[X86][SSE] Add PACKUS support to LowerTruncate
Similar to the existing code to lower to PACKSS, we can use PACKUS if the input vector's leading zero bits extend all the way to the packed/truncated value.
We have to account for pre-SSE41 targets not supporting PACKUSDW
llvm-svn: 317128
diff --git a/llvm/test/CodeGen/X86/avg.ll b/llvm/test/CodeGen/X86/avg.ll
index 5b36ccf..508f10e 100644
--- a/llvm/test/CodeGen/X86/avg.ll
+++ b/llvm/test/CodeGen/X86/avg.ll
@@ -2905,17 +2905,16 @@
; AVX2-NEXT: vpsrld $1, %ymm1, %ymm1
; AVX2-NEXT: vpsrld $1, %ymm2, %ymm2
; AVX2-NEXT: vpsrld $1, %ymm3, %ymm3
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
-; AVX2-NEXT: vpshufb %ymm4, %ymm3, %ymm3
-; AVX2-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT: vpshufb %ymm4, %ymm2, %ymm2
-; AVX2-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT: vinserti128 $1, %xmm2, %ymm3, %ymm2
-; AVX2-NEXT: vpshufb %ymm4, %ymm1, %ymm1
-; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT: vpshufb %ymm4, %ymm0, %ymm0
-; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm4
+; AVX2-NEXT: vpackusdw %xmm4, %xmm3, %xmm3
+; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm4
+; AVX2-NEXT: vpackusdw %xmm4, %xmm2, %xmm2
+; AVX2-NEXT: vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm3
+; AVX2-NEXT: vpackusdw %xmm3, %xmm1, %xmm1
+; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm3
+; AVX2-NEXT: vpackusdw %xmm3, %xmm0, %xmm0
+; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX2-NEXT: vmovdqu %ymm0, (%rax)
; AVX2-NEXT: vmovdqu %ymm2, (%rax)
; AVX2-NEXT: vzeroupper
diff --git a/llvm/test/CodeGen/X86/combine-srl.ll b/llvm/test/CodeGen/X86/combine-srl.ll
index 1d064b3..9f7f8a97 100644
--- a/llvm/test/CodeGen/X86/combine-srl.ll
+++ b/llvm/test/CodeGen/X86/combine-srl.ll
@@ -181,9 +181,8 @@
; AVX-LABEL: combine_vec_lshr_trunc_lshr0:
; AVX: # BB#0:
; AVX-NEXT: vpsrlq $48, %ymm0, %ymm0
-; AVX-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
-; AVX-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX-NEXT: vpackusdw %xmm1, %xmm0, %xmm0
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
%1 = lshr <4 x i64> %x, <i64 32, i64 32, i64 32, i64 32>
diff --git a/llvm/test/CodeGen/X86/pr31773.ll b/llvm/test/CodeGen/X86/pr31773.ll
index 5b1ea9a..d7ae04b 100644
--- a/llvm/test/CodeGen/X86/pr31773.ll
+++ b/llvm/test/CodeGen/X86/pr31773.ll
@@ -7,14 +7,11 @@
define <16 x i8> @usat_trunc_wb_256(<16 x i16> %i) {
; AVX-LABEL: usat_trunc_wb_256:
; AVX: # BB#0:
-; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
-; AVX-NEXT: vpminuw %xmm1, %xmm0, %xmm2
-; AVX-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX-NEXT: vpminuw %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
-; AVX-NEXT: vpshufb %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpshufb %xmm1, %xmm2, %xmm1
-; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX-NEXT: vextractf128 $1, %ymm0, %xmm1
+; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; AVX-NEXT: vpminuw %xmm2, %xmm1, %xmm1
+; AVX-NEXT: vpminuw %xmm2, %xmm0, %xmm0
+; AVX-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
;
@@ -32,14 +29,11 @@
define <8 x i16> @usat_trunc_dw_256(<8 x i32> %i) {
; AVX-LABEL: usat_trunc_dw_256:
; AVX: # BB#0:
-; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
-; AVX-NEXT: vpminud %xmm1, %xmm0, %xmm2
-; AVX-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX-NEXT: vpminud %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX-NEXT: vpshufb %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpshufb %xmm1, %xmm2, %xmm1
-; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX-NEXT: vextractf128 $1, %ymm0, %xmm1
+; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = [65535,65535,65535,65535]
+; AVX-NEXT: vpminud %xmm2, %xmm1, %xmm1
+; AVX-NEXT: vpminud %xmm2, %xmm0, %xmm0
+; AVX-NEXT: vpackusdw %xmm1, %xmm0, %xmm0
; AVX-NEXT: vzeroupper
; AVX-NEXT: retq
;
diff --git a/llvm/test/CodeGen/X86/psubus.ll b/llvm/test/CodeGen/X86/psubus.ll
index e8bb3b8..6e38f06 100644
--- a/llvm/test/CodeGen/X86/psubus.ll
+++ b/llvm/test/CodeGen/X86/psubus.ll
@@ -1555,14 +1555,11 @@
;
; AVX1-LABEL: psubus_8i32_max:
; AVX1: # BB#0: # %vector.ph
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [65535,65535,65535,65535]
-; AVX1-NEXT: vpminud %xmm4, %xmm3, %xmm3
-; AVX1-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX1-NEXT: vpminud %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
+; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [65535,65535,65535,65535]
+; AVX1-NEXT: vpminud %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpminud %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpackusdw %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vpsubusw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
@@ -1571,8 +1568,8 @@
; AVX2: # BB#0: # %vector.ph
; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm2 = [65535,65535,65535,65535,65535,65535,65535,65535]
; AVX2-NEXT: vpminud %ymm2, %ymm1, %ymm1
-; AVX2-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
-; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT: vpackusdw %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpsubusw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
@@ -2101,20 +2098,15 @@
;
; AVX1-LABEL: psubus_16i32_max:
; AVX1: # BB#0: # %vector.ph
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [65535,65535,65535,65535]
-; AVX1-NEXT: vpminud %xmm5, %xmm4, %xmm4
-; AVX1-NEXT: vpshufb %xmm3, %xmm4, %xmm4
-; AVX1-NEXT: vpminud %xmm5, %xmm2, %xmm2
-; AVX1-NEXT: vpshufb %xmm3, %xmm2, %xmm2
-; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm4[0]
-; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT: vpminud %xmm5, %xmm4, %xmm4
-; AVX1-NEXT: vpshufb %xmm3, %xmm4, %xmm4
-; AVX1-NEXT: vpminud %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm4[0]
+; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [65535,65535,65535,65535]
+; AVX1-NEXT: vpminud %xmm4, %xmm3, %xmm3
+; AVX1-NEXT: vpminud %xmm4, %xmm2, %xmm2
+; AVX1-NEXT: vpackusdw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT: vpminud %xmm4, %xmm3, %xmm3
+; AVX1-NEXT: vpminud %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpackusdw %xmm3, %xmm1, %xmm1
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
; AVX1-NEXT: vpsubusw %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpsubusw %xmm1, %xmm0, %xmm0
@@ -2130,23 +2122,22 @@
;
; AVX2-LABEL: psubus_16i32_max:
; AVX2: # BB#0: # %vector.ph
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
-; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm4 = [65535,65535,65535,65535,65535,65535,65535,65535]
-; AVX2-NEXT: vpminud %ymm4, %ymm1, %ymm1
-; AVX2-NEXT: vpshufb %ymm3, %ymm1, %ymm1
-; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT: vpminud %ymm4, %ymm2, %ymm2
-; AVX2-NEXT: vpshufb %ymm3, %ymm2, %ymm2
-; AVX2-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
+; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm3 = [65535,65535,65535,65535,65535,65535,65535,65535]
+; AVX2-NEXT: vpminud %ymm3, %ymm1, %ymm1
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm4
+; AVX2-NEXT: vpackusdw %xmm4, %xmm1, %xmm1
+; AVX2-NEXT: vpminud %ymm3, %ymm2, %ymm2
+; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
+; AVX2-NEXT: vpackusdw %xmm3, %xmm2, %xmm2
; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT: vpackusdw %ymm0, %ymm3, %ymm3
-; AVX2-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
+; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm4
+; AVX2-NEXT: vpackusdw %xmm4, %xmm3, %xmm3
+; AVX2-NEXT: vpsubusw %xmm1, %xmm3, %xmm1
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT: vpmovzxwd {{.*#+}} 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
-; AVX2-NEXT: vpackusdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm3
+; AVX2-NEXT: vpackusdw %xmm3, %xmm0, %xmm0
; AVX2-NEXT: vpsubusw %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vpsubusw %xmm1, %xmm3, %xmm1
; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
@@ -2239,14 +2230,11 @@
;
; AVX1-LABEL: psubus_i16_i32_max_swapped:
; AVX1: # BB#0: # %vector.ph
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [65535,65535,65535,65535]
-; AVX1-NEXT: vpminud %xmm4, %xmm3, %xmm3
-; AVX1-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX1-NEXT: vpminud %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
+; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [65535,65535,65535,65535]
+; AVX1-NEXT: vpminud %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpminud %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpackusdw %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vpsubusw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
@@ -2255,8 +2243,8 @@
; AVX2: # BB#0: # %vector.ph
; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm2 = [65535,65535,65535,65535,65535,65535,65535,65535]
; AVX2-NEXT: vpminud %ymm2, %ymm1, %ymm1
-; AVX2-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
-; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT: vpackusdw %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpsubusw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
@@ -2353,14 +2341,11 @@
;
; AVX1-LABEL: psubus_i16_i32_min:
; AVX1: # BB#0: # %vector.ph
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [65535,65535,65535,65535]
-; AVX1-NEXT: vpminud %xmm4, %xmm3, %xmm3
-; AVX1-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX1-NEXT: vpminud %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
+; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [65535,65535,65535,65535]
+; AVX1-NEXT: vpminud %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpminud %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpackusdw %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vpsubusw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
@@ -2369,8 +2354,8 @@
; AVX2: # BB#0: # %vector.ph
; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm2 = [65535,65535,65535,65535,65535,65535,65535,65535]
; AVX2-NEXT: vpminud %ymm2, %ymm1, %ymm1
-; AVX2-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
-; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT: vpackusdw %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpsubusw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
diff --git a/llvm/test/CodeGen/X86/vector-trunc.ll b/llvm/test/CodeGen/X86/vector-trunc.ll
index e9da847..dc08d88 100644
--- a/llvm/test/CodeGen/X86/vector-trunc.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc.ll
@@ -483,9 +483,8 @@
; AVX2-LABEL: trunc8i32_8i16_lshr:
; AVX2: # BB#0: # %entry
; AVX2-NEXT: vpsrld $16, %ymm0, %ymm0
-; AVX2-NEXT: vpackusdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT: vpackusdw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
@@ -814,12 +813,12 @@
;
; AVX2-LABEL: trunc16i32_16i16_lshr:
; AVX2: # BB#0: # %entry
-; AVX2-NEXT: vpsrld $16, %ymm1, %ymm1
; AVX2-NEXT: vpsrld $16, %ymm0, %ymm0
-; AVX2-NEXT: vpackusdw %ymm0, %ymm0, %ymm0
-; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT: vpackusdw %ymm0, %ymm1, %ymm1
-; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT: vpsrld $16, %ymm1, %ymm1
+; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT: vpackusdw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-NEXT: vpackusdw %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX2-NEXT: vmovdqu %ymm0, (%rax)
; AVX2-NEXT: vzeroupper
@@ -1901,8 +1900,8 @@
; AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2
; AVX1-NEXT: vpsrlw $8, %xmm1, %xmm1
; AVX1-NEXT: vpackuswb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT: vmovups %ymm0, (%rsi)
+; AVX1-NEXT: vmovdqu %xmm0, (%rsi)
+; AVX1-NEXT: vmovdqu %xmm1, 16(%rsi)
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@@ -1916,8 +1915,8 @@
; AVX2-NEXT: vpackuswb %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-NEXT: vpackuswb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-NEXT: vmovdqu %ymm0, (%rsi)
+; AVX2-NEXT: vmovdqu %xmm0, (%rsi)
+; AVX2-NEXT: vmovdqu %xmm1, 16(%rsi)
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;