[X86][SSE] Add support for VZEXT constant folding

llvm-svn: 265646

4 files changed, 21 insertions, 6 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index a93eaa5e1e5..836caa9cb05 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -29609,11 +29609,29 @@ static SDValue combineVZext(SDNode *N, SelectionDAG &DAG,
                             const X86Subtarget &Subtarget) {
   SDLoc DL(N);
   MVT VT = N->getSimpleValueType(0);
+  MVT SVT = VT.getVectorElementType();
   SDValue Op = N->getOperand(0);
   MVT OpVT = Op.getSimpleValueType();
   MVT OpEltVT = OpVT.getVectorElementType();
   unsigned InputBits = OpEltVT.getSizeInBits() * VT.getVectorNumElements();
 
+  // Perform any constant folding.
+  if (ISD::isBuildVectorOfConstantSDNodes(Op.getNode())) {
+    SmallVector<SDValue, 4> Vals;
+    for (int i = 0, e = VT.getVectorNumElements(); i != e; ++i) {
+      SDValue OpElt = Op.getOperand(i);
+      if (OpElt.getOpcode() == ISD::UNDEF) {
+        Vals.push_back(DAG.getUNDEF(SVT));
+        continue;
+      }
+      APInt Cst = cast<ConstantSDNode>(OpElt.getNode())->getAPIntValue();
+      assert(Cst.getBitWidth() == OpEltVT.getSizeInBits());
+      Cst = Cst.zextOrTrunc(SVT.getSizeInBits());
+      Vals.push_back(DAG.getConstant(Cst, DL, SVT));
+    }
+    return DAG.getNode(ISD::BUILD_VECTOR, DL, VT, Vals);
+  }
+
   // (vzext (bitcast (vzext (x)) -> (vzext x)
   SDValue V = peekThroughBitcasts(Op);
   if (V != Op && V.getOpcode() == X86ISD::VZEXT) {
diff --git a/llvm/test/CodeGen/X86/vector-rotate-128.ll b/llvm/test/CodeGen/X86/vector-rotate-128.ll
index 9fbc82922c8..e3e7168fc59 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-128.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-128.ll
@@ -990,8 +990,7 @@ define <8 x i16> @constant_rotate_v8i16(<8 x i16> %a) nounwind {
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm1
 ; 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 {{.*#+}} ymm2 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
-; AVX2-NEXT:    vpsrlvd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
diff --git a/llvm/test/CodeGen/X86/vector-shift-ashr-128.ll b/llvm/test/CodeGen/X86/vector-shift-ashr-128.ll
index 4d9471f7640..3920dd9f240 100644
--- a/llvm/test/CodeGen/X86/vector-shift-ashr-128.ll
+++ b/llvm/test/CodeGen/X86/vector-shift-ashr-128.ll
@@ -1215,8 +1215,7 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) nounwind {
 ; AVX2-LABEL: constant_shift_v8i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpmovsxwd %xmm0, %ymm0
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
-; AVX2-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpsravd {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vzeroupper
diff --git a/llvm/test/CodeGen/X86/vector-shift-lshr-128.ll b/llvm/test/CodeGen/X86/vector-shift-lshr-128.ll
index 1106a7e93d8..bbe7887b091 100644
--- a/llvm/test/CodeGen/X86/vector-shift-lshr-128.ll
+++ b/llvm/test/CodeGen/X86/vector-shift-lshr-128.ll
@@ -949,8 +949,7 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) nounwind {
 ; AVX2-LABEL: constant_shift_v8i16:
 ; AVX2:       # BB#0:
 ; 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 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
-; AVX2-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vzeroupper