td pattern). Currently we only model the immediate operand variation of
VPERMILPS and VPERMILPD, we should make that clear in the pseudos used.
Will be adding support for the variable mask variant in my next commit.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218282
91177308-0d34-0410-b5e6-
96231b3b80d8
case X86ISD::MOVSD:
case X86ISD::UNPCKL:
case X86ISD::UNPCKH:
case X86ISD::MOVSD:
case X86ISD::UNPCKL:
case X86ISD::UNPCKH:
+ case X86ISD::VPERMILPI:
case X86ISD::VPERM2X128:
case X86ISD::VPERMI:
return true;
case X86ISD::VPERM2X128:
case X86ISD::VPERMI:
return true;
case X86ISD::PSHUFD:
case X86ISD::PSHUFHW:
case X86ISD::PSHUFLW:
case X86ISD::PSHUFD:
case X86ISD::PSHUFHW:
case X86ISD::PSHUFLW:
+ case X86ISD::VPERMILPI:
case X86ISD::VPERMI:
return DAG.getNode(Opc, dl, VT, V1, DAG.getConstant(TargetMask, MVT::i8));
}
case X86ISD::VPERMI:
return DAG.getNode(Opc, dl, VT, V1, DAG.getConstant(TargetMask, MVT::i8));
}
DecodePALIGNRMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
break;
case X86ISD::PSHUFD:
DecodePALIGNRMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
break;
case X86ISD::PSHUFD:
+ case X86ISD::VPERMILPI:
ImmN = N->getOperand(N->getNumOperands()-1);
DecodePSHUFMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
IsUnary = true;
ImmN = N->getOperand(N->getNumOperands()-1);
DecodePSHUFMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
IsUnary = true;
if (Subtarget->hasAVX()) {
// If we have AVX, we can use VPERMILPS which will allow folding a load
// into the shuffle.
if (Subtarget->hasAVX()) {
// If we have AVX, we can use VPERMILPS which will allow folding a load
// into the shuffle.
- return DAG.getNode(X86ISD::VPERMILP, DL, MVT::v2f64, V1,
+ return DAG.getNode(X86ISD::VPERMILPI, DL, MVT::v2f64, V1,
DAG.getConstant(SHUFPDMask, MVT::i8));
}
DAG.getConstant(SHUFPDMask, MVT::i8));
}
if (Subtarget->hasAVX()) {
// If we have AVX, we can use VPERMILPS which will allow folding a load
// into the shuffle.
if (Subtarget->hasAVX()) {
// If we have AVX, we can use VPERMILPS which will allow folding a load
// into the shuffle.
- return DAG.getNode(X86ISD::VPERMILP, DL, MVT::v4f32, V1,
+ return DAG.getNode(X86ISD::VPERMILPI, DL, MVT::v4f32, V1,
getV4X86ShuffleImm8ForMask(Mask, DAG));
}
getV4X86ShuffleImm8ForMask(Mask, DAG));
}
// interleaved permutation.
unsigned VPERMILPMask = (Mask[0] == 1) | ((Mask[1] == 1) << 1) |
((Mask[2] == 3) << 2) | ((Mask[3] == 3) << 3);
// interleaved permutation.
unsigned VPERMILPMask = (Mask[0] == 1) | ((Mask[1] == 1) << 1) |
((Mask[2] == 3) << 2) | ((Mask[3] == 3) << 3);
- return DAG.getNode(X86ISD::VPERMILP, DL, MVT::v4f64, V1,
+ return DAG.getNode(X86ISD::VPERMILPI, DL, MVT::v4f64, V1,
DAG.getConstant(VPERMILPMask, MVT::i8));
}
DAG.getConstant(VPERMILPMask, MVT::i8));
}
if (is128BitLaneRepeatedShuffleMask(MVT::v8f32, Mask)) {
ArrayRef<int> LoMask = Mask.slice(0, 4);
if (isSingleInputShuffleMask(Mask))
if (is128BitLaneRepeatedShuffleMask(MVT::v8f32, Mask)) {
ArrayRef<int> LoMask = Mask.slice(0, 4);
if (isSingleInputShuffleMask(Mask))
- return DAG.getNode(X86ISD::VPERMILP, DL, MVT::v8f32, V1,
+ return DAG.getNode(X86ISD::VPERMILPI, DL, MVT::v8f32, V1,
getV4X86ShuffleImm8ForMask(LoMask, DAG));
// Use dedicated unpack instructions for masks that match their pattern.
getV4X86ShuffleImm8ForMask(LoMask, DAG));
// Use dedicated unpack instructions for masks that match their pattern.
M -= 4;
SDValue Lo = V1, Hi = V1;
if (!isNoopShuffleMask(LoMask))
M -= 4;
SDValue Lo = V1, Hi = V1;
if (!isNoopShuffleMask(LoMask))
- Lo = DAG.getNode(X86ISD::VPERMILP, DL, MVT::v8f32, Lo,
+ Lo = DAG.getNode(X86ISD::VPERMILPI, DL, MVT::v8f32, Lo,
getV4X86ShuffleImm8ForMask(LoMask, DAG));
if (!isNoopShuffleMask(HiMask))
getV4X86ShuffleImm8ForMask(LoMask, DAG));
if (!isNoopShuffleMask(HiMask))
- Hi = DAG.getNode(X86ISD::VPERMILP, DL, MVT::v8f32, Hi,
+ Hi = DAG.getNode(X86ISD::VPERMILPI, DL, MVT::v8f32, Hi,
getV4X86ShuffleImm8ForMask(HiMask, DAG));
unsigned BlendMask = 1 << 4 | 1 << 5 | 1 << 6 | 1 << 7;
return DAG.getNode(X86ISD::BLENDI, DL, MVT::v8f32, Lo, Hi,
getV4X86ShuffleImm8ForMask(HiMask, DAG));
unsigned BlendMask = 1 << 4 | 1 << 5 | 1 << 6 | 1 << 7;
return DAG.getNode(X86ISD::BLENDI, DL, MVT::v8f32, Lo, Hi,
return getTargetShuffleNode(X86ISD::PSHUFD, dl, VT, V1, TargetMask, DAG);
if (HasFp256 && (VT == MVT::v4f32 || VT == MVT::v2f64))
return getTargetShuffleNode(X86ISD::PSHUFD, dl, VT, V1, TargetMask, DAG);
if (HasFp256 && (VT == MVT::v4f32 || VT == MVT::v2f64))
- return getTargetShuffleNode(X86ISD::VPERMILP, dl, VT, V1, TargetMask,
+ return getTargetShuffleNode(X86ISD::VPERMILPI, dl, VT, V1, TargetMask,
DAG);
return getTargetShuffleNode(X86ISD::SHUFP, dl, VT, V1, V1,
DAG);
return getTargetShuffleNode(X86ISD::SHUFP, dl, VT, V1, V1,
if ((HasInt256 && VT == MVT::v8i32) || VT == MVT::v16i32)
return getTargetShuffleNode(X86ISD::PSHUFD, dl, VT, V1,
getShuffleSHUFImmediate(SVOp), DAG);
if ((HasInt256 && VT == MVT::v8i32) || VT == MVT::v16i32)
return getTargetShuffleNode(X86ISD::PSHUFD, dl, VT, V1,
getShuffleSHUFImmediate(SVOp), DAG);
- return getTargetShuffleNode(X86ISD::VPERMILP, dl, VT, V1,
+ return getTargetShuffleNode(X86ISD::VPERMILPI, dl, VT, V1,
getShuffleSHUFImmediate(SVOp), DAG);
}
getShuffleSHUFImmediate(SVOp), DAG);
}
case X86ISD::VBROADCAST: return "X86ISD::VBROADCAST";
case X86ISD::VBROADCASTM: return "X86ISD::VBROADCASTM";
case X86ISD::VEXTRACT: return "X86ISD::VEXTRACT";
case X86ISD::VBROADCAST: return "X86ISD::VBROADCAST";
case X86ISD::VBROADCASTM: return "X86ISD::VBROADCASTM";
case X86ISD::VEXTRACT: return "X86ISD::VEXTRACT";
- case X86ISD::VPERMILP: return "X86ISD::VPERMILP";
+ case X86ISD::VPERMILPI: return "X86ISD::VPERMILPI";
case X86ISD::VPERM2X128: return "X86ISD::VPERM2X128";
case X86ISD::VPERMV: return "X86ISD::VPERMV";
case X86ISD::VPERMV3: return "X86ISD::VPERMV3";
case X86ISD::VPERM2X128: return "X86ISD::VPERM2X128";
case X86ISD::VPERMV: return "X86ISD::VPERMV";
case X86ISD::VPERMV3: return "X86ISD::VPERMV3";
case X86ISD::PSHUFLW:
case X86ISD::MOVSS:
case X86ISD::MOVSD:
case X86ISD::PSHUFLW:
case X86ISD::MOVSS:
case X86ISD::MOVSD:
+ case X86ISD::VPERMILPI:
case X86ISD::VPERM2X128:
case ISD::VECTOR_SHUFFLE: return PerformShuffleCombine(N, DAG, DCI,Subtarget);
case ISD::FMA: return PerformFMACombine(N, DAG, Subtarget);
case X86ISD::VPERM2X128:
case ISD::VECTOR_SHUFFLE: return PerformShuffleCombine(N, DAG, DCI,Subtarget);
case ISD::FMA: return PerformFMACombine(N, DAG, Subtarget);
VPERMV,
VPERMV3,
VPERMIV3,
VPERMV,
VPERMV3,
VPERMIV3,
i512mem, v16i32>, PD, EVEX_V512, EVEX_CD8<32, CD8VF>;
let ExeDomain = SSEPackedSingle in
i512mem, v16i32>, PD, EVEX_V512, EVEX_CD8<32, CD8VF>;
let ExeDomain = SSEPackedSingle in
-defm VPERMILPSZ : avx512_pshuf_imm<0x04, "vpermilps", VR512, X86VPermilp,
+defm VPERMILPSZ : avx512_pshuf_imm<0x04, "vpermilps", VR512, X86VPermilpi,
memopv16f32, i512mem, v16f32>, TAPD, EVEX_V512,
EVEX_CD8<32, CD8VF>;
let ExeDomain = SSEPackedDouble in
memopv16f32, i512mem, v16f32>, TAPD, EVEX_V512,
EVEX_CD8<32, CD8VF>;
let ExeDomain = SSEPackedDouble in
-defm VPERMILPDZ : avx512_pshuf_imm<0x05, "vpermilpd", VR512, X86VPermilp,
+defm VPERMILPDZ : avx512_pshuf_imm<0x05, "vpermilpd", VR512, X86VPermilpi,
memopv8f64, i512mem, v8f64>, TAPD, EVEX_V512,
VEX_W, EVEX_CD8<32, CD8VF>;
memopv8f64, i512mem, v8f64>, TAPD, EVEX_V512,
VEX_W, EVEX_CD8<32, CD8VF>;
-def : Pat<(v16i32 (X86VPermilp VR512:$src1, (i8 imm:$imm))),
+def : Pat<(v16i32 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
(VPERMILPSZri VR512:$src1, imm:$imm)>;
(VPERMILPSZri VR512:$src1, imm:$imm)>;
-def : Pat<(v8i64 (X86VPermilp VR512:$src1, (i8 imm:$imm))),
+def : Pat<(v8i64 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
(VPERMILPDZri VR512:$src1, imm:$imm)>;
//===----------------------------------------------------------------------===//
(VPERMILPDZri VR512:$src1, imm:$imm)>;
//===----------------------------------------------------------------------===//
def X86Unpckl : SDNode<"X86ISD::UNPCKL", SDTShuff2Op>;
def X86Unpckh : SDNode<"X86ISD::UNPCKH", SDTShuff2Op>;
def X86Unpckl : SDNode<"X86ISD::UNPCKL", SDTShuff2Op>;
def X86Unpckh : SDNode<"X86ISD::UNPCKH", SDTShuff2Op>;
-def X86VPermilp : SDNode<"X86ISD::VPERMILP", SDTShuff2OpI>;
-def X86VPermv : SDNode<"X86ISD::VPERMV", SDTShuff2Op>;
-def X86VPermi : SDNode<"X86ISD::VPERMI", SDTShuff2OpI>;
-def X86VPermv3 : SDNode<"X86ISD::VPERMV3", SDTShuff3Op>;
+def X86VPermilpi : SDNode<"X86ISD::VPERMILPI", SDTShuff2OpI>;
+def X86VPermv : SDNode<"X86ISD::VPERMV", SDTShuff2Op>;
+def X86VPermi : SDNode<"X86ISD::VPERMI", SDTShuff2OpI>;
+def X86VPermv3 : SDNode<"X86ISD::VPERMV3", SDTShuff3Op>;
def X86VPermiv3 : SDNode<"X86ISD::VPERMIV3", SDTShuff3Op>;
def X86VPerm2x128 : SDNode<"X86ISD::VPERM2X128", SDTShuff3OpI>;
def X86VPermiv3 : SDNode<"X86ISD::VPERMIV3", SDTShuff3Op>;
def X86VPerm2x128 : SDNode<"X86ISD::VPERM2X128", SDTShuff3OpI>;
def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX,
+ [(set RC:$dst, (vt (X86VPermilpi RC:$src1, (i8 imm:$src2))))]>, VEX,
Sched<[WriteFShuffle]>;
def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
(ins x86memop_f:$src1, i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
Sched<[WriteFShuffle]>;
def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
(ins x86memop_f:$src1, i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
- (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX,
+ (vt (X86VPermilpi (memop addr:$src1), (i8 imm:$src2))))]>, VEX,
Sched<[WriteFShuffleLd]>;
}
Sched<[WriteFShuffleLd]>;
}
}
let Predicates = [HasAVX] in {
}
let Predicates = [HasAVX] in {
-def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
+def : Pat<(v8i32 (X86VPermilpi VR256:$src1, (i8 imm:$imm))),
(VPERMILPSYri VR256:$src1, imm:$imm)>;
(VPERMILPSYri VR256:$src1, imm:$imm)>;
-def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
+def : Pat<(v4i64 (X86VPermilpi VR256:$src1, (i8 imm:$imm))),
(VPERMILPDYri VR256:$src1, imm:$imm)>;
(VPERMILPDYri VR256:$src1, imm:$imm)>;
-def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (loadv4i64 addr:$src1)),
+def : Pat<(v8i32 (X86VPermilpi (bc_v8i32 (loadv4i64 addr:$src1)),
(i8 imm:$imm))),
(VPERMILPSYmi addr:$src1, imm:$imm)>;
(i8 imm:$imm))),
(VPERMILPSYmi addr:$src1, imm:$imm)>;
-def : Pat<(v4i64 (X86VPermilp (loadv4i64 addr:$src1), (i8 imm:$imm))),
+def : Pat<(v4i64 (X86VPermilpi (loadv4i64 addr:$src1), (i8 imm:$imm))),
(VPERMILPDYmi addr:$src1, imm:$imm)>;
(VPERMILPDYmi addr:$src1, imm:$imm)>;
-def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
+def : Pat<(v2i64 (X86VPermilpi VR128:$src1, (i8 imm:$imm))),
(VPERMILPDri VR128:$src1, imm:$imm)>;
(VPERMILPDri VR128:$src1, imm:$imm)>;
-def : Pat<(v2i64 (X86VPermilp (loadv2i64 addr:$src1), (i8 imm:$imm))),
+def : Pat<(v2i64 (X86VPermilpi (loadv2i64 addr:$src1), (i8 imm:$imm))),
(VPERMILPDmi addr:$src1, imm:$imm)>;
}
(VPERMILPDmi addr:$src1, imm:$imm)>;
}