Merge vpermps/vpermd and vpermpd/vpermq SD nodes.

llvm-svn: 154782
This commit is contained in:
Craig Topper 2012-04-16 00:41:45 +00:00
parent b04fe34030
commit b86fa404d3
4 changed files with 20 additions and 31 deletions

View File

@ -2935,8 +2935,7 @@ static SDValue getTargetShuffleNode(unsigned Opc, DebugLoc dl, EVT VT,
case X86ISD::PSHUFHW:
case X86ISD::PSHUFLW:
case X86ISD::VPERMILP:
case X86ISD::VPERMQ:
case X86ISD::VPERMPD:
case X86ISD::VPERMI:
return DAG.getNode(Opc, dl, VT, V1, DAG.getConstant(TargetMask, MVT::i8));
}
}
@ -6656,16 +6655,14 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
for (unsigned i = 0; i != 8; ++i) {
permclMask.push_back(DAG.getConstant((M[i]>=0) ? M[i] : 0, MVT::i32));
}
return DAG.getNode(VT.isInteger()? X86ISD::VPERMD:X86ISD::VPERMPS, dl, VT,
return DAG.getNode(X86ISD::VPERMV, dl, VT,
DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32,
&permclMask[0], 8), V1);
}
if (V2IsUndef && HasAVX2 && (VT == MVT::v4i64 || VT == MVT::v4f64)) {
unsigned Opcode = VT.isInteger() ? X86ISD::VPERMQ : X86ISD::VPERMPD;
return getTargetShuffleNode(Opcode, dl, VT, V1,
if (V2IsUndef && HasAVX2 && (VT == MVT::v4i64 || VT == MVT::v4f64))
return getTargetShuffleNode(X86ISD::VPERMI, dl, VT, V1,
getShuffleCLImmediate(SVOp), DAG);
}
//===--------------------------------------------------------------------===//
@ -11181,10 +11178,8 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
case X86ISD::VBROADCAST: return "X86ISD::VBROADCAST";
case X86ISD::VPERMILP: return "X86ISD::VPERMILP";
case X86ISD::VPERM2X128: return "X86ISD::VPERM2X128";
case X86ISD::VPERMD: return "X86ISD::VPERMD";
case X86ISD::VPERMQ: return "X86ISD::VPERMQ";
case X86ISD::VPERMPS: return "X86ISD::VPERMPS";
case X86ISD::VPERMPD: return "X86ISD::VPERMPD";
case X86ISD::VPERMV: return "X86ISD::VPERMV";
case X86ISD::VPERMI: return "X86ISD::VPERMI";
case X86ISD::PMULUDQ: return "X86ISD::PMULUDQ";
case X86ISD::VASTART_SAVE_XMM_REGS: return "X86ISD::VASTART_SAVE_XMM_REGS";
case X86ISD::VAARG_64: return "X86ISD::VAARG_64";

View File

@ -285,10 +285,8 @@ namespace llvm {
UNPCKL,
UNPCKH,
VPERMILP,
VPERMD,
VPERMQ,
VPERMPS,
VPERMPD,
VPERMV,
VPERMI,
VPERM2X128,
VBROADCAST,

View File

@ -157,10 +157,8 @@ def X86Unpckl : SDNode<"X86ISD::UNPCKL", SDTShuff2Op>;
def X86Unpckh : SDNode<"X86ISD::UNPCKH", SDTShuff2Op>;
def X86VPermilp : SDNode<"X86ISD::VPERMILP", SDTShuff2OpI>;
def X86VPermd : SDNode<"X86ISD::VPERMD", SDTVPermv>;
def X86VPermps : SDNode<"X86ISD::VPERMPS", SDTVPermv>;
def X86VPermq : SDNode<"X86ISD::VPERMQ", SDTShuff2OpI>;
def X86VPermpd : SDNode<"X86ISD::VPERMPD", SDTShuff2OpI>;
def X86VPermv : SDNode<"X86ISD::VPERMV", SDTVPermv>;
def X86VPermi : SDNode<"X86ISD::VPERMI", SDTShuff2OpI>;
def X86VPerm2x128 : SDNode<"X86ISD::VPERM2X128", SDTShuff3OpI>;

View File

@ -7755,37 +7755,35 @@ let ExeDomain = SSEPackedSingle in
defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
SDNode OpNode, ValueType OpVT> {
ValueType OpVT> {
def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(OpVT (OpNode VR256:$src1, (i8 imm:$src2))))]>, VEX;
(OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
(ins i256mem:$src1, i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(OpVT (OpNode (mem_frag addr:$src1), (i8 imm:$src2))))]>,
VEX;
(OpVT (X86VPermi (mem_frag addr:$src1),
(i8 imm:$src2))))]>, VEX;
}
defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, X86VPermq, v4i64>,
VEX_W;
defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
let ExeDomain = SSEPackedDouble in
defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, X86VPermpd, v4f64>,
VEX_W;
defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
let Predicates = [HasAVX2] in {
def : Pat<(v8i32 (X86VPermd VR256:$src1, VR256:$src2)),
def : Pat<(v8i32 (X86VPermv VR256:$src1, VR256:$src2)),
(VPERMDYrr VR256:$src1, VR256:$src2)>;
def : Pat<(v8f32 (X86VPermps VR256:$src1, VR256:$src2)),
def : Pat<(v8f32 (X86VPermv VR256:$src1, VR256:$src2)),
(VPERMPSYrr VR256:$src1, VR256:$src2)>;
def : Pat<(v8i32 (X86VPermd VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
def : Pat<(v8i32 (X86VPermv VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
(VPERMDYrm VR256:$src1, addr:$src2)>;
def : Pat<(v8f32 (X86VPermps VR256:$src1, (memopv8f32 addr:$src2))),
def : Pat<(v8f32 (X86VPermv VR256:$src1, (memopv8f32 addr:$src2))),
(VPERMPSYrm VR256:$src1, addr:$src2)>;
}