[X86][AVX2] Combine extract(broadcast(scalar_value)) --> scalar_value

As it has a scalar source we don't treat it as a target shuffle so needs special handling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@321610 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 19bff03f26ab695e2a1a4ecad746792f0ad8c45d..b8eea1ca2be410651a0459f6518ac2d06965fee0 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -30923,6 +30923,11 @@ static SDValue combineExtractWithShuffle(SDNode *N, SelectionDAG &DAG,
   if (SrcSVT == MVT::i1 || !isa<ConstantSDNode>(Idx))
     return SDValue();
 
+  // Handle extract(broadcast(scalar_value)), it doesn't matter what index is.
+  if (X86ISD::VBROADCAST == Src.getOpcode() &&
+      Src.getOperand(0).getValueType() == VT)
+    return Src.getOperand(0);
+
   // Resolve the target shuffle inputs and mask.
   SmallVector<int, 16> Mask;
   SmallVector<SDValue, 2> Ops;

diff --git a/test/CodeGen/X86/avx-vbroadcast.ll b/test/CodeGen/X86/avx-vbroadcast.ll
index 5a9f23007d86f7b1f610bc6d8a4b14911733a053..e2dc40c6f0ef1e0505c9e0d6c127d73bf85f41b8 100644 (file)
--- a/test/CodeGen/X86/avx-vbroadcast.ll
+++ b/test/CodeGen/X86/avx-vbroadcast.ll
@@ -874,39 +874,33 @@ define float @broadcast_lifetime() nounwind {
 ; X32-LABEL: broadcast_lifetime:
 ; X32:       ## %bb.0:
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    subl $56, %esp
+; X32-NEXT:    subl $40, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, (%esp)
 ; X32-NEXT:    calll _gfunc
 ; X32-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X32-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%esp) ## 16-byte Spill
+; X32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp) ## 4-byte Spill
 ; X32-NEXT:    movl %esi, (%esp)
 ; X32-NEXT:    calll _gfunc
 ; X32-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X32-NEXT:    vpermilps $0, {{[0-9]+}}(%esp), %xmm1 ## 16-byte Folded Reload
-; X32-NEXT:    ## xmm1 = mem[0,0,0,0]
-; X32-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; X32-NEXT:    vsubss %xmm1, %xmm0, %xmm0
+; X32-NEXT:    vsubss {{[0-9]+}}(%esp), %xmm0, %xmm0 ## 4-byte Folded Reload
 ; X32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp)
 ; X32-NEXT:    flds {{[0-9]+}}(%esp)
-; X32-NEXT:    addl $56, %esp
+; X32-NEXT:    addl $40, %esp
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: broadcast_lifetime:
 ; X64:       ## %bb.0:
 ; X64-NEXT:    subq $40, %rsp
-; X64-NEXT:    movq %rsp, %rdi
+; X64-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
 ; X64-NEXT:    callq _gfunc
 ; X64-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X64-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) ## 16-byte Spill
-; X64-NEXT:    movq %rsp, %rdi
+; X64-NEXT:    vmovss %xmm0, {{[0-9]+}}(%rsp) ## 4-byte Spill
+; X64-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
 ; X64-NEXT:    callq _gfunc
 ; X64-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X64-NEXT:    vpermilps $0, {{[0-9]+}}(%rsp), %xmm1 ## 16-byte Folded Reload
-; X64-NEXT:    ## xmm1 = mem[0,0,0,0]
-; X64-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; X64-NEXT:    vsubss %xmm1, %xmm0, %xmm0
+; X64-NEXT:    vsubss {{[0-9]+}}(%rsp), %xmm0, %xmm0 ## 4-byte Folded Reload
 ; X64-NEXT:    addq $40, %rsp
 ; X64-NEXT:    retq
   %1 = alloca <4 x float>, align 16

diff --git a/test/CodeGen/X86/vector-pcmp.ll b/test/CodeGen/X86/vector-pcmp.ll
index b06b667ec07a6688329dca9b7153226c5bd5da2a..b2c0a4d096c2b925a74dd221463bf335ba450376 100644 (file)
--- a/test/CodeGen/X86/vector-pcmp.ll
+++ b/test/CodeGen/X86/vector-pcmp.ll
@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE42
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSE42
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX2
 
 ; Lower common integer comparisons such as 'isPositive' efficiently:
 ; https://llvm.org/bugs/show_bug.cgi?id=26701
@@ -84,31 +84,13 @@ define <2 x i64> @test_pcmpgtq(<2 x i64> %x) {
 }
 
 define <1 x i128> @test_strange_type(<1 x i128> %x) {
-; SSE-LABEL: test_strange_type:
-; SSE:       # %bb.0:
-; SSE-NEXT:    sarq $63, %rsi
-; SSE-NEXT:    notq %rsi
-; SSE-NEXT:    movq %rsi, %rax
-; SSE-NEXT:    movq %rsi, %rdx
-; SSE-NEXT:    retq
-;
-; AVX1-LABEL: test_strange_type:
-; AVX1:       # %bb.0:
-; AVX1-NEXT:    sarq $63, %rsi
-; AVX1-NEXT:    notq %rsi
-; AVX1-NEXT:    movq %rsi, %rax
-; AVX1-NEXT:    movq %rsi, %rdx
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test_strange_type:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    sarq $63, %rsi
-; AVX2-NEXT:    notq %rsi
-; AVX2-NEXT:    vmovq %rsi, %xmm0
-; AVX2-NEXT:    vpbroadcastq %xmm0, %xmm0
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rdx
-; AVX2-NEXT:    retq
+; CHECK-LABEL: test_strange_type:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    sarq $63, %rsi
+; CHECK-NEXT:    notq %rsi
+; CHECK-NEXT:    movq %rsi, %rax
+; CHECK-NEXT:    movq %rsi, %rdx
+; CHECK-NEXT:    retq
   %sign = ashr <1 x i128> %x, <i128 127>
   %not = xor <1 x i128> %sign, <i128 -1>
   ret <1 x i128> %not