[X86,AVX-512] Use INSERT_SUBREG instead of SUBREG_TO_REG when the input is not the...

SUBREG_TO_REG is supposed to indicate that the super register has been zeroed, but we can't prove that if we don't know where it came from.

llvm-svn: 281885

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 11b5246..99fe48a 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -1421,15 +1421,15 @@ def : Pat<(v8f32 (vselect (v8i1 VK8WM:$mask), (v8f32 VR256X:$src1),
                             (v8f32 VR256X:$src2))),
             (EXTRACT_SUBREG
               (v16f32 (VBLENDMPSZrrk (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
-            (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)),
-            (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
+            (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm)),
+            (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)))), sub_ymm)>;
 
 def : Pat<(v8i32 (vselect (v8i1 VK8WM:$mask), (v8i32 VR256X:$src1),
                             (v8i32 VR256X:$src2))),
             (EXTRACT_SUBREG
                 (v16i32 (VPBLENDMDZrrk (COPY_TO_REGCLASS VK8WM:$mask, VK16WM),
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)),
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm)),
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)))), sub_ymm)>;
 }
 //===----------------------------------------------------------------------===//
 // Compare Instructions
@@ -1636,13 +1636,13 @@ defm VPCMPGTQ : avx512_icmp_packed_rmb_vl<0x37, "vpcmpgtq", X86pcmpgtm,
 let Predicates = [HasAVX512, NoVLX] in {
 def : Pat<(v8i1 (X86pcmpgtm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2))),
             (COPY_TO_REGCLASS (VPCMPGTDZrr
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)),
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm))), VK8)>;
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)),
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm))), VK8)>;
 
 def : Pat<(v8i1 (X86pcmpeqm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2))),
             (COPY_TO_REGCLASS (VPCMPEQDZrr
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)),
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm))), VK8)>;
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)),
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm))), VK8)>;
 }
 
 multiclass avx512_icmp_cc<bits<8> opc, string Suffix, SDNode OpNode,
@@ -1897,18 +1897,18 @@ defm VCMPPS : avx512_vcmp<avx512vl_f32_info>,
 
 def : Pat<(v8i1 (X86cmpm (v8f32 VR256X:$src1), (v8f32 VR256X:$src2), imm:$cc)),
           (COPY_TO_REGCLASS (VCMPPSZrri
-            (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)),
-            (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)),
+            (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)),
+            (v16f32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm)),
             imm:$cc), VK8)>;
 def : Pat<(v8i1 (X86cmpm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2), imm:$cc)),
           (COPY_TO_REGCLASS (VPCMPDZrri
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)),
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)),
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)),
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm)),
             imm:$cc), VK8)>;
 def : Pat<(v8i1 (X86cmpmu (v8i32 VR256X:$src1), (v8i32 VR256X:$src2), imm:$cc)),
           (COPY_TO_REGCLASS (VPCMPUDZrri
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)),
-            (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)),
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src1, sub_ymm)),
+            (v16i32 (INSERT_SUBREG (IMPLICIT_DEF), VR256X:$src2, sub_ymm)),
             imm:$cc), VK8)>;
 
 // ----------------------------------------------------------------
@@ -2097,7 +2097,7 @@ def : Pat<(i8 (bitconvert (v8i1 VK8:$src))),
 def : Pat<(i32 (zext (i16 (bitconvert (v16i1 VK16:$src))))),
           (KMOVWrk VK16:$src)>;
 def : Pat<(i32 (anyext (i16 (bitconvert (v16i1 VK16:$src))))),
-          (i32 (SUBREG_TO_REG (i64 0),
+          (i32 (INSERT_SUBREG (IMPLICIT_DEF),
                 (i16 (COPY_TO_REGCLASS VK16:$src, GR16)), sub_16bit))>;
 
 def : Pat<(i32 (zext (i8 (bitconvert (v8i1 VK8:$src))))),
@@ -2105,7 +2105,7 @@ def : Pat<(i32 (zext (i8 (bitconvert (v8i1 VK8:$src))))),
 def : Pat<(i32 (zext (i8 (bitconvert (v8i1 VK8:$src))))),
           (KMOVBrk VK8:$src)>, Requires<[HasDQI]>;
 def : Pat<(i32 (anyext (i8 (bitconvert (v8i1 VK8:$src))))),
-          (i32 (SUBREG_TO_REG (i64 0),
+          (i32 (INSERT_SUBREG (IMPLICIT_DEF),
                 (i8 (COPY_TO_REGCLASS VK8:$src, GR8)), sub_8bit))>;
 
 def : Pat<(v32i1 (bitconvert (i32 GR32:$src))),
@@ -2194,12 +2194,14 @@ let Predicates = [HasAVX512] in {
 
   def : Pat<(i1 (trunc (i8 GR8:$src))),
        (COPY_TO_REGCLASS
-        (KMOVWkr (AND32ri8 (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit), (i32 1))),
+        (KMOVWkr (AND32ri8 (INSERT_SUBREG (i32 (IMPLICIT_DEF)),
+                                          GR8:$src, sub_8bit), (i32 1))),
        VK1)>;
 
   def : Pat<(i1 (trunc (i16 GR16:$src))),
        (COPY_TO_REGCLASS
-        (KMOVWkr (AND32ri8 (SUBREG_TO_REG (i32 0), $src, sub_16bit), (i32 1))),
+        (KMOVWkr (AND32ri8 (INSERT_SUBREG (i32 (IMPLICIT_DEF)),
+                                          GR16:$src, sub_16bit), (i32 1))),
        VK1)>;
 
   def : Pat<(i32 (zext VK1:$src)),
@@ -2221,7 +2223,7 @@ let Predicates = [HasAVX512] in {
              (KMOVWrk (COPY_TO_REGCLASS VK1:$src, VK16)), sub_32bit), (i64 1))>;
 
   def : Pat<(i64 (anyext VK1:$src)),
-            (SUBREG_TO_REG (i64 0),
+            (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
              (i32 (COPY_TO_REGCLASS VK1:$src, GR32)), sub_32bit)>;
 
   def : Pat<(i16 (zext VK1:$src)),
@@ -6188,27 +6190,33 @@ defm VCVTUQQ2PS : avx512_cvtqq2ps<0x7A, "vcvtuqq2ps", uint_to_fp,
 let Predicates = [HasAVX512, NoVLX] in {
 def : Pat<(v8i32 (fp_to_uint (v8f32 VR256X:$src1))),
           (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
-           (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
+           (v16f32 (INSERT_SUBREG (IMPLICIT_DEF),
+                                  VR256X:$src1, sub_ymm)))), sub_ymm)>;
 
 def : Pat<(v4i32 (fp_to_uint (v4f32 VR128X:$src1))),
           (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
-           (v16f32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_xmm)>;
+           (v16f32 (INSERT_SUBREG (IMPLICIT_DEF),
+                                  VR128X:$src1, sub_xmm)))), sub_xmm)>;
 
 def : Pat<(v4i32 (fp_to_uint (v4f64 VR256X:$src1))),
           (EXTRACT_SUBREG (v8i32 (VCVTTPD2UDQZrr
-           (v8f64 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_xmm)>;
+           (v8f64 (INSERT_SUBREG (IMPLICIT_DEF),
+                                 VR256X:$src1, sub_ymm)))), sub_xmm)>;
 
 def : Pat<(v8f32 (uint_to_fp (v8i32 VR256X:$src1))),
           (EXTRACT_SUBREG (v16f32 (VCVTUDQ2PSZrr
-           (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
+           (v16i32 (INSERT_SUBREG (IMPLICIT_DEF),
+                                  VR256X:$src1, sub_ymm)))), sub_ymm)>;
 
 def : Pat<(v4f32 (uint_to_fp (v4i32 VR128X:$src1))),
           (EXTRACT_SUBREG (v16f32 (VCVTUDQ2PSZrr
-           (v16i32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_xmm)>;
+           (v16i32 (INSERT_SUBREG (IMPLICIT_DEF),
+                                  VR128X:$src1, sub_xmm)))), sub_xmm)>;
 
 def : Pat<(v4f64 (uint_to_fp (v4i32 VR128X:$src1))),
           (EXTRACT_SUBREG (v8f64 (VCVTUDQ2PDZrr
-           (v8i32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_ymm)>;
+           (v8i32 (INSERT_SUBREG (IMPLICIT_DEF),
+                                 VR128X:$src1, sub_xmm)))), sub_ymm)>;
 }
 
 let Predicates = [HasAVX512] in {
@@ -6951,17 +6959,17 @@ defm VPMOVUSWB  : avx512_trunc_sat_wb<0x10, "us",  X86vtruncus>;
 let Predicates = [HasAVX512, NoVLX] in {
 def: Pat<(v8i16 (X86vtrunc (v8i32 VR256X:$src))),
          (v8i16 (EXTRACT_SUBREG
-                 (v16i16 (VPMOVDWZrr (v16i32 (SUBREG_TO_REG (i32 0),
+                 (v16i16 (VPMOVDWZrr (v16i32 (INSERT_SUBREG (IMPLICIT_DEF),
                                           VR256X:$src, sub_ymm)))), sub_xmm))>;
 def: Pat<(v4i32 (X86vtrunc (v4i64 VR256X:$src))),
          (v4i32 (EXTRACT_SUBREG
-                 (v8i32 (VPMOVQDZrr (v8i64 (SUBREG_TO_REG (i32 0),
+                 (v8i32 (VPMOVQDZrr (v8i64 (INSERT_SUBREG (IMPLICIT_DEF),
                                            VR256X:$src, sub_ymm)))), sub_xmm))>;
 }
 
 let Predicates = [HasBWI, NoVLX] in {
 def: Pat<(v16i8 (X86vtrunc (v16i16 VR256X:$src))),
-         (v16i8 (EXTRACT_SUBREG  (VPMOVWBZrr (v32i16 (SUBREG_TO_REG (i32 0),
+         (v16i8 (EXTRACT_SUBREG (VPMOVWBZrr (v32i16 (INSERT_SUBREG (IMPLICIT_DEF),
                                             VR256X:$src, sub_ymm))), sub_xmm))>;
 }
 

diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index ebbed64..d02f833 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -8362,20 +8362,24 @@ let Predicates = [HasAVX2, NoVLX] in {
 let Predicates = [HasAVX2, NoVLX_Or_NoBWI], AddedComplexity = 20 in {
   def : Pat<(v16i8 (X86VBroadcast GR8:$src)),
         (VPBROADCASTBrr (COPY_TO_REGCLASS
-                         (i32 (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)),
+                         (i32 (INSERT_SUBREG (i32 (IMPLICIT_DEF)),
+                                             GR8:$src, sub_8bit)),
                          VR128))>;
   def : Pat<(v32i8 (X86VBroadcast GR8:$src)),
         (VPBROADCASTBYrr (COPY_TO_REGCLASS
-                          (i32 (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)),
+                          (i32 (INSERT_SUBREG (i32 (IMPLICIT_DEF)),
+                                              GR8:$src, sub_8bit)),
                           VR128))>;
 
   def : Pat<(v8i16 (X86VBroadcast GR16:$src)),
         (VPBROADCASTWrr (COPY_TO_REGCLASS
-                         (i32 (SUBREG_TO_REG (i32 0), GR16:$src, sub_16bit)),
+                         (i32 (INSERT_SUBREG (i32 (IMPLICIT_DEF)),
+                                             GR16:$src, sub_16bit)),
                          VR128))>;
   def : Pat<(v16i16 (X86VBroadcast GR16:$src)),
         (VPBROADCASTWYrr (COPY_TO_REGCLASS
-                          (i32 (SUBREG_TO_REG (i32 0), GR16:$src, sub_16bit)),
+                          (i32 (INSERT_SUBREG (i32 (IMPLICIT_DEF)),
+                                              GR16:$src, sub_16bit)),
                           VR128))>;
 }
 let Predicates = [HasAVX2, NoVLX], AddedComplexity = 20 in {

diff --git a/llvm/test/CodeGen/X86/avx512-fsel.ll b/llvm/test/CodeGen/X86/avx512-fsel.ll
index 42469e5..2c9c42b 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512-fsel.ll
+++ b/llvm/test/CodeGen/X86/avx512-fsel.ll
@@ -17,10 +17,12 @@ define i32 @test(float %a, float %b)  {
 ; CHECK-NEXT:    setp %sil
 ; CHECK-NEXT:    setne %dil
 ; CHECK-NEXT:    andb %cl, %dl
+; CHECK-NEXT:    ## implicit-def: %R8D
 ; CHECK-NEXT:    movb %dl, %r8b
 ; CHECK-NEXT:    andl $1, %r8d
 ; CHECK-NEXT:    kmovw %r8d, %k0
 ; CHECK-NEXT:    orb %sil, %dil
+; CHECK-NEXT:    ## implicit-def: %R8D
 ; CHECK-NEXT:    movb %dil, %r8b
 ; CHECK-NEXT:    andl $1, %r8d
 ; CHECK-NEXT:    kmovw %r8d, %k1

diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics.ll b/llvm/test/CodeGen/X86/avx512-intrinsics.ll
index 0e8e9f4..3c74aec 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512-intrinsics.ll
@@ -1094,9 +1094,9 @@ define <8 x i8> @test_cmp_q_512(<8 x i64> %a0, <8 x i64> %a1) {
 ; CHECK-NEXT:    vpcmpnltq %zmm1, %zmm0, %k2
 ; CHECK-NEXT:    vpcmpnleq %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vpcmpordq %zmm1, %zmm0, %k0
-; CHECK-NEXT:    kmovw %k3, %eax
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0
 ; CHECK-NEXT:    kmovw %k4, %eax
+; CHECK-NEXT:    kmovw %k3, %ecx
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
 ; CHECK-NEXT:    kmovw %k5, %eax
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
@@ -1142,9 +1142,9 @@ define <8 x i8> @test_mask_cmp_q_512(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
 ; CHECK-NEXT:    vpcmpnltq %zmm1, %zmm0, %k2 {%k3}
 ; CHECK-NEXT:    vpcmpnleq %zmm1, %zmm0, %k1 {%k3}
 ; CHECK-NEXT:    vpcmpordq %zmm1, %zmm0, %k3 {%k3}
-; CHECK-NEXT:    kmovw %k4, %eax
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0
 ; CHECK-NEXT:    kmovw %k5, %eax
+; CHECK-NEXT:    kmovw %k4, %ecx
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
 ; CHECK-NEXT:    kmovw %k6, %eax
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
@@ -1191,9 +1191,9 @@ define <8 x i8> @test_ucmp_q_512(<8 x i64> %a0, <8 x i64> %a1) {
 ; CHECK-NEXT:    vpcmpnltuq %zmm1, %zmm0, %k2
 ; CHECK-NEXT:    vpcmpnleuq %zmm1, %zmm0, %k1
 ; CHECK-NEXT:    vpcmporduq %zmm1, %zmm0, %k0
-; CHECK-NEXT:    kmovw %k3, %eax
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0
 ; CHECK-NEXT:    kmovw %k4, %eax
+; CHECK-NEXT:    kmovw %k3, %ecx
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
 ; CHECK-NEXT:    kmovw %k5, %eax
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
@@ -1239,9 +1239,9 @@ define <8 x i8> @test_mask_ucmp_q_512(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
 ; CHECK-NEXT:    vpcmpnltuq %zmm1, %zmm0, %k2 {%k3}
 ; CHECK-NEXT:    vpcmpnleuq %zmm1, %zmm0, %k1 {%k3}
 ; CHECK-NEXT:    vpcmporduq %zmm1, %zmm0, %k3 {%k3}
-; CHECK-NEXT:    kmovw %k4, %eax
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0
 ; CHECK-NEXT:    kmovw %k5, %eax
+; CHECK-NEXT:    kmovw %k4, %ecx
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
 ; CHECK-NEXT:    kmovw %k6, %eax
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0

diff --git a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
index 3c12018..f0c4e65 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
@@ -694,9 +694,9 @@ define <8 x i8> @test_cmp_w_128(<8 x i16> %a0, <8 x i16> %a1) {
 ; CHECK-NEXT:    vpcmpnltw %xmm1, %xmm0, %k2 ## encoding: [0x62,0xf3,0xfd,0x08,0x3f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnlew %xmm1, %xmm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x08,0x3f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordw %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf3,0xfd,0x08,0x3f,0xc1,0x07]
-; CHECK-NEXT:    kmovw %k3, %eax ## encoding: [0xc5,0xf8,0x93,0xc3]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x00]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
+; CHECK-NEXT:    kmovw %k3, %ecx ## encoding: [0xc5,0xf8,0x93,0xcb]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x02]
 ; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x04]
@@ -742,9 +742,9 @@ define <8 x i8> @test_mask_cmp_w_128(<8 x i16> %a0, <8 x i16> %a1, i8 %mask) {
 ; CHECK-NEXT:    vpcmpnltw %xmm1, %xmm0, %k2 {%k3} ## encoding: [0x62,0xf3,0xfd,0x0b,0x3f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnlew %xmm1, %xmm0, %k1 {%k3} ## encoding: [0x62,0xf3,0xfd,0x0b,0x3f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordw %xmm1, %xmm0, %k3 {%k3} ## encoding: [0x62,0xf3,0xfd,0x0b,0x3f,0xd9,0x07]
-; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x00]
 ; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k4, %ecx ## encoding: [0xc5,0xf8,0x93,0xcc]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x02]
 ; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x04]
@@ -791,9 +791,9 @@ define <8 x i8> @test_ucmp_w_128(<8 x i16> %a0, <8 x i16> %a1) {
 ; CHECK-NEXT:    vpcmpnltuw %xmm1, %xmm0, %k2 ## encoding: [0x62,0xf3,0xfd,0x08,0x3e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleuw %xmm1, %xmm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x08,0x3e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmporduw %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf3,0xfd,0x08,0x3e,0xc1,0x07]
-; CHECK-NEXT:    kmovw %k3, %eax ## encoding: [0xc5,0xf8,0x93,0xc3]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x00]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
+; CHECK-NEXT:    kmovw %k3, %ecx ## encoding: [0xc5,0xf8,0x93,0xcb]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x02]
 ; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x04]
@@ -839,9 +839,9 @@ define <8 x i8> @test_mask_ucmp_w_128(<8 x i16> %a0, <8 x i16> %a1, i8 %mask) {
 ; CHECK-NEXT:    vpcmpnltuw %xmm1, %xmm0, %k2 {%k3} ## encoding: [0x62,0xf3,0xfd,0x0b,0x3e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleuw %xmm1, %xmm0, %k1 {%k3} ## encoding: [0x62,0xf3,0xfd,0x0b,0x3e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmporduw %xmm1, %xmm0, %k3 {%k3} ## encoding: [0x62,0xf3,0xfd,0x0b,0x3e,0xd9,0x07]
-; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x00]
 ; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k4, %ecx ## encoding: [0xc5,0xf8,0x93,0xcc]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x02]
 ; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0x62,0xf3,0x7d,0x08,0x20,0xc0,0x04]

diff --git a/llvm/test/CodeGen/X86/avx512dq-intrinsics.ll b/llvm/test/CodeGen/X86/avx512dq-intrinsics.ll
index e684ad9..f9a22d8 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512dq-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512dq-intrinsics.ll
@@ -469,9 +469,9 @@ define i16@test_int_x86_avx512_mask_fpclass_ps_512(<16 x float> %x0, i16 %x1) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_fpclass_ps_512:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1
-; CHECK-NEXT:    vfpclassps $4, %zmm0, %k0 {%k1}
-; CHECK-NEXT:    kmovw %k0, %ecx
 ; CHECK-NEXT:    vfpclassps $4, %zmm0, %k0
+; CHECK-NEXT:    kmovw %k0, %ecx
+; CHECK-NEXT:    vfpclassps $4, %zmm0, %k0 {%k1}
 ; CHECK-NEXT:    kmovw %k0, %eax
 ; CHECK-NEXT:    addl %ecx, %eax
 ; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>

diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
index 5932e14..bf780e6 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
@@ -14,9 +14,9 @@ define <8 x i8> @test_cmp_d_256(<8 x i32> %a0, <8 x i32> %a1) {
 ; CHECK-NEXT:    vpcmpnltd %ymm1, %ymm0, %k2 ## encoding: [0x62,0xf3,0x7d,0x28,0x1f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnled %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x28,0x1f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordd %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf3,0x7d,0x28,0x1f,0xc1,0x07]
-; CHECK-NEXT:    kmovw %k3, %eax ## encoding: [0xc5,0xf8,0x93,0xc3]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
+; CHECK-NEXT:    kmovw %k3, %ecx ## encoding: [0xc5,0xf8,0x93,0xcb]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
@@ -62,9 +62,9 @@ define <8 x i8> @test_mask_cmp_d_256(<8 x i32> %a0, <8 x i32> %a1, i8 %mask) {
 ; CHECK-NEXT:    vpcmpnltd %ymm1, %ymm0, %k2 {%k3} ## encoding: [0x62,0xf3,0x7d,0x2b,0x1f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnled %ymm1, %ymm0, %k1 {%k3} ## encoding: [0x62,0xf3,0x7d,0x2b,0x1f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordd %ymm1, %ymm0, %k3 {%k3} ## encoding: [0x62,0xf3,0x7d,0x2b,0x1f,0xd9,0x07]
-; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
 ; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k4, %ecx ## encoding: [0xc5,0xf8,0x93,0xcc]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
@@ -111,9 +111,9 @@ define <8 x i8> @test_ucmp_d_256(<8 x i32> %a0, <8 x i32> %a1) {
 ; CHECK-NEXT:    vpcmpnltud %ymm1, %ymm0, %k2 ## encoding: [0x62,0xf3,0x7d,0x28,0x1e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleud %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x28,0x1e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordud %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf3,0x7d,0x28,0x1e,0xc1,0x07]
-; CHECK-NEXT:    kmovw %k3, %eax ## encoding: [0xc5,0xf8,0x93,0xc3]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
+; CHECK-NEXT:    kmovw %k3, %ecx ## encoding: [0xc5,0xf8,0x93,0xcb]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
@@ -159,9 +159,9 @@ define <8 x i8> @test_mask_ucmp_d_256(<8 x i32> %a0, <8 x i32> %a1, i8 %mask) {
 ; CHECK-NEXT:    vpcmpnltud %ymm1, %ymm0, %k2 {%k3} ## encoding: [0x62,0xf3,0x7d,0x2b,0x1e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleud %ymm1, %ymm0, %k1 {%k3} ## encoding: [0x62,0xf3,0x7d,0x2b,0x1e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordud %ymm1, %ymm0, %k3 {%k3} ## encoding: [0x62,0xf3,0x7d,0x2b,0x1e,0xd9,0x07]
-; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
 ; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k4, %ecx ## encoding: [0xc5,0xf8,0x93,0xcc]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
 ; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
@@ -200,19 +200,19 @@ declare i8 @llvm.x86.avx512.mask.ucmp.d.256(<8 x i32>, <8 x i32>, i32, i8) nounw
 define <8 x i8> @test_cmp_q_256(<4 x i64> %a0, <4 x i64> %a1) {
 ; CHECK-LABEL: test_cmp_q_256:
 ; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpcmpeqq %ymm1, %ymm0, %k7 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xf9,0x00]
-; CHECK-NEXT:    vpcmpltq %ymm1, %ymm0, %k6 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleq %ymm1, %ymm0, %k5 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xe9,0x02]
+; CHECK-NEXT:    vpcmpeqq %ymm1, %ymm0, %k5 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltq %ymm1, %ymm0, %k7 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xf9,0x01]
+; CHECK-NEXT:    vpcmpleq %ymm1, %ymm0, %k6 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunordq %ymm1, %ymm0, %k4 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpneqq %ymm1, %ymm0, %k3 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltq %ymm1, %ymm0, %k2 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleq %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordq %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf3,0xfd,0x28,0x1f,0xc1,0x07]
 ; CHECK-NEXT:    kmovw %k7, %eax ## encoding: [0xc5,0xf8,0x93,0xc7]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -248,19 +248,19 @@ define <8 x i8> @test_mask_cmp_q_256(<4 x i64> %a0, <4 x i64> %a1, i8 %mask) {
 ; CHECK-LABEL: test_mask_cmp_q_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k7 ## encoding: [0xc5,0xf8,0x92,0xff]
-; CHECK-NEXT:    vpcmpeqq %ymm1, %ymm0, %k0 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xc1,0x00]
-; CHECK-NEXT:    vpcmpltq %ymm1, %ymm0, %k6 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleq %ymm1, %ymm0, %k5 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xe9,0x02]
+; CHECK-NEXT:    vpcmpeqq %ymm1, %ymm0, %k5 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltq %ymm1, %ymm0, %k0 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xc1,0x01]
+; CHECK-NEXT:    vpcmpleq %ymm1, %ymm0, %k6 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunordq %ymm1, %ymm0, %k4 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpneqq %ymm1, %ymm0, %k3 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltq %ymm1, %ymm0, %k2 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleq %ymm1, %ymm0, %k1 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordq %ymm1, %ymm0, %k7 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1f,0xf9,0x07]
 ; CHECK-NEXT:    kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -297,19 +297,19 @@ declare i8 @llvm.x86.avx512.mask.cmp.q.256(<4 x i64>, <4 x i64>, i32, i8) nounwi
 define <8 x i8> @test_ucmp_q_256(<4 x i64> %a0, <4 x i64> %a1) {
 ; CHECK-LABEL: test_ucmp_q_256:
 ; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpcmpequq %ymm1, %ymm0, %k7 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xf9,0x00]
-; CHECK-NEXT:    vpcmpltuq %ymm1, %ymm0, %k6 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleuq %ymm1, %ymm0, %k5 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xe9,0x02]
+; CHECK-NEXT:    vpcmpequq %ymm1, %ymm0, %k5 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltuq %ymm1, %ymm0, %k7 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xf9,0x01]
+; CHECK-NEXT:    vpcmpleuq %ymm1, %ymm0, %k6 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunorduq %ymm1, %ymm0, %k4 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpnequq %ymm1, %ymm0, %k3 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltuq %ymm1, %ymm0, %k2 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleuq %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmporduq %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf3,0xfd,0x28,0x1e,0xc1,0x07]
 ; CHECK-NEXT:    kmovw %k7, %eax ## encoding: [0xc5,0xf8,0x93,0xc7]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -345,19 +345,19 @@ define <8 x i8> @test_mask_ucmp_q_256(<4 x i64> %a0, <4 x i64> %a1, i8 %mask) {
 ; CHECK-LABEL: test_mask_ucmp_q_256:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k7 ## encoding: [0xc5,0xf8,0x92,0xff]
-; CHECK-NEXT:    vpcmpequq %ymm1, %ymm0, %k0 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xc1,0x00]
-; CHECK-NEXT:    vpcmpltuq %ymm1, %ymm0, %k6 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleuq %ymm1, %ymm0, %k5 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xe9,0x02]
+; CHECK-NEXT:    vpcmpequq %ymm1, %ymm0, %k5 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltuq %ymm1, %ymm0, %k0 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xc1,0x01]
+; CHECK-NEXT:    vpcmpleuq %ymm1, %ymm0, %k6 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunorduq %ymm1, %ymm0, %k4 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpnequq %ymm1, %ymm0, %k3 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltuq %ymm1, %ymm0, %k2 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleuq %ymm1, %ymm0, %k1 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmporduq %ymm1, %ymm0, %k7 {%k7} ## encoding: [0x62,0xf3,0xfd,0x2f,0x1e,0xf9,0x07]
 ; CHECK-NEXT:    kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -396,19 +396,19 @@ declare i8 @llvm.x86.avx512.mask.ucmp.q.256(<4 x i64>, <4 x i64>, i32, i8) nounw
 define <8 x i8> @test_cmp_d_128(<4 x i32> %a0, <4 x i32> %a1) {
 ; CHECK-LABEL: test_cmp_d_128:
 ; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpcmpeqd %xmm1, %xmm0, %k7 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xf9,0x00]
-; CHECK-NEXT:    vpcmpltd %xmm1, %xmm0, %k6 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xf1,0x01]
-; CHECK-NEXT:    vpcmpled %xmm1, %xmm0, %k5 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xe9,0x02]
+; CHECK-NEXT:    vpcmpeqd %xmm1, %xmm0, %k5 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltd %xmm1, %xmm0, %k7 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xf9,0x01]
+; CHECK-NEXT:    vpcmpled %xmm1, %xmm0, %k6 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunordd %xmm1, %xmm0, %k4 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpneqd %xmm1, %xmm0, %k3 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltd %xmm1, %xmm0, %k2 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnled %xmm1, %xmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordd %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf3,0x7d,0x08,0x1f,0xc1,0x07]
 ; CHECK-NEXT:    kmovw %k7, %eax ## encoding: [0xc5,0xf8,0x93,0xc7]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -444,19 +444,19 @@ define <8 x i8> @test_mask_cmp_d_128(<4 x i32> %a0, <4 x i32> %a1, i8 %mask) {
 ; CHECK-LABEL: test_mask_cmp_d_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k7 ## encoding: [0xc5,0xf8,0x92,0xff]
-; CHECK-NEXT:    vpcmpeqd %xmm1, %xmm0, %k0 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xc1,0x00]
-; CHECK-NEXT:    vpcmpltd %xmm1, %xmm0, %k6 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xf1,0x01]
-; CHECK-NEXT:    vpcmpled %xmm1, %xmm0, %k5 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xe9,0x02]
+; CHECK-NEXT:    vpcmpeqd %xmm1, %xmm0, %k5 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltd %xmm1, %xmm0, %k0 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xc1,0x01]
+; CHECK-NEXT:    vpcmpled %xmm1, %xmm0, %k6 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunordd %xmm1, %xmm0, %k4 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpneqd %xmm1, %xmm0, %k3 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltd %xmm1, %xmm0, %k2 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnled %xmm1, %xmm0, %k1 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordd %xmm1, %xmm0, %k7 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1f,0xf9,0x07]
 ; CHECK-NEXT:    kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -493,19 +493,19 @@ declare i8 @llvm.x86.avx512.mask.cmp.d.128(<4 x i32>, <4 x i32>, i32, i8) nounwi
 define <8 x i8> @test_ucmp_d_128(<4 x i32> %a0, <4 x i32> %a1) {
 ; CHECK-LABEL: test_ucmp_d_128:
 ; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpcmpequd %xmm1, %xmm0, %k7 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xf9,0x00]
-; CHECK-NEXT:    vpcmpltud %xmm1, %xmm0, %k6 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleud %xmm1, %xmm0, %k5 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xe9,0x02]
+; CHECK-NEXT:    vpcmpequd %xmm1, %xmm0, %k5 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltud %xmm1, %xmm0, %k7 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xf9,0x01]
+; CHECK-NEXT:    vpcmpleud %xmm1, %xmm0, %k6 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunordud %xmm1, %xmm0, %k4 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpnequd %xmm1, %xmm0, %k3 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltud %xmm1, %xmm0, %k2 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleud %xmm1, %xmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordud %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf3,0x7d,0x08,0x1e,0xc1,0x07]
 ; CHECK-NEXT:    kmovw %k7, %eax ## encoding: [0xc5,0xf8,0x93,0xc7]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -541,19 +541,19 @@ define <8 x i8> @test_mask_ucmp_d_128(<4 x i32> %a0, <4 x i32> %a1, i8 %mask) {
 ; CHECK-LABEL: test_mask_ucmp_d_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k7 ## encoding: [0xc5,0xf8,0x92,0xff]
-; CHECK-NEXT:    vpcmpequd %xmm1, %xmm0, %k0 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xc1,0x00]
-; CHECK-NEXT:    vpcmpltud %xmm1, %xmm0, %k6 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleud %xmm1, %xmm0, %k5 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xe9,0x02]
+; CHECK-NEXT:    vpcmpequd %xmm1, %xmm0, %k5 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltud %xmm1, %xmm0, %k0 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xc1,0x01]
+; CHECK-NEXT:    vpcmpleud %xmm1, %xmm0, %k6 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunordud %xmm1, %xmm0, %k4 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpnequd %xmm1, %xmm0, %k3 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltud %xmm1, %xmm0, %k2 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleud %xmm1, %xmm0, %k1 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordud %xmm1, %xmm0, %k7 {%k7} ## encoding: [0x62,0xf3,0x7d,0x0f,0x1e,0xf9,0x07]
 ; CHECK-NEXT:    kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -590,19 +590,19 @@ declare i8 @llvm.x86.avx512.mask.ucmp.d.128(<4 x i32>, <4 x i32>, i32, i8) nounw
 define <8 x i8> @test_cmp_q_128(<2 x i64> %a0, <2 x i64> %a1) {
 ; CHECK-LABEL: test_cmp_q_128:
 ; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpcmpeqq %xmm1, %xmm0, %k7 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xf9,0x00]
-; CHECK-NEXT:    vpcmpltq %xmm1, %xmm0, %k6 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleq %xmm1, %xmm0, %k5 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xe9,0x02]
+; CHECK-NEXT:    vpcmpeqq %xmm1, %xmm0, %k5 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltq %xmm1, %xmm0, %k7 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xf9,0x01]
+; CHECK-NEXT:    vpcmpleq %xmm1, %xmm0, %k6 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunordq %xmm1, %xmm0, %k4 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpneqq %xmm1, %xmm0, %k3 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltq %xmm1, %xmm0, %k2 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleq %xmm1, %xmm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordq %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf3,0xfd,0x08,0x1f,0xc1,0x07]
 ; CHECK-NEXT:    kmovw %k7, %eax ## encoding: [0xc5,0xf8,0x93,0xc7]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -638,19 +638,19 @@ define <8 x i8> @test_mask_cmp_q_128(<2 x i64> %a0, <2 x i64> %a1, i8 %mask) {
 ; CHECK-LABEL: test_mask_cmp_q_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k7 ## encoding: [0xc5,0xf8,0x92,0xff]
-; CHECK-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xc1,0x00]
-; CHECK-NEXT:    vpcmpltq %xmm1, %xmm0, %k6 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleq %xmm1, %xmm0, %k5 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xe9,0x02]
+; CHECK-NEXT:    vpcmpeqq %xmm1, %xmm0, %k5 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltq %xmm1, %xmm0, %k0 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xc1,0x01]
+; CHECK-NEXT:    vpcmpleq %xmm1, %xmm0, %k6 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunordq %xmm1, %xmm0, %k4 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpneqq %xmm1, %xmm0, %k3 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltq %xmm1, %xmm0, %k2 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleq %xmm1, %xmm0, %k1 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xc9,0x06]
 ; CHECK-NEXT:    vpcmpordq %xmm1, %xmm0, %k7 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1f,0xf9,0x07]
 ; CHECK-NEXT:    kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -687,19 +687,19 @@ declare i8 @llvm.x86.avx512.mask.cmp.q.128(<2 x i64>, <2 x i64>, i32, i8) nounwi
 define <8 x i8> @test_ucmp_q_128(<2 x i64> %a0, <2 x i64> %a1) {
 ; CHECK-LABEL: test_ucmp_q_128:
 ; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpcmpequq %xmm1, %xmm0, %k7 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xf9,0x00]
-; CHECK-NEXT:    vpcmpltuq %xmm1, %xmm0, %k6 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleuq %xmm1, %xmm0, %k5 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xe9,0x02]
+; CHECK-NEXT:    vpcmpequq %xmm1, %xmm0, %k5 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltuq %xmm1, %xmm0, %k7 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xf9,0x01]
+; CHECK-NEXT:    vpcmpleuq %xmm1, %xmm0, %k6 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunorduq %xmm1, %xmm0, %k4 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpnequq %xmm1, %xmm0, %k3 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltuq %xmm1, %xmm0, %k2 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmporduq %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf3,0xfd,0x08,0x1e,0xc1,0x07]
 ; CHECK-NEXT:    kmovw %k7, %eax ## encoding: [0xc5,0xf8,0x93,0xc7]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]
@@ -735,19 +735,19 @@ define <8 x i8> @test_mask_ucmp_q_128(<2 x i64> %a0, <2 x i64> %a1, i8 %mask) {
 ; CHECK-LABEL: test_mask_ucmp_q_128:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k7 ## encoding: [0xc5,0xf8,0x92,0xff]
-; CHECK-NEXT:    vpcmpequq %xmm1, %xmm0, %k0 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xc1,0x00]
-; CHECK-NEXT:    vpcmpltuq %xmm1, %xmm0, %k6 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xf1,0x01]
-; CHECK-NEXT:    vpcmpleuq %xmm1, %xmm0, %k5 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xe9,0x02]
+; CHECK-NEXT:    vpcmpequq %xmm1, %xmm0, %k5 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xe9,0x00]
+; CHECK-NEXT:    vpcmpltuq %xmm1, %xmm0, %k0 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xc1,0x01]
+; CHECK-NEXT:    vpcmpleuq %xmm1, %xmm0, %k6 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xf1,0x02]
 ; CHECK-NEXT:    vpcmpunorduq %xmm1, %xmm0, %k4 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xe1,0x03]
 ; CHECK-NEXT:    vpcmpnequq %xmm1, %xmm0, %k3 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xd9,0x04]
 ; CHECK-NEXT:    vpcmpnltuq %xmm1, %xmm0, %k2 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xd1,0x05]
 ; CHECK-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k1 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xc9,0x06]
 ; CHECK-NEXT:    vpcmporduq %xmm1, %xmm0, %k7 {%k7} ## encoding: [0x62,0xf3,0xfd,0x0f,0x1e,0xf9,0x07]
 ; CHECK-NEXT:    kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT:    vpinsrb $0, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x00]
-; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
+; CHECK-NEXT:    kmovw %k5, %ecx ## encoding: [0xc5,0xf8,0x93,0xcd]
+; CHECK-NEXT:    vpinsrb $0, %ecx, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc1,0x00]
 ; CHECK-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x02]
-; CHECK-NEXT:    kmovw %k5, %eax ## encoding: [0xc5,0xf8,0x93,0xc5]
+; CHECK-NEXT:    kmovw %k6, %eax ## encoding: [0xc5,0xf8,0x93,0xc6]
 ; CHECK-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x04]
 ; CHECK-NEXT:    kmovw %k4, %eax ## encoding: [0xc5,0xf8,0x93,0xc4]
 ; CHECK-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x20,0xc0,0x06]

diff --git a/llvm/test/CodeGen/X86/pr27591.ll b/llvm/test/CodeGen/X86/pr27591.ll
index 53a23b3..3331a93 100644 (file)
--- a/llvm/test/CodeGen/X86/pr27591.ll
+++ b/llvm/test/CodeGen/X86/pr27591.ll
@@ -8,6 +8,7 @@ define void @test1(i32 %x) #0 {
 ; CHECK-NEXT:    pushq %rax
 ; CHECK-NEXT:    testl %edi, %edi
 ; CHECK-NEXT:    setne %al
+; CHECK-NEXT:  # implicit-def: %EDI
 ; CHECK-NEXT:    movb %al, %dil
 ; CHECK-NEXT:    andl $1, %edi
 ; CHECK-NEXT:    kmovw %edi, %k0
@@ -29,6 +30,7 @@ define void @test2(i32 %x) #0 {
 ; CHECK-NEXT:    pushq %rax
 ; CHECK-NEXT:    testl %edi, %edi
 ; CHECK-NEXT:    setne %al
+; CHECK-NEXT:  # implicit-def: %EDI
 ; CHECK-NEXT:    movb %al, %dil
 ; CHECK-NEXT:    andl $1, %edi
 ; CHECK-NEXT:    kmovw %edi, %k0