[DAGCombiner] allow hoisting vector bitwise logic ahead of extends

llvm-svn: 350032

2 files changed, 48 insertions, 85 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 81bb73bf28e..93af1f4cc14 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -3731,13 +3731,12 @@ SDValue DAGCombiner::hoistLogicOpWithSameOpcodeHands(SDNode *N) {
     if (!N0.hasOneUse() && !N1.hasOneUse())
       return SDValue();
     // We need matching integer source types.
-    // Do not hoist logic op inside of a vector extend, since it may combine
-    // into a vsetcc.
-    // TODO: Should the vector check apply to truncate though?
-    if (VT.isVector() || XVT != Y.getValueType())
+    if (XVT != Y.getValueType())
       return SDValue();
-    // Don't create an illegal op during or after legalization.
-    if (LegalOperations && !TLI.isOperationLegal(LogicOpcode, XVT))
+    // Don't create an illegal op during or after legalization. Don't ever
+    // create an unsupported vector op.
+    if ((VT.isVector() || LegalOperations) &&
+        !TLI.isOperationLegalOrCustom(LogicOpcode, XVT))
       return SDValue();
     // Avoid infinite looping with PromoteIntBinOp.
     // TODO: Should we apply desirable/legal constraints to all opcodes?
diff --git a/llvm/test/CodeGen/X86/vector-ext-logic.ll b/llvm/test/CodeGen/X86/vector-ext-logic.ll
index 67b6f6c5af7..01c6c1a2fd5 100644
--- a/llvm/test/CodeGen/X86/vector-ext-logic.ll
+++ b/llvm/test/CodeGen/X86/vector-ext-logic.ll
@@ -5,23 +5,19 @@
 define <8 x i32> @zext_and_v8i32(<8 x i16> %x, <8 x i16> %y) {
 ; SSE2-LABEL: zext_and_v8i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
-; SSE2-NEXT:    pand %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    pand %xmm1, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
 ; SSE2-NEXT:    movdqa %xmm2, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; AVX2-LABEL: zext_and_v8i32:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %xz = zext <8 x i16> %x to <8 x i32>
   %yz = zext <8 x i16> %y to <8 x i32>
@@ -32,23 +28,19 @@ define <8 x i32> @zext_and_v8i32(<8 x i16> %x, <8 x i16> %y) {
 define <8 x i32> @zext_or_v8i32(<8 x i16> %x, <8 x i16> %y) {
 ; SSE2-LABEL: zext_or_v8i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm4 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; SSE2-NEXT:    por %xmm4, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    por %xmm1, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
 ; SSE2-NEXT:    movdqa %xmm2, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; AVX2-LABEL: zext_or_v8i32:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %xz = zext <8 x i16> %x to <8 x i32>
   %yz = zext <8 x i16> %y to <8 x i32>
@@ -59,23 +51,19 @@ define <8 x i32> @zext_or_v8i32(<8 x i16> %x, <8 x i16> %y) {
 define <8 x i32> @zext_xor_v8i32(<8 x i16> %x, <8 x i16> %y) {
 ; SSE2-LABEL: zext_xor_v8i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm4 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; SSE2-NEXT:    pxor %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
 ; SSE2-NEXT:    movdqa %xmm2, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; AVX2-LABEL: zext_xor_v8i32:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %xz = zext <8 x i16> %x to <8 x i32>
   %yz = zext <8 x i16> %y to <8 x i32>
@@ -86,24 +74,18 @@ define <8 x i32> @zext_xor_v8i32(<8 x i16> %x, <8 x i16> %y) {
 define <8 x i32> @sext_and_v8i32(<8 x i16> %x, <8 x i16> %y) {
 ; SSE2-LABEL: sext_and_v8i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    psrad $16, %xmm3
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
-; SSE2-NEXT:    psrad $16, %xmm4
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
+; SSE2-NEXT:    pand %xmm1, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
 ; SSE2-NEXT:    psrad $16, %xmm2
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    psrad $16, %xmm0
-; SSE2-NEXT:    pand %xmm4, %xmm0
-; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSE2-NEXT:    psrad $16, %xmm1
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX2-LABEL: sext_and_v8i32:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
-; AVX2-NEXT:    vpmovsxwd %xmm1, %ymm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %xs = sext <8 x i16> %x to <8 x i32>
   %ys = sext <8 x i16> %y to <8 x i32>
@@ -114,24 +96,18 @@ define <8 x i32> @sext_and_v8i32(<8 x i16> %x, <8 x i16> %y) {
 define <8 x i32> @sext_or_v8i32(<8 x i16> %x, <8 x i16> %y) {
 ; SSE2-LABEL: sext_or_v8i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    psrad $16, %xmm3
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
-; SSE2-NEXT:    psrad $16, %xmm4
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
 ; SSE2-NEXT:    psrad $16, %xmm2
-; SSE2-NEXT:    por %xmm3, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    psrad $16, %xmm0
-; SSE2-NEXT:    por %xmm4, %xmm0
-; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSE2-NEXT:    psrad $16, %xmm1
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX2-LABEL: sext_or_v8i32:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
-; AVX2-NEXT:    vpmovsxwd %xmm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %xs = sext <8 x i16> %x to <8 x i32>
   %ys = sext <8 x i16> %y to <8 x i32>
@@ -142,24 +118,18 @@ define <8 x i32> @sext_or_v8i32(<8 x i16> %x, <8 x i16> %y) {
 define <8 x i32> @sext_xor_v8i32(<8 x i16> %x, <8 x i16> %y) {
 ; SSE2-LABEL: sext_xor_v8i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    psrad $16, %xmm3
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
-; SSE2-NEXT:    psrad $16, %xmm4
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
+; SSE2-NEXT:    pxor %xmm1, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
 ; SSE2-NEXT:    psrad $16, %xmm2
-; SSE2-NEXT:    pxor %xmm3, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    psrad $16, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSE2-NEXT:    psrad $16, %xmm1
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX2-LABEL: sext_xor_v8i32:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
-; AVX2-NEXT:    vpmovsxwd %xmm1, %ymm1
-; AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %xs = sext <8 x i16> %x to <8 x i32>
   %ys = sext <8 x i16> %y to <8 x i32>
@@ -342,12 +312,9 @@ define <8 x i32> @bool_zext_or(<8 x i1> %x, <8 x i1> %y) {
 ;
 ; AVX2-LABEL: bool_zext_or:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
-; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %xz = zext <8 x i1> %x to <8 x i32>
   %yz = zext <8 x i1> %y to <8 x i32>
@@ -375,12 +342,9 @@ define <8 x i32> @bool_zext_xor(<8 x i1> %x, <8 x i1> %y) {
 ;
 ; AVX2-LABEL: bool_zext_xor:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
-; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %xz = zext <8 x i1> %x to <8 x i32>
   %yz = zext <8 x i1> %y to <8 x i32>