Skip to content

Commit a7b115b

Browse files
rotaterightrotateright
committedDec 13, 2018
[DAGCombiner] after simplifying demanded elements of vector operand of extract, revisit the extract
Differential Revision: https://reviews.llvm.org/D55655 llvm-svn: 349051
·
llvmorg-21-initllvmorg-8.0.0-rc1
1 parent 4110967 commit a7b115b

9 files changed

+14
-31
lines changed
 

‎llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

+6-1
Original file line numberDiff line numberDiff line change
@@ -15698,8 +15698,13 @@ SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) {
1569815698
if (CstElt->getAPIntValue().ult(VT.getVectorNumElements()))
1569915699
DemandedElts.setBit(CstElt->getZExtValue());
1570015700
}
15701-
if (SimplifyDemandedVectorElts(InVec, DemandedElts, true))
15701+
if (SimplifyDemandedVectorElts(InVec, DemandedElts, true)) {
15702+
// We simplified the operands of this extract element, but if this extract
15703+
// is still alive, visit it again to ensure that it is folded properly.
15704+
if (!N->use_empty())
15705+
AddToWorklist(N);
1570215706
return SDValue(N, 0);
15707+
}
1570315708
}
1570415709

1570515710
bool BCNumEltsChanged = false;

‎llvm/test/CodeGen/X86/extractelement-load.ll

+1-2
Original file line numberDiff line numberDiff line change
@@ -93,8 +93,7 @@ define i64 @t4(<2 x double>* %a) {
9393
;
9494
; X64-SSSE3-LABEL: t4:
9595
; X64-SSSE3: # %bb.0:
96-
; X64-SSSE3-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
97-
; X64-SSSE3-NEXT: movq %xmm0, %rax
96+
; X64-SSSE3-NEXT: movq (%rdi), %rax
9897
; X64-SSSE3-NEXT: retq
9998
;
10099
; X64-AVX-LABEL: t4:

‎llvm/test/CodeGen/X86/insertps-combine.ll

+2-3
Original file line numberDiff line numberDiff line change
@@ -284,13 +284,12 @@ define float @extract_lane_insertps_5123(<4 x float> %a0, <4 x float> *%p1) {
284284
define float @extract_lane_insertps_6123(<4 x float> %a0, <4 x float> *%p1) {
285285
; SSE-LABEL: extract_lane_insertps_6123:
286286
; SSE: # %bb.0:
287-
; SSE-NEXT: movaps (%rdi), %xmm0
288-
; SSE-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
287+
; SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
289288
; SSE-NEXT: retq
290289
;
291290
; AVX-LABEL: extract_lane_insertps_6123:
292291
; AVX: # %bb.0:
293-
; AVX-NEXT: vpermilpd {{.*#+}} xmm0 = mem[1,0]
292+
; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
294293
; AVX-NEXT: retq
295294
%a1 = load <4 x float>, <4 x float> *%p1
296295
%res = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a0, <4 x float> %a1, i8 128)

‎llvm/test/CodeGen/X86/masked_load.ll

+1-1
Original file line numberDiff line numberDiff line change
@@ -556,7 +556,7 @@ define <8 x float> @load_v8f32_v8i32(<8 x i32> %trigger, <8 x float>* %addr, <8
556556
; SSE2-NEXT: movss {{.*#+}} xmm5 = mem[0],zero,zero,zero
557557
; SSE2-NEXT: movss {{.*#+}} xmm2 = xmm5[0],xmm2[1,2,3]
558558
; SSE2-NEXT: LBB6_2: ## %else
559-
; SSE2-NEXT: packssdw %xmm0, %xmm4
559+
; SSE2-NEXT: psrlq $16, %xmm4
560560
; SSE2-NEXT: movd %xmm4, %eax
561561
; SSE2-NEXT: shrl $16, %eax
562562
; SSE2-NEXT: testb $1, %al

‎llvm/test/CodeGen/X86/masked_store.ll

+1-1
Original file line numberDiff line numberDiff line change
@@ -145,7 +145,7 @@ define void @store_v8i32_v8i32(<8 x i32> %trigger, <8 x i32>* %addr, <8 x i32> %
145145
; SSE2-NEXT: ## %bb.1: ## %cond.store
146146
; SSE2-NEXT: movd %xmm2, (%rdi)
147147
; SSE2-NEXT: LBB2_2: ## %else
148-
; SSE2-NEXT: packssdw %xmm0, %xmm4
148+
; SSE2-NEXT: psrlq $16, %xmm4
149149
; SSE2-NEXT: movd %xmm4, %eax
150150
; SSE2-NEXT: shrl $16, %eax
151151
; SSE2-NEXT: testb $1, %al

‎llvm/test/CodeGen/X86/shrink_vmul-widen.ll

-6
Original file line numberDiff line numberDiff line change
@@ -1175,7 +1175,6 @@ define void @mul_2xi16_sext_zext(i8* nocapture readonly %a, i8* nocapture readon
11751175
; X86-SSE-NEXT: pmuludq %xmm0, %xmm1
11761176
; X86-SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
11771177
; X86-SSE-NEXT: pmuludq %xmm2, %xmm0
1178-
; X86-SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
11791178
; X86-SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
11801179
; X86-SSE-NEXT: movq %xmm1, (%esi,%ecx,4)
11811180
; X86-SSE-NEXT: popl %esi
@@ -1210,7 +1209,6 @@ define void @mul_2xi16_sext_zext(i8* nocapture readonly %a, i8* nocapture readon
12101209
; X64-SSE-NEXT: pmuludq %xmm0, %xmm1
12111210
; X64-SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
12121211
; X64-SSE-NEXT: pmuludq %xmm2, %xmm0
1213-
; X64-SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
12141212
; X64-SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
12151213
; X64-SSE-NEXT: movq %xmm1, (%rax,%rdx,4)
12161214
; X64-SSE-NEXT: retq
@@ -1965,7 +1963,6 @@ define void @mul_2xi16_varconst3(i8* nocapture readonly %a, i64 %index) {
19651963
; X86-SSE-NEXT: pmuludq %xmm1, %xmm0
19661964
; X86-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
19671965
; X86-SSE-NEXT: pmuludq %xmm2, %xmm1
1968-
; X86-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
19691966
; X86-SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
19701967
; X86-SSE-NEXT: movq %xmm0, (%edx,%eax,4)
19711968
; X86-SSE-NEXT: retl
@@ -1992,7 +1989,6 @@ define void @mul_2xi16_varconst3(i8* nocapture readonly %a, i64 %index) {
19921989
; X64-SSE-NEXT: pmuludq %xmm1, %xmm0
19931990
; X64-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
19941991
; X64-SSE-NEXT: pmuludq %xmm2, %xmm1
1995-
; X64-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
19961992
; X64-SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
19971993
; X64-SSE-NEXT: movq %xmm0, (%rax,%rsi,4)
19981994
; X64-SSE-NEXT: retq
@@ -2037,7 +2033,6 @@ define void @mul_2xi16_varconst4(i8* nocapture readonly %a, i64 %index) {
20372033
; X86-SSE-NEXT: pmuludq %xmm1, %xmm0
20382034
; X86-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
20392035
; X86-SSE-NEXT: pmuludq %xmm2, %xmm1
2040-
; X86-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
20412036
; X86-SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
20422037
; X86-SSE-NEXT: movq %xmm0, (%edx,%eax,4)
20432038
; X86-SSE-NEXT: retl
@@ -2064,7 +2059,6 @@ define void @mul_2xi16_varconst4(i8* nocapture readonly %a, i64 %index) {
20642059
; X64-SSE-NEXT: pmuludq %xmm1, %xmm0
20652060
; X64-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
20662061
; X64-SSE-NEXT: pmuludq %xmm2, %xmm1
2067-
; X64-SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
20682062
; X64-SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
20692063
; X64-SSE-NEXT: movq %xmm0, (%rax,%rsi,4)
20702064
; X64-SSE-NEXT: retq

‎llvm/test/CodeGen/X86/vec_extract-sse4.ll

+2-6
Original file line numberDiff line numberDiff line change
@@ -25,17 +25,13 @@ define void @t1(float* %R, <4 x float>* %P1) nounwind {
2525
define float @t2(<4 x float>* %P1) nounwind {
2626
; X32-LABEL: t2:
2727
; X32: # %bb.0:
28-
; X32-NEXT: pushl %eax
2928
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
30-
; X32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
31-
; X32-NEXT: movss %xmm0, (%esp)
32-
; X32-NEXT: flds (%esp)
33-
; X32-NEXT: popl %eax
29+
; X32-NEXT: flds 8(%eax)
3430
; X32-NEXT: retl
3531
;
3632
; X64-LABEL: t2:
3733
; X64: # %bb.0:
38-
; X64-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
34+
; X64-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
3935
; X64-NEXT: retq
4036
%X = load <4 x float>, <4 x float>* %P1
4137
%tmp = extractelement <4 x float> %X, i32 2

‎llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll

-2
Original file line numberDiff line numberDiff line change
@@ -1863,8 +1863,6 @@ define i8 @test_v16i8(<16 x i8> %a0) {
18631863
; 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
18641864
; 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
18651865
; AVX2-NEXT: vpmullw %xmm1, %xmm0, %xmm0
1866-
; AVX2-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
1867-
; AVX2-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
18681866
; AVX2-NEXT: vpextrb $0, %xmm0, %eax
18691867
; AVX2-NEXT: # kill: def $al killed $al killed $eax
18701868
; AVX2-NEXT: vzeroupper

‎llvm/test/CodeGen/X86/vector-reduce-mul.ll

+1-9
Original file line numberDiff line numberDiff line change
@@ -808,7 +808,6 @@ define i32 @test_v2i32(<2 x i32> %a0) {
808808
; AVX512DQ: # %bb.0:
809809
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
810810
; AVX512DQ-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
811-
; AVX512DQ-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
812811
; AVX512DQ-NEXT: vpmullq %zmm1, %zmm0, %zmm0
813812
; AVX512DQ-NEXT: vmovd %xmm0, %eax
814813
; AVX512DQ-NEXT: vzeroupper
@@ -817,7 +816,6 @@ define i32 @test_v2i32(<2 x i32> %a0) {
817816
; AVX512DQVL-LABEL: test_v2i32:
818817
; AVX512DQVL: # %bb.0:
819818
; AVX512DQVL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
820-
; AVX512DQVL-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
821819
; AVX512DQVL-NEXT: vpmullq %xmm1, %xmm0, %xmm0
822820
; AVX512DQVL-NEXT: vmovd %xmm0, %eax
823821
; AVX512DQVL-NEXT: retq
@@ -1178,7 +1176,6 @@ define i16 @test_v2i16(<2 x i16> %a0) {
11781176
; AVX512DQ: # %bb.0:
11791177
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
11801178
; AVX512DQ-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1181-
; AVX512DQ-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
11821179
; AVX512DQ-NEXT: vpmullq %zmm1, %zmm0, %zmm0
11831180
; AVX512DQ-NEXT: vmovd %xmm0, %eax
11841181
; AVX512DQ-NEXT: # kill: def $ax killed $ax killed $eax
@@ -1188,7 +1185,6 @@ define i16 @test_v2i16(<2 x i16> %a0) {
11881185
; AVX512DQVL-LABEL: test_v2i16:
11891186
; AVX512DQVL: # %bb.0:
11901187
; AVX512DQVL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1191-
; AVX512DQVL-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
11921188
; AVX512DQVL-NEXT: vpmullq %xmm1, %xmm0, %xmm0
11931189
; AVX512DQVL-NEXT: vmovd %xmm0, %eax
11941190
; AVX512DQVL-NEXT: # kill: def $ax killed $ax killed $eax
@@ -1210,7 +1206,7 @@ define i16 @test_v4i16(<4 x i16> %a0) {
12101206
; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
12111207
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
12121208
; SSE2-NEXT: pmuludq %xmm0, %xmm1
1213-
; SSE2-NEXT: movd %xmm1, %eax
1209+
; SSE2-NEXT: pextrw $0, %xmm1, %eax
12141210
; SSE2-NEXT: # kill: def $ax killed $ax killed $eax
12151211
; SSE2-NEXT: retq
12161212
;
@@ -1658,7 +1654,6 @@ define i8 @test_v2i8(<2 x i8> %a0) {
16581654
; AVX512DQ: # %bb.0:
16591655
; AVX512DQ-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
16601656
; AVX512DQ-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1661-
; AVX512DQ-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
16621657
; AVX512DQ-NEXT: vpmullq %zmm1, %zmm0, %zmm0
16631658
; AVX512DQ-NEXT: vpextrb $0, %xmm0, %eax
16641659
; AVX512DQ-NEXT: # kill: def $al killed $al killed $eax
@@ -1668,7 +1663,6 @@ define i8 @test_v2i8(<2 x i8> %a0) {
16681663
; AVX512DQVL-LABEL: test_v2i8:
16691664
; AVX512DQVL: # %bb.0:
16701665
; AVX512DQVL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
1671-
; AVX512DQVL-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
16721666
; AVX512DQVL-NEXT: vpmullq %xmm1, %xmm0, %xmm0
16731667
; AVX512DQVL-NEXT: vpextrb $0, %xmm0, %eax
16741668
; AVX512DQVL-NEXT: # kill: def $al killed $al killed $eax
@@ -1899,8 +1893,6 @@ define i8 @test_v16i8(<16 x i8> %a0) {
18991893
; 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
19001894
; 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
19011895
; AVX2-NEXT: vpmullw %xmm1, %xmm0, %xmm0
1902-
; AVX2-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
1903-
; AVX2-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
19041896
; AVX2-NEXT: vpextrb $0, %xmm0, %eax
19051897
; AVX2-NEXT: # kill: def $al killed $al killed $eax
19061898
; AVX2-NEXT: vzeroupper

0 commit comments

Comments
 (0)
Please sign in to comment.