//===-- VOP3Instructions.td - Vector Instruction Definitions --------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
// Special case for v_div_fmas_{f32|f64}, since it seems to be the
// only VOP instruction that implicitly reads VCC.
let Asm64 = " $vdst, $src0_modifiers, $src1_modifiers, $src2_modifiers$clamp$omod" in {
def VOP_F32_F32_F32_F32_VCC : VOPProfile<[f32, f32, f32, f32]> {
let Outs64 = (outs DstRC.RegClass:$vdst);
let HasExtVOP3DPP = 0;
let HasExtDPP = 0;
let IsSingle = 1;
}
def VOP_F64_F64_F64_F64_VCC : VOPProfile<[f64, f64, f64, f64]> {
let Outs64 = (outs DstRC.RegClass:$vdst);
let IsSingle = 1;
}
}
class VOP3b_Profile<ValueType vt> : VOPProfile<[vt, vt, vt, vt]> {
let Outs64 = (outs DstRC:$vdst, VOPDstS64orS32:$sdst);
let Asm64 = "$vdst, $sdst, $src0_modifiers, $src1_modifiers, $src2_modifiers$clamp$omod";
let IsSingle = 1;
let HasExtVOP3DPP = 0;
let HasExtDPP = 0;
}
def VOP3b_F32_I1_F32_F32_F32 : VOP3b_Profile<f32>;
def VOP3b_F64_I1_F64_F64_F64 : VOP3b_Profile<f64>;
def VOP3b_I64_I1_I32_I32_I64 : VOPProfile<[i64, i32, i32, i64]> {
let HasClamp = 1;
let IsSingle = 1;
let Outs64 = (outs DstRC:$vdst, VOPDstS64orS32:$sdst);
let Asm64 = "$vdst, $sdst, $src0, $src1, $src2$clamp";
}
class V_MUL_PROF<VOPProfile P> : VOP3_Profile<P> {
let HasExtVOP3DPP = 0;
let HasExtDPP = 0;
}
def DIV_FIXUP_F32_PROF : VOP3_Profile<VOP_F32_F32_F32_F32> {
let HasExtVOP3DPP = 0;
let HasExtDPP = 0;
}
//===----------------------------------------------------------------------===//
// VOP3 INTERP
//===----------------------------------------------------------------------===//
class VOP3Interp<string OpName, VOPProfile P, list<dag> pattern = []> :
VOP3_Pseudo<OpName, P, pattern> {
let AsmMatchConverter = "cvtVOP3Interp";
let mayRaiseFPException = 0;
}
def VOP3_INTERP : VOPProfile<[f32, f32, i32, untyped]> {
let Src0Mod = FPVRegInputMods;
let Ins64 = (ins Src0Mod:$src0_modifiers, VRegSrc_32:$src0,
InterpAttr:$attr, InterpAttrChan:$attrchan,
Clamp0:$clamp, omod0:$omod);
let Asm64 = "$vdst, $src0_modifiers, $attr$attrchan$clamp$omod";
}
def VOP3_INTERP_MOV : VOPProfile<[f32, i32, i32, untyped]> {
let Ins64 = (ins InterpSlot:$src0,
InterpAttr:$attr, InterpAttrChan:$attrchan,
Clamp0:$clamp, omod0:$omod);
let Asm64 = "$vdst, $src0, $attr$attrchan$clamp$omod";
let HasClamp = 1;
let HasSrc0Mods = 0;
}
class getInterp16Asm <bit HasSrc2, bit HasOMod> {
string src2 = !if(HasSrc2, ", $src2_modifiers", "");
string omod = !if(HasOMod, "$omod", "");
string ret =
" $vdst, $src0_modifiers, $attr$attrchan"#src2#"$high$clamp"#omod;
}
class getInterp16Ins <bit HasSrc2, bit HasOMod,
Operand Src0Mod, Operand Src2Mod> {
dag ret = !if(HasSrc2,
!if(HasOMod,
(ins Src0Mod:$src0_modifiers, VRegSrc_32:$src0,
InterpAttr:$attr, InterpAttrChan:$attrchan,
Src2Mod:$src2_modifiers, VRegSrc_32:$src2,
highmod:$high, Clamp0:$clamp, omod0:$omod),
(ins Src0Mod:$src0_modifiers, VRegSrc_32:$src0,
InterpAttr:$attr, InterpAttrChan:$attrchan,
Src2Mod:$src2_modifiers, VRegSrc_32:$src2,
highmod:$high, Clamp0:$clamp)
),
(ins Src0Mod:$src0_modifiers, VRegSrc_32:$src0,
InterpAttr:$attr, InterpAttrChan:$attrchan,
highmod:$high, Clamp0:$clamp, omod0:$omod)
);
}
class VOP3_INTERP16 <list<ValueType> ArgVT> : VOPProfile<ArgVT> {
let IsSingle = 1;
let HasOMod = !ne(DstVT.Value, f16.Value);
let HasHigh = 1;
let Src0Mod = FPVRegInputMods;
let Src2Mod = FPVRegInputMods;
let Outs64 = (outs DstRC.RegClass:$vdst);
let Ins64 = getInterp16Ins<HasSrc2, HasOMod, Src0Mod, Src2Mod>.ret;
let Asm64 = getInterp16Asm<HasSrc2, HasOMod>.ret;
}
//===----------------------------------------------------------------------===//
// VOP3 Instructions
//===----------------------------------------------------------------------===//
let isCommutable = 1 in {
let isReMaterializable = 1 in {
let mayRaiseFPException = 0 in {
let SubtargetPredicate = HasMadMacF32Insts in {
defm V_MAD_LEGACY_F32 : VOP3Inst <"v_mad_legacy_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
defm V_MAD_F32 : VOP3Inst <"v_mad_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, any_fmad>;
} // End SubtargetPredicate = HasMadMacInsts
let SubtargetPredicate = HasFmaLegacy32 in
defm V_FMA_LEGACY_F32 : VOP3Inst <"v_fma_legacy_f32",
VOP3_Profile<VOP_F32_F32_F32_F32>,
int_amdgcn_fma_legacy>;
}
defm V_MAD_I32_I24 : VOP3Inst <"v_mad_i32_i24", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
defm V_MAD_U32_U24 : VOP3Inst <"v_mad_u32_u24", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
defm V_FMA_F32 : VOP3Inst <"v_fma_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, any_fma>;
defm V_LERP_U8 : VOP3Inst <"v_lerp_u8", VOP3_Profile<VOP_I32_I32_I32_I32>, int_amdgcn_lerp>;
let SchedRW = [WriteDoubleAdd] in {
let FPDPRounding = 1 in {
defm V_FMA_F64 : VOP3Inst <"v_fma_f64", VOP3_Profile<VOP_F64_F64_F64_F64>, any_fma>;
let SubtargetPredicate = isNotGFX12Plus in {
defm V_ADD_F64 : VOP3Inst <"v_add_f64", VOP3_Profile<VOP_F64_F64_F64>, any_fadd>;
defm V_MUL_F64 : VOP3Inst <"v_mul_f64", VOP3_Profile<VOP_F64_F64_F64>, any_fmul>;
} // End SubtargetPredicate = isNotGFX12Plus
} // End FPDPRounding = 1
let SubtargetPredicate = isNotGFX12Plus in {
defm V_MIN_F64 : VOP3Inst <"v_min_f64", VOP3_Profile<VOP_F64_F64_F64>, fminnum_like>;
defm V_MAX_F64 : VOP3Inst <"v_max_f64", VOP3_Profile<VOP_F64_F64_F64>, fmaxnum_like>;
} // End SubtargetPredicate = isNotGFX12Plus
} // End SchedRW = [WriteDoubleAdd]
let SchedRW = [WriteIntMul] in {
defm V_MUL_LO_U32 : VOP3Inst <"v_mul_lo_u32", V_MUL_PROF<VOP_I32_I32_I32>, DivergentBinFrag<mul>>;
defm V_MUL_HI_U32 : VOP3Inst <"v_mul_hi_u32", V_MUL_PROF<VOP_I32_I32_I32>, mulhu>;
defm V_MUL_LO_I32 : VOP3Inst <"v_mul_lo_i32", V_MUL_PROF<VOP_I32_I32_I32>>;
defm V_MUL_HI_I32 : VOP3Inst <"v_mul_hi_i32", V_MUL_PROF<VOP_I32_I32_I32>, mulhs>;
} // End SchedRW = [WriteIntMul]
let SubtargetPredicate = isGFX12Plus, ReadsModeReg = 0 in {
defm V_MINIMUM_F32 : VOP3Inst <"v_minimum_f32", VOP3_Profile<VOP_F32_F32_F32>, DivergentBinFrag<fminimum>>;
defm V_MAXIMUM_F32 : VOP3Inst <"v_maximum_f32", VOP3_Profile<VOP_F32_F32_F32>, DivergentBinFrag<fmaximum>>;
defm V_MINIMUM_F16 : VOP3Inst <"v_minimum_f16", VOP3_Profile<VOP_F16_F16_F16>, DivergentBinFrag<fminimum>>;
defm V_MAXIMUM_F16 : VOP3Inst <"v_maximum_f16", VOP3_Profile<VOP_F16_F16_F16>, DivergentBinFrag<fmaximum>>;
let SchedRW = [WriteDoubleAdd] in {
defm V_MINIMUM_F64 : VOP3Inst <"v_minimum_f64", VOP3_Profile<VOP_F64_F64_F64>, fminimum>;
defm V_MAXIMUM_F64 : VOP3Inst <"v_maximum_f64", VOP3_Profile<VOP_F64_F64_F64>, fmaximum>;
} // End SchedRW = [WriteDoubleAdd]
} // End SubtargetPredicate = isGFX12Plus, ReadsModeReg = 0
} // End isReMaterializable = 1
let Uses = [MODE, VCC, EXEC] in {
// v_div_fmas_f32:
// result = src0 * src1 + src2
// if (vcc)
// result *= 2^32
//
let SchedRW = [WriteFloatFMA] in
defm V_DIV_FMAS_F32 : VOP3Inst_Pseudo_Wrapper <"v_div_fmas_f32", VOP_F32_F32_F32_F32_VCC, []>;
// v_div_fmas_f64:
// result = src0 * src1 + src2
// if (vcc)
// result *= 2^64
//
let SchedRW = [WriteDouble], FPDPRounding = 1 in
defm V_DIV_FMAS_F64 : VOP3Inst_Pseudo_Wrapper <"v_div_fmas_f64", VOP_F64_F64_F64_F64_VCC, []>;
} // End Uses = [MODE, VCC, EXEC]
} // End isCommutable = 1
let isReMaterializable = 1 in {
let mayRaiseFPException = 0 in {
defm V_CUBEID_F32 : VOP3Inst <"v_cubeid_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, int_amdgcn_cubeid>;
defm V_CUBESC_F32 : VOP3Inst <"v_cubesc_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, int_amdgcn_cubesc>;
defm V_CUBETC_F32 : VOP3Inst <"v_cubetc_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, int_amdgcn_cubetc>;
defm V_CUBEMA_F32 : VOP3Inst <"v_cubema_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, int_amdgcn_cubema>;
} // End mayRaiseFPException
defm V_BFE_U32 : VOP3Inst <"v_bfe_u32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUbfe_u32>;
defm V_BFE_I32 : VOP3Inst <"v_bfe_i32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUbfe_i32>;
defm V_BFI_B32 : VOP3Inst <"v_bfi_b32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUbfi>;
defm V_ALIGNBIT_B32 : VOP3Inst <"v_alignbit_b32", VOP3_Profile<VOP_I32_I32_I32_I32>, fshr>;
defm V_ALIGNBYTE_B32 : VOP3Inst <"v_alignbyte_b32", VOP3_Profile<VOP_I32_I32_I32_I32>, int_amdgcn_alignbyte>;
// XXX - No FPException seems suspect but manual doesn't say it does
let mayRaiseFPException = 0 in {
let isCommutable = 1 in {
defm V_MIN3_I32 : VOP3Inst <"v_min3_i32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUsmin3>;
defm V_MIN3_U32 : VOP3Inst <"v_min3_u32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUumin3>;
defm V_MAX3_I32 : VOP3Inst <"v_max3_i32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUsmax3>;
defm V_MAX3_U32 : VOP3Inst <"v_max3_u32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUumax3>;
defm V_MED3_I32 : VOP3Inst <"v_med3_i32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUsmed3>;
defm V_MED3_U32 : VOP3Inst <"v_med3_u32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUumed3>;
} // End isCommutable = 1
defm V_MIN3_F32 : VOP3Inst <"v_min3_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, AMDGPUfmin3>;
defm V_MAX3_F32 : VOP3Inst <"v_max3_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, AMDGPUfmax3>;
defm V_MED3_F32 : VOP3Inst <"v_med3_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, AMDGPUfmed3>;
} // End mayRaiseFPException = 0
let SubtargetPredicate = isGFX12Plus, ReadsModeReg = 0 in {
defm V_MINIMUM3_F32 : VOP3Inst <"v_minimum3_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, AMDGPUfminimum3>;
defm V_MAXIMUM3_F32 : VOP3Inst <"v_maximum3_f32", VOP3_Profile<VOP_F32_F32_F32_F32>, AMDGPUfmaximum3>;
} // End SubtargetPredicate = isGFX12Plus, ReadsModeReg = 0
let isCommutable = 1 in {
defm V_SAD_U8 : VOP3Inst <"v_sad_u8", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
defm V_SAD_HI_U8 : VOP3Inst <"v_sad_hi_u8", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
defm V_SAD_U16 : VOP3Inst <"v_sad_u16", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
defm V_SAD_U32 : VOP3Inst <"v_sad_u32", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
} // End isCommutable = 1
defm V_CVT_PK_U8_F32 : VOP3Inst<"v_cvt_pk_u8_f32", VOP3_Profile<VOP_I32_F32_I32_I32>, int_amdgcn_cvt_pk_u8_f32>;
defm V_DIV_FIXUP_F32 : VOP3Inst <"v_div_fixup_f32", DIV_FIXUP_F32_PROF, AMDGPUdiv_fixup>;
let SchedRW = [WriteDoubleAdd], FPDPRounding = 1 in {
defm V_DIV_FIXUP_F64 : VOP3Inst <"v_div_fixup_f64", VOP3_Profile<VOP_F64_F64_F64_F64>, AMDGPUdiv_fixup>;
defm V_LDEXP_F64 : VOP3Inst <"v_ldexp_f64", VOP3_Profile<VOP_F64_F64_I32>, any_fldexp>;
} // End SchedRW = [WriteDoubleAdd], FPDPRounding = 1
} // End isReMaterializable = 1
let mayRaiseFPException = 0 in { // Seems suspicious but manual doesn't say it does.
let SchedRW = [WriteFloatFMA, WriteSALU] in
defm V_DIV_SCALE_F32 : VOP3Inst_Pseudo_Wrapper <"v_div_scale_f32", VOP3b_F32_I1_F32_F32_F32> ;
// Double precision division pre-scale.
let SchedRW = [WriteDouble, WriteSALU], FPDPRounding = 1 in
defm V_DIV_SCALE_F64 : VOP3Inst_Pseudo_Wrapper <"v_div_scale_f64", VOP3b_F64_I1_F64_F64_F64>;
} // End mayRaiseFPException = 0
let isReMaterializable = 1 in
defm V_MSAD_U8 : VOP3Inst <"v_msad_u8", VOP3_Profile<VOP_I32_I32_I32_I32, VOP3_CLAMP>>;
let Constraints = "@earlyclobber $vdst" in {
defm V_MQSAD_PK_U16_U8 : VOP3Inst <"v_mqsad_pk_u16_u8", VOP3_Profile<VOP_I64_I64_I32_I64, VOP3_CLAMP>>;
} // End Constraints = "@earlyclobber $vdst"
let isReMaterializable = 1 in {
let SchedRW = [WriteDouble] in {
defm V_TRIG_PREOP_F64 : VOP3Inst <"v_trig_preop_f64", VOP3_Profile<VOP_F64_F64_I32>, int_amdgcn_trig_preop>;
} // End SchedRW = [WriteDouble]
let SchedRW = [Write64Bit] in {
let SubtargetPredicate = isGFX6GFX7 in {
defm V_LSHL_B64 : VOP3Inst <"v_lshl_b64", VOP3_Profile<VOP_I64_I64_I32>, cshl_64>;
defm V_LSHR_B64 : VOP3Inst <"v_lshr_b64", VOP3_Profile<VOP_I64_I64_I32>, csrl_64>;
defm V_ASHR_I64 : VOP3Inst <"v_ashr_i64", VOP3_Profile<VOP_I64_I64_I32>, csra_64>;
} // End SubtargetPredicate = isGFX6GFX7
let SubtargetPredicate = isGFX8Plus in {
defm V_LSHRREV_B64 : VOP3Inst <"v_lshrrev_b64", VOP3_Profile<VOP_I64_I32_I64>, clshr_rev_64>;
defm V_ASHRREV_I64 : VOP3Inst <"v_ashrrev_i64", VOP3_Profile<VOP_I64_I32_I64>, cashr_rev_64>;
} // End SubtargetPredicate = isGFX8Plus
let SubtargetPredicate = isGFX8GFX9GFX10GFX11 in {
defm V_LSHLREV_B64 : VOP3Inst <"v_lshlrev_b64", VOP3_Profile<VOP_I64_I32_I64>, clshl_rev_64>;
} // End SubtargetPredicate = isGFX8GFX9GFX10GFX11
} // End SchedRW = [Write64Bit]
} // End isReMaterializable = 1
def : GCNPat<
(i32 (DivergentUnaryFrag<sext> i16:$src)),
(i32 (V_BFE_I32_e64 i16:$src, (i32 0), (i32 0x10)))
>;
let isReMaterializable = 1 in {
let SubtargetPredicate = isGFX6GFX7GFX10Plus in {
defm V_MULLIT_F32 : VOP3Inst <"v_mullit_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
} // End SubtargetPredicate = isGFX6GFX7GFX10Plus
let SchedRW = [Write32Bit] in {
let SubtargetPredicate = isGFX8Plus in {
defm V_PERM_B32 : VOP3Inst <"v_perm_b32", VOP3_Profile<VOP_I32_I32_I32_I32>, AMDGPUperm>;
} // End SubtargetPredicate = isGFX8Plus
} // End SchedRW = [Write32Bit]
} // End isReMaterializable = 1
def VOPProfileMQSAD : VOP3_Profile<VOP_V4I32_I64_I32_V4I32, VOP3_CLAMP> {
let HasModifiers = 0;
}
let SubtargetPredicate = isGFX7Plus in {
let Constraints = "@earlyclobber $vdst", SchedRW = [WriteQuarterRate32] in {
defm V_QSAD_PK_U16_U8 : VOP3Inst <"v_qsad_pk_u16_u8", VOP3_Profile<VOP_I64_I64_I32_I64, VOP3_CLAMP>>;
defm V_MQSAD_U32_U8 : VOP3Inst <"v_mqsad_u32_u8", VOPProfileMQSAD>;
} // End Constraints = "@earlyclobber $vdst", SchedRW = [WriteQuarterRate32]
} // End SubtargetPredicate = isGFX7Plus
let isCommutable = 1, SchedRW = [WriteIntMul, WriteSALU] in {
let SubtargetPredicate = isGFX7Plus, OtherPredicates = [HasNotMADIntraFwdBug] in {
defm V_MAD_U64_U32 : VOP3Inst <"v_mad_u64_u32", VOP3b_I64_I1_I32_I32_I64>;
defm V_MAD_I64_I32 : VOP3Inst <"v_mad_i64_i32", VOP3b_I64_I1_I32_I32_I64>;
}
let SubtargetPredicate = isGFX11Only, OtherPredicates = [HasMADIntraFwdBug],
Constraints = "@earlyclobber $vdst" in {
defm V_MAD_U64_U32_gfx11 : VOP3Inst <"v_mad_u64_u32", VOP3b_I64_I1_I32_I32_I64>;
defm V_MAD_I64_I32_gfx11 : VOP3Inst <"v_mad_i64_i32", VOP3b_I64_I1_I32_I32_I64>;
}
} // End isCommutable = 1, SchedRW = [WriteIntMul, WriteSALU]
let FPDPRounding = 1 in {
let Predicates = [Has16BitInsts, isGFX8Only] in {
defm V_DIV_FIXUP_F16 : VOP3Inst <"v_div_fixup_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, AMDGPUdiv_fixup>;
defm V_FMA_F16 : VOP3Inst <"v_fma_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, any_fma>;
} // End Predicates = [Has16BitInsts, isGFX8Only]
let SubtargetPredicate = isGFX9Plus in {
defm V_DIV_FIXUP_F16_gfx9 : VOP3Inst <"v_div_fixup_f16_gfx9",
VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, AMDGPUdiv_fixup>;
defm V_FMA_F16_gfx9 : VOP3Inst <"v_fma_f16_gfx9", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, any_fma>;
} // End SubtargetPredicate = isGFX9Plus
} // End FPDPRounding = 1
let SubtargetPredicate = Has16BitInsts, isCommutable = 1 in {
defm V_MAD_U16 : VOP3Inst <"v_mad_u16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_CLAMP>>;
defm V_MAD_I16 : VOP3Inst <"v_mad_i16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_CLAMP>>;
let FPDPRounding = 1 in {
defm V_MAD_F16 : VOP3Inst <"v_mad_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, any_fmad>;
let Uses = [MODE, M0, EXEC] in {
let OtherPredicates = [isNotGFX90APlus] in
// For some reason the intrinsic operands are in a different order
// from the instruction operands.
def V_INTERP_P2_F16 : VOP3Interp <"v_interp_p2_f16", VOP3_INTERP16<[f16, f32, i32, f32]>,
[(set f16:$vdst,
(int_amdgcn_interp_p2_f16 (VOP3Mods f32:$src2, i32:$src2_modifiers),
(VOP3Mods f32:$src0, i32:$src0_modifiers),
(i32 timm:$attrchan),
(i32 timm:$attr),
(i1 timm:$high),
M0))]>;
} // End Uses = [M0, MODE, EXEC]
} // End FPDPRounding = 1
let SubtargetPredicate = isGFX9Only, FPDPRounding = 1 in {
defm V_MAD_F16_gfx9 : VOP3Inst <"v_mad_f16_gfx9", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>> ;
} // End SubtargetPredicate = isGFX9Only, FPDPRounding = 1
let SubtargetPredicate = isGFX9Plus in {
defm V_MAD_U16_gfx9 : VOP3Inst <"v_mad_u16_gfx9", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>>;
defm V_MAD_I16_gfx9 : VOP3Inst <"v_mad_i16_gfx9", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>>;
let OtherPredicates = [isNotGFX90APlus] in
def V_INTERP_P2_F16_gfx9 : VOP3Interp <"v_interp_p2_f16_gfx9", VOP3_INTERP16<[f16, f32, i32, f32]>>;
} // End SubtargetPredicate = isGFX9Plus
// This predicate should only apply to the selection pattern. The
// instruction still exists and should decode on subtargets with
// other bank counts.
let OtherPredicates = [isNotGFX90APlus, has32BankLDS], Uses = [MODE, M0, EXEC], FPDPRounding = 1 in {
def V_INTERP_P1LL_F16 : VOP3Interp <"v_interp_p1ll_f16", VOP3_INTERP16<[f32, f32, i32, untyped]>,
[(set f32:$vdst, (int_amdgcn_interp_p1_f16 (VOP3Mods f32:$src0, i32:$src0_modifiers),
(i32 timm:$attrchan),
(i32 timm:$attr),
(i1 timm:$high), M0))]>;
} // End OtherPredicates = [isNotGFX90APlus, has32BankLDS], Uses = [MODE, M0, EXEC], FPDPRounding = 1
let OtherPredicates = [isNotGFX90APlus], Uses = [MODE, M0, EXEC], FPDPRounding = 1 in {
def V_INTERP_P1LV_F16 : VOP3Interp <"v_interp_p1lv_f16", VOP3_INTERP16<[f32, f32, i32, f16]>>;
} // End OtherPredicates = [isNotGFX90APlus], Uses = [MODE, M0, EXEC], FPDPRounding = 1
} // End SubtargetPredicate = Has16BitInsts, isCommutable = 1
def : GCNPat<
(i64 (DivergentUnaryFrag<sext> i16:$src)),
(REG_SEQUENCE VReg_64,
(i32 (V_BFE_I32_e64 $src, (S_MOV_B32 (i32 0)), (S_MOV_B32 (i32 0x10)))), sub0,
(i32 (COPY_TO_REGCLASS
(V_ASHRREV_I32_e32 (S_MOV_B32 (i32 0x1f)), (i32 (V_BFE_I32_e64 $src, (S_MOV_B32 (i32 0)), (S_MOV_B32 (i32 0x10))))
), VGPR_32)), sub1)
>;
let SubtargetPredicate = isGFX8Plus, Uses = [MODE, M0, EXEC], OtherPredicates = [isNotGFX90APlus] in {
def V_INTERP_P1_F32_e64 : VOP3Interp <"v_interp_p1_f32", VOP3_INTERP>;
def V_INTERP_P2_F32_e64 : VOP3Interp <"v_interp_p2_f32", VOP3_INTERP>;
def V_INTERP_MOV_F32_e64 : VOP3Interp <"v_interp_mov_f32", VOP3_INTERP_MOV>;
} // End SubtargetPredicate = isGFX8Plus, Uses = [MODE, M0, EXEC], OtherPredicates = [isNotGFX90APlus]
// Note: 16-bit instructions produce a 0 result in the high 16-bits
// on GFX8 and GFX9 and preserve high 16 bits on GFX10+
multiclass Arithmetic_i16_0Hi_TernaryPats <SDPatternOperator op, Instruction inst> {
def : GCNPat<
(i32 (zext (op i16:$src0, i16:$src1, i16:$src2))),
(inst VSrc_b16:$src0, VSrc_b16:$src1, VSrc_b16:$src2)
>;
}
let Predicates = [Has16BitInsts, isGFX8GFX9] in {
defm : Arithmetic_i16_0Hi_TernaryPats<imad, V_MAD_U16_e64>;
}
let Predicates = [Has16BitInsts, isGFX6GFX7GFX8GFX9] in {
// FIXME: Should be able to just pass imad to the instruction
// definition pattern, but the implied clamp input interferes.
multiclass Ternary_i16_Pats <SDPatternOperator op, Instruction inst> {
def : GCNPat <
(op i16:$src0, i16:$src1, i16:$src2),
(inst i16:$src0, i16:$src1, i16:$src2, (i1 0))
>;
}
defm: Ternary_i16_Pats<imad, V_MAD_U16_e64>;
} // End Predicates = [Has16BitInsts, isGFX6GFX7GFX8GFX9]
class Ternary_i16_Pats_gfx9<SDPatternOperator op1, SDPatternOperator op2,
Instruction inst> : GCNPat <
(op2 (op1 i16:$src0, i16:$src1), i16:$src2),
(inst SRCMODS.NONE, $src0, SRCMODS.NONE, $src1, SRCMODS.NONE, $src2, DSTCLAMP.NONE)
>;
let Predicates = [Has16BitInsts, isGFX10Plus] in {
def: Ternary_i16_Pats_gfx9<mul, add, V_MAD_U16_gfx9_e64>;
} // End Predicates = [Has16BitInsts, isGFX10Plus]
class ThreeOpFragSDAG<SDPatternOperator op1, SDPatternOperator op2> : PatFrag<
(ops node:$x, node:$y, node:$z),
// When the inner operation is used multiple times, selecting 3-op
// instructions may still be beneficial -- if the other users can be
// combined similarly. Let's be conservative for now.
(op2 (HasOneUseBinOp<op1> node:$x, node:$y), node:$z),
[{
// Only use VALU ops when the result is divergent.
if (!N->isDivergent())
return false;
// Check constant bus limitations.
//
// Note: Use !isDivergent as a conservative proxy for whether the value
// is in an SGPR (uniform values can end up in VGPRs as well).
unsigned ConstantBusUses = 0;
for (unsigned i = 0; i < 3; ++i) {
if (!Operands[i]->isDivergent() &&
!isInlineImmediate(Operands[i].getNode())) {
ConstantBusUses++;
// This uses AMDGPU::V_ADD3_U32_e64, but all three operand instructions
// have the same constant bus limit.
if (ConstantBusUses > Subtarget->getConstantBusLimit(AMDGPU::V_ADD3_U32_e64))
return false;
}
}
return true;
}]> {
let PredicateCodeUsesOperands = 1;
}
class ThreeOpFrag<SDPatternOperator op1, SDPatternOperator op2> : ThreeOpFragSDAG<op1, op2> {
// The divergence predicate is irrelevant in GlobalISel, as we have
// proper register bank checks. We just need to verify the constant
// bus restriction when all the sources are considered.
//
// FIXME: With unlucky SGPR operands, we could penalize code by
// blocking folding SGPR->VGPR copies later.
// FIXME: There's no register bank verifier
let GISelPredicateCode = [{
const int ConstantBusLimit = Subtarget->getConstantBusLimit(AMDGPU::V_ADD3_U32_e64);
int ConstantBusUses = 0;
for (unsigned i = 0; i < 3; ++i) {
const RegisterBank *RegBank = RBI.getRegBank(Operands[i]->getReg(), MRI, TRI);
if (RegBank->getID() == AMDGPU::SGPRRegBankID) {
if (++ConstantBusUses > ConstantBusLimit)
return false;
}
}
return true;
}];
}
def shl_0_to_4 : PatFrag<
(ops node:$src0, node:$src1), (shl node:$src0, node:$src1),
[{
if (auto *C = dyn_cast<ConstantSDNode>(N->getOperand(1))) {
return C->getZExtValue() <= 4;
}
return false;
}]> {
let GISelPredicateCode = [{
int64_t Imm = 0;
if (!mi_match(MI.getOperand(2).getReg(), MRI, m_ICst(Imm)) &&
!mi_match(MI.getOperand(2).getReg(), MRI, m_Copy(m_ICst(Imm))))
return false;
return (uint64_t)Imm <= 4;
}];
}
def VOP3_CVT_PK_F8_F32_Profile : VOP3_Profile<VOP_I32_F32_F32, VOP3_OPSEL> {
let InsVOP3OpSel = (ins FP32InputMods:$src0_modifiers, Src0RC64:$src0,
FP32InputMods:$src1_modifiers, Src1RC64:$src1,
VGPR_32:$vdst_in, op_sel0:$op_sel);
let InsVOP3DPP = (ins VGPR_32:$old,
FP32InputMods:$src0_modifiers, Src0VOP3DPP:$src0,
FP32InputMods:$src1_modifiers, Src1VOP3DPP:$src1,
VGPR_32:$vdst_in, op_sel0:$op_sel,
dpp_ctrl:$dpp_ctrl, DppRowMask:$row_mask,
DppBankMask:$bank_mask, DppBoundCtrl:$bound_ctrl);
let InsVOP3DPP16 = (ins VGPR_32:$old,
FP32InputMods:$src0_modifiers, Src0VOP3DPP:$src0,
FP32InputMods:$src1_modifiers, Src1VOP3DPP:$src1,
VGPR_32:$vdst_in, op_sel0:$op_sel,
dpp_ctrl:$dpp_ctrl, DppRowMask:$row_mask,
DppBankMask:$bank_mask, DppBoundCtrl:$bound_ctrl, Dpp16FI:$fi);
let InsVOP3DPP8 = (ins VGPR_32:$old,
FP32InputMods:$src0_modifiers, Src0VOP3DPP:$src0,
FP32InputMods:$src1_modifiers, Src1VOP3DPP:$src1,
VGPR_32:$vdst_in, op_sel0:$op_sel, dpp8:$dpp8, Dpp8FI:$fi);
let HasClamp = 0;
let HasExtVOP3DPP = 1;
}
def VOP3_CVT_SR_F8_F32_Profile : VOP3_Profile<VOPProfile<[i32, f32, i32, f32]>,
VOP3_OPSEL> {
let InsVOP3OpSel = (ins FP32InputMods:$src0_modifiers, Src0RC64:$src0,
FP32InputMods:$src1_modifiers, Src1RC64:$src1,
FP32InputMods:$src2_modifiers, VGPR_32:$src2,
op_sel0:$op_sel);
let InsVOP3DPP16 = (ins VGPR_32:$old,
FP32InputMods:$src0_modifiers, Src0VOP3DPP:$src0,
FP32InputMods:$src1_modifiers, Src1VOP3DPP:$src1,
FP32InputMods:$src2_modifiers, VGPR_32:$src2,
op_sel0:$op_sel, dpp_ctrl:$dpp_ctrl, DppRowMask:$row_mask,
DppBankMask:$bank_mask, DppBoundCtrl:$bound_ctrl, Dpp16FI:$fi);
let InsVOP3DPP8 = (ins VGPR_32:$old,
FP32InputMods:$src0_modifiers, Src0VOP3DPP:$src0,
FP32InputMods:$src1_modifiers, Src1VOP3DPP:$src1,
FP32InputMods:$src2_modifiers, VGPR_32:$src2,
op_sel0:$op_sel, dpp8:$dpp8, Dpp8FI:$fi);
let HasClamp = 0;
let HasSrc2 = 0;
let HasSrc2Mods = 1;
let HasExtVOP3DPP = 1;
let HasOpSel = 1;
let HasFP8DstByteSel = 1;
let AsmVOP3OpSel = !subst(", $src2_modifiers", "",
getAsmVOP3OpSel<3, HasClamp, HasOMod,
HasSrc0FloatMods, HasSrc1FloatMods,
HasSrc2FloatMods>.ret);
let AsmVOP3DPP16 = !subst(", $src2_modifiers", "",
getAsmVOP3DPP16<getAsmVOP3Base<3, 1, HasClamp, 1,
HasOMod, 0, 1, HasSrc0FloatMods,
HasSrc1FloatMods,
HasSrc2FloatMods>.ret>.ret);
let AsmVOP3DPP8 = !subst(", $src2_modifiers", "",
getAsmVOP3DPP8<getAsmVOP3Base<3, 1, HasClamp, 1,
HasOMod, 0, 1, HasSrc0FloatMods,
HasSrc1FloatMods,
HasSrc2FloatMods>.ret>.ret);
}
class VOP3_CVT_SR_F8_ByteSel_Profile<ValueType SrcVT> :
VOP3_Profile<VOPProfile<[i32, SrcVT, i32, untyped]>> {
let IsFP8DstByteSel = 1;
let HasFP8DstByteSel = 1;
let HasClamp = 0;
defvar bytesel = (ins VGPR_32:$vdst_in, ByteSel:$byte_sel);
let Ins64 = !con(getIns64<Src0RC64, Src1RC64, Src2RC64, NumSrcArgs,
HasClamp, HasModifiers, HasSrc2Mods,
HasOMod, Src0Mod, Src1Mod, Src2Mod>.ret,
bytesel);
let InsVOP3Base = !con(
getInsVOP3Base<Src0VOP3DPP, Src1VOP3DPP,
Src2VOP3DPP, NumSrcArgs, HasClamp, HasModifiers, HasSrc2Mods, HasOMod,
Src0ModVOP3DPP, Src1ModVOP3DPP, Src2ModVOP3DPP, HasOpSel>.ret,
bytesel);
}
def IsPow2Plus1: PatLeaf<(i32 imm), [{
uint32_t V = N->getZExtValue();
return isPowerOf2_32(V - 1);
}]>;
def Log2_32: SDNodeXForm<imm, [{
uint32_t V = N->getZExtValue();
return CurDAG->getTargetConstant(Log2_32(V - 1), SDLoc(N), MVT::i32);
}]>;
let SubtargetPredicate = isGFX9Plus in {
let isCommutable = 1, isReMaterializable = 1 in {
defm V_ADD3_U32 : VOP3Inst <"v_add3_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_AND_OR_B32 : VOP3Inst <"v_and_or_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_OR3_B32 : VOP3Inst <"v_or3_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_XAD_U32 : VOP3Inst <"v_xad_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_ADD_I32 : VOP3Inst <"v_add_i32", VOP3_Profile<VOP_I32_I32_I32_ARITH>>;
defm V_ADD_LSHL_U32 : VOP3Inst <"v_add_lshl_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
} // End isCommutable = 1, isReMaterializable = 1
// TODO src0 contains the opsel bit for dst, so if we commute, need to mask and swap this
// to the new src0.
defm V_MED3_F16 : VOP3Inst <"v_med3_f16", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, AMDGPUfmed3>;
defm V_MED3_I16 : VOP3Inst <"v_med3_i16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>, AMDGPUsmed3>;
defm V_MED3_U16 : VOP3Inst <"v_med3_u16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>, AMDGPUumed3>;
defm V_MIN3_F16 : VOP3Inst <"v_min3_f16", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, AMDGPUfmin3>;
defm V_MIN3_I16 : VOP3Inst <"v_min3_i16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>, AMDGPUsmin3>;
defm V_MIN3_U16 : VOP3Inst <"v_min3_u16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>, AMDGPUumin3>;
defm V_MAX3_F16 : VOP3Inst <"v_max3_f16", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, AMDGPUfmax3>;
defm V_MAX3_I16 : VOP3Inst <"v_max3_i16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>, AMDGPUsmax3>;
defm V_MAX3_U16 : VOP3Inst <"v_max3_u16", VOP3_Profile<VOP_I16_I16_I16_I16, VOP3_OPSEL>, AMDGPUumax3>;
let SubtargetPredicate = isGFX12Plus, ReadsModeReg = 0 in {
defm V_MINIMUM3_F16 : VOP3Inst <"v_minimum3_f16", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, AMDGPUfminimum3>;
defm V_MAXIMUM3_F16 : VOP3Inst <"v_maximum3_f16", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>, AMDGPUfmaximum3>;
} // End SubtargetPredicate = isGFX12Plus, ReadsModeReg = 0
defm V_ADD_I16 : VOP3Inst <"v_add_i16", VOP3_Profile<VOP_I16_I16_I16, VOP3_OPSEL>>;
defm V_SUB_I16 : VOP3Inst <"v_sub_i16", VOP3_Profile<VOP_I16_I16_I16, VOP3_OPSEL>>;
defm V_MAD_U32_U16 : VOP3Inst <"v_mad_u32_u16", VOP3_Profile<VOP_I32_I16_I16_I32, VOP3_OPSEL>>;
defm V_MAD_I32_I16 : VOP3Inst <"v_mad_i32_i16", VOP3_Profile<VOP_I32_I16_I16_I32, VOP3_OPSEL>>;
defm V_CVT_PKNORM_I16_F16 : VOP3Inst <"v_cvt_pknorm_i16_f16", VOP3_Profile<VOP_B32_F16_F16, VOP3_OPSEL>>;
defm V_CVT_PKNORM_U16_F16 : VOP3Inst <"v_cvt_pknorm_u16_f16", VOP3_Profile<VOP_B32_F16_F16, VOP3_OPSEL>>;
defm V_PACK_B32_F16 : VOP3Inst <"v_pack_b32_f16", VOP3_Profile<VOP_B32_F16_F16, VOP3_OPSEL>>;
let isReMaterializable = 1 in {
defm V_SUB_I32 : VOP3Inst <"v_sub_i32", VOP3_Profile<VOP_I32_I32_I32_ARITH>>;
defm V_LSHL_ADD_U32 : VOP3Inst <"v_lshl_add_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_LSHL_OR_B32 : VOP3Inst <"v_lshl_or_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
} // End isReMaterializable = 1
// V_LSHL_ADD_U64: D0.u64 = (S0.u64 << S1.u[2:0]) + S2.u64
// src0 is shifted left by 0-4 (use “0” to get ADD_U64).
let SubtargetPredicate = isGFX940Plus in
defm V_LSHL_ADD_U64 : VOP3Inst <"v_lshl_add_u64", VOP3_Profile<VOP_I64_I64_I32_I64>>;
let OtherPredicates = [HasFP8ConversionInsts], mayRaiseFPException = 0,
SchedRW = [WriteFloatCvt] in {
let Constraints = "$vdst = $vdst_in", DisableEncoding = "$vdst_in" in {
defm V_CVT_PK_FP8_F32 : VOP3Inst<"v_cvt_pk_fp8_f32", VOP3_CVT_PK_F8_F32_Profile>;
defm V_CVT_PK_BF8_F32 : VOP3Inst<"v_cvt_pk_bf8_f32", VOP3_CVT_PK_F8_F32_Profile>;
let SubtargetPredicate = isGFX12Plus in {
defm V_CVT_SR_FP8_F32_gfx12 : VOP3Inst<"v_cvt_sr_fp8_f32_gfx12", VOP3_CVT_SR_F8_ByteSel_Profile<f32>>;
defm V_CVT_SR_BF8_F32_gfx12 : VOP3Inst<"v_cvt_sr_bf8_f32_gfx12", VOP3_CVT_SR_F8_ByteSel_Profile<f32>>;
}
}
// These instructions have non-standard use of op_sel. In particular they are
// using op_sel bits 2 and 3 while only having two sources. Therefore dummy
// src2 is used to hold the op_sel value.
let Constraints = "$vdst = $src2", DisableEncoding = "$src2", SubtargetPredicate = isGFX940Plus in {
defm V_CVT_SR_FP8_F32 : VOP3Inst<"v_cvt_sr_fp8_f32", VOP3_CVT_SR_F8_F32_Profile>;
defm V_CVT_SR_BF8_F32 : VOP3Inst<"v_cvt_sr_bf8_f32", VOP3_CVT_SR_F8_F32_Profile>;
}
}
class Cvt_PK_F8_F32_Pat<SDPatternOperator node, int index, VOP3_Pseudo inst> : GCNPat<
(i32 (node f32:$src0, f32:$src1, i32:$old, index)),
(inst !if(index, SRCMODS.DST_OP_SEL, 0), $src0, 0, $src1, $old, 0)
>;
class Cvt_SR_F8_F32_Pat<SDPatternOperator node, bits<2> index, VOP3_Pseudo inst> : GCNPat<
(i32 (node f32:$src0, i32:$src1, i32:$old, index)),
(inst !if(index{1}, SRCMODS.DST_OP_SEL, 0), $src0, 0, $src1,
!if(index{0}, SRCMODS.OP_SEL_0, 0), $old, 0)
>;
class Cvt_SR_F8_ByteSel_Pat<SDPatternOperator node, VOP3_Pseudo inst, ValueType SrcVT> : GCNPat<
(i32 (node (VOP3Mods SrcVT:$src0, i32:$src0_modifiers), (VOP3Mods i32:$src1, i32:$src1_modifiers),
i32:$old, timm:$byte_sel)),
(inst $src0_modifiers, $src0, $src1_modifiers, $src1, $old, (as_i32timm $byte_sel))
>;
let OtherPredicates = [HasFP8ConversionInsts] in {
foreach Index = [0, -1] in {
def : Cvt_PK_F8_F32_Pat<int_amdgcn_cvt_pk_fp8_f32, Index, V_CVT_PK_FP8_F32_e64>;
def : Cvt_PK_F8_F32_Pat<int_amdgcn_cvt_pk_bf8_f32, Index, V_CVT_PK_BF8_F32_e64>;
}
let SubtargetPredicate = isGFX940Plus in {
foreach Index = [0, 1, 2, 3] in {
def : Cvt_SR_F8_F32_Pat<int_amdgcn_cvt_sr_fp8_f32, Index, V_CVT_SR_FP8_F32_e64>;
def : Cvt_SR_F8_F32_Pat<int_amdgcn_cvt_sr_bf8_f32, Index, V_CVT_SR_BF8_F32_e64>;
}
}
let SubtargetPredicate = isGFX12Plus in {
def : Cvt_SR_F8_ByteSel_Pat<int_amdgcn_cvt_sr_fp8_f32, V_CVT_SR_FP8_F32_gfx12_e64, f32>;
def : Cvt_SR_F8_ByteSel_Pat<int_amdgcn_cvt_sr_bf8_f32, V_CVT_SR_BF8_F32_gfx12_e64, f32>;
}
}
class ThreeOp_i32_Pats <SDPatternOperator op1, SDPatternOperator op2, Instruction inst> : GCNPat <
// This matches (op2 (op1 i32:$src0, i32:$src1), i32:$src2) with conditions.
(ThreeOpFrag<op1, op2> i32:$src0, i32:$src1, i32:$src2),
(inst VSrc_b32:$src0, VSrc_b32:$src1, VSrc_b32:$src2)
>;
def : ThreeOp_i32_Pats<cshl_32, add, V_LSHL_ADD_U32_e64>;
def : ThreeOp_i32_Pats<add, cshl_32, V_ADD_LSHL_U32_e64>;
def : ThreeOp_i32_Pats<add, add, V_ADD3_U32_e64>;
def : ThreeOp_i32_Pats<ptradd, ptradd, V_ADD3_U32_e64>;
def : ThreeOp_i32_Pats<cshl_32, or, V_LSHL_OR_B32_e64>;
def : ThreeOp_i32_Pats<and, or, V_AND_OR_B32_e64>;
def : ThreeOp_i32_Pats<or, or, V_OR3_B32_e64>;
def : ThreeOp_i32_Pats<xor, add, V_XAD_U32_e64>;
def : GCNPat<
(DivergentBinFrag<mul> i32:$src0, IsPow2Plus1:$src1),
(V_LSHL_ADD_U32_e64 i32:$src0, (i32 (Log2_32 imm:$src1)), i32:$src0)>;
let SubtargetPredicate = isGFX940Plus in
def : GCNPat<
(ThreeOpFrag<shl_0_to_4, add> i64:$src0, i32:$src1, i64:$src2),
(V_LSHL_ADD_U64_e64 VSrc_b64:$src0, VSrc_b32:$src1, VSrc_b64:$src2)
>;
def : VOPBinOpClampPat<saddsat, V_ADD_I32_e64, i32>;
def : VOPBinOpClampPat<ssubsat, V_SUB_I32_e64, i32>;
def : GCNPat<(DivergentBinFrag<or> (or_oneuse i64:$src0, i64:$src1), i64:$src2),
(REG_SEQUENCE VReg_64,
(V_OR3_B32_e64 (i32 (EXTRACT_SUBREG $src0, sub0)),
(i32 (EXTRACT_SUBREG $src1, sub0)),
(i32 (EXTRACT_SUBREG $src2, sub0))), sub0,
(V_OR3_B32_e64 (i32 (EXTRACT_SUBREG $src0, sub1)),
(i32 (EXTRACT_SUBREG $src1, sub1)),
(i32 (EXTRACT_SUBREG $src2, sub1))), sub1)>;
// FIXME: Probably should hardcode clamp bit in pseudo and avoid this.
class OpSelBinOpClampPat<SDPatternOperator node,
Instruction inst> : GCNPat<
(node (i16 (VOP3OpSel i16:$src0, i32:$src0_modifiers)),
(i16 (VOP3OpSel i16:$src1, i32:$src1_modifiers))),
(inst $src0_modifiers, $src0, $src1_modifiers, $src1, DSTCLAMP.ENABLE, 0)
>;
def : OpSelBinOpClampPat<saddsat, V_ADD_I16_e64>;
def : OpSelBinOpClampPat<ssubsat, V_SUB_I16_e64>;
} // End SubtargetPredicate = isGFX9Plus
multiclass IMAD32_Pats <VOP3_Pseudo inst> {
def : GCNPat <
(ThreeOpFrag<mul, add> i32:$src0, i32:$src1, i32:$src2),
(EXTRACT_SUBREG (inst i32:$src0, i32:$src1,
(REG_SEQUENCE SReg_64, // Use scalar and let it be legalized
$src2, sub0,
(i32 (IMPLICIT_DEF)), sub1),
0 /* clamp */),
sub0)
>;
// GISel-specific pattern that avoids creating a SGPR->VGPR copy if
// $src2 is a VGPR.
def : GCNPat <
(ThreeOpFrag<mul, add> i32:$src0, i32:$src1, VGPR_32:$src2),
(EXTRACT_SUBREG (inst i32:$src0, i32:$src1,
(REG_SEQUENCE VReg_64,
$src2, sub0,
(i32 (IMPLICIT_DEF)), sub1),
0 /* clamp */),
sub0)
>;
// Immediate src2 in the pattern above will not fold because it would be partially
// undef. Hence define specialized pattern for this case.
def : GCNPat <
(ThreeOpFrag<mul, add> i32:$src0, i32:$src1, (i32 imm:$src2)),
(EXTRACT_SUBREG (inst i32:$src0, i32:$src1, (i64 (as_i64imm $src2)), 0 /* clamp */), sub0)
>;
}
// Handle cases where amdgpu-codegenprepare-mul24 made a mul24 instead of a normal mul.
// We need to separate this because otherwise OtherPredicates would be overriden.
class IMAD32_Mul24_Pat<VOP3_Pseudo inst>: GCNPat <
(i64 (add (i64 (AMDGPUmul_u24 i32:$src0, i32:$src1)), i64:$src2)),
(inst $src0, $src1, $src2, 0 /* clamp */)
>;
// exclude pre-GFX9 where it was slow
let OtherPredicates = [HasNotMADIntraFwdBug], SubtargetPredicate = isGFX9Plus in {
defm : IMAD32_Pats<V_MAD_U64_U32_e64>;
def : IMAD32_Mul24_Pat<V_MAD_U64_U32_e64>;
}
let OtherPredicates = [HasMADIntraFwdBug], SubtargetPredicate = isGFX11Only in {
defm : IMAD32_Pats<V_MAD_U64_U32_gfx11_e64>;
def : IMAD32_Mul24_Pat<V_MAD_U64_U32_gfx11_e64>;
}
def VOP3_PERMLANE_Profile : VOP3_Profile<VOPProfile <[i32, i32, i32, i32]>, VOP3_OPSEL> {
let InsVOP3OpSel = (ins IntOpSelMods:$src0_modifiers, VRegSrc_32:$src0,
IntOpSelMods:$src1_modifiers, SSrc_b32:$src1,
IntOpSelMods:$src2_modifiers, SSrc_b32:$src2,
VGPR_32:$vdst_in, op_sel0:$op_sel);
let HasClamp = 0;
let HasExtVOP3DPP = 0;
let HasExtDPP = 0;
}
def VOP3_PERMLANE_VAR_Profile : VOP3_Profile<VOPProfile <[i32, i32, i32, untyped]>, VOP3_OPSEL> {
let InsVOP3OpSel = (ins IntOpSelMods:$src0_modifiers, VRegSrc_32:$src0,
IntOpSelMods:$src1_modifiers, VRegSrc_32:$src1,
VGPR_32:$vdst_in, op_sel0:$op_sel);
let HasClamp = 0;
let HasExtVOP3DPP = 0;
let HasExtDPP = 0;
}
def opsel_i1timm : SDNodeXForm<timm, [{
return CurDAG->getTargetConstant(
N->getZExtValue() ? SISrcMods::OP_SEL_0 : SISrcMods::NONE,
SDLoc(N), MVT::i32);
}]>;
def gi_opsel_i1timm : GICustomOperandRenderer<"renderOpSelTImm">,
GISDNodeXFormEquiv<opsel_i1timm>;
class PermlanePat<SDPatternOperator permlane,
Instruction inst, ValueType vt> : GCNPat<
(vt (permlane vt:$vdst_in, vt:$src0, i32:$src1, i32:$src2,
timm:$fi, timm:$bc)),
(inst (opsel_i1timm $fi), VGPR_32:$src0, (opsel_i1timm $bc),
SCSrc_b32:$src1, 0, SCSrc_b32:$src2, VGPR_32:$vdst_in)
>;
class PermlaneVarPat<SDPatternOperator permlane,
Instruction inst> : GCNPat<
(permlane i32:$vdst_in, i32:$src0, i32:$src1,
timm:$fi, timm:$bc),
(inst (opsel_i1timm $fi), VGPR_32:$src0, (opsel_i1timm $bc),
VGPR_32:$src1, VGPR_32:$vdst_in)
>;
let SubtargetPredicate = isGFX10Plus in {
let isCommutable = 1, isReMaterializable = 1 in {
defm V_XOR3_B32 : VOP3Inst <"v_xor3_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
} // End isCommutable = 1, isReMaterializable = 1
def : ThreeOp_i32_Pats<xor, xor, V_XOR3_B32_e64>;
let Constraints = "$vdst = $vdst_in", DisableEncoding="$vdst_in" in {
defm V_PERMLANE16_B32 : VOP3Inst<"v_permlane16_b32", VOP3_PERMLANE_Profile>;
defm V_PERMLANEX16_B32 : VOP3Inst<"v_permlanex16_b32", VOP3_PERMLANE_Profile>;
} // End $vdst = $vdst_in, DisableEncoding $vdst_in
foreach vt = Reg32Types.types in {
def : PermlanePat<int_amdgcn_permlane16, V_PERMLANE16_B32_e64, vt>;
def : PermlanePat<int_amdgcn_permlanex16, V_PERMLANEX16_B32_e64, vt>;
}
defm V_ADD_NC_U16 : VOP3Inst <"v_add_nc_u16", VOP3_Profile<VOP_I16_I16_I16, VOP3_OPSEL>, add>;
defm V_SUB_NC_U16 : VOP3Inst <"v_sub_nc_u16", VOP3_Profile<VOP_I16_I16_I16, VOP3_OPSEL>, sub>;
def : OpSelBinOpClampPat<uaddsat, V_ADD_NC_U16_e64>;
def : OpSelBinOpClampPat<usubsat, V_SUB_NC_U16_e64>;
// Undo sub x, c -> add x, -c canonicalization since c is more likely
// an inline immediate than -c.
def : GCNPat<
(add i16:$src0, (i16 NegSubInlineIntConst16:$src1)),
(V_SUB_NC_U16_e64 0, VSrc_b16:$src0, 0, NegSubInlineIntConst16:$src1, 0, 0)
>;
} // End SubtargetPredicate = isGFX10Plus
let SubtargetPredicate = isGFX12Plus in {
let Constraints = "$vdst = $vdst_in", DisableEncoding="$vdst_in" in {
defm V_PERMLANE16_VAR_B32 : VOP3Inst<"v_permlane16_var_b32", VOP3_PERMLANE_VAR_Profile>;
defm V_PERMLANEX16_VAR_B32 : VOP3Inst<"v_permlanex16_var_b32", VOP3_PERMLANE_VAR_Profile>;
} // End $vdst = $vdst_in, DisableEncoding $vdst_in
def : PermlaneVarPat<int_amdgcn_permlane16_var, V_PERMLANE16_VAR_B32_e64>;
def : PermlaneVarPat<int_amdgcn_permlanex16_var, V_PERMLANEX16_VAR_B32_e64>;
} // End SubtargetPredicate = isGFX12Plus
class DivFmasPat<ValueType vt, Instruction inst, Register CondReg> : GCNPat<
(AMDGPUdiv_fmas (vt (VOP3Mods vt:$src0, i32:$src0_modifiers)),
(vt (VOP3Mods vt:$src1, i32:$src1_modifiers)),
(vt (VOP3Mods vt:$src2, i32:$src2_modifiers)),
(i1 CondReg)),
(inst $src0_modifiers, $src0, $src1_modifiers, $src1, $src2_modifiers, $src2)
>;
let WaveSizePredicate = isWave64 in {
def : DivFmasPat<f32, V_DIV_FMAS_F32_e64, VCC>;
def : DivFmasPat<f64, V_DIV_FMAS_F64_e64, VCC>;
}
let WaveSizePredicate = isWave32 in {
def : DivFmasPat<f32, V_DIV_FMAS_F32_e64, VCC_LO>;
def : DivFmasPat<f64, V_DIV_FMAS_F64_e64, VCC_LO>;
}
class VOP3_DOT_Profile<VOPProfile P> : VOP3_Profile<P, VOP3_OPSEL> {
let HasClamp = 0;
let HasOMod = 0;
}
let SubtargetPredicate = isGFX11Plus in {
defm V_MAXMIN_F32 : VOP3Inst<"v_maxmin_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
defm V_MINMAX_F32 : VOP3Inst<"v_minmax_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
defm V_MAXMIN_F16 : VOP3Inst<"v_maxmin_f16", VOP3_Profile<VOP_F16_F16_F16_F16>>;
defm V_MINMAX_F16 : VOP3Inst<"v_minmax_f16", VOP3_Profile<VOP_F16_F16_F16_F16>>;
defm V_MAXMIN_U32 : VOP3Inst<"v_maxmin_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_MINMAX_U32 : VOP3Inst<"v_minmax_u32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_MAXMIN_I32 : VOP3Inst<"v_maxmin_i32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_MINMAX_I32 : VOP3Inst<"v_minmax_i32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
defm V_CVT_PK_I16_F32 : VOP3Inst<"v_cvt_pk_i16_f32", VOP3_Profile<VOP_V2I16_F32_F32>>;
defm V_CVT_PK_U16_F32 : VOP3Inst<"v_cvt_pk_u16_f32", VOP3_Profile<VOP_V2I16_F32_F32>>;
} // End SubtargetPredicate = isGFX11Plus
let SubtargetPredicate = isGFX12Plus, ReadsModeReg = 0 in {
defm V_MAXIMUMMINIMUM_F32 : VOP3Inst<"v_maximumminimum_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
defm V_MINIMUMMAXIMUM_F32 : VOP3Inst<"v_minimummaximum_f32", VOP3_Profile<VOP_F32_F32_F32_F32>>;
defm V_MAXIMUMMINIMUM_F16 : VOP3Inst<"v_maximumminimum_f16", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>>;
defm V_MINIMUMMAXIMUM_F16 : VOP3Inst<"v_minimummaximum_f16", VOP3_Profile<VOP_F16_F16_F16_F16, VOP3_OPSEL>>;
} // End SubtargetPredicate = isGFX12Plus, ReadsModeReg = 0
let OtherPredicates = [HasDot9Insts], IsDOT=1 in {
defm V_DOT2_F16_F16 : VOP3Inst<"v_dot2_f16_f16", VOP3_DOT_Profile<VOP_F16_V2F16_V2F16_F16>, int_amdgcn_fdot2_f16_f16>;
defm V_DOT2_BF16_BF16 : VOP3Inst<"v_dot2_bf16_bf16", VOP3_DOT_Profile<VOP_BF16_V2BF16_V2BF16_BF16>, int_amdgcn_fdot2_bf16_bf16>;
}
class VOP_Pseudo_Scalar<RegisterClass Dst, RegisterOperand SrcOp,
ValueType dstVt, ValueType srcVt = dstVt>
: VOPProfile<[dstVt, srcVt, untyped, untyped]> {
let DstRC = VOPDstOperand<Dst>;
let Src0RC64 = SrcOp;
let HasOMod = 1;
let HasModifiers = 1;
}
def VOP_Pseudo_Scalar_F32 : VOP_Pseudo_Scalar<SReg_32_XEXEC, SSrc_f32, f32>;
def VOP_Pseudo_Scalar_F16 : VOP_Pseudo_Scalar<SReg_32_XEXEC, SSrc_f16, f32, f16>;
let SubtargetPredicate = HasPseudoScalarTrans, TRANS = 1,
isReMaterializable = 1, SchedRW = [WritePseudoScalarTrans] in {
defm V_S_EXP_F32 : VOP3PseudoScalarInst<"v_s_exp_f32", VOP_Pseudo_Scalar_F32, AMDGPUexp>;
defm V_S_EXP_F16 : VOP3PseudoScalarInst<"v_s_exp_f16", VOP_Pseudo_Scalar_F16>;
defm V_S_LOG_F32 : VOP3PseudoScalarInst<"v_s_log_f32", VOP_Pseudo_Scalar_F32, AMDGPUlog>;
defm V_S_LOG_F16 : VOP3PseudoScalarInst<"v_s_log_f16", VOP_Pseudo_Scalar_F16>;
defm V_S_RCP_F32 : VOP3PseudoScalarInst<"v_s_rcp_f32", VOP_Pseudo_Scalar_F32, AMDGPUrcp>;
defm V_S_RCP_F16 : VOP3PseudoScalarInst<"v_s_rcp_f16", VOP_Pseudo_Scalar_F16>;
defm V_S_RSQ_F32 : VOP3PseudoScalarInst<"v_s_rsq_f32", VOP_Pseudo_Scalar_F32, AMDGPUrsq>;
defm V_S_RSQ_F16 : VOP3PseudoScalarInst<"v_s_rsq_f16", VOP_Pseudo_Scalar_F16>;
defm V_S_SQRT_F32 : VOP3PseudoScalarInst<"v_s_sqrt_f32", VOP_Pseudo_Scalar_F32, any_amdgcn_sqrt>;
defm V_S_SQRT_F16 : VOP3PseudoScalarInst<"v_s_sqrt_f16", VOP_Pseudo_Scalar_F16>;
}
class PseudoScalarPatF16<SDPatternOperator node, VOP3_Pseudo inst> : GCNPat <
(f16 (UniformUnaryFrag<node> (f16 (VOP3Mods0 f16:$src0, i32:$src0_modifiers,
i1:$clamp, i32:$omod)))),
(f16 (COPY_TO_REGCLASS (f32 (inst i32:$src0_modifiers, f16:$src0, i1:$clamp,
i32:$omod)),
SReg_32_XEXEC))
>;
let SubtargetPredicate = HasPseudoScalarTrans in {
def : PseudoScalarPatF16<AMDGPUexpf16, V_S_EXP_F16_e64>;
def : PseudoScalarPatF16<AMDGPUlogf16, V_S_LOG_F16_e64>;
def : PseudoScalarPatF16<AMDGPUrcp, V_S_RCP_F16_e64>;
def : PseudoScalarPatF16<AMDGPUrsq, V_S_RSQ_F16_e64>;
def : PseudoScalarPatF16<any_amdgcn_sqrt, V_S_SQRT_F16_e64>;
}
//===----------------------------------------------------------------------===//
// Integer Clamp Patterns
//===----------------------------------------------------------------------===//
class getClampPat<VOPProfile P, SDPatternOperator node> {
dag ret3 = (P.DstVT (node P.Src0VT:$src0, P.Src1VT:$src1, P.Src2VT:$src2));
dag ret2 = (P.DstVT (node P.Src0VT:$src0, P.Src1VT:$src1));
dag ret1 = (P.DstVT (node P.Src0VT:$src0));
dag ret = !if(!eq(P.NumSrcArgs, 3), ret3,
!if(!eq(P.NumSrcArgs, 2), ret2,
ret1));
}
class getClampRes<VOPProfile P, Instruction inst> {
dag ret3 = (inst P.Src0VT:$src0, P.Src1VT:$src1, P.Src2VT:$src2, (i1 0));
dag ret2 = (inst P.Src0VT:$src0, P.Src1VT:$src1, (i1 0));
dag ret1 = (inst P.Src0VT:$src0, (i1 0));
dag ret = !if(!eq(P.NumSrcArgs, 3), ret3,
!if(!eq(P.NumSrcArgs, 2), ret2,
ret1));
}
class IntClampPat<VOP3InstBase inst, SDPatternOperator node> : GCNPat<
getClampPat<inst.Pfl, node>.ret,
getClampRes<inst.Pfl, inst>.ret
>;
def : IntClampPat<V_MAD_I32_I24_e64, AMDGPUmad_i24>;
def : IntClampPat<V_MAD_U32_U24_e64, AMDGPUmad_u24>;
def : IntClampPat<V_SAD_U8_e64, int_amdgcn_sad_u8>;
def : IntClampPat<V_SAD_HI_U8_e64, int_amdgcn_sad_hi_u8>;
def : IntClampPat<V_SAD_U16_e64, int_amdgcn_sad_u16>;
def : IntClampPat<V_MSAD_U8_e64, int_amdgcn_msad_u8>;
def : IntClampPat<V_MQSAD_PK_U16_U8_e64, int_amdgcn_mqsad_pk_u16_u8>;
def : IntClampPat<V_QSAD_PK_U16_U8_e64, int_amdgcn_qsad_pk_u16_u8>;
def : IntClampPat<V_MQSAD_U32_U8_e64, int_amdgcn_mqsad_u32_u8>;
//===----------------------------------------------------------------------===//
// Target-specific instruction encodings.
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// GFX12.
//===----------------------------------------------------------------------===//
defm V_MIN3_NUM_F32 : VOP3_Realtriple_with_name_gfx12<0x229, "V_MIN3_F32", "v_min3_num_f32">;
defm V_MAX3_NUM_F32 : VOP3_Realtriple_with_name_gfx12<0x22a, "V_MAX3_F32", "v_max3_num_f32">;
defm V_MIN3_NUM_F16 : VOP3_Realtriple_with_name_gfx12<0x22b, "V_MIN3_F16", "v_min3_num_f16">;
defm V_MAX3_NUM_F16 : VOP3_Realtriple_with_name_gfx12<0x22c, "V_MAX3_F16", "v_max3_num_f16">;
defm V_MINIMUM3_F32 : VOP3Only_Realtriple_gfx12<0x22d>;
defm V_MAXIMUM3_F32 : VOP3Only_Realtriple_gfx12<0x22e>;
defm V_MINIMUM3_F16 : VOP3Only_Realtriple_t16_gfx12<0x22f>;
defm V_MAXIMUM3_F16 : VOP3Only_Realtriple_t16_gfx12<0x230>;
defm V_MED3_NUM_F32 : VOP3_Realtriple_with_name_gfx12<0x231, "V_MED3_F32", "v_med3_num_f32">;
defm V_MED3_NUM_F16 : VOP3_Realtriple_with_name_gfx12<0x232, "V_MED3_F16", "v_med3_num_f16">;
defm V_MINMAX_NUM_F32 : VOP3_Realtriple_with_name_gfx12<0x268, "V_MINMAX_F32", "v_minmax_num_f32">;
defm V_MAXMIN_NUM_F32 : VOP3_Realtriple_with_name_gfx12<0x269, "V_MAXMIN_F32", "v_maxmin_num_f32">;
defm V_MINMAX_NUM_F16 : VOP3_Realtriple_with_name_gfx12<0x26a, "V_MINMAX_F16", "v_minmax_num_f16">;
defm V_MAXMIN_NUM_F16 : VOP3_Realtriple_with_name_gfx12<0x26b, "V_MAXMIN_F16", "v_maxmin_num_f16">;
defm V_MINIMUMMAXIMUM_F32 : VOP3Only_Realtriple_gfx12<0x26c>;
defm V_MAXIMUMMINIMUM_F32 : VOP3Only_Realtriple_gfx12<0x26d>;
defm V_MINIMUMMAXIMUM_F16 : VOP3Only_Realtriple_t16_gfx12<0x26e>;
defm V_MAXIMUMMINIMUM_F16 : VOP3Only_Realtriple_t16_gfx12<0x26f>;
defm V_S_EXP_F32 : VOP3Only_Real_Base_gfx12<0x280>;
defm V_S_EXP_F16 : VOP3Only_Real_Base_gfx12<0x281>;
defm V_S_LOG_F32 : VOP3Only_Real_Base_gfx12<0x282>;
defm V_S_LOG_F16 : VOP3Only_Real_Base_gfx12<0x283>;
defm V_S_RCP_F32 : VOP3Only_Real_Base_gfx12<0x284>;
defm V_S_RCP_F16 : VOP3Only_Real_Base_gfx12<0x285>;
defm V_S_RSQ_F32 : VOP3Only_Real_Base_gfx12<0x286>;
defm V_S_RSQ_F16 : VOP3Only_Real_Base_gfx12<0x287>;
defm V_S_SQRT_F32 : VOP3Only_Real_Base_gfx12<0x288>;
defm V_S_SQRT_F16 : VOP3Only_Real_Base_gfx12<0x289>;
defm V_MAD_CO_U64_U32 : VOP3be_Real_with_name_gfx12<0x2fe, "V_MAD_U64_U32", "v_mad_co_u64_u32">;
defm V_MAD_CO_I64_I32 : VOP3be_Real_with_name_gfx12<0x2ff, "V_MAD_I64_I32", "v_mad_co_i64_i32">;
defm V_MINIMUM_F64 : VOP3Only_Real_Base_gfx12<0x341>;
defm V_MAXIMUM_F64 : VOP3Only_Real_Base_gfx12<0x342>;
defm V_MINIMUM_F32 : VOP3Only_Realtriple_gfx12<0x365>;
defm V_MAXIMUM_F32 : VOP3Only_Realtriple_gfx12<0x366>;
defm V_MINIMUM_F16 : VOP3Only_Realtriple_t16_gfx12<0x367>;
defm V_MAXIMUM_F16 : VOP3Only_Realtriple_t16_gfx12<0x368>;
defm V_PERMLANE16_VAR_B32 : VOP3Only_Real_Base_gfx12<0x30f>;
defm V_PERMLANEX16_VAR_B32 : VOP3Only_Real_Base_gfx12<0x310>;
defm V_CVT_PK_FP8_F32 : VOP3Only_Realtriple_gfx12<0x369>;
defm V_CVT_PK_BF8_F32 : VOP3Only_Realtriple_gfx12<0x36a>;
defm V_CVT_SR_FP8_F32_gfx12 : VOP3_Realtriple_with_name_gfx12<0x36b, "V_CVT_SR_FP8_F32_gfx12", "v_cvt_sr_fp8_f32" >;
defm V_CVT_SR_BF8_F32_gfx12 : VOP3_Realtriple_with_name_gfx12<0x36c, "V_CVT_SR_BF8_F32_gfx12", "v_cvt_sr_bf8_f32">;
//===----------------------------------------------------------------------===//
// GFX11, GFX12
//===----------------------------------------------------------------------===//
multiclass VOP3_Real_with_name_gfx11_gfx12<bits<10> op, string opName,
string asmName> :
VOP3_Real_with_name<GFX11Gen, op, opName, asmName>,
VOP3_Real_with_name<GFX12Gen, op, opName, asmName>;
multiclass VOP3_Realtriple_gfx11_gfx12<bits<10> op> :
VOP3_Realtriple<GFX11Gen, op>, VOP3_Realtriple<GFX12Gen, op>;
multiclass VOP3_Real_Base_gfx11_gfx12<bits<10> op> :
VOP3_Real_Base<GFX11Gen, op>, VOP3_Real_Base<GFX12Gen, op>;
multiclass VOP3_Realtriple_with_name_gfx11_gfx12<bits<10> op, string opName,
string asmName> :
VOP3_Realtriple_with_name<GFX11Gen, op, opName, asmName>,
VOP3_Realtriple_with_name<GFX12Gen, op, opName, asmName>;
multiclass VOP3Dot_Realtriple_gfx11_gfx12<bits<10> op> :
VOP3Dot_Realtriple<GFX11Gen, op>, VOP3Dot_Realtriple<GFX12Gen, op>;
multiclass VOP3be_Real_gfx11_gfx12<bits<10> op, string opName, string asmName> :
VOP3be_Real<GFX11Gen, op, opName, asmName>,
VOP3be_Real<GFX12Gen, op, opName, asmName>;
multiclass VOP3_Real_No_Suffix_gfx11_gfx12<bits<10> op> :
VOP3_Real_No_Suffix<GFX11Gen, op>, VOP3_Real_No_Suffix<GFX12Gen, op>;
defm V_FMA_DX9_ZERO_F32 : VOP3_Real_with_name_gfx11_gfx12<0x209, "V_FMA_LEGACY_F32", "v_fma_dx9_zero_f32">;
defm V_MAD_I32_I24 : VOP3_Realtriple_gfx11_gfx12<0x20a>;
defm V_MAD_U32_U24 : VOP3_Realtriple_gfx11_gfx12<0x20b>;
defm V_CUBEID_F32 : VOP3_Realtriple_gfx11_gfx12<0x20c>;
defm V_CUBESC_F32 : VOP3_Realtriple_gfx11_gfx12<0x20d>;
defm V_CUBETC_F32 : VOP3_Realtriple_gfx11_gfx12<0x20e>;
defm V_CUBEMA_F32 : VOP3_Realtriple_gfx11_gfx12<0x20f>;
defm V_BFE_U32 : VOP3_Realtriple_gfx11_gfx12<0x210>;
defm V_BFE_I32 : VOP3_Realtriple_gfx11_gfx12<0x211>;
defm V_BFI_B32 : VOP3_Realtriple_gfx11_gfx12<0x212>;
defm V_FMA_F32 : VOP3_Realtriple_gfx11_gfx12<0x213>;
defm V_FMA_F64 : VOP3_Real_Base_gfx11_gfx12<0x214>;
defm V_LERP_U8 : VOP3_Realtriple_gfx11_gfx12<0x215>;
defm V_ALIGNBIT_B32 : VOP3_Realtriple_gfx11_gfx12<0x216>;
defm V_ALIGNBYTE_B32 : VOP3_Realtriple_gfx11_gfx12<0x217>;
defm V_MULLIT_F32 : VOP3_Realtriple_gfx11_gfx12<0x218>;
defm V_MIN3_F32 : VOP3_Realtriple_gfx11<0x219>;
defm V_MIN3_I32 : VOP3_Realtriple_gfx11_gfx12<0x21a>;
defm V_MIN3_U32 : VOP3_Realtriple_gfx11_gfx12<0x21b>;
defm V_MAX3_F32 : VOP3_Realtriple_gfx11<0x21c>;
defm V_MAX3_I32 : VOP3_Realtriple_gfx11_gfx12<0x21d>;
defm V_MAX3_U32 : VOP3_Realtriple_gfx11_gfx12<0x21e>;
defm V_MED3_F32 : VOP3_Realtriple_gfx11<0x21f>;
defm V_MED3_I32 : VOP3_Realtriple_gfx11_gfx12<0x220>;
defm V_MED3_U32 : VOP3_Realtriple_gfx11_gfx12<0x221>;
defm V_SAD_U8 : VOP3_Realtriple_gfx11_gfx12<0x222>;
defm V_SAD_HI_U8 : VOP3_Realtriple_gfx11_gfx12<0x223>;
defm V_SAD_U16 : VOP3_Realtriple_gfx11_gfx12<0x224>;
defm V_SAD_U32 : VOP3_Realtriple_gfx11_gfx12<0x225>;
defm V_CVT_PK_U8_F32 : VOP3_Realtriple_gfx11_gfx12<0x226>;
defm V_DIV_FIXUP_F32 : VOP3_Real_Base_gfx11_gfx12<0x227>;
defm V_DIV_FIXUP_F64 : VOP3_Real_Base_gfx11_gfx12<0x228>;
defm V_DIV_FMAS_F32 : VOP3_Real_Base_gfx11_gfx12<0x237>;
defm V_DIV_FMAS_F64 : VOP3_Real_Base_gfx11_gfx12<0x238>;
defm V_MSAD_U8 : VOP3_Realtriple_gfx11_gfx12<0x239>;
defm V_QSAD_PK_U16_U8 : VOP3_Real_Base_gfx11_gfx12<0x23a>;
defm V_MQSAD_PK_U16_U8 : VOP3_Real_Base_gfx11_gfx12<0x23b>;
defm V_MQSAD_U32_U8 : VOP3_Real_Base_gfx11_gfx12<0x23d>;
defm V_XOR3_B32 : VOP3_Realtriple_gfx11_gfx12<0x240>;
defm V_MAD_U16 : VOP3_Realtriple_with_name_gfx11_gfx12<0x241, "V_MAD_U16_gfx9", "v_mad_u16">;
defm V_PERM_B32 : VOP3_Realtriple_gfx11_gfx12<0x244>;
defm V_XAD_U32 : VOP3_Realtriple_gfx11_gfx12<0x245>;
defm V_LSHL_ADD_U32 : VOP3_Realtriple_gfx11_gfx12<0x246>;
defm V_ADD_LSHL_U32 : VOP3_Realtriple_gfx11_gfx12<0x247>;
defm V_FMA_F16 : VOP3_Realtriple_with_name_gfx11_gfx12<0x248, "V_FMA_F16_gfx9", "v_fma_f16">;
defm V_MIN3_F16 : VOP3_Realtriple_gfx11<0x249>;
defm V_MIN3_I16 : VOP3_Realtriple_gfx11_gfx12<0x24a>;
defm V_MIN3_U16 : VOP3_Realtriple_gfx11_gfx12<0x24b>;
defm V_MAX3_F16 : VOP3_Realtriple_gfx11<0x24c>;
defm V_MAX3_I16 : VOP3_Realtriple_gfx11_gfx12<0x24d>;
defm V_MAX3_U16 : VOP3_Realtriple_gfx11_gfx12<0x24e>;
defm V_MED3_F16 : VOP3_Realtriple_gfx11<0x24f>;
defm V_MED3_I16 : VOP3_Realtriple_gfx11_gfx12<0x250>;
defm V_MED3_U16 : VOP3_Realtriple_gfx11_gfx12<0x251>;
defm V_MAD_I16 : VOP3_Realtriple_with_name_gfx11_gfx12<0x253, "V_MAD_I16_gfx9", "v_mad_i16">;
defm V_DIV_FIXUP_F16 : VOP3_Realtriple_with_name_gfx11_gfx12<0x254, "V_DIV_FIXUP_F16_gfx9", "v_div_fixup_f16">;
defm V_ADD3_U32 : VOP3_Realtriple_gfx11_gfx12<0x255>;
defm V_LSHL_OR_B32 : VOP3_Realtriple_gfx11_gfx12<0x256>;
defm V_AND_OR_B32 : VOP3_Realtriple_gfx11_gfx12<0x257>;
defm V_OR3_B32 : VOP3_Realtriple_gfx11_gfx12<0x258>;
defm V_MAD_U32_U16 : VOP3_Realtriple_gfx11_gfx12<0x259>;
defm V_MAD_I32_I16 : VOP3_Realtriple_gfx11_gfx12<0x25a>;
defm V_PERMLANE16_B32 : VOP3_Real_Base_gfx11_gfx12<0x25b>;
defm V_PERMLANEX16_B32 : VOP3_Real_Base_gfx11_gfx12<0x25c>;
defm V_MAXMIN_F32 : VOP3_Realtriple_gfx11<0x25e>;
defm V_MINMAX_F32 : VOP3_Realtriple_gfx11<0x25f>;
defm V_MAXMIN_F16 : VOP3_Realtriple_gfx11<0x260>;
defm V_MINMAX_F16 : VOP3_Realtriple_gfx11<0x261>;
defm V_MAXMIN_U32 : VOP3_Realtriple_gfx11_gfx12<0x262>;
defm V_MINMAX_U32 : VOP3_Realtriple_gfx11_gfx12<0x263>;
defm V_MAXMIN_I32 : VOP3_Realtriple_gfx11_gfx12<0x264>;
defm V_MINMAX_I32 : VOP3_Realtriple_gfx11_gfx12<0x265>;
defm V_DOT2_F16_F16 : VOP3Dot_Realtriple_gfx11_gfx12<0x266>;
defm V_DOT2_BF16_BF16 : VOP3Dot_Realtriple_gfx11_gfx12<0x267>;
defm V_DIV_SCALE_F32 : VOP3be_Real_gfx11_gfx12<0x2fc, "V_DIV_SCALE_F32", "v_div_scale_f32">;
defm V_DIV_SCALE_F64 : VOP3be_Real_gfx11_gfx12<0x2fd, "V_DIV_SCALE_F64", "v_div_scale_f64">;
defm V_MAD_U64_U32_gfx11 : VOP3be_Real_gfx11<0x2fe, "V_MAD_U64_U32_gfx11", "v_mad_u64_u32">;
defm V_MAD_I64_I32_gfx11 : VOP3be_Real_gfx11<0x2ff, "V_MAD_I64_I32_gfx11", "v_mad_i64_i32">;
defm V_ADD_NC_U16 : VOP3Only_Realtriple_gfx11_gfx12<0x303>;
defm V_SUB_NC_U16 : VOP3Only_Realtriple_gfx11_gfx12<0x304>;
defm V_MUL_LO_U16 : VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<0x305, "v_mul_lo_u16">;
defm V_CVT_PK_I16_F32 : VOP3_Realtriple_gfx11_gfx12<0x306>;
defm V_CVT_PK_U16_F32 : VOP3_Realtriple_gfx11_gfx12<0x307>;
defm V_MAX_U16 : VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<0x309, "v_max_u16">;
defm V_MAX_I16 : VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<0x30a, "v_max_i16">;
defm V_MIN_U16 : VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<0x30b, "v_min_u16">;
defm V_MIN_I16 : VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<0x30c, "v_min_i16">;
defm V_ADD_NC_I16 : VOP3_Realtriple_with_name_gfx11_gfx12<0x30d, "V_ADD_I16", "v_add_nc_i16">;
defm V_SUB_NC_I16 : VOP3_Realtriple_with_name_gfx11_gfx12<0x30e, "V_SUB_I16", "v_sub_nc_i16">;
defm V_PACK_B32_F16 : VOP3_Realtriple_gfx11_gfx12<0x311>;
defm V_CVT_PK_NORM_I16_F16 : VOP3_Realtriple_with_name_gfx11_gfx12<0x312, "V_CVT_PKNORM_I16_F16" , "v_cvt_pk_norm_i16_f16" >;
defm V_CVT_PK_NORM_U16_F16 : VOP3_Realtriple_with_name_gfx11_gfx12<0x313, "V_CVT_PKNORM_U16_F16" , "v_cvt_pk_norm_u16_f16" >;
defm V_SUB_NC_I32 : VOP3_Realtriple_with_name_gfx11_gfx12<0x325, "V_SUB_I32", "v_sub_nc_i32">;
defm V_ADD_NC_I32 : VOP3_Realtriple_with_name_gfx11_gfx12<0x326, "V_ADD_I32", "v_add_nc_i32">;
defm V_ADD_F64 : VOP3_Real_Base_gfx11<0x327>;
defm V_MUL_F64 : VOP3_Real_Base_gfx11<0x328>;
defm V_MIN_F64 : VOP3_Real_Base_gfx11<0x329>;
defm V_MAX_F64 : VOP3_Real_Base_gfx11<0x32a>;
defm V_LDEXP_F64 : VOP3_Real_Base_gfx11_gfx12<0x32b>;
defm V_MUL_LO_U32 : VOP3_Real_Base_gfx11_gfx12<0x32c>;
defm V_MUL_HI_U32 : VOP3_Real_Base_gfx11_gfx12<0x32d>;
defm V_MUL_HI_I32 : VOP3_Real_Base_gfx11_gfx12<0x32e>;
defm V_TRIG_PREOP_F64 : VOP3_Real_Base_gfx11_gfx12<0x32f>;
defm V_LSHLREV_B16 : VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<0x338, "v_lshlrev_b16">;
defm V_LSHRREV_B16 : VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<0x339, "v_lshrrev_b16">;
defm V_ASHRREV_I16 : VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12<0x33a, "v_ashrrev_i16">;
defm V_LSHLREV_B64 : VOP3_Real_Base_gfx11<0x33c>;
defm V_LSHRREV_B64 : VOP3_Real_Base_gfx11_gfx12<0x33d>;
defm V_ASHRREV_I64 : VOP3_Real_Base_gfx11_gfx12<0x33e>;
defm V_READLANE_B32 : VOP3_Real_No_Suffix_gfx11_gfx12<0x360>; // Pseudo in VOP2
let InOperandList = (ins SSrcOrLds_b32:$src0, SCSrc_b32:$src1, VGPR_32:$vdst_in) in {
defm V_WRITELANE_B32 : VOP3_Real_No_Suffix_gfx11_gfx12<0x361>; // Pseudo in VOP2
} // End InOperandList = (ins SSrcOrLds_b32:$src0, SCSrc_b32:$src1, VGPR_32:$vdst_in)
defm V_AND_B16_t16 : VOP3Only_Realtriple_t16_gfx11_gfx12<0x362, "v_and_b16">;
defm V_AND_B16_fake16 : VOP3Only_Realtriple_t16_gfx11_gfx12<0x362, "v_and_b16">;
defm V_OR_B16_t16 : VOP3Only_Realtriple_t16_gfx11_gfx12<0x363, "v_or_b16">;
defm V_OR_B16_fake16 : VOP3Only_Realtriple_t16_gfx11_gfx12<0x363, "v_or_b16">;
defm V_XOR_B16_t16 : VOP3Only_Realtriple_t16_gfx11_gfx12<0x364, "v_xor_b16">;
defm V_XOR_B16_fake16 : VOP3Only_Realtriple_t16_gfx11_gfx12<0x364, "v_xor_b16">;
//===----------------------------------------------------------------------===//
// GFX10.
//===----------------------------------------------------------------------===//
let AssemblerPredicate = isGFX10Only, DecoderNamespace = "GFX10" in {
multiclass VOP3_Real_gfx10<bits<10> op> {
def _gfx10 :
VOP3_Real<!cast<VOP_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX10>,
VOP3e_gfx10<op, !cast<VOP_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3_Real_No_Suffix_gfx10<bits<10> op> {
def _gfx10 :
VOP3_Real<!cast<VOP_Pseudo>(NAME), SIEncodingFamily.GFX10>,
VOP3e_gfx10<op, !cast<VOP_Pseudo>(NAME).Pfl>;
}
multiclass VOP3_Real_gfx10_with_name<bits<10> op, string opName,
string asmName> {
def _gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(opName#"_e64"), SIEncodingFamily.GFX10>,
VOP3e_gfx10<op, !cast<VOP3_Pseudo>(opName#"_e64").Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(opName#"_e64");
let AsmString = asmName # ps.AsmOperands;
let IsSingle = 1;
}
}
multiclass VOP3be_Real_gfx10<bits<10> op> {
def _gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX10>,
VOP3be_gfx10<op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3Interp_Real_gfx10<bits<10> op> {
def _gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.GFX10>,
VOP3Interp_gfx10<op, !cast<VOP3_Pseudo>(NAME).Pfl>;
}
multiclass VOP3OpSel_Real_gfx10<bits<10> op> {
def _gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX10>,
VOP3OpSel_gfx10<op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3OpSel_Real_gfx10_with_name<bits<10> op, string opName,
string asmName> {
def _gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(opName#"_e64"), SIEncodingFamily.GFX10>,
VOP3OpSel_gfx10<op, !cast<VOP3_Pseudo>(opName#"_e64").Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(opName#"_e64");
let AsmString = asmName # ps.AsmOperands;
}
}
} // End AssemblerPredicate = isGFX10Only, DecoderNamespace = "GFX10"
defm V_READLANE_B32 : VOP3_Real_No_Suffix_gfx10<0x360>;
let InOperandList = (ins SSrcOrLds_b32:$src0, SCSrc_b32:$src1, VGPR_32:$vdst_in) in {
defm V_WRITELANE_B32 : VOP3_Real_No_Suffix_gfx10<0x361>;
} // End InOperandList = (ins SSrcOrLds_b32:$src0, SCSrc_b32:$src1, VGPR_32:$vdst_in)
let SubtargetPredicate = isGFX10Before1030 in {
defm V_MUL_LO_I32 : VOP3_Real_gfx10<0x16b>;
}
defm V_XOR3_B32 : VOP3_Real_gfx10<0x178>;
defm V_LSHLREV_B64 : VOP3_Real_gfx10<0x2ff>;
defm V_LSHRREV_B64 : VOP3_Real_gfx10<0x300>;
defm V_ASHRREV_I64 : VOP3_Real_gfx10<0x301>;
defm V_PERM_B32 : VOP3_Real_gfx10<0x344>;
defm V_XAD_U32 : VOP3_Real_gfx10<0x345>;
defm V_LSHL_ADD_U32 : VOP3_Real_gfx10<0x346>;
defm V_ADD_LSHL_U32 : VOP3_Real_gfx10<0x347>;
defm V_ADD3_U32 : VOP3_Real_gfx10<0x36d>;
defm V_LSHL_OR_B32 : VOP3_Real_gfx10<0x36f>;
defm V_AND_OR_B32 : VOP3_Real_gfx10<0x371>;
defm V_OR3_B32 : VOP3_Real_gfx10<0x372>;
// TODO-GFX10: add MC tests for v_add/sub_nc_i16
defm V_ADD_NC_I16 :
VOP3OpSel_Real_gfx10_with_name<0x30d, "V_ADD_I16", "v_add_nc_i16">;
defm V_SUB_NC_I16 :
VOP3OpSel_Real_gfx10_with_name<0x30e, "V_SUB_I16", "v_sub_nc_i16">;
defm V_SUB_NC_I32 :
VOP3_Real_gfx10_with_name<0x376, "V_SUB_I32", "v_sub_nc_i32">;
defm V_ADD_NC_I32 :
VOP3_Real_gfx10_with_name<0x37f, "V_ADD_I32", "v_add_nc_i32">;
defm V_INTERP_P1_F32_e64 : VOP3Interp_Real_gfx10<0x200>;
defm V_INTERP_P2_F32_e64 : VOP3Interp_Real_gfx10<0x201>;
defm V_INTERP_MOV_F32_e64 : VOP3Interp_Real_gfx10<0x202>;
defm V_INTERP_P1LL_F16 : VOP3Interp_Real_gfx10<0x342>;
defm V_INTERP_P1LV_F16 : VOP3Interp_Real_gfx10<0x343>;
defm V_INTERP_P2_F16 : VOP3Interp_Real_gfx10<0x35a>;
defm V_PACK_B32_F16 : VOP3OpSel_Real_gfx10<0x311>;
defm V_CVT_PKNORM_I16_F16 : VOP3OpSel_Real_gfx10<0x312>;
defm V_CVT_PKNORM_U16_F16 : VOP3OpSel_Real_gfx10<0x313>;
defm V_MIN3_F16 : VOP3OpSel_Real_gfx10<0x351>;
defm V_MIN3_I16 : VOP3OpSel_Real_gfx10<0x352>;
defm V_MIN3_U16 : VOP3OpSel_Real_gfx10<0x353>;
defm V_MAX3_F16 : VOP3OpSel_Real_gfx10<0x354>;
defm V_MAX3_I16 : VOP3OpSel_Real_gfx10<0x355>;
defm V_MAX3_U16 : VOP3OpSel_Real_gfx10<0x356>;
defm V_MED3_F16 : VOP3OpSel_Real_gfx10<0x357>;
defm V_MED3_I16 : VOP3OpSel_Real_gfx10<0x358>;
defm V_MED3_U16 : VOP3OpSel_Real_gfx10<0x359>;
defm V_MAD_U32_U16 : VOP3OpSel_Real_gfx10<0x373>;
defm V_MAD_I32_I16 : VOP3OpSel_Real_gfx10<0x375>;
defm V_MAD_U16 :
VOP3OpSel_Real_gfx10_with_name<0x340, "V_MAD_U16_gfx9", "v_mad_u16">;
defm V_FMA_F16 :
VOP3OpSel_Real_gfx10_with_name<0x34b, "V_FMA_F16_gfx9", "v_fma_f16">;
defm V_MAD_I16 :
VOP3OpSel_Real_gfx10_with_name<0x35e, "V_MAD_I16_gfx9", "v_mad_i16">;
defm V_DIV_FIXUP_F16 :
VOP3OpSel_Real_gfx10_with_name<0x35f, "V_DIV_FIXUP_F16_gfx9", "v_div_fixup_f16">;
defm V_ADD_NC_U16 : VOP3OpSel_Real_gfx10<0x303>;
defm V_SUB_NC_U16 : VOP3OpSel_Real_gfx10<0x304>;
// FIXME-GFX10-OPSEL: Need to add "selective" opsel support to some of these
// (they do not support SDWA or DPP).
defm V_MUL_LO_U16 : VOP3_Real_gfx10_with_name<0x305, "V_MUL_LO_U16", "v_mul_lo_u16">;
defm V_LSHRREV_B16 : VOP3_Real_gfx10_with_name<0x307, "V_LSHRREV_B16", "v_lshrrev_b16">;
defm V_ASHRREV_I16 : VOP3_Real_gfx10_with_name<0x308, "V_ASHRREV_I16", "v_ashrrev_i16">;
defm V_MAX_U16 : VOP3_Real_gfx10_with_name<0x309, "V_MAX_U16", "v_max_u16">;
defm V_MAX_I16 : VOP3_Real_gfx10_with_name<0x30a, "V_MAX_I16", "v_max_i16">;
defm V_MIN_U16 : VOP3_Real_gfx10_with_name<0x30b, "V_MIN_U16", "v_min_u16">;
defm V_MIN_I16 : VOP3_Real_gfx10_with_name<0x30c, "V_MIN_I16", "v_min_i16">;
defm V_LSHLREV_B16 : VOP3_Real_gfx10_with_name<0x314, "V_LSHLREV_B16", "v_lshlrev_b16">;
defm V_PERMLANE16_B32 : VOP3OpSel_Real_gfx10<0x377>;
defm V_PERMLANEX16_B32 : VOP3OpSel_Real_gfx10<0x378>;
//===----------------------------------------------------------------------===//
// GFX7, GFX10.
//===----------------------------------------------------------------------===//
let AssemblerPredicate = isGFX7Only, DecoderNamespace = "GFX7" in {
multiclass VOP3_Real_gfx7<bits<10> op> {
def _gfx7 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
VOP3e_gfx6_gfx7<op{8-0}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3be_Real_gfx7<bits<10> op> {
def _gfx7 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
VOP3be_gfx6_gfx7<op{8-0}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
} // End AssemblerPredicate = isGFX7Only, DecoderNamespace = "GFX7"
multiclass VOP3_Real_gfx7_gfx10<bits<10> op> :
VOP3_Real_gfx7<op>, VOP3_Real_gfx10<op>;
multiclass VOP3be_Real_gfx7_gfx10<bits<10> op> :
VOP3be_Real_gfx7<op>, VOP3be_Real_gfx10<op>;
defm V_QSAD_PK_U16_U8 : VOP3_Real_gfx7_gfx10<0x172>;
defm V_MQSAD_U32_U8 : VOP3_Real_gfx7_gfx10<0x175>;
defm V_MAD_U64_U32 : VOP3be_Real_gfx7_gfx10<0x176>;
defm V_MAD_I64_I32 : VOP3be_Real_gfx7_gfx10<0x177>;
//===----------------------------------------------------------------------===//
// GFX6, GFX7, GFX10.
//===----------------------------------------------------------------------===//
let AssemblerPredicate = isGFX6GFX7, DecoderNamespace = "GFX6GFX7" in {
multiclass VOP3_Real_gfx6_gfx7<bits<10> op> {
def _gfx6_gfx7 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
VOP3e_gfx6_gfx7<op{8-0}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3be_Real_gfx6_gfx7<bits<10> op> {
def _gfx6_gfx7 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
VOP3be_gfx6_gfx7<op{8-0}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
} // End AssemblerPredicate = isGFX6GFX7, DecoderNamespace = "GFX6GFX7"
multiclass VOP3_Real_gfx6_gfx7_gfx10<bits<10> op> :
VOP3_Real_gfx6_gfx7<op>, VOP3_Real_gfx10<op>;
multiclass VOP3be_Real_gfx6_gfx7_gfx10<bits<10> op> :
VOP3be_Real_gfx6_gfx7<op>, VOP3be_Real_gfx10<op>;
defm V_LSHL_B64 : VOP3_Real_gfx6_gfx7<0x161>;
defm V_LSHR_B64 : VOP3_Real_gfx6_gfx7<0x162>;
defm V_ASHR_I64 : VOP3_Real_gfx6_gfx7<0x163>;
defm V_MUL_LO_I32 : VOP3_Real_gfx6_gfx7<0x16b>;
defm V_MAD_LEGACY_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x140>;
defm V_MAD_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x141>;
defm V_MAD_I32_I24 : VOP3_Real_gfx6_gfx7_gfx10<0x142>;
defm V_MAD_U32_U24 : VOP3_Real_gfx6_gfx7_gfx10<0x143>;
defm V_CUBEID_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x144>;
defm V_CUBESC_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x145>;
defm V_CUBETC_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x146>;
defm V_CUBEMA_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x147>;
defm V_BFE_U32 : VOP3_Real_gfx6_gfx7_gfx10<0x148>;
defm V_BFE_I32 : VOP3_Real_gfx6_gfx7_gfx10<0x149>;
defm V_BFI_B32 : VOP3_Real_gfx6_gfx7_gfx10<0x14a>;
defm V_FMA_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x14b>;
defm V_FMA_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x14c>;
defm V_LERP_U8 : VOP3_Real_gfx6_gfx7_gfx10<0x14d>;
defm V_ALIGNBIT_B32 : VOP3_Real_gfx6_gfx7_gfx10<0x14e>;
defm V_ALIGNBYTE_B32 : VOP3_Real_gfx6_gfx7_gfx10<0x14f>;
defm V_MULLIT_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x150>;
defm V_MIN3_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x151>;
defm V_MIN3_I32 : VOP3_Real_gfx6_gfx7_gfx10<0x152>;
defm V_MIN3_U32 : VOP3_Real_gfx6_gfx7_gfx10<0x153>;
defm V_MAX3_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x154>;
defm V_MAX3_I32 : VOP3_Real_gfx6_gfx7_gfx10<0x155>;
defm V_MAX3_U32 : VOP3_Real_gfx6_gfx7_gfx10<0x156>;
defm V_MED3_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x157>;
defm V_MED3_I32 : VOP3_Real_gfx6_gfx7_gfx10<0x158>;
defm V_MED3_U32 : VOP3_Real_gfx6_gfx7_gfx10<0x159>;
defm V_SAD_U8 : VOP3_Real_gfx6_gfx7_gfx10<0x15a>;
defm V_SAD_HI_U8 : VOP3_Real_gfx6_gfx7_gfx10<0x15b>;
defm V_SAD_U16 : VOP3_Real_gfx6_gfx7_gfx10<0x15c>;
defm V_SAD_U32 : VOP3_Real_gfx6_gfx7_gfx10<0x15d>;
defm V_CVT_PK_U8_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x15e>;
defm V_DIV_FIXUP_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x15f>;
defm V_DIV_FIXUP_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x160>;
defm V_ADD_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x164>;
defm V_MUL_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x165>;
defm V_MIN_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x166>;
defm V_MAX_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x167>;
defm V_LDEXP_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x168>;
defm V_MUL_LO_U32 : VOP3_Real_gfx6_gfx7_gfx10<0x169>;
defm V_MUL_HI_U32 : VOP3_Real_gfx6_gfx7_gfx10<0x16a>;
defm V_MUL_HI_I32 : VOP3_Real_gfx6_gfx7_gfx10<0x16c>;
defm V_DIV_FMAS_F32 : VOP3_Real_gfx6_gfx7_gfx10<0x16f>;
defm V_DIV_FMAS_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x170>;
defm V_MSAD_U8 : VOP3_Real_gfx6_gfx7_gfx10<0x171>;
defm V_MQSAD_PK_U16_U8 : VOP3_Real_gfx6_gfx7_gfx10<0x173>;
defm V_TRIG_PREOP_F64 : VOP3_Real_gfx6_gfx7_gfx10<0x174>;
defm V_DIV_SCALE_F32 : VOP3be_Real_gfx6_gfx7_gfx10<0x16d>;
defm V_DIV_SCALE_F64 : VOP3be_Real_gfx6_gfx7_gfx10<0x16e>;
// NB: Same opcode as v_mad_legacy_f32
let DecoderNamespace = "GFX10_B" in
defm V_FMA_LEGACY_F32 : VOP3_Real_gfx10<0x140>;
//===----------------------------------------------------------------------===//
// GFX8, GFX9 (VI).
//===----------------------------------------------------------------------===//
let AssemblerPredicate = isGFX8GFX9, DecoderNamespace = "GFX8" in {
multiclass VOP3_Real_vi<bits<10> op> {
def _vi : VOP3_Real<!cast<VOP_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>,
VOP3e_vi <op, !cast<VOP_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3_Real_No_Suffix_vi<bits<10> op> {
def _vi : VOP3_Real<!cast<VOP_Pseudo>(NAME), SIEncodingFamily.VI>,
VOP3e_vi <op, !cast<VOP_Pseudo>(NAME).Pfl>;
}
multiclass VOP3be_Real_vi<bits<10> op> {
def _vi : VOP3_Real<!cast<VOP_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>,
VOP3be_vi <op, !cast<VOP_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3OpSel_Real_gfx9<bits<10> op> {
def _vi : VOP3_Real<!cast<VOP_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>,
VOP3OpSel_gfx9 <op, !cast<VOP_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3OpSel_Real_gfx9_forced_opsel2<bits<10> op> {
def _vi : VOP3_Real<!cast<VOP_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>,
VOP3OpSel_gfx9 <op, !cast<VOP_Pseudo>(NAME#"_e64").Pfl> {
let Inst{13} = src2_modifiers{2}; // op_sel(2)
}
}
multiclass VOP3Interp_Real_vi<bits<10> op> {
def _vi : VOP3_Real<!cast<VOP_Pseudo>(NAME), SIEncodingFamily.VI>,
VOP3Interp_vi <op, !cast<VOP_Pseudo>(NAME).Pfl>;
}
} // End AssemblerPredicate = isGFX8GFX9, DecoderNamespace = "GFX8"
let AssemblerPredicate = isGFX8Only, DecoderNamespace = "GFX8" in {
multiclass VOP3_F16_Real_vi<bits<10> op> {
def _vi : VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>,
VOP3e_vi <op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP3Interp_F16_Real_vi<bits<10> op> {
def _vi : VOP3_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.VI>,
VOP3Interp_vi <op, !cast<VOP3_Pseudo>(NAME).Pfl>;
}
} // End AssemblerPredicate = isGFX8Only, DecoderNamespace = "GFX8"
let AssemblerPredicate = isGFX9Only, DecoderNamespace = "GFX9" in {
multiclass VOP3_F16_Real_gfx9<bits<10> op, string OpName, string AsmName> {
def _gfx9 : VOP3_Real<!cast<VOP3_Pseudo>(OpName#"_e64"), SIEncodingFamily.GFX9>,
VOP3e_vi <op, !cast<VOP3_Pseudo>(OpName#"_e64").Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(OpName#"_e64");
let AsmString = AsmName # ps.AsmOperands;
}
}
multiclass VOP3OpSel_F16_Real_gfx9<bits<10> op, string AsmName> {
def _gfx9 : VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX9>,
VOP3OpSel_gfx9 <op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(NAME#"_e64");
let AsmString = AsmName # ps.AsmOperands;
}
}
multiclass VOP3Interp_F16_Real_gfx9<bits<10> op, string OpName, string AsmName> {
def _gfx9 : VOP3_Real<!cast<VOP3_Pseudo>(OpName), SIEncodingFamily.GFX9>,
VOP3Interp_vi <op, !cast<VOP3_Pseudo>(OpName).Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(OpName);
let AsmString = AsmName # ps.AsmOperands;
}
}
multiclass VOP3_Real_gfx9<bits<10> op, string AsmName> {
def _gfx9 : VOP3_Real<!cast<VOP_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX9>,
VOP3e_vi <op, !cast<VOP_Pseudo>(NAME#"_e64").Pfl> {
VOP_Pseudo ps = !cast<VOP_Pseudo>(NAME#"_e64");
let AsmString = AsmName # ps.AsmOperands;
}
}
} // End AssemblerPredicate = isGFX9Only, DecoderNamespace = "GFX9"
defm V_MAD_U64_U32 : VOP3be_Real_vi <0x1E8>;
defm V_MAD_I64_I32 : VOP3be_Real_vi <0x1E9>;
defm V_MAD_LEGACY_F32 : VOP3_Real_vi <0x1c0>;
defm V_MAD_F32 : VOP3_Real_vi <0x1c1>;
defm V_MAD_I32_I24 : VOP3_Real_vi <0x1c2>;
defm V_MAD_U32_U24 : VOP3_Real_vi <0x1c3>;
defm V_CUBEID_F32 : VOP3_Real_vi <0x1c4>;
defm V_CUBESC_F32 : VOP3_Real_vi <0x1c5>;
defm V_CUBETC_F32 : VOP3_Real_vi <0x1c6>;
defm V_CUBEMA_F32 : VOP3_Real_vi <0x1c7>;
defm V_BFE_U32 : VOP3_Real_vi <0x1c8>;
defm V_BFE_I32 : VOP3_Real_vi <0x1c9>;
defm V_BFI_B32 : VOP3_Real_vi <0x1ca>;
defm V_FMA_F32 : VOP3_Real_vi <0x1cb>;
defm V_FMA_F64 : VOP3_Real_vi <0x1cc>;
defm V_LERP_U8 : VOP3_Real_vi <0x1cd>;
defm V_ALIGNBIT_B32 : VOP3_Real_vi <0x1ce>;
defm V_ALIGNBYTE_B32 : VOP3_Real_vi <0x1cf>;
defm V_MIN3_F32 : VOP3_Real_vi <0x1d0>;
defm V_MIN3_I32 : VOP3_Real_vi <0x1d1>;
defm V_MIN3_U32 : VOP3_Real_vi <0x1d2>;
defm V_MAX3_F32 : VOP3_Real_vi <0x1d3>;
defm V_MAX3_I32 : VOP3_Real_vi <0x1d4>;
defm V_MAX3_U32 : VOP3_Real_vi <0x1d5>;
defm V_MED3_F32 : VOP3_Real_vi <0x1d6>;
defm V_MED3_I32 : VOP3_Real_vi <0x1d7>;
defm V_MED3_U32 : VOP3_Real_vi <0x1d8>;
defm V_SAD_U8 : VOP3_Real_vi <0x1d9>;
defm V_SAD_HI_U8 : VOP3_Real_vi <0x1da>;
defm V_SAD_U16 : VOP3_Real_vi <0x1db>;
defm V_SAD_U32 : VOP3_Real_vi <0x1dc>;
defm V_CVT_PK_U8_F32 : VOP3_Real_vi <0x1dd>;
defm V_DIV_FIXUP_F32 : VOP3_Real_vi <0x1de>;
defm V_DIV_FIXUP_F64 : VOP3_Real_vi <0x1df>;
defm V_DIV_SCALE_F32 : VOP3be_Real_vi <0x1e0>;
defm V_DIV_SCALE_F64 : VOP3be_Real_vi <0x1e1>;
defm V_DIV_FMAS_F32 : VOP3_Real_vi <0x1e2>;
defm V_DIV_FMAS_F64 : VOP3_Real_vi <0x1e3>;
defm V_MSAD_U8 : VOP3_Real_vi <0x1e4>;
defm V_QSAD_PK_U16_U8 : VOP3_Real_vi <0x1e5>;
defm V_MQSAD_PK_U16_U8 : VOP3_Real_vi <0x1e6>;
defm V_MQSAD_U32_U8 : VOP3_Real_vi <0x1e7>;
defm V_PERM_B32 : VOP3_Real_vi <0x1ed>;
defm V_MAD_F16 : VOP3_F16_Real_vi <0x1ea>;
defm V_MAD_U16 : VOP3_F16_Real_vi <0x1eb>;
defm V_MAD_I16 : VOP3_F16_Real_vi <0x1ec>;
defm V_FMA_F16 : VOP3_F16_Real_vi <0x1ee>;
defm V_DIV_FIXUP_F16 : VOP3_F16_Real_vi <0x1ef>;
defm V_INTERP_P2_F16 : VOP3Interp_F16_Real_vi <0x276>;
let FPDPRounding = 1 in {
defm V_MAD_LEGACY_F16 : VOP3_F16_Real_gfx9 <0x1ea, "V_MAD_F16", "v_mad_legacy_f16">;
defm V_FMA_LEGACY_F16 : VOP3_F16_Real_gfx9 <0x1ee, "V_FMA_F16", "v_fma_legacy_f16">;
defm V_DIV_FIXUP_LEGACY_F16 : VOP3_F16_Real_gfx9 <0x1ef, "V_DIV_FIXUP_F16", "v_div_fixup_legacy_f16">;
defm V_INTERP_P2_LEGACY_F16 : VOP3Interp_F16_Real_gfx9 <0x276, "V_INTERP_P2_F16", "v_interp_p2_legacy_f16">;
} // End FPDPRounding = 1
defm V_MAD_LEGACY_U16 : VOP3_F16_Real_gfx9 <0x1eb, "V_MAD_U16", "v_mad_legacy_u16">;
defm V_MAD_LEGACY_I16 : VOP3_F16_Real_gfx9 <0x1ec, "V_MAD_I16", "v_mad_legacy_i16">;
defm V_MAD_F16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x203, "v_mad_f16">;
defm V_MAD_U16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x204, "v_mad_u16">;
defm V_MAD_I16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x205, "v_mad_i16">;
defm V_FMA_F16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x206, "v_fma_f16">;
defm V_DIV_FIXUP_F16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x207, "v_div_fixup_f16">;
defm V_INTERP_P2_F16_gfx9 : VOP3Interp_F16_Real_gfx9 <0x277, "V_INTERP_P2_F16_gfx9", "v_interp_p2_f16">;
defm V_ADD_I32 : VOP3_Real_vi <0x29c>;
defm V_SUB_I32 : VOP3_Real_vi <0x29d>;
defm V_INTERP_P1_F32_e64 : VOP3Interp_Real_vi <0x270>;
defm V_INTERP_P2_F32_e64 : VOP3Interp_Real_vi <0x271>;
defm V_INTERP_MOV_F32_e64 : VOP3Interp_Real_vi <0x272>;
defm V_INTERP_P1LL_F16 : VOP3Interp_Real_vi <0x274>;
defm V_INTERP_P1LV_F16 : VOP3Interp_Real_vi <0x275>;
defm V_ADD_F64 : VOP3_Real_vi <0x280>;
defm V_MUL_F64 : VOP3_Real_vi <0x281>;
defm V_MIN_F64 : VOP3_Real_vi <0x282>;
defm V_MAX_F64 : VOP3_Real_vi <0x283>;
defm V_LDEXP_F64 : VOP3_Real_vi <0x284>;
defm V_MUL_LO_U32 : VOP3_Real_vi <0x285>;
// removed from VI as identical to V_MUL_LO_U32
let isAsmParserOnly = 1 in {
defm V_MUL_LO_I32 : VOP3_Real_vi <0x285>;
}
defm V_MUL_HI_U32 : VOP3_Real_vi <0x286>;
defm V_MUL_HI_I32 : VOP3_Real_vi <0x287>;
defm V_READLANE_B32 : VOP3_Real_No_Suffix_vi <0x289>;
defm V_WRITELANE_B32 : VOP3_Real_No_Suffix_vi <0x28a>;
defm V_LSHLREV_B64 : VOP3_Real_vi <0x28f>;
defm V_LSHRREV_B64 : VOP3_Real_vi <0x290>;
defm V_ASHRREV_I64 : VOP3_Real_vi <0x291>;
defm V_TRIG_PREOP_F64 : VOP3_Real_vi <0x292>;
defm V_LSHL_ADD_U32 : VOP3_Real_vi <0x1fd>;
defm V_ADD_LSHL_U32 : VOP3_Real_vi <0x1fe>;
defm V_ADD3_U32 : VOP3_Real_vi <0x1ff>;
defm V_LSHL_OR_B32 : VOP3_Real_vi <0x200>;
defm V_AND_OR_B32 : VOP3_Real_vi <0x201>;
defm V_OR3_B32 : VOP3_Real_vi <0x202>;
defm V_PACK_B32_F16 : VOP3OpSel_Real_gfx9 <0x2a0>;
defm V_XAD_U32 : VOP3_Real_vi <0x1f3>;
defm V_MIN3_F16 : VOP3OpSel_Real_gfx9 <0x1f4>;
defm V_MIN3_I16 : VOP3OpSel_Real_gfx9 <0x1f5>;
defm V_MIN3_U16 : VOP3OpSel_Real_gfx9 <0x1f6>;
defm V_MAX3_F16 : VOP3OpSel_Real_gfx9 <0x1f7>;
defm V_MAX3_I16 : VOP3OpSel_Real_gfx9 <0x1f8>;
defm V_MAX3_U16 : VOP3OpSel_Real_gfx9 <0x1f9>;
defm V_MED3_F16 : VOP3OpSel_Real_gfx9 <0x1fa>;
defm V_MED3_I16 : VOP3OpSel_Real_gfx9 <0x1fb>;
defm V_MED3_U16 : VOP3OpSel_Real_gfx9 <0x1fc>;
defm V_ADD_I16 : VOP3OpSel_Real_gfx9 <0x29e>;
defm V_SUB_I16 : VOP3OpSel_Real_gfx9 <0x29f>;
defm V_MAD_U32_U16 : VOP3OpSel_Real_gfx9 <0x1f1>;
defm V_MAD_I32_I16 : VOP3OpSel_Real_gfx9 <0x1f2>;
defm V_CVT_PKNORM_I16_F16 : VOP3OpSel_Real_gfx9 <0x299>;
defm V_CVT_PKNORM_U16_F16 : VOP3OpSel_Real_gfx9 <0x29a>;
defm V_LSHL_ADD_U64 : VOP3_Real_vi <0x208>;
defm V_CVT_PK_FP8_F32 : VOP3OpSel_Real_gfx9 <0x2a2>;
defm V_CVT_PK_BF8_F32 : VOP3OpSel_Real_gfx9 <0x2a3>;
defm V_CVT_SR_FP8_F32 : VOP3OpSel_Real_gfx9_forced_opsel2 <0x2a4>;
defm V_CVT_SR_BF8_F32 : VOP3OpSel_Real_gfx9_forced_opsel2 <0x2a5>;
Codebase Browser
llvm