[DAGCombiner] after simplifying demanded elements of vector operand of extract, revisit the extract

Differential Revision: https://reviews.llvm.org/D55655

llvm-svn: 349051

9 files changed, 14 insertions, 31 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 7768d12c75c..b364b1baea7 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -15698,8 +15698,13 @@ SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) {
       if (CstElt->getAPIntValue().ult(VT.getVectorNumElements()))
         DemandedElts.setBit(CstElt->getZExtValue());
     }
-    if (SimplifyDemandedVectorElts(InVec, DemandedElts, true))
+    if (SimplifyDemandedVectorElts(InVec, DemandedElts, true)) {
+      // We simplified the operands of this extract element, but if this extract
+      // is still alive, visit it again to ensure that it is folded properly.
+      if (!N->use_empty())
+        AddToWorklist(N);
       return SDValue(N, 0);
+    }
   }
 
   bool BCNumEltsChanged = false;
diff --git a/llvm/test/CodeGen/X86/extractelement-load.ll b/llvm/test/CodeGen/X86/extractelement-load.ll
index 9eb161f5804..c4a597b90ed 100644
--- a/llvm/test/CodeGen/X86/extractelement-load.ll
+++ b/llvm/test/CodeGen/X86/extractelement-load.ll
@@ -93,8 +93,7 @@ define i64 @t4(<2 x double>* %a) {
 ;
 ; X64-SSSE3-LABEL: t4:
 ; X64-SSSE3:       # %bb.0:
-; X64-SSSE3-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
-; X64-SSSE3-NEXT:    movq %xmm0, %rax
+; X64-SSSE3-NEXT:    movq (%rdi), %rax
 ; X64-SSSE3-NEXT:    retq
 ;
 ; X64-AVX-LABEL: t4:
diff --git a/llvm/test/CodeGen/X86/insertps-combine.ll b/llvm/test/CodeGen/X86/insertps-combine.ll
index 6807048a05e..b99605ad50b 100644
--- a/llvm/test/CodeGen/X86/insertps-combine.ll
+++ b/llvm/test/CodeGen/X86/insertps-combine.ll
@@ -284,13 +284,12 @@ define float @extract_lane_insertps_5123(<4 x float> %a0, <4 x float> *%p1) {
 define float @extract_lane_insertps_6123(<4 x float> %a0, <4 x float> *%p1) {
 ; SSE-LABEL: extract_lane_insertps_6123:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movaps (%rdi), %xmm0
-; SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
+; SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: extract_lane_insertps_6123:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm0 = mem[1,0]
+; AVX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; AVX-NEXT:    retq
   %a1 = load <4 x float>, <4 x float> *%p1
   %res = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a0, <4 x float> %a1, i8 128)
diff --git a/llvm/test/CodeGen/X86/masked_load.ll b/llvm/test/CodeGen/X86/masked_load.ll
index ef67b94f70a..5a456bd2d71 100644
--- a/llvm/test/CodeGen/X86/masked_load.ll
+++ b/llvm/test/CodeGen/X86/masked_load.ll
@@ -556,7 +556,7 @@ define <8 x float> @load_v8f32_v8i32(<8 x i32> %trigger, <8 x float>* %addr, <8
 ; SSE2-NEXT:    movss {{.*#+}} xmm5 = mem[0],zero,zero,zero
 ; SSE2-NEXT:    movss {{.*#+}} xmm2 = xmm5[0],xmm2[1,2,3]
 ; SSE2-NEXT:  LBB6_2: ## %else
-; SSE2-NEXT:    packssdw %xmm0, %xmm4
+; SSE2-NEXT:    psrlq $16, %xmm4
 ; SSE2-NEXT:    movd %xmm4, %eax
 ; SSE2-NEXT:    shrl $16, %eax
 ; SSE2-NEXT:    testb $1, %al
diff --git a/llvm/test/CodeGen/X86/masked_store.ll b/llvm/test/CodeGen/X86/masked_store.ll
index 931ec7dc0e8..61edacdba95 100644
--- a/llvm/test/CodeGen/X86/masked_store.ll
+++ b/llvm/test/CodeGen/X86/masked_store.ll
@@ -145,7 +145,7 @@ define void @store_v8i32_v8i32(<8 x i32> %trigger, <8 x i32>* %addr, <8 x i32> %
 ; SSE2-NEXT:  ## %bb.1: ## %cond.store
 ; SSE2-NEXT:    movd %xmm2, (%rdi)
 ; SSE2-NEXT:  LBB2_2: ## %else
-; SSE2-NEXT:    packssdw %xmm0, %xmm4
+; SSE2-NEXT:    psrlq $16, %xmm4
 ; SSE2-NEXT:    movd %xmm4, %eax
 ; SSE2-NEXT:    shrl $16, %eax
 ; SSE2-NEXT:    testb $1, %al
diff --git a/llvm/test/CodeGen/X86/shrink_vmul-widen.ll b/llvm/test/CodeGen/X86/shrink_vmul-widen.ll
index 47b91912c95..3c047dae80a 100644
--- a/llvm/test/CodeGen/X86/shrink_vmul-widen.ll
+++ b/llvm/test/CodeGen/X86/shrink_vmul-widen.ll
@@ -1175,7 +1175,6 @@ define void @mul_2xi16_sext_zext(i8* nocapture readonly %a, i8* nocapture readon
 ; X86-SSE-NEXT:    pmuludq %xmm0, %xmm1
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
 ; X86-SSE-NEXT:    pmuludq %xmm2, %xmm0
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
 ; X86-SSE-NEXT:    movq %xmm1, (%esi,%ecx,4)
 ; X86-SSE-NEXT:    popl %esi
@@ -1210,7 +1209,6 @@ define void @mul_2xi16_sext_zext(i8* nocapture readonly %a, i8* nocapture readon
 ; X64-SSE-NEXT:    pmuludq %xmm0, %xmm1
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
 ; X64-SSE-NEXT:    pmuludq %xmm2, %xmm0
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
 ; X64-SSE-NEXT:    movq %xmm1, (%rax,%rdx,4)
 ; X64-SSE-NEXT:    retq
@@ -1965,7 +1963,6 @@ define void @mul_2xi16_varconst3(i8* nocapture readonly %a, i64 %index) {
 ; X86-SSE-NEXT:    pmuludq %xmm1, %xmm0
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; X86-SSE-NEXT:    pmuludq %xmm2, %xmm1
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; X86-SSE-NEXT:    movq %xmm0, (%edx,%eax,4)
 ; X86-SSE-NEXT:    retl
@@ -1992,7 +1989,6 @@ define void @mul_2xi16_varconst3(i8* nocapture readonly %a, i64 %index) {
 ; X64-SSE-NEXT:    pmuludq %xmm1, %xmm0
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; X64-SSE-NEXT:    pmuludq %xmm2, %xmm1
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; X64-SSE-NEXT:    movq %xmm0, (%rax,%rsi,4)
 ; X64-SSE-NEXT:    retq
@@ -2037,7 +2033,6 @@ define void @mul_2xi16_varconst4(i8* nocapture readonly %a, i64 %index) {
 ; X86-SSE-NEXT:    pmuludq %xmm1, %xmm0
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; X86-SSE-NEXT:    pmuludq %xmm2, %xmm1
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; X86-SSE-NEXT:    movq %xmm0, (%edx,%eax,4)
 ; X86-SSE-NEXT:    retl
@@ -2064,7 +2059,6 @@ define void @mul_2xi16_varconst4(i8* nocapture readonly %a, i64 %index) {
 ; X64-SSE-NEXT:    pmuludq %xmm1, %xmm0
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
 ; X64-SSE-NEXT:    pmuludq %xmm2, %xmm1
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; X64-SSE-NEXT:    movq %xmm0, (%rax,%rsi,4)
 ; X64-SSE-NEXT:    retq
diff --git a/llvm/test/CodeGen/X86/vec_extract-sse4.ll b/llvm/test/CodeGen/X86/vec_extract-sse4.ll
index 2933c567fbb..ca3d87c4c61 100644
--- a/llvm/test/CodeGen/X86/vec_extract-sse4.ll
+++ b/llvm/test/CodeGen/X86/vec_extract-sse4.ll
@@ -25,17 +25,13 @@ define void @t1(float* %R, <4 x float>* %P1) nounwind {
 define float @t2(<4 x float>* %P1) nounwind {
 ; X32-LABEL: t2:
 ; X32:       # %bb.0:
-; X32-NEXT:    pushl %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
-; X32-NEXT:    movss %xmm0, (%esp)
-; X32-NEXT:    flds (%esp)
-; X32-NEXT:    popl %eax
+; X32-NEXT:    flds 8(%eax)
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: t2:
 ; X64:       # %bb.0:
-; X64-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X64-NEXT:    retq
   %X = load <4 x float>, <4 x float>* %P1
   %tmp = extractelement <4 x float> %X, i32 2
diff --git a/llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll b/llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll
index 09db9c5d420..7fbd7245c41 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll
@@ -1863,8 +1863,6 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
diff --git a/llvm/test/CodeGen/X86/vector-reduce-mul.ll b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
index 728aabd51bc..ea80a5be2a6 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-mul.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
@@ -808,7 +808,6 @@ define i32 @test_v2i32(<2 x i32> %a0) {
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512DQ-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX512DQ-NEXT:    vpmullq %zmm1, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQ-NEXT:    vzeroupper
@@ -817,7 +816,6 @@ define i32 @test_v2i32(<2 x i32> %a0) {
 ; AVX512DQVL-LABEL: test_v2i32:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512DQVL-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX512DQVL-NEXT:    vpmullq %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    retq
@@ -1178,7 +1176,6 @@ define i16 @test_v2i16(<2 x i16> %a0) {
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512DQ-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX512DQ-NEXT:    vpmullq %zmm1, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $ax killed $ax killed $eax
@@ -1188,7 +1185,6 @@ define i16 @test_v2i16(<2 x i16> %a0) {
 ; AVX512DQVL-LABEL: test_v2i16:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512DQVL-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX512DQVL-NEXT:    vpmullq %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $ax killed $ax killed $eax
@@ -1210,7 +1206,7 @@ define i16 @test_v4i16(<4 x i16> %a0) {
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pmuludq %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    pextrw $0, %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
@@ -1658,7 +1654,6 @@ define i8 @test_v2i8(<2 x i8> %a0) {
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512DQ-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX512DQ-NEXT:    vpmullq %zmm1, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
@@ -1668,7 +1663,6 @@ define i8 @test_v2i8(<2 x i8> %a0) {
 ; AVX512DQVL-LABEL: test_v2i8:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512DQVL-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX512DQVL-NEXT:    vpmullq %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
@@ -1899,8 +1893,6 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} 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,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper