OSchip
/
llvm-project
mirror of https://github.com/THU-DSP-LAB/llvm-project.git
13
0
Fork 6
Code Issues Packages Projects Releases Wiki Activity

[X86] SRL non-LSB extracts when folding to truncating broadcasts. 

Now that we recognize this, we can support it instead of bailing out.
That is, we can fold:
  (v8i16 (shufflevector
    (v8i16 (bitcast (v4i32 (build_vector X, Y, ...)))),
    <1,1,...,1>))
into:
  (v8i16 (vbroadcast (i16 (trunc (srl Y, 16)))))

llvm-svn: 252362
This commit is contained in:
Ahmed Bougacha 2015-11-06 23:16:43 +00:00
parent 68614a36d1
commit 05a0514b12
5 changed files with 119 additions and 62 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -7896,15 +7896,20 @@ static SDValue lowerVectorShuffleAsTruncBroadcast(SDLoc DL, MVT VT, SDValue V0,
const unsigned Scale = V0EltSize / EltSize; const unsigned Scale = V0EltSize / EltSize;
const unsigned V0BroadcastIdx = BroadcastIdx / Scale; const unsigned V0BroadcastIdx = BroadcastIdx / Scale;
// If we're extracting non-least-significant bits, this isn't a truncation.
if (BroadcastIdx % Scale)
return SDValue();
if ((V0Opc != ISD::SCALAR_TO_VECTOR || V0BroadcastIdx != 0) && if ((V0Opc != ISD::SCALAR_TO_VECTOR || V0BroadcastIdx != 0) &&
V0Opc != ISD::BUILD_VECTOR) V0Opc != ISD::BUILD_VECTOR)
return SDValue(); return SDValue();
SDValue Scalar = V0.getOperand(V0BroadcastIdx); SDValue Scalar = V0.getOperand(V0BroadcastIdx);
// If we're extracting non-least-significant bits, shift so we can truncate.
// Hopefully, we can fold away the trunc/srl/load into the broadcast.
// Even if we can't (and !isShuffleFoldableLoad(Scalar)), prefer
// vpbroadcast+vmovd+shr to vpshufb(m)+vmovd.
if (const int OffsetIdx = BroadcastIdx % Scale)
Scalar = DAG.getNode(ISD::SRL, DL, Scalar.getValueType(), Scalar,
DAG.getConstant(OffsetIdx * EltSize, DL, Scalar.getValueType()));
return DAG.getNode(X86ISD::VBROADCAST, DL, VT, return DAG.getNode(X86ISD::VBROADCAST, DL, VT,
DAG.getNode(ISD::TRUNCATE, DL, EltVT, Scalar)); DAG.getNode(ISD::TRUNCATE, DL, EltVT, Scalar));
} }

70
llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll
View File

@ -1482,11 +1482,16 @@ define <16 x i8> @insert_dup_elt1_mem_v16i8_i32(i32* %ptr) {
; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] ; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; SSE41-NEXT: retq ; SSE41-NEXT: retq
; ;
; AVX-LABEL: insert_dup_elt1_mem_v16i8_i32: ; AVX1-LABEL: insert_dup_elt1_mem_v16i8_i32:
; AVX: # BB#0: ; AVX1: # BB#0:
; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero ; AVX1-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX-NEXT: retq ; AVX1-NEXT: retq
;
; AVX2-LABEL: insert_dup_elt1_mem_v16i8_i32:
; AVX2: # BB#0:
; AVX2-NEXT: vpbroadcastb 1(%rdi), %xmm0
; AVX2-NEXT: retq
%tmp = load i32, i32* %ptr, align 4 %tmp = load i32, i32* %ptr, align 4
%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
%tmp2 = bitcast <4 x i32> %tmp1 to <16 x i8> %tmp2 = bitcast <4 x i32> %tmp1 to <16 x i8>
@ -1516,11 +1521,16 @@ define <16 x i8> @insert_dup_elt2_mem_v16i8_i32(i32* %ptr) {
; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2] ; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
; SSE41-NEXT: retq ; SSE41-NEXT: retq
; ;
; AVX-LABEL: insert_dup_elt2_mem_v16i8_i32: ; AVX1-LABEL: insert_dup_elt2_mem_v16i8_i32:
; AVX: # BB#0: ; AVX1: # BB#0:
; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero ; AVX1-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2] ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
; AVX-NEXT: retq ; AVX1-NEXT: retq
;
; AVX2-LABEL: insert_dup_elt2_mem_v16i8_i32:
; AVX2: # BB#0:
; AVX2-NEXT: vpbroadcastb 2(%rdi), %xmm0
; AVX2-NEXT: retq
%tmp = load i32, i32* %ptr, align 4 %tmp = load i32, i32* %ptr, align 4
%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
%tmp2 = bitcast <4 x i32> %tmp1 to <16 x i8> %tmp2 = bitcast <4 x i32> %tmp1 to <16 x i8>
@ -1553,12 +1563,20 @@ define <16 x i8> @insert_dup_elt1_mem_v16i8_sext_i8(i8* %ptr) {
; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] ; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; SSE41-NEXT: retq ; SSE41-NEXT: retq
; ;
; AVX-LABEL: insert_dup_elt1_mem_v16i8_sext_i8: ; AVX1-LABEL: insert_dup_elt1_mem_v16i8_sext_i8:
; AVX: # BB#0: ; AVX1: # BB#0:
; AVX-NEXT: movsbl (%rdi), %eax ; AVX1-NEXT: movsbl (%rdi), %eax
; AVX-NEXT: vmovd %eax, %xmm0 ; AVX1-NEXT: vmovd %eax, %xmm0
; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX-NEXT: retq ; AVX1-NEXT: retq
;
; AVX2-LABEL: insert_dup_elt1_mem_v16i8_sext_i8:
; AVX2: # BB#0:
; AVX2-NEXT: movsbl (%rdi), %eax
; AVX2-NEXT: shrl $8, %eax
; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vpbroadcastb %xmm0, %xmm0
; AVX2-NEXT: retq
%tmp = load i8, i8* %ptr, align 1 %tmp = load i8, i8* %ptr, align 1
%tmp1 = sext i8 %tmp to i32 %tmp1 = sext i8 %tmp to i32
%tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 0 %tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 0
@ -1592,12 +1610,20 @@ define <16 x i8> @insert_dup_elt2_mem_v16i8_sext_i8(i8* %ptr) {
; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2] ; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
; SSE41-NEXT: retq ; SSE41-NEXT: retq
; ;
; AVX-LABEL: insert_dup_elt2_mem_v16i8_sext_i8: ; AVX1-LABEL: insert_dup_elt2_mem_v16i8_sext_i8:
; AVX: # BB#0: ; AVX1: # BB#0:
; AVX-NEXT: movsbl (%rdi), %eax ; AVX1-NEXT: movsbl (%rdi), %eax
; AVX-NEXT: vmovd %eax, %xmm0 ; AVX1-NEXT: vmovd %eax, %xmm0
; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2] ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
; AVX-NEXT: retq ; AVX1-NEXT: retq
;
; AVX2-LABEL: insert_dup_elt2_mem_v16i8_sext_i8:
; AVX2: # BB#0:
; AVX2-NEXT: movsbl (%rdi), %eax
; AVX2-NEXT: shrl $16, %eax
; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vpbroadcastb %xmm0, %xmm0
; AVX2-NEXT: retq
%tmp = load i8, i8* %ptr, align 1 %tmp = load i8, i8* %ptr, align 1
%tmp1 = sext i8 %tmp to i32 %tmp1 = sext i8 %tmp to i32
%tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 0 %tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 0

74
llvm/test/CodeGen/X86/vector-shuffle-128-v8.ll
View File

@ -2250,11 +2250,18 @@ define <8 x i16> @insert_dup_elt1_mem_v8i16_i32(i32* %ptr) {
; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]
; SSE41-NEXT: retq ; SSE41-NEXT: retq
; ;
; AVX-LABEL: insert_dup_elt1_mem_v8i16_i32: ; AVX1-LABEL: insert_dup_elt1_mem_v8i16_i32:
; AVX: # BB#0: ; AVX1: # BB#0:
; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero ; AVX1-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]
; AVX-NEXT: retq ; AVX1-NEXT: retq
;
; AVX2-LABEL: insert_dup_elt1_mem_v8i16_i32:
; AVX2: # BB#0:
; AVX2-NEXT: movzwl 2(%rdi), %eax
; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vpbroadcastw %xmm0, %xmm0
; AVX2-NEXT: retq
%tmp = load i32, i32* %ptr, align 4 %tmp = load i32, i32* %ptr, align 4
%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
%tmp2 = bitcast <4 x i32> %tmp1 to <8 x i16> %tmp2 = bitcast <4 x i32> %tmp1 to <8 x i16>
@ -2283,11 +2290,18 @@ define <8 x i16> @insert_dup_elt3_mem_v8i16_i32(i32* %ptr) {
; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]
; SSE41-NEXT: retq ; SSE41-NEXT: retq
; ;
; AVX-LABEL: insert_dup_elt3_mem_v8i16_i32: ; AVX1-LABEL: insert_dup_elt3_mem_v8i16_i32:
; AVX: # BB#0: ; AVX1: # BB#0:
; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero ; AVX1-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]
; AVX-NEXT: retq ; AVX1-NEXT: retq
;
; AVX2-LABEL: insert_dup_elt3_mem_v8i16_i32:
; AVX2: # BB#0:
; AVX2-NEXT: movzwl 2(%rdi), %eax
; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vpbroadcastw %xmm0, %xmm0
; AVX2-NEXT: retq
%tmp = load i32, i32* %ptr, align 4 %tmp = load i32, i32* %ptr, align 4
%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 1 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 1
%tmp2 = bitcast <4 x i32> %tmp1 to <8 x i16> %tmp2 = bitcast <4 x i32> %tmp1 to <8 x i16>
@ -2319,12 +2333,20 @@ define <8 x i16> @insert_dup_elt1_mem_v8i16_sext_i16(i16* %ptr) {
; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]
; SSE41-NEXT: retq ; SSE41-NEXT: retq
; ;
; AVX-LABEL: insert_dup_elt1_mem_v8i16_sext_i16: ; AVX1-LABEL: insert_dup_elt1_mem_v8i16_sext_i16:
; AVX: # BB#0: ; AVX1: # BB#0:
; AVX-NEXT: movswl (%rdi), %eax ; AVX1-NEXT: movswl (%rdi), %eax
; AVX-NEXT: vmovd %eax, %xmm0 ; AVX1-NEXT: vmovd %eax, %xmm0
; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]
; AVX-NEXT: retq ; AVX1-NEXT: retq
;
; AVX2-LABEL: insert_dup_elt1_mem_v8i16_sext_i16:
; AVX2: # BB#0:
; AVX2-NEXT: movswl (%rdi), %eax
; AVX2-NEXT: shrl $16, %eax
; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vpbroadcastw %xmm0, %xmm0
; AVX2-NEXT: retq
%tmp = load i16, i16* %ptr, align 2 %tmp = load i16, i16* %ptr, align 2
%tmp1 = sext i16 %tmp to i32 %tmp1 = sext i16 %tmp to i32
%tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 0 %tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 0
@ -2357,12 +2379,20 @@ define <8 x i16> @insert_dup_elt3_mem_v8i16_sext_i16(i16* %ptr) {
; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; SSE41-NEXT: pshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]
; SSE41-NEXT: retq ; SSE41-NEXT: retq
; ;
; AVX-LABEL: insert_dup_elt3_mem_v8i16_sext_i16: ; AVX1-LABEL: insert_dup_elt3_mem_v8i16_sext_i16:
; AVX: # BB#0: ; AVX1: # BB#0:
; AVX-NEXT: movswl (%rdi), %eax ; AVX1-NEXT: movswl (%rdi), %eax
; AVX-NEXT: vmovd %eax, %xmm0 ; AVX1-NEXT: vmovd %eax, %xmm0
; AVX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]
; AVX-NEXT: retq ; AVX1-NEXT: retq
;
; AVX2-LABEL: insert_dup_elt3_mem_v8i16_sext_i16:
; AVX2: # BB#0:
; AVX2-NEXT: movswl (%rdi), %eax
; AVX2-NEXT: shrl $16, %eax
; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vpbroadcastw %xmm0, %xmm0
; AVX2-NEXT: retq
%tmp = load i16, i16* %ptr, align 2 %tmp = load i16, i16* %ptr, align 2
%tmp1 = sext i16 %tmp to i32 %tmp1 = sext i16 %tmp to i32
%tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 1 %tmp2 = insertelement <4 x i32> zeroinitializer, i32 %tmp1, i32 1

12
llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
View File

@ -3342,9 +3342,9 @@ define <16 x i16> @insert_dup_elt1_mem_v16i16_i32(i32* %ptr) #0 {
; ;
; AVX2-LABEL: insert_dup_elt1_mem_v16i16_i32: ; AVX2-LABEL: insert_dup_elt1_mem_v16i16_i32:
; AVX2: # BB#0: ; AVX2: # BB#0:
; AVX2-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero ; AVX2-NEXT: movzwl 2(%rdi), %eax
; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0 ; AVX2-NEXT: vpbroadcastw %xmm0, %ymm0
; AVX2-NEXT: retq ; AVX2-NEXT: retq
%tmp = load i32, i32* %ptr, align 4 %tmp = load i32, i32* %ptr, align 4
%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
@ -3363,9 +3363,9 @@ define <16 x i16> @insert_dup_elt3_mem_v16i16_i32(i32* %ptr) #0 {
; ;
; AVX2-LABEL: insert_dup_elt3_mem_v16i16_i32: ; AVX2-LABEL: insert_dup_elt3_mem_v16i16_i32:
; AVX2: # BB#0: ; AVX2: # BB#0:
; AVX2-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero ; AVX2-NEXT: movzwl 2(%rdi), %eax
; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3] ; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0 ; AVX2-NEXT: vpbroadcastw %xmm0, %ymm0
; AVX2-NEXT: retq ; AVX2-NEXT: retq
%tmp = load i32, i32* %ptr, align 4 %tmp = load i32, i32* %ptr, align 4
%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 1 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 1

12
llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
View File

@ -2029,9 +2029,7 @@ define <32 x i8> @insert_dup_elt1_mem_v32i8_i32(i32* %ptr) {
; ;
; AVX2-LABEL: insert_dup_elt1_mem_v32i8_i32: ; AVX2-LABEL: insert_dup_elt1_mem_v32i8_i32:
; AVX2: # BB#0: ; AVX2: # BB#0:
; AVX2-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero ; AVX2-NEXT: vpbroadcastb 1(%rdi), %ymm0
; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-NEXT: retq ; AVX2-NEXT: retq
%tmp = load i32, i32* %ptr, align 4 %tmp = load i32, i32* %ptr, align 4
%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
@ -2050,9 +2048,7 @@ define <32 x i8> @insert_dup_elt3_mem_v32i8_i32(i32* %ptr) {
; ;
; AVX2-LABEL: insert_dup_elt3_mem_v32i8_i32: ; AVX2-LABEL: insert_dup_elt3_mem_v32i8_i32:
; AVX2: # BB#0: ; AVX2: # BB#0:
; AVX2-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero ; AVX2-NEXT: vpbroadcastb 3(%rdi), %ymm0
; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3]
; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-NEXT: retq ; AVX2-NEXT: retq
%tmp = load i32, i32* %ptr, align 4 %tmp = load i32, i32* %ptr, align 4
%tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0 %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
@ -2073,9 +2069,9 @@ define <32 x i8> @insert_dup_elt1_mem_v32i8_sext_i8(i8* %ptr) {
; AVX2-LABEL: insert_dup_elt1_mem_v32i8_sext_i8: ; AVX2-LABEL: insert_dup_elt1_mem_v32i8_sext_i8:
; AVX2: # BB#0: ; AVX2: # BB#0:
; AVX2-NEXT: movsbl (%rdi), %eax ; AVX2-NEXT: movsbl (%rdi), %eax
; AVX2-NEXT: shrl $8, %eax
; AVX2-NEXT: vmovd %eax, %xmm0 ; AVX2-NEXT: vmovd %eax, %xmm0
; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] ; AVX2-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-NEXT: retq ; AVX2-NEXT: retq
%tmp = load i8, i8* %ptr, align 1 %tmp = load i8, i8* %ptr, align 1
%tmp1 = sext i8 %tmp to i32 %tmp1 = sext i8 %tmp to i32