[X86][SSE] Fold SIGN_EXTEND(SIGN_EXTEND_VECTOR_INREG(X)) -> SIGN_EXTEND_VECTOR_INREG(X)

It should be possible to make this generic, but we're not great at checking legality of *_EXTEND_VECTOR_INREG ops so I'm conservatively putting this inside X86ISelLowering.cpp

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 26333c6..f032758 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -46769,10 +46769,14 @@ static SDValue combineSext(SDNode *N, SelectionDAG &DAG,
   if (SDValue V = combineToExtendBoolVectorInReg(N, DAG, DCI, Subtarget))
     return V;
 
-  if (VT.isVector())
+  if (VT.isVector()) {
     if (SDValue R = PromoteMaskArithmetic(N, DAG, Subtarget))
       return R;
 
+    if (N0.getOpcode() == ISD::SIGN_EXTEND_VECTOR_INREG)
+      return DAG.getNode(N0.getOpcode(), DL, VT, N0.getOperand(0));
+  }
+
   if (SDValue NewAdd = promoteExtBeforeAdd(N, DAG, Subtarget))
     return NewAdd;
 

diff --git a/llvm/test/CodeGen/X86/masked_load.ll b/llvm/test/CodeGen/X86/masked_load.ll
index 3a63608..c0b274d 100644 (file)
--- a/llvm/test/CodeGen/X86/masked_load.ll
+++ b/llvm/test/CodeGen/X86/masked_load.ll
@@ -462,9 +462,8 @@ define <8 x double> @load_v8f64_v8i16(<8 x i16> %trigger, <8 x double>* %addr, <
 ; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpmovsxwq %xmm3, %xmm5
-; AVX1-NEXT:    vpmovsxwd %xmm3, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,2,3]
-; AVX1-NEXT:    vpmovsxdq %xmm3, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,1,1]
+; AVX1-NEXT:    vpmovsxwq %xmm3, %xmm3
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
 ; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovsxwq %xmm0, %xmm4
@@ -482,8 +481,7 @@ define <8 x double> @load_v8f64_v8i16(<8 x i16> %trigger, <8 x double>* %addr, <
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpcmpeqw %xmm4, %xmm3, %xmm3
-; AVX2-NEXT:    vpmovsxwd %xmm3, %xmm3
-; AVX2-NEXT:    vpmovsxdq %xmm3, %ymm3
+; AVX2-NEXT:    vpmovsxwq %xmm3, %ymm3
 ; AVX2-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovsxwq %xmm0, %ymm0
 ; AVX2-NEXT:    vmaskmovpd (%rdi), %ymm0, %ymm4
@@ -1782,9 +1780,8 @@ define <8 x i64> @load_v8i64_v8i16(<8 x i16> %trigger, <8 x i64>* %addr, <8 x i6
 ; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpmovsxwq %xmm3, %xmm5
-; AVX1-NEXT:    vpmovsxwd %xmm3, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,2,3]
-; AVX1-NEXT:    vpmovsxdq %xmm3, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,1,1]
+; AVX1-NEXT:    vpmovsxwq %xmm3, %xmm3
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
 ; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovsxwq %xmm0, %xmm4
@@ -1802,8 +1799,7 @@ define <8 x i64> @load_v8i64_v8i16(<8 x i16> %trigger, <8 x i64>* %addr, <8 x i6
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpcmpeqw %xmm4, %xmm3, %xmm3
-; AVX2-NEXT:    vpmovsxwd %xmm3, %xmm3
-; AVX2-NEXT:    vpmovsxdq %xmm3, %ymm3
+; AVX2-NEXT:    vpmovsxwq %xmm3, %ymm3
 ; AVX2-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovsxwq %xmm0, %ymm0
 ; AVX2-NEXT:    vpmaskmovq (%rdi), %ymm0, %ymm4