summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/vector-trunc-packus.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/vector-trunc-packus.ll')
-rw-r--r--llvm/test/CodeGen/X86/vector-trunc-packus.ll138
1 files changed, 44 insertions, 94 deletions
diff --git a/llvm/test/CodeGen/X86/vector-trunc-packus.ll b/llvm/test/CodeGen/X86/vector-trunc-packus.ll
index a95a81f435e..d382e0db3d3 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-packus.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-packus.ll
@@ -119,47 +119,42 @@ define <2 x i32> @trunc_packus_v2i64_v2i32(<2 x i64> %a0) {
; AVX512F-LABEL: trunc_packus_v2i64_v2i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT: vmovdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512F-NEXT: vpmovusqd %zmm0, %ymm0
+; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: trunc_packus_v2i64_v2i32:
; AVX512VL: # %bb.0:
-; AVX512VL-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512VL-NEXT: vpmovusqd %xmm0, %xmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: trunc_packus_v2i64_v2i32:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512BW-NEXT: vpmovusqd %zmm0, %ymm0
+; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
; AVX512BWVL-LABEL: trunc_packus_v2i64_v2i32:
; AVX512BWVL: # %bb.0:
-; AVX512BWVL-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; AVX512BWVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BWVL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512BWVL-NEXT: vpmovusqd %xmm0, %xmm0
; AVX512BWVL-NEXT: retq
;
; SKX-LABEL: trunc_packus_v2i64_v2i32:
; SKX: # %bb.0:
-; SKX-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; SKX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; SKX-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; SKX-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SKX-NEXT: vpmovusqd %xmm0, %xmm0
; SKX-NEXT: retq
%1 = icmp slt <2 x i64> %a0, <i64 4294967295, i64 4294967295>
%2 = select <2 x i1> %1, <2 x i64> %a0, <2 x i64> <i64 4294967295, i64 4294967295>
@@ -277,11 +272,9 @@ define void @trunc_packus_v2i64_v2i32_store(<2 x i64> %a0, <2 x i32>* %p1) {
; AVX512F-LABEL: trunc_packus_v2i64_v2i32_store:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT: vmovdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512F-NEXT: vpmovusqd %zmm0, %ymm0
; AVX512F-NEXT: vmovq %xmm0, (%rdi)
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
@@ -296,11 +289,9 @@ define void @trunc_packus_v2i64_v2i32_store(<2 x i64> %a0, <2 x i32>* %p1) {
; AVX512BW-LABEL: trunc_packus_v2i64_v2i32_store:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512BW-NEXT: vpmovusqd %zmm0, %ymm0
; AVX512BW-NEXT: vmovq %xmm0, (%rdi)
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
@@ -536,10 +527,9 @@ define <4 x i32> @trunc_packus_v4i64_v4i32(<4 x i64> %a0) {
; AVX512F-LABEL: trunc_packus_v4i64_v4i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT: vpmovusqd %zmm0, %ymm0
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
@@ -555,10 +545,9 @@ define <4 x i32> @trunc_packus_v4i64_v4i32(<4 x i64> %a0) {
; AVX512BW-LABEL: trunc_packus_v4i64_v4i32:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT: vpmovusqd %zmm0, %ymm0
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
@@ -1132,48 +1121,40 @@ define <2 x i16> @trunc_packus_v2i64_v2i16(<2 x i64> %a0) {
; AVX512F-LABEL: trunc_packus_v2i64_v2i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT: vmovdqa {{.*#+}} xmm1 = [65535,65535]
-; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512F-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512F-NEXT: vpmovusqw %zmm0, %xmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: trunc_packus_v2i64_v2i16:
; AVX512VL: # %bb.0:
-; AVX512VL-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX512VL-NEXT: vpmovusqw %xmm0, %xmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: trunc_packus_v2i64_v2i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = [65535,65535]
-; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX512BW-NEXT: vpmovusqw %zmm0, %xmm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
; AVX512BWVL-LABEL: trunc_packus_v2i64_v2i16:
; AVX512BWVL: # %bb.0:
-; AVX512BWVL-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; AVX512BWVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BWVL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX512BWVL-NEXT: vpmovusqw %xmm0, %xmm0
; AVX512BWVL-NEXT: retq
;
; SKX-LABEL: trunc_packus_v2i64_v2i16:
; SKX: # %bb.0:
-; SKX-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; SKX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; SKX-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; SKX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
+; SKX-NEXT: vpmovusqw %xmm0, %xmm0
; SKX-NEXT: retq
%1 = icmp slt <2 x i64> %a0, <i64 65535, i64 65535>
%2 = select <2 x i1> %1, <2 x i64> %a0, <2 x i64> <i64 65535, i64 65535>
@@ -1320,12 +1301,9 @@ define void @trunc_packus_v2i64_v2i16_store(<2 x i64> %a0, <2 x i16> *%p1) {
; AVX512F-LABEL: trunc_packus_v2i64_v2i16_store:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT: vmovdqa {{.*#+}} xmm1 = [65535,65535]
-; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512F-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512F-NEXT: vpmovusqw %zmm0, %xmm0
; AVX512F-NEXT: vmovd %xmm0, (%rdi)
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
@@ -1340,11 +1318,9 @@ define void @trunc_packus_v2i64_v2i16_store(<2 x i64> %a0, <2 x i16> *%p1) {
; AVX512BW-LABEL: trunc_packus_v2i64_v2i16_store:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = [65535,65535]
-; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX512BW-NEXT: vpmovusqw %zmm0, %xmm0
; AVX512BW-NEXT: vmovd %xmm0, (%rdi)
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
@@ -1602,10 +1578,9 @@ define <4 x i16> @trunc_packus_v4i64_v4i16(<4 x i64> %a0) {
; AVX512F-LABEL: trunc_packus_v4i64_v4i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT: vpmovusqw %zmm0, %xmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
@@ -1620,10 +1595,9 @@ define <4 x i16> @trunc_packus_v4i64_v4i16(<4 x i64> %a0) {
; AVX512BW-LABEL: trunc_packus_v4i64_v4i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpmovqw %zmm0, %xmm0
+; AVX512BW-NEXT: vpmovusqw %zmm0, %xmm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
@@ -1887,10 +1861,9 @@ define void @trunc_packus_v4i64_v4i16_store(<4 x i64> %a0, <4 x i16> *%p1) {
; AVX512F-LABEL: trunc_packus_v4i64_v4i16_store:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT: vpmovusqw %zmm0, %xmm0
; AVX512F-NEXT: vmovq %xmm0, (%rdi)
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
@@ -1906,10 +1879,9 @@ define void @trunc_packus_v4i64_v4i16_store(<4 x i64> %a0, <4 x i16> *%p1) {
; AVX512BW-LABEL: trunc_packus_v4i64_v4i16_store:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpmovqw %zmm0, %xmm0
+; AVX512BW-NEXT: vpmovusqw %zmm0, %xmm0
; AVX512BW-NEXT: vmovq %xmm0, (%rdi)
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
@@ -2878,47 +2850,40 @@ define <2 x i8> @trunc_packus_v2i64_v2i8(<2 x i64> %a0) {
; AVX512F-LABEL: trunc_packus_v2i64_v2i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT: vmovdqa {{.*#+}} xmm1 = [255,255]
-; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512F-NEXT: vpmovusqb %zmm0, %xmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: trunc_packus_v2i64_v2i8:
; AVX512VL: # %bb.0:
-; AVX512VL-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512VL-NEXT: vpmovusqb %xmm0, %xmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: trunc_packus_v2i64_v2i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = [255,255]
-; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512BW-NEXT: vpmovusqb %zmm0, %xmm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
; AVX512BWVL-LABEL: trunc_packus_v2i64_v2i8:
; AVX512BWVL: # %bb.0:
-; AVX512BWVL-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; AVX512BWVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BWVL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512BWVL-NEXT: vpmovusqb %xmm0, %xmm0
; AVX512BWVL-NEXT: retq
;
; SKX-LABEL: trunc_packus_v2i64_v2i8:
; SKX: # %bb.0:
-; SKX-NEXT: vpminsq {{.*}}(%rip), %xmm0, %xmm0
; SKX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; SKX-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
-; SKX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; SKX-NEXT: vpmovusqb %xmm0, %xmm0
; SKX-NEXT: retq
%1 = icmp slt <2 x i64> %a0, <i64 255, i64 255>
%2 = select <2 x i1> %1, <2 x i64> %a0, <2 x i64> <i64 255, i64 255>
@@ -3041,11 +3006,9 @@ define void @trunc_packus_v2i64_v2i8_store(<2 x i64> %a0, <2 x i8> *%p1) {
; AVX512F-LABEL: trunc_packus_v2i64_v2i8_store:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT: vmovdqa {{.*#+}} xmm1 = [255,255]
-; AVX512F-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512F-NEXT: vpmovusqb %zmm0, %xmm0
; AVX512F-NEXT: vpextrw $0, %xmm0, (%rdi)
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
@@ -3060,11 +3023,9 @@ define void @trunc_packus_v2i64_v2i8_store(<2 x i64> %a0, <2 x i8> *%p1) {
; AVX512BW-LABEL: trunc_packus_v2i64_v2i8_store:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm1 = [255,255]
-; AVX512BW-NEXT: vpminsq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512BW-NEXT: vpmovusqb %zmm0, %xmm0
; AVX512BW-NEXT: vpextrw $0, %xmm0, (%rdi)
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
@@ -3303,10 +3264,9 @@ define <4 x i8> @trunc_packus_v4i64_v4i8(<4 x i64> %a0) {
; AVX512F-LABEL: trunc_packus_v4i64_v4i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpmovqb %zmm0, %xmm0
+; AVX512F-NEXT: vpmovusqb %zmm0, %xmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
@@ -3321,10 +3281,9 @@ define <4 x i8> @trunc_packus_v4i64_v4i8(<4 x i64> %a0) {
; AVX512BW-LABEL: trunc_packus_v4i64_v4i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpmovqb %zmm0, %xmm0
+; AVX512BW-NEXT: vpmovusqb %zmm0, %xmm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
@@ -3567,10 +3526,9 @@ define void @trunc_packus_v4i64_v4i8_store(<4 x i64> %a0, <4 x i8> *%p1) {
; AVX512F-LABEL: trunc_packus_v4i64_v4i8_store:
; AVX512F: # %bb.0:
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT: vpmovqb %zmm0, %xmm0
+; AVX512F-NEXT: vpmovusqb %zmm0, %xmm0
; AVX512F-NEXT: vmovd %xmm0, (%rdi)
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
@@ -3586,10 +3544,9 @@ define void @trunc_packus_v4i64_v4i8_store(<4 x i64> %a0, <4 x i8> *%p1) {
; AVX512BW-LABEL: trunc_packus_v4i64_v4i8_store:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT: vpminsq {{.*}}(%rip){1to8}, %zmm0, %zmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT: vpmovqb %zmm0, %xmm0
+; AVX512BW-NEXT: vpmovusqb %zmm0, %xmm0
; AVX512BW-NEXT: vmovd %xmm0, (%rdi)
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
@@ -5279,44 +5236,39 @@ define <4 x i8> @trunc_packus_v4i32_v4i8(<4 x i32> %a0) "min-legal-vector-width"
;
; AVX512F-LABEL: trunc_packus_v4i32_v4i8:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vpbroadcastd {{.*#+}} xmm1 = [255,255,255,255]
-; AVX512F-NEXT: vpminsd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
-; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512F-NEXT: vpmovusdb %zmm0, %xmm0
+; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: trunc_packus_v4i32_v4i8:
; AVX512VL: # %bb.0:
-; AVX512VL-NEXT: vpminsd {{.*}}(%rip){1to4}, %xmm0, %xmm0
; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512VL-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512VL-NEXT: vpmovusdb %xmm0, %xmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: trunc_packus_v4i32_v4i8:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpbroadcastd {{.*#+}} xmm1 = [255,255,255,255]
-; AVX512BW-NEXT: vpminsd %xmm1, %xmm0, %xmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512BW-NEXT: vpmovusdb %zmm0, %xmm0
+; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
; AVX512BWVL-LABEL: trunc_packus_v4i32_v4i8:
; AVX512BWVL: # %bb.0:
-; AVX512BWVL-NEXT: vpminsd {{.*}}(%rip){1to4}, %xmm0, %xmm0
; AVX512BWVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BWVL-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512BWVL-NEXT: vpmovusdb %xmm0, %xmm0
; AVX512BWVL-NEXT: retq
;
; SKX-LABEL: trunc_packus_v4i32_v4i8:
; SKX: # %bb.0:
-; SKX-NEXT: vpminsd {{.*}}(%rip){1to4}, %xmm0, %xmm0
; SKX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; SKX-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
-; SKX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; SKX-NEXT: vpmovusdb %xmm0, %xmm0
; SKX-NEXT: retq
%1 = icmp slt <4 x i32> %a0, <i32 255, i32 255, i32 255, i32 255>
%2 = select <4 x i1> %1, <4 x i32> %a0, <4 x i32> <i32 255, i32 255, i32 255, i32 255>
@@ -5391,12 +5343,11 @@ define void @trunc_packus_v4i32_v4i8_store(<4 x i32> %a0, <4 x i8> *%p1) {
;
; AVX512F-LABEL: trunc_packus_v4i32_v4i8_store:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vpbroadcastd {{.*#+}} xmm1 = [255,255,255,255]
-; AVX512F-NEXT: vpminsd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512F-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
-; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512F-NEXT: vpmovusdb %zmm0, %xmm0
; AVX512F-NEXT: vmovd %xmm0, (%rdi)
+; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: trunc_packus_v4i32_v4i8_store:
@@ -5408,12 +5359,11 @@ define void @trunc_packus_v4i32_v4i8_store(<4 x i32> %a0, <4 x i8> *%p1) {
;
; AVX512BW-LABEL: trunc_packus_v4i32_v4i8_store:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpbroadcastd {{.*#+}} xmm1 = [255,255,255,255]
-; AVX512BW-NEXT: vpminsd %xmm1, %xmm0, %xmm0
; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512BW-NEXT: vpmovusdb %zmm0, %xmm0
; AVX512BW-NEXT: vmovd %xmm0, (%rdi)
+; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
; AVX512BWVL-LABEL: trunc_packus_v4i32_v4i8_store:
OpenPOWER on IntegriCloud