MIN_RTN_U32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicUMinU32, true, error);
|
||||
case Decoder::Opcode::DsMaxU32:
|
||||
case Decoder::Opcode::DS_MAX_U32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicUMaxU32, false, error);
|
||||
case Decoder::Opcode::DsMaxRtnU32:
|
||||
case Decoder::Opcode::DS_MAX_RTN_U32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicUMaxU32, true, error);
|
||||
case Decoder::Opcode::DsAndB32:
|
||||
case Decoder::Opcode::DS_AND_B32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicAndU32, false, error);
|
||||
case Decoder::Opcode::DsAndRtnB32:
|
||||
case Decoder::Opcode::DS_AND_RTN_B32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicAndU32, true, error);
|
||||
case Decoder::Opcode::DsOrB32:
|
||||
case Decoder::Opcode::DS_OR_B32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicOrU32, false, error);
|
||||
case Decoder::Opcode::DsOrRtnB32:
|
||||
case Decoder::Opcode::DS_OR_RTN_B32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicOrU32, true, error);
|
||||
case Decoder::Opcode::DsXorB32:
|
||||
case Decoder::Opcode::DS_XOR_B32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicXorU32, false, error);
|
||||
case Decoder::Opcode::DsXorRtnB32:
|
||||
case Decoder::Opcode::DS_XOR_RTN_B32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicXorU32, true, error);
|
||||
case Decoder::Opcode::DsWrxchgRtnB32:
|
||||
case Decoder::Opcode::DS_WRXCHG_RTN_B32:
|
||||
return LowerDsAtomicU32(decoded, block, Opcode::AtomicSwapU32, true, error);
|
||||
case Decoder::Opcode::DsMinF32:
|
||||
case Decoder::Opcode::DS_MIN_F32:
|
||||
return LowerDsFloatMinMaxF32(decoded, block, Opcode::DsMinF32, error);
|
||||
case Decoder::Opcode::DsMaxF32:
|
||||
case Decoder::Opcode::DS_MAX_F32:
|
||||
return LowerDsFloatMinMaxF32(decoded, block, Opcode::DsMaxF32, error);
|
||||
case Decoder::Opcode::DsSwizzleB32: return LowerDsSwizzleB32(decoded, block, error);
|
||||
case Decoder::Opcode::DsConsume:
|
||||
case Decoder::Opcode::DS_SWIZZLE_B32: return LowerDsSwizzleB32(decoded, block, error);
|
||||
case Decoder::Opcode::DS_CONSUME:
|
||||
return LowerDsAppendConsume(decoded, block, Opcode::DsConsume, error);
|
||||
case Decoder::Opcode::DsAppend:
|
||||
case Decoder::Opcode::DS_APPEND:
|
||||
return LowerDsAppendConsume(decoded, block, Opcode::DsAppend, error);
|
||||
case Decoder::Opcode::DsWriteAddtidB32: return LowerDsWriteAddtidB32(decoded, block, error);
|
||||
case Decoder::Opcode::DsReadAddtidB32: return LowerDsReadAddtidB32(decoded, block, error);
|
||||
case Decoder::Opcode::DsRead2B32: return LowerDsRead2(decoded, block, 1, error);
|
||||
case Decoder::Opcode::DsRead2B64:
|
||||
case Decoder::Opcode::DsRead2St64B64: return LowerDsRead2(decoded, block, 2, error);
|
||||
case Decoder::Opcode::DsReadSbyte:
|
||||
case Decoder::Opcode::DsReadUbyte:
|
||||
case Decoder::Opcode::DsReadSshort:
|
||||
case Decoder::Opcode::DsReadUshort:
|
||||
case Decoder::Opcode::DsReadB32:
|
||||
case Decoder::Opcode::DsReadB64:
|
||||
case Decoder::Opcode::DsReadB96:
|
||||
case Decoder::Opcode::DsReadB128: return LowerDsRead(decoded, block, error);
|
||||
case Decoder::Opcode::DsWrite2B32:
|
||||
case Decoder::Opcode::DsWrite2St64B32: return LowerDsWrite2(decoded, block, 1, error);
|
||||
case Decoder::Opcode::DsWrite2B64:
|
||||
case Decoder::Opcode::DsWrite2St64B64: return LowerDsWrite2(decoded, block, 2, error);
|
||||
case Decoder::Opcode::DsWriteByte:
|
||||
case Decoder::Opcode::DsWriteShort:
|
||||
case Decoder::Opcode::DsWriteB32:
|
||||
case Decoder::Opcode::DsWriteB64:
|
||||
case Decoder::Opcode::DsWriteB96:
|
||||
case Decoder::Opcode::DsWriteB128: return LowerDsWrite(decoded, block, error);
|
||||
case Decoder::Opcode::ImageGetResinfo:
|
||||
case Decoder::Opcode::ImageGetLod:
|
||||
case Decoder::Opcode::ImageLoad:
|
||||
case Decoder::Opcode::ImageLoadMip:
|
||||
case Decoder::Opcode::ImageStore:
|
||||
case Decoder::Opcode::ImageStoreMip:
|
||||
case Decoder::Opcode::ImageGather4Lz:
|
||||
case Decoder::Opcode::ImageGather4C:
|
||||
case Decoder::Opcode::ImageGather4CLz:
|
||||
case Decoder::Opcode::ImageGather4LzO:
|
||||
case Decoder::Opcode::ImageGather4CO:
|
||||
case Decoder::Opcode::ImageGather4CLzO:
|
||||
case Decoder::Opcode::ImageGather4H:
|
||||
case Decoder::Opcode::ImageSample: return LowerImageOperation(decoded, block, error);
|
||||
case Decoder::Opcode::ImageAtomicAdd:
|
||||
case Decoder::Opcode::ImageAtomicUMin:
|
||||
case Decoder::Opcode::ImageAtomicUMax:
|
||||
case Decoder::Opcode::ImageAtomicAnd:
|
||||
case Decoder::Opcode::ImageAtomicOr:
|
||||
case Decoder::Opcode::ImageAtomicXor: return LowerImageAtomicU32(decoded, block, error);
|
||||
case Decoder::Opcode::DS_WRITE_ADDTID_B32:
|
||||
return LowerDsWriteAddtidB32(decoded, block, error);
|
||||
case Decoder::Opcode::DS_READ_ADDTID_B32:
|
||||
return LowerDsReadAddtidB32(decoded, block, error);
|
||||
case Decoder::Opcode::DS_READ2_B32: return LowerDsRead2(decoded, block, 1, error);
|
||||
case Decoder::Opcode::DS_READ2_B64:
|
||||
case Decoder::Opcode::DS_READ2ST64_B64: return LowerDsRead2(decoded, block, 2, error);
|
||||
case Decoder::Opcode::DS_READ_I8:
|
||||
case Decoder::Opcode::DS_READ_U8:
|
||||
case Decoder::Opcode::DS_READ_I16:
|
||||
case Decoder::Opcode::DS_READ_U16:
|
||||
case Decoder::Opcode::DS_READ_B32:
|
||||
case Decoder::Opcode::DS_READ_B64:
|
||||
case Decoder::Opcode::DS_READ_B96:
|
||||
case Decoder::Opcode::DS_READ_B128: return LowerDsRead(decoded, block, error);
|
||||
case Decoder::Opcode::DS_WRITE2_B32:
|
||||
case Decoder::Opcode::DS_WRITE2ST64_B32: return LowerDsWrite2(decoded, block, 1, error);
|
||||
case Decoder::Opcode::DS_WRITE2_B64:
|
||||
case Decoder::Opcode::DS_WRITE2ST64_B64: return LowerDsWrite2(decoded, block, 2, error);
|
||||
case Decoder::Opcode::DS_WRITE_B8:
|
||||
case Decoder::Opcode::DS_WRITE_B16:
|
||||
case Decoder::Opcode::DS_WRITE_B32:
|
||||
case Decoder::Opcode::DS_WRITE_B64:
|
||||
case Decoder::Opcode::DS_WRITE_B96:
|
||||
case Decoder::Opcode::DS_WRITE_B128: return LowerDsWrite(decoded, block, error);
|
||||
case Decoder::Opcode::IMAGE_GET_RESINFO:
|
||||
case Decoder::Opcode::IMAGE_GET_LOD:
|
||||
case Decoder::Opcode::IMAGE_LOAD:
|
||||
case Decoder::Opcode::IMAGE_LOAD_MIP:
|
||||
case Decoder::Opcode::IMAGE_STORE:
|
||||
case Decoder::Opcode::IMAGE_STORE_MIP:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_LZ:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_LZ:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_LZ_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_LZ_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4H:
|
||||
case Decoder::Opcode::IMAGE_SAMPLE: return LowerImageOperation(decoded, block, error);
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_ADD:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_UMIN:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_UMAX:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_AND:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_OR:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_XOR: return LowerImageAtomicU32(decoded, block, error);
|
||||
default:
|
||||
if (error != nullptr) {
|
||||
*error = fmt::format("memory opcode has no specialized IR lowering: {}",
|
||||
Decoder::OpcodeToString(decoded.opcode));
|
||||
magic_enum::enum_name(decoded.opcode));
|
||||
}
|
||||
return false;
|
||||
}
|
||||
@@ -690,139 +693,139 @@ bool LowerMemoryInstruction(const Decoder::Instruction& decoded, BasicBlock& blo
|
||||
|
||||
bool IsMemoryOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SLoadDword:
|
||||
case Decoder::Opcode::SLoadDwordx2:
|
||||
case Decoder::Opcode::SLoadDwordx4:
|
||||
case Decoder::Opcode::SLoadDwordx8:
|
||||
case Decoder::Opcode::SLoadDwordx16:
|
||||
case Decoder::Opcode::SBufferLoadDword:
|
||||
case Decoder::Opcode::SBufferLoadDwordx2:
|
||||
case Decoder::Opcode::SBufferLoadDwordx4:
|
||||
case Decoder::Opcode::SBufferLoadDwordx8:
|
||||
case Decoder::Opcode::SBufferLoadDwordx16:
|
||||
case Decoder::Opcode::BufferLoadUbyte:
|
||||
case Decoder::Opcode::BufferLoadSbyte:
|
||||
case Decoder::Opcode::BufferLoadUshort:
|
||||
case Decoder::Opcode::BufferLoadSshort:
|
||||
case Decoder::Opcode::BufferLoadDword:
|
||||
case Decoder::Opcode::BufferLoadDwordx2:
|
||||
case Decoder::Opcode::BufferLoadDwordx3:
|
||||
case Decoder::Opcode::BufferLoadDwordx4:
|
||||
case Decoder::Opcode::BufferLoadFormatX:
|
||||
case Decoder::Opcode::BufferLoadFormatXy:
|
||||
case Decoder::Opcode::BufferLoadFormatXyz:
|
||||
case Decoder::Opcode::BufferLoadFormatXyzw:
|
||||
case Decoder::Opcode::TBufferLoadFormatX:
|
||||
case Decoder::Opcode::TBufferLoadFormatXy:
|
||||
case Decoder::Opcode::TBufferLoadFormatXyz:
|
||||
case Decoder::Opcode::TBufferLoadFormatXyzw:
|
||||
case Decoder::Opcode::BufferStoreDword:
|
||||
case Decoder::Opcode::BufferStoreDwordx2:
|
||||
case Decoder::Opcode::BufferStoreDwordx3:
|
||||
case Decoder::Opcode::BufferStoreDwordx4:
|
||||
case Decoder::Opcode::BufferStoreByte:
|
||||
case Decoder::Opcode::BufferStoreShort:
|
||||
case Decoder::Opcode::BufferStoreFormatX:
|
||||
case Decoder::Opcode::BufferStoreFormatXy:
|
||||
case Decoder::Opcode::BufferStoreFormatXyz:
|
||||
case Decoder::Opcode::BufferStoreFormatXyzw:
|
||||
case Decoder::Opcode::TBufferStoreFormatX:
|
||||
case Decoder::Opcode::TBufferStoreFormatXy:
|
||||
case Decoder::Opcode::TBufferStoreFormatXyz:
|
||||
case Decoder::Opcode::TBufferStoreFormatXyzw:
|
||||
case Decoder::Opcode::BufferAtomicSwap:
|
||||
case Decoder::Opcode::BufferAtomicAdd:
|
||||
case Decoder::Opcode::BufferAtomicSub:
|
||||
case Decoder::Opcode::BufferAtomicSMin:
|
||||
case Decoder::Opcode::BufferAtomicUMin:
|
||||
case Decoder::Opcode::BufferAtomicSMax:
|
||||
case Decoder::Opcode::BufferAtomicUMax:
|
||||
case Decoder::Opcode::BufferAtomicAnd:
|
||||
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:
|
||||
case Decoder::Opcode::FlatLoadSshort:
|
||||
case Decoder::Opcode::FlatLoadDword:
|
||||
case Decoder::Opcode::FlatLoadDwordx2:
|
||||
case Decoder::Opcode::FlatLoadDwordx3:
|
||||
case Decoder::Opcode::FlatLoadDwordx4:
|
||||
case Decoder::Opcode::FlatStoreByte:
|
||||
case Decoder::Opcode::FlatStoreShort:
|
||||
case Decoder::Opcode::FlatStoreDword:
|
||||
case Decoder::Opcode::FlatStoreDwordx2:
|
||||
case Decoder::Opcode::FlatStoreDwordx3:
|
||||
case Decoder::Opcode::FlatStoreDwordx4:
|
||||
case Decoder::Opcode::DsAddU32:
|
||||
case Decoder::Opcode::DsAddRtnU32:
|
||||
case Decoder::Opcode::DsSubU32:
|
||||
case Decoder::Opcode::DsSubRtnU32:
|
||||
case Decoder::Opcode::DsMinI32:
|
||||
case Decoder::Opcode::DsMinRtnI32:
|
||||
case Decoder::Opcode::DsMaxI32:
|
||||
case Decoder::Opcode::DsMaxRtnI32:
|
||||
case Decoder::Opcode::DsMinU32:
|
||||
case Decoder::Opcode::DsMinRtnU32:
|
||||
case Decoder::Opcode::DsMaxU32:
|
||||
case Decoder::Opcode::DsMaxRtnU32:
|
||||
case Decoder::Opcode::DsAndB32:
|
||||
case Decoder::Opcode::DsAndRtnB32:
|
||||
case Decoder::Opcode::DsOrB32:
|
||||
case Decoder::Opcode::DsOrRtnB32:
|
||||
case Decoder::Opcode::DsXorB32:
|
||||
case Decoder::Opcode::DsXorRtnB32:
|
||||
case Decoder::Opcode::DsWrxchgRtnB32:
|
||||
case Decoder::Opcode::DsMinF32:
|
||||
case Decoder::Opcode::DsMaxF32:
|
||||
case Decoder::Opcode::DsSwizzleB32:
|
||||
case Decoder::Opcode::DsConsume:
|
||||
case Decoder::Opcode::DsAppend:
|
||||
case Decoder::Opcode::DsReadSbyte:
|
||||
case Decoder::Opcode::DsReadUbyte:
|
||||
case Decoder::Opcode::DsReadSshort:
|
||||
case Decoder::Opcode::DsReadUshort:
|
||||
case Decoder::Opcode::DsRead2B32:
|
||||
case Decoder::Opcode::DsReadB32:
|
||||
case Decoder::Opcode::DsReadB64:
|
||||
case Decoder::Opcode::DsRead2B64:
|
||||
case Decoder::Opcode::DsRead2St64B64:
|
||||
case Decoder::Opcode::DsReadB96:
|
||||
case Decoder::Opcode::DsReadB128:
|
||||
case Decoder::Opcode::DsWriteByte:
|
||||
case Decoder::Opcode::DsWriteShort:
|
||||
case Decoder::Opcode::DsWrite2B32:
|
||||
case Decoder::Opcode::DsWrite2St64B32:
|
||||
case Decoder::Opcode::DsWrite2B64:
|
||||
case Decoder::Opcode::DsWrite2St64B64:
|
||||
case Decoder::Opcode::DsWriteB32:
|
||||
case Decoder::Opcode::DsWriteB64:
|
||||
case Decoder::Opcode::DsWriteB96:
|
||||
case Decoder::Opcode::DsWriteB128:
|
||||
case Decoder::Opcode::DsWriteAddtidB32:
|
||||
case Decoder::Opcode::DsReadAddtidB32:
|
||||
case Decoder::Opcode::ImageGetResinfo:
|
||||
case Decoder::Opcode::ImageGetLod:
|
||||
case Decoder::Opcode::ImageLoad:
|
||||
case Decoder::Opcode::ImageLoadMip:
|
||||
case Decoder::Opcode::ImageStore:
|
||||
case Decoder::Opcode::ImageStoreMip:
|
||||
case Decoder::Opcode::ImageAtomicAdd:
|
||||
case Decoder::Opcode::ImageAtomicUMin:
|
||||
case Decoder::Opcode::ImageAtomicUMax:
|
||||
case Decoder::Opcode::ImageAtomicAnd:
|
||||
case Decoder::Opcode::ImageAtomicOr:
|
||||
case Decoder::Opcode::ImageAtomicXor:
|
||||
case Decoder::Opcode::ImageGather4Lz:
|
||||
case Decoder::Opcode::ImageGather4C:
|
||||
case Decoder::Opcode::ImageGather4CLz:
|
||||
case Decoder::Opcode::ImageGather4LzO:
|
||||
case Decoder::Opcode::ImageGather4CO:
|
||||
case Decoder::Opcode::ImageGather4CLzO:
|
||||
case Decoder::Opcode::ImageGather4H:
|
||||
case Decoder::Opcode::ImageSample: return true;
|
||||
case Decoder::Opcode::S_LOAD_DWORD:
|
||||
case Decoder::Opcode::S_LOAD_DWORDX2:
|
||||
case Decoder::Opcode::S_LOAD_DWORDX4:
|
||||
case Decoder::Opcode::S_LOAD_DWORDX8:
|
||||
case Decoder::Opcode::S_LOAD_DWORDX16:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORD:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORDX2:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORDX4:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORDX8:
|
||||
case Decoder::Opcode::S_BUFFER_LOAD_DWORDX16:
|
||||
case Decoder::Opcode::BUFFER_LOAD_UBYTE:
|
||||
case Decoder::Opcode::BUFFER_LOAD_SBYTE:
|
||||
case Decoder::Opcode::BUFFER_LOAD_USHORT:
|
||||
case Decoder::Opcode::BUFFER_LOAD_SSHORT:
|
||||
case Decoder::Opcode::BUFFER_LOAD_DWORD:
|
||||
case Decoder::Opcode::BUFFER_LOAD_DWORDX2:
|
||||
case Decoder::Opcode::BUFFER_LOAD_DWORDX3:
|
||||
case Decoder::Opcode::BUFFER_LOAD_DWORDX4:
|
||||
case Decoder::Opcode::BUFFER_LOAD_FORMAT_X:
|
||||
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XY:
|
||||
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XYZ:
|
||||
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XYZW:
|
||||
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_X:
|
||||
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_XY:
|
||||
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_XYZ:
|
||||
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_XYZW:
|
||||
case Decoder::Opcode::BUFFER_STORE_DWORD:
|
||||
case Decoder::Opcode::BUFFER_STORE_DWORDX2:
|
||||
case Decoder::Opcode::BUFFER_STORE_DWORDX3:
|
||||
case Decoder::Opcode::BUFFER_STORE_DWORDX4:
|
||||
case Decoder::Opcode::BUFFER_STORE_BYTE:
|
||||
case Decoder::Opcode::BUFFER_STORE_SHORT:
|
||||
case Decoder::Opcode::BUFFER_STORE_FORMAT_X:
|
||||
case Decoder::Opcode::BUFFER_STORE_FORMAT_XY:
|
||||
case Decoder::Opcode::BUFFER_STORE_FORMAT_XYZ:
|
||||
case Decoder::Opcode::BUFFER_STORE_FORMAT_XYZW:
|
||||
case Decoder::Opcode::TBUFFER_STORE_FORMAT_X:
|
||||
case Decoder::Opcode::TBUFFER_STORE_FORMAT_XY:
|
||||
case Decoder::Opcode::TBUFFER_STORE_FORMAT_XYZ:
|
||||
case Decoder::Opcode::TBUFFER_STORE_FORMAT_XYZW:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_SWAP:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_ADD:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_SUB:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_SMIN:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_UMIN:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_SMAX:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_UMAX:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_AND:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_OR:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_XOR:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_FMIN:
|
||||
case Decoder::Opcode::BUFFER_ATOMIC_FMAX:
|
||||
case Decoder::Opcode::FLAT_LOAD_UBYTE:
|
||||
case Decoder::Opcode::FLAT_LOAD_SBYTE:
|
||||
case Decoder::Opcode::FLAT_LOAD_USHORT:
|
||||
case Decoder::Opcode::FLAT_LOAD_SSHORT:
|
||||
case Decoder::Opcode::FLAT_LOAD_DWORD:
|
||||
case Decoder::Opcode::FLAT_LOAD_DWORDX2:
|
||||
case Decoder::Opcode::FLAT_LOAD_DWORDX3:
|
||||
case Decoder::Opcode::FLAT_LOAD_DWORDX4:
|
||||
case Decoder::Opcode::FLAT_STORE_BYTE:
|
||||
case Decoder::Opcode::FLAT_STORE_SHORT:
|
||||
case Decoder::Opcode::FLAT_STORE_DWORD:
|
||||
case Decoder::Opcode::FLAT_STORE_DWORDX2:
|
||||
case Decoder::Opcode::FLAT_STORE_DWORDX3:
|
||||
case Decoder::Opcode::FLAT_STORE_DWORDX4:
|
||||
case Decoder::Opcode::DS_ADD_U32:
|
||||
case Decoder::Opcode::DS_ADD_RTN_U32:
|
||||
case Decoder::Opcode::DS_SUB_U32:
|
||||
case Decoder::Opcode::DS_SUB_RTN_U32:
|
||||
case Decoder::Opcode::DS_MIN_I32:
|
||||
case Decoder::Opcode::DS_MIN_RTN_I32:
|
||||
case Decoder::Opcode::DS_MAX_I32:
|
||||
case Decoder::Opcode::DS_MAX_RTN_I32:
|
||||
case Decoder::Opcode::DS_MIN_U32:
|
||||
case Decoder::Opcode::DS_MIN_RTN_U32:
|
||||
case Decoder::Opcode::DS_MAX_U32:
|
||||
case Decoder::Opcode::DS_MAX_RTN_U32:
|
||||
case Decoder::Opcode::DS_AND_B32:
|
||||
case Decoder::Opcode::DS_AND_RTN_B32:
|
||||
case Decoder::Opcode::DS_OR_B32:
|
||||
case Decoder::Opcode::DS_OR_RTN_B32:
|
||||
case Decoder::Opcode::DS_XOR_B32:
|
||||
case Decoder::Opcode::DS_XOR_RTN_B32:
|
||||
case Decoder::Opcode::DS_WRXCHG_RTN_B32:
|
||||
case Decoder::Opcode::DS_MIN_F32:
|
||||
case Decoder::Opcode::DS_MAX_F32:
|
||||
case Decoder::Opcode::DS_SWIZZLE_B32:
|
||||
case Decoder::Opcode::DS_CONSUME:
|
||||
case Decoder::Opcode::DS_APPEND:
|
||||
case Decoder::Opcode::DS_READ_I8:
|
||||
case Decoder::Opcode::DS_READ_U8:
|
||||
case Decoder::Opcode::DS_READ_I16:
|
||||
case Decoder::Opcode::DS_READ_U16:
|
||||
case Decoder::Opcode::DS_READ2_B32:
|
||||
case Decoder::Opcode::DS_READ_B32:
|
||||
case Decoder::Opcode::DS_READ_B64:
|
||||
case Decoder::Opcode::DS_READ2_B64:
|
||||
case Decoder::Opcode::DS_READ2ST64_B64:
|
||||
case Decoder::Opcode::DS_READ_B96:
|
||||
case Decoder::Opcode::DS_READ_B128:
|
||||
case Decoder::Opcode::DS_WRITE_B8:
|
||||
case Decoder::Opcode::DS_WRITE_B16:
|
||||
case Decoder::Opcode::DS_WRITE2_B32:
|
||||
case Decoder::Opcode::DS_WRITE2ST64_B32:
|
||||
case Decoder::Opcode::DS_WRITE2_B64:
|
||||
case Decoder::Opcode::DS_WRITE2ST64_B64:
|
||||
case Decoder::Opcode::DS_WRITE_B32:
|
||||
case Decoder::Opcode::DS_WRITE_B64:
|
||||
case Decoder::Opcode::DS_WRITE_B96:
|
||||
case Decoder::Opcode::DS_WRITE_B128:
|
||||
case Decoder::Opcode::DS_WRITE_ADDTID_B32:
|
||||
case Decoder::Opcode::DS_READ_ADDTID_B32:
|
||||
case Decoder::Opcode::IMAGE_GET_RESINFO:
|
||||
case Decoder::Opcode::IMAGE_GET_LOD:
|
||||
case Decoder::Opcode::IMAGE_LOAD:
|
||||
case Decoder::Opcode::IMAGE_LOAD_MIP:
|
||||
case Decoder::Opcode::IMAGE_STORE:
|
||||
case Decoder::Opcode::IMAGE_STORE_MIP:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_ADD:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_UMIN:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_UMAX:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_AND:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_OR:
|
||||
case Decoder::Opcode::IMAGE_ATOMIC_XOR:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_LZ:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_LZ:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_LZ_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4_C_LZ_O:
|
||||
case Decoder::Opcode::IMAGE_GATHER4H:
|
||||
case Decoder::Opcode::IMAGE_SAMPLE: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
|
||||
@@ -11,430 +11,430 @@ namespace Libs::Graphics::ShaderRecompiler::IR {
|
||||
namespace {
|
||||
|
||||
struct LowerMap {
|
||||
Decoder::Opcode decoded = Decoder::Opcode::Unknown;
|
||||
Decoder::Opcode decoded = Decoder::Opcode::UNKNOWN;
|
||||
Opcode ir = Opcode::MoveU32;
|
||||
};
|
||||
|
||||
constexpr LowerMap LOWER_OPS[] = {
|
||||
{Decoder::Opcode::SMovB32, Opcode::MoveU32},
|
||||
{Decoder::Opcode::SMovB64, Opcode::MoveU64},
|
||||
{Decoder::Opcode::SWqmB64, Opcode::WqmB64},
|
||||
{Decoder::Opcode::SMovkI32, Opcode::MoveU32},
|
||||
{Decoder::Opcode::SAbsI32, Opcode::AbsI32},
|
||||
{Decoder::Opcode::SBrevB32, Opcode::BitReverseU32},
|
||||
{Decoder::Opcode::SBcnt1I32B32, Opcode::BitCountU32},
|
||||
{Decoder::Opcode::SBcnt1I32B64, Opcode::BitCountU64},
|
||||
{Decoder::Opcode::SFf1I32B32, Opcode::FindLsbU32},
|
||||
{Decoder::Opcode::SFf1I32B64, Opcode::FindLsbU64},
|
||||
{Decoder::Opcode::SFlbitI32B64, Opcode::FindMsbFromHighU64},
|
||||
{Decoder::Opcode::SBitreplicateB64B32, Opcode::BitReplicateB64B32},
|
||||
{Decoder::Opcode::SNotB32, Opcode::BitwiseNotU32},
|
||||
{Decoder::Opcode::SNotB64, Opcode::BitwiseNotU64},
|
||||
{Decoder::Opcode::SAddU32, Opcode::IAddU32},
|
||||
{Decoder::Opcode::SSubU32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::SAddI32, Opcode::IAddU32},
|
||||
{Decoder::Opcode::SSubI32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::SMinI32, Opcode::IMinI32},
|
||||
{Decoder::Opcode::SMaxI32, Opcode::IMaxI32},
|
||||
{Decoder::Opcode::SMinU32, Opcode::UMinU32},
|
||||
{Decoder::Opcode::SMaxU32, Opcode::UMaxU32},
|
||||
{Decoder::Opcode::SAndB32, Opcode::BitwiseAndU32},
|
||||
{Decoder::Opcode::SAndB64, Opcode::BitwiseAndU64},
|
||||
{Decoder::Opcode::SAndn2B32, Opcode::BitwiseAndNotU32},
|
||||
{Decoder::Opcode::SAndn2B64, Opcode::BitwiseAndNotU64},
|
||||
{Decoder::Opcode::SOrB32, Opcode::BitwiseOrU32},
|
||||
{Decoder::Opcode::SOrB64, Opcode::BitwiseOrU64},
|
||||
{Decoder::Opcode::SOrn2B32, Opcode::BitwiseOrNotU32},
|
||||
{Decoder::Opcode::SOrn2B64, Opcode::BitwiseOrNotU64},
|
||||
{Decoder::Opcode::SXorB32, Opcode::BitwiseXorU32},
|
||||
{Decoder::Opcode::SXorB64, Opcode::BitwiseXorU64},
|
||||
{Decoder::Opcode::SNandB32, Opcode::BitwiseNandU32},
|
||||
{Decoder::Opcode::SNandB64, Opcode::BitwiseNandU64},
|
||||
{Decoder::Opcode::SNorB32, Opcode::BitwiseNorU32},
|
||||
{Decoder::Opcode::SNorB64, Opcode::BitwiseNorU64},
|
||||
{Decoder::Opcode::SXnorB32, Opcode::BitwiseXnorU32},
|
||||
{Decoder::Opcode::SXnorB64, Opcode::BitwiseXnorU64},
|
||||
{Decoder::Opcode::SLshlB32, Opcode::ShiftLeftLogicalU32},
|
||||
{Decoder::Opcode::SLshlB64, Opcode::ShiftLeftLogicalU64},
|
||||
{Decoder::Opcode::SLshrB32, Opcode::ShiftRightLogicalU32},
|
||||
{Decoder::Opcode::SLshrB64, Opcode::ShiftRightLogicalU64},
|
||||
{Decoder::Opcode::SAshrI32, Opcode::ShiftRightArithmeticI32},
|
||||
{Decoder::Opcode::SMulI32, Opcode::IMulU32},
|
||||
{Decoder::Opcode::SMulHiU32, Opcode::UMulHighU32},
|
||||
{Decoder::Opcode::SMulkI32, Opcode::IMulU32},
|
||||
{Decoder::Opcode::SBfeU32, Opcode::BitFieldExtractU32},
|
||||
{Decoder::Opcode::SBfeI32, Opcode::BitFieldExtractI32},
|
||||
{Decoder::Opcode::SBfeU64, Opcode::BitFieldExtractU64},
|
||||
{Decoder::Opcode::SBfmB32, Opcode::BitFieldMaskU32},
|
||||
{Decoder::Opcode::SBfmB64, Opcode::BitFieldMaskU64},
|
||||
{Decoder::Opcode::SBitcmp0B32, Opcode::BitCompare0B32},
|
||||
{Decoder::Opcode::SBitcmp1B32, Opcode::BitCompare1B32},
|
||||
{Decoder::Opcode::SPackLlB32B16, Opcode::PackLowLowU16},
|
||||
{Decoder::Opcode::SPackLhB32B16, Opcode::PackLowHighU16},
|
||||
{Decoder::Opcode::SPackHhB32B16, Opcode::PackHighHighU16},
|
||||
{Decoder::Opcode::SCmpEqI32, Opcode::CompareEqI32},
|
||||
{Decoder::Opcode::SCmpLgI32, Opcode::CompareNeI32},
|
||||
{Decoder::Opcode::SCmpGtI32, Opcode::CompareGtI32},
|
||||
{Decoder::Opcode::SCmpGeI32, Opcode::CompareGeI32},
|
||||
{Decoder::Opcode::SCmpLtI32, Opcode::CompareLtI32},
|
||||
{Decoder::Opcode::SCmpLeI32, Opcode::CompareLeI32},
|
||||
{Decoder::Opcode::SCmpEqU32, Opcode::CompareEqU32},
|
||||
{Decoder::Opcode::SCmpLgU32, Opcode::CompareNeU32},
|
||||
{Decoder::Opcode::SCmpGtU32, Opcode::CompareGtU32},
|
||||
{Decoder::Opcode::SCmpGeU32, Opcode::CompareGeU32},
|
||||
{Decoder::Opcode::SCmpLtU32, Opcode::CompareLtU32},
|
||||
{Decoder::Opcode::SCmpLeU32, Opcode::CompareLeU32},
|
||||
{Decoder::Opcode::SCmpEqU64, Opcode::CompareEqU64},
|
||||
{Decoder::Opcode::SCmpLgU64, Opcode::CompareNeU64},
|
||||
{Decoder::Opcode::VReadfirstlaneB32, Opcode::ReadFirstLaneU32},
|
||||
{Decoder::Opcode::VReadlaneB32, Opcode::ReadLaneU32},
|
||||
{Decoder::Opcode::VWritelaneB32, Opcode::WriteLaneU32},
|
||||
{Decoder::Opcode::VPermlane16B32, Opcode::Permlane16B32},
|
||||
{Decoder::Opcode::VPermlanex16B32, Opcode::Permlanex16B32},
|
||||
{Decoder::Opcode::VCvtF32I32, Opcode::ConvertI32ToF32},
|
||||
{Decoder::Opcode::VCvtF32U32, Opcode::ConvertU32ToF32},
|
||||
{Decoder::Opcode::VCvtU32F32, Opcode::ConvertF32ToU32},
|
||||
{Decoder::Opcode::VCvtI32F32, Opcode::ConvertF32ToI32},
|
||||
{Decoder::Opcode::VCvtF16F32, Opcode::ConvertF32ToF16},
|
||||
{Decoder::Opcode::VCvtF32F16, Opcode::ConvertF16ToF32},
|
||||
{Decoder::Opcode::VCvtF16U16, Opcode::ConvertU16ToF16},
|
||||
{Decoder::Opcode::VCvtU16F16, Opcode::ConvertF16ToU16},
|
||||
{Decoder::Opcode::VCvtF16I16, Opcode::ConvertI16ToF16},
|
||||
{Decoder::Opcode::VCvtI16F16, Opcode::ConvertF16ToI16},
|
||||
{Decoder::Opcode::VCvtRpiI32F32, Opcode::ConvertRoundPlusInfF32ToI32},
|
||||
{Decoder::Opcode::VCvtFlrI32F32, Opcode::ConvertFloorF32ToI32},
|
||||
{Decoder::Opcode::VFrexpExpI32F32, Opcode::FrexpExpI32F32},
|
||||
{Decoder::Opcode::VFrexpMantF32, Opcode::FrexpMantF32},
|
||||
{Decoder::Opcode::VCvtOffF32I4, Opcode::ConvertI4ToOffsetF32},
|
||||
{Decoder::Opcode::VRcpF32, Opcode::RcpF32},
|
||||
{Decoder::Opcode::VRcpIflagF32, Opcode::RcpIflagF32},
|
||||
{Decoder::Opcode::VFractF32, Opcode::FractF32},
|
||||
{Decoder::Opcode::VTruncF32, Opcode::TruncF32},
|
||||
{Decoder::Opcode::VCeilF32, Opcode::CeilF32},
|
||||
{Decoder::Opcode::VRndneF32, Opcode::RoundEvenF32},
|
||||
{Decoder::Opcode::VFloorF32, Opcode::FloorF32},
|
||||
{Decoder::Opcode::VExpF32, Opcode::Exp2F32},
|
||||
{Decoder::Opcode::VLogF32, Opcode::Log2F32},
|
||||
{Decoder::Opcode::VRsqF32, Opcode::InverseSqrtF32},
|
||||
{Decoder::Opcode::VSqrtF32, Opcode::SqrtF32},
|
||||
{Decoder::Opcode::VRcpF16, Opcode::RcpF16},
|
||||
{Decoder::Opcode::VSqrtF16, Opcode::SqrtF16},
|
||||
{Decoder::Opcode::VRsqF16, Opcode::InverseSqrtF16},
|
||||
{Decoder::Opcode::VLogF16, Opcode::Log2F16},
|
||||
{Decoder::Opcode::VExpF16, Opcode::Exp2F16},
|
||||
{Decoder::Opcode::VFloorF16, Opcode::FloorF16},
|
||||
{Decoder::Opcode::VCeilF16, Opcode::CeilF16},
|
||||
{Decoder::Opcode::VTruncF16, Opcode::TruncF16},
|
||||
{Decoder::Opcode::VRndneF16, Opcode::RoundEvenF16},
|
||||
{Decoder::Opcode::VSinF32, Opcode::SinF32},
|
||||
{Decoder::Opcode::VCosF32, Opcode::CosF32},
|
||||
{Decoder::Opcode::VNotB32, Opcode::BitwiseNotU32},
|
||||
{Decoder::Opcode::VBfrevB32, Opcode::BitReverseU32},
|
||||
{Decoder::Opcode::VFfblB32, Opcode::FindLsbU32},
|
||||
{Decoder::Opcode::VFfbhU32, Opcode::FindMsbFromHighU32},
|
||||
{Decoder::Opcode::VAddF32, Opcode::FAddF32},
|
||||
{Decoder::Opcode::VSubF32, Opcode::FSubF32},
|
||||
{Decoder::Opcode::VSubrevF32, Opcode::FSubF32},
|
||||
{Decoder::Opcode::VMulF32, Opcode::FMulF32},
|
||||
{Decoder::Opcode::VMacF32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::VMadmkF32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::VMadakF32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::VMinF32, Opcode::FMinF32},
|
||||
{Decoder::Opcode::VMaxF32, Opcode::FMaxF32},
|
||||
{Decoder::Opcode::VMinI32, Opcode::IMinI32},
|
||||
{Decoder::Opcode::VMaxI32, Opcode::IMaxI32},
|
||||
{Decoder::Opcode::VMadF32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::VMadI32I24, Opcode::IMadI24U32},
|
||||
{Decoder::Opcode::VMadU32U24, Opcode::UMadU24U32},
|
||||
{Decoder::Opcode::VCubeidF32, Opcode::CubeIdF32},
|
||||
{Decoder::Opcode::VCubescF32, Opcode::CubeScF32},
|
||||
{Decoder::Opcode::VCubetcF32, Opcode::CubeTcF32},
|
||||
{Decoder::Opcode::VCubemaF32, Opcode::CubeMaF32},
|
||||
{Decoder::Opcode::VFmaF32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::VBfeU32, Opcode::BitFieldExtract3U32},
|
||||
{Decoder::Opcode::VBfeI32, Opcode::BitFieldExtract3I32},
|
||||
{Decoder::Opcode::VBfiB32, Opcode::BitFieldInsertSelectU32},
|
||||
{Decoder::Opcode::VAlignbitB32, Opcode::AlignBitU32},
|
||||
{Decoder::Opcode::VAlignbyteB32, Opcode::AlignByteU32},
|
||||
{Decoder::Opcode::VMin3F32, Opcode::FMin3F32},
|
||||
{Decoder::Opcode::VMin3I32, Opcode::IMin3I32},
|
||||
{Decoder::Opcode::VMin3U32, Opcode::UMin3U32},
|
||||
{Decoder::Opcode::VMin3F16, Opcode::Min3F16},
|
||||
{Decoder::Opcode::VMax3F32, Opcode::FMax3F32},
|
||||
{Decoder::Opcode::VMax3I32, Opcode::IMax3I32},
|
||||
{Decoder::Opcode::VMax3U32, Opcode::UMax3U32},
|
||||
{Decoder::Opcode::VMax3F16, Opcode::Max3F16},
|
||||
{Decoder::Opcode::VMed3F32, Opcode::FMed3F32},
|
||||
{Decoder::Opcode::VMed3I32, Opcode::IMed3I32},
|
||||
{Decoder::Opcode::VMed3U32, Opcode::UMed3U32},
|
||||
{Decoder::Opcode::VMed3F16, Opcode::Med3F16},
|
||||
{Decoder::Opcode::VMed3I16, Opcode::IMed3I16},
|
||||
{Decoder::Opcode::VSadU32, Opcode::SadU32},
|
||||
{Decoder::Opcode::VAdd3U32, Opcode::IAdd3U32},
|
||||
{Decoder::Opcode::VLshlAddU32, Opcode::ShiftLeftAddU32},
|
||||
{Decoder::Opcode::VAddLshlU32, Opcode::AddShiftLeftU32},
|
||||
{Decoder::Opcode::VXadU32, Opcode::XorAddU32},
|
||||
{Decoder::Opcode::VLshlOrB32, Opcode::ShiftLeftOrU32},
|
||||
{Decoder::Opcode::VAndOrB32, Opcode::BitwiseAndOrU32},
|
||||
{Decoder::Opcode::VOr3B32, Opcode::BitwiseOr3U32},
|
||||
{Decoder::Opcode::VXor3B32, Opcode::BitwiseXor3U32},
|
||||
{Decoder::Opcode::VAddNcU32, Opcode::IAddU32},
|
||||
{Decoder::Opcode::VSubNcU32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::VSubrevNcU32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::VAddNcU16, Opcode::IAddU16},
|
||||
{Decoder::Opcode::VSubNcU16, Opcode::ISubI16},
|
||||
{Decoder::Opcode::VMaxU16, Opcode::UMaxU16},
|
||||
{Decoder::Opcode::VMaxI16, Opcode::IMaxI16},
|
||||
{Decoder::Opcode::VMinU16, Opcode::UMinU16},
|
||||
{Decoder::Opcode::VMinI16, Opcode::IMinI16},
|
||||
{Decoder::Opcode::VAddNcI16, Opcode::IAddU16},
|
||||
{Decoder::Opcode::VSubNcI16, Opcode::ISubI16},
|
||||
{Decoder::Opcode::VMulI32I24, Opcode::IMulI24U32},
|
||||
{Decoder::Opcode::VMulU32U24, Opcode::UMulU24U32},
|
||||
{Decoder::Opcode::VMulLoU32, Opcode::IMulU32},
|
||||
{Decoder::Opcode::VMulHiU32, Opcode::UMulHighU32},
|
||||
{Decoder::Opcode::VMulLoI32, Opcode::IMulU32},
|
||||
{Decoder::Opcode::VMulHiI32, Opcode::SMulHighI32},
|
||||
{Decoder::Opcode::VAddI32, Opcode::IAddU32},
|
||||
{Decoder::Opcode::VSubI32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::VSubrevI32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::VBfmB32, Opcode::BitFieldMaskU32},
|
||||
{Decoder::Opcode::VCvtPkrtzF16F32, Opcode::PackF32ToF16Rtz},
|
||||
{Decoder::Opcode::VLdexpF32, Opcode::LdexpF32},
|
||||
{Decoder::Opcode::VCvtPknormI16F32, Opcode::PackSnorm2x16F32},
|
||||
{Decoder::Opcode::VCvtPknormU16F32, Opcode::PackUnorm2x16F32},
|
||||
{Decoder::Opcode::VCvtPkU16U32, Opcode::PackU16U32},
|
||||
{Decoder::Opcode::VCvtPkI16I32, Opcode::PackU16U32},
|
||||
{Decoder::Opcode::VCvtPkU8F32, Opcode::PackU8F32},
|
||||
{Decoder::Opcode::VPackB32F16, Opcode::PackB32F16},
|
||||
{Decoder::Opcode::VPkMadI16, Opcode::PackedMadI16},
|
||||
{Decoder::Opcode::VPkMulLoU16, Opcode::PackedMulLoU16},
|
||||
{Decoder::Opcode::VPkAddI16, Opcode::PackedAddI16},
|
||||
{Decoder::Opcode::VPkSubI16, Opcode::PackedSubI16},
|
||||
{Decoder::Opcode::VPkLshlrevB16, Opcode::PackedLshlrevB16},
|
||||
{Decoder::Opcode::VPkLshrrevB16, Opcode::PackedLshrrevB16},
|
||||
{Decoder::Opcode::VPkAshrrevI16, Opcode::PackedAshrrevI16},
|
||||
{Decoder::Opcode::VPkMaxI16, Opcode::PackedMaxI16},
|
||||
{Decoder::Opcode::VPkMinI16, Opcode::PackedMinI16},
|
||||
{Decoder::Opcode::VPkMadU16, Opcode::PackedMadU16},
|
||||
{Decoder::Opcode::VPkAddU16, Opcode::PackedAddU16},
|
||||
{Decoder::Opcode::VPkSubU16, Opcode::PackedSubU16},
|
||||
{Decoder::Opcode::VPkMaxU16, Opcode::PackedMaxU16},
|
||||
{Decoder::Opcode::VPkMinU16, Opcode::PackedMinU16},
|
||||
{Decoder::Opcode::VPkAddF16, Opcode::PackedAddF16},
|
||||
{Decoder::Opcode::VPkMulF16, Opcode::PackedMulF16},
|
||||
{Decoder::Opcode::VPkMinF16, Opcode::PackedMinF16},
|
||||
{Decoder::Opcode::VPkMaxF16, Opcode::PackedMaxF16},
|
||||
{Decoder::Opcode::VPkFmaF16, Opcode::PackedFmaF16},
|
||||
{Decoder::Opcode::VPkFmacF16, Opcode::PackedFmaF16},
|
||||
{Decoder::Opcode::VAddF16, Opcode::AddF16},
|
||||
{Decoder::Opcode::VSubF16, Opcode::SubF16},
|
||||
{Decoder::Opcode::VSubrevF16, Opcode::SubF16},
|
||||
{Decoder::Opcode::VMulF16, Opcode::MulF16},
|
||||
{Decoder::Opcode::VFmacF16, Opcode::FmaF16},
|
||||
{Decoder::Opcode::VFmamkF16, Opcode::FmaF16},
|
||||
{Decoder::Opcode::VFmaakF16, Opcode::FmaF16},
|
||||
{Decoder::Opcode::VFmaF16, Opcode::FmaF16},
|
||||
{Decoder::Opcode::VMadMixloF16, Opcode::MadMixF16},
|
||||
{Decoder::Opcode::VMadMixhiF16, Opcode::MadMixF16},
|
||||
{Decoder::Opcode::VAndB32, Opcode::BitwiseAndU32},
|
||||
{Decoder::Opcode::VOrB32, Opcode::BitwiseOrU32},
|
||||
{Decoder::Opcode::VXorB32, Opcode::BitwiseXorU32},
|
||||
{Decoder::Opcode::VXnorB32, Opcode::BitwiseXnorU32},
|
||||
{Decoder::Opcode::VBcntU32B32, Opcode::BitCountAddU32},
|
||||
{Decoder::Opcode::VMbcntLoU32B32, Opcode::MaskedBitCountLowU32},
|
||||
{Decoder::Opcode::VMbcntHiU32B32, Opcode::MaskedBitCountHighU32},
|
||||
{Decoder::Opcode::VLshlB32, Opcode::ShiftLeftLogicalU32},
|
||||
{Decoder::Opcode::VLshlrevB32, Opcode::ShiftLeftLogicalU32},
|
||||
{Decoder::Opcode::VLshlrevB16, Opcode::ShiftLeftLogicalU16},
|
||||
{Decoder::Opcode::VLshrB32, Opcode::ShiftRightLogicalU32},
|
||||
{Decoder::Opcode::VLshrrevB32, Opcode::ShiftRightLogicalU32},
|
||||
{Decoder::Opcode::VLshrrevB16, Opcode::ShiftRightLogicalU16},
|
||||
{Decoder::Opcode::VAshrI32, Opcode::ShiftRightArithmeticI32},
|
||||
{Decoder::Opcode::VAshrrevI32, Opcode::ShiftRightArithmeticI32},
|
||||
{Decoder::Opcode::VAshrrevI16, Opcode::ShiftRightArithmeticI16},
|
||||
{Decoder::Opcode::VMinU32, Opcode::UMinU32},
|
||||
{Decoder::Opcode::VMaxU32, Opcode::UMaxU32},
|
||||
{Decoder::Opcode::VMaxF16, Opcode::MaxF16},
|
||||
{Decoder::Opcode::VMinF16, Opcode::MinF16},
|
||||
{Decoder::Opcode::VCmpFF32, Opcode::CompareFalse},
|
||||
{Decoder::Opcode::VCmpLtF32, Opcode::CompareLtF32},
|
||||
{Decoder::Opcode::VCmpEqF32, Opcode::CompareEqF32},
|
||||
{Decoder::Opcode::VCmpLeF32, Opcode::CompareLeF32},
|
||||
{Decoder::Opcode::VCmpGtF32, Opcode::CompareGtF32},
|
||||
{Decoder::Opcode::VCmpLgF32, Opcode::CompareNeF32},
|
||||
{Decoder::Opcode::VCmpGeF32, Opcode::CompareGeF32},
|
||||
{Decoder::Opcode::VCmpOF32, Opcode::CompareOrderedF32},
|
||||
{Decoder::Opcode::VCmpUF32, Opcode::CompareUnorderedF32},
|
||||
{Decoder::Opcode::VCmpNgeF32, Opcode::CompareUnordLtF32},
|
||||
{Decoder::Opcode::VCmpNlgF32, Opcode::CompareUnordEqF32},
|
||||
{Decoder::Opcode::VCmpNgtF32, Opcode::CompareUnordLeF32},
|
||||
{Decoder::Opcode::VCmpNleF32, Opcode::CompareUnordGtF32},
|
||||
{Decoder::Opcode::VCmpNeqF32, Opcode::CompareUnordNeF32},
|
||||
{Decoder::Opcode::VCmpNltF32, Opcode::CompareUnordGeF32},
|
||||
{Decoder::Opcode::VCmpTruF32, Opcode::CompareTrue},
|
||||
{Decoder::Opcode::VCmpClassF32, Opcode::CompareClassF32},
|
||||
{Decoder::Opcode::VCmpLtF16, Opcode::CompareLtF16},
|
||||
{Decoder::Opcode::VCmpEqF16, Opcode::CompareEqF16},
|
||||
{Decoder::Opcode::VCmpLeF16, Opcode::CompareLeF16},
|
||||
{Decoder::Opcode::VCmpGtF16, Opcode::CompareGtF16},
|
||||
{Decoder::Opcode::VCmpLgF16, Opcode::CompareNeF16},
|
||||
{Decoder::Opcode::VCmpGeF16, Opcode::CompareGeF16},
|
||||
{Decoder::Opcode::VCmpNeqF16, Opcode::CompareUnordNeF16},
|
||||
{Decoder::Opcode::VCmpxLtF16, Opcode::CompareMaskLtF16},
|
||||
{Decoder::Opcode::VCmpxEqF16, Opcode::CompareMaskEqF16},
|
||||
{Decoder::Opcode::VCmpxLeF16, Opcode::CompareMaskLeF16},
|
||||
{Decoder::Opcode::VCmpxGtF16, Opcode::CompareMaskGtF16},
|
||||
{Decoder::Opcode::VCmpxGeF16, Opcode::CompareMaskGeF16},
|
||||
{Decoder::Opcode::VCmpxNeqF16, Opcode::CompareMaskUnordNeF16},
|
||||
{Decoder::Opcode::VCmpxNltF16, Opcode::CompareMaskUnordGeF16},
|
||||
{Decoder::Opcode::VCmpxLtF32, Opcode::CompareMaskLtF32},
|
||||
{Decoder::Opcode::VCmpxEqF32, Opcode::CompareMaskEqF32},
|
||||
{Decoder::Opcode::VCmpxLeF32, Opcode::CompareMaskLeF32},
|
||||
{Decoder::Opcode::VCmpxGtF32, Opcode::CompareMaskGtF32},
|
||||
{Decoder::Opcode::VCmpxLgF32, Opcode::CompareMaskNeF32},
|
||||
{Decoder::Opcode::VCmpxGeF32, Opcode::CompareMaskGeF32},
|
||||
{Decoder::Opcode::VCmpxNgeF32, Opcode::CompareMaskUnordLtF32},
|
||||
{Decoder::Opcode::VCmpxNlgF32, Opcode::CompareMaskUnordEqF32},
|
||||
{Decoder::Opcode::VCmpxNgtF32, Opcode::CompareMaskUnordLeF32},
|
||||
{Decoder::Opcode::VCmpxNleF32, Opcode::CompareMaskUnordGtF32},
|
||||
{Decoder::Opcode::VCmpxNeqF32, Opcode::CompareMaskUnordNeF32},
|
||||
{Decoder::Opcode::VCmpxNltF32, Opcode::CompareMaskUnordGeF32},
|
||||
{Decoder::Opcode::VCmpFI32, Opcode::CompareFalse},
|
||||
{Decoder::Opcode::VCmpLtI32, Opcode::CompareLtI32},
|
||||
{Decoder::Opcode::VCmpEqI32, Opcode::CompareEqI32},
|
||||
{Decoder::Opcode::VCmpLeI32, Opcode::CompareLeI32},
|
||||
{Decoder::Opcode::VCmpGtI32, Opcode::CompareGtI32},
|
||||
{Decoder::Opcode::VCmpNeI32, Opcode::CompareNeI32},
|
||||
{Decoder::Opcode::VCmpGeI32, Opcode::CompareGeI32},
|
||||
{Decoder::Opcode::VCmpTI32, Opcode::CompareTrue},
|
||||
{Decoder::Opcode::VCmpLtI16, Opcode::CompareLtI16},
|
||||
{Decoder::Opcode::VCmpEqI16, Opcode::CompareEqI16},
|
||||
{Decoder::Opcode::VCmpLeI16, Opcode::CompareLeI16},
|
||||
{Decoder::Opcode::VCmpGtI16, Opcode::CompareGtI16},
|
||||
{Decoder::Opcode::VCmpNeI16, Opcode::CompareNeI16},
|
||||
{Decoder::Opcode::VCmpGeI16, Opcode::CompareGeI16},
|
||||
{Decoder::Opcode::VCmpxLtI32, Opcode::CompareMaskLtI32},
|
||||
{Decoder::Opcode::VCmpxEqI32, Opcode::CompareMaskEqI32},
|
||||
{Decoder::Opcode::VCmpxLeI32, Opcode::CompareMaskLeI32},
|
||||
{Decoder::Opcode::VCmpxGtI32, Opcode::CompareMaskGtI32},
|
||||
{Decoder::Opcode::VCmpxNeI32, Opcode::CompareMaskNeI32},
|
||||
{Decoder::Opcode::VCmpxGeI32, Opcode::CompareMaskGeI32},
|
||||
{Decoder::Opcode::VCmpLtU16, Opcode::CompareLtU16},
|
||||
{Decoder::Opcode::VCmpEqU16, Opcode::CompareEqU16},
|
||||
{Decoder::Opcode::VCmpLeU16, Opcode::CompareLeU16},
|
||||
{Decoder::Opcode::VCmpGtU16, Opcode::CompareGtU16},
|
||||
{Decoder::Opcode::VCmpNeU16, Opcode::CompareNeU16},
|
||||
{Decoder::Opcode::VCmpGeU16, Opcode::CompareGeU16},
|
||||
{Decoder::Opcode::VCmpFU32, Opcode::CompareFalse},
|
||||
{Decoder::Opcode::VCmpLtU32, Opcode::CompareLtU32},
|
||||
{Decoder::Opcode::VCmpEqU32, Opcode::CompareEqU32},
|
||||
{Decoder::Opcode::VCmpLeU32, Opcode::CompareLeU32},
|
||||
{Decoder::Opcode::VCmpGtU32, Opcode::CompareGtU32},
|
||||
{Decoder::Opcode::VCmpNeU32, Opcode::CompareNeU32},
|
||||
{Decoder::Opcode::VCmpGeU32, Opcode::CompareGeU32},
|
||||
{Decoder::Opcode::VCmpTU32, Opcode::CompareTrue},
|
||||
{Decoder::Opcode::VCmpEqI64, Opcode::CompareEqU64},
|
||||
{Decoder::Opcode::VCmpGtU64, Opcode::CompareGtU64},
|
||||
{Decoder::Opcode::VCmpNeU64, Opcode::CompareNeU64},
|
||||
{Decoder::Opcode::VCmpxLtU32, Opcode::CompareMaskLtU32},
|
||||
{Decoder::Opcode::VCmpxEqU32, Opcode::CompareMaskEqU32},
|
||||
{Decoder::Opcode::VCmpxLeU32, Opcode::CompareMaskLeU32},
|
||||
{Decoder::Opcode::VCmpxGtU32, Opcode::CompareMaskGtU32},
|
||||
{Decoder::Opcode::VCmpxNeU32, Opcode::CompareMaskNeU32},
|
||||
{Decoder::Opcode::VCmpxGeU32, Opcode::CompareMaskGeU32},
|
||||
{Decoder::Opcode::SLoadDword, Opcode::SLoadDword},
|
||||
{Decoder::Opcode::SBufferLoadDword, Opcode::SBufferLoadDword},
|
||||
{Decoder::Opcode::BufferLoadDword, Opcode::BufferLoadDword},
|
||||
{Decoder::Opcode::BufferLoadSbyte, Opcode::BufferLoadSbyte},
|
||||
{Decoder::Opcode::BufferLoadSshort, Opcode::BufferLoadSshort},
|
||||
{Decoder::Opcode::BufferStoreByte, Opcode::BufferStoreByte},
|
||||
{Decoder::Opcode::BufferStoreShort, Opcode::BufferStoreShort},
|
||||
{Decoder::Opcode::BufferStoreDword, Opcode::BufferStoreDword},
|
||||
{Decoder::Opcode::BufferAtomicSwap, Opcode::AtomicSwapU32},
|
||||
{Decoder::Opcode::BufferAtomicAdd, Opcode::AtomicAddU32},
|
||||
{Decoder::Opcode::BufferAtomicSub, Opcode::AtomicSubU32},
|
||||
{Decoder::Opcode::BufferAtomicSMin, Opcode::AtomicSMinI32},
|
||||
{Decoder::Opcode::BufferAtomicUMin, Opcode::AtomicUMinU32},
|
||||
{Decoder::Opcode::BufferAtomicSMax, Opcode::AtomicSMaxI32},
|
||||
{Decoder::Opcode::BufferAtomicUMax, Opcode::AtomicUMaxU32},
|
||||
{Decoder::Opcode::BufferAtomicAnd, Opcode::AtomicAndU32},
|
||||
{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},
|
||||
{Decoder::Opcode::FlatStoreByte, Opcode::FlatStoreByte},
|
||||
{Decoder::Opcode::FlatStoreShort, Opcode::FlatStoreShort},
|
||||
{Decoder::Opcode::FlatStoreDword, Opcode::FlatStoreDword},
|
||||
{Decoder::Opcode::DsAddU32, Opcode::AtomicAddU32},
|
||||
{Decoder::Opcode::DsAddRtnU32, Opcode::AtomicAddU32},
|
||||
{Decoder::Opcode::DsSubU32, Opcode::AtomicSubU32},
|
||||
{Decoder::Opcode::DsSubRtnU32, Opcode::AtomicSubU32},
|
||||
{Decoder::Opcode::DsMinI32, Opcode::AtomicSMinI32},
|
||||
{Decoder::Opcode::DsMinRtnI32, Opcode::AtomicSMinI32},
|
||||
{Decoder::Opcode::DsMaxI32, Opcode::AtomicSMaxI32},
|
||||
{Decoder::Opcode::DsMaxRtnI32, Opcode::AtomicSMaxI32},
|
||||
{Decoder::Opcode::DsMinU32, Opcode::AtomicUMinU32},
|
||||
{Decoder::Opcode::DsMinRtnU32, Opcode::AtomicUMinU32},
|
||||
{Decoder::Opcode::DsMaxU32, Opcode::AtomicUMaxU32},
|
||||
{Decoder::Opcode::DsMaxRtnU32, Opcode::AtomicUMaxU32},
|
||||
{Decoder::Opcode::DsAndB32, Opcode::AtomicAndU32},
|
||||
{Decoder::Opcode::DsAndRtnB32, Opcode::AtomicAndU32},
|
||||
{Decoder::Opcode::DsOrB32, Opcode::AtomicOrU32},
|
||||
{Decoder::Opcode::DsOrRtnB32, Opcode::AtomicOrU32},
|
||||
{Decoder::Opcode::DsXorB32, Opcode::AtomicXorU32},
|
||||
{Decoder::Opcode::DsXorRtnB32, Opcode::AtomicXorU32},
|
||||
{Decoder::Opcode::DsWrxchgRtnB32, Opcode::AtomicSwapU32},
|
||||
{Decoder::Opcode::DsReadSbyte, Opcode::DsReadSbyte},
|
||||
{Decoder::Opcode::DsReadUbyte, Opcode::DsReadUbyte},
|
||||
{Decoder::Opcode::DsReadSshort, Opcode::DsReadSshort},
|
||||
{Decoder::Opcode::DsReadUshort, Opcode::DsReadUshort},
|
||||
{Decoder::Opcode::DsRead2B32, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DsReadB32, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DsReadB64, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DsRead2B64, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DsRead2St64B64, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DsReadB96, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DsReadB128, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DsWriteByte, Opcode::DsWriteByte},
|
||||
{Decoder::Opcode::DsWriteShort, Opcode::DsWriteShort},
|
||||
{Decoder::Opcode::DsWrite2B32, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DsWrite2St64B32, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DsWrite2B64, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DsWrite2St64B64, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DsWriteB32, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DsWriteB64, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DsWriteB96, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DsWriteB128, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DsMinF32, Opcode::DsMinF32},
|
||||
{Decoder::Opcode::DsMaxF32, Opcode::DsMaxF32},
|
||||
{Decoder::Opcode::DsConsume, Opcode::DsConsume},
|
||||
{Decoder::Opcode::DsAppend, Opcode::DsAppend},
|
||||
{Decoder::Opcode::DsWriteAddtidB32, Opcode::DsWriteAddtidB32},
|
||||
{Decoder::Opcode::DsReadAddtidB32, Opcode::DsReadAddtidB32},
|
||||
{Decoder::Opcode::ImageGetResinfo, Opcode::ImageGetResinfo},
|
||||
{Decoder::Opcode::ImageGetLod, Opcode::ImageGetLod},
|
||||
{Decoder::Opcode::ImageLoad, Opcode::ImageLoad},
|
||||
{Decoder::Opcode::ImageLoadMip, Opcode::ImageLoad},
|
||||
{Decoder::Opcode::ImageStore, Opcode::ImageStore},
|
||||
{Decoder::Opcode::ImageStoreMip, Opcode::ImageStore},
|
||||
{Decoder::Opcode::ImageAtomicAdd, Opcode::AtomicAddU32},
|
||||
{Decoder::Opcode::ImageAtomicUMin, Opcode::AtomicUMinU32},
|
||||
{Decoder::Opcode::ImageAtomicUMax, Opcode::AtomicUMaxU32},
|
||||
{Decoder::Opcode::ImageAtomicAnd, Opcode::AtomicAndU32},
|
||||
{Decoder::Opcode::ImageAtomicOr, Opcode::AtomicOrU32},
|
||||
{Decoder::Opcode::ImageAtomicXor, Opcode::AtomicXorU32},
|
||||
{Decoder::Opcode::ImageSample, Opcode::ImageSample},
|
||||
{Decoder::Opcode::ImageGather4Lz, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::ImageGather4C, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::ImageGather4CLz, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::ImageGather4LzO, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::ImageGather4CO, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::ImageGather4CLzO, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::ImageGather4H, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::S_MOV_B32, Opcode::MoveU32},
|
||||
{Decoder::Opcode::S_MOV_B64, Opcode::MoveU64},
|
||||
{Decoder::Opcode::S_WQM_B64, Opcode::WqmB64},
|
||||
{Decoder::Opcode::S_MOVK_I32, Opcode::MoveU32},
|
||||
{Decoder::Opcode::S_ABS_I32, Opcode::AbsI32},
|
||||
{Decoder::Opcode::S_BREV_B32, Opcode::BitReverseU32},
|
||||
{Decoder::Opcode::S_BCNT1_I32_B32, Opcode::BitCountU32},
|
||||
{Decoder::Opcode::S_BCNT1_I32_B64, Opcode::BitCountU64},
|
||||
{Decoder::Opcode::S_FF1_I32_B32, Opcode::FindLsbU32},
|
||||
{Decoder::Opcode::S_FF1_I32_B64, Opcode::FindLsbU64},
|
||||
{Decoder::Opcode::S_FLBIT_I32_B64, Opcode::FindMsbFromHighU64},
|
||||
{Decoder::Opcode::S_BITREPLICATE_B64_B32, Opcode::BitReplicateB64B32},
|
||||
{Decoder::Opcode::S_NOT_B32, Opcode::BitwiseNotU32},
|
||||
{Decoder::Opcode::S_NOT_B64, Opcode::BitwiseNotU64},
|
||||
{Decoder::Opcode::S_ADD_U32, Opcode::IAddU32},
|
||||
{Decoder::Opcode::S_SUB_U32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::S_ADD_I32, Opcode::IAddU32},
|
||||
{Decoder::Opcode::S_SUB_I32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::S_MIN_I32, Opcode::IMinI32},
|
||||
{Decoder::Opcode::S_MAX_I32, Opcode::IMaxI32},
|
||||
{Decoder::Opcode::S_MIN_U32, Opcode::UMinU32},
|
||||
{Decoder::Opcode::S_MAX_U32, Opcode::UMaxU32},
|
||||
{Decoder::Opcode::S_AND_B32, Opcode::BitwiseAndU32},
|
||||
{Decoder::Opcode::S_AND_B64, Opcode::BitwiseAndU64},
|
||||
{Decoder::Opcode::S_ANDN2_B32, Opcode::BitwiseAndNotU32},
|
||||
{Decoder::Opcode::S_ANDN2_B64, Opcode::BitwiseAndNotU64},
|
||||
{Decoder::Opcode::S_OR_B32, Opcode::BitwiseOrU32},
|
||||
{Decoder::Opcode::S_OR_B64, Opcode::BitwiseOrU64},
|
||||
{Decoder::Opcode::S_ORN2_B32, Opcode::BitwiseOrNotU32},
|
||||
{Decoder::Opcode::S_ORN2_B64, Opcode::BitwiseOrNotU64},
|
||||
{Decoder::Opcode::S_XOR_B32, Opcode::BitwiseXorU32},
|
||||
{Decoder::Opcode::S_XOR_B64, Opcode::BitwiseXorU64},
|
||||
{Decoder::Opcode::S_NAND_B32, Opcode::BitwiseNandU32},
|
||||
{Decoder::Opcode::S_NAND_B64, Opcode::BitwiseNandU64},
|
||||
{Decoder::Opcode::S_NOR_B32, Opcode::BitwiseNorU32},
|
||||
{Decoder::Opcode::S_NOR_B64, Opcode::BitwiseNorU64},
|
||||
{Decoder::Opcode::S_XNOR_B32, Opcode::BitwiseXnorU32},
|
||||
{Decoder::Opcode::S_XNOR_B64, Opcode::BitwiseXnorU64},
|
||||
{Decoder::Opcode::S_LSHL_B32, Opcode::ShiftLeftLogicalU32},
|
||||
{Decoder::Opcode::S_LSHL_B64, Opcode::ShiftLeftLogicalU64},
|
||||
{Decoder::Opcode::S_LSHR_B32, Opcode::ShiftRightLogicalU32},
|
||||
{Decoder::Opcode::S_LSHR_B64, Opcode::ShiftRightLogicalU64},
|
||||
{Decoder::Opcode::S_ASHR_I32, Opcode::ShiftRightArithmeticI32},
|
||||
{Decoder::Opcode::S_MUL_I32, Opcode::IMulU32},
|
||||
{Decoder::Opcode::S_MUL_HI_U32, Opcode::UMulHighU32},
|
||||
{Decoder::Opcode::S_MULK_I32, Opcode::IMulU32},
|
||||
{Decoder::Opcode::S_BFE_U32, Opcode::BitFieldExtractU32},
|
||||
{Decoder::Opcode::S_BFE_I32, Opcode::BitFieldExtractI32},
|
||||
{Decoder::Opcode::S_BFE_U64, Opcode::BitFieldExtractU64},
|
||||
{Decoder::Opcode::S_BFM_B32, Opcode::BitFieldMaskU32},
|
||||
{Decoder::Opcode::S_BFM_B64, Opcode::BitFieldMaskU64},
|
||||
{Decoder::Opcode::S_BITCMP0_B32, Opcode::BitCompare0B32},
|
||||
{Decoder::Opcode::S_BITCMP1_B32, Opcode::BitCompare1B32},
|
||||
{Decoder::Opcode::S_PACK_LL_B32_B16, Opcode::PackLowLowU16},
|
||||
{Decoder::Opcode::S_PACK_LH_B32_B16, Opcode::PackLowHighU16},
|
||||
{Decoder::Opcode::S_PACK_HH_B32_B16, Opcode::PackHighHighU16},
|
||||
{Decoder::Opcode::S_CMP_EQ_I32, Opcode::CompareEqI32},
|
||||
{Decoder::Opcode::S_CMP_LG_I32, Opcode::CompareNeI32},
|
||||
{Decoder::Opcode::S_CMP_GT_I32, Opcode::CompareGtI32},
|
||||
{Decoder::Opcode::S_CMP_GE_I32, Opcode::CompareGeI32},
|
||||
{Decoder::Opcode::S_CMP_LT_I32, Opcode::CompareLtI32},
|
||||
{Decoder::Opcode::S_CMP_LE_I32, Opcode::CompareLeI32},
|
||||
{Decoder::Opcode::S_CMP_EQ_U32, Opcode::CompareEqU32},
|
||||
{Decoder::Opcode::S_CMP_LG_U32, Opcode::CompareNeU32},
|
||||
{Decoder::Opcode::S_CMP_GT_U32, Opcode::CompareGtU32},
|
||||
{Decoder::Opcode::S_CMP_GE_U32, Opcode::CompareGeU32},
|
||||
{Decoder::Opcode::S_CMP_LT_U32, Opcode::CompareLtU32},
|
||||
{Decoder::Opcode::S_CMP_LE_U32, Opcode::CompareLeU32},
|
||||
{Decoder::Opcode::S_CMP_EQ_U64, Opcode::CompareEqU64},
|
||||
{Decoder::Opcode::S_CMP_LG_U64, Opcode::CompareNeU64},
|
||||
{Decoder::Opcode::V_READFIRSTLANE_B32, Opcode::ReadFirstLaneU32},
|
||||
{Decoder::Opcode::V_READLANE_B32, Opcode::ReadLaneU32},
|
||||
{Decoder::Opcode::V_WRITELANE_B32, Opcode::WriteLaneU32},
|
||||
{Decoder::Opcode::V_PERMLANE16_B32, Opcode::Permlane16B32},
|
||||
{Decoder::Opcode::V_PERMLANEX16_B32, Opcode::Permlanex16B32},
|
||||
{Decoder::Opcode::V_CVT_F32_I32, Opcode::ConvertI32ToF32},
|
||||
{Decoder::Opcode::V_CVT_F32_U32, Opcode::ConvertU32ToF32},
|
||||
{Decoder::Opcode::V_CVT_U32_F32, Opcode::ConvertF32ToU32},
|
||||
{Decoder::Opcode::V_CVT_I32_F32, Opcode::ConvertF32ToI32},
|
||||
{Decoder::Opcode::V_CVT_F16_F32, Opcode::ConvertF32ToF16},
|
||||
{Decoder::Opcode::V_CVT_F32_F16, Opcode::ConvertF16ToF32},
|
||||
{Decoder::Opcode::V_CVT_F16_U16, Opcode::ConvertU16ToF16},
|
||||
{Decoder::Opcode::V_CVT_U16_F16, Opcode::ConvertF16ToU16},
|
||||
{Decoder::Opcode::V_CVT_F16_I16, Opcode::ConvertI16ToF16},
|
||||
{Decoder::Opcode::V_CVT_I16_F16, Opcode::ConvertF16ToI16},
|
||||
{Decoder::Opcode::V_CVT_RPI_I32_F32, Opcode::ConvertRoundPlusInfF32ToI32},
|
||||
{Decoder::Opcode::V_CVT_FLR_I32_F32, Opcode::ConvertFloorF32ToI32},
|
||||
{Decoder::Opcode::V_FREXP_EXP_I32_F32, Opcode::FrexpExpI32F32},
|
||||
{Decoder::Opcode::V_FREXP_MANT_F32, Opcode::FrexpMantF32},
|
||||
{Decoder::Opcode::V_CVT_OFF_F32_I4, Opcode::ConvertI4ToOffsetF32},
|
||||
{Decoder::Opcode::V_RCP_F32, Opcode::RcpF32},
|
||||
{Decoder::Opcode::V_RCP_IFLAG_F32, Opcode::RcpIflagF32},
|
||||
{Decoder::Opcode::V_FRACT_F32, Opcode::FractF32},
|
||||
{Decoder::Opcode::V_TRUNC_F32, Opcode::TruncF32},
|
||||
{Decoder::Opcode::V_CEIL_F32, Opcode::CeilF32},
|
||||
{Decoder::Opcode::V_RNDNE_F32, Opcode::RoundEvenF32},
|
||||
{Decoder::Opcode::V_FLOOR_F32, Opcode::FloorF32},
|
||||
{Decoder::Opcode::V_EXP_F32, Opcode::Exp2F32},
|
||||
{Decoder::Opcode::V_LOG_F32, Opcode::Log2F32},
|
||||
{Decoder::Opcode::V_RSQ_F32, Opcode::InverseSqrtF32},
|
||||
{Decoder::Opcode::V_SQRT_F32, Opcode::SqrtF32},
|
||||
{Decoder::Opcode::V_RCP_F16, Opcode::RcpF16},
|
||||
{Decoder::Opcode::V_SQRT_F16, Opcode::SqrtF16},
|
||||
{Decoder::Opcode::V_RSQ_F16, Opcode::InverseSqrtF16},
|
||||
{Decoder::Opcode::V_LOG_F16, Opcode::Log2F16},
|
||||
{Decoder::Opcode::V_EXP_F16, Opcode::Exp2F16},
|
||||
{Decoder::Opcode::V_FLOOR_F16, Opcode::FloorF16},
|
||||
{Decoder::Opcode::V_CEIL_F16, Opcode::CeilF16},
|
||||
{Decoder::Opcode::V_TRUNC_F16, Opcode::TruncF16},
|
||||
{Decoder::Opcode::V_RNDNE_F16, Opcode::RoundEvenF16},
|
||||
{Decoder::Opcode::V_SIN_F32, Opcode::SinF32},
|
||||
{Decoder::Opcode::V_COS_F32, Opcode::CosF32},
|
||||
{Decoder::Opcode::V_NOT_B32, Opcode::BitwiseNotU32},
|
||||
{Decoder::Opcode::V_BFREV_B32, Opcode::BitReverseU32},
|
||||
{Decoder::Opcode::V_FFBL_B32, Opcode::FindLsbU32},
|
||||
{Decoder::Opcode::V_FFBH_U32, Opcode::FindMsbFromHighU32},
|
||||
{Decoder::Opcode::V_ADD_F32, Opcode::FAddF32},
|
||||
{Decoder::Opcode::V_SUB_F32, Opcode::FSubF32},
|
||||
{Decoder::Opcode::V_SUBREV_F32, Opcode::FSubF32},
|
||||
{Decoder::Opcode::V_MUL_F32, Opcode::FMulF32},
|
||||
{Decoder::Opcode::V_MAC_F32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::V_MADMK_F32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::V_MADAK_F32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::V_MIN_F32, Opcode::FMinF32},
|
||||
{Decoder::Opcode::V_MAX_F32, Opcode::FMaxF32},
|
||||
{Decoder::Opcode::V_MIN_I32, Opcode::IMinI32},
|
||||
{Decoder::Opcode::V_MAX_I32, Opcode::IMaxI32},
|
||||
{Decoder::Opcode::V_MAD_F32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::V_MAD_I32_I24, Opcode::IMadI24U32},
|
||||
{Decoder::Opcode::V_MAD_U32_U24, Opcode::UMadU24U32},
|
||||
{Decoder::Opcode::V_CUBEID_F32, Opcode::CubeIdF32},
|
||||
{Decoder::Opcode::V_CUBESC_F32, Opcode::CubeScF32},
|
||||
{Decoder::Opcode::V_CUBETC_F32, Opcode::CubeTcF32},
|
||||
{Decoder::Opcode::V_CUBEMA_F32, Opcode::CubeMaF32},
|
||||
{Decoder::Opcode::V_FMA_F32, Opcode::FMadF32},
|
||||
{Decoder::Opcode::V_BFE_U32, Opcode::BitFieldExtract3U32},
|
||||
{Decoder::Opcode::V_BFE_I32, Opcode::BitFieldExtract3I32},
|
||||
{Decoder::Opcode::V_BFI_B32, Opcode::BitFieldInsertSelectU32},
|
||||
{Decoder::Opcode::V_ALIGNBIT_B32, Opcode::AlignBitU32},
|
||||
{Decoder::Opcode::V_ALIGNBYTE_B32, Opcode::AlignByteU32},
|
||||
{Decoder::Opcode::V_MIN3_F32, Opcode::FMin3F32},
|
||||
{Decoder::Opcode::V_MIN3_I32, Opcode::IMin3I32},
|
||||
{Decoder::Opcode::V_MIN3_U32, Opcode::UMin3U32},
|
||||
{Decoder::Opcode::V_MIN3_F16, Opcode::Min3F16},
|
||||
{Decoder::Opcode::V_MAX3_F32, Opcode::FMax3F32},
|
||||
{Decoder::Opcode::V_MAX3_I32, Opcode::IMax3I32},
|
||||
{Decoder::Opcode::V_MAX3_U32, Opcode::UMax3U32},
|
||||
{Decoder::Opcode::V_MAX3_F16, Opcode::Max3F16},
|
||||
{Decoder::Opcode::V_MED3_F32, Opcode::FMed3F32},
|
||||
{Decoder::Opcode::V_MED3_I32, Opcode::IMed3I32},
|
||||
{Decoder::Opcode::V_MED3_U32, Opcode::UMed3U32},
|
||||
{Decoder::Opcode::V_MED3_F16, Opcode::Med3F16},
|
||||
{Decoder::Opcode::V_MED3_I16, Opcode::IMed3I16},
|
||||
{Decoder::Opcode::V_SAD_U32, Opcode::SadU32},
|
||||
{Decoder::Opcode::V_ADD3_U32, Opcode::IAdd3U32},
|
||||
{Decoder::Opcode::V_LSHL_ADD_U32, Opcode::ShiftLeftAddU32},
|
||||
{Decoder::Opcode::V_ADD_LSHL_U32, Opcode::AddShiftLeftU32},
|
||||
{Decoder::Opcode::V_XAD_U32, Opcode::XorAddU32},
|
||||
{Decoder::Opcode::V_LSHL_OR_B32, Opcode::ShiftLeftOrU32},
|
||||
{Decoder::Opcode::V_AND_OR_B32, Opcode::BitwiseAndOrU32},
|
||||
{Decoder::Opcode::V_OR3_B32, Opcode::BitwiseOr3U32},
|
||||
{Decoder::Opcode::V_XOR3_B32, Opcode::BitwiseXor3U32},
|
||||
{Decoder::Opcode::V_ADD_NC_U32, Opcode::IAddU32},
|
||||
{Decoder::Opcode::V_SUB_NC_U32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::V_SUBREV_NC_U32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::V_ADD_NC_U16, Opcode::IAddU16},
|
||||
{Decoder::Opcode::V_SUB_NC_U16, Opcode::ISubI16},
|
||||
{Decoder::Opcode::V_MAX_U16, Opcode::UMaxU16},
|
||||
{Decoder::Opcode::V_MAX_I16, Opcode::IMaxI16},
|
||||
{Decoder::Opcode::V_MIN_U16, Opcode::UMinU16},
|
||||
{Decoder::Opcode::V_MIN_I16, Opcode::IMinI16},
|
||||
{Decoder::Opcode::V_ADD_NC_I16, Opcode::IAddU16},
|
||||
{Decoder::Opcode::V_SUB_NC_I16, Opcode::ISubI16},
|
||||
{Decoder::Opcode::V_MUL_I32_I24, Opcode::IMulI24U32},
|
||||
{Decoder::Opcode::V_MUL_U32_U24, Opcode::UMulU24U32},
|
||||
{Decoder::Opcode::V_MUL_LO_U32, Opcode::IMulU32},
|
||||
{Decoder::Opcode::V_MUL_HI_U32, Opcode::UMulHighU32},
|
||||
{Decoder::Opcode::V_MUL_LO_I32, Opcode::IMulU32},
|
||||
{Decoder::Opcode::V_MUL_HI_I32, Opcode::SMulHighI32},
|
||||
{Decoder::Opcode::V_ADD_I32, Opcode::IAddU32},
|
||||
{Decoder::Opcode::V_SUB_I32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::V_SUBREV_I32, Opcode::ISubU32},
|
||||
{Decoder::Opcode::V_BFM_B32, Opcode::BitFieldMaskU32},
|
||||
{Decoder::Opcode::V_CVT_PKRTZ_F16_F32, Opcode::PackF32ToF16Rtz},
|
||||
{Decoder::Opcode::V_LDEXP_F32, Opcode::LdexpF32},
|
||||
{Decoder::Opcode::V_CVT_PKNORM_I16_F32, Opcode::PackSnorm2x16F32},
|
||||
{Decoder::Opcode::V_CVT_PKNORM_U16_F32, Opcode::PackUnorm2x16F32},
|
||||
{Decoder::Opcode::V_CVT_PK_U16_U32, Opcode::PackU16U32},
|
||||
{Decoder::Opcode::V_CVT_PK_I16_I32, Opcode::PackU16U32},
|
||||
{Decoder::Opcode::V_CVT_PK_U8_F32, Opcode::PackU8F32},
|
||||
{Decoder::Opcode::V_PACK_B32_F16, Opcode::PackB32F16},
|
||||
{Decoder::Opcode::V_PK_MAD_I16, Opcode::PackedMadI16},
|
||||
{Decoder::Opcode::V_PK_MUL_LO_U16, Opcode::PackedMulLoU16},
|
||||
{Decoder::Opcode::V_PK_ADD_I16, Opcode::PackedAddI16},
|
||||
{Decoder::Opcode::V_PK_SUB_I16, Opcode::PackedSubI16},
|
||||
{Decoder::Opcode::V_PK_LSHLREV_B16, Opcode::PackedLshlrevB16},
|
||||
{Decoder::Opcode::V_PK_LSHRREV_B16, Opcode::PackedLshrrevB16},
|
||||
{Decoder::Opcode::V_PK_ASHRREV_I16, Opcode::PackedAshrrevI16},
|
||||
{Decoder::Opcode::V_PK_MAX_I16, Opcode::PackedMaxI16},
|
||||
{Decoder::Opcode::V_PK_MIN_I16, Opcode::PackedMinI16},
|
||||
{Decoder::Opcode::V_PK_MAD_U16, Opcode::PackedMadU16},
|
||||
{Decoder::Opcode::V_PK_ADD_U16, Opcode::PackedAddU16},
|
||||
{Decoder::Opcode::V_PK_SUB_U16, Opcode::PackedSubU16},
|
||||
{Decoder::Opcode::V_PK_MAX_U16, Opcode::PackedMaxU16},
|
||||
{Decoder::Opcode::V_PK_MIN_U16, Opcode::PackedMinU16},
|
||||
{Decoder::Opcode::V_PK_ADD_F16, Opcode::PackedAddF16},
|
||||
{Decoder::Opcode::V_PK_MUL_F16, Opcode::PackedMulF16},
|
||||
{Decoder::Opcode::V_PK_MIN_F16, Opcode::PackedMinF16},
|
||||
{Decoder::Opcode::V_PK_MAX_F16, Opcode::PackedMaxF16},
|
||||
{Decoder::Opcode::V_PK_FMA_F16, Opcode::PackedFmaF16},
|
||||
{Decoder::Opcode::V_PK_FMAC_F16, Opcode::PackedFmaF16},
|
||||
{Decoder::Opcode::V_ADD_F16, Opcode::AddF16},
|
||||
{Decoder::Opcode::V_SUB_F16, Opcode::SubF16},
|
||||
{Decoder::Opcode::V_SUBREV_F16, Opcode::SubF16},
|
||||
{Decoder::Opcode::V_MUL_F16, Opcode::MulF16},
|
||||
{Decoder::Opcode::V_FMAC_F16, Opcode::FmaF16},
|
||||
{Decoder::Opcode::V_FMAMK_F16, Opcode::FmaF16},
|
||||
{Decoder::Opcode::V_FMAAK_F16, Opcode::FmaF16},
|
||||
{Decoder::Opcode::V_FMA_F16, Opcode::FmaF16},
|
||||
{Decoder::Opcode::V_MAD_MIXLO_F16, Opcode::MadMixF16},
|
||||
{Decoder::Opcode::V_MAD_MIXHI_F16, Opcode::MadMixF16},
|
||||
{Decoder::Opcode::V_AND_B32, Opcode::BitwiseAndU32},
|
||||
{Decoder::Opcode::V_OR_B32, Opcode::BitwiseOrU32},
|
||||
{Decoder::Opcode::V_XOR_B32, Opcode::BitwiseXorU32},
|
||||
{Decoder::Opcode::V_XNOR_B32, Opcode::BitwiseXnorU32},
|
||||
{Decoder::Opcode::V_BCNT_U32_B32, Opcode::BitCountAddU32},
|
||||
{Decoder::Opcode::V_MBCNT_LO_U32_B32, Opcode::MaskedBitCountLowU32},
|
||||
{Decoder::Opcode::V_MBCNT_HI_U32_B32, Opcode::MaskedBitCountHighU32},
|
||||
{Decoder::Opcode::V_LSHL_B32, Opcode::ShiftLeftLogicalU32},
|
||||
{Decoder::Opcode::V_LSHLREV_B32, Opcode::ShiftLeftLogicalU32},
|
||||
{Decoder::Opcode::V_LSHLREV_B16, Opcode::ShiftLeftLogicalU16},
|
||||
{Decoder::Opcode::V_LSHR_B32, Opcode::ShiftRightLogicalU32},
|
||||
{Decoder::Opcode::V_LSHRREV_B32, Opcode::ShiftRightLogicalU32},
|
||||
{Decoder::Opcode::V_LSHRREV_B16, Opcode::ShiftRightLogicalU16},
|
||||
{Decoder::Opcode::V_ASHR_I32, Opcode::ShiftRightArithmeticI32},
|
||||
{Decoder::Opcode::V_ASHRREV_I32, Opcode::ShiftRightArithmeticI32},
|
||||
{Decoder::Opcode::V_ASHRREV_I16, Opcode::ShiftRightArithmeticI16},
|
||||
{Decoder::Opcode::V_MIN_U32, Opcode::UMinU32},
|
||||
{Decoder::Opcode::V_MAX_U32, Opcode::UMaxU32},
|
||||
{Decoder::Opcode::V_MAX_F16, Opcode::MaxF16},
|
||||
{Decoder::Opcode::V_MIN_F16, Opcode::MinF16},
|
||||
{Decoder::Opcode::V_CMP_F_F32, Opcode::CompareFalse},
|
||||
{Decoder::Opcode::V_CMP_LT_F32, Opcode::CompareLtF32},
|
||||
{Decoder::Opcode::V_CMP_EQ_F32, Opcode::CompareEqF32},
|
||||
{Decoder::Opcode::V_CMP_LE_F32, Opcode::CompareLeF32},
|
||||
{Decoder::Opcode::V_CMP_GT_F32, Opcode::CompareGtF32},
|
||||
{Decoder::Opcode::V_CMP_LG_F32, Opcode::CompareNeF32},
|
||||
{Decoder::Opcode::V_CMP_GE_F32, Opcode::CompareGeF32},
|
||||
{Decoder::Opcode::V_CMP_O_F32, Opcode::CompareOrderedF32},
|
||||
{Decoder::Opcode::V_CMP_U_F32, Opcode::CompareUnorderedF32},
|
||||
{Decoder::Opcode::V_CMP_NGE_F32, Opcode::CompareUnordLtF32},
|
||||
{Decoder::Opcode::V_CMP_NLG_F32, Opcode::CompareUnordEqF32},
|
||||
{Decoder::Opcode::V_CMP_NGT_F32, Opcode::CompareUnordLeF32},
|
||||
{Decoder::Opcode::V_CMP_NLE_F32, Opcode::CompareUnordGtF32},
|
||||
{Decoder::Opcode::V_CMP_NEQ_F32, Opcode::CompareUnordNeF32},
|
||||
{Decoder::Opcode::V_CMP_NLT_F32, Opcode::CompareUnordGeF32},
|
||||
{Decoder::Opcode::V_CMP_TRU_F32, Opcode::CompareTrue},
|
||||
{Decoder::Opcode::V_CMP_CLASS_F32, Opcode::CompareClassF32},
|
||||
{Decoder::Opcode::V_CMP_LT_F16, Opcode::CompareLtF16},
|
||||
{Decoder::Opcode::V_CMP_EQ_F16, Opcode::CompareEqF16},
|
||||
{Decoder::Opcode::V_CMP_LE_F16, Opcode::CompareLeF16},
|
||||
{Decoder::Opcode::V_CMP_GT_F16, Opcode::CompareGtF16},
|
||||
{Decoder::Opcode::V_CMP_LG_F16, Opcode::CompareNeF16},
|
||||
{Decoder::Opcode::V_CMP_GE_F16, Opcode::CompareGeF16},
|
||||
{Decoder::Opcode::V_CMP_NEQ_F16, Opcode::CompareUnordNeF16},
|
||||
{Decoder::Opcode::V_CMPX_LT_F16, Opcode::CompareMaskLtF16},
|
||||
{Decoder::Opcode::V_CMPX_EQ_F16, Opcode::CompareMaskEqF16},
|
||||
{Decoder::Opcode::V_CMPX_LE_F16, Opcode::CompareMaskLeF16},
|
||||
{Decoder::Opcode::V_CMPX_GT_F16, Opcode::CompareMaskGtF16},
|
||||
{Decoder::Opcode::V_CMPX_GE_F16, Opcode::CompareMaskGeF16},
|
||||
{Decoder::Opcode::V_CMPX_NEQ_F16, Opcode::CompareMaskUnordNeF16},
|
||||
{Decoder::Opcode::V_CMPX_NLT_F16, Opcode::CompareMaskUnordGeF16},
|
||||
{Decoder::Opcode::V_CMPX_LT_F32, Opcode::CompareMaskLtF32},
|
||||
{Decoder::Opcode::V_CMPX_EQ_F32, Opcode::CompareMaskEqF32},
|
||||
{Decoder::Opcode::V_CMPX_LE_F32, Opcode::CompareMaskLeF32},
|
||||
{Decoder::Opcode::V_CMPX_GT_F32, Opcode::CompareMaskGtF32},
|
||||
{Decoder::Opcode::V_CMPX_LG_F32, Opcode::CompareMaskNeF32},
|
||||
{Decoder::Opcode::V_CMPX_GE_F32, Opcode::CompareMaskGeF32},
|
||||
{Decoder::Opcode::V_CMPX_NGE_F32, Opcode::CompareMaskUnordLtF32},
|
||||
{Decoder::Opcode::V_CMPX_NLG_F32, Opcode::CompareMaskUnordEqF32},
|
||||
{Decoder::Opcode::V_CMPX_NGT_F32, Opcode::CompareMaskUnordLeF32},
|
||||
{Decoder::Opcode::V_CMPX_NLE_F32, Opcode::CompareMaskUnordGtF32},
|
||||
{Decoder::Opcode::V_CMPX_NEQ_F32, Opcode::CompareMaskUnordNeF32},
|
||||
{Decoder::Opcode::V_CMPX_NLT_F32, Opcode::CompareMaskUnordGeF32},
|
||||
{Decoder::Opcode::V_CMP_F_I32, Opcode::CompareFalse},
|
||||
{Decoder::Opcode::V_CMP_LT_I32, Opcode::CompareLtI32},
|
||||
{Decoder::Opcode::V_CMP_EQ_I32, Opcode::CompareEqI32},
|
||||
{Decoder::Opcode::V_CMP_LE_I32, Opcode::CompareLeI32},
|
||||
{Decoder::Opcode::V_CMP_GT_I32, Opcode::CompareGtI32},
|
||||
{Decoder::Opcode::V_CMP_NE_I32, Opcode::CompareNeI32},
|
||||
{Decoder::Opcode::V_CMP_GE_I32, Opcode::CompareGeI32},
|
||||
{Decoder::Opcode::V_CMP_T_I32, Opcode::CompareTrue},
|
||||
{Decoder::Opcode::V_CMP_LT_I16, Opcode::CompareLtI16},
|
||||
{Decoder::Opcode::V_CMP_EQ_I16, Opcode::CompareEqI16},
|
||||
{Decoder::Opcode::V_CMP_LE_I16, Opcode::CompareLeI16},
|
||||
{Decoder::Opcode::V_CMP_GT_I16, Opcode::CompareGtI16},
|
||||
{Decoder::Opcode::V_CMP_NE_I16, Opcode::CompareNeI16},
|
||||
{Decoder::Opcode::V_CMP_GE_I16, Opcode::CompareGeI16},
|
||||
{Decoder::Opcode::V_CMPX_LT_I32, Opcode::CompareMaskLtI32},
|
||||
{Decoder::Opcode::V_CMPX_EQ_I32, Opcode::CompareMaskEqI32},
|
||||
{Decoder::Opcode::V_CMPX_LE_I32, Opcode::CompareMaskLeI32},
|
||||
{Decoder::Opcode::V_CMPX_GT_I32, Opcode::CompareMaskGtI32},
|
||||
{Decoder::Opcode::V_CMPX_NE_I32, Opcode::CompareMaskNeI32},
|
||||
{Decoder::Opcode::V_CMPX_GE_I32, Opcode::CompareMaskGeI32},
|
||||
{Decoder::Opcode::V_CMP_LT_U16, Opcode::CompareLtU16},
|
||||
{Decoder::Opcode::V_CMP_EQ_U16, Opcode::CompareEqU16},
|
||||
{Decoder::Opcode::V_CMP_LE_U16, Opcode::CompareLeU16},
|
||||
{Decoder::Opcode::V_CMP_GT_U16, Opcode::CompareGtU16},
|
||||
{Decoder::Opcode::V_CMP_NE_U16, Opcode::CompareNeU16},
|
||||
{Decoder::Opcode::V_CMP_GE_U16, Opcode::CompareGeU16},
|
||||
{Decoder::Opcode::V_CMP_F_U32, Opcode::CompareFalse},
|
||||
{Decoder::Opcode::V_CMP_LT_U32, Opcode::CompareLtU32},
|
||||
{Decoder::Opcode::V_CMP_EQ_U32, Opcode::CompareEqU32},
|
||||
{Decoder::Opcode::V_CMP_LE_U32, Opcode::CompareLeU32},
|
||||
{Decoder::Opcode::V_CMP_GT_U32, Opcode::CompareGtU32},
|
||||
{Decoder::Opcode::V_CMP_NE_U32, Opcode::CompareNeU32},
|
||||
{Decoder::Opcode::V_CMP_GE_U32, Opcode::CompareGeU32},
|
||||
{Decoder::Opcode::V_CMP_T_U32, Opcode::CompareTrue},
|
||||
{Decoder::Opcode::V_CMP_EQ_I64, Opcode::CompareEqU64},
|
||||
{Decoder::Opcode::V_CMP_GT_U64, Opcode::CompareGtU64},
|
||||
{Decoder::Opcode::V_CMP_NE_U64, Opcode::CompareNeU64},
|
||||
{Decoder::Opcode::V_CMPX_LT_U32, Opcode::CompareMaskLtU32},
|
||||
{Decoder::Opcode::V_CMPX_EQ_U32, Opcode::CompareMaskEqU32},
|
||||
{Decoder::Opcode::V_CMPX_LE_U32, Opcode::CompareMaskLeU32},
|
||||
{Decoder::Opcode::V_CMPX_GT_U32, Opcode::CompareMaskGtU32},
|
||||
{Decoder::Opcode::V_CMPX_NE_U32, Opcode::CompareMaskNeU32},
|
||||
{Decoder::Opcode::V_CMPX_GE_U32, Opcode::CompareMaskGeU32},
|
||||
{Decoder::Opcode::S_LOAD_DWORD, Opcode::SLoadDword},
|
||||
{Decoder::Opcode::S_BUFFER_LOAD_DWORD, Opcode::SBufferLoadDword},
|
||||
{Decoder::Opcode::BUFFER_LOAD_DWORD, Opcode::BufferLoadDword},
|
||||
{Decoder::Opcode::BUFFER_LOAD_SBYTE, Opcode::BufferLoadSbyte},
|
||||
{Decoder::Opcode::BUFFER_LOAD_SSHORT, Opcode::BufferLoadSshort},
|
||||
{Decoder::Opcode::BUFFER_STORE_BYTE, Opcode::BufferStoreByte},
|
||||
{Decoder::Opcode::BUFFER_STORE_SHORT, Opcode::BufferStoreShort},
|
||||
{Decoder::Opcode::BUFFER_STORE_DWORD, Opcode::BufferStoreDword},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_SWAP, Opcode::AtomicSwapU32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_ADD, Opcode::AtomicAddU32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_SUB, Opcode::AtomicSubU32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_SMIN, Opcode::AtomicSMinI32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_UMIN, Opcode::AtomicUMinU32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_SMAX, Opcode::AtomicSMaxI32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_UMAX, Opcode::AtomicUMaxU32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_AND, Opcode::AtomicAndU32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_OR, Opcode::AtomicOrU32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_XOR, Opcode::AtomicXorU32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_FMIN, Opcode::AtomicFMinF32},
|
||||
{Decoder::Opcode::BUFFER_ATOMIC_FMAX, Opcode::AtomicFMaxF32},
|
||||
{Decoder::Opcode::FLAT_LOAD_UBYTE, Opcode::FlatLoadUbyte},
|
||||
{Decoder::Opcode::FLAT_LOAD_SBYTE, Opcode::FlatLoadSbyte},
|
||||
{Decoder::Opcode::FLAT_LOAD_SSHORT, Opcode::FlatLoadSshort},
|
||||
{Decoder::Opcode::FLAT_STORE_BYTE, Opcode::FlatStoreByte},
|
||||
{Decoder::Opcode::FLAT_STORE_SHORT, Opcode::FlatStoreShort},
|
||||
{Decoder::Opcode::FLAT_STORE_DWORD, Opcode::FlatStoreDword},
|
||||
{Decoder::Opcode::DS_ADD_U32, Opcode::AtomicAddU32},
|
||||
{Decoder::Opcode::DS_ADD_RTN_U32, Opcode::AtomicAddU32},
|
||||
{Decoder::Opcode::DS_SUB_U32, Opcode::AtomicSubU32},
|
||||
{Decoder::Opcode::DS_SUB_RTN_U32, Opcode::AtomicSubU32},
|
||||
{Decoder::Opcode::DS_MIN_I32, Opcode::AtomicSMinI32},
|
||||
{Decoder::Opcode::DS_MIN_RTN_I32, Opcode::AtomicSMinI32},
|
||||
{Decoder::Opcode::DS_MAX_I32, Opcode::AtomicSMaxI32},
|
||||
{Decoder::Opcode::DS_MAX_RTN_I32, Opcode::AtomicSMaxI32},
|
||||
{Decoder::Opcode::DS_MIN_U32, Opcode::AtomicUMinU32},
|
||||
{Decoder::Opcode::DS_MIN_RTN_U32, Opcode::AtomicUMinU32},
|
||||
{Decoder::Opcode::DS_MAX_U32, Opcode::AtomicUMaxU32},
|
||||
{Decoder::Opcode::DS_MAX_RTN_U32, Opcode::AtomicUMaxU32},
|
||||
{Decoder::Opcode::DS_AND_B32, Opcode::AtomicAndU32},
|
||||
{Decoder::Opcode::DS_AND_RTN_B32, Opcode::AtomicAndU32},
|
||||
{Decoder::Opcode::DS_OR_B32, Opcode::AtomicOrU32},
|
||||
{Decoder::Opcode::DS_OR_RTN_B32, Opcode::AtomicOrU32},
|
||||
{Decoder::Opcode::DS_XOR_B32, Opcode::AtomicXorU32},
|
||||
{Decoder::Opcode::DS_XOR_RTN_B32, Opcode::AtomicXorU32},
|
||||
{Decoder::Opcode::DS_WRXCHG_RTN_B32, Opcode::AtomicSwapU32},
|
||||
{Decoder::Opcode::DS_READ_I8, Opcode::DsReadSbyte},
|
||||
{Decoder::Opcode::DS_READ_U8, Opcode::DsReadUbyte},
|
||||
{Decoder::Opcode::DS_READ_I16, Opcode::DsReadSshort},
|
||||
{Decoder::Opcode::DS_READ_U16, Opcode::DsReadUshort},
|
||||
{Decoder::Opcode::DS_READ2_B32, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DS_READ_B32, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DS_READ_B64, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DS_READ2_B64, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DS_READ2ST64_B64, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DS_READ_B96, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DS_READ_B128, Opcode::DsReadB32},
|
||||
{Decoder::Opcode::DS_WRITE_B8, Opcode::DsWriteByte},
|
||||
{Decoder::Opcode::DS_WRITE_B16, Opcode::DsWriteShort},
|
||||
{Decoder::Opcode::DS_WRITE2_B32, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DS_WRITE2ST64_B32, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DS_WRITE2_B64, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DS_WRITE2ST64_B64, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DS_WRITE_B32, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DS_WRITE_B64, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DS_WRITE_B96, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DS_WRITE_B128, Opcode::DsWriteB32},
|
||||
{Decoder::Opcode::DS_MIN_F32, Opcode::DsMinF32},
|
||||
{Decoder::Opcode::DS_MAX_F32, Opcode::DsMaxF32},
|
||||
{Decoder::Opcode::DS_CONSUME, Opcode::DsConsume},
|
||||
{Decoder::Opcode::DS_APPEND, Opcode::DsAppend},
|
||||
{Decoder::Opcode::DS_WRITE_ADDTID_B32, Opcode::DsWriteAddtidB32},
|
||||
{Decoder::Opcode::DS_READ_ADDTID_B32, Opcode::DsReadAddtidB32},
|
||||
{Decoder::Opcode::IMAGE_GET_RESINFO, Opcode::ImageGetResinfo},
|
||||
{Decoder::Opcode::IMAGE_GET_LOD, Opcode::ImageGetLod},
|
||||
{Decoder::Opcode::IMAGE_LOAD, Opcode::ImageLoad},
|
||||
{Decoder::Opcode::IMAGE_LOAD_MIP, Opcode::ImageLoad},
|
||||
{Decoder::Opcode::IMAGE_STORE, Opcode::ImageStore},
|
||||
{Decoder::Opcode::IMAGE_STORE_MIP, Opcode::ImageStore},
|
||||
{Decoder::Opcode::IMAGE_ATOMIC_ADD, Opcode::AtomicAddU32},
|
||||
{Decoder::Opcode::IMAGE_ATOMIC_UMIN, Opcode::AtomicUMinU32},
|
||||
{Decoder::Opcode::IMAGE_ATOMIC_UMAX, Opcode::AtomicUMaxU32},
|
||||
{Decoder::Opcode::IMAGE_ATOMIC_AND, Opcode::AtomicAndU32},
|
||||
{Decoder::Opcode::IMAGE_ATOMIC_OR, Opcode::AtomicOrU32},
|
||||
{Decoder::Opcode::IMAGE_ATOMIC_XOR, Opcode::AtomicXorU32},
|
||||
{Decoder::Opcode::IMAGE_SAMPLE, Opcode::ImageSample},
|
||||
{Decoder::Opcode::IMAGE_GATHER4_LZ, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::IMAGE_GATHER4_C, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::IMAGE_GATHER4_C_LZ, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::IMAGE_GATHER4_LZ_O, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::IMAGE_GATHER4_C_O, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::IMAGE_GATHER4_C_LZ_O, Opcode::ImageGather4},
|
||||
{Decoder::Opcode::IMAGE_GATHER4H, Opcode::ImageGather4},
|
||||
};
|
||||
|
||||
} // namespace
|
||||
@@ -456,80 +456,80 @@ std::optional<Opcode> LookupIrOpcode(Decoder::Opcode opcode) {
|
||||
|
||||
bool IsReversedBinary(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::VSubrevF32:
|
||||
case Decoder::Opcode::VSubrevF16:
|
||||
case Decoder::Opcode::VSubrevNcU32:
|
||||
case Decoder::Opcode::VLshlrevB32:
|
||||
case Decoder::Opcode::VLshrrevB32:
|
||||
case Decoder::Opcode::VAshrrevI32:
|
||||
case Decoder::Opcode::VLshlrevB16:
|
||||
case Decoder::Opcode::VLshrrevB16:
|
||||
case Decoder::Opcode::VAshrrevI16:
|
||||
case Decoder::Opcode::VSubrevI32: return true;
|
||||
case Decoder::Opcode::V_SUBREV_F32:
|
||||
case Decoder::Opcode::V_SUBREV_F16:
|
||||
case Decoder::Opcode::V_SUBREV_NC_U32:
|
||||
case Decoder::Opcode::V_LSHLREV_B32:
|
||||
case Decoder::Opcode::V_LSHRREV_B32:
|
||||
case Decoder::Opcode::V_ASHRREV_I32:
|
||||
case Decoder::Opcode::V_LSHLREV_B16:
|
||||
case Decoder::Opcode::V_LSHRREV_B16:
|
||||
case Decoder::Opcode::V_ASHRREV_I16:
|
||||
case Decoder::Opcode::V_SUBREV_I32: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
|
||||
bool IsVectorCarryOutOpcode(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::VAddI32:
|
||||
case Decoder::Opcode::VSubI32:
|
||||
case Decoder::Opcode::VSubrevI32: return true;
|
||||
case Decoder::Opcode::V_ADD_I32:
|
||||
case Decoder::Opcode::V_SUB_I32:
|
||||
case Decoder::Opcode::V_SUBREV_I32: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
|
||||
bool ScalarResultWritesSccNonZero(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SAbsI32:
|
||||
case Decoder::Opcode::SBcnt1I32B32:
|
||||
case Decoder::Opcode::SBcnt1I32B64:
|
||||
case Decoder::Opcode::SAndB32:
|
||||
case Decoder::Opcode::SAndn2B32:
|
||||
case Decoder::Opcode::SOrB32:
|
||||
case Decoder::Opcode::SOrn2B32:
|
||||
case Decoder::Opcode::SXorB32:
|
||||
case Decoder::Opcode::SNandB32:
|
||||
case Decoder::Opcode::SNorB32:
|
||||
case Decoder::Opcode::SXnorB32:
|
||||
case Decoder::Opcode::SNotB32:
|
||||
case Decoder::Opcode::SLshlB32:
|
||||
case Decoder::Opcode::SLshrB32:
|
||||
case Decoder::Opcode::SAshrI32:
|
||||
case Decoder::Opcode::SBfeU32:
|
||||
case Decoder::Opcode::SBfeI32:
|
||||
case Decoder::Opcode::SAndB64:
|
||||
case Decoder::Opcode::SAndn2B64:
|
||||
case Decoder::Opcode::SNotB64:
|
||||
case Decoder::Opcode::SOrB64:
|
||||
case Decoder::Opcode::SOrn2B64:
|
||||
case Decoder::Opcode::SXorB64:
|
||||
case Decoder::Opcode::SNandB64:
|
||||
case Decoder::Opcode::SNorB64:
|
||||
case Decoder::Opcode::SXnorB64:
|
||||
case Decoder::Opcode::SLshlB64:
|
||||
case Decoder::Opcode::SLshrB64:
|
||||
case Decoder::Opcode::SBfeU64:
|
||||
case Decoder::Opcode::SWqmB64: return true;
|
||||
case Decoder::Opcode::S_ABS_I32:
|
||||
case Decoder::Opcode::S_BCNT1_I32_B32:
|
||||
case Decoder::Opcode::S_BCNT1_I32_B64:
|
||||
case Decoder::Opcode::S_AND_B32:
|
||||
case Decoder::Opcode::S_ANDN2_B32:
|
||||
case Decoder::Opcode::S_OR_B32:
|
||||
case Decoder::Opcode::S_ORN2_B32:
|
||||
case Decoder::Opcode::S_XOR_B32:
|
||||
case Decoder::Opcode::S_NAND_B32:
|
||||
case Decoder::Opcode::S_NOR_B32:
|
||||
case Decoder::Opcode::S_XNOR_B32:
|
||||
case Decoder::Opcode::S_NOT_B32:
|
||||
case Decoder::Opcode::S_LSHL_B32:
|
||||
case Decoder::Opcode::S_LSHR_B32:
|
||||
case Decoder::Opcode::S_ASHR_I32:
|
||||
case Decoder::Opcode::S_BFE_U32:
|
||||
case Decoder::Opcode::S_BFE_I32:
|
||||
case Decoder::Opcode::S_AND_B64:
|
||||
case Decoder::Opcode::S_ANDN2_B64:
|
||||
case Decoder::Opcode::S_NOT_B64:
|
||||
case Decoder::Opcode::S_OR_B64:
|
||||
case Decoder::Opcode::S_ORN2_B64:
|
||||
case Decoder::Opcode::S_XOR_B64:
|
||||
case Decoder::Opcode::S_NAND_B64:
|
||||
case Decoder::Opcode::S_NOR_B64:
|
||||
case Decoder::Opcode::S_XNOR_B64:
|
||||
case Decoder::Opcode::S_LSHL_B64:
|
||||
case Decoder::Opcode::S_LSHR_B64:
|
||||
case Decoder::Opcode::S_BFE_U64:
|
||||
case Decoder::Opcode::S_WQM_B64: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
|
||||
bool ScalarResultIs64Bit(Decoder::Opcode opcode) {
|
||||
switch (opcode) {
|
||||
case Decoder::Opcode::SAndB64:
|
||||
case Decoder::Opcode::SAndn2B64:
|
||||
case Decoder::Opcode::SNotB64:
|
||||
case Decoder::Opcode::SOrB64:
|
||||
case Decoder::Opcode::SOrn2B64:
|
||||
case Decoder::Opcode::SXorB64:
|
||||
case Decoder::Opcode::SNandB64:
|
||||
case Decoder::Opcode::SNorB64:
|
||||
case Decoder::Opcode::SXnorB64:
|
||||
case Decoder::Opcode::SLshlB64:
|
||||
case Decoder::Opcode::SLshrB64:
|
||||
case Decoder::Opcode::SBfeU64:
|
||||
case Decoder::Opcode::SWqmB64: return true;
|
||||
case Decoder::Opcode::S_AND_B64:
|
||||
case Decoder::Opcode::S_ANDN2_B64:
|
||||
case Decoder::Opcode::S_NOT_B64:
|
||||
case Decoder::Opcode::S_OR_B64:
|
||||
case Decoder::Opcode::S_ORN2_B64:
|
||||
case Decoder::Opcode::S_XOR_B64:
|
||||
case Decoder::Opcode::S_NAND_B64:
|
||||
case Decoder::Opcode::S_NOR_B64:
|
||||
case Decoder::Opcode::S_XNOR_B64:
|
||||
case Decoder::Opcode::S_LSHL_B64:
|
||||
case Decoder::Opcode::S_LSHR_B64:
|
||||
case Decoder::Opcode::S_BFE_U64:
|
||||
case Decoder::Opcode::S_WQM_B64: return true;
|
||||
default: return false;
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user