Skip to content

Commit

Permalink
Merge X86 SHUFPS and SHUFPD node types.
Browse files Browse the repository at this point in the history 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@147394 91177308-0d34-0410-b5e6-96231b3b80d8
  • Loading branch information
@topperc
topperc committed Dec 31, 2011
1 parent 3ee6d22 commit b3982da
Show file tree
Hide file tree
Showing 4 changed files with 35 additions and 58 deletions.
39 changes: 9 additions & 30 deletions lib/Target/X86/X86ISelLowering.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2869,9 +2869,8 @@ static bool isTargetShuffle(unsigned Opcode) {
case X86ISD::PSHUFD:
case X86ISD::PSHUFHW:
case X86ISD::PSHUFLW:
case X86ISD::SHUFPD:
case X86ISD::SHUFP:
case X86ISD::PALIGN:
case X86ISD::SHUFPS:
case X86ISD::MOVLHPS:
case X86ISD::MOVLHPD:
case X86ISD::MOVHLPS:
Expand Down Expand Up @@ -2923,8 +2922,7 @@ static SDValue getTargetShuffleNode(unsigned Opc, DebugLoc dl, EVT VT,
switch(Opc) {
default: llvm_unreachable("Unknown x86 shuffle node");
case X86ISD::PALIGN:
case X86ISD::SHUFPD:
case X86ISD::SHUFPS:
case X86ISD::SHUFP:
case X86ISD::VPERM2X128:
return DAG.getNode(Opc, dl, VT, V1, V2,
DAG.getConstant(TargetMask, MVT::i8));
Expand Down Expand Up @@ -4495,8 +4493,7 @@ static SDValue getShuffleScalarElt(SDNode *N, int Index, SelectionDAG &DAG,
SDValue ImmN;

switch(Opcode) {
case X86ISD::SHUFPS:
case X86ISD::SHUFPD:
case X86ISD::SHUFP:
ImmN = N->getOperand(N->getNumOperands()-1);
DecodeSHUFPMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(),
ShuffleMask);
Expand Down Expand Up @@ -6346,22 +6343,6 @@ SDValue getMOVHighToLow(SDValue &Op, DebugLoc &dl, SelectionDAG &DAG) {
return getTargetShuffleNode(X86ISD::MOVHLPS, dl, VT, V1, V2, DAG);
}

static inline unsigned getSHUFPOpcode(EVT VT) {
switch(VT.getSimpleVT().SimpleTy) {
case MVT::v8i32: // Use fp unit for int unpack.
case MVT::v8f32:
case MVT::v4i32: // Use fp unit for int unpack.
case MVT::v4f32: return X86ISD::SHUFPS;
case MVT::v4i64: // Use fp unit for int unpack.
case MVT::v4f64:
case MVT::v2i64: // Use fp unit for int unpack.
case MVT::v2f64: return X86ISD::SHUFPD;
default:
llvm_unreachable("Unknown type for shufp*");
}
return 0;
}

static
SDValue getMOVLP(SDValue &Op, DebugLoc &dl, SelectionDAG &DAG, bool HasXMMInt) {
SDValue V1 = Op.getOperand(0);
Expand Down Expand Up @@ -6415,7 +6396,7 @@ SDValue getMOVLP(SDValue &Op, DebugLoc &dl, SelectionDAG &DAG, bool HasXMMInt) {
assert(VT != MVT::v4i32 && "unsupported shuffle type");

// Invert the operand order and use SHUFPS to match it.
return getTargetShuffleNode(getSHUFPOpcode(VT), dl, VT, V2, V1,
return getTargetShuffleNode(X86ISD::SHUFP, dl, VT, V2, V1,
X86::getShuffleSHUFImmediate(SVOp), DAG);
}

Expand Down Expand Up @@ -6557,7 +6538,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
if (HasXMMInt && (VT == MVT::v4f32 || VT == MVT::v4i32))
return getTargetShuffleNode(X86ISD::PSHUFD, dl, VT, V1, TargetMask, DAG);

return getTargetShuffleNode(getSHUFPOpcode(VT), dl, VT, V1, V1,
return getTargetShuffleNode(X86ISD::SHUFP, dl, VT, V1, V1,
TargetMask, DAG);
}

Expand Down Expand Up @@ -6707,7 +6688,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
DAG);

if (isSHUFPMask(M, VT))
return getTargetShuffleNode(getSHUFPOpcode(VT), dl, VT, V1, V2,
return getTargetShuffleNode(X86ISD::SHUFP, dl, VT, V1, V2,
X86::getShuffleSHUFImmediate(SVOp), DAG);

if (isUNPCKL_v_undef_Mask(M, VT, HasAVX2))
Expand Down Expand Up @@ -6736,7 +6717,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {

// Handle VSHUFPS/DY permutations
if (isVSHUFPYMask(M, VT, HasAVX))
return getTargetShuffleNode(getSHUFPOpcode(VT), dl, VT, V1, V2,
return getTargetShuffleNode(X86ISD::SHUFP, dl, VT, V1, V2,
getShuffleVSHUFPYImmediate(SVOp), DAG);

//===--------------------------------------------------------------------===//
Expand Down Expand Up @@ -11031,8 +11012,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
case X86ISD::PSHUFHW_LD: return "X86ISD::PSHUFHW_LD";
case X86ISD::PSHUFLW: return "X86ISD::PSHUFLW";
case X86ISD::PSHUFLW_LD: return "X86ISD::PSHUFLW_LD";
case X86ISD::SHUFPS: return "X86ISD::SHUFPS";
case X86ISD::SHUFPD: return "X86ISD::SHUFPD";
case X86ISD::SHUFP: return "X86ISD::SHUFP";
case X86ISD::MOVLHPS: return "X86ISD::MOVLHPS";
case X86ISD::MOVLHPD: return "X86ISD::MOVLHPD";
case X86ISD::MOVHLPS: return "X86ISD::MOVHLPS";
Expand Down Expand Up @@ -14639,8 +14619,7 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
case X86ISD::VZEXT_MOVL: return PerformVZEXT_MOVLCombine(N, DAG);
case ISD::ZERO_EXTEND: return PerformZExtCombine(N, DAG);
case X86ISD::SETCC: return PerformSETCCCombine(N, DAG);
case X86ISD::SHUFPS: // Handle all target specific shuffles
case X86ISD::SHUFPD:
case X86ISD::SHUFP: // Handle all target specific shuffles
case X86ISD::PALIGN:
case X86ISD::UNPCKH:
case X86ISD::UNPCKL:
Expand Down
3 changes: 1 addition & 2 deletions lib/Target/X86/X86ISelLowering.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -258,8 +258,7 @@ namespace llvm {
PSHUFLW,
PSHUFHW_LD,
PSHUFLW_LD,
SHUFPD,
SHUFPS,
SHUFP,
MOVDDUP,
MOVSHDUP,
MOVSLDUP,
Expand Down
3 changes: 1 addition & 2 deletions lib/Target/X86/X86InstrFragmentsSIMD.td
View file
Original file line number Diff line number Diff line change
Expand Up @@ -112,8 +112,7 @@ def X86PShufd : SDNode<"X86ISD::PSHUFD", SDTShuff2OpI>;
def X86PShufhw : SDNode<"X86ISD::PSHUFHW", SDTShuff2OpI>;
def X86PShuflw : SDNode<"X86ISD::PSHUFLW", SDTShuff2OpI>;

def X86Shufpd : SDNode<"X86ISD::SHUFPD", SDTShuff3OpI>;
def X86Shufps : SDNode<"X86ISD::SHUFPS", SDTShuff3OpI>;
def X86Shufp : SDNode<"X86ISD::SHUFP", SDTShuff3OpI>;

def X86Movddup : SDNode<"X86ISD::MOVDDUP", SDTShuff1Op>;
def X86Movshdup : SDNode<"X86ISD::MOVSHDUP", SDTShuff1Op>;
Expand Down
48 changes: 24 additions & 24 deletions lib/Target/X86/X86InstrSSE.td
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2306,15 +2306,15 @@ let Constraints = "$src1 = $dst" in {
}

let Predicates = [HasSSE1] in {
def : Pat<(v4f32 (X86Shufps VR128:$src1,
def : Pat<(v4f32 (X86Shufp VR128:$src1,
(memopv4f32 addr:$src2), (i8 imm:$imm))),
(SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
def : Pat<(v4i32 (X86Shufps VR128:$src1,
def : Pat<(v4i32 (X86Shufp VR128:$src1,
(bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
(SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
// vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
// fall back to this for SSE1)
Expand Down Expand Up @@ -2348,28 +2348,28 @@ let Predicates = [HasSSE2] in {
(SHUFPDrri VR128:$src1, VR128:$src2,
(SHUFFLE_get_shuf_imm VR128:$src3))>;
// Generic SHUFPD patterns
def : Pat<(v2i64 (X86Shufpd VR128:$src1,
def : Pat<(v2i64 (X86Shufp VR128:$src1,
(memopv2i64 addr:$src2), (i8 imm:$imm))),
(SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v2f64 (X86Shufpd VR128:$src1,
def : Pat<(v2f64 (X86Shufp VR128:$src1,
(memopv2f64 addr:$src2), (i8 imm:$imm))),
(SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
}

let Predicates = [HasAVX] in {
def : Pat<(v4f32 (X86Shufps VR128:$src1,
def : Pat<(v4f32 (X86Shufp VR128:$src1,
(memopv4f32 addr:$src2), (i8 imm:$imm))),
(VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
def : Pat<(v4i32 (X86Shufps VR128:$src1,
def : Pat<(v4i32 (X86Shufp VR128:$src1,
(bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
(VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
// vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
// fall back to this for SSE1)
Expand Down Expand Up @@ -2400,39 +2400,39 @@ let Predicates = [HasAVX] in {
(VSHUFPDrri VR128:$src1, VR128:$src2,
(SHUFFLE_get_shuf_imm VR128:$src3))>;

def : Pat<(v2i64 (X86Shufpd VR128:$src1,
def : Pat<(v2i64 (X86Shufp VR128:$src1,
(memopv2i64 addr:$src2), (i8 imm:$imm))),
(VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v2f64 (X86Shufpd VR128:$src1,
def : Pat<(v2f64 (X86Shufp VR128:$src1,
(memopv2f64 addr:$src2), (i8 imm:$imm))),
(VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;

// 256-bit patterns
def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v8i32 (X86Shufps VR256:$src1,
def : Pat<(v8i32 (X86Shufp VR256:$src1,
(bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
(VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;

def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v8f32 (X86Shufps VR256:$src1,
def : Pat<(v8f32 (X86Shufp VR256:$src1,
(memopv8f32 addr:$src2), (i8 imm:$imm))),
(VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;

def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v4i64 (X86Shufpd VR256:$src1,
def : Pat<(v4i64 (X86Shufp VR256:$src1,
(memopv4i64 addr:$src2), (i8 imm:$imm))),
(VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;

def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v4f64 (X86Shufpd VR256:$src1,
def : Pat<(v4f64 (X86Shufp VR256:$src1,
(memopv4f64 addr:$src2), (i8 imm:$imm))),
(VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
}
Expand Down

0 comments on commit b3982da

Please sign in to comment.