[X86] Teach how to combine a vselect into a movss/movsd

Add target specific rules for combining vselect dag nodes into movss/movsd
when possible.

If the vector type of the vselect dag node in input is either MVT::v4i13 or
MVT::v4f32, then try to fold according to rules:

  1) fold (vselect (build_vector (0, -1, -1, -1)), A, B) -> (movss A, B)
  2) fold (vselect (build_vector (-1, 0, 0, 0)), A, B) -> (movss B, A)

If the vector type of the vselect dag node in input is either MVT::v2i64 or
MVT::v2f64 (and we have SSE2), then try to fold according to rules:

  3) fold (vselect (build_vector (0, -1)), A, B) -> (movsd A, B)
  4) fold (vselect (build_vector (-1, 0)), A, B) -> (movsd B, A)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199683 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 195f11a82641f2ab42bf846517da7847d1c2f864..82ac9f90826113e72cd6d02b0e25322eee8368ef 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -17155,6 +17155,41 @@ static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG,
     }
   }
 
+  // Try to fold this VSELECT into a MOVSS/MOVSD
+  if (N->getOpcode() == ISD::VSELECT &&
+      Cond.getOpcode() == ISD::BUILD_VECTOR && !DCI.isBeforeLegalize()) {
+    if (VT == MVT::v4i32 || VT == MVT::v4f32 ||
+        (Subtarget->hasSSE2() && (VT == MVT::v2i64 || VT == MVT::v2f64))) {
+      bool CanFold = false;
+      unsigned NumElems = Cond.getNumOperands();
+      SDValue A = LHS;
+      SDValue B = RHS;
+      
+      if (isZero(Cond.getOperand(0))) {
+        CanFold = true;
+
+        // fold (vselect <0,-1,-1,-1>, A, B) -> (movss A, B)
+        // fold (vselect <0,-1> -> (movsd A, B)
+        for (unsigned i = 1, e = NumElems; i != e && CanFold; ++i)
+          CanFold = isAllOnes(Cond.getOperand(i));
+      } else if (isAllOnes(Cond.getOperand(0))) {
+        CanFold = true;
+        std::swap(A, B);
+
+        // fold (vselect <-1,0,0,0>, A, B) -> (movss B, A)
+        // fold (vselect <-1,0> -> (movsd B, A)
+        for (unsigned i = 1, e = NumElems; i != e && CanFold; ++i)
+          CanFold = isZero(Cond.getOperand(i));
+      }
+
+      if (CanFold) {
+        if (VT == MVT::v4i32 || VT == MVT::v4f32)
+          return getTargetShuffleNode(X86ISD::MOVSS, DL, VT, A, B, DAG);
+        return getTargetShuffleNode(X86ISD::MOVSD, DL, VT, A, B, DAG);
+      }
+    }
+  }
+
   // If we know that this node is legal then we know that it is going to be
   // matched by one of the SSE/AVX BLEND instructions. These instructions only
   // depend on the highest bit in each word. Try to use SimplifyDemandedBits

diff --git a/test/CodeGen/X86/avx-blend.ll b/test/CodeGen/X86/avx-blend.ll
index a98e0761ce3131c5f075176e3801fc0411c46554..e9bfce663f6af8b355a8ba78fa2094f95c3f2fbb 100644 (file)
--- a/test/CodeGen/X86/avx-blend.ll
+++ b/test/CodeGen/X86/avx-blend.ll
@@ -6,7 +6,7 @@
 ;CHECK: vblendvps
 ;CHECK: ret
 define <4 x float> @vsel_float(<4 x float> %v1, <4 x float> %v2) {
-  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x float> %v1, <4 x float> %v2
+  %vsel = select <4 x i1> <i1 true, i1 false, i1 true, i1 false>, <4 x float> %v1, <4 x float> %v2
   ret <4 x float> %vsel
 }
 
@@ -15,13 +15,13 @@ define <4 x float> @vsel_float(<4 x float> %v1, <4 x float> %v2) {
 ;CHECK: vblendvps
 ;CHECK: ret
 define <4 x i32> @vsel_i32(<4 x i32> %v1, <4 x i32> %v2) {
-  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x i32> %v1, <4 x i32> %v2
+  %vsel = select <4 x i1> <i1 true, i1 false, i1 true, i1 false>, <4 x i32> %v1, <4 x i32> %v2
   ret <4 x i32> %vsel
 }
 
 
 ;CHECK-LABEL: vsel_double:
-;CHECK: vblendvpd
+;CHECK: vmovsd
 ;CHECK: ret
 define <2 x double> @vsel_double(<2 x double> %v1, <2 x double> %v2) {
   %vsel = select <2 x i1> <i1 true, i1 false>, <2 x double> %v1, <2 x double> %v2
@@ -30,7 +30,7 @@ define <2 x double> @vsel_double(<2 x double> %v1, <2 x double> %v2) {
 
 
 ;CHECK-LABEL: vsel_i64:
-;CHECK: vblendvpd
+;CHECK: vmovsd
 ;CHECK: ret
 define <2 x i64> @vsel_i64(<2 x i64> %v1, <2 x i64> %v2) {
   %vsel = select <2 x i1> <i1 true, i1 false>, <2 x i64> %v1, <2 x i64> %v2

diff --git a/test/CodeGen/X86/blend-msb.ll b/test/CodeGen/X86/blend-msb.ll
index 0485a42eb7e5dd89110b070e9c83b8e3bc319ede..6b465963292e7c56f9f7ad670c47e3b0be539d4d 100644 (file)
--- a/test/CodeGen/X86/blend-msb.ll
+++ b/test/CodeGen/X86/blend-msb.ll
@@ -1,13 +1,11 @@
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7 -mattr=+sse4.1 | FileCheck %s
 
 
-; In this test we check that sign-extend of the mask bit is performed by
-; shifting the needed bit to the MSB, and not using shl+sra.
+; Verify that we produce movss instead of blendvps when possible.
 
 ;CHECK-LABEL: vsel_float:
-;CHECK: movl $-1
-;CHECK-NEXT: movd
-;CHECK-NEXT: blendvps
+;CHECK-NOT: blendvps
+;CHECK: movss
 ;CHECK: ret
 define <4 x float> @vsel_float(<4 x float> %v1, <4 x float> %v2) {
   %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x float> %v1, <4 x float> %v2
@@ -15,9 +13,8 @@ define <4 x float> @vsel_float(<4 x float> %v1, <4 x float> %v2) {
 }
 
 ;CHECK-LABEL: vsel_4xi8:
-;CHECK: movl $-1
-;CHECK-NEXT: movd
-;CHECK-NEXT: blendvps
+;CHECK-NOT: blendvps
+;CHECK: movss
 ;CHECK: ret
 define <4 x i8> @vsel_4xi8(<4 x i8> %v1, <4 x i8> %v2) {
   %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x i8> %v1, <4 x i8> %v2

diff --git a/test/CodeGen/X86/sse-scalar-fp-arith-2.ll b/test/CodeGen/X86/sse-scalar-fp-arith-2.ll
index 59685993f5d4d76da6acc700d03803adfba3a3ee..600ee1b7b1e59e46114a3ad329a817ef5b8398ed 100644 (file)
--- a/test/CodeGen/X86/sse-scalar-fp-arith-2.ll
+++ b/test/CodeGen/X86/sse-scalar-fp-arith-2.ll
@@ -213,3 +213,211 @@ define <2 x double> @test2_div_sd(<2 x double> %a, <2 x double> %b) {
 ; CHECK-NOT: movsd
 ; CHECK: ret
 
+
+define <4 x float> @test3_add_ss(<4 x float> %a, <4 x float> %b) {
+  %1 = fadd <4 x float> %a, %b
+  %2 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %a, <4 x float> %1
+  ret <4 x float> %2
+}
+
+; CHECK-LABEL: test3_add_ss
+; SSE2: addss   %xmm1, %xmm0
+; AVX: vaddss   %xmm1, %xmm0, %xmm0
+; CHECK-NOT: movss
+; CHECK: ret
+
+
+define <4 x float> @test3_sub_ss(<4 x float> %a, <4 x float> %b) {
+  %1 = fsub <4 x float> %a, %b
+  %2 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %a, <4 x float> %1
+  ret <4 x float> %2
+}
+
+; CHECK-LABEL: test3_sub_ss
+; SSE2: subss   %xmm1, %xmm0
+; AVX: vsubss   %xmm1, %xmm0, %xmm0
+; CHECK-NOT: movss
+; CHECK: ret
+
+
+define <4 x float> @test3_mul_ss(<4 x float> %a, <4 x float> %b) {
+  %1 = fmul <4 x float> %a, %b
+  %2 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %a, <4 x float> %1
+  ret <4 x float> %2
+}
+
+; CHECK-LABEL: test3_mul_ss
+; SSE2: mulss   %xmm1, %xmm0
+; AVX: vmulss   %xmm1, %xmm0, %xmm0
+; CHECK-NOT: movss
+; CHECK: ret
+
+
+define <4 x float> @test3_div_ss(<4 x float> %a, <4 x float> %b) {
+  %1 = fdiv <4 x float> %a, %b
+  %2 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %a, <4 x float> %1
+  ret <4 x float> %2
+}
+
+; CHECK-LABEL: test3_div_ss
+; SSE2: divss   %xmm1, %xmm0
+; AVX: vdivss   %xmm1, %xmm0, %xmm0
+; CHECK-NOT: movss
+; CHECK: ret
+
+
+define <2 x double> @test3_add_sd(<2 x double> %a, <2 x double> %b) {
+  %1 = fadd <2 x double> %a, %b
+  %2 = select <2 x i1> <i1 false, i1 true>, <2 x double> %a, <2 x double> %1
+  ret <2 x double> %2
+}
+
+; CHECK-LABEL: test3_add_sd
+; SSE2: addsd   %xmm1, %xmm0
+; AVX: vaddsd   %xmm1, %xmm0, %xmm0
+; CHECK-NOT: movsd
+; CHECK: ret
+
+
+define <2 x double> @test3_sub_sd(<2 x double> %a, <2 x double> %b) {
+  %1 = fsub <2 x double> %a, %b
+  %2 = select <2 x i1> <i1 false, i1 true>, <2 x double> %a, <2 x double> %1
+  ret <2 x double> %2
+}
+
+; CHECK-LABEL: test3_sub_sd
+; SSE2: subsd   %xmm1, %xmm0
+; AVX: vsubsd   %xmm1, %xmm0, %xmm0
+; CHECK-NOT: movsd
+; CHECK: ret
+
+
+define <2 x double> @test3_mul_sd(<2 x double> %a, <2 x double> %b) {
+  %1 = fmul <2 x double> %a, %b
+  %2 = select <2 x i1> <i1 false, i1 true>, <2 x double> %a, <2 x double> %1
+  ret <2 x double> %2
+}
+
+; CHECK-LABEL: test3_mul_sd
+; SSE2: mulsd   %xmm1, %xmm0
+; AVX: vmulsd   %xmm1, %xmm0, %xmm0
+; CHECK-NOT: movsd
+; CHECK: ret
+
+
+define <2 x double> @test3_div_sd(<2 x double> %a, <2 x double> %b) {
+  %1 = fdiv <2 x double> %a, %b
+  %2 = select <2 x i1> <i1 false, i1 true>, <2 x double> %a, <2 x double> %1
+  ret <2 x double> %2
+}
+
+; CHECK-LABEL: test3_div_sd
+; SSE2: divsd   %xmm1, %xmm0
+; AVX: vdivsd   %xmm1, %xmm0, %xmm0
+; CHECK-NOT: movsd
+; CHECK: ret
+
+
+define <4 x float> @test4_add_ss(<4 x float> %a, <4 x float> %b) {
+  %1 = fadd <4 x float> %b, %a
+  %2 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %b, <4 x float> %1
+  ret <4 x float> %2
+}
+
+; CHECK-LABEL: test4_add_ss
+; SSE2: addss   %xmm0, %xmm1
+; AVX: vaddss   %xmm0, %xmm1, %xmm0
+; CHECK-NOT: movss
+; CHECK: ret
+
+
+define <4 x float> @test4_sub_ss(<4 x float> %a, <4 x float> %b) {
+  %1 = fsub <4 x float> %b, %a
+  %2 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %b, <4 x float> %1
+  ret <4 x float> %2
+}
+
+; CHECK-LABEL: test4_sub_ss
+; SSE2: subss   %xmm0, %xmm1
+; AVX: vsubss   %xmm0, %xmm1, %xmm0
+; CHECK-NOT: movss
+; CHECK: ret
+
+
+define <4 x float> @test4_mul_ss(<4 x float> %a, <4 x float> %b) {
+  %1 = fmul <4 x float> %b, %a
+  %2 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %b, <4 x float> %1
+  ret <4 x float> %2
+}
+
+; CHECK-LABEL: test4_mul_ss
+; SSE2: mulss   %xmm0, %xmm1
+; AVX: vmulss   %xmm0, %xmm1, %xmm0
+; CHECK-NOT: movss
+; CHECK: ret
+
+
+define <4 x float> @test4_div_ss(<4 x float> %a, <4 x float> %b) {
+  %1 = fdiv <4 x float> %b, %a
+  %2 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %b, <4 x float> %1
+  ret <4 x float> %2
+}
+
+; CHECK-LABEL: test4_div_ss
+; SSE2: divss   %xmm0, %xmm1
+; AVX: vdivss   %xmm0, %xmm1, %xmm0
+; CHECK-NOT: movss
+; CHECK: ret
+
+
+define <2 x double> @test4_add_sd(<2 x double> %a, <2 x double> %b) {
+  %1 = fadd <2 x double> %b, %a
+  %2 = select <2 x i1> <i1 false, i1 true>, <2 x double> %b, <2 x double> %1
+  ret <2 x double> %2
+}
+
+; CHECK-LABEL: test4_add_sd
+; SSE2: addsd   %xmm0, %xmm1
+; AVX: vaddsd   %xmm0, %xmm1, %xmm0
+; CHECK-NOT: movsd
+; CHECK: ret
+
+
+define <2 x double> @test4_sub_sd(<2 x double> %a, <2 x double> %b) {
+  %1 = fsub <2 x double> %b, %a
+  %2 = select <2 x i1> <i1 false, i1 true>, <2 x double> %b, <2 x double> %1
+  ret <2 x double> %2
+}
+
+; CHECK-LABEL: test4_sub_sd
+; SSE2: subsd   %xmm0, %xmm1
+; AVX: vsubsd   %xmm0, %xmm1, %xmm0
+; CHECK-NOT: movsd
+; CHECK: ret
+
+
+define <2 x double> @test4_mul_sd(<2 x double> %a, <2 x double> %b) {
+  %1 = fmul <2 x double> %b, %a
+  %2 = select <2 x i1> <i1 false, i1 true>, <2 x double> %b, <2 x double> %1
+  ret <2 x double> %2
+}
+
+; CHECK-LABEL: test4_mul_sd
+; SSE2: mulsd   %xmm0, %xmm1
+; AVX: vmulsd   %xmm0, %xmm1, %xmm0
+; CHECK-NOT: movsd
+; CHECK: ret
+
+
+define <2 x double> @test4_div_sd(<2 x double> %a, <2 x double> %b) {
+  %1 = fdiv <2 x double> %b, %a
+  %2 = select <2 x i1> <i1 false, i1 true>, <2 x double> %b, <2 x double> %1
+  ret <2 x double> %2
+}
+
+; CHECK-LABEL: test4_div_sd
+; SSE2: divsd   %xmm0, %xmm1
+; AVX: vdivsd   %xmm0, %xmm1, %xmm0
+; CHECK-NOT: movsd
+; CHECK: ret
+

diff --git a/test/CodeGen/X86/sse2-blend.ll b/test/CodeGen/X86/sse2-blend.ll
index 968595c383af618fc19bf62df99e0758adb65182..c63ff72b48012fc9411e8060772f211c0c367f3b 100644 (file)
--- a/test/CodeGen/X86/sse2-blend.ll
+++ b/test/CodeGen/X86/sse2-blend.ll
@@ -1,22 +1,22 @@
 ; RUN: llc < %s -march=x86 -mcpu=yonah -mattr=+sse2,-sse4.1 | FileCheck %s
 
-; CHECK: vsel_float
-; CHECK: xorps
+; CHECK-LABEL: vsel_float
+; CHECK-NOT: xorps
 ; CHECK: movss
-; CHECK: orps
+; CHECK-NOT: orps
 ; CHECK: ret
 define void@vsel_float(<4 x float>* %v1, <4 x float>* %v2) {
   %A = load <4 x float>* %v1
   %B = load <4 x float>* %v2
-  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x float> %A, <4 x float> %B
+  %vsel = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %A, <4 x float> %B
   store <4 x float > %vsel, <4 x float>* %v1
   ret void
 }
 
-; CHECK: vsel_i32
-; CHECK: xorps
+; CHECK-LABEL: vsel_i32
+; CHECK-NOT: xorps
 ; CHECK: movss
-; CHECK: orps
+; CHECK-NOT: orps
 ; CHECK: ret
 define void@vsel_i32(<4 x i32>* %v1, <4 x i32>* %v2) {
   %A = load <4 x i32>* %v1
@@ -27,7 +27,7 @@ define void@vsel_i32(<4 x i32>* %v1, <4 x i32>* %v2) {
 }
 
 ; Without forcing instructions, fall back to the preferred PS domain.
-; CHECK: vsel_i64
+; CHECK-LABEL: vsel_i64
 ; CHECK: andnps
 ; CHECK: orps
 ; CHECK: ret
@@ -41,7 +41,7 @@ define void@vsel_i64(<2 x i64>* %v1, <2 x i64>* %v2) {
 }
 
 ; Without forcing instructions, fall back to the preferred PS domain.
-; CHECK: vsel_double
+; CHECK-LABEL: vsel_double
 ; CHECK: andnps
 ; CHECK: orps
 ; CHECK: ret

diff --git a/test/CodeGen/X86/sse41-blend.ll b/test/CodeGen/X86/sse41-blend.ll
index a32f5de30a6cef3bf1629dbbf3bb3f2f3a053253..597852c36904066bad0fd3851782425477e77077 100644 (file)
--- a/test/CodeGen/X86/sse41-blend.ll
+++ b/test/CodeGen/X86/sse41-blend.ll
@@ -4,7 +4,7 @@
 ;CHECK: blendvps
 ;CHECK: ret
 define <4 x float> @vsel_float(<4 x float> %v1, <4 x float> %v2) {
-  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x float> %v1, <4 x float> %v2
+  %vsel = select <4 x i1> <i1 true, i1 false, i1 true, i1 true>, <4 x float> %v1, <4 x float> %v2
   ret <4 x float> %vsel
 }
 
@@ -13,7 +13,7 @@ define <4 x float> @vsel_float(<4 x float> %v1, <4 x float> %v2) {
 ;CHECK: blendvps
 ;CHECK: ret
 define <4 x i8> @vsel_4xi8(<4 x i8> %v1, <4 x i8> %v2) {
-  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x i8> %v1, <4 x i8> %v2
+  %vsel = select <4 x i1> <i1 true, i1 true, i1 false, i1 false>, <4 x i8> %v1, <4 x i8> %v2
   ret <4 x i8> %vsel
 }
 
@@ -21,7 +21,7 @@ define <4 x i8> @vsel_4xi8(<4 x i8> %v1, <4 x i8> %v2) {
 ;CHECK: blendvps
 ;CHECK: ret
 define <4 x i16> @vsel_4xi16(<4 x i16> %v1, <4 x i16> %v2) {
-  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x i16> %v1, <4 x i16> %v2
+  %vsel = select <4 x i1> <i1 true, i1 false, i1 true, i1 true>, <4 x i16> %v1, <4 x i16> %v2
   ret <4 x i16> %vsel
 }
 
@@ -30,13 +30,13 @@ define <4 x i16> @vsel_4xi16(<4 x i16> %v1, <4 x i16> %v2) {
 ;CHECK: blendvps
 ;CHECK: ret
 define <4 x i32> @vsel_i32(<4 x i32> %v1, <4 x i32> %v2) {
-  %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x i32> %v1, <4 x i32> %v2
+  %vsel = select <4 x i1> <i1 true, i1 true, i1 false, i1 false>, <4 x i32> %v1, <4 x i32> %v2
   ret <4 x i32> %vsel
 }
 
 
 ;CHECK-LABEL: vsel_double:
-;CHECK: blendvpd
+;CHECK: movsd
 ;CHECK: ret
 define <4 x double> @vsel_double(<4 x double> %v1, <4 x double> %v2) {
   %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x double> %v1, <4 x double> %v2
@@ -45,7 +45,7 @@ define <4 x double> @vsel_double(<4 x double> %v1, <4 x double> %v2) {
 
 
 ;CHECK-LABEL: vsel_i64:
-;CHECK: blendvpd
+;CHECK: movsd
 ;CHECK: ret
 define <4 x i64> @vsel_i64(<4 x i64> %v1, <4 x i64> %v2) {
   %vsel = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x i64> %v1, <4 x i64> %v2

diff --git a/test/CodeGen/X86/vselect.ll b/test/CodeGen/X86/vselect.ll
index 36c79838ff2cc0af90978411ffb0006dd27bb83a..0cf03fc5d62c96f5375548448c60e62d19a59c56 100644 (file)
--- a/test/CodeGen/X86/vselect.ll
+++ b/test/CodeGen/X86/vselect.ll
@@ -174,3 +174,91 @@ define <8 x i16> @test17(<8 x i16> %a, <8 x i16> %b) {
 ; CHECK-NOT: xorps
 ; CHECK: ret
 
+define <4 x float> @test18(<4 x float> %a, <4 x float> %b) {
+  %1 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x float> %a, <4 x float> %b
+  ret <4 x float> %1
+}
+; CHECK-LABEL: test18
+; CHECK-NOT: psllw
+; CHECK-NOT: psraw
+; CHECK-NOT: xorps
+; CHECK: movss
+; CHECK: ret
+
+define <4 x i32> @test19(<4 x i32> %a, <4 x i32> %b) {
+  %1 = select <4 x i1> <i1 false, i1 true, i1 true, i1 true>, <4 x i32> %a, <4 x i32> %b
+  ret <4 x i32> %1
+}
+; CHECK-LABEL: test19
+; CHECK-NOT: psllw
+; CHECK-NOT: psraw
+; CHECK-NOT: xorps
+; CHECK: movss
+; CHECK: ret
+
+define <2 x double> @test20(<2 x double> %a, <2 x double> %b) {
+  %1 = select <2 x i1> <i1 false, i1 true>, <2 x double> %a, <2 x double> %b
+  ret <2 x double> %1
+}
+; CHECK-LABEL: test20
+; CHECK-NOT: psllw
+; CHECK-NOT: psraw
+; CHECK-NOT: xorps
+; CHECK: movsd
+; CHECK: ret
+
+define <2 x i64> @test21(<2 x i64> %a, <2 x i64> %b) {
+  %1 = select <2 x i1> <i1 false, i1 true>, <2 x i64> %a, <2 x i64> %b
+  ret <2 x i64> %1
+}
+; CHECK-LABEL: test21
+; CHECK-NOT: psllw
+; CHECK-NOT: psraw
+; CHECK-NOT: xorps
+; CHECK: movsd
+; CHECK: ret
+
+define <4 x float> @test22(<4 x float> %a, <4 x float> %b) {
+  %1 = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x float> %a, <4 x float> %b
+  ret <4 x float> %1
+}
+; CHECK-LABEL: test22
+; CHECK-NOT: psllw
+; CHECK-NOT: psraw
+; CHECK-NOT: xorps
+; CHECK: movss
+; CHECK: ret
+
+define <4 x i32> @test23(<4 x i32> %a, <4 x i32> %b) {
+  %1 = select <4 x i1> <i1 true, i1 false, i1 false, i1 false>, <4 x i32> %a, <4 x i32> %b
+  ret <4 x i32> %1
+}
+; CHECK-LABEL: test23
+; CHECK-NOT: psllw
+; CHECK-NOT: psraw
+; CHECK-NOT: xorps
+; CHECK: movss
+; CHECK: ret
+
+define <2 x double> @test24(<2 x double> %a, <2 x double> %b) {
+  %1 = select <2 x i1> <i1 true, i1 false>, <2 x double> %a, <2 x double> %b
+  ret <2 x double> %1
+}
+; CHECK-LABEL: test24
+; CHECK-NOT: psllw
+; CHECK-NOT: psraw
+; CHECK-NOT: xorps
+; CHECK: movsd
+; CHECK: ret
+
+define <2 x i64> @test25(<2 x i64> %a, <2 x i64> %b) {
+  %1 = select <2 x i1> <i1 true, i1 false>, <2 x i64> %a, <2 x i64> %b
+  ret <2 x i64> %1
+}
+; CHECK-LABEL: test25
+; CHECK-NOT: psllw
+; CHECK-NOT: psraw
+; CHECK-NOT: xorps
+; CHECK: movsd
+; CHECK: ret
+