[X86][SSE] Add missing AVX512 rotation tests

Increase coverage to make sure we're not doing anything stupid without AVX512BW llvm-svn: 335746
author: Simon Pilgrim <llvm-dev@redking.me.uk> 2018-06-27 16:00:53 +0000
committer: Simon Pilgrim <llvm-dev@redking.me.uk> 2018-06-27 16:00:53 +0000
commit: 8a02b25313cb631f6523c3f57a1e5a916fdb3e00 (patch)
tree: 9edad83816c8f19e084b2492ef8c2fef9da3f98f /llvm/test
parent: 851f36369114a77618d03246cfd4f225ea1281c3 (diff)
download: bcm5719-llvm-8a02b25313cb631f6523c3f57a1e5a916fdb3e00.tar.gz
bcm5719-llvm-8a02b25313cb631f6523c3f57a1e5a916fdb3e00.zip
3 files changed, 943 insertions, 230 deletions
diff --git a/llvm/test/CodeGen/X86/vector-rotate-128.ll b/llvm/test/CodeGen/X86/vector-rotate-128.ll
index dab12689dcf..d7544aa344e 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-128.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-128.ll
@@ -1,15 +1,17 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE41
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefix=ALL --check-prefix=XOP --check-prefix=XOPAVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=XOP --check-prefix=XOPAVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VLBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefixes=XOP,XOPAVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=XOP,XOPAVX2
 
 ; Just one 32-bit run to make sure we do reasonable things for i64 rotates.
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=ALL --check-prefix=X32-SSE --check-prefix=X32-SSE2
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=X32-SSE,X32-SSE2
 
 ;
 ; Variable Rotates
@@ -76,6 +78,20 @@ define <2 x i64> @var_rotate_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
 ; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: var_rotate_v2i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vprolvq %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_rotate_v2i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolvq %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: var_rotate_v2i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
@@ -85,10 +101,10 @@ define <2 x i64> @var_rotate_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: var_rotate_v2i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolvq %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: var_rotate_v2i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolvq %xmm1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: var_rotate_v2i64:
 ; XOP:       # %bb.0:
@@ -179,6 +195,20 @@ define <4 x i32> @var_rotate_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: var_rotate_v4i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_rotate_v4i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: var_rotate_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
@@ -188,10 +218,10 @@ define <4 x i32> @var_rotate_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: var_rotate_v4i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: var_rotate_v4i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: var_rotate_v4i32:
 ; XOP:       # %bb.0:
@@ -304,6 +334,36 @@ define <8 x i16> @var_rotate_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: var_rotate_v8i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX512F-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
+; AVX512F-NEXT:    vpsllvd %ymm2, %ymm0, %ymm2
+; AVX512F-NEXT:    vpmovdw %zmm2, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX512F-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} 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
+; AVX512F-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_rotate_v8i16:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX512VL-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
+; AVX512VL-NEXT:    vpsllvd %ymm2, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpmovdw %ymm2, %xmm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX512VL-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} 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
+; AVX512VL-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: var_rotate_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
@@ -316,14 +376,14 @@ define <8 x i16> @var_rotate_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: var_rotate_v8i16:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllvw %xmm1, %xmm0, %xmm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
-; AVX512VL-NEXT:    vpsrlvw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: var_rotate_v8i16:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpsllvw %xmm1, %xmm0, %xmm2
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX512VLBW-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX512VLBW-NEXT:    vpsrlvw %xmm1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: var_rotate_v8i16:
 ; XOP:       # %bb.0:
@@ -500,6 +560,36 @@ define <16 x i8> @var_rotate_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
+; AVX512F-LABEL: var_rotate_v16i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512F-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512F-NEXT:    vpsllvd %zmm1, %zmm0, %zmm1
+; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512F-NEXT:    vpsrlvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_rotate_v16i8:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VL-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsllvd %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: var_rotate_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
@@ -515,20 +605,20 @@ define <16 x i8> @var_rotate_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: var_rotate_v16i8:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
-; AVX512VL-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
-; AVX512VL-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
-; AVX512VL-NEXT:    vpsllvw %ymm1, %ymm0, %ymm1
-; AVX512VL-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VL-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VL-NEXT:    vpsrlvw %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: var_rotate_v16i8:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
+; AVX512VLBW-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
+; AVX512VLBW-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
+; AVX512VLBW-NEXT:    vpsllvw %ymm1, %ymm0, %ymm1
+; AVX512VLBW-NEXT:    vpmovwb %ymm1, %xmm1
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLBW-NEXT:    vpsrlvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vzeroupper
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: var_rotate_v16i8:
 ; XOP:       # %bb.0:
@@ -637,6 +727,21 @@ define <2 x i64> @splatvar_rotate_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
 ; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatvar_rotate_v2i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512F-NEXT:    vprolvq %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatvar_rotate_v2i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512VL-NEXT:    vprolvq %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatvar_rotate_v2i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -646,11 +751,11 @@ define <2 x i64> @splatvar_rotate_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatvar_rotate_v2i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm1, %xmm1
-; AVX512VL-NEXT:    vprolvq %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatvar_rotate_v2i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vprolvq %xmm1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v2i64:
 ; XOPAVX1:       # %bb.0:
@@ -738,6 +843,21 @@ define <4 x i32> @splatvar_rotate_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatvar_rotate_v4i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX512F-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatvar_rotate_v4i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX512VL-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatvar_rotate_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -747,11 +867,11 @@ define <4 x i32> @splatvar_rotate_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatvar_rotate_v4i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm1, %xmm1
-; AVX512VL-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatvar_rotate_v4i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v4i32:
 ; XOPAVX1:       # %bb.0:
@@ -843,6 +963,30 @@ define <8 x i16> @splatvar_rotate_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatvar_rotate_v8i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512F-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX512F-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX512F-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatvar_rotate_v8i16:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX512VL-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX512VL-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatvar_rotate_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -856,16 +1000,16 @@ define <8 x i16> @splatvar_rotate_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatvar_rotate_v8i16:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
-; AVX512VL-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
-; AVX512VL-NEXT:    vpsrlvw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm0, %xmm2, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatvar_rotate_v8i16:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX512VLBW-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX512VLBW-NEXT:    vpsrlvw %xmm1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v8i16:
 ; XOPAVX1:       # %bb.0:
@@ -1079,6 +1223,38 @@ define <16 x i8> @splatvar_rotate_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatvar_rotate_v16i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %xmm1
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512F-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512F-NEXT:    vpsllvd %zmm1, %zmm0, %zmm1
+; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512F-NEXT:    vpsrlvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatvar_rotate_v16i8:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VL-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsllvd %zmm1, %zmm0, %zmm1
+; AVX512VL-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatvar_rotate_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
@@ -1095,21 +1271,21 @@ define <16 x i8> @splatvar_rotate_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatvar_rotate_v16i8:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
-; AVX512VL-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
-; AVX512VL-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
-; AVX512VL-NEXT:    vpsllvw %ymm1, %ymm0, %ymm1
-; AVX512VL-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VL-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VL-NEXT:    vpsrlvw %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatvar_rotate_v16i8:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
+; AVX512VLBW-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
+; AVX512VLBW-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
+; AVX512VLBW-NEXT:    vpsllvw %ymm1, %ymm0, %ymm1
+; AVX512VLBW-NEXT:    vpmovwb %ymm1, %xmm1
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLBW-NEXT:    vpsrlvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vzeroupper
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v16i8:
 ; XOPAVX1:       # %bb.0:
@@ -1245,6 +1421,20 @@ define <2 x i64> @constant_rotate_v2i64(<2 x i64> %a) nounwind {
 ; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: constant_rotate_v2i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,14]
+; AVX512F-NEXT:    vprolvq %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: constant_rotate_v2i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolvq {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: constant_rotate_v2i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -1254,10 +1444,10 @@ define <2 x i64> @constant_rotate_v2i64(<2 x i64> %a) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: constant_rotate_v2i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolvq {{.*}}(%rip), %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: constant_rotate_v2i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolvq {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: constant_rotate_v2i64:
 ; XOP:       # %bb.0:
@@ -1335,6 +1525,20 @@ define <4 x i32> @constant_rotate_v4i32(<4 x i32> %a) nounwind {
 ; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: constant_rotate_v4i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,5,6,7]
+; AVX512F-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: constant_rotate_v4i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolvd {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: constant_rotate_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -1344,10 +1548,10 @@ define <4 x i32> @constant_rotate_v4i32(<4 x i32> %a) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: constant_rotate_v4i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolvd {{.*}}(%rip), %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: constant_rotate_v4i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolvd {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: constant_rotate_v4i32:
 ; XOP:       # %bb.0:
@@ -1393,6 +1597,22 @@ define <8 x i16> @constant_rotate_v8i16(<8 x i16> %a) nounwind {
 ; AVX-NEXT:    vpor %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
+; AVX512F-LABEL: constant_rotate_v8i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,2,4,8,16,32,64,128]
+; AVX512F-NEXT:    vpmulhuw %xmm1, %xmm0, %xmm2
+; AVX512F-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: constant_rotate_v8i16:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    vpmulhuw %xmm1, %xmm0, %xmm2
+; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: constant_rotate_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -1404,12 +1624,12 @@ define <8 x i16> @constant_rotate_v8i16(<8 x i16> %a) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: constant_rotate_v8i16:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsrlvw {{.*}}(%rip), %xmm0, %xmm1
-; AVX512VL-NEXT:    vpsllvw {{.*}}(%rip), %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: constant_rotate_v8i16:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %xmm0, %xmm1
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: constant_rotate_v8i16:
 ; XOP:       # %bb.0:
@@ -1554,6 +1774,28 @@ define <16 x i8> @constant_rotate_v16i8(<16 x i8> %a) nounwind {
 ; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
+; AVX512F-LABEL: constant_rotate_v16i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512F-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm1
+; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512F-NEXT:    vpsrlvd {{.*}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: constant_rotate_v16i8:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm1
+; AVX512VL-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: constant_rotate_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
@@ -1567,16 +1809,16 @@ define <16 x i8> @constant_rotate_v16i8(<16 x i8> %a) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: constant_rotate_v16i8:
-; AVX512VL:       # %bb.0:
-; AVX512VL-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
-; AVX512VL-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm1
-; AVX512VL-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VL-NEXT:    vpsrlvw {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: constant_rotate_v16i8:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-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
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm1
+; AVX512VLBW-NEXT:    vpmovwb %ymm1, %xmm1
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vzeroupper
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: constant_rotate_v16i8:
 ; XOP:       # %bb.0:
@@ -1666,6 +1908,19 @@ define <2 x i64> @splatconstant_rotate_v2i64(<2 x i64> %a) nounwind {
 ; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
+; AVX512F-LABEL: splatconstant_rotate_v2i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vprolq $14, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatconstant_rotate_v2i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolq $14, %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatconstant_rotate_v2i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -1674,10 +1929,10 @@ define <2 x i64> @splatconstant_rotate_v2i64(<2 x i64> %a) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatconstant_rotate_v2i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolq $14, %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatconstant_rotate_v2i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolq $14, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: splatconstant_rotate_v2i64:
 ; XOP:       # %bb.0:
@@ -1713,6 +1968,19 @@ define <4 x i32> @splatconstant_rotate_v4i32(<4 x i32> %a) nounwind {
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
+; AVX512F-LABEL: splatconstant_rotate_v4i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vprold $4, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatconstant_rotate_v4i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprold $4, %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatconstant_rotate_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -1721,10 +1989,10 @@ define <4 x i32> @splatconstant_rotate_v4i32(<4 x i32> %a) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatconstant_rotate_v4i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprold $4, %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatconstant_rotate_v4i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprold $4, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: splatconstant_rotate_v4i32:
 ; XOP:       # %bb.0:
@@ -1851,6 +2119,20 @@ define <2 x i64> @splatconstant_rotate_mask_v2i64(<2 x i64> %a) nounwind {
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
+; AVX512F-LABEL: splatconstant_rotate_mask_v2i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vprolq $15, %zmm0, %zmm0
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatconstant_rotate_mask_v2i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolq $15, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v2i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -1859,11 +2141,11 @@ define <2 x i64> @splatconstant_rotate_mask_v2i64(<2 x i64> %a) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatconstant_rotate_mask_v2i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolq $15, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatconstant_rotate_mask_v2i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolq $15, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: splatconstant_rotate_mask_v2i64:
 ; XOP:       # %bb.0:
@@ -1902,6 +2184,20 @@ define <4 x i32> @splatconstant_rotate_mask_v4i32(<4 x i32> %a) nounwind {
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
+; AVX512F-LABEL: splatconstant_rotate_mask_v4i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vprold $4, %zmm0, %zmm0
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatconstant_rotate_mask_v4i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprold $4, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
@@ -1910,11 +2206,11 @@ define <4 x i32> @splatconstant_rotate_mask_v4i32(<4 x i32> %a) nounwind {
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatconstant_rotate_mask_v4i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprold $4, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatconstant_rotate_mask_v4i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprold $4, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: splatconstant_rotate_mask_v4i32:
 ; XOP:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-rotate-256.ll b/llvm/test/CodeGen/X86/vector-rotate-256.ll
index 0e45f0625f1..c8174dea46c 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-256.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-256.ll
@@ -1,10 +1,12 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefix=ALL --check-prefix=XOP --check-prefix=XOPAVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=XOP --check-prefix=XOPAVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VLBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefixes=XOP,XOPAVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=XOP,XOPAVX2
 
 ;
 ; Variable Rotates
@@ -48,6 +50,19 @@ define <4 x i64> @var_rotate_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: var_rotate_v4i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vprolvq %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_rotate_v4i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolvq %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: var_rotate_v4i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
@@ -56,10 +71,10 @@ define <4 x i64> @var_rotate_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: var_rotate_v4i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolvq %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: var_rotate_v4i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolvq %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_rotate_v4i64:
 ; XOPAVX1:       # %bb.0:
@@ -127,6 +142,19 @@ define <8 x i32> @var_rotate_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: var_rotate_v8i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_rotate_v8i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolvd %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: var_rotate_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
@@ -135,10 +163,10 @@ define <8 x i32> @var_rotate_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: var_rotate_v8i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolvd %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: var_rotate_v8i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolvd %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_rotate_v8i32:
 ; XOPAVX1:       # %bb.0:
@@ -222,6 +250,34 @@ define <16 x i16> @var_rotate_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm3, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: var_rotate_v16i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm2
+; AVX512F-NEXT:    vpmovdw %zmm2, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512F-NEXT:    vpsubw %ymm1, %ymm3, %ymm1
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512F-NEXT:    vpsrlvd %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_rotate_v16i16:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm2
+; AVX512VL-NEXT:    vpmovdw %zmm2, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VL-NEXT:    vpsubw %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512VL-NEXT:    vpsrlvd %zmm1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: var_rotate_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
@@ -233,14 +289,14 @@ define <16 x i16> @var_rotate_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
 ; AVX512BW-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: var_rotate_v16i16:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllvw %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vpsrlvw %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm0, %ymm2, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: var_rotate_v16i16:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpsllvw %ymm1, %ymm0, %ymm2
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-NEXT:    vpsubw %ymm1, %ymm3, %ymm1
+; AVX512VLBW-NEXT:    vpsrlvw %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_rotate_v16i16:
 ; XOPAVX1:       # %bb.0:
@@ -360,19 +416,93 @@ define <32 x i8> @var_rotate_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: var_rotate_v32i8:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
-; AVX512-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512-NEXT:    vpsllvw %zmm1, %zmm0, %zmm1
-; AVX512-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
-; AVX512-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: var_rotate_v32i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512F-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllw $5, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm3
+; AVX512F-NEXT:    vpsllw $2, %ymm3, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm1
+; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $2, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $1, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_rotate_v32i8:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VL-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllw $5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $2, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $1, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    retq
+;
+; AVX512BW-LABEL: var_rotate_v32i8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512BW-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512BW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512VLBW-LABEL: var_rotate_v32i8:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VLBW-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLBW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpmovwb %zmm1, %ymm1
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLBW-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_rotate_v32i8:
 ; XOPAVX1:       # %bb.0:
@@ -428,6 +558,20 @@ define <4 x i64> @splatvar_rotate_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatvar_rotate_v4i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vpbroadcastq %xmm1, %ymm1
+; AVX512F-NEXT:    vprolvq %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatvar_rotate_v4i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastq %xmm1, %ymm1
+; AVX512VL-NEXT:    vprolvq %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatvar_rotate_v4i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -436,11 +580,11 @@ define <4 x i64> @splatvar_rotate_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatvar_rotate_v4i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm1, %ymm1
-; AVX512VL-NEXT:    vprolvq %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatvar_rotate_v4i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm1, %ymm1
+; AVX512VLBW-NEXT:    vprolvq %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v4i64:
 ; XOPAVX1:       # %bb.0:
@@ -497,6 +641,20 @@ define <8 x i32> @splatvar_rotate_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatvar_rotate_v8i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vpbroadcastd %xmm1, %ymm1
+; AVX512F-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatvar_rotate_v8i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastd %xmm1, %ymm1
+; AVX512VL-NEXT:    vprolvd %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatvar_rotate_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -505,11 +663,11 @@ define <8 x i32> @splatvar_rotate_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatvar_rotate_v8i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm1, %ymm1
-; AVX512VL-NEXT:    vprolvd %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatvar_rotate_v8i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm1, %ymm1
+; AVX512VLBW-NEXT:    vprolvd %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v8i32:
 ; XOPAVX1:       # %bb.0:
@@ -568,6 +726,30 @@ define <16 x i16> @splatvar_rotate_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatvar_rotate_v16i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpbroadcastw %xmm1, %ymm2
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm1, %ymm0, %ymm1
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512F-NEXT:    vpsubw %ymm2, %ymm3, %ymm2
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatvar_rotate_v16i16:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastw %xmm1, %ymm2
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm1, %ymm0, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VL-NEXT:    vpsubw %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatvar_rotate_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -580,16 +762,16 @@ define <16 x i16> @splatvar_rotate_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ; AVX512BW-NEXT:    vpor %ymm2, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatvar_rotate_v16i16:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm1, %ymm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm1, %ymm0, %ymm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpsrlvw %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatvar_rotate_v16i16:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %ymm2
+; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpsllw %xmm1, %ymm0, %ymm1
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-NEXT:    vpsubw %ymm2, %ymm3, %ymm2
+; AVX512VLBW-NEXT:    vpsrlvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v16i16:
 ; XOPAVX1:       # %bb.0:
@@ -707,20 +889,97 @@ define <32 x i8> @splatvar_rotate_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: splatvar_rotate_v32i8:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastb %xmm1, %ymm1
-; AVX512-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
-; AVX512-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512-NEXT:    vpsllvw %zmm1, %zmm0, %zmm1
-; AVX512-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
-; AVX512-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: splatvar_rotate_v32i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512F-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw $5, %ymm1, %ymm1
+; AVX512F-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm3
+; AVX512F-NEXT:    vpsllw $2, %ymm3, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm1
+; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $2, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $1, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatvar_rotate_v32i8:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VL-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw $5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $2, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $1, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    retq
+;
+; AVX512BW-LABEL: splatvar_rotate_v32i8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512BW-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512BW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512VLBW-LABEL: splatvar_rotate_v32i8:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VLBW-NEXT:    vpsubb %ymm1, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VLBW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpmovwb %zmm1, %ymm1
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLBW-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v32i8:
 ; XOPAVX1:       # %bb.0:
@@ -781,6 +1040,19 @@ define <4 x i64> @constant_rotate_v4i64(<4 x i64> %a) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: constant_rotate_v4i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,14,50,60]
+; AVX512F-NEXT:    vprolvq %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: constant_rotate_v4i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolvq {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: constant_rotate_v4i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -789,10 +1061,10 @@ define <4 x i64> @constant_rotate_v4i64(<4 x i64> %a) nounwind {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: constant_rotate_v4i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolvq {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: constant_rotate_v4i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolvq {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_rotate_v4i64:
 ; XOPAVX1:       # %bb.0:
@@ -849,6 +1121,19 @@ define <8 x i32> @constant_rotate_v8i32(<8 x i32> %a) nounwind {
 ; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: constant_rotate_v8i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,5,6,7,8,9,10,11]
+; AVX512F-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: constant_rotate_v8i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolvd {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: constant_rotate_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -857,10 +1142,10 @@ define <8 x i32> @constant_rotate_v8i32(<8 x i32> %a) nounwind {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: constant_rotate_v8i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolvd {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: constant_rotate_v8i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolvd {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_rotate_v8i32:
 ; XOPAVX1:       # %bb.0:
@@ -906,6 +1191,22 @@ define <16 x i16> @constant_rotate_v16i16(<16 x i16> %a) nounwind {
 ; AVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: constant_rotate_v16i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768]
+; AVX512F-NEXT:    vpmulhuw %ymm1, %ymm0, %ymm2
+; AVX512F-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: constant_rotate_v16i16:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768]
+; AVX512VL-NEXT:    vpmulhuw %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: constant_rotate_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -916,12 +1217,12 @@ define <16 x i16> @constant_rotate_v16i16(<16 x i16> %a) nounwind {
 ; AVX512BW-NEXT:    vpor %ymm2, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: constant_rotate_v16i16:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsrlvw {{.*}}(%rip), %ymm0, %ymm1
-; AVX512VL-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: constant_rotate_v16i16:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm0, %ymm1
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_rotate_v16i16:
 ; XOPAVX1:       # %bb.0:
@@ -1026,15 +1327,81 @@ define <32 x i8> @constant_rotate_v32i8(<32 x i8> %a) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: constant_rotate_v32i8:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm1
-; AVX512-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
-; AVX512-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: constant_rotate_v32i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm1
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm1
+; AVX512F-NEXT:    vpsllw $2, %ymm1, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536]
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $2, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $1, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: constant_rotate_v32i8:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm1
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm1
+; AVX512VL-NEXT:    vpsllw $2, %ymm1, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536]
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $2, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $1, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    retq
+;
+; AVX512BW-LABEL: constant_rotate_v32i8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm1
+; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
+; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512VLBW-LABEL: constant_rotate_v32i8:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpmovwb %zmm1, %ymm1
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_rotate_v32i8:
 ; XOPAVX1:       # %bb.0:
@@ -1083,6 +1450,18 @@ define <4 x i64> @splatconstant_rotate_v4i64(<4 x i64> %a) nounwind {
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatconstant_rotate_v4i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vprolq $14, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatconstant_rotate_v4i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolq $14, %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatconstant_rotate_v4i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -1090,10 +1469,10 @@ define <4 x i64> @splatconstant_rotate_v4i64(<4 x i64> %a) nounwind {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatconstant_rotate_v4i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolq $14, %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatconstant_rotate_v4i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolq $14, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatconstant_rotate_v4i64:
 ; XOPAVX1:       # %bb.0:
@@ -1136,6 +1515,18 @@ define <8 x i32> @splatconstant_rotate_v8i32(<8 x i32> %a) nounwind {
 ; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatconstant_rotate_v8i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vprold $4, %zmm0, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatconstant_rotate_v8i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprold $4, %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatconstant_rotate_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -1143,10 +1534,10 @@ define <8 x i32> @splatconstant_rotate_v8i32(<8 x i32> %a) nounwind {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatconstant_rotate_v8i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprold $4, %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatconstant_rotate_v8i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprold $4, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatconstant_rotate_v8i32:
 ; XOPAVX1:       # %bb.0:
@@ -1295,6 +1686,19 @@ define <4 x i64> @splatconstant_rotate_mask_v4i64(<4 x i64> %a) nounwind {
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatconstant_rotate_mask_v4i64:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vprolq $15, %zmm0, %zmm0
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatconstant_rotate_mask_v4i64:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprolq $15, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v4i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -1302,11 +1706,11 @@ define <4 x i64> @splatconstant_rotate_mask_v4i64(<4 x i64> %a) nounwind {
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatconstant_rotate_mask_v4i64:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprolq $15, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatconstant_rotate_mask_v4i64:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprolq $15, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatconstant_rotate_mask_v4i64:
 ; XOPAVX1:       # %bb.0:
@@ -1355,6 +1759,19 @@ define <8 x i32> @splatconstant_rotate_mask_v8i32(<8 x i32> %a) nounwind {
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
+; AVX512F-LABEL: splatconstant_rotate_mask_v8i32:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vprold $4, %zmm0, %zmm0
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: splatconstant_rotate_mask_v8i32:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vprold $4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
@@ -1362,11 +1779,11 @@ define <8 x i32> @splatconstant_rotate_mask_v8i32(<8 x i32> %a) nounwind {
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
-; AVX512VL-LABEL: splatconstant_rotate_mask_v8i32:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vprold $4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VLBW-LABEL: splatconstant_rotate_mask_v8i32:
+; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vprold $4, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatconstant_rotate_mask_v8i32:
 ; XOPAVX1:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-rotate-512.ll b/llvm/test/CodeGen/X86/vector-rotate-512.ll
index 47774203641..ba4abf276ee 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-512.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-512.ll
@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512VLBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VLBW
 
 ;
 ; Variable Rotates
author	Simon Pilgrim <llvm-dev@redking.me.uk>	2018-06-27 16:00:53 +0000
committer	Simon Pilgrim <llvm-dev@redking.me.uk>	2018-06-27 16:00:53 +0000
commit	8a02b25313cb631f6523c3f57a1e5a916fdb3e00 (patch)
tree	9edad83816c8f19e084b2492ef8c2fef9da3f98f /llvm/test
parent	851f36369114a77618d03246cfd4f225ea1281c3 (diff)
download	bcm5719-llvm-8a02b25313cb631f6523c3f57a1e5a916fdb3e00.tar.gz bcm5719-llvm-8a02b25313cb631f6523c3f57a1e5a916fdb3e00.zip