[llvm] r312458 - [X86] Combine inserting a vector of zeros into a vector of zeros just the larger vector.

[Craig Topper via llvm-commits]

Author: ctopper
Date: Sun Sep  3 15:25:49 2017
New Revision: 312458

URL: http://llvm.org/viewvc/llvm-project?rev=312458&view=rev
Log:
[X86] Combine inserting a vector of zeros into a vector of zeros just the larger vector.

Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
    llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=312458&r1=312457&r2=312458&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sun Sep  3 15:25:49 2017
@@ -35656,6 +35656,11 @@ static SDValue combineInsertSubvector(SD
   unsigned IdxVal = cast<ConstantSDNode>(Idx)->getZExtValue();
   MVT SubVecVT = SubVec.getSimpleValueType();
 
+  // Inserting zeros into zeros is a nop.
+  if (ISD::isBuildVectorAllZeros(Vec.getNode()) &&
+      ISD::isBuildVectorAllZeros(SubVec.getNode()))
+    return Vec;
+
   // If this is an insert of an extract, combine to a shuffle. Don't do this
   // if the insert or extract can be represented with a subregister operation.
   if (SubVec.getOpcode() == ISD::EXTRACT_SUBVECTOR &&

Modified: llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll?rev=312458&r1=312457&r2=312458&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll (original)
+++ llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll Sun Sep  3 15:25:49 2017
@@ -1134,17 +1134,13 @@ define <8 x double> @test_mm512_zextpd12
 ; X32-LABEL: test_mm512_zextpd128_pd512:
 ; X32:       # BB#0:
 ; X32-NEXT:    vmovaps %xmm0, %xmm0
-; X32-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; X32-NEXT:    vmovaps %xmm1, %xmm1
-; X32-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
+; X32-NEXT:    vmovaps %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_mm512_zextpd128_pd512:
 ; X64:       # BB#0:
 ; X64-NEXT:    vmovaps %xmm0, %xmm0
-; X64-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; X64-NEXT:    vmovaps %xmm1, %xmm1
-; X64-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
+; X64-NEXT:    vmovaps %ymm0, %ymm0
 ; X64-NEXT:    retq
   %res = shufflevector <2 x double> %a0, <2 x double> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   ret <8 x double> %res
@@ -1169,7 +1165,6 @@ define <16 x float> @test_mm512_zextps12
 ; X32:       # BB#0:
 ; X32-NEXT:    vmovaps %xmm0, %xmm0
 ; X32-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; X32-NEXT:    vmovaps %xmm1, %xmm1
 ; X32-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
 ; X32-NEXT:    retl
 ;
@@ -1177,7 +1172,6 @@ define <16 x float> @test_mm512_zextps12
 ; X64:       # BB#0:
 ; X64-NEXT:    vmovaps %xmm0, %xmm0
 ; X64-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; X64-NEXT:    vmovaps %xmm1, %xmm1
 ; X64-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
 ; X64-NEXT:    retq
   %res = shufflevector <4 x float> %a0, <4 x float> zeroinitializer, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
@@ -1202,17 +1196,13 @@ define <8 x i64> @test_mm512_zextsi128_s
 ; X32-LABEL: test_mm512_zextsi128_si512:
 ; X32:       # BB#0:
 ; X32-NEXT:    vmovaps %xmm0, %xmm0
-; X32-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; X32-NEXT:    vmovaps %xmm1, %xmm1
-; X32-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
+; X32-NEXT:    vmovaps %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_mm512_zextsi128_si512:
 ; X64:       # BB#0:
 ; X64-NEXT:    vmovaps %xmm0, %xmm0
-; X64-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; X64-NEXT:    vmovaps %xmm1, %xmm1
-; X64-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
+; X64-NEXT:    vmovaps %ymm0, %ymm0
 ; X64-NEXT:    retq
   %res = shufflevector <2 x i64> %a0, <2 x i64> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
   ret <8 x i64> %res