[AVX-512] Make VBMI instruction set enabling imply that the BWI instruction set is also enabled.

[craig.topper]

Summary

This is needed to make the v64i8 and v32i16 types legal for the 512-bit VBMI instructions. Fixes PR30912.

This is the LLVM side part of this. Clang part is here https://reviews.llvm.org/D26306

I removed the -mcpu=skx from the vbmi intrinsic test cases so that we could be sure vbmi was enabling bwi to make the types legal. Somehow that changed the order of the addb instructions. Maybe something to do with the scheduler model?

[zvi]

LGTM

llvm/trunk/lib/Target/X86/X86.td

def FeatureDQI     : SubtargetFeature<"avx512dq", "HasDQI", "true",
                                       [FeatureAVX512]>;
 def FeatureBWI     : SubtargetFeature<"avx512bw", "HasBWI", "true",
                       "Enable AVX-512 Byte and Word Instructions",
                                       [FeatureAVX512]>;
 def FeatureVLX     : SubtargetFeature<"avx512vl", "HasVLX", "true",
                       "Enable AVX-512 Vector Length eXtensions",
                                       [FeatureAVX512]>;
 def FeatureVBMI     : SubtargetFeature<"avx512vbmi", "HasVBMI", "true",
-                      "Enable AVX-512 Vector Bit Manipulation Instructions",
-                                      [FeatureAVX512]>;
+                      "Enable AVX-512 Vector Byte Manipulation Instructions",
+                                      [FeatureBWI]>;
 def FeatureIFMA     : SubtargetFeature<"avx512ifma", "HasIFMA", "true",
                       "Enable AVX-512 Integer Fused Multiple-Add",
                                       [FeatureAVX512]>;
 def FeaturePKU   : SubtargetFeature<"pku", "HasPKU", "true",
                       "Enable protection keys">;
 def FeaturePCLMUL  : SubtargetFeature<"pclmul", "HasPCLMUL", "true",
                          "Enable packed carry-less multiplication instructions",
                                [FeatureSSE2]>;

llvm/trunk/test/CodeGen/X86/avx512vbmi-intrinsics.ll

 ; 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
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -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:    vpaddb %zmm0, %zmm3, %zmm0
+; CHECK-NEXT:    vpaddb %zmm0, %zmm2, %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)
 
 define <64 x i8>@test_int_x86_avx512_mask_pmultishift_qb_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_pmultishift_qb_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovq %rdi, %k1
 ; CHECK-NEXT:    vpmultishiftqb %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpmultishiftqb %zmm1, %zmm0, %zmm3 {%k1} {z}
 ; CHECK-NEXT:    vpmultishiftqb %zmm1, %zmm0, %zmm0
-; CHECK-NEXT:    vpaddb %zmm3, %zmm2, %zmm1
-; CHECK-NEXT:    vpaddb %zmm0, %zmm1, %zmm0
+; CHECK-NEXT:    vpaddb %zmm0, %zmm3, %zmm0
+; CHECK-NEXT:    vpaddb %zmm0, %zmm2, %zmm0
 ; CHECK-NEXT:    retq
   %res = call <64 x i8> @llvm.x86.avx512.mask.pmultishift.qb.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
   %res1 = call <64 x i8> @llvm.x86.avx512.mask.pmultishift.qb.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> zeroinitializer, i64 %x3)
   %res2 = call <64 x i8> @llvm.x86.avx512.mask.pmultishift.qb.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.vpermi2var.qi.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
 
 define <64 x i8>@test_int_x86_avx512_mask_vpermi2var_qi_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_qi_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovq %rdi, %k1
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm3
 ; CHECK-NEXT:    vpermi2b %zmm2, %zmm0, %zmm3 {%k1}
 ; CHECK-NEXT:    vpermi2b %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vpxord %zmm4, %zmm4, %zmm4
 ; CHECK-NEXT:    vpermi2b %zmm2, %zmm0, %zmm4 {%k1} {z}
-; CHECK-NEXT:    vpaddb %zmm1, %zmm3, %zmm0
-; CHECK-NEXT:    vpaddb %zmm0, %zmm4, %zmm0
+; CHECK-NEXT:    vpaddb %zmm1, %zmm4, %zmm0
+; CHECK-NEXT:    vpaddb %zmm0, %zmm3, %zmm0
 ; CHECK-NEXT:    retq
   %res = call <64 x i8> @llvm.x86.avx512.mask.vpermi2var.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.vpermi2var.qi.512(<64 x i8> %x0, <64 x i8> zeroinitializer, <64 x i8> %x2, i64 %x3)
   %res2 = call <64 x i8> @llvm.x86.avx512.mask.vpermi2var.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.vpermt2var.qi.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
 
 define <64 x i8>@test_int_x86_avx512_mask_vpermt2var_qi_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_qi_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovq %rdi, %k1
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm3
 ; CHECK-NEXT:    vpermt2b %zmm2, %zmm0, %zmm3 {%k1}
 ; CHECK-NEXT:    vpermt2b %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vpxord %zmm4, %zmm4, %zmm4
 ; CHECK-NEXT:    vpermt2b %zmm2, %zmm0, %zmm4 {%k1} {z}
-; CHECK-NEXT:    vpaddb %zmm1, %zmm3, %zmm0
-; CHECK-NEXT:    vpaddb %zmm0, %zmm4, %zmm0
+; CHECK-NEXT:    vpaddb %zmm1, %zmm4, %zmm0
+; CHECK-NEXT:    vpaddb %zmm0, %zmm3, %zmm0
 ; CHECK-NEXT:    retq
   %res = call <64 x i8> @llvm.x86.avx512.mask.vpermt2var.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.vpermt2var.qi.512(<64 x i8> %x0, <64 x i8> zeroinitializer, <64 x i8> %x2, i64 %x3)
   %res2 = call <64 x i8> @llvm.x86.avx512.mask.vpermt2var.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
 }

llvm/trunk/test/CodeGen/X86/avx512vbmivl-intrinsics.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; NOTE: Assertions have been autogenerated by update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx -mattr=+avx512vl -mattr=+avx512vbmi --show-mc-encoding| FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512vl -mattr=+avx512vbmi --show-mc-encoding| FileCheck %s
 declare <16 x i8> @llvm.x86.avx512.mask.permvar.qi.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
 
 define <16 x i8>@test_int_x86_avx512_mask_permvar_qi_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3) {
 ; 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 %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:    vpaddb %xmm0, %xmm3, %xmm0 ## encoding: [0x62,0xf1,0x65,0x08,0xfc,0xc0]
+; CHECK-NEXT:    vpaddb %xmm0, %xmm2, %xmm0 ## encoding: [0x62,0xf1,0x6d,0x08,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <16 x i8> @llvm.x86.avx512.mask.permvar.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3)
   %res1 = call <16 x i8> @llvm.x86.avx512.mask.permvar.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> zeroinitializer, i16 %x3)
   %res2 = call <16 x i8> @llvm.x86.avx512.mask.permvar.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 -1)
   %res3 = add <16 x i8> %res, %res1
   %res4 = add <16 x i8> %res3, %res2
   ret <16 x i8> %res4
 }
 
 declare <32 x i8> @llvm.x86.avx512.mask.permvar.qi.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
 
 define <32 x i8>@test_int_x86_avx512_mask_permvar_qi_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
 ; 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 %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:    vpaddb %ymm0, %ymm3, %ymm0 ## encoding: [0x62,0xf1,0x65,0x28,0xfc,0xc0]
+; CHECK-NEXT:    vpaddb %ymm0, %ymm2, %ymm0 ## encoding: [0x62,0xf1,0x6d,0x28,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <32 x i8> @llvm.x86.avx512.mask.permvar.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
   %res1 = call <32 x i8> @llvm.x86.avx512.mask.permvar.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> zeroinitializer, i32 %x3)
   %res2 = call <32 x i8> @llvm.x86.avx512.mask.permvar.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
   %res3 = add <32 x i8> %res, %res1
   %res4 = add <32 x i8> %res3, %res2
   ret <32 x i8> %res4
 }
 
 declare <16 x i8> @llvm.x86.avx512.mask.pmultishift.qb.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
 
 define <16 x i8>@test_int_x86_avx512_mask_pmultishift_qb_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_pmultishift_qb_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
 ; CHECK-NEXT:    vpmultishiftqb %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0x83,0xd1]
 ; CHECK-NEXT:    vpmultishiftqb %xmm1, %xmm0, %xmm3 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0x89,0x83,0xd9]
 ; CHECK-NEXT:    vpmultishiftqb %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf2,0xfd,0x08,0x83,0xc1]
-; 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:    vpaddb %xmm0, %xmm3, %xmm0 ## encoding: [0x62,0xf1,0x65,0x08,0xfc,0xc0]
+; CHECK-NEXT:    vpaddb %xmm0, %xmm2, %xmm0 ## encoding: [0x62,0xf1,0x6d,0x08,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <16 x i8> @llvm.x86.avx512.mask.pmultishift.qb.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3)
   %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmultishift.qb.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> zeroinitializer, i16 %x3)
   %res2 = call <16 x i8> @llvm.x86.avx512.mask.pmultishift.qb.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 -1)
   %res3 = add <16 x i8> %res, %res1
   %res4 = add <16 x i8> %res3, %res2
   ret <16 x i8> %res4
 }
 
 declare <32 x i8> @llvm.x86.avx512.mask.pmultishift.qb.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
 
 define <32 x i8>@test_int_x86_avx512_mask_pmultishift_qb_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_pmultishift_qb_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
 ; CHECK-NEXT:    vpmultishiftqb %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x83,0xd1]
 ; CHECK-NEXT:    vpmultishiftqb %ymm1, %ymm0, %ymm3 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0xa9,0x83,0xd9]
 ; CHECK-NEXT:    vpmultishiftqb %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0xfd,0x28,0x83,0xc1]
-; 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:    vpaddb %ymm0, %ymm3, %ymm0 ## encoding: [0x62,0xf1,0x65,0x28,0xfc,0xc0]
+; CHECK-NEXT:    vpaddb %ymm0, %ymm2, %ymm0 ## encoding: [0x62,0xf1,0x6d,0x28,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <32 x i8> @llvm.x86.avx512.mask.pmultishift.qb.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
   %res1 = call <32 x i8> @llvm.x86.avx512.mask.pmultishift.qb.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> zeroinitializer, i32 %x3)
   %res2 = call <32 x i8> @llvm.x86.avx512.mask.pmultishift.qb.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
   %res3 = add <32 x i8> %res, %res1
   %res4 = add <32 x i8> %res3, %res2
   ret <32 x i8> %res4
 }
 
 declare <16 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
 
 define <16 x i8>@test_int_x86_avx512_mask_vpermi2var_qi_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_qi_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
 ; CHECK-NEXT:    vmovdqa64 %xmm1, %xmm3 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xd9]
 ; CHECK-NEXT:    vpermi2b %xmm2, %xmm0, %xmm3 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0x75,0xda]
 ; CHECK-NEXT:    vpermi2b %xmm2, %xmm0, %xmm1 ## encoding: [0x62,0xf2,0x7d,0x08,0x75,0xca]
 ; CHECK-NEXT:    vpxord %xmm4, %xmm4, %xmm4 ## encoding: [0x62,0xf1,0x5d,0x08,0xef,0xe4]
 ; CHECK-NEXT:    vpermi2b %xmm2, %xmm0, %xmm4 {%k1} {z} ## encoding: [0x62,0xf2,0x7d,0x89,0x75,0xe2]
-; CHECK-NEXT:    vpaddb %xmm1, %xmm3, %xmm0 ## encoding: [0x62,0xf1,0x65,0x08,0xfc,0xc1]
-; CHECK-NEXT:    vpaddb %xmm0, %xmm4, %xmm0 ## encoding: [0x62,0xf1,0x5d,0x08,0xfc,0xc0]
+; CHECK-NEXT:    vpaddb %xmm1, %xmm4, %xmm0 ## encoding: [0x62,0xf1,0x5d,0x08,0xfc,0xc1]
+; CHECK-NEXT:    vpaddb %xmm0, %xmm3, %xmm0 ## encoding: [0x62,0xf1,0x65,0x08,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <16 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3)
   %res1 = call <16 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.128(<16 x i8> %x0, <16 x i8> zeroinitializer, <16 x i8> %x2, i16 %x3)
   %res2 = call <16 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 -1)
   %res3 = add <16 x i8> %res, %res1
   %res4 = add <16 x i8> %res3, %res2
   ret <16 x i8> %res4
 }
 
 declare <32 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
 
 define <32 x i8>@test_int_x86_avx512_mask_vpermi2var_qi_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_qi_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
 ; CHECK-NEXT:    vmovdqa64 %ymm1, %ymm3 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xd9]
 ; CHECK-NEXT:    vpermi2b %ymm2, %ymm0, %ymm3 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x75,0xda]
 ; CHECK-NEXT:    vpermi2b %ymm2, %ymm0, %ymm1 ## encoding: [0x62,0xf2,0x7d,0x28,0x75,0xca]
 ; CHECK-NEXT:    vpxord %ymm4, %ymm4, %ymm4 ## encoding: [0x62,0xf1,0x5d,0x28,0xef,0xe4]
 ; CHECK-NEXT:    vpermi2b %ymm2, %ymm0, %ymm4 {%k1} {z} ## encoding: [0x62,0xf2,0x7d,0xa9,0x75,0xe2]
-; CHECK-NEXT:    vpaddb %ymm1, %ymm3, %ymm0 ## encoding: [0x62,0xf1,0x65,0x28,0xfc,0xc1]
-; CHECK-NEXT:    vpaddb %ymm0, %ymm4, %ymm0 ## encoding: [0x62,0xf1,0x5d,0x28,0xfc,0xc0]
+; CHECK-NEXT:    vpaddb %ymm1, %ymm4, %ymm0 ## encoding: [0x62,0xf1,0x5d,0x28,0xfc,0xc1]
+; CHECK-NEXT:    vpaddb %ymm0, %ymm3, %ymm0 ## encoding: [0x62,0xf1,0x65,0x28,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <32 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
   %res1 = call <32 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.256(<32 x i8> %x0, <32 x i8> zeroinitializer, <32 x i8> %x2, i32 %x3)
   %res2 = call <32 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
   %res3 = add <32 x i8> %res, %res1
   %res4 = add <32 x i8> %res3, %res2
   ret <32 x i8> %res4
 }
 
 declare <16 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
 
 define <16 x i8>@test_int_x86_avx512_mask_vpermt2var_qi_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_qi_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
 ; CHECK-NEXT:    vmovdqa64 %xmm1, %xmm3 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xd9]
 ; CHECK-NEXT:    vpermt2b %xmm2, %xmm0, %xmm3 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0x7d,0xda]
 ; CHECK-NEXT:    vpermt2b %xmm2, %xmm0, %xmm1 ## encoding: [0x62,0xf2,0x7d,0x08,0x7d,0xca]
 ; CHECK-NEXT:    vpxord %xmm4, %xmm4, %xmm4 ## encoding: [0x62,0xf1,0x5d,0x08,0xef,0xe4]
 ; CHECK-NEXT:    vpermt2b %xmm2, %xmm0, %xmm4 {%k1} {z} ## encoding: [0x62,0xf2,0x7d,0x89,0x7d,0xe2]
-; CHECK-NEXT:    vpaddb %xmm1, %xmm3, %xmm0 ## encoding: [0x62,0xf1,0x65,0x08,0xfc,0xc1]
-; CHECK-NEXT:    vpaddb %xmm0, %xmm4, %xmm0 ## encoding: [0x62,0xf1,0x5d,0x08,0xfc,0xc0]
+; CHECK-NEXT:    vpaddb %xmm1, %xmm4, %xmm0 ## encoding: [0x62,0xf1,0x5d,0x08,0xfc,0xc1]
+; CHECK-NEXT:    vpaddb %xmm0, %xmm3, %xmm0 ## encoding: [0x62,0xf1,0x65,0x08,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <16 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3)
   %res1 = call <16 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.128(<16 x i8> %x0, <16 x i8> zeroinitializer, <16 x i8> %x2, i16 %x3)
   %res2 = call <16 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 -1)
   %res3 = add <16 x i8> %res, %res1
   %res4 = add <16 x i8> %res3, %res2
   ret <16 x i8> %res4
 }
 
 declare <32 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
 
 define <32 x i8>@test_int_x86_avx512_mask_vpermt2var_qi_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_qi_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
 ; CHECK-NEXT:    vmovdqa64 %ymm1, %ymm3 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xd9]
 ; CHECK-NEXT:    vpermt2b %ymm2, %ymm0, %ymm3 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x7d,0xda]
 ; CHECK-NEXT:    vpermt2b %ymm2, %ymm0, %ymm1 ## encoding: [0x62,0xf2,0x7d,0x28,0x7d,0xca]
 ; CHECK-NEXT:    vpxord %ymm4, %ymm4, %ymm4 ## encoding: [0x62,0xf1,0x5d,0x28,0xef,0xe4]
 ; CHECK-NEXT:    vpermt2b %ymm2, %ymm0, %ymm4 {%k1} {z} ## encoding: [0x62,0xf2,0x7d,0xa9,0x7d,0xe2]
-; CHECK-NEXT:    vpaddb %ymm1, %ymm3, %ymm0 ## encoding: [0x62,0xf1,0x65,0x28,0xfc,0xc1]
-; CHECK-NEXT:    vpaddb %ymm0, %ymm4, %ymm0 ## encoding: [0x62,0xf1,0x5d,0x28,0xfc,0xc0]
+; CHECK-NEXT:    vpaddb %ymm1, %ymm4, %ymm0 ## encoding: [0x62,0xf1,0x5d,0x28,0xfc,0xc1]
+; CHECK-NEXT:    vpaddb %ymm0, %ymm3, %ymm0 ## encoding: [0x62,0xf1,0x65,0x28,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <32 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
   %res1 = call <32 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.256(<32 x i8> %x0, <32 x i8> zeroinitializer, <32 x i8> %x2, i32 %x3)
   %res2 = call <32 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
   %res3 = add <32 x i8> %res, %res1
   %res4 = add <32 x i8> %res3, %res2
   ret <32 x i8> %res4
 }