llvm-project/llvm/lib/Target/AMDGPU/VOP2Instructions.td

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1589 lines
64 KiB
TableGen
Raw Normal View History

//===-- VOP2Instructions.td - Vector Instruction Defintions ---------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// VOP2 Classes
//===----------------------------------------------------------------------===//
class VOP2e <bits<6> op, VOPProfile P> : Enc32 {
bits<8> vdst;
bits<9> src0;
bits<8> src1;
let Inst{8-0} = !if(P.HasSrc0, src0, 0);
let Inst{16-9} = !if(P.HasSrc1, src1, 0);
let Inst{24-17} = !if(P.EmitDst, vdst, 0);
let Inst{30-25} = op;
let Inst{31} = 0x0; //encoding
}
class VOP2_MADKe <bits<6> op, VOPProfile P> : Enc64 {
bits<8> vdst;
bits<9> src0;
bits<8> src1;
bits<32> imm;
let Inst{8-0} = !if(P.HasSrc0, src0, 0);
let Inst{16-9} = !if(P.HasSrc1, src1, 0);
let Inst{24-17} = !if(P.EmitDst, vdst, 0);
let Inst{30-25} = op;
let Inst{31} = 0x0; // encoding
let Inst{63-32} = imm;
}
class VOP2_SDWAe <bits<6> op, VOPProfile P> : VOP_SDWAe <P> {
bits<8> vdst;
bits<8> src1;
let Inst{8-0} = 0xf9; // sdwa
let Inst{16-9} = !if(P.HasSrc1, src1{7-0}, 0);
let Inst{24-17} = !if(P.EmitDst, vdst{7-0}, 0);
let Inst{30-25} = op;
let Inst{31} = 0x0; // encoding
}
class VOP2_SDWA9Ae <bits<6> op, VOPProfile P> : VOP_SDWA9Ae <P> {
bits<8> vdst;
bits<9> src1;
let Inst{8-0} = 0xf9; // sdwa
let Inst{16-9} = !if(P.HasSrc1, src1{7-0}, 0);
let Inst{24-17} = !if(P.EmitDst, vdst{7-0}, 0);
let Inst{30-25} = op;
let Inst{31} = 0x0; // encoding
let Inst{63} = !if(P.HasSrc1, src1{8}, 0); // src1_sgpr
}
class VOP2_Pseudo <string opName, VOPProfile P, list<dag> pattern=[], string suffix = "_e32"> :
VOP_Pseudo <opName, suffix, P, P.Outs32, P.Ins32, "", pattern> {
let AsmOperands = P.Asm32;
let Size = 4;
let mayLoad = 0;
let mayStore = 0;
let hasSideEffects = 0;
let VOP2 = 1;
let VALU = 1;
let Uses = [EXEC];
let AsmVariantName = AMDGPUAsmVariants.Default;
}
class VOP2_Real <VOP2_Pseudo ps, int EncodingFamily> :
InstSI <ps.OutOperandList, ps.InOperandList, ps.Mnemonic # ps.AsmOperands, []>,
SIMCInstr <ps.PseudoInstr, EncodingFamily> {
let isPseudo = 0;
let isCodeGenOnly = 0;
let Constraints = ps.Constraints;
let DisableEncoding = ps.DisableEncoding;
// copy relevant pseudo op flags
let SubtargetPredicate = ps.SubtargetPredicate;
let AsmMatchConverter = ps.AsmMatchConverter;
let AsmVariantName = ps.AsmVariantName;
let Constraints = ps.Constraints;
let DisableEncoding = ps.DisableEncoding;
let TSFlags = ps.TSFlags;
let UseNamedOperandTable = ps.UseNamedOperandTable;
let Uses = ps.Uses;
let Defs = ps.Defs;
}
class VOP2_SDWA_Pseudo <string OpName, VOPProfile P, list<dag> pattern=[]> :
VOP_SDWA_Pseudo <OpName, P, pattern> {
let AsmMatchConverter = "cvtSdwaVOP2";
}
class VOP2_DPP_Pseudo <string OpName, VOPProfile P, list<dag> pattern=[]> :
VOP_DPP_Pseudo <OpName, P, pattern> {
}
class getVOP2Pat64 <SDPatternOperator node, VOPProfile P> : LetDummies {
list<dag> ret = !if(P.HasModifiers,
[(set P.DstVT:$vdst,
(node (P.Src0VT
!if(P.HasOMod,
(VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod),
(VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp))),
(P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers))))],
[(set P.DstVT:$vdst, (node P.Src0VT:$src0, P.Src1VT:$src1))]);
}
multiclass VOP2Inst_e32<string opName,
VOPProfile P,
SDPatternOperator node = null_frag,
string revOp = opName,
bit GFX9Renamed = 0> {
let renamedInGFX9 = GFX9Renamed in {
def _e32 : VOP2_Pseudo <opName, P, VOPPatOrNull<node,P>.ret>,
Commutable_REV<revOp#"_e32", !eq(revOp, opName)>;
} // End renamedInGFX9 = GFX9Renamed
}
multiclass VOP2Inst_e64<string opName,
VOPProfile P,
SDPatternOperator node = null_frag,
string revOp = opName,
bit GFX9Renamed = 0> {
let renamedInGFX9 = GFX9Renamed in {
def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<node, P>.ret>,
Commutable_REV<revOp#"_e64", !eq(revOp, opName)>;
} // End renamedInGFX9 = GFX9Renamed
}
multiclass VOP2Inst_sdwa<string opName,
VOPProfile P,
SDPatternOperator node = null_frag,
string revOp = opName,
bit GFX9Renamed = 0> {
let renamedInGFX9 = GFX9Renamed in {
foreach _ = BoolToList<P.HasExtSDWA>.ret in
def _sdwa : VOP2_SDWA_Pseudo <opName, P>;
} // End renamedInGFX9 = GFX9Renamed
}
multiclass VOP2Inst<string opName,
VOPProfile P,
SDPatternOperator node = null_frag,
string revOp = opName,
bit GFX9Renamed = 0> :
VOP2Inst_e32<opName, P, node, revOp, GFX9Renamed>,
VOP2Inst_e64<opName, P, node, revOp, GFX9Renamed>,
VOP2Inst_sdwa<opName, P, node, revOp, GFX9Renamed> {
let renamedInGFX9 = GFX9Renamed in {
foreach _ = BoolToList<P.HasExtDPP>.ret in
def _dpp : VOP2_DPP_Pseudo <opName, P>;
}
}
multiclass VOP2bInst <string opName,
VOPProfile P,
SDPatternOperator node = null_frag,
string revOp = opName,
bit GFX9Renamed = 0,
bit useSGPRInput = !eq(P.NumSrcArgs, 3)> {
let renamedInGFX9 = GFX9Renamed in {
let SchedRW = [Write32Bit, WriteSALU] in {
let Uses = !if(useSGPRInput, [VCC, EXEC], [EXEC]), Defs = [VCC] in {
def _e32 : VOP2_Pseudo <opName, P, VOPPatOrNull<node,P>.ret>,
Commutable_REV<revOp#"_e32", !eq(revOp, opName)> {
let usesCustomInserter = !eq(P.NumSrcArgs, 2);
}
foreach _ = BoolToList<P.HasExtSDWA>.ret in
def _sdwa : VOP2_SDWA_Pseudo <opName, P> {
let AsmMatchConverter = "cvtSdwaVOP2b";
}
foreach _ = BoolToList<P.HasExtDPP>.ret in
def _dpp : VOP2_DPP_Pseudo <opName, P>;
}
def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<node, P>.ret>,
Commutable_REV<revOp#"_e64", !eq(revOp, opName)>;
}
}
}
class VOP2bInstAlias <VOP2_Pseudo ps, Instruction inst,
string OpName, string opnd> :
InstAlias <OpName#" "#!subst("vcc", opnd, ps.Pfl.Asm32),
(inst ps.Pfl.DstRC:$vdst, ps.Pfl.Src0RC32:$src0,
ps.Pfl.Src1RC32:$src1)>,
PredicateControl {
}
multiclass VOP2bInstAliases<VOP2_Pseudo ps, VOP2_Real inst, string OpName> {
let WaveSizePredicate = isWave32 in {
def : VOP2bInstAlias<ps, inst, OpName, "vcc_lo">;
}
let WaveSizePredicate = isWave64 in {
def : VOP2bInstAlias<ps, inst, OpName, "vcc">;
}
}
multiclass VOP2eInst <string opName,
VOPProfile P,
SDPatternOperator node = null_frag,
string revOp = opName,
bit useSGPRInput = !eq(P.NumSrcArgs, 3)> {
let SchedRW = [Write32Bit] in {
let Uses = !if(useSGPRInput, [VCC, EXEC], [EXEC]) in {
def _e32 : VOP2_Pseudo <opName, P>,
Commutable_REV<revOp#"_e32", !eq(revOp, opName)>;
foreach _ = BoolToList<P.HasExtSDWA>.ret in
def _sdwa : VOP2_SDWA_Pseudo <opName, P> {
let AsmMatchConverter = "cvtSdwaVOP2e";
}
foreach _ = BoolToList<P.HasExtDPP>.ret in
def _dpp : VOP2_DPP_Pseudo <opName, P>;
}
def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<node, P>.ret>,
Commutable_REV<revOp#"_e64", !eq(revOp, opName)>;
}
}
class VOP2eInstAlias <VOP2_Pseudo ps, Instruction inst, string opnd> :
InstAlias <ps.OpName#" "#ps.Pfl.Asm32#", "#opnd,
(inst ps.Pfl.DstRC:$vdst, ps.Pfl.Src0RC32:$src0,
ps.Pfl.Src1RC32:$src1)>,
PredicateControl {
}
multiclass VOP2eInstAliases<VOP2_Pseudo ps, VOP2_Real inst> {
let WaveSizePredicate = isWave32 in {
def : VOP2eInstAlias<ps, inst, "vcc_lo">;
}
let WaveSizePredicate = isWave64 in {
def : VOP2eInstAlias<ps, inst, "vcc">;
}
}
class VOP_MADAK <ValueType vt> : VOPProfile <[vt, vt, vt, vt]> {
field Operand ImmOpType = !if(!eq(vt.Size, 32), f32kimm, f16kimm);
field dag Ins32 = !if(!eq(vt.Size, 32),
(ins VCSrc_f32:$src0, VGPR_32:$src1, ImmOpType:$imm),
(ins VCSrc_f16:$src0, VGPR_32:$src1, ImmOpType:$imm));
field bit HasExt = 0;
// Hack to stop printing _e64
let DstRC = RegisterOperand<VGPR_32>;
field string Asm32 = " $vdst, $src0, $src1, $imm";
}
def VOP_MADAK_F16 : VOP_MADAK <f16>;
def VOP_MADAK_F32 : VOP_MADAK <f32>;
class VOP_MADMK <ValueType vt> : VOPProfile <[vt, vt, vt, vt]> {
field Operand ImmOpType = !if(!eq(vt.Size, 32), f32kimm, f16kimm);
field dag Ins32 = (ins VCSrc_f32:$src0, ImmOpType:$imm, VGPR_32:$src1);
field bit HasExt = 0;
// Hack to stop printing _e64
let DstRC = RegisterOperand<VGPR_32>;
field string Asm32 = " $vdst, $src0, $imm, $src1";
}
def VOP_MADMK_F16 : VOP_MADMK <f16>;
def VOP_MADMK_F32 : VOP_MADMK <f32>;
// FIXME: Remove src2_modifiers. It isn't used, so is wasting memory
// and processing time but it makes it easier to convert to mad.
class VOP_MAC <ValueType vt0, ValueType vt1=vt0> : VOPProfile <[vt0, vt1, vt1, vt0]> {
let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1, VGPR_32:$src2);
let Ins64 = getIns64<Src0RC64, Src1RC64, RegisterOperand<VGPR_32>, 3,
0, HasModifiers, HasModifiers, HasOMod,
Src0Mod, Src1Mod, Src2Mod>.ret;
let InsDPP = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0,
Src1ModDPP:$src1_modifiers, Src1DPP:$src1,
VGPR_32:$src2, // stub argument
dpp_ctrl:$dpp_ctrl, row_mask:$row_mask,
bank_mask:$bank_mask, bound_ctrl:$bound_ctrl);
let InsDPP16 = !con(InsDPP, (ins FI:$fi));
let InsDPP8 = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0,
Src1ModDPP:$src1_modifiers, Src1DPP:$src1,
VGPR_32:$src2, // stub argument
dpp8:$dpp8, FI:$fi);
let InsSDWA = (ins Src0ModSDWA:$src0_modifiers, Src0SDWA:$src0,
Src1ModSDWA:$src1_modifiers, Src1SDWA:$src1,
VGPR_32:$src2, // stub argument
clampmod:$clamp, omod:$omod,
dst_sel:$dst_sel, dst_unused:$dst_unused,
src0_sel:$src0_sel, src1_sel:$src1_sel);
let Asm32 = getAsm32<1, 2, vt0>.ret;
let Asm64 = getAsm64<1, 2, 0, HasModifiers, HasOMod, vt0>.ret;
let AsmDPP = getAsmDPP<1, 2, HasModifiers, vt0>.ret;
let AsmDPP16 = getAsmDPP16<1, 2, HasModifiers, vt0>.ret;
let AsmDPP8 = getAsmDPP8<1, 2, 0, vt0>.ret;
let AsmSDWA = getAsmSDWA<1, 2, vt0>.ret;
let AsmSDWA9 = getAsmSDWA9<1, 1, 2, vt0>.ret;
let HasSrc2 = 0;
let HasSrc2Mods = 0;
let HasExt = 1;
let HasExtDPP = 1;
let HasExtSDWA = 1;
let HasExtSDWA9 = 0;
let TieRegDPP = "$src2";
}
def VOP_MAC_F16 : VOP_MAC <f16>;
def VOP_MAC_F32 : VOP_MAC <f32>;
class VOP_DOT_ACC<ValueType vt0, ValueType vt1> : VOP_MAC<vt0, vt1> {
let HasClamp = 0;
let HasExtSDWA = 0;
let HasModifiers = 1;
let HasOpSel = 0;
let IsPacked = 0;
}
def VOP_DOT_ACC_F32_V2F16 : VOP_DOT_ACC<f32, v2f16> {
let Src0ModDPP = FPVRegInputMods;
let Src1ModDPP = FPVRegInputMods;
}
def VOP_DOT_ACC_I32_I32 : VOP_DOT_ACC<i32, i32>;
// Write out to vcc or arbitrary SGPR.
def VOP2b_I32_I1_I32_I32 : VOPProfile<[i32, i32, i32, untyped], 0, /*EnableClamp=*/1> {
let Asm32 = "$vdst, vcc, $src0, $src1";
let Asm64 = "$vdst, $sdst, $src0, $src1$clamp";
let AsmSDWA = "$vdst, vcc, $src0_modifiers, $src1_modifiers$clamp $dst_sel $dst_unused $src0_sel $src1_sel";
let AsmSDWA9 = "$vdst, vcc, $src0_modifiers, $src1_modifiers$clamp $dst_sel $dst_unused $src0_sel $src1_sel";
let AsmDPP = "$vdst, vcc, $src0, $src1 $dpp_ctrl$row_mask$bank_mask$bound_ctrl";
let AsmDPP8 = "$vdst, vcc, $src0, $src1 $dpp8$fi";
let AsmDPP16 = AsmDPP#"$fi";
let Outs32 = (outs DstRC:$vdst);
let Outs64 = (outs DstRC:$vdst, VOPDstS64orS32:$sdst);
}
// Write out to vcc or arbitrary SGPR and read in from vcc or
// arbitrary SGPR.
def VOP2b_I32_I1_I32_I32_I1 : VOPProfile<[i32, i32, i32, i1], 0, /*EnableClamp=*/1> {
let Asm32 = "$vdst, vcc, $src0, $src1, vcc";
let Asm64 = "$vdst, $sdst, $src0, $src1, $src2$clamp";
let AsmSDWA = "$vdst, vcc, $src0_modifiers, $src1_modifiers, vcc $clamp $dst_sel $dst_unused $src0_sel $src1_sel";
let AsmSDWA9 = "$vdst, vcc, $src0_modifiers, $src1_modifiers, vcc $clamp $dst_sel $dst_unused $src0_sel $src1_sel";
let AsmDPP = "$vdst, vcc, $src0, $src1, vcc $dpp_ctrl$row_mask$bank_mask$bound_ctrl";
let AsmDPP8 = "$vdst, vcc, $src0, $src1, vcc $dpp8$fi";
let AsmDPP16 = AsmDPP#"$fi";
let Outs32 = (outs DstRC:$vdst);
let Outs64 = (outs DstRC:$vdst, VOPDstS64orS32:$sdst);
// Suppress src2 implied by type since the 32-bit encoding uses an
// implicit VCC use.
let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1);
let InsSDWA = (ins Src0ModSDWA:$src0_modifiers, Src0SDWA:$src0,
Src1ModSDWA:$src1_modifiers, Src1SDWA:$src1,
clampmod:$clamp,
dst_sel:$dst_sel, dst_unused:$dst_unused,
src0_sel:$src0_sel, src1_sel:$src1_sel);
[AMDGPU] Add pseudo "old" source to all DPP instructions Summary: All instructions with the DPP modifier may not write to certain lanes of the output if bound_ctrl=1 is set or any bits in bank_mask or row_mask aren't set, so the destination register may be both defined and modified. The right way to handle this is to add a constraint that the destination register is the same as one of the inputs. We could tie the destination to the first source, but that would be too restrictive for some use-cases where we want the destination to be some other value before the instruction executes. Instead, add a fake "old" source and tie it to the destination. Effectively, the "old" source defines what value unwritten lanes will get. We'll expose this functionality to users with a new intrinsic later. Also, we want to use DPP instructions for computing derivatives, which means we need to set WQM for them. We also need to enable the entire wavefront when using DPP intrinsics to implement nonuniform subgroup reductions, since otherwise we'll get incorrect results in some cases. To accomodate this, add a new operand to all DPP instructions which will be interpreted by the SI WQM pass. This will be exposed with a new intrinsic later. We'll also add support for Whole Wavefront Mode later. I also fixed llvm.amdgcn.mov.dpp to overwrite the source and fixed up the test. However, I could also keep the old behavior (where lanes that aren't written are undefined) if people want it. Reviewers: tstellar, arsenm Subscribers: kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye Differential Revision: https://reviews.llvm.org/D34716 llvm-svn: 310283
2017-08-08 03:10:56 +08:00
let InsDPP = (ins DstRCDPP:$old,
Src0DPP:$src0,
Src1DPP:$src1,
dpp_ctrl:$dpp_ctrl, row_mask:$row_mask,
bank_mask:$bank_mask, bound_ctrl:$bound_ctrl);
let InsDPP16 = !con(InsDPP, (ins FI:$fi));
let HasExt = 1;
let HasExtDPP = 1;
let HasExtSDWA = 1;
let HasExtSDWA9 = 1;
}
// Read in from vcc or arbitrary SGPR.
def VOP2e_I32_I32_I32_I1 : VOPProfile<[i32, i32, i32, i1], /*EnableF32SrcMods=*/1> {
let Asm32 = "$vdst, $src0, $src1";
let Asm64 = "$vdst, $src0_modifiers, $src1_modifiers, $src2";
let AsmSDWA = "$vdst, $src0_modifiers, $src1_modifiers, vcc $clamp $dst_sel $dst_unused $src0_sel $src1_sel";
let AsmSDWA9 = "$vdst, $src0_modifiers, $src1_modifiers, vcc $clamp $dst_sel $dst_unused $src0_sel $src1_sel";
let AsmDPP = "$vdst, $src0, $src1, vcc $dpp_ctrl$row_mask$bank_mask$bound_ctrl";
let AsmDPP8 = "$vdst, $src0, $src1, vcc $dpp8$fi";
let AsmDPP16 = AsmDPP#"$fi";
let Outs32 = (outs DstRC:$vdst);
let Outs64 = (outs DstRC:$vdst);
// Suppress src2 implied by type since the 32-bit encoding uses an
// implicit VCC use.
let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1);
let InsSDWA = (ins Src0ModSDWA:$src0_modifiers, Src0SDWA:$src0,
Src1ModSDWA:$src1_modifiers, Src1SDWA:$src1,
clampmod:$clamp,
dst_sel:$dst_sel, dst_unused:$dst_unused,
src0_sel:$src0_sel, src1_sel:$src1_sel);
let InsDPP = (ins DstRCDPP:$old,
Src0ModDPP:$src0_modifiers, Src0DPP:$src0,
Src1ModDPP:$src1_modifiers, Src1DPP:$src1,
dpp_ctrl:$dpp_ctrl, row_mask:$row_mask,
bank_mask:$bank_mask, bound_ctrl:$bound_ctrl);
let InsDPP16 = !con(InsDPP, (ins FI:$fi));
let HasExt = 1;
let HasExtDPP = 1;
let HasExtSDWA = 1;
let HasExtSDWA9 = 1;
}
def VOP_READLANE : VOPProfile<[i32, i32, i32]> {
let Outs32 = (outs SReg_32:$vdst);
let Outs64 = Outs32;
let Ins32 = (ins VRegOrLds_32:$src0, SCSrc_b32:$src1);
let Ins64 = Ins32;
let Asm32 = " $vdst, $src0, $src1";
let Asm64 = Asm32;
let HasExt = 0;
let HasExtDPP = 0;
let HasExtSDWA = 0;
let HasExtSDWA9 = 0;
}
def VOP_WRITELANE : VOPProfile<[i32, i32, i32, i32]> {
let Outs32 = (outs VGPR_32:$vdst);
let Outs64 = Outs32;
let Ins32 = (ins SCSrc_b32:$src0, SCSrc_b32:$src1, VGPR_32:$vdst_in);
let Ins64 = Ins32;
let Asm32 = " $vdst, $src0, $src1";
let Asm64 = Asm32;
let HasSrc2 = 0;
let HasSrc2Mods = 0;
let HasExt = 0;
let HasExtDPP = 0;
let HasExtSDWA = 0;
let HasExtSDWA9 = 0;
}
//===----------------------------------------------------------------------===//
// VOP2 Instructions
//===----------------------------------------------------------------------===//
defm V_CNDMASK_B32 : VOP2eInst <"v_cndmask_b32", VOP2e_I32_I32_I32_I1>;
def V_MADMK_F32 : VOP2_Pseudo <"v_madmk_f32", VOP_MADMK_F32, []>;
let isCommutable = 1 in {
defm V_ADD_F32 : VOP2Inst <"v_add_f32", VOP_F32_F32_F32, fadd>;
defm V_SUB_F32 : VOP2Inst <"v_sub_f32", VOP_F32_F32_F32, fsub>;
defm V_SUBREV_F32 : VOP2Inst <"v_subrev_f32", VOP_F32_F32_F32, null_frag, "v_sub_f32">;
defm V_MUL_LEGACY_F32 : VOP2Inst <"v_mul_legacy_f32", VOP_F32_F32_F32, AMDGPUfmul_legacy>;
defm V_MUL_F32 : VOP2Inst <"v_mul_f32", VOP_F32_F32_F32, fmul>;
defm V_MUL_I32_I24 : VOP2Inst <"v_mul_i32_i24", VOP_PAT_GEN<VOP_I32_I32_I32, 2>, AMDGPUmul_i24>;
defm V_MUL_HI_I32_I24 : VOP2Inst <"v_mul_hi_i32_i24", VOP_PAT_GEN<VOP_I32_I32_I32, 2>, AMDGPUmulhi_i24>;
defm V_MUL_U32_U24 : VOP2Inst <"v_mul_u32_u24", VOP_PAT_GEN<VOP_I32_I32_I32, 2>, AMDGPUmul_u24>;
defm V_MUL_HI_U32_U24 : VOP2Inst <"v_mul_hi_u32_u24", VOP_PAT_GEN<VOP_I32_I32_I32, 2>, AMDGPUmulhi_u24>;
defm V_MIN_F32 : VOP2Inst <"v_min_f32", VOP_F32_F32_F32, fminnum_like>;
defm V_MAX_F32 : VOP2Inst <"v_max_f32", VOP_F32_F32_F32, fmaxnum_like>;
defm V_MIN_I32 : VOP2Inst <"v_min_i32", VOP_PAT_GEN<VOP_I32_I32_I32>, smin>;
defm V_MAX_I32 : VOP2Inst <"v_max_i32", VOP_PAT_GEN<VOP_I32_I32_I32>, smax>;
defm V_MIN_U32 : VOP2Inst <"v_min_u32", VOP_PAT_GEN<VOP_I32_I32_I32>, umin>;
defm V_MAX_U32 : VOP2Inst <"v_max_u32", VOP_PAT_GEN<VOP_I32_I32_I32>, umax>;
defm V_LSHRREV_B32 : VOP2Inst <"v_lshrrev_b32", VOP_I32_I32_I32, lshr_rev, "v_lshr_b32">;
defm V_ASHRREV_I32 : VOP2Inst <"v_ashrrev_i32", VOP_I32_I32_I32, ashr_rev, "v_ashr_i32">;
defm V_LSHLREV_B32 : VOP2Inst <"v_lshlrev_b32", VOP_I32_I32_I32, lshl_rev, "v_lshl_b32">;
defm V_AND_B32 : VOP2Inst <"v_and_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, and>;
defm V_OR_B32 : VOP2Inst <"v_or_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, or>;
defm V_XOR_B32 : VOP2Inst <"v_xor_b32", VOP_PAT_GEN<VOP_I32_I32_I32>, xor>;
let Constraints = "$vdst = $src2", DisableEncoding="$src2",
isConvertibleToThreeAddress = 1 in {
defm V_MAC_F32 : VOP2Inst <"v_mac_f32", VOP_MAC_F32>;
}
def V_MADAK_F32 : VOP2_Pseudo <"v_madak_f32", VOP_MADAK_F32, []>;
// No patterns so that the scalar instructions are always selected.
// The scalar versions will be replaced with vector when needed later.
// V_ADD_I32, V_SUB_I32, and V_SUBREV_I32 where renamed to *_U32 in VI,
// but the VI instructions behave the same as the SI versions.
defm V_ADD_I32 : VOP2bInst <"v_add_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_add_i32", 1>;
defm V_SUB_I32 : VOP2bInst <"v_sub_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_sub_i32", 1>;
defm V_SUBREV_I32 : VOP2bInst <"v_subrev_i32", VOP2b_I32_I1_I32_I32, null_frag, "v_sub_i32", 1>;
defm V_ADDC_U32 : VOP2bInst <"v_addc_u32", VOP2b_I32_I1_I32_I32_I1, null_frag, "v_addc_u32", 1>;
defm V_SUBB_U32 : VOP2bInst <"v_subb_u32", VOP2b_I32_I1_I32_I32_I1, null_frag, "v_subb_u32", 1>;
defm V_SUBBREV_U32 : VOP2bInst <"v_subbrev_u32", VOP2b_I32_I1_I32_I32_I1, null_frag, "v_subb_u32", 1>;
let SubtargetPredicate = HasAddNoCarryInsts in {
defm V_ADD_U32 : VOP2Inst <"v_add_u32", VOP_I32_I32_I32_ARITH, null_frag, "v_add_u32", 1>;
defm V_SUB_U32 : VOP2Inst <"v_sub_u32", VOP_I32_I32_I32_ARITH, null_frag, "v_sub_u32", 1>;
defm V_SUBREV_U32 : VOP2Inst <"v_subrev_u32", VOP_I32_I32_I32_ARITH, null_frag, "v_sub_u32", 1>;
}
} // End isCommutable = 1
// These are special and do not read the exec mask.
let isConvergent = 1, Uses = []<Register> in {
def V_READLANE_B32 : VOP2_Pseudo<"v_readlane_b32", VOP_READLANE,
[(set i32:$vdst, (int_amdgcn_readlane i32:$src0, i32:$src1))]>;
let Constraints = "$vdst = $vdst_in", DisableEncoding="$vdst_in" in {
def V_WRITELANE_B32 : VOP2_Pseudo<"v_writelane_b32", VOP_WRITELANE,
[(set i32:$vdst, (int_amdgcn_writelane i32:$src0, i32:$src1, i32:$vdst_in))]>;
} // End $vdst = $vdst_in, DisableEncoding $vdst_in
} // End isConvergent = 1
defm V_BFM_B32 : VOP2Inst <"v_bfm_b32", VOP_NO_EXT<VOP_I32_I32_I32>>;
defm V_BCNT_U32_B32 : VOP2Inst <"v_bcnt_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>, add_ctpop>;
defm V_MBCNT_LO_U32_B32 : VOP2Inst <"v_mbcnt_lo_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>, int_amdgcn_mbcnt_lo>;
defm V_MBCNT_HI_U32_B32 : VOP2Inst <"v_mbcnt_hi_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>, int_amdgcn_mbcnt_hi>;
defm V_LDEXP_F32 : VOP2Inst <"v_ldexp_f32", VOP_NO_EXT<VOP_F32_F32_I32>, AMDGPUldexp>;
defm V_CVT_PKACCUM_U8_F32 : VOP2Inst <"v_cvt_pkaccum_u8_f32", VOP_NO_EXT<VOP_I32_F32_I32>>; // TODO: set "Uses = dst"
defm V_CVT_PKNORM_I16_F32 : VOP2Inst <"v_cvt_pknorm_i16_f32", VOP_NO_EXT<VOP_V2I16_F32_F32>, AMDGPUpknorm_i16_f32>;
defm V_CVT_PKNORM_U16_F32 : VOP2Inst <"v_cvt_pknorm_u16_f32", VOP_NO_EXT<VOP_V2I16_F32_F32>, AMDGPUpknorm_u16_f32>;
defm V_CVT_PKRTZ_F16_F32 : VOP2Inst <"v_cvt_pkrtz_f16_f32", VOP_NO_EXT<VOP_V2F16_F32_F32>, AMDGPUpkrtz_f16_f32>;
defm V_CVT_PK_U16_U32 : VOP2Inst <"v_cvt_pk_u16_u32", VOP_NO_EXT<VOP_V2I16_I32_I32>, AMDGPUpk_u16_u32>;
defm V_CVT_PK_I16_I32 : VOP2Inst <"v_cvt_pk_i16_i32", VOP_NO_EXT<VOP_V2I16_I32_I32>, AMDGPUpk_i16_i32>;
let SubtargetPredicate = isGFX6GFX7 in {
defm V_MIN_LEGACY_F32 : VOP2Inst <"v_min_legacy_f32", VOP_F32_F32_F32, AMDGPUfmin_legacy>;
defm V_MAX_LEGACY_F32 : VOP2Inst <"v_max_legacy_f32", VOP_F32_F32_F32, AMDGPUfmax_legacy>;
} // End SubtargetPredicate = isGFX6GFX7
let SubtargetPredicate = isGFX6GFX7GFX10 in {
let isCommutable = 1 in {
defm V_MAC_LEGACY_F32 : VOP2Inst <"v_mac_legacy_f32", VOP_F32_F32_F32>;
defm V_LSHR_B32 : VOP2Inst <"v_lshr_b32", VOP_I32_I32_I32, srl>;
defm V_ASHR_I32 : VOP2Inst <"v_ashr_i32", VOP_I32_I32_I32, sra>;
defm V_LSHL_B32 : VOP2Inst <"v_lshl_b32", VOP_I32_I32_I32, shl>;
} // End isCommutable = 1
} // End SubtargetPredicate = isGFX6GFX7GFX10
class DivergentBinOp<SDPatternOperator Op, VOP_Pseudo Inst> :
GCNPat<
(getDivergentFrag<Op>.ret Inst.Pfl.Src0VT:$src0, Inst.Pfl.Src1VT:$src1),
!if(!cast<Commutable_REV>(Inst).IsOrig,
(Inst $src0, $src1),
(Inst $src1, $src0)
)
>;
class DivergentClampingBinOp<SDPatternOperator Op, VOP_Pseudo Inst> :
GCNPat<
(getDivergentFrag<Op>.ret Inst.Pfl.Src0VT:$src0, Inst.Pfl.Src1VT:$src1),
!if(!cast<Commutable_REV>(Inst).IsOrig,
(Inst $src0, $src1, 0),
(Inst $src1, $src0, 0)
)
>;
def : DivergentBinOp<srl, V_LSHRREV_B32_e64>;
def : DivergentBinOp<sra, V_ASHRREV_I32_e64>;
def : DivergentBinOp<shl, V_LSHLREV_B32_e64>;
let SubtargetPredicate = HasAddNoCarryInsts in {
def : DivergentClampingBinOp<add, V_ADD_U32_e64>;
def : DivergentClampingBinOp<sub, V_SUB_U32_e64>;
}
let SubtargetPredicate = isGFX6GFX7GFX8GFX9, Predicates = [isGFX6GFX7GFX8GFX9] in {
def : DivergentClampingBinOp<add, V_ADD_I32_e64>;
def : DivergentClampingBinOp<sub, V_SUB_I32_e64>;
}
def : DivergentBinOp<adde, V_ADDC_U32_e32>;
def : DivergentBinOp<sube, V_SUBB_U32_e32>;
class divergent_i64_BinOp <SDPatternOperator Op, Instruction Inst> :
GCNPat<
(getDivergentFrag<Op>.ret i64:$src0, i64:$src1),
(REG_SEQUENCE VReg_64,
(Inst
(i32 (EXTRACT_SUBREG $src0, sub0)),
(i32 (EXTRACT_SUBREG $src1, sub0))
), sub0,
(Inst
(i32 (EXTRACT_SUBREG $src0, sub1)),
(i32 (EXTRACT_SUBREG $src1, sub1))
), sub1
)
>;
def : divergent_i64_BinOp <and, V_AND_B32_e32>;
def : divergent_i64_BinOp <or, V_OR_B32_e32>;
def : divergent_i64_BinOp <xor, V_XOR_B32_e32>;
let SubtargetPredicate = Has16BitInsts in {
let FPDPRounding = 1 in {
def V_MADMK_F16 : VOP2_Pseudo <"v_madmk_f16", VOP_MADMK_F16, [], "">;
defm V_LDEXP_F16 : VOP2Inst <"v_ldexp_f16", VOP_F16_F16_I32, AMDGPUldexp>;
} // End FPDPRounding = 1
defm V_LSHLREV_B16 : VOP2Inst <"v_lshlrev_b16", VOP_I16_I16_I16, lshl_rev>;
defm V_LSHRREV_B16 : VOP2Inst <"v_lshrrev_b16", VOP_I16_I16_I16, lshr_rev>;
defm V_ASHRREV_I16 : VOP2Inst <"v_ashrrev_i16", VOP_I16_I16_I16, ashr_rev>;
let isCommutable = 1 in {
let FPDPRounding = 1 in {
defm V_ADD_F16 : VOP2Inst <"v_add_f16", VOP_F16_F16_F16, fadd>;
defm V_SUB_F16 : VOP2Inst <"v_sub_f16", VOP_F16_F16_F16, fsub>;
defm V_SUBREV_F16 : VOP2Inst <"v_subrev_f16", VOP_F16_F16_F16, null_frag, "v_sub_f16">;
defm V_MUL_F16 : VOP2Inst <"v_mul_f16", VOP_F16_F16_F16, fmul>;
def V_MADAK_F16 : VOP2_Pseudo <"v_madak_f16", VOP_MADAK_F16, [], "">;
} // End FPDPRounding = 1
defm V_ADD_U16 : VOP2Inst <"v_add_u16", VOP_I16_I16_I16, add>;
defm V_SUB_U16 : VOP2Inst <"v_sub_u16" , VOP_I16_I16_I16, sub>;
defm V_SUBREV_U16 : VOP2Inst <"v_subrev_u16", VOP_I16_I16_I16, null_frag, "v_sub_u16">;
defm V_MUL_LO_U16 : VOP2Inst <"v_mul_lo_u16", VOP_I16_I16_I16, mul>;
defm V_MAX_F16 : VOP2Inst <"v_max_f16", VOP_F16_F16_F16, fmaxnum_like>;
defm V_MIN_F16 : VOP2Inst <"v_min_f16", VOP_F16_F16_F16, fminnum_like>;
defm V_MAX_U16 : VOP2Inst <"v_max_u16", VOP_I16_I16_I16, umax>;
defm V_MAX_I16 : VOP2Inst <"v_max_i16", VOP_I16_I16_I16, smax>;
defm V_MIN_U16 : VOP2Inst <"v_min_u16", VOP_I16_I16_I16, umin>;
defm V_MIN_I16 : VOP2Inst <"v_min_i16", VOP_I16_I16_I16, smin>;
let Constraints = "$vdst = $src2", DisableEncoding="$src2",
isConvertibleToThreeAddress = 1 in {
defm V_MAC_F16 : VOP2Inst <"v_mac_f16", VOP_MAC_F16>;
}
} // End isCommutable = 1
} // End SubtargetPredicate = Has16BitInsts
let SubtargetPredicate = HasDLInsts in {
defm V_XNOR_B32 : VOP2Inst <"v_xnor_b32", VOP_I32_I32_I32>;
let Constraints = "$vdst = $src2",
DisableEncoding="$src2",
isConvertibleToThreeAddress = 1,
isCommutable = 1 in {
defm V_FMAC_F32 : VOP2Inst <"v_fmac_f32", VOP_MAC_F32>;
}
} // End SubtargetPredicate = HasDLInsts
let Constraints = "$vdst = $src2",
DisableEncoding="$src2",
isConvertibleToThreeAddress = 1,
isCommutable = 1,
IsDOT = 1 in {
let SubtargetPredicate = HasDot5Insts in
defm V_DOT2C_F32_F16 : VOP2Inst<"v_dot2c_f32_f16", VOP_DOT_ACC_F32_V2F16>;
let SubtargetPredicate = HasDot6Insts in
defm V_DOT4C_I32_I8 : VOP2Inst<"v_dot4c_i32_i8", VOP_DOT_ACC_I32_I32>;
let SubtargetPredicate = HasDot4Insts in
defm V_DOT2C_I32_I16 : VOP2Inst<"v_dot2c_i32_i16", VOP_DOT_ACC_I32_I32>;
let SubtargetPredicate = HasDot3Insts in
defm V_DOT8C_I32_I4 : VOP2Inst<"v_dot8c_i32_i4", VOP_DOT_ACC_I32_I32>;
}
let AddedComplexity = 30 in {
def : GCNPat<
(f32 (AMDGPUfdot2 v2f16:$src0, v2f16:$src1, f32:$src2, (i1 DSTCLAMP.NONE))),
(f32 (V_DOT2C_F32_F16_e32 $src0, $src1, $src2))
> {
let SubtargetPredicate = HasDot5Insts;
}
def : GCNPat<
(i32 (int_amdgcn_sdot4 i32:$src0, i32:$src1, i32:$src2, (i1 DSTCLAMP.NONE))),
(i32 (V_DOT4C_I32_I8_e32 $src0, $src1, $src2))
> {
let SubtargetPredicate = HasDot6Insts;
}
def : GCNPat<
(i32 (int_amdgcn_sdot2 v2i16:$src0, v2i16:$src1, i32:$src2, (i1 DSTCLAMP.NONE))),
(i32 (V_DOT2C_I32_I16_e32 $src0, $src1, $src2))
> {
let SubtargetPredicate = HasDot4Insts;
}
def : GCNPat<
(i32 (int_amdgcn_sdot8 i32:$src0, i32:$src1, i32:$src2, (i1 DSTCLAMP.NONE))),
(i32 (V_DOT8C_I32_I4_e32 $src0, $src1, $src2))
> {
let SubtargetPredicate = HasDot3Insts;
}
} // End AddedComplexity = 30
let SubtargetPredicate = isGFX10Plus in {
def V_FMAMK_F32 : VOP2_Pseudo<"v_fmamk_f32", VOP_MADMK_F32, [], "">;
let FPDPRounding = 1 in
def V_FMAMK_F16 : VOP2_Pseudo <"v_fmamk_f16", VOP_MADMK_F16, [], "">;
let isCommutable = 1 in {
def V_FMAAK_F32 : VOP2_Pseudo<"v_fmaak_f32", VOP_MADAK_F32, [], "">;
let FPDPRounding = 1 in
def V_FMAAK_F16 : VOP2_Pseudo <"v_fmaak_f16", VOP_MADAK_F16, [], "">;
} // End isCommutable = 1
let Constraints = "$vdst = $src2",
DisableEncoding="$src2",
isConvertibleToThreeAddress = 1,
isCommutable = 1 in {
defm V_FMAC_F16 : VOP2Inst <"v_fmac_f16", VOP_MAC_F16>;
}
} // End SubtargetPredicate = isGFX10Plus
let SubtargetPredicate = HasPkFmacF16Inst in {
defm V_PK_FMAC_F16 : VOP2Inst<"v_pk_fmac_f16", VOP_V2F16_V2F16_V2F16>;
} // End SubtargetPredicate = HasPkFmacF16Inst
// 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_Pats <SDPatternOperator op, Instruction inst> {
def : GCNPat<
(i32 (zext (op i16:$src0, i16:$src1))),
(inst $src0, $src1)
>;
def : GCNPat<
(i64 (zext (op i16:$src0, i16:$src1))),
(REG_SEQUENCE VReg_64,
(inst $src0, $src1), sub0,
(V_MOV_B32_e32 (i32 0)), sub1)
>;
}
class ZExt_i16_i1_Pat <SDNode ext> : GCNPat <
(i16 (ext i1:$src)),
(V_CNDMASK_B32_e64 (i32 0/*src0mod*/), (i32 0/*src0*/),
(i32 0/*src1mod*/), (i32 1/*src1*/),
$src)
>;
foreach vt = [i16, v2i16] in {
def : GCNPat <
(and vt:$src0, vt:$src1),
(V_AND_B32_e64 VSrc_b32:$src0, VSrc_b32:$src1)
>;
def : GCNPat <
(or vt:$src0, vt:$src1),
(V_OR_B32_e64 VSrc_b32:$src0, VSrc_b32:$src1)
>;
def : GCNPat <
(xor vt:$src0, vt:$src1),
(V_XOR_B32_e64 VSrc_b32:$src0, VSrc_b32:$src1)
>;
}
let Predicates = [Has16BitInsts] in {
let Predicates = [Has16BitInsts, isGFX7GFX8GFX9] in {
defm : Arithmetic_i16_0Hi_Pats<add, V_ADD_U16_e64>;
defm : Arithmetic_i16_0Hi_Pats<mul, V_MUL_LO_U16_e64>;
defm : Arithmetic_i16_0Hi_Pats<sub, V_SUB_U16_e64>;
defm : Arithmetic_i16_0Hi_Pats<smin, V_MIN_I16_e64>;
defm : Arithmetic_i16_0Hi_Pats<smax, V_MAX_I16_e64>;
defm : Arithmetic_i16_0Hi_Pats<umin, V_MIN_U16_e64>;
defm : Arithmetic_i16_0Hi_Pats<umax, V_MAX_U16_e64>;
defm : Arithmetic_i16_0Hi_Pats<lshl_rev, V_LSHLREV_B16_e64>;
defm : Arithmetic_i16_0Hi_Pats<lshr_rev, V_LSHRREV_B16_e64>;
defm : Arithmetic_i16_0Hi_Pats<ashr_rev, V_ASHRREV_I16_e64>;
}
def : ZExt_i16_i1_Pat<zext>;
def : ZExt_i16_i1_Pat<anyext>;
def : GCNPat <
(i16 (sext i1:$src)),
(V_CNDMASK_B32_e64 /*src0mod*/(i32 0), /*src0*/(i32 0),
/*src1mod*/(i32 0), /*src1*/(i32 -1), $src)
>;
// Undo sub x, c -> add x, -c canonicalization since c is more likely
// an inline immediate than -c.
// TODO: Also do for 64-bit.
def : GCNPat<
(add i16:$src0, (i16 NegSubInlineConst16:$src1)),
(V_SUB_U16_e64 $src0, NegSubInlineConst16:$src1)
>;
} // End Predicates = [Has16BitInsts, isGFX7GFX8GFX9]
//===----------------------------------------------------------------------===//
// Target-specific instruction encodings.
//===----------------------------------------------------------------------===//
class VOP2_DPP<bits<6> op, VOP2_DPP_Pseudo ps,
string opName = ps.OpName, VOPProfile p = ps.Pfl,
bit IsDPP16 = 0> :
VOP_DPP<opName, p, IsDPP16> {
let hasSideEffects = ps.hasSideEffects;
let Defs = ps.Defs;
let SchedRW = ps.SchedRW;
let Uses = ps.Uses;
bits<8> vdst;
bits<8> src1;
let Inst{8-0} = 0xfa;
let Inst{16-9} = !if(p.HasSrc1, src1{7-0}, 0);
let Inst{24-17} = !if(p.EmitDst, vdst{7-0}, 0);
let Inst{30-25} = op;
let Inst{31} = 0x0;
}
class Base_VOP2_DPP16<bits<6> op, VOP2_DPP_Pseudo ps,
string opName = ps.OpName, VOPProfile p = ps.Pfl> :
VOP2_DPP<op, ps, opName, p, 1> {
let AssemblerPredicate = !if(p.HasExt, HasDPP16, DisableInst);
let SubtargetPredicate = HasDPP16;
}
class VOP2_DPP16<bits<6> op, VOP2_DPP_Pseudo ps,
string opName = ps.OpName, VOPProfile p = ps.Pfl> :
Base_VOP2_DPP16<op, ps, opName, p>,
SIMCInstr <ps.PseudoInstr, SIEncodingFamily.GFX10>;
class VOP2_DPP8<bits<6> op, VOP2_Pseudo ps,
string opName = ps.OpName, VOPProfile p = ps.Pfl> :
VOP_DPP8<ps.OpName, p> {
let hasSideEffects = ps.hasSideEffects;
let Defs = ps.Defs;
let SchedRW = ps.SchedRW;
let Uses = ps.Uses;
bits<8> vdst;
bits<8> src1;
let Inst{8-0} = fi;
let Inst{16-9} = !if(p.HasSrc1, src1{7-0}, 0);
let Inst{24-17} = !if(p.EmitDst, vdst{7-0}, 0);
let Inst{30-25} = op;
let Inst{31} = 0x0;
let AssemblerPredicate = !if(p.HasExt, HasDPP8, DisableInst);
let SubtargetPredicate = HasDPP8;
}
//===----------------------------------------------------------------------===//
// GFX10.
//===----------------------------------------------------------------------===//
let AssemblerPredicate = isGFX10Plus, DecoderNamespace = "GFX10" in {
//===------------------------------- VOP2 -------------------------------===//
multiclass VOP2Only_Real_MADK_gfx10<bits<6> op> {
def _gfx10 :
VOP2_Real<!cast<VOP2_Pseudo>(NAME), SIEncodingFamily.GFX10>,
VOP2_MADKe<op{5-0}, !cast<VOP2_Pseudo>(NAME).Pfl>;
}
multiclass VOP2Only_Real_MADK_gfx10_with_name<bits<6> op, string opName,
string asmName> {
def _gfx10 :
VOP2_Real<!cast<VOP2_Pseudo>(opName), SIEncodingFamily.GFX10>,
VOP2_MADKe<op{5-0}, !cast<VOP2_Pseudo>(opName).Pfl> {
VOP2_Pseudo ps = !cast<VOP2_Pseudo>(opName);
let AsmString = asmName # ps.AsmOperands;
}
}
multiclass VOP2_Real_e32_gfx10<bits<6> op> {
def _e32_gfx10 :
VOP2_Real<!cast<VOP2_Pseudo>(NAME#"_e32"), SIEncodingFamily.GFX10>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(NAME#"_e32").Pfl>;
}
multiclass VOP2_Real_e64_gfx10<bits<6> op> {
def _e64_gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX10>,
VOP3e_gfx10<{0, 1, 0, 0, op{5-0}}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP2_Real_sdwa_gfx10<bits<6> op> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(NAME#"_e32").Pfl.HasExtSDWA9>.ret in
def _sdwa_gfx10 :
VOP_SDWA10_Real<!cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa")>,
VOP2_SDWA9Ae<op{5-0}, !cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa").Pfl> {
let DecoderNamespace = "SDWA10";
}
}
multiclass VOP2_Real_dpp_gfx10<bits<6> op> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(NAME#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_gfx10 : VOP2_DPP16<op, !cast<VOP2_DPP_Pseudo>(NAME#"_dpp")> {
let DecoderNamespace = "SDWA10";
}
}
multiclass VOP2_Real_dpp8_gfx10<bits<6> op> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(NAME#"_e32").Pfl.HasExtDPP>.ret in
def _dpp8_gfx10 : VOP2_DPP8<op, !cast<VOP2_Pseudo>(NAME#"_e32")> {
let DecoderNamespace = "DPP8";
}
}
//===------------------------- VOP2 (with name) -------------------------===//
multiclass VOP2_Real_e32_gfx10_with_name<bits<6> op, string opName,
string asmName> {
def _e32_gfx10 :
VOP2_Real<!cast<VOP2_Pseudo>(opName#"_e32"), SIEncodingFamily.GFX10>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(opName#"_e32").Pfl> {
VOP2_Pseudo ps = !cast<VOP2_Pseudo>(opName#"_e32");
let AsmString = asmName # ps.AsmOperands;
}
}
multiclass VOP2_Real_e64_gfx10_with_name<bits<6> op, string opName,
string asmName> {
def _e64_gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(opName#"_e64"), SIEncodingFamily.GFX10>,
VOP3e_gfx10<{0, 1, 0, 0, op{5-0}},
!cast<VOP3_Pseudo>(opName#"_e64").Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(opName#"_e64");
let AsmString = asmName # ps.AsmOperands;
}
}
let DecoderNamespace = "SDWA10" in {
multiclass VOP2_Real_sdwa_gfx10_with_name<bits<6> op, string opName,
string asmName> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtSDWA9>.ret in
def _sdwa_gfx10 :
VOP_SDWA10_Real<!cast<VOP2_SDWA_Pseudo>(opName#"_sdwa")>,
VOP2_SDWA9Ae<op{5-0}, !cast<VOP2_SDWA_Pseudo>(opName#"_sdwa").Pfl> {
VOP2_SDWA_Pseudo ps = !cast<VOP2_SDWA_Pseudo>(opName#"_sdwa");
let AsmString = asmName # ps.AsmOperands;
}
}
multiclass VOP2_Real_dpp_gfx10_with_name<bits<6> op, string opName,
string asmName> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_gfx10 : VOP2_DPP16<op, !cast<VOP2_DPP_Pseudo>(opName#"_dpp")> {
VOP2_Pseudo ps = !cast<VOP2_Pseudo>(opName#"_e32");
let AsmString = asmName # ps.Pfl.AsmDPP16;
}
}
multiclass VOP2_Real_dpp8_gfx10_with_name<bits<6> op, string opName,
string asmName> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp8_gfx10 : VOP2_DPP8<op, !cast<VOP2_Pseudo>(opName#"_e32")> {
VOP2_Pseudo ps = !cast<VOP2_Pseudo>(opName#"_e32");
let AsmString = asmName # ps.Pfl.AsmDPP8;
let DecoderNamespace = "DPP8";
}
}
} // End DecoderNamespace = "SDWA10"
//===------------------------------ VOP2be ------------------------------===//
multiclass VOP2be_Real_e32_gfx10<bits<6> op, string opName, string asmName> {
def _e32_gfx10 :
VOP2_Real<!cast<VOP2_Pseudo>(opName#"_e32"), SIEncodingFamily.GFX10>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(opName#"_e32").Pfl> {
VOP2_Pseudo Ps = !cast<VOP2_Pseudo>(opName#"_e32");
let AsmString = asmName # !subst(", vcc", "", Ps.AsmOperands);
}
}
multiclass VOP2be_Real_e64_gfx10<bits<6> op, string opName, string asmName> {
def _e64_gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(opName#"_e64"), SIEncodingFamily.GFX10>,
VOP3be_gfx10<{0, 1, 0, 0, op{5-0}},
!cast<VOP3_Pseudo>(opName#"_e64").Pfl> {
VOP3_Pseudo Ps = !cast<VOP3_Pseudo>(opName#"_e64");
let AsmString = asmName # Ps.AsmOperands;
}
}
multiclass VOP2be_Real_sdwa_gfx10<bits<6> op, string opName, string asmName> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtSDWA9>.ret in
def _sdwa_gfx10 :
VOP_SDWA10_Real<!cast<VOP2_SDWA_Pseudo>(opName#"_sdwa")>,
VOP2_SDWA9Ae<op{5-0}, !cast<VOP2_SDWA_Pseudo>(opName#"_sdwa").Pfl> {
VOP2_SDWA_Pseudo Ps = !cast<VOP2_SDWA_Pseudo>(opName#"_sdwa");
let AsmString = asmName # !subst(", vcc", "", Ps.AsmOperands);
let DecoderNamespace = "SDWA10";
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtSDWA9>.ret in
def _sdwa_w32_gfx10 :
Base_VOP_SDWA10_Real<!cast<VOP2_SDWA_Pseudo>(opName#"_sdwa")>,
VOP2_SDWA9Ae<op{5-0}, !cast<VOP2_SDWA_Pseudo>(opName#"_sdwa").Pfl> {
VOP2_SDWA_Pseudo Ps = !cast<VOP2_SDWA_Pseudo>(opName#"_sdwa");
let AsmString = asmName # !subst("vcc", "vcc_lo", Ps.AsmOperands);
let isAsmParserOnly = 1;
let DecoderNamespace = "SDWA10";
let WaveSizePredicate = isWave32;
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtSDWA9>.ret in
def _sdwa_w64_gfx10 :
Base_VOP_SDWA10_Real<!cast<VOP2_SDWA_Pseudo>(opName#"_sdwa")>,
VOP2_SDWA9Ae<op{5-0}, !cast<VOP2_SDWA_Pseudo>(opName#"_sdwa").Pfl> {
VOP2_SDWA_Pseudo Ps = !cast<VOP2_SDWA_Pseudo>(opName#"_sdwa");
let AsmString = asmName # Ps.AsmOperands;
let isAsmParserOnly = 1;
let DecoderNamespace = "SDWA10";
let WaveSizePredicate = isWave64;
}
}
multiclass VOP2be_Real_dpp_gfx10<bits<6> op, string opName, string asmName> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_gfx10 :
VOP2_DPP16<op, !cast<VOP2_DPP_Pseudo>(opName#"_dpp"), asmName> {
string AsmDPP = !cast<VOP2_Pseudo>(opName#"_e32").Pfl.AsmDPP16;
let AsmString = asmName # !subst(", vcc", "", AsmDPP);
let DecoderNamespace = "SDWA10";
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_w32_gfx10 :
Base_VOP2_DPP16<op, !cast<VOP2_DPP_Pseudo>(opName#"_dpp"), asmName> {
string AsmDPP = !cast<VOP2_Pseudo>(opName#"_e32").Pfl.AsmDPP16;
let AsmString = asmName # !subst("vcc", "vcc_lo", AsmDPP);
let isAsmParserOnly = 1;
let WaveSizePredicate = isWave32;
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_w64_gfx10 :
Base_VOP2_DPP16<op, !cast<VOP2_DPP_Pseudo>(opName#"_dpp"), asmName> {
string AsmDPP = !cast<VOP2_Pseudo>(opName#"_e32").Pfl.AsmDPP16;
let AsmString = asmName # AsmDPP;
let isAsmParserOnly = 1;
let WaveSizePredicate = isWave64;
}
}
multiclass VOP2be_Real_dpp8_gfx10<bits<6> op, string opName, string asmName> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp8_gfx10 :
VOP2_DPP8<op, !cast<VOP2_Pseudo>(opName#"_e32"), asmName> {
string AsmDPP8 = !cast<VOP2_Pseudo>(opName#"_e32").Pfl.AsmDPP8;
let AsmString = asmName # !subst(", vcc", "", AsmDPP8);
let DecoderNamespace = "DPP8";
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp8_w32_gfx10 :
VOP2_DPP8<op, !cast<VOP2_Pseudo>(opName#"_e32"), asmName> {
string AsmDPP8 = !cast<VOP2_Pseudo>(opName#"_e32").Pfl.AsmDPP8;
let AsmString = asmName # !subst("vcc", "vcc_lo", AsmDPP8);
let isAsmParserOnly = 1;
let WaveSizePredicate = isWave32;
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(opName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp8_w64_gfx10 :
VOP2_DPP8<op, !cast<VOP2_Pseudo>(opName#"_e32"), asmName> {
string AsmDPP8 = !cast<VOP2_Pseudo>(opName#"_e32").Pfl.AsmDPP8;
let AsmString = asmName # AsmDPP8;
let isAsmParserOnly = 1;
let WaveSizePredicate = isWave64;
}
}
//===----------------------------- VOP3Only -----------------------------===//
multiclass VOP3Only_Real_gfx10<bits<10> op> {
def _e64_gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX10>,
VOP3e_gfx10<op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
//===---------------------------- VOP3beOnly ----------------------------===//
multiclass VOP3beOnly_Real_gfx10<bits<10> op, string opName, string asmName> {
def _e64_gfx10 :
VOP3_Real<!cast<VOP3_Pseudo>(opName#"_e64"), SIEncodingFamily.GFX10>,
VOP3be_gfx10<op, !cast<VOP3_Pseudo>(opName#"_e64").Pfl> {
VOP3_Pseudo Ps = !cast<VOP3_Pseudo>(opName#"_e64");
let AsmString = asmName # Ps.AsmOperands;
}
}
} // End AssemblerPredicate = isGFX10Plus, DecoderNamespace = "GFX10"
multiclass VOP2be_Real_gfx10<bits<6> op, string opName, string asmName> :
VOP2be_Real_e32_gfx10<op, opName, asmName>,
VOP2be_Real_e64_gfx10<op, opName, asmName>,
VOP2be_Real_sdwa_gfx10<op, opName, asmName>,
VOP2be_Real_dpp_gfx10<op, opName, asmName>,
VOP2be_Real_dpp8_gfx10<op, opName, asmName>;
multiclass VOP2e_Real_gfx10<bits<6> op, string opName, string asmName> :
VOP2_Real_e32_gfx10<op>,
VOP2_Real_e64_gfx10<op>,
VOP2be_Real_sdwa_gfx10<op, opName, asmName>,
VOP2be_Real_dpp_gfx10<op, opName, asmName>,
VOP2be_Real_dpp8_gfx10<op, opName, asmName>;
multiclass VOP2_Real_gfx10<bits<6> op> :
VOP2_Real_e32_gfx10<op>, VOP2_Real_e64_gfx10<op>,
VOP2_Real_sdwa_gfx10<op>, VOP2_Real_dpp_gfx10<op>, VOP2_Real_dpp8_gfx10<op>;
multiclass VOP2_Real_gfx10_with_name<bits<6> op, string opName,
string asmName> :
VOP2_Real_e32_gfx10_with_name<op, opName, asmName>,
VOP2_Real_e64_gfx10_with_name<op, opName, asmName>,
VOP2_Real_sdwa_gfx10_with_name<op, opName, asmName>,
VOP2_Real_dpp_gfx10_with_name<op, opName, asmName>,
VOP2_Real_dpp8_gfx10_with_name<op, opName, asmName>;
defm V_XNOR_B32 : VOP2_Real_gfx10<0x01e>;
defm V_FMAC_F32 : VOP2_Real_gfx10<0x02b>;
defm V_FMAMK_F32 : VOP2Only_Real_MADK_gfx10<0x02c>;
defm V_FMAAK_F32 : VOP2Only_Real_MADK_gfx10<0x02d>;
defm V_ADD_F16 : VOP2_Real_gfx10<0x032>;
defm V_SUB_F16 : VOP2_Real_gfx10<0x033>;
defm V_SUBREV_F16 : VOP2_Real_gfx10<0x034>;
defm V_MUL_F16 : VOP2_Real_gfx10<0x035>;
defm V_FMAC_F16 : VOP2_Real_gfx10<0x036>;
defm V_FMAMK_F16 : VOP2Only_Real_MADK_gfx10<0x037>;
defm V_FMAAK_F16 : VOP2Only_Real_MADK_gfx10<0x038>;
defm V_MAX_F16 : VOP2_Real_gfx10<0x039>;
defm V_MIN_F16 : VOP2_Real_gfx10<0x03a>;
defm V_LDEXP_F16 : VOP2_Real_gfx10<0x03b>;
defm V_PK_FMAC_F16 : VOP2_Real_e32_gfx10<0x03c>;
// VOP2 no carry-in, carry-out.
defm V_ADD_NC_U32 :
VOP2_Real_gfx10_with_name<0x025, "V_ADD_U32", "v_add_nc_u32">;
defm V_SUB_NC_U32 :
VOP2_Real_gfx10_with_name<0x026, "V_SUB_U32", "v_sub_nc_u32">;
defm V_SUBREV_NC_U32 :
VOP2_Real_gfx10_with_name<0x027, "V_SUBREV_U32", "v_subrev_nc_u32">;
// VOP2 carry-in, carry-out.
defm V_ADD_CO_CI_U32 :
VOP2be_Real_gfx10<0x028, "V_ADDC_U32", "v_add_co_ci_u32">;
defm V_SUB_CO_CI_U32 :
VOP2be_Real_gfx10<0x029, "V_SUBB_U32", "v_sub_co_ci_u32">;
defm V_SUBREV_CO_CI_U32 :
VOP2be_Real_gfx10<0x02a, "V_SUBBREV_U32", "v_subrev_co_ci_u32">;
defm V_CNDMASK_B32 :
VOP2e_Real_gfx10<0x001, "V_CNDMASK_B32", "v_cndmask_b32">;
// VOP3 only.
defm V_BFM_B32 : VOP3Only_Real_gfx10<0x363>;
defm V_BCNT_U32_B32 : VOP3Only_Real_gfx10<0x364>;
defm V_MBCNT_LO_U32_B32 : VOP3Only_Real_gfx10<0x365>;
defm V_MBCNT_HI_U32_B32 : VOP3Only_Real_gfx10<0x366>;
defm V_LDEXP_F32 : VOP3Only_Real_gfx10<0x362>;
defm V_CVT_PKNORM_I16_F32 : VOP3Only_Real_gfx10<0x368>;
defm V_CVT_PKNORM_U16_F32 : VOP3Only_Real_gfx10<0x369>;
defm V_CVT_PK_U16_U32 : VOP3Only_Real_gfx10<0x36a>;
defm V_CVT_PK_I16_I32 : VOP3Only_Real_gfx10<0x36b>;
// VOP3 carry-in, carry-out.
defm V_ADD_CO_U32 :
VOP3beOnly_Real_gfx10<0x30f, "V_ADD_I32", "v_add_co_u32">;
defm V_SUB_CO_U32 :
VOP3beOnly_Real_gfx10<0x310, "V_SUB_I32", "v_sub_co_u32">;
defm V_SUBREV_CO_U32 :
VOP3beOnly_Real_gfx10<0x319, "V_SUBREV_I32", "v_subrev_co_u32">;
let SubtargetPredicate = isGFX10Plus in {
defm : VOP2eInstAliases<V_CNDMASK_B32_e32, V_CNDMASK_B32_e32_gfx10>;
defm : VOP2bInstAliases<
V_ADDC_U32_e32, V_ADD_CO_CI_U32_e32_gfx10, "v_add_co_ci_u32">;
defm : VOP2bInstAliases<
V_SUBB_U32_e32, V_SUB_CO_CI_U32_e32_gfx10, "v_sub_co_ci_u32">;
defm : VOP2bInstAliases<
V_SUBBREV_U32_e32, V_SUBREV_CO_CI_U32_e32_gfx10, "v_subrev_co_ci_u32">;
} // End SubtargetPredicate = isGFX10Plus
//===----------------------------------------------------------------------===//
// GFX6, GFX7, GFX10.
//===----------------------------------------------------------------------===//
class VOP2_DPPe <bits<6> op, VOP2_DPP_Pseudo ps, VOPProfile P = ps.Pfl> :
VOP_DPPe <P> {
bits<8> vdst;
bits<8> src1;
let Inst{8-0} = 0xfa; //dpp
let Inst{16-9} = !if(P.HasSrc1, src1{7-0}, 0);
let Inst{24-17} = !if(P.EmitDst, vdst{7-0}, 0);
let Inst{30-25} = op;
let Inst{31} = 0x0; //encoding
}
let AssemblerPredicate = isGFX6GFX7, DecoderNamespace = "GFX6GFX7" in {
multiclass VOP2Only_Real_gfx6_gfx7<bits<6> op> {
def _gfx6_gfx7 :
VOP2_Real<!cast<VOP2_Pseudo>(NAME), SIEncodingFamily.SI>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(NAME).Pfl>;
}
multiclass VOP2Only_Real_MADK_gfx6_gfx7<bits<6> op> {
def _gfx6_gfx7 :
VOP2_Real<!cast<VOP2_Pseudo>(NAME), SIEncodingFamily.SI>,
VOP2_MADKe<op{5-0}, !cast<VOP2_Pseudo>(NAME).Pfl>;
}
multiclass VOP2_Real_e32_gfx6_gfx7<bits<6> op> {
def _e32_gfx6_gfx7 :
VOP2_Real<!cast<VOP2_Pseudo>(NAME#"_e32"), SIEncodingFamily.SI>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(NAME#"_e32").Pfl>;
}
multiclass VOP2_Real_e64_gfx6_gfx7<bits<6> op> {
def _e64_gfx6_gfx7 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
VOP3e_gfx6_gfx7<{1, 0, 0, op{5-0}}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP2be_Real_e64_gfx6_gfx7<bits<6> op> {
def _e64_gfx6_gfx7 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.SI>,
VOP3be_gfx6_gfx7<{1, 0, 0, op{5-0}}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
} // End AssemblerPredicate = isGFX6GFX7, DecoderNamespace = "GFX6GFX7"
multiclass VOP2Only_Real_MADK_gfx6_gfx7_gfx10<bits<6> op> :
VOP2Only_Real_MADK_gfx6_gfx7<op>, VOP2Only_Real_MADK_gfx10<op>;
multiclass VOP2_Real_gfx6_gfx7<bits<6> op> :
VOP2_Real_e32_gfx6_gfx7<op>, VOP2_Real_e64_gfx6_gfx7<op>;
multiclass VOP2_Real_gfx6_gfx7_gfx10<bits<6> op> :
VOP2_Real_gfx6_gfx7<op>, VOP2_Real_gfx10<op>;
multiclass VOP2be_Real_gfx6_gfx7<bits<6> op> :
VOP2_Real_e32_gfx6_gfx7<op>, VOP2be_Real_e64_gfx6_gfx7<op>;
defm V_CNDMASK_B32 : VOP2_Real_gfx6_gfx7<0x000>;
defm V_MIN_LEGACY_F32 : VOP2_Real_gfx6_gfx7<0x00d>;
defm V_MAX_LEGACY_F32 : VOP2_Real_gfx6_gfx7<0x00e>;
defm V_LSHR_B32 : VOP2_Real_gfx6_gfx7<0x015>;
defm V_ASHR_I32 : VOP2_Real_gfx6_gfx7<0x017>;
defm V_LSHL_B32 : VOP2_Real_gfx6_gfx7<0x019>;
defm V_BFM_B32 : VOP2_Real_gfx6_gfx7<0x01e>;
defm V_BCNT_U32_B32 : VOP2_Real_gfx6_gfx7<0x022>;
defm V_MBCNT_LO_U32_B32 : VOP2_Real_gfx6_gfx7<0x023>;
defm V_MBCNT_HI_U32_B32 : VOP2_Real_gfx6_gfx7<0x024>;
defm V_LDEXP_F32 : VOP2_Real_gfx6_gfx7<0x02b>;
defm V_CVT_PKACCUM_U8_F32 : VOP2_Real_gfx6_gfx7<0x02c>;
defm V_CVT_PKNORM_I16_F32 : VOP2_Real_gfx6_gfx7<0x02d>;
defm V_CVT_PKNORM_U16_F32 : VOP2_Real_gfx6_gfx7<0x02e>;
defm V_CVT_PK_U16_U32 : VOP2_Real_gfx6_gfx7<0x030>;
defm V_CVT_PK_I16_I32 : VOP2_Real_gfx6_gfx7<0x031>;
defm V_ADD_I32 : VOP2be_Real_gfx6_gfx7<0x025>;
defm V_SUB_I32 : VOP2be_Real_gfx6_gfx7<0x026>;
defm V_SUBREV_I32 : VOP2be_Real_gfx6_gfx7<0x027>;
defm V_ADDC_U32 : VOP2be_Real_gfx6_gfx7<0x028>;
defm V_SUBB_U32 : VOP2be_Real_gfx6_gfx7<0x029>;
defm V_SUBBREV_U32 : VOP2be_Real_gfx6_gfx7<0x02a>;
defm V_READLANE_B32 : VOP2Only_Real_gfx6_gfx7<0x001>;
let InOperandList = (ins SSrcOrLds_b32:$src0, SCSrc_b32:$src1, VSrc_b32:$vdst_in) in {
defm V_WRITELANE_B32 : VOP2Only_Real_gfx6_gfx7<0x002>;
} // End InOperandList = (ins SSrcOrLds_b32:$src0, SCSrc_b32:$src1, VSrc_b32:$vdst_in)
let SubtargetPredicate = isGFX6GFX7 in {
defm : VOP2eInstAliases<V_CNDMASK_B32_e32, V_CNDMASK_B32_e32_gfx6_gfx7>;
} // End SubtargetPredicate = isGFX6GFX7
defm V_ADD_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x003>;
defm V_SUB_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x004>;
defm V_SUBREV_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x005>;
defm V_MAC_LEGACY_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x006>;
defm V_MUL_LEGACY_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x007>;
defm V_MUL_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x008>;
defm V_MUL_I32_I24 : VOP2_Real_gfx6_gfx7_gfx10<0x009>;
defm V_MUL_HI_I32_I24 : VOP2_Real_gfx6_gfx7_gfx10<0x00a>;
defm V_MUL_U32_U24 : VOP2_Real_gfx6_gfx7_gfx10<0x00b>;
defm V_MUL_HI_U32_U24 : VOP2_Real_gfx6_gfx7_gfx10<0x00c>;
defm V_MIN_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x00f>;
defm V_MAX_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x010>;
defm V_MIN_I32 : VOP2_Real_gfx6_gfx7_gfx10<0x011>;
defm V_MAX_I32 : VOP2_Real_gfx6_gfx7_gfx10<0x012>;
defm V_MIN_U32 : VOP2_Real_gfx6_gfx7_gfx10<0x013>;
defm V_MAX_U32 : VOP2_Real_gfx6_gfx7_gfx10<0x014>;
defm V_LSHRREV_B32 : VOP2_Real_gfx6_gfx7_gfx10<0x016>;
defm V_ASHRREV_I32 : VOP2_Real_gfx6_gfx7_gfx10<0x018>;
defm V_LSHLREV_B32 : VOP2_Real_gfx6_gfx7_gfx10<0x01a>;
defm V_AND_B32 : VOP2_Real_gfx6_gfx7_gfx10<0x01b>;
defm V_OR_B32 : VOP2_Real_gfx6_gfx7_gfx10<0x01c>;
defm V_XOR_B32 : VOP2_Real_gfx6_gfx7_gfx10<0x01d>;
defm V_MAC_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x01f>;
defm V_CVT_PKRTZ_F16_F32 : VOP2_Real_gfx6_gfx7_gfx10<0x02f>;
defm V_MADMK_F32 : VOP2Only_Real_MADK_gfx6_gfx7_gfx10<0x020>;
defm V_MADAK_F32 : VOP2Only_Real_MADK_gfx6_gfx7_gfx10<0x021>;
//===----------------------------------------------------------------------===//
// GFX8, GFX9 (VI).
//===----------------------------------------------------------------------===//
let AssemblerPredicates = [isGFX8GFX9], DecoderNamespace = "GFX8" in {
multiclass VOP2_Real_MADK_vi <bits<6> op> {
def _vi : VOP2_Real<!cast<VOP2_Pseudo>(NAME), SIEncodingFamily.VI>,
VOP2_MADKe<op{5-0}, !cast<VOP2_Pseudo>(NAME).Pfl>;
}
multiclass VOP2_Real_e32_vi <bits<6> op> {
def _e32_vi :
VOP2_Real<!cast<VOP2_Pseudo>(NAME#"_e32"), SIEncodingFamily.VI>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(NAME#"_e32").Pfl>;
}
multiclass VOP2_Real_e64_vi <bits<10> op> {
def _e64_vi :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>,
VOP3e_vi <op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>;
}
multiclass VOP2_Real_e64only_vi <bits<10> op> {
def _e64_vi :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>,
VOP3e_vi <op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl> {
// Hack to stop printing _e64
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(NAME#"_e64");
let OutOperandList = (outs VGPR_32:$vdst);
let AsmString = ps.Mnemonic # " " # ps.AsmOperands;
}
}
multiclass Base_VOP2_Real_e32e64_vi <bits<6> op> :
VOP2_Real_e32_vi<op>,
VOP2_Real_e64_vi<{0, 1, 0, 0, op{5-0}}>;
} // End AssemblerPredicates = [isGFX8GFX9], DecoderNamespace = "GFX8"
multiclass VOP2_SDWA_Real <bits<6> op> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(NAME#"_e32").Pfl.HasExtSDWA>.ret in
def _sdwa_vi :
VOP_SDWA_Real <!cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa")>,
VOP2_SDWAe <op{5-0}, !cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa").Pfl>;
}
multiclass VOP2_SDWA9_Real <bits<6> op> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(NAME#"_e32").Pfl.HasExtSDWA9>.ret in
def _sdwa_gfx9 :
VOP_SDWA9_Real <!cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa")>,
VOP2_SDWA9Ae <op{5-0}, !cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa").Pfl>;
}
let AssemblerPredicates = [isGFX8Only] in {
multiclass VOP2be_Real_e32e64_vi_only <bits<6> op, string OpName, string AsmName> {
def _e32_vi :
VOP2_Real<!cast<VOP2_Pseudo>(OpName#"_e32"), SIEncodingFamily.VI>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(OpName#"_e32").Pfl> {
VOP2_Pseudo ps = !cast<VOP2_Pseudo>(OpName#"_e32");
let AsmString = AsmName # ps.AsmOperands;
let DecoderNamespace = "GFX8";
}
def _e64_vi :
VOP3_Real<!cast<VOP3_Pseudo>(OpName#"_e64"), SIEncodingFamily.VI>,
VOP3be_vi <{0, 1, 0, 0, op{5-0}}, !cast<VOP3_Pseudo>(OpName#"_e64").Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(OpName#"_e64");
let AsmString = AsmName # ps.AsmOperands;
let DecoderNamespace = "GFX8";
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(OpName#"_e32").Pfl.HasExtSDWA>.ret in
def _sdwa_vi :
VOP_SDWA_Real <!cast<VOP2_SDWA_Pseudo>(OpName#"_sdwa")>,
VOP2_SDWAe <op{5-0}, !cast<VOP2_SDWA_Pseudo>(OpName#"_sdwa").Pfl> {
VOP2_SDWA_Pseudo ps = !cast<VOP2_SDWA_Pseudo>(OpName#"_sdwa");
let AsmString = AsmName # ps.AsmOperands;
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(OpName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_vi :
VOP_DPP_Real<!cast<VOP2_DPP_Pseudo>(OpName#"_dpp"), SIEncodingFamily.VI>,
VOP2_DPPe<op, !cast<VOP2_DPP_Pseudo>(OpName#"_dpp")> {
VOP2_DPP_Pseudo ps = !cast<VOP2_DPP_Pseudo>(OpName#"_dpp");
let AsmString = AsmName # ps.AsmOperands;
}
}
}
let AssemblerPredicates = [isGFX9Only] in {
multiclass VOP2be_Real_e32e64_gfx9 <bits<6> op, string OpName, string AsmName> {
def _e32_gfx9 :
VOP2_Real<!cast<VOP2_Pseudo>(OpName#"_e32"), SIEncodingFamily.GFX9>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(OpName#"_e32").Pfl> {
VOP2_Pseudo ps = !cast<VOP2_Pseudo>(OpName#"_e32");
let AsmString = AsmName # ps.AsmOperands;
let DecoderNamespace = "GFX9";
}
def _e64_gfx9 :
VOP3_Real<!cast<VOP3_Pseudo>(OpName#"_e64"), SIEncodingFamily.GFX9>,
VOP3be_vi <{0, 1, 0, 0, op{5-0}}, !cast<VOP3_Pseudo>(OpName#"_e64").Pfl> {
VOP3_Pseudo ps = !cast<VOP3_Pseudo>(OpName#"_e64");
let AsmString = AsmName # ps.AsmOperands;
let DecoderNamespace = "GFX9";
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(OpName#"_e32").Pfl.HasExtSDWA9>.ret in
def _sdwa_gfx9 :
VOP_SDWA9_Real <!cast<VOP2_SDWA_Pseudo>(OpName#"_sdwa")>,
VOP2_SDWA9Ae <op{5-0}, !cast<VOP2_SDWA_Pseudo>(OpName#"_sdwa").Pfl> {
VOP2_SDWA_Pseudo ps = !cast<VOP2_SDWA_Pseudo>(OpName#"_sdwa");
let AsmString = AsmName # ps.AsmOperands;
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(OpName#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_gfx9 :
VOP_DPP_Real<!cast<VOP2_DPP_Pseudo>(OpName#"_dpp"), SIEncodingFamily.GFX9>,
VOP2_DPPe<op, !cast<VOP2_DPP_Pseudo>(OpName#"_dpp")> {
VOP2_DPP_Pseudo ps = !cast<VOP2_DPP_Pseudo>(OpName#"_dpp");
let AsmString = AsmName # ps.AsmOperands;
let DecoderNamespace = "SDWA9";
}
}
multiclass VOP2_Real_e32e64_gfx9 <bits<6> op> {
def _e32_gfx9 :
VOP2_Real<!cast<VOP2_Pseudo>(NAME#"_e32"), SIEncodingFamily.GFX9>,
VOP2e<op{5-0}, !cast<VOP2_Pseudo>(NAME#"_e32").Pfl>{
let DecoderNamespace = "GFX9";
}
def _e64_gfx9 :
VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX9>,
VOP3e_vi <{0, 1, 0, 0, op{5-0}}, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl> {
let DecoderNamespace = "GFX9";
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(NAME#"_e32").Pfl.HasExtSDWA9>.ret in
def _sdwa_gfx9 :
VOP_SDWA9_Real <!cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa")>,
VOP2_SDWA9Ae <op{5-0}, !cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa").Pfl> {
}
foreach _ = BoolToList<!cast<VOP2_Pseudo>(NAME#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_gfx9 :
VOP_DPP_Real<!cast<VOP2_DPP_Pseudo>(NAME#"_dpp"), SIEncodingFamily.GFX9>,
VOP2_DPPe<op, !cast<VOP2_DPP_Pseudo>(NAME#"_dpp")> {
let DecoderNamespace = "SDWA9";
}
}
} // AssemblerPredicates = [isGFX9Only]
multiclass VOP2_Real_e32e64_vi <bits<6> op> :
Base_VOP2_Real_e32e64_vi<op>, VOP2_SDWA_Real<op>, VOP2_SDWA9_Real<op> {
foreach _ = BoolToList<!cast<VOP2_Pseudo>(NAME#"_e32").Pfl.HasExtDPP>.ret in
def _dpp_vi :
VOP_DPP_Real<!cast<VOP2_DPP_Pseudo>(NAME#"_dpp"), SIEncodingFamily.VI>,
VOP2_DPPe<op, !cast<VOP2_DPP_Pseudo>(NAME#"_dpp")>;
}
defm V_CNDMASK_B32 : VOP2_Real_e32e64_vi <0x0>;
defm V_ADD_F32 : VOP2_Real_e32e64_vi <0x1>;
defm V_SUB_F32 : VOP2_Real_e32e64_vi <0x2>;
defm V_SUBREV_F32 : VOP2_Real_e32e64_vi <0x3>;
defm V_MUL_LEGACY_F32 : VOP2_Real_e32e64_vi <0x4>;
defm V_MUL_F32 : VOP2_Real_e32e64_vi <0x5>;
defm V_MUL_I32_I24 : VOP2_Real_e32e64_vi <0x6>;
defm V_MUL_HI_I32_I24 : VOP2_Real_e32e64_vi <0x7>;
defm V_MUL_U32_U24 : VOP2_Real_e32e64_vi <0x8>;
defm V_MUL_HI_U32_U24 : VOP2_Real_e32e64_vi <0x9>;
defm V_MIN_F32 : VOP2_Real_e32e64_vi <0xa>;
defm V_MAX_F32 : VOP2_Real_e32e64_vi <0xb>;
defm V_MIN_I32 : VOP2_Real_e32e64_vi <0xc>;
defm V_MAX_I32 : VOP2_Real_e32e64_vi <0xd>;
defm V_MIN_U32 : VOP2_Real_e32e64_vi <0xe>;
defm V_MAX_U32 : VOP2_Real_e32e64_vi <0xf>;
defm V_LSHRREV_B32 : VOP2_Real_e32e64_vi <0x10>;
defm V_ASHRREV_I32 : VOP2_Real_e32e64_vi <0x11>;
defm V_LSHLREV_B32 : VOP2_Real_e32e64_vi <0x12>;
defm V_AND_B32 : VOP2_Real_e32e64_vi <0x13>;
defm V_OR_B32 : VOP2_Real_e32e64_vi <0x14>;
defm V_XOR_B32 : VOP2_Real_e32e64_vi <0x15>;
defm V_MAC_F32 : VOP2_Real_e32e64_vi <0x16>;
defm V_MADMK_F32 : VOP2_Real_MADK_vi <0x17>;
defm V_MADAK_F32 : VOP2_Real_MADK_vi <0x18>;
defm V_ADD_U32 : VOP2be_Real_e32e64_vi_only <0x19, "V_ADD_I32", "v_add_u32">;
defm V_SUB_U32 : VOP2be_Real_e32e64_vi_only <0x1a, "V_SUB_I32", "v_sub_u32">;
defm V_SUBREV_U32 : VOP2be_Real_e32e64_vi_only <0x1b, "V_SUBREV_I32", "v_subrev_u32">;
defm V_ADDC_U32 : VOP2be_Real_e32e64_vi_only <0x1c, "V_ADDC_U32", "v_addc_u32">;
defm V_SUBB_U32 : VOP2be_Real_e32e64_vi_only <0x1d, "V_SUBB_U32", "v_subb_u32">;
defm V_SUBBREV_U32 : VOP2be_Real_e32e64_vi_only <0x1e, "V_SUBBREV_U32", "v_subbrev_u32">;
defm V_ADD_CO_U32 : VOP2be_Real_e32e64_gfx9 <0x19, "V_ADD_I32", "v_add_co_u32">;
defm V_SUB_CO_U32 : VOP2be_Real_e32e64_gfx9 <0x1a, "V_SUB_I32", "v_sub_co_u32">;
defm V_SUBREV_CO_U32 : VOP2be_Real_e32e64_gfx9 <0x1b, "V_SUBREV_I32", "v_subrev_co_u32">;
defm V_ADDC_CO_U32 : VOP2be_Real_e32e64_gfx9 <0x1c, "V_ADDC_U32", "v_addc_co_u32">;
defm V_SUBB_CO_U32 : VOP2be_Real_e32e64_gfx9 <0x1d, "V_SUBB_U32", "v_subb_co_u32">;
defm V_SUBBREV_CO_U32 : VOP2be_Real_e32e64_gfx9 <0x1e, "V_SUBBREV_U32", "v_subbrev_co_u32">;
defm V_ADD_U32 : VOP2_Real_e32e64_gfx9 <0x34>;
defm V_SUB_U32 : VOP2_Real_e32e64_gfx9 <0x35>;
defm V_SUBREV_U32 : VOP2_Real_e32e64_gfx9 <0x36>;
defm V_BFM_B32 : VOP2_Real_e64only_vi <0x293>;
defm V_BCNT_U32_B32 : VOP2_Real_e64only_vi <0x28b>;
defm V_MBCNT_LO_U32_B32 : VOP2_Real_e64only_vi <0x28c>;
defm V_MBCNT_HI_U32_B32 : VOP2_Real_e64only_vi <0x28d>;
defm V_LDEXP_F32 : VOP2_Real_e64only_vi <0x288>;
defm V_CVT_PKACCUM_U8_F32 : VOP2_Real_e64only_vi <0x1f0>;
defm V_CVT_PKNORM_I16_F32 : VOP2_Real_e64only_vi <0x294>;
defm V_CVT_PKNORM_U16_F32 : VOP2_Real_e64only_vi <0x295>;
defm V_CVT_PKRTZ_F16_F32 : VOP2_Real_e64only_vi <0x296>;
defm V_CVT_PK_U16_U32 : VOP2_Real_e64only_vi <0x297>;
defm V_CVT_PK_I16_I32 : VOP2_Real_e64only_vi <0x298>;
defm V_ADD_F16 : VOP2_Real_e32e64_vi <0x1f>;
defm V_SUB_F16 : VOP2_Real_e32e64_vi <0x20>;
defm V_SUBREV_F16 : VOP2_Real_e32e64_vi <0x21>;
defm V_MUL_F16 : VOP2_Real_e32e64_vi <0x22>;
defm V_MAC_F16 : VOP2_Real_e32e64_vi <0x23>;
defm V_MADMK_F16 : VOP2_Real_MADK_vi <0x24>;
defm V_MADAK_F16 : VOP2_Real_MADK_vi <0x25>;
defm V_ADD_U16 : VOP2_Real_e32e64_vi <0x26>;
defm V_SUB_U16 : VOP2_Real_e32e64_vi <0x27>;
defm V_SUBREV_U16 : VOP2_Real_e32e64_vi <0x28>;
defm V_MUL_LO_U16 : VOP2_Real_e32e64_vi <0x29>;
defm V_LSHLREV_B16 : VOP2_Real_e32e64_vi <0x2a>;
defm V_LSHRREV_B16 : VOP2_Real_e32e64_vi <0x2b>;
defm V_ASHRREV_I16 : VOP2_Real_e32e64_vi <0x2c>;
defm V_MAX_F16 : VOP2_Real_e32e64_vi <0x2d>;
defm V_MIN_F16 : VOP2_Real_e32e64_vi <0x2e>;
defm V_MAX_U16 : VOP2_Real_e32e64_vi <0x2f>;
defm V_MAX_I16 : VOP2_Real_e32e64_vi <0x30>;
defm V_MIN_U16 : VOP2_Real_e32e64_vi <0x31>;
defm V_MIN_I16 : VOP2_Real_e32e64_vi <0x32>;
defm V_LDEXP_F16 : VOP2_Real_e32e64_vi <0x33>;
let SubtargetPredicate = isGFX8GFX9 in {
// Aliases to simplify matching of floating-point instructions that
// are VOP2 on SI and VOP3 on VI.
class SI2_VI3Alias <string name, VOP3_Real inst> : InstAlias <
name#" $dst, $src0, $src1",
!if(inst.Pfl.HasOMod,
(inst VGPR_32:$dst, 0, VCSrc_f32:$src0, 0, VCSrc_f32:$src1, 0, 0),
(inst VGPR_32:$dst, 0, VCSrc_f32:$src0, 0, VCSrc_f32:$src1, 0))
>, PredicateControl {
let UseInstAsmMatchConverter = 0;
let AsmVariantName = AMDGPUAsmVariants.VOP3;
}
def : SI2_VI3Alias <"v_ldexp_f32", V_LDEXP_F32_e64_vi>;
def : SI2_VI3Alias <"v_cvt_pkaccum_u8_f32", V_CVT_PKACCUM_U8_F32_e64_vi>;
def : SI2_VI3Alias <"v_cvt_pknorm_i16_f32", V_CVT_PKNORM_I16_F32_e64_vi>;
def : SI2_VI3Alias <"v_cvt_pknorm_u16_f32", V_CVT_PKNORM_U16_F32_e64_vi>;
def : SI2_VI3Alias <"v_cvt_pkrtz_f16_f32", V_CVT_PKRTZ_F16_F32_e64_vi>;
defm : VOP2eInstAliases<V_CNDMASK_B32_e32, V_CNDMASK_B32_e32_vi>;
} // End SubtargetPredicate = isGFX8GFX9
let SubtargetPredicate = isGFX9Only in {
defm : VOP2bInstAliases<V_ADD_I32_e32, V_ADD_CO_U32_e32_gfx9, "v_add_co_u32">;
defm : VOP2bInstAliases<V_ADDC_U32_e32, V_ADDC_CO_U32_e32_gfx9, "v_addc_co_u32">;
defm : VOP2bInstAliases<V_SUB_I32_e32, V_SUB_CO_U32_e32_gfx9, "v_sub_co_u32">;
defm : VOP2bInstAliases<V_SUBB_U32_e32, V_SUBB_CO_U32_e32_gfx9, "v_subb_co_u32">;
defm : VOP2bInstAliases<V_SUBREV_I32_e32, V_SUBREV_CO_U32_e32_gfx9, "v_subrev_co_u32">;
defm : VOP2bInstAliases<V_SUBBREV_U32_e32, V_SUBBREV_CO_U32_e32_gfx9, "v_subbrev_co_u32">;
} // End SubtargetPredicate = isGFX9Only
let SubtargetPredicate = HasDLInsts in {
defm V_FMAC_F32 : VOP2_Real_e32e64_vi <0x3b>;
defm V_XNOR_B32 : VOP2_Real_e32e64_vi <0x3d>;
} // End SubtargetPredicate = HasDLInsts
multiclass VOP2_Real_DOT_ACC_gfx9<bits<6> op> : VOP2_Real_e32_vi<op> {
def _dpp_vi : VOP2_DPP<op, !cast<VOP2_DPP_Pseudo>(NAME#"_dpp")>;
}
multiclass VOP2_Real_DOT_ACC_gfx10<bits<6> op> :
VOP2_Real_e32_gfx10<op>,
VOP2_Real_dpp_gfx10<op>,
VOP2_Real_dpp8_gfx10<op>;
let SubtargetPredicate = HasDot5Insts in {
defm V_DOT2C_F32_F16 : VOP2_Real_DOT_ACC_gfx9<0x37>;
// NB: Opcode conflicts with V_DOT8C_I32_I4
// This opcode exists in gfx 10.1* only
defm V_DOT2C_F32_F16 : VOP2_Real_DOT_ACC_gfx10<0x02>;
}
let SubtargetPredicate = HasDot6Insts in {
defm V_DOT4C_I32_I8 : VOP2_Real_DOT_ACC_gfx9<0x39>;
defm V_DOT4C_I32_I8 : VOP2_Real_DOT_ACC_gfx10<0x0d>;
}
let SubtargetPredicate = HasDot4Insts in {
defm V_DOT2C_I32_I16 : VOP2_Real_DOT_ACC_gfx9<0x38>;
}
let SubtargetPredicate = HasDot3Insts in {
defm V_DOT8C_I32_I4 : VOP2_Real_DOT_ACC_gfx9<0x3a>;
}
let SubtargetPredicate = HasPkFmacF16Inst in {
defm V_PK_FMAC_F16 : VOP2_Real_e32_vi<0x3c>;
} // End SubtargetPredicate = HasPkFmacF16Inst