[llvm][AVX512][intrinsics] Fix vperm{b|w|d|q|ps|pd} intrinsics. Index is second argument to buildin function but it is first instruction operand.

Differential Revision: http://reviews.llvm.org/D20515

llvm-svn: 270548

6 files changed, 63 insertions, 63 deletions

diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics.ll b/llvm/test/CodeGen/X86/avx512-intrinsics.ll
index 8f2f2ea6a8b..d18fd987ca4 100644
--- a/llvm/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512-intrinsics.ll
@@ -7035,9 +7035,9 @@ define <8 x double>@test_int_x86_avx512_mask_permvar_df_512(<8 x double> %x0, <8
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_df_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1
-; CHECK-NEXT:    vpermpd %zmm1, %zmm0, %zmm2 {%k1}
-; CHECK-NEXT:    vpermpd %zmm1, %zmm0, %zmm3 {%k1} {z}
-; CHECK-NEXT:    vpermpd %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    vpermpd %zmm0, %zmm1, %zmm2 {%k1}
+; CHECK-NEXT:    vpermpd %zmm0, %zmm1, %zmm3 {%k1} {z}
+; CHECK-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    vaddpd %zmm3, %zmm2, %zmm1
 ; CHECK-NEXT:    vaddpd %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
@@ -7055,9 +7055,9 @@ define <8 x i64>@test_int_x86_avx512_mask_permvar_di_512(<8 x i64> %x0, <8 x i64
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_di_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1
-; CHECK-NEXT:    vpermq %zmm1, %zmm0, %zmm2 {%k1}
-; CHECK-NEXT:    vpermq %zmm1, %zmm0, %zmm3 {%k1} {z}
-; CHECK-NEXT:    vpermq %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    vpermq %zmm0, %zmm1, %zmm2 {%k1}
+; CHECK-NEXT:    vpermq %zmm0, %zmm1, %zmm3 {%k1} {z}
+; CHECK-NEXT:    vpermq %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    vpaddq %zmm3, %zmm2, %zmm1
 ; CHECK-NEXT:    vpaddq %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
@@ -7075,9 +7075,9 @@ define <16 x float>@test_int_x86_avx512_mask_permvar_sf_512(<16 x float> %x0, <1
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_sf_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1
-; CHECK-NEXT:    vpermps %zmm1, %zmm0, %zmm2 {%k1}
-; CHECK-NEXT:    vpermps %zmm1, %zmm0, %zmm3 {%k1} {z}
-; CHECK-NEXT:    vpermps %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    vpermps %zmm0, %zmm1, %zmm2 {%k1}
+; CHECK-NEXT:    vpermps %zmm0, %zmm1, %zmm3 {%k1} {z}
+; CHECK-NEXT:    vpermps %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    vaddps %zmm3, %zmm2, %zmm1
 ; CHECK-NEXT:    vaddps %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
@@ -7095,9 +7095,9 @@ define <16 x i32>@test_int_x86_avx512_mask_permvar_si_512(<16 x i32> %x0, <16 x
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_si_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1
-; CHECK-NEXT:    vpermd %zmm1, %zmm0, %zmm2 {%k1}
-; CHECK-NEXT:    vpermd %zmm1, %zmm0, %zmm3 {%k1} {z}
-; CHECK-NEXT:    vpermd %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    vpermd %zmm0, %zmm1, %zmm2 {%k1}
+; CHECK-NEXT:    vpermd %zmm0, %zmm1, %zmm3 {%k1} {z}
+; CHECK-NEXT:    vpermd %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    vpaddd %zmm3, %zmm2, %zmm1
 ; CHECK-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
diff --git a/llvm/test/CodeGen/X86/avx512bw-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bw-intrinsics.ll
index c5b707e9f0a..e3ca747e7c5 100644
--- a/llvm/test/CodeGen/X86/avx512bw-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512bw-intrinsics.ll
@@ -3323,9 +3323,9 @@ define <32 x i16>@test_int_x86_avx512_mask_permvar_hi_512(<32 x i16> %x0, <32 x
 ; AVX512BW-LABEL: test_int_x86_avx512_mask_permvar_hi_512:
 ; AVX512BW:       ## BB#0:
 ; AVX512BW-NEXT:    kmovd %edi, %k1
-; AVX512BW-NEXT:    vpermw %zmm1, %zmm0, %zmm2 {%k1}
-; AVX512BW-NEXT:    vpermw %zmm1, %zmm0, %zmm3 {%k1} {z}
-; AVX512BW-NEXT:    vpermw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm2 {%k1}
+; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm3 {%k1} {z}
+; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpaddw %zmm3, %zmm2, %zmm1
 ; AVX512BW-NEXT:    vpaddw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
@@ -3333,9 +3333,9 @@ define <32 x i16>@test_int_x86_avx512_mask_permvar_hi_512(<32 x i16> %x0, <32 x
 ; AVX512F-32-LABEL: test_int_x86_avx512_mask_permvar_hi_512:
 ; AVX512F-32:       # BB#0:
 ; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
-; AVX512F-32-NEXT:    vpermw %zmm1, %zmm0, %zmm2 {%k1}
-; AVX512F-32-NEXT:    vpermw %zmm1, %zmm0, %zmm3 {%k1} {z}
-; AVX512F-32-NEXT:    vpermw %zmm1, %zmm0, %zmm0
+; AVX512F-32-NEXT:    vpermw %zmm0, %zmm1, %zmm2 {%k1}
+; AVX512F-32-NEXT:    vpermw %zmm0, %zmm1, %zmm3 {%k1} {z}
+; AVX512F-32-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; AVX512F-32-NEXT:    vpaddw %zmm3, %zmm2, %zmm1
 ; AVX512F-32-NEXT:    vpaddw %zmm0, %zmm1, %zmm0
 ; AVX512F-32-NEXT:    retl
diff --git a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
index 087381edffb..29e46d21691 100644
--- a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
@@ -6430,9 +6430,9 @@ define <8 x i16>@test_int_x86_avx512_mask_permvar_hi_128(<8 x i16> %x0, <8 x i16
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_hi_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT:    vpermw %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0x8d,0xd1]
-; CHECK-NEXT:    vpermw %xmm1, %xmm0, %xmm3 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0x89,0x8d,0xd9]
-; CHECK-NEXT:    vpermw %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf2,0xfd,0x08,0x8d,0xc1]
+; CHECK-NEXT:    vpermw %xmm0, %xmm1, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0x8d,0xd0]
+; CHECK-NEXT:    vpermw %xmm0, %xmm1, %xmm3 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0x8d,0xd8]
+; CHECK-NEXT:    vpermw %xmm0, %xmm1, %xmm0 ## encoding: [0x62,0xf2,0xf5,0x08,0x8d,0xc0]
 ; CHECK-NEXT:    vpaddw %xmm3, %xmm2, %xmm1 ## encoding: [0x62,0xf1,0x6d,0x08,0xfd,0xcb]
 ; CHECK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x75,0x08,0xfd,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
@@ -6450,9 +6450,9 @@ define <16 x i16>@test_int_x86_avx512_mask_permvar_hi_256(<16 x i16> %x0, <16 x
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_hi_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT:    vpermw %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x8d,0xd1]
-; CHECK-NEXT:    vpermw %ymm1, %ymm0, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0xa9,0x8d,0xd9]
-; CHECK-NEXT:    vpermw %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0xfd,0x28,0x8d,0xc1]
+; CHECK-NEXT:    vpermw %ymm0, %ymm1, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x8d,0xd0]
+; CHECK-NEXT:    vpermw %ymm0, %ymm1, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0x8d,0xd8]
+; CHECK-NEXT:    vpermw %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf2,0xf5,0x28,0x8d,0xc0]
 ; CHECK-NEXT:    vpaddw %ymm3, %ymm2, %ymm1 ## encoding: [0x62,0xf1,0x6d,0x28,0xfd,0xcb]
 ; CHECK-NEXT:    vpaddw %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x75,0x28,0xfd,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
diff --git a/llvm/test/CodeGen/X86/avx512vbmi-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vbmi-intrinsics.ll
index 49fcaf4632d..ce999855d1f 100644
--- a/llvm/test/CodeGen/X86/avx512vbmi-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512vbmi-intrinsics.ll
@@ -1,24 +1,24 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py

-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx  -mattr=+avx512vbmi --show-mc-encoding| FileCheck %s

-declare <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)

-

-define <64 x i8>@test_int_x86_avx512_mask_permvar_qi_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {

-; CHECK-LABEL: test_int_x86_avx512_mask_permvar_qi_512:

-; CHECK:       ## BB#0:

-; CHECK-NEXT:    kmovq %rdi, %k1 

-; CHECK-NEXT:    vpermb %zmm1, %zmm0, %zmm2 {%k1} 

-; CHECK-NEXT:    vpermb %zmm1, %zmm0, %zmm3 {%k1} {z} 

-; CHECK-NEXT:    vpermb %zmm1, %zmm0, %zmm0 

-; CHECK-NEXT:    vpaddb %zmm3, %zmm2, %zmm1 

-; CHECK-NEXT:    vpaddb %zmm0, %zmm1, %zmm0 

-; CHECK-NEXT:    retq 

- %res = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)

- %res1 = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> zeroinitializer, i64 %x3)

- %res2 = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)

- %res3 = add <64 x i8> %res, %res1

- %res4 = add <64 x i8> %res3, %res2

- ret <64 x i8> %res4

-}

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx  -mattr=+avx512vbmi | FileCheck %s
+declare <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8>@test_int_x86_avx512_mask_permvar_qi_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
+; CHECK-LABEL: test_int_x86_avx512_mask_permvar_qi_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovq %rdi, %k1
+; CHECK-NEXT:    vpermb %zmm0, %zmm1, %zmm2 {%k1}
+; CHECK-NEXT:    vpermb %zmm0, %zmm1, %zmm3 {%k1} {z}
+; CHECK-NEXT:    vpermb %zmm0, %zmm1, %zmm0
+; CHECK-NEXT:    vpaddb %zmm3, %zmm2, %zmm1
+; CHECK-NEXT:    vpaddb %zmm0, %zmm1, %zmm0
+; CHECK-NEXT:    retq
+ %res = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+ %res1 = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> zeroinitializer, i64 %x3)
+ %res2 = call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+ %res3 = add <64 x i8> %res, %res1
+ %res4 = add <64 x i8> %res3, %res2
+ ret <64 x i8> %res4
+}
 
 declare <64 x i8> @llvm.x86.avx512.mask.pmultishift.qb.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
 
diff --git a/llvm/test/CodeGen/X86/avx512vbmivl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vbmivl-intrinsics.ll
index 6f7abca0a72..a09523b2992 100644
--- a/llvm/test/CodeGen/X86/avx512vbmivl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512vbmivl-intrinsics.ll
@@ -6,9 +6,9 @@ define <16 x i8>@test_int_x86_avx512_mask_permvar_qi_128(<16 x i8> %x0, <16 x i8
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_qi_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT:    vpermb %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0x8d,0xd1]
-; CHECK-NEXT:    vpermb %xmm1, %xmm0, %xmm3 {%k1} {z} ## encoding: [0x62,0xf2,0x7d,0x89,0x8d,0xd9]
-; CHECK-NEXT:    vpermb %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf2,0x7d,0x08,0x8d,0xc1]
+; CHECK-NEXT:    vpermb %xmm0, %xmm1, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0x8d,0xd0]
+; CHECK-NEXT:    vpermb %xmm0, %xmm1, %xmm3 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0x8d,0xd8]
+; CHECK-NEXT:    vpermb %xmm0, %xmm1, %xmm0 ## encoding: [0x62,0xf2,0x75,0x08,0x8d,0xc0]
 ; CHECK-NEXT:    vpaddb %xmm3, %xmm2, %xmm1 ## encoding: [0x62,0xf1,0x6d,0x08,0xfc,0xcb]
 ; CHECK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x75,0x08,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
@@ -26,9 +26,9 @@ define <32 x i8>@test_int_x86_avx512_mask_permvar_qi_256(<32 x i8> %x0, <32 x i8
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_qi_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT:    vpermb %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x8d,0xd1]
-; CHECK-NEXT:    vpermb %ymm1, %ymm0, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0x7d,0xa9,0x8d,0xd9]
-; CHECK-NEXT:    vpermb %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0x7d,0x28,0x8d,0xc1]
+; CHECK-NEXT:    vpermb %ymm0, %ymm1, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x8d,0xd0]
+; CHECK-NEXT:    vpermb %ymm0, %ymm1, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0x8d,0xd8]
+; CHECK-NEXT:    vpermb %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf2,0x75,0x28,0x8d,0xc0]
 ; CHECK-NEXT:    vpaddb %ymm3, %ymm2, %ymm1 ## encoding: [0x62,0xf1,0x6d,0x28,0xfc,0xcb]
 ; CHECK-NEXT:    vpaddb %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x75,0x28,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
index c97262f8daf..ed428bb754e 100644
--- a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
@@ -9209,9 +9209,9 @@ define <4 x double>@test_int_x86_avx512_mask_permvar_df_256(<4 x double> %x0, <4
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_df_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT:    vpermpd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x16,0xd1]
-; CHECK-NEXT:    vpermpd %ymm1, %ymm0, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0xa9,0x16,0xd9]
-; CHECK-NEXT:    vpermpd %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0xfd,0x28,0x16,0xc1]
+; CHECK-NEXT:    vpermpd %ymm0, %ymm1, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x16,0xd0]
+; CHECK-NEXT:    vpermpd %ymm0, %ymm1, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0x16,0xd8]
+; CHECK-NEXT:    vpermpd %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf2,0xf5,0x28,0x16,0xc0]
 ; CHECK-NEXT:    vaddpd %ymm3, %ymm2, %ymm1 ## encoding: [0x62,0xf1,0xed,0x28,0x58,0xcb]
 ; CHECK-NEXT:    vaddpd %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xf5,0x28,0x58,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
@@ -9229,9 +9229,9 @@ define <4 x i64>@test_int_x86_avx512_mask_permvar_di_256(<4 x i64> %x0, <4 x i64
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_di_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT:    vpermq %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x36,0xd1]
-; CHECK-NEXT:    vpermq %ymm1, %ymm0, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0xa9,0x36,0xd9]
-; CHECK-NEXT:    vpermq %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0xfd,0x28,0x36,0xc1]
+; CHECK-NEXT:    vpermq %ymm0, %ymm1, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x36,0xd0]
+; CHECK-NEXT:    vpermq %ymm0, %ymm1, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0x36,0xd8]
+; CHECK-NEXT:    vpermq %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf2,0xf5,0x28,0x36,0xc0]
 ; CHECK-NEXT:    vpaddq %ymm3, %ymm2, %ymm1 ## encoding: [0x62,0xf1,0xed,0x28,0xd4,0xcb]
 ; CHECK-NEXT:    vpaddq %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xf5,0x28,0xd4,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
@@ -9249,9 +9249,9 @@ define <8 x float>@test_int_x86_avx512_mask_permvar_sf_256(<8 x float> %x0, <8 x
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_sf_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT:    vpermps %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x16,0xd1]
-; CHECK-NEXT:    vpermps %ymm1, %ymm0, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0x7d,0xa9,0x16,0xd9]
-; CHECK-NEXT:    vpermps %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0x7d,0x28,0x16,0xc1]
+; CHECK-NEXT:    vpermps %ymm0, %ymm1, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x16,0xd0]
+; CHECK-NEXT:    vpermps %ymm0, %ymm1, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0x16,0xd8]
+; CHECK-NEXT:    vpermps %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf2,0x75,0x28,0x16,0xc0]
 ; CHECK-NEXT:    vaddps %ymm3, %ymm2, %ymm1 ## encoding: [0x62,0xf1,0x6c,0x28,0x58,0xcb]
 ; CHECK-NEXT:    vaddps %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x74,0x28,0x58,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
@@ -9269,9 +9269,9 @@ define <8 x i32>@test_int_x86_avx512_mask_permvar_si_256(<8 x i32> %x0, <8 x i32
 ; CHECK-LABEL: test_int_x86_avx512_mask_permvar_si_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT:    vpermd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x36,0xd1]
-; CHECK-NEXT:    vpermd %ymm1, %ymm0, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0x7d,0xa9,0x36,0xd9]
-; CHECK-NEXT:    vpermd %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0x7d,0x28,0x36,0xc1]
+; CHECK-NEXT:    vpermd %ymm0, %ymm1, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x36,0xd0]
+; CHECK-NEXT:    vpermd %ymm0, %ymm1, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0x36,0xd8]
+; CHECK-NEXT:    vpermd %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf2,0x75,0x28,0x36,0xc0]
 ; CHECK-NEXT:    vpaddd %ymm3, %ymm2, %ymm1 ## encoding: [0x62,0xf1,0x6d,0x28,0xfe,0xcb]
 ; CHECK-NEXT:    vpaddd %ymm0, %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x75,0x28,0xfe,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]