[ScalarizeMaskedMemIntrin] Bitcast the mask to the scalar domain and use scalar bit tests for the branches.

[craig.topper]

Summary

X86 at least is able to use movmsk or kmov to move the mask to the scalar
domain. Then we can just use test instructions to test individual bits.

This is more efficient than extracting each mask element
individually.

I special cased v1i1 to use the previous behavior. This avoids
poor type legalization of bitcast of v1i1 to i1.

I've skipped expandload/compressstore as I think we need to
handle constant masks for those better first.

Many tests end up with duplicate test instructions due to tail
duplication in the branch folding pass. But the same thing
happens when constructing similar code in C. So its not unique
to the scalarization.

Not sure if this lowering code will also be good for other targets,
but we're only testing X86 today.

[RKSimon]

I think this covers all of https://reviews.llvm.org/D65319 ?

[craig.topper]

In D65319#1604232, @RKSimon wrote:

I think this covers all of https://reviews.llvm.org/D65319 ?

Wrong link? That’s just a link to this review

[RKSimon]

In D65319#1604552, @craig.topper wrote:

In D65319#1604232, @RKSimon wrote:

I think this covers all of https://reviews.llvm.org/D65319 ?

Wrong link? That’s just a link to this review

I don't like Mondays - https://bugs.llvm.org/show_bug.cgi?id=39665 - which goes beyond masked memory ops so probably isn't covered by this patch along.

[RKSimon]

LGTM - I think this is the way forward for the scalarization of a lot of masked ops and I've already raised some similar bugs for arm/aarch64/ppc (PR41634, PR41635, PR41636) to better handle bitcasts of <X x i1> to iX for reductions - this patch should be able to make use of those upcoming improvements.

Please can you raise bugs for the expandload/compressstore improvements you mentioned and the duplication of test+branch instructions.

llvm/trunk/lib/CodeGen/ScalarizeMaskedMemIntrin.cpp

for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
       LoadInst *Load = Builder.CreateAlignedLoad(EltTy, Gep, AlignVal);
       VResult = Builder.CreateInsertElement(VResult, Load, Idx);
     }
     CI->replaceAllUsesWith(VResult);
     CI->eraseFromParent();
     return;
   }
 
+  // If the mask is not v1i1, use scalar bit test operations. This generates
+  // better results on X86 at least.
+  Value *SclrMask;
+  if (VectorWidth != 1) {
+    Type *SclrMaskTy = Builder.getIntNTy(VectorWidth);
+    SclrMask = Builder.CreateBitCast(Mask, SclrMaskTy, "scalar_mask");
+  }
+
   for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
     // Fill the "else" block, created in the previous iteration
     //
     //  %res.phi.else3 = phi <16 x i32> [ %11, %cond.load1 ], [ %res.phi.else, %else ]
-    //  %mask_1 = extractelement <16 x i1> %mask, i32 Idx
+    //  %mask_1 = and i16 %scalar_mask, i32 1 << Idx
+    //  %cond = icmp ne i16 %mask_1, 0
     //  br i1 %mask_1, label %cond.load, label %else
     //
-
-    Value *Predicate = Builder.CreateExtractElement(Mask, Idx);
+    Value *Predicate;
+    if (VectorWidth != 1) {
+      Value *Mask = Builder.getInt(APInt::getOneBitSet(VectorWidth, Idx));
+      Predicate = Builder.CreateICmpNE(Builder.CreateAnd(SclrMask, Mask),
+                                       Builder.getIntN(VectorWidth, 0));
+    } else {
+      Predicate = Builder.CreateExtractElement(Mask, Idx);
+    }
 
     // Create "cond" block
     //
     //  %EltAddr = getelementptr i32* %1, i32 0
     //  %Elt = load i32* %EltAddr
     //  VResult = insertelement <16 x i32> VResult, i32 %Elt, i32 Idx
     //
     BasicBlock *CondBlock = IfBlock->splitBasicBlock(InsertPt->getIterator(),
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
       Value *OneElt = Builder.CreateExtractElement(Src, Idx);
       Value *Gep = Builder.CreateConstInBoundsGEP1_32(EltTy, FirstEltPtr, Idx);
       Builder.CreateAlignedStore(OneElt, Gep, AlignVal);
     }
     CI->eraseFromParent();
     return;
   }
 
+  // If the mask is not v1i1, use scalar bit test operations. This generates
+  // better results on X86 at least.
+  Value *SclrMask;
+  if (VectorWidth != 1) {
+    Type *SclrMaskTy = Builder.getIntNTy(VectorWidth);
+    SclrMask = Builder.CreateBitCast(Mask, SclrMaskTy, "scalar_mask");
+  }
+
   for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
     // Fill the "else" block, created in the previous iteration
     //
-    //  %mask_1 = extractelement <16 x i1> %mask, i32 Idx
+    //  %mask_1 = and i16 %scalar_mask, i32 1 << Idx
+    //  %cond = icmp ne i16 %mask_1, 0
     //  br i1 %mask_1, label %cond.store, label %else
     //
-    Value *Predicate = Builder.CreateExtractElement(Mask, Idx);
+    Value *Predicate;
+    if (VectorWidth != 1) {
+      Value *Mask = Builder.getInt(APInt::getOneBitSet(VectorWidth, Idx));
+      Predicate = Builder.CreateICmpNE(Builder.CreateAnd(SclrMask, Mask),
+                                       Builder.getIntN(VectorWidth, 0));
+    } else {
+      Predicate = Builder.CreateExtractElement(Mask, Idx);
+    }
 
     // Create "cond" block
     //
     //  %OneElt = extractelement <16 x i32> %Src, i32 Idx
     //  %EltAddr = getelementptr i32* %1, i32 0
     //  %store i32 %OneElt, i32* %EltAddr
     //
     BasicBlock *CondBlock =
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
       VResult =
           Builder.CreateInsertElement(VResult, Load, Idx, "Res" + Twine(Idx));
     }
     CI->replaceAllUsesWith(VResult);
     CI->eraseFromParent();
     return;
   }
 
+  // If the mask is not v1i1, use scalar bit test operations. This generates
+  // better results on X86 at least.
+  Value *SclrMask;
+  if (VectorWidth != 1) {
+    Type *SclrMaskTy = Builder.getIntNTy(VectorWidth);
+    SclrMask = Builder.CreateBitCast(Mask, SclrMaskTy, "scalar_mask");
+  }
+
   for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
     // Fill the "else" block, created in the previous iteration
     //
-    //  %Mask1 = extractelement <16 x i1> %Mask, i32 1
+    //  %Mask1 = and i16 %scalar_mask, i32 1 << Idx
+    //  %cond = icmp ne i16 %mask_1, 0
     //  br i1 %Mask1, label %cond.load, label %else
     //
 
-    Value *Predicate =
-        Builder.CreateExtractElement(Mask, Idx, "Mask" + Twine(Idx));
+    Value *Predicate;
+    if (VectorWidth != 1) {
+      Value *Mask = Builder.getInt(APInt::getOneBitSet(VectorWidth, Idx));
+      Predicate = Builder.CreateICmpNE(Builder.CreateAnd(SclrMask, Mask),
+                                       Builder.getIntN(VectorWidth, 0));
+    } else {
+      Predicate = Builder.CreateExtractElement(Mask, Idx, "Mask" + Twine(Idx));
+    }
 
     // Create "cond" block
     //
     //  %EltAddr = getelementptr i32* %1, i32 0
     //  %Elt = load i32* %EltAddr
     //  VResult = insertelement <16 x i32> VResult, i32 %Elt, i32 Idx
     //
     BasicBlock *CondBlock = IfBlock->splitBasicBlock(InsertPt, "cond.load");
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
           Builder.CreateExtractElement(Src, Idx, "Elt" + Twine(Idx));
       Value *Ptr = Builder.CreateExtractElement(Ptrs, Idx, "Ptr" + Twine(Idx));
       Builder.CreateAlignedStore(OneElt, Ptr, AlignVal);
     }
     CI->eraseFromParent();
     return;
   }
 
+  // If the mask is not v1i1, use scalar bit test operations. This generates
+  // better results on X86 at least.
+  Value *SclrMask;
+  if (VectorWidth != 1) {
+    Type *SclrMaskTy = Builder.getIntNTy(VectorWidth);
+    SclrMask = Builder.CreateBitCast(Mask, SclrMaskTy, "scalar_mask");
+  }
+
   for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
     // Fill the "else" block, created in the previous iteration
     //
-    //  %Mask1 = extractelement <16 x i1> %Mask, i32 Idx
+    //  %Mask1 = and i16 %scalar_mask, i32 1 << Idx
+    //  %cond = icmp ne i16 %mask_1, 0
     //  br i1 %Mask1, label %cond.store, label %else
     //
-    Value *Predicate =
-        Builder.CreateExtractElement(Mask, Idx, "Mask" + Twine(Idx));
+    Value *Predicate;
+    if (VectorWidth != 1) {
+      Value *Mask = Builder.getInt(APInt::getOneBitSet(VectorWidth, Idx));
+      Predicate = Builder.CreateICmpNE(Builder.CreateAnd(SclrMask, Mask),
+                                       Builder.getIntN(VectorWidth, 0));
+    } else {
+      Predicate = Builder.CreateExtractElement(Mask, Idx, "Mask" + Twine(Idx));
+    }
 
     // Create "cond" block
     //
     //  %Elt1 = extractelement <16 x i32> %Src, i32 1
     //  %Ptr1 = extractelement <16 x i32*> %Ptrs, i32 1
     //  %store i32 %Elt1, i32* %Ptr1
     //
     BasicBlock *CondBlock = IfBlock->splitBasicBlock(InsertPt, "cond.store");

llvm/trunk/test/CodeGen/X86/avx2-masked-gather.ll

 ; X64-NEXT:    vpslld $31, %xmm0, %xmm0
 ; X64-NEXT:    vpgatherqd %xmm0, (,%xmm2), %xmm1
 ; X64-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v2i32:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %xmm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB0_2
-; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    movl (%rax), %eax
-; NOGATHER-NEXT:    vpinsrq $0, %rax, %xmm1, %xmm1
-; NOGATHER-NEXT:  .LBB0_2: # %else
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
+; NOGATHER-NEXT:    vpsllq $63, %xmm0, %xmm0
+; NOGATHER-NEXT:    vmovmskpd %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    jne .LBB0_1
+; NOGATHER-NEXT:  # %bb.2: # %else
+; NOGATHER-NEXT:    testb $2, %al
+; NOGATHER-NEXT:    jne .LBB0_3
+; NOGATHER-NEXT:  .LBB0_4: # %else2
+; NOGATHER-NEXT:    vmovdqa %xmm1, %xmm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB0_1: # %cond.load
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    movl (%rcx), %ecx
+; NOGATHER-NEXT:    vpinsrq $0, %rcx, %xmm1, %xmm1
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB0_4
-; NOGATHER-NEXT:  # %bb.3: # %cond.load1
+; NOGATHER-NEXT:  .LBB0_3: # %cond.load1
 ; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
 ; NOGATHER-NEXT:    movl (%rax), %eax
 ; NOGATHER-NEXT:    vpinsrq $1, %rax, %xmm1, %xmm1
-; NOGATHER-NEXT:  .LBB0_4: # %else2
 ; NOGATHER-NEXT:    vmovdqa %xmm1, %xmm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <2 x i32*>, <2 x i32*>* %ptr
   %res = call <2 x i32> @llvm.masked.gather.v2i32(<2 x i32*> %ld, i32 0, <2 x i1> %masks, <2 x i32> %passthro)
   ret <2 x i32> %res
 }
 

 ; X64-NEXT:    vpslld $31, %xmm0, %xmm0
 ; X64-NEXT:    vpgatherqd %xmm0, (,%xmm2), %xmm1
 ; X64-NEXT:    vmovdqa %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v2i32_concat:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %xmm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB1_2
-; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    movl (%rax), %eax
-; NOGATHER-NEXT:    vpinsrq $0, %rax, %xmm1, %xmm1
-; NOGATHER-NEXT:  .LBB1_2: # %else
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
+; NOGATHER-NEXT:    vpsllq $63, %xmm0, %xmm0
+; NOGATHER-NEXT:    vmovmskpd %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    jne .LBB1_1
+; NOGATHER-NEXT:  # %bb.2: # %else
+; NOGATHER-NEXT:    testb $2, %al
+; NOGATHER-NEXT:    jne .LBB1_3
+; NOGATHER-NEXT:  .LBB1_4: # %else2
+; NOGATHER-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB1_1: # %cond.load
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    movl (%rcx), %ecx
+; NOGATHER-NEXT:    vpinsrq $0, %rcx, %xmm1, %xmm1
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB1_4
-; NOGATHER-NEXT:  # %bb.3: # %cond.load1
+; NOGATHER-NEXT:  .LBB1_3: # %cond.load1
 ; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
 ; NOGATHER-NEXT:    movl (%rax), %eax
 ; NOGATHER-NEXT:    vpinsrq $1, %rax, %xmm1, %xmm1
-; NOGATHER-NEXT:  .LBB1_4: # %else2
 ; NOGATHER-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,2,2,3]
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <2 x i32*>, <2 x i32*>* %ptr
   %res = call <2 x i32> @llvm.masked.gather.v2i32(<2 x i32*> %ld, i32 0, <2 x i1> %masks, <2 x i32> %passthro)
   %res2 = shufflevector <2 x i32> %res, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   ret <4 x i32> %res2
 }

 ; X64-NEXT:    vpslld $31, %xmm0, %xmm0
 ; X64-NEXT:    vgatherqps %xmm0, (,%xmm2), %xmm1
 ; X64-NEXT:    vmovaps %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v2float:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %xmm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB2_2
-; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; NOGATHER-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1,2,3]
-; NOGATHER-NEXT:  .LBB2_2: # %else
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
+; NOGATHER-NEXT:    vpsllq $63, %xmm0, %xmm0
+; NOGATHER-NEXT:    vmovmskpd %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    jne .LBB2_1
+; NOGATHER-NEXT:  # %bb.2: # %else
+; NOGATHER-NEXT:    testb $2, %al
+; NOGATHER-NEXT:    jne .LBB2_3
+; NOGATHER-NEXT:  .LBB2_4: # %else2
+; NOGATHER-NEXT:    vmovaps %xmm1, %xmm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB2_1: # %cond.load
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; NOGATHER-NEXT:    vblendps {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB2_4
-; NOGATHER-NEXT:  # %bb.3: # %cond.load1
+; NOGATHER-NEXT:  .LBB2_3: # %cond.load1
 ; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
 ; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[2,3]
-; NOGATHER-NEXT:  .LBB2_4: # %else2
 ; NOGATHER-NEXT:    vmovaps %xmm1, %xmm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <2 x float*>, <2 x float*>* %ptr
   %res = call <2 x float> @llvm.masked.gather.v2float(<2 x float*> %ld, i32 0, <2 x i1> %masks, <2 x float> %passthro)
   ret <2 x float> %res
 }
 

 ; X64-NEXT:    vpslld $31, %xmm0, %xmm0
 ; X64-NEXT:    vgatherqps %xmm0, (,%xmm2), %xmm1
 ; X64-NEXT:    vmovaps %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v2float_concat:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %xmm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB3_2
-; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; NOGATHER-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1,2,3]
-; NOGATHER-NEXT:  .LBB3_2: # %else
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
+; NOGATHER-NEXT:    vpsllq $63, %xmm0, %xmm0
+; NOGATHER-NEXT:    vmovmskpd %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    jne .LBB3_1
+; NOGATHER-NEXT:  # %bb.2: # %else
+; NOGATHER-NEXT:    testb $2, %al
+; NOGATHER-NEXT:    jne .LBB3_3
+; NOGATHER-NEXT:  .LBB3_4: # %else2
+; NOGATHER-NEXT:    vmovaps %xmm1, %xmm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB3_1: # %cond.load
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; NOGATHER-NEXT:    vblendps {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB3_4
-; NOGATHER-NEXT:  # %bb.3: # %cond.load1
+; NOGATHER-NEXT:  .LBB3_3: # %cond.load1
 ; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
 ; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[2,3]
-; NOGATHER-NEXT:  .LBB3_4: # %else2
 ; NOGATHER-NEXT:    vmovaps %xmm1, %xmm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <2 x float*>, <2 x float*>* %ptr
   %res = call <2 x float> @llvm.masked.gather.v2float(<2 x float*> %ld, i32 0, <2 x i1> %masks, <2 x float> %passthro)
   %res2 = shufflevector <2 x float> %res, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   ret <4 x float> %res2
 }

 ; X64-NEXT:    vpslld $31, %xmm1, %xmm1
 ; X64-NEXT:    vpgatherqd %xmm1, (,%ymm0), %xmm2
 ; X64-NEXT:    vmovdqa %xmm2, %xmm0
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v4i32:
 ; NOGATHER:       # %bb.0: # %entry
-; NOGATHER-NEXT:    vpextrb $0, %xmm1, %eax
+; NOGATHER-NEXT:    vpslld $31, %xmm1, %xmm1
+; NOGATHER-NEXT:    vmovmskps %xmm1, %eax
 ; NOGATHER-NEXT:    testb $1, %al
 ; NOGATHER-NEXT:    je .LBB4_2
 ; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm0, %rax
-; NOGATHER-NEXT:    vpinsrd $0, (%rax), %xmm2, %xmm2
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vpinsrd $0, (%rcx), %xmm2, %xmm2
 ; NOGATHER-NEXT:  .LBB4_2: # %else
-; NOGATHER-NEXT:    vpextrb $4, %xmm1, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB4_4
 ; NOGATHER-NEXT:  # %bb.3: # %cond.load1
-; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
-; NOGATHER-NEXT:    vpinsrd $1, (%rax), %xmm2, %xmm2
+; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rcx
+; NOGATHER-NEXT:    vpinsrd $1, (%rcx), %xmm2, %xmm2
 ; NOGATHER-NEXT:  .LBB4_4: # %else2
-; NOGATHER-NEXT:    vpextrb $8, %xmm1, %eax
 ; NOGATHER-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB4_6
-; NOGATHER-NEXT:  # %bb.5: # %cond.load4
-; NOGATHER-NEXT:    vmovq %xmm0, %rax
-; NOGATHER-NEXT:    vpinsrd $2, (%rax), %xmm2, %xmm2
-; NOGATHER-NEXT:  .LBB4_6: # %else5
-; NOGATHER-NEXT:    vpextrb $12, %xmm1, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $4, %al
+; NOGATHER-NEXT:    jne .LBB4_5
+; NOGATHER-NEXT:  # %bb.6: # %else5
+; NOGATHER-NEXT:    testb $8, %al
+; NOGATHER-NEXT:    jne .LBB4_7
+; NOGATHER-NEXT:  .LBB4_8: # %else8
+; NOGATHER-NEXT:    vmovdqa %xmm2, %xmm0
+; NOGATHER-NEXT:    vzeroupper
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB4_5: # %cond.load4
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vpinsrd $2, (%rcx), %xmm2, %xmm2
+; NOGATHER-NEXT:    testb $8, %al
 ; NOGATHER-NEXT:    je .LBB4_8
-; NOGATHER-NEXT:  # %bb.7: # %cond.load7
+; NOGATHER-NEXT:  .LBB4_7: # %cond.load7
 ; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
 ; NOGATHER-NEXT:    vpinsrd $3, (%rax), %xmm2, %xmm2
-; NOGATHER-NEXT:  .LBB4_8: # %else8
 ; NOGATHER-NEXT:    vmovdqa %xmm2, %xmm0
 ; NOGATHER-NEXT:    vzeroupper
 ; NOGATHER-NEXT:    retq
 entry:
   %res = call <4 x i32> @llvm.masked.gather.v4i32(<4 x i32*> %ptrs, i32 0, <4 x i1> %masks, <4 x i32> %passthro)
   ret <4 x i32> %res
 }
 

 ; X64-NEXT:    vpslld $31, %xmm1, %xmm1
 ; X64-NEXT:    vgatherqps %xmm1, (,%ymm0), %xmm2
 ; X64-NEXT:    vmovaps %xmm2, %xmm0
 ; X64-NEXT:    vzeroupper
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v4float:
 ; NOGATHER:       # %bb.0: # %entry
-; NOGATHER-NEXT:    vpextrb $0, %xmm1, %eax
+; NOGATHER-NEXT:    vpslld $31, %xmm1, %xmm1
+; NOGATHER-NEXT:    vmovmskps %xmm1, %eax
 ; NOGATHER-NEXT:    testb $1, %al
 ; NOGATHER-NEXT:    je .LBB5_2
 ; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm0, %rax
-; NOGATHER-NEXT:    vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; NOGATHER-NEXT:    vblendps {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3]
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; NOGATHER-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
 ; NOGATHER-NEXT:  .LBB5_2: # %else
-; NOGATHER-NEXT:    vpextrb $4, %xmm1, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB5_4
 ; NOGATHER-NEXT:  # %bb.3: # %cond.load1
-; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
+; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rcx
 ; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
 ; NOGATHER-NEXT:  .LBB5_4: # %else2
-; NOGATHER-NEXT:    vpextrb $8, %xmm1, %eax
 ; NOGATHER-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB5_6
-; NOGATHER-NEXT:  # %bb.5: # %cond.load4
-; NOGATHER-NEXT:    vmovq %xmm0, %rax
+; NOGATHER-NEXT:    testb $4, %al
+; NOGATHER-NEXT:    jne .LBB5_5
+; NOGATHER-NEXT:  # %bb.6: # %else5
+; NOGATHER-NEXT:    testb $8, %al
+; NOGATHER-NEXT:    jne .LBB5_7
+; NOGATHER-NEXT:  .LBB5_8: # %else8
+; NOGATHER-NEXT:    vmovaps %xmm2, %xmm0
+; NOGATHER-NEXT:    vzeroupper
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB5_5: # %cond.load4
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
 ; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; NOGATHER-NEXT:  .LBB5_6: # %else5
-; NOGATHER-NEXT:    vpextrb $12, %xmm1, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $8, %al
 ; NOGATHER-NEXT:    je .LBB5_8
-; NOGATHER-NEXT:  # %bb.7: # %cond.load7
+; NOGATHER-NEXT:  .LBB5_7: # %cond.load7
 ; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
 ; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; NOGATHER-NEXT:  .LBB5_8: # %else8
 ; NOGATHER-NEXT:    vmovaps %xmm2, %xmm0
 ; NOGATHER-NEXT:    vzeroupper
 ; NOGATHER-NEXT:    retq
 entry:
   %res = call <4 x float> @llvm.masked.gather.v4float(<4 x float*> %ptrs, i32 0, <4 x i1> %masks, <4 x float> %passthro)
   ret <4 x float> %res
 }
 

 ; X64-NEXT:    vpgatherqd %xmm0, (,%ymm2), %xmm1
 ; X64-NEXT:    vinserti128 $1, %xmm4, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v8i32:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %ymm3
 ; NOGATHER-NEXT:    vmovdqa 32(%rdi), %ymm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
+; NOGATHER-NEXT:    vpsllw $15, %xmm0, %xmm0
+; NOGATHER-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
+; NOGATHER-NEXT:    vpmovmskb %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
 ; NOGATHER-NEXT:    je .LBB6_2
 ; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm3, %rax
-; NOGATHER-NEXT:    vpinsrd $0, (%rax), %xmm1, %xmm4
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    vmovq %xmm3, %rcx
+; NOGATHER-NEXT:    vpinsrd $0, (%rcx), %xmm1, %xmm0
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; NOGATHER-NEXT:  .LBB6_2: # %else
-; NOGATHER-NEXT:    vpextrb $2, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB6_4
 ; NOGATHER-NEXT:  # %bb.3: # %cond.load1
-; NOGATHER-NEXT:    vpextrq $1, %xmm3, %rax
-; NOGATHER-NEXT:    vpinsrd $1, (%rax), %xmm1, %xmm4
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    vpextrq $1, %xmm3, %rcx
+; NOGATHER-NEXT:    vpinsrd $1, (%rcx), %xmm1, %xmm0
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; NOGATHER-NEXT:  .LBB6_4: # %else2
-; NOGATHER-NEXT:    vpextrb $4, %xmm0, %eax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm3, %xmm3
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB6_6
-; NOGATHER-NEXT:  # %bb.5: # %cond.load4
-; NOGATHER-NEXT:    vmovq %xmm3, %rax
-; NOGATHER-NEXT:    vpinsrd $2, (%rax), %xmm1, %xmm4
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7]
-; NOGATHER-NEXT:  .LBB6_6: # %else5
-; NOGATHER-NEXT:    vpextrb $6, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB6_8
-; NOGATHER-NEXT:  # %bb.7: # %cond.load7
-; NOGATHER-NEXT:    vpextrq $1, %xmm3, %rax
-; NOGATHER-NEXT:    vpinsrd $3, (%rax), %xmm1, %xmm3
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    vextractf128 $1, %ymm3, %xmm0
+; NOGATHER-NEXT:    testb $4, %al
+; NOGATHER-NEXT:    jne .LBB6_5
+; NOGATHER-NEXT:  # %bb.6: # %else5
+; NOGATHER-NEXT:    testb $8, %al
+; NOGATHER-NEXT:    jne .LBB6_7
 ; NOGATHER-NEXT:  .LBB6_8: # %else8
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB6_10
-; NOGATHER-NEXT:  # %bb.9: # %cond.load10
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; NOGATHER-NEXT:    vpinsrd $0, (%rax), %xmm3, %xmm3
-; NOGATHER-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; NOGATHER-NEXT:    testb $16, %al
+; NOGATHER-NEXT:    jne .LBB6_9
 ; NOGATHER-NEXT:  .LBB6_10: # %else11
-; NOGATHER-NEXT:    vpextrb $10, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $32, %al
 ; NOGATHER-NEXT:    je .LBB6_12
-; NOGATHER-NEXT:  # %bb.11: # %cond.load13
-; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; NOGATHER-NEXT:    vpinsrd $1, (%rax), %xmm3, %xmm3
-; NOGATHER-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; NOGATHER-NEXT:  .LBB6_11: # %cond.load13
+; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rcx
+; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm0
+; NOGATHER-NEXT:    vpinsrd $1, (%rcx), %xmm0, %xmm0
+; NOGATHER-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
 ; NOGATHER-NEXT:  .LBB6_12: # %else14
-; NOGATHER-NEXT:    vpextrb $12, %xmm0, %eax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm2, %xmm2
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB6_14
-; NOGATHER-NEXT:  # %bb.13: # %cond.load16
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; NOGATHER-NEXT:    vpinsrd $2, (%rax), %xmm3, %xmm3
-; NOGATHER-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; NOGATHER-NEXT:  .LBB6_14: # %else17
-; NOGATHER-NEXT:    vpextrb $14, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    vextractf128 $1, %ymm2, %xmm0
+; NOGATHER-NEXT:    testb $64, %al
+; NOGATHER-NEXT:    jne .LBB6_13
+; NOGATHER-NEXT:  # %bb.14: # %else17
+; NOGATHER-NEXT:    testb $-128, %al
+; NOGATHER-NEXT:    jne .LBB6_15
+; NOGATHER-NEXT:  .LBB6_16: # %else20
+; NOGATHER-NEXT:    vmovaps %ymm1, %ymm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB6_5: # %cond.load4
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vpinsrd $2, (%rcx), %xmm1, %xmm3
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    testb $8, %al
+; NOGATHER-NEXT:    je .LBB6_8
+; NOGATHER-NEXT:  .LBB6_7: # %cond.load7
+; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rcx
+; NOGATHER-NEXT:    vpinsrd $3, (%rcx), %xmm1, %xmm0
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    testb $16, %al
+; NOGATHER-NEXT:    je .LBB6_10
+; NOGATHER-NEXT:  .LBB6_9: # %cond.load10
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm0
+; NOGATHER-NEXT:    vpinsrd $0, (%rcx), %xmm0, %xmm0
+; NOGATHER-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
+; NOGATHER-NEXT:    testb $32, %al
+; NOGATHER-NEXT:    jne .LBB6_11
+; NOGATHER-NEXT:    jmp .LBB6_12
+; NOGATHER-NEXT:  .LBB6_13: # %cond.load16
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; NOGATHER-NEXT:    vpinsrd $2, (%rcx), %xmm2, %xmm2
+; NOGATHER-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; NOGATHER-NEXT:    testb $-128, %al
 ; NOGATHER-NEXT:    je .LBB6_16
-; NOGATHER-NEXT:  # %bb.15: # %cond.load19
-; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
+; NOGATHER-NEXT:  .LBB6_15: # %cond.load19
+; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
 ; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm0
 ; NOGATHER-NEXT:    vpinsrd $3, (%rax), %xmm0, %xmm0
 ; NOGATHER-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
-; NOGATHER-NEXT:  .LBB6_16: # %else20
 ; NOGATHER-NEXT:    vmovaps %ymm1, %ymm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <8 x i32*>, <8 x i32*>* %ptr
   %res = call <8 x i32> @llvm.masked.gather.v8i32(<8 x i32*> %ld, i32 0, <8 x i1> %masks, <8 x i32> %passthro)
   ret <8 x i32> %res
 }
 

 ; X64-NEXT:    vgatherqps %xmm0, (,%ymm2), %xmm1
 ; X64-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v8float:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %ymm3
 ; NOGATHER-NEXT:    vmovdqa 32(%rdi), %ymm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
+; NOGATHER-NEXT:    vpsllw $15, %xmm0, %xmm0
+; NOGATHER-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
+; NOGATHER-NEXT:    vpmovmskb %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
 ; NOGATHER-NEXT:    je .LBB7_2
 ; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm3, %rax
-; NOGATHER-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0],ymm1[1,2,3,4,5,6,7]
+; NOGATHER-NEXT:    vmovq %xmm3, %rcx
+; NOGATHER-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7]
 ; NOGATHER-NEXT:  .LBB7_2: # %else
-; NOGATHER-NEXT:    vpextrb $2, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB7_4
 ; NOGATHER-NEXT:  # %bb.3: # %cond.load1
-; NOGATHER-NEXT:    vpextrq $1, %xmm3, %rax
-; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm4 = xmm1[0],mem[0],xmm1[2,3]
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    vpextrq $1, %xmm3, %rcx
+; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],mem[0],xmm1[2,3]
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; NOGATHER-NEXT:  .LBB7_4: # %else2
-; NOGATHER-NEXT:    vpextrb $4, %xmm0, %eax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm3, %xmm3
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB7_6
-; NOGATHER-NEXT:  # %bb.5: # %cond.load4
-; NOGATHER-NEXT:    vmovq %xmm3, %rax
-; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm4 = xmm1[0,1],mem[0],xmm1[3]
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7]
-; NOGATHER-NEXT:  .LBB7_6: # %else5
-; NOGATHER-NEXT:    vpextrb $6, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB7_8
-; NOGATHER-NEXT:  # %bb.7: # %cond.load7
-; NOGATHER-NEXT:    vpextrq $1, %xmm3, %rax
-; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm3 = xmm1[0,1,2],mem[0]
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    vextractf128 $1, %ymm3, %xmm0
+; NOGATHER-NEXT:    testb $4, %al
+; NOGATHER-NEXT:    jne .LBB7_5
+; NOGATHER-NEXT:  # %bb.6: # %else5
+; NOGATHER-NEXT:    testb $8, %al
+; NOGATHER-NEXT:    jne .LBB7_7
 ; NOGATHER-NEXT:  .LBB7_8: # %else8
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB7_10
-; NOGATHER-NEXT:  # %bb.9: # %cond.load10
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; NOGATHER-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; NOGATHER-NEXT:    vblendps {{.*#+}} xmm3 = xmm4[0],xmm3[1,2,3]
-; NOGATHER-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; NOGATHER-NEXT:    testb $16, %al
+; NOGATHER-NEXT:    jne .LBB7_9
 ; NOGATHER-NEXT:  .LBB7_10: # %else11
-; NOGATHER-NEXT:    vpextrb $10, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $32, %al
 ; NOGATHER-NEXT:    je .LBB7_12
-; NOGATHER-NEXT:  # %bb.11: # %cond.load13
-; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm3 = xmm3[0],mem[0],xmm3[2,3]
-; NOGATHER-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; NOGATHER-NEXT:  .LBB7_11: # %cond.load13
+; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rcx
+; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm0
+; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
+; NOGATHER-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
 ; NOGATHER-NEXT:  .LBB7_12: # %else14
-; NOGATHER-NEXT:    vpextrb $12, %xmm0, %eax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm2, %xmm2
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB7_14
-; NOGATHER-NEXT:  # %bb.13: # %cond.load16
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3]
-; NOGATHER-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; NOGATHER-NEXT:  .LBB7_14: # %else17
-; NOGATHER-NEXT:    vpextrb $14, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    vextractf128 $1, %ymm2, %xmm0
+; NOGATHER-NEXT:    testb $64, %al
+; NOGATHER-NEXT:    jne .LBB7_13
+; NOGATHER-NEXT:  # %bb.14: # %else17
+; NOGATHER-NEXT:    testb $-128, %al
+; NOGATHER-NEXT:    jne .LBB7_15
+; NOGATHER-NEXT:  .LBB7_16: # %else20
+; NOGATHER-NEXT:    vmovaps %ymm1, %ymm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB7_5: # %cond.load4
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm3 = xmm1[0,1],mem[0],xmm1[3]
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    testb $8, %al
+; NOGATHER-NEXT:    je .LBB7_8
+; NOGATHER-NEXT:  .LBB7_7: # %cond.load7
+; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rcx
+; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],mem[0]
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    testb $16, %al
+; NOGATHER-NEXT:    je .LBB7_10
+; NOGATHER-NEXT:  .LBB7_9: # %cond.load10
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm0
+; NOGATHER-NEXT:    vmovd {{.*#+}} xmm3 = mem[0],zero,zero,zero
+; NOGATHER-NEXT:    vpblendw {{.*#+}} xmm0 = xmm3[0,1],xmm0[2,3,4,5,6,7]
+; NOGATHER-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
+; NOGATHER-NEXT:    testb $32, %al
+; NOGATHER-NEXT:    jne .LBB7_11
+; NOGATHER-NEXT:    jmp .LBB7_12
+; NOGATHER-NEXT:  .LBB7_13: # %cond.load16
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
+; NOGATHER-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; NOGATHER-NEXT:    testb $-128, %al
 ; NOGATHER-NEXT:    je .LBB7_16
-; NOGATHER-NEXT:  # %bb.15: # %cond.load19
-; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
+; NOGATHER-NEXT:  .LBB7_15: # %cond.load19
+; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
 ; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm0
 ; NOGATHER-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0]
 ; NOGATHER-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
-; NOGATHER-NEXT:  .LBB7_16: # %else20
 ; NOGATHER-NEXT:    vmovaps %ymm1, %ymm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <8 x float*>, <8 x float*>* %ptr
   %res = call <8 x float> @llvm.masked.gather.v8float(<8 x float*> %ld, i32 0, <8 x i1> %masks, <8 x float> %passthro)
   ret <8 x float> %res
 }
 

 ; X64-NEXT:    vmovdqa (%rdi), %ymm2
 ; X64-NEXT:    vpgatherqq %ymm0, (,%ymm2), %ymm1
 ; X64-NEXT:    vmovdqa %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v4i64:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %ymm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
+; NOGATHER-NEXT:    vpslld $31, %xmm0, %xmm0
+; NOGATHER-NEXT:    vmovmskps %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
 ; NOGATHER-NEXT:    je .LBB8_2
 ; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vpinsrq $0, (%rax), %xmm1, %xmm3
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    vpinsrq $0, (%rcx), %xmm1, %xmm0
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; NOGATHER-NEXT:  .LBB8_2: # %else
-; NOGATHER-NEXT:    vpextrb $4, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB8_4
 ; NOGATHER-NEXT:  # %bb.3: # %cond.load1
-; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
-; NOGATHER-NEXT:    vpinsrq $1, (%rax), %xmm1, %xmm3
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rcx
+; NOGATHER-NEXT:    vpinsrq $1, (%rcx), %xmm1, %xmm0
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; NOGATHER-NEXT:  .LBB8_4: # %else2
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm2, %xmm2
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB8_6
-; NOGATHER-NEXT:  # %bb.5: # %cond.load4
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; NOGATHER-NEXT:    vpinsrq $0, (%rax), %xmm3, %xmm3
-; NOGATHER-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; NOGATHER-NEXT:  .LBB8_6: # %else5
-; NOGATHER-NEXT:    vpextrb $12, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    vextractf128 $1, %ymm2, %xmm0
+; NOGATHER-NEXT:    testb $4, %al
+; NOGATHER-NEXT:    jne .LBB8_5
+; NOGATHER-NEXT:  # %bb.6: # %else5
+; NOGATHER-NEXT:    testb $8, %al
+; NOGATHER-NEXT:    jne .LBB8_7
+; NOGATHER-NEXT:  .LBB8_8: # %else8
+; NOGATHER-NEXT:    vmovaps %ymm1, %ymm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB8_5: # %cond.load4
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; NOGATHER-NEXT:    vpinsrq $0, (%rcx), %xmm2, %xmm2
+; NOGATHER-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; NOGATHER-NEXT:    testb $8, %al
 ; NOGATHER-NEXT:    je .LBB8_8
-; NOGATHER-NEXT:  # %bb.7: # %cond.load7
-; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
+; NOGATHER-NEXT:  .LBB8_7: # %cond.load7
+; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
 ; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm0
 ; NOGATHER-NEXT:    vpinsrq $1, (%rax), %xmm0, %xmm0
 ; NOGATHER-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
-; NOGATHER-NEXT:  .LBB8_8: # %else8
 ; NOGATHER-NEXT:    vmovaps %ymm1, %ymm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <4 x i64*>, <4 x i64*>* %ptr
   %res = call <4 x i64> @llvm.masked.gather.v4i64(<4 x i64*> %ld, i32 0, <4 x i1> %masks, <4 x i64> %passthro)
   ret <4 x i64> %res
 }
 

 ; X64-NEXT:    vmovapd (%rdi), %ymm2
 ; X64-NEXT:    vgatherqpd %ymm0, (,%ymm2), %ymm1
 ; X64-NEXT:    vmovapd %ymm1, %ymm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v4double:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %ymm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
+; NOGATHER-NEXT:    vpslld $31, %xmm0, %xmm0
+; NOGATHER-NEXT:    vmovmskps %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
 ; NOGATHER-NEXT:    je .LBB9_2
 ; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1],ymm1[2,3,4,5,6,7]
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1],ymm1[2,3,4,5,6,7]
 ; NOGATHER-NEXT:  .LBB9_2: # %else
-; NOGATHER-NEXT:    vpextrb $4, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB9_4
 ; NOGATHER-NEXT:  # %bb.3: # %cond.load1
-; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
-; NOGATHER-NEXT:    vmovhps {{.*#+}} xmm3 = xmm1[0,1],mem[0,1]
-; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rcx
+; NOGATHER-NEXT:    vmovhps {{.*#+}} xmm0 = xmm1[0,1],mem[0,1]
+; NOGATHER-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; NOGATHER-NEXT:  .LBB9_4: # %else2
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm2, %xmm2
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB9_6
-; NOGATHER-NEXT:  # %bb.5: # %cond.load4
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; NOGATHER-NEXT:    vmovlps {{.*#+}} xmm3 = mem[0,1],xmm3[2,3]
-; NOGATHER-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; NOGATHER-NEXT:  .LBB9_6: # %else5
-; NOGATHER-NEXT:    vpextrb $12, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    vextractf128 $1, %ymm2, %xmm0
+; NOGATHER-NEXT:    testb $4, %al
+; NOGATHER-NEXT:    jne .LBB9_5
+; NOGATHER-NEXT:  # %bb.6: # %else5
+; NOGATHER-NEXT:    testb $8, %al
+; NOGATHER-NEXT:    jne .LBB9_7
+; NOGATHER-NEXT:  .LBB9_8: # %else8
+; NOGATHER-NEXT:    vmovaps %ymm1, %ymm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB9_5: # %cond.load4
+; NOGATHER-NEXT:    vmovq %xmm0, %rcx
+; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; NOGATHER-NEXT:    vmovlps {{.*#+}} xmm2 = mem[0,1],xmm2[2,3]
+; NOGATHER-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; NOGATHER-NEXT:    testb $8, %al
 ; NOGATHER-NEXT:    je .LBB9_8
-; NOGATHER-NEXT:  # %bb.7: # %cond.load7
-; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
+; NOGATHER-NEXT:  .LBB9_7: # %cond.load7
+; NOGATHER-NEXT:    vpextrq $1, %xmm0, %rax
 ; NOGATHER-NEXT:    vextractf128 $1, %ymm1, %xmm0
 ; NOGATHER-NEXT:    vmovhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
 ; NOGATHER-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
-; NOGATHER-NEXT:  .LBB9_8: # %else8
 ; NOGATHER-NEXT:    vmovaps %ymm1, %ymm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <4 x double*>, <4 x double*>* %ptr
   %res = call <4 x double> @llvm.masked.gather.v4double(<4 x double*> %ld, i32 0, <4 x i1> %masks, <4 x double> %passthro)
   ret <4 x double> %res
 }
 

 ; X64-NEXT:    vmovdqa (%rdi), %xmm2
 ; X64-NEXT:    vpgatherqq %xmm0, (,%xmm2), %xmm1
 ; X64-NEXT:    vmovdqa %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v2i64:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %xmm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB10_2
-; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
-; NOGATHER-NEXT:    vpinsrq $0, (%rax), %xmm1, %xmm1
-; NOGATHER-NEXT:  .LBB10_2: # %else
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
+; NOGATHER-NEXT:    vpsllq $63, %xmm0, %xmm0
+; NOGATHER-NEXT:    vmovmskpd %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    jne .LBB10_1
+; NOGATHER-NEXT:  # %bb.2: # %else
+; NOGATHER-NEXT:    testb $2, %al
+; NOGATHER-NEXT:    jne .LBB10_3
+; NOGATHER-NEXT:  .LBB10_4: # %else2
+; NOGATHER-NEXT:    vmovdqa %xmm1, %xmm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB10_1: # %cond.load
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
+; NOGATHER-NEXT:    vpinsrq $0, (%rcx), %xmm1, %xmm1
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB10_4
-; NOGATHER-NEXT:  # %bb.3: # %cond.load1
+; NOGATHER-NEXT:  .LBB10_3: # %cond.load1
 ; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
 ; NOGATHER-NEXT:    vpinsrq $1, (%rax), %xmm1, %xmm1
-; NOGATHER-NEXT:  .LBB10_4: # %else2
 ; NOGATHER-NEXT:    vmovdqa %xmm1, %xmm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <2 x i64*>, <2 x i64*>* %ptr
   %res = call <2 x i64> @llvm.masked.gather.v2i64(<2 x i64*> %ld, i32 0, <2 x i1> %masks, <2 x i64> %passthro)
   ret <2 x i64> %res
 }
 

 ; X64-NEXT:    vmovapd (%rdi), %xmm2
 ; X64-NEXT:    vgatherqpd %xmm0, (,%xmm2), %xmm1
 ; X64-NEXT:    vmovapd %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; NOGATHER-LABEL: masked_gather_v2double:
 ; NOGATHER:       # %bb.0: # %entry
 ; NOGATHER-NEXT:    vmovdqa (%rdi), %xmm2
-; NOGATHER-NEXT:    vpextrb $0, %xmm0, %eax
+; NOGATHER-NEXT:    vpsllq $63, %xmm0, %xmm0
+; NOGATHER-NEXT:    vmovmskpd %xmm0, %eax
 ; NOGATHER-NEXT:    testb $1, %al
-; NOGATHER-NEXT:    je .LBB11_2
-; NOGATHER-NEXT:  # %bb.1: # %cond.load
-; NOGATHER-NEXT:    vmovq %xmm2, %rax
+; NOGATHER-NEXT:    jne .LBB11_1
+; NOGATHER-NEXT:  # %bb.2: # %else
+; NOGATHER-NEXT:    testb $2, %al
+; NOGATHER-NEXT:    jne .LBB11_3
+; NOGATHER-NEXT:  .LBB11_4: # %else2
+; NOGATHER-NEXT:    vmovaps %xmm1, %xmm0
+; NOGATHER-NEXT:    retq
+; NOGATHER-NEXT:  .LBB11_1: # %cond.load
+; NOGATHER-NEXT:    vmovq %xmm2, %rcx
 ; NOGATHER-NEXT:    vmovlps {{.*#+}} xmm1 = mem[0,1],xmm1[2,3]
-; NOGATHER-NEXT:  .LBB11_2: # %else
-; NOGATHER-NEXT:    vpextrb $8, %xmm0, %eax
-; NOGATHER-NEXT:    testb $1, %al
+; NOGATHER-NEXT:    testb $2, %al
 ; NOGATHER-NEXT:    je .LBB11_4
-; NOGATHER-NEXT:  # %bb.3: # %cond.load1
+; NOGATHER-NEXT:  .LBB11_3: # %cond.load1
 ; NOGATHER-NEXT:    vpextrq $1, %xmm2, %rax
 ; NOGATHER-NEXT:    vmovhps {{.*#+}} xmm1 = xmm1[0,1],mem[0,1]
-; NOGATHER-NEXT:  .LBB11_4: # %else2
 ; NOGATHER-NEXT:    vmovaps %xmm1, %xmm0
 ; NOGATHER-NEXT:    retq
 entry:
   %ld  = load <2 x double*>, <2 x double*>* %ptr
   %res = call <2 x double> @llvm.masked.gather.v2double(<2 x double*> %ld, i32 0, <2 x i1> %masks, <2 x double> %passthro)
   ret <2 x double> %res
 }
 

llvm/trunk/test/CodeGen/X86/avx512-masked_memop-16-8.ll

 declare void @llvm.masked.store.v32i16.p0v32i16(<32 x i16>, <32 x i16>*, i32, <32 x i1>)
 
 ; Make sure we scalarize masked loads of f16.
 define <16 x half> @test_mask_load_16xf16(<16 x i1> %mask, <16 x half>* %addr, <16 x half> %val) {
 ; CHECK-LABEL: test_mask_load_16xf16:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    movq %rdi, %rax
 ; CHECK-NEXT:    vpsllw $7, %xmm0, %xmm0
-; CHECK-NEXT:    vpmovb2m %xmm0, %k0
-; CHECK-NEXT:    kmovd %k0, %ecx
+; CHECK-NEXT:    vpmovmskb %xmm0, %ecx
 ; CHECK-NEXT:    testb $1, %cl
 ; CHECK-NEXT:    je LBB12_1
 ; CHECK-NEXT:  ## %bb.2: ## %cond.load
-; CHECK-NEXT:    movswl (%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
+; CHECK-NEXT:    movswl (%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
 ; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm8
 ; CHECK-NEXT:    jmp LBB12_3
 ; CHECK-NEXT:  LBB12_1:
 ; CHECK-NEXT:    vxorps %xmm8, %xmm8, %xmm8
 ; CHECK-NEXT:  LBB12_3: ## %else
 ; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
 ; CHECK-NEXT:    vxorps %xmm9, %xmm9, %xmm9
-; CHECK-NEXT:    kshiftrw $1, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
+; CHECK-NEXT:    testb $2, %cl
 ; CHECK-NEXT:    je LBB12_4
 ; CHECK-NEXT:  ## %bb.5: ## %cond.load1
-; CHECK-NEXT:    movswl 2(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
+; CHECK-NEXT:    movswl 2(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm1
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm7
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm6
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm5
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm4
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm3
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm16
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm15
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm14
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm13
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm12
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm11
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm10
 ; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm2
-; CHECK-NEXT:    jmp LBB12_6
+; CHECK-NEXT:    testb $4, %cl
+; CHECK-NEXT:    jne LBB12_7
+; CHECK-NEXT:    jmp LBB12_8
 ; CHECK-NEXT:  LBB12_4:
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm1
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm7
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm6
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm5
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm4
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm3
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm16
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm15
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm14
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm13
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm12
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm11
 ; CHECK-NEXT:    vmovaps %xmm2, %xmm10
-; CHECK-NEXT:  LBB12_6: ## %else2
-; CHECK-NEXT:    kshiftrw $2, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
+; CHECK-NEXT:    testb $4, %cl
 ; CHECK-NEXT:    je LBB12_8
-; CHECK-NEXT:  ## %bb.7: ## %cond.load4
-; CHECK-NEXT:    movswl 4(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
+; CHECK-NEXT:  LBB12_7: ## %cond.load4
+; CHECK-NEXT:    movswl 4(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
 ; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm1
 ; CHECK-NEXT:  LBB12_8: ## %else5
-; CHECK-NEXT:    kshiftrw $3, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_10
-; CHECK-NEXT:  ## %bb.9: ## %cond.load7
-; CHECK-NEXT:    movswl 6(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm7
-; CHECK-NEXT:  LBB12_10: ## %else8
-; CHECK-NEXT:    kshiftrw $4, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_12
-; CHECK-NEXT:  ## %bb.11: ## %cond.load10
-; CHECK-NEXT:    movswl 8(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm6
+; CHECK-NEXT:    testb $8, %cl
+; CHECK-NEXT:    jne LBB12_9
+; CHECK-NEXT:  ## %bb.10: ## %else8
+; CHECK-NEXT:    testb $16, %cl
+; CHECK-NEXT:    jne LBB12_11
 ; CHECK-NEXT:  LBB12_12: ## %else11
-; CHECK-NEXT:    kshiftrw $5, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_14
-; CHECK-NEXT:  ## %bb.13: ## %cond.load13
-; CHECK-NEXT:    movswl 10(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm5
+; CHECK-NEXT:    testb $32, %cl
+; CHECK-NEXT:    jne LBB12_13
 ; CHECK-NEXT:  LBB12_14: ## %else14
-; CHECK-NEXT:    kshiftrw $6, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_16
-; CHECK-NEXT:  ## %bb.15: ## %cond.load16
-; CHECK-NEXT:    movswl 12(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm4
+; CHECK-NEXT:    testb $64, %cl
+; CHECK-NEXT:    jne LBB12_15
 ; CHECK-NEXT:  LBB12_16: ## %else17
-; CHECK-NEXT:    kshiftrw $7, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_18
-; CHECK-NEXT:  ## %bb.17: ## %cond.load19
-; CHECK-NEXT:    movswl 14(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm3
+; CHECK-NEXT:    testb $-128, %cl
+; CHECK-NEXT:    jne LBB12_17
 ; CHECK-NEXT:  LBB12_18: ## %else20
-; CHECK-NEXT:    kshiftrw $8, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_20
-; CHECK-NEXT:  ## %bb.19: ## %cond.load22
-; CHECK-NEXT:    movswl 16(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm16
+; CHECK-NEXT:    testl $256, %ecx ## imm = 0x100
+; CHECK-NEXT:    jne LBB12_19
 ; CHECK-NEXT:  LBB12_20: ## %else23
-; CHECK-NEXT:    kshiftrw $9, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_22
-; CHECK-NEXT:  ## %bb.21: ## %cond.load25
-; CHECK-NEXT:    movswl 18(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm15
+; CHECK-NEXT:    testl $512, %ecx ## imm = 0x200
+; CHECK-NEXT:    jne LBB12_21
 ; CHECK-NEXT:  LBB12_22: ## %else26
-; CHECK-NEXT:    kshiftrw $10, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_24
-; CHECK-NEXT:  ## %bb.23: ## %cond.load28
-; CHECK-NEXT:    movswl 20(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm14
+; CHECK-NEXT:    testl $1024, %ecx ## imm = 0x400
+; CHECK-NEXT:    jne LBB12_23
 ; CHECK-NEXT:  LBB12_24: ## %else29
-; CHECK-NEXT:    kshiftrw $11, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_26
-; CHECK-NEXT:  ## %bb.25: ## %cond.load31
-; CHECK-NEXT:    movswl 22(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm13
+; CHECK-NEXT:    testl $2048, %ecx ## imm = 0x800
+; CHECK-NEXT:    jne LBB12_25
 ; CHECK-NEXT:  LBB12_26: ## %else32
-; CHECK-NEXT:    kshiftrw $12, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_28
-; CHECK-NEXT:  ## %bb.27: ## %cond.load34
-; CHECK-NEXT:    movswl 24(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm12
+; CHECK-NEXT:    testl $4096, %ecx ## imm = 0x1000
+; CHECK-NEXT:    jne LBB12_27
 ; CHECK-NEXT:  LBB12_28: ## %else35
-; CHECK-NEXT:    kshiftrw $13, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_30
-; CHECK-NEXT:  ## %bb.29: ## %cond.load37
-; CHECK-NEXT:    movswl 26(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm11
+; CHECK-NEXT:    testl $8192, %ecx ## imm = 0x2000
+; CHECK-NEXT:    jne LBB12_29
 ; CHECK-NEXT:  LBB12_30: ## %else38
-; CHECK-NEXT:    kshiftrw $14, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %ecx
-; CHECK-NEXT:    testb $1, %cl
-; CHECK-NEXT:    je LBB12_32
-; CHECK-NEXT:  ## %bb.31: ## %cond.load40
-; CHECK-NEXT:    movswl 28(%rsi), %ecx
-; CHECK-NEXT:    vmovd %ecx, %xmm0
-; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm10
+; CHECK-NEXT:    testl $16384, %ecx ## imm = 0x4000
+; CHECK-NEXT:    jne LBB12_31
 ; CHECK-NEXT:  LBB12_32: ## %else41
-; CHECK-NEXT:    kshiftrw $15, %k0, %k0
-; CHECK-NEXT:    kmovd %k0, %ecx
-; CHECK-NEXT:    testb $1, %cl
+; CHECK-NEXT:    testl $32768, %ecx ## imm = 0x8000
 ; CHECK-NEXT:    je LBB12_34
-; CHECK-NEXT:  ## %bb.33: ## %cond.load43
+; CHECK-NEXT:  LBB12_33: ## %cond.load43
 ; CHECK-NEXT:    movswl 30(%rsi), %ecx
 ; CHECK-NEXT:    vmovd %ecx, %xmm0
 ; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm9
 ; CHECK-NEXT:  LBB12_34: ## %else44
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm8, %xmm0
 ; CHECK-NEXT:    vmovd %xmm0, %ecx
 ; CHECK-NEXT:    movw %cx, (%rax)
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm2, %xmm0

 ; CHECK-NEXT:    movw %cx, 26(%rax)
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm10, %xmm0
 ; CHECK-NEXT:    vmovd %xmm0, %ecx
 ; CHECK-NEXT:    movw %cx, 28(%rax)
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm9, %xmm0
 ; CHECK-NEXT:    vmovd %xmm0, %ecx
 ; CHECK-NEXT:    movw %cx, 30(%rax)
 ; CHECK-NEXT:    retq
+; CHECK-NEXT:  LBB12_9: ## %cond.load7
+; CHECK-NEXT:    movswl 6(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm7
+; CHECK-NEXT:    testb $16, %cl
+; CHECK-NEXT:    je LBB12_12
+; CHECK-NEXT:  LBB12_11: ## %cond.load10
+; CHECK-NEXT:    movswl 8(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm6
+; CHECK-NEXT:    testb $32, %cl
+; CHECK-NEXT:    je LBB12_14
+; CHECK-NEXT:  LBB12_13: ## %cond.load13
+; CHECK-NEXT:    movswl 10(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm5
+; CHECK-NEXT:    testb $64, %cl
+; CHECK-NEXT:    je LBB12_16
+; CHECK-NEXT:  LBB12_15: ## %cond.load16
+; CHECK-NEXT:    movswl 12(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm4
+; CHECK-NEXT:    testb $-128, %cl
+; CHECK-NEXT:    je LBB12_18
+; CHECK-NEXT:  LBB12_17: ## %cond.load19
+; CHECK-NEXT:    movswl 14(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm3
+; CHECK-NEXT:    testl $256, %ecx ## imm = 0x100
+; CHECK-NEXT:    je LBB12_20
+; CHECK-NEXT:  LBB12_19: ## %cond.load22
+; CHECK-NEXT:    movswl 16(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm16
+; CHECK-NEXT:    testl $512, %ecx ## imm = 0x200
+; CHECK-NEXT:    je LBB12_22
+; CHECK-NEXT:  LBB12_21: ## %cond.load25
+; CHECK-NEXT:    movswl 18(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm15
+; CHECK-NEXT:    testl $1024, %ecx ## imm = 0x400
+; CHECK-NEXT:    je LBB12_24
+; CHECK-NEXT:  LBB12_23: ## %cond.load28
+; CHECK-NEXT:    movswl 20(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm14
+; CHECK-NEXT:    testl $2048, %ecx ## imm = 0x800
+; CHECK-NEXT:    je LBB12_26
+; CHECK-NEXT:  LBB12_25: ## %cond.load31
+; CHECK-NEXT:    movswl 22(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm13
+; CHECK-NEXT:    testl $4096, %ecx ## imm = 0x1000
+; CHECK-NEXT:    je LBB12_28
+; CHECK-NEXT:  LBB12_27: ## %cond.load34
+; CHECK-NEXT:    movswl 24(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm12
+; CHECK-NEXT:    testl $8192, %ecx ## imm = 0x2000
+; CHECK-NEXT:    je LBB12_30
+; CHECK-NEXT:  LBB12_29: ## %cond.load37
+; CHECK-NEXT:    movswl 26(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm11
+; CHECK-NEXT:    testl $16384, %ecx ## imm = 0x4000
+; CHECK-NEXT:    je LBB12_32
+; CHECK-NEXT:  LBB12_31: ## %cond.load40
+; CHECK-NEXT:    movswl 28(%rsi), %edx
+; CHECK-NEXT:    vmovd %edx, %xmm0
+; CHECK-NEXT:    vcvtph2ps %xmm0, %xmm10
+; CHECK-NEXT:    testl $32768, %ecx ## imm = 0x8000
+; CHECK-NEXT:    jne LBB12_33
+; CHECK-NEXT:    jmp LBB12_34
   %res = call <16 x half> @llvm.masked.load.v16f16(<16 x half>* %addr, i32 4, <16 x i1>%mask, <16 x half> zeroinitializer)
   ret <16 x half> %res
 }
 declare <16 x half> @llvm.masked.load.v16f16(<16 x half>*, i32, <16 x i1>, <16 x half>)
 
 ; Make sure we scalarize masked stores of f16.
 define void @test_mask_store_16xf16(<16 x i1> %mask, <16 x half>* %addr, <16 x half> %val) {
 ; CHECK-LABEL: test_mask_store_16xf16:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vpsllw $7, %xmm0, %xmm0
-; CHECK-NEXT:    vpmovb2m %xmm0, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
+; CHECK-NEXT:    vpmovmskb %xmm0, %eax
 ; CHECK-NEXT:    testb $1, %al
-; CHECK-NEXT:    je LBB13_2
-; CHECK-NEXT:  ## %bb.1: ## %cond.store
+; CHECK-NEXT:    jne LBB13_1
+; CHECK-NEXT:  ## %bb.2: ## %else
+; CHECK-NEXT:    testb $2, %al
+; CHECK-NEXT:    jne LBB13_3
+; CHECK-NEXT:  LBB13_4: ## %else2
+; CHECK-NEXT:    testb $4, %al
+; CHECK-NEXT:    jne LBB13_5
+; CHECK-NEXT:  LBB13_6: ## %else4
+; CHECK-NEXT:    testb $8, %al
+; CHECK-NEXT:    jne LBB13_7
+; CHECK-NEXT:  LBB13_8: ## %else6
+; CHECK-NEXT:    testb $16, %al
+; CHECK-NEXT:    jne LBB13_9
+; CHECK-NEXT:  LBB13_10: ## %else8
+; CHECK-NEXT:    testb $32, %al
+; CHECK-NEXT:    jne LBB13_11
+; CHECK-NEXT:  LBB13_12: ## %else10
+; CHECK-NEXT:    testb $64, %al
+; CHECK-NEXT:    jne LBB13_13
+; CHECK-NEXT:  LBB13_14: ## %else12
+; CHECK-NEXT:    testb $-128, %al
+; CHECK-NEXT:    jne LBB13_15
+; CHECK-NEXT:  LBB13_16: ## %else14
+; CHECK-NEXT:    testl $256, %eax ## imm = 0x100
+; CHECK-NEXT:    jne LBB13_17
+; CHECK-NEXT:  LBB13_18: ## %else16
+; CHECK-NEXT:    testl $512, %eax ## imm = 0x200
+; CHECK-NEXT:    jne LBB13_19
+; CHECK-NEXT:  LBB13_20: ## %else18
+; CHECK-NEXT:    testl $1024, %eax ## imm = 0x400
+; CHECK-NEXT:    jne LBB13_21
+; CHECK-NEXT:  LBB13_22: ## %else20
+; CHECK-NEXT:    testl $2048, %eax ## imm = 0x800
+; CHECK-NEXT:    jne LBB13_23
+; CHECK-NEXT:  LBB13_24: ## %else22
+; CHECK-NEXT:    testl $4096, %eax ## imm = 0x1000
+; CHECK-NEXT:    jne LBB13_25
+; CHECK-NEXT:  LBB13_26: ## %else24
+; CHECK-NEXT:    testl $8192, %eax ## imm = 0x2000
+; CHECK-NEXT:    jne LBB13_27
+; CHECK-NEXT:  LBB13_28: ## %else26
+; CHECK-NEXT:    testl $16384, %eax ## imm = 0x4000
+; CHECK-NEXT:    jne LBB13_29
+; CHECK-NEXT:  LBB13_30: ## %else28
+; CHECK-NEXT:    testl $32768, %eax ## imm = 0x8000
+; CHECK-NEXT:    jne LBB13_31
+; CHECK-NEXT:  LBB13_32: ## %else30
+; CHECK-NEXT:    retq
+; CHECK-NEXT:  LBB13_1: ## %cond.store
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm1, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, (%rdi)
-; CHECK-NEXT:  LBB13_2: ## %else
-; CHECK-NEXT:    kshiftrw $1, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, (%rdi)
+; CHECK-NEXT:    testb $2, %al
 ; CHECK-NEXT:    je LBB13_4
-; CHECK-NEXT:  ## %bb.3: ## %cond.store1
+; CHECK-NEXT:  LBB13_3: ## %cond.store1
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm2, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 2(%rdi)
-; CHECK-NEXT:  LBB13_4: ## %else2
-; CHECK-NEXT:    kshiftrw $2, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 2(%rdi)
+; CHECK-NEXT:    testb $4, %al
 ; CHECK-NEXT:    je LBB13_6
-; CHECK-NEXT:  ## %bb.5: ## %cond.store3
+; CHECK-NEXT:  LBB13_5: ## %cond.store3
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm3, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 4(%rdi)
-; CHECK-NEXT:  LBB13_6: ## %else4
-; CHECK-NEXT:    kshiftrw $3, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 4(%rdi)
+; CHECK-NEXT:    testb $8, %al
 ; CHECK-NEXT:    je LBB13_8
-; CHECK-NEXT:  ## %bb.7: ## %cond.store5
+; CHECK-NEXT:  LBB13_7: ## %cond.store5
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm4, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 6(%rdi)
-; CHECK-NEXT:  LBB13_8: ## %else6
-; CHECK-NEXT:    kshiftrw $4, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 6(%rdi)
+; CHECK-NEXT:    testb $16, %al
 ; CHECK-NEXT:    je LBB13_10
-; CHECK-NEXT:  ## %bb.9: ## %cond.store7
+; CHECK-NEXT:  LBB13_9: ## %cond.store7
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm5, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 8(%rdi)
-; CHECK-NEXT:  LBB13_10: ## %else8
-; CHECK-NEXT:    kshiftrw $5, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 8(%rdi)
+; CHECK-NEXT:    testb $32, %al
 ; CHECK-NEXT:    je LBB13_12
-; CHECK-NEXT:  ## %bb.11: ## %cond.store9
+; CHECK-NEXT:  LBB13_11: ## %cond.store9
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm6, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 10(%rdi)
-; CHECK-NEXT:  LBB13_12: ## %else10
-; CHECK-NEXT:    kshiftrw $6, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 10(%rdi)
+; CHECK-NEXT:    testb $64, %al
 ; CHECK-NEXT:    je LBB13_14
-; CHECK-NEXT:  ## %bb.13: ## %cond.store11
+; CHECK-NEXT:  LBB13_13: ## %cond.store11
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm7, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 12(%rdi)
-; CHECK-NEXT:  LBB13_14: ## %else12
-; CHECK-NEXT:    kshiftrw $7, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 12(%rdi)
+; CHECK-NEXT:    testb $-128, %al
 ; CHECK-NEXT:    je LBB13_16
-; CHECK-NEXT:  ## %bb.15: ## %cond.store13
+; CHECK-NEXT:  LBB13_15: ## %cond.store13
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 14(%rdi)
-; CHECK-NEXT:  LBB13_16: ## %else14
-; CHECK-NEXT:    kshiftrw $8, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 14(%rdi)
+; CHECK-NEXT:    testl $256, %eax ## imm = 0x100
 ; CHECK-NEXT:    je LBB13_18
-; CHECK-NEXT:  ## %bb.17: ## %cond.store15
+; CHECK-NEXT:  LBB13_17: ## %cond.store15
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 16(%rdi)
-; CHECK-NEXT:  LBB13_18: ## %else16
-; CHECK-NEXT:    kshiftrw $9, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 16(%rdi)
+; CHECK-NEXT:    testl $512, %eax ## imm = 0x200
 ; CHECK-NEXT:    je LBB13_20
-; CHECK-NEXT:  ## %bb.19: ## %cond.store17
+; CHECK-NEXT:  LBB13_19: ## %cond.store17
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 18(%rdi)
-; CHECK-NEXT:  LBB13_20: ## %else18
-; CHECK-NEXT:    kshiftrw $10, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 18(%rdi)
+; CHECK-NEXT:    testl $1024, %eax ## imm = 0x400
 ; CHECK-NEXT:    je LBB13_22
-; CHECK-NEXT:  ## %bb.21: ## %cond.store19
+; CHECK-NEXT:  LBB13_21: ## %cond.store19
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 20(%rdi)
-; CHECK-NEXT:  LBB13_22: ## %else20
-; CHECK-NEXT:    kshiftrw $11, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 20(%rdi)
+; CHECK-NEXT:    testl $2048, %eax ## imm = 0x800
 ; CHECK-NEXT:    je LBB13_24
-; CHECK-NEXT:  ## %bb.23: ## %cond.store21
+; CHECK-NEXT:  LBB13_23: ## %cond.store21
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 22(%rdi)
-; CHECK-NEXT:  LBB13_24: ## %else22
-; CHECK-NEXT:    kshiftrw $12, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 22(%rdi)
+; CHECK-NEXT:    testl $4096, %eax ## imm = 0x1000
 ; CHECK-NEXT:    je LBB13_26
-; CHECK-NEXT:  ## %bb.25: ## %cond.store23
+; CHECK-NEXT:  LBB13_25: ## %cond.store23
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 24(%rdi)
-; CHECK-NEXT:  LBB13_26: ## %else24
-; CHECK-NEXT:    kshiftrw $13, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 24(%rdi)
+; CHECK-NEXT:    testl $8192, %eax ## imm = 0x2000
 ; CHECK-NEXT:    je LBB13_28
-; CHECK-NEXT:  ## %bb.27: ## %cond.store25
+; CHECK-NEXT:  LBB13_27: ## %cond.store25
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 26(%rdi)
-; CHECK-NEXT:  LBB13_28: ## %else26
-; CHECK-NEXT:    kshiftrw $14, %k0, %k1
-; CHECK-NEXT:    kmovd %k1, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 26(%rdi)
+; CHECK-NEXT:    testl $16384, %eax ## imm = 0x4000
 ; CHECK-NEXT:    je LBB13_30
-; CHECK-NEXT:  ## %bb.29: ## %cond.store27
+; CHECK-NEXT:  LBB13_29: ## %cond.store27
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
-; CHECK-NEXT:    vmovd %xmm0, %eax
-; CHECK-NEXT:    movw %ax, 28(%rdi)
-; CHECK-NEXT:  LBB13_30: ## %else28
-; CHECK-NEXT:    kshiftrw $15, %k0, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    testb $1, %al
+; CHECK-NEXT:    vmovd %xmm0, %ecx
+; CHECK-NEXT:    movw %cx, 28(%rdi)
+; CHECK-NEXT:    testl $32768, %eax ## imm = 0x8000
 ; CHECK-NEXT:    je LBB13_32
-; CHECK-NEXT:  ## %bb.31: ## %cond.store29
+; CHECK-NEXT:  LBB13_31: ## %cond.store29
 ; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
 ; CHECK-NEXT:    vmovd %xmm0, %eax
 ; CHECK-NEXT:    movw %ax, 30(%rdi)
-; CHECK-NEXT:  LBB13_32: ## %else30
 ; CHECK-NEXT:    retq
   call void @llvm.masked.store.v16f16.p0v16f16(<16 x half> %val, <16 x half>* %addr, i32 4, <16 x i1>%mask)
   ret void
 }
 declare void @llvm.masked.store.v16f16.p0v16f16(<16 x half>, <16 x half>*, i32, <16 x i1>)

llvm/trunk/test/CodeGen/X86/masked_gather.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+sse4.2 < %s | FileCheck %s --check-prefixes=ALL,SSE,SSE42
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+avx < %s | FileCheck %s --check-prefixes=ALL,AVX,AVX1
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+avx2 < %s | FileCheck %s --check-prefixes=ALL,AVX,AVX2
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu  -mattr=+avx512f < %s | FileCheck %s --check-prefixes=ALL,AVX512
 
 ;
 ; vXf32
 ;
 
 define <4 x float> @gather_v4f32_ptr_v4i32(<4 x float*> %ptr, <4 x i32> %trigger, <4 x float> %passthru) {
 ; SSE-LABEL: gather_v4f32_ptr_v4i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pxor %xmm4, %xmm4
 ; SSE-NEXT:    pcmpeqd %xmm2, %xmm4
-; SSE-NEXT:    pextrb $0, %xmm4, %eax
-; SSE-NEXT:    testb $1, %al
-; SSE-NEXT:    je .LBB0_2
-; SSE-NEXT:  # %bb.1: # %cond.load
-; SSE-NEXT:    movq %xmm0, %rax
-; SSE-NEXT:    movss {{.*#+}} xmm5 = mem[0],zero,zero,zero
-; SSE-NEXT:    blendps {{.*#+}} xmm3 = xmm5[0],xmm3[1,2,3]
-; SSE-NEXT:  .LBB0_2: # %else
-; SSE-NEXT:    pextrb $4, %xmm4, %eax
+; SSE-NEXT:    movmskps %xmm4, %eax
 ; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    jne .LBB0_1
+; SSE-NEXT:  # %bb.2: # %else
+; SSE-NEXT:    testb $2, %al
+; SSE-NEXT:    jne .LBB0_3
+; SSE-NEXT:  .LBB0_4: # %else2
+; SSE-NEXT:    testb $4, %al
+; SSE-NEXT:    jne .LBB0_5
+; SSE-NEXT:  .LBB0_6: # %else5
+; SSE-NEXT:    testb $8, %al
+; SSE-NEXT:    jne .LBB0_7
+; SSE-NEXT:  .LBB0_8: # %else8
+; SSE-NEXT:    movaps %xmm3, %xmm0
+; SSE-NEXT:    retq
+; SSE-NEXT:  .LBB0_1: # %cond.load
+; SSE-NEXT:    movq %xmm0, %rcx
+; SSE-NEXT:    movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
+; SSE-NEXT:    pblendw {{.*#+}} xmm3 = xmm2[0,1],xmm3[2,3,4,5,6,7]
+; SSE-NEXT:    testb $2, %al
 ; SSE-NEXT:    je .LBB0_4
-; SSE-NEXT:  # %bb.3: # %cond.load1
-; SSE-NEXT:    pextrq $1, %xmm0, %rax
+; SSE-NEXT:  .LBB0_3: # %cond.load1
+; SSE-NEXT:    pextrq $1, %xmm0, %rcx
 ; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0],mem[0],xmm3[2,3]
-; SSE-NEXT:  .LBB0_4: # %else2
-; SSE-NEXT:    pxor %xmm0, %xmm0
-; SSE-NEXT:    pcmpeqd %xmm0, %xmm2
-; SSE-NEXT:    pextrb $8, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $4, %al
 ; SSE-NEXT:    je .LBB0_6
-; SSE-NEXT:  # %bb.5: # %cond.load4
-; SSE-NEXT:    movq %xmm1, %rax
+; SSE-NEXT:  .LBB0_5: # %cond.load4
+; SSE-NEXT:    movq %xmm1, %rcx
 ; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3]
-; SSE-NEXT:  .LBB0_6: # %else5
-; SSE-NEXT:    pextrb $12, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $8, %al
 ; SSE-NEXT:    je .LBB0_8
-; SSE-NEXT:  # %bb.7: # %cond.load7
+; SSE-NEXT:  .LBB0_7: # %cond.load7
 ; SSE-NEXT:    pextrq $1, %xmm1, %rax
 ; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0,1,2],mem[0]
-; SSE-NEXT:  .LBB0_8: # %else8
 ; SSE-NEXT:    movaps %xmm3, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: gather_v4f32_ptr_v4i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm3
-; AVX1-NEXT:    vpextrb $0, %xmm3, %eax
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vmovmskps %xmm1, %eax
 ; AVX1-NEXT:    testb $1, %al
 ; AVX1-NEXT:    je .LBB0_2
 ; AVX1-NEXT:  # %bb.1: # %cond.load
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX1-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0],xmm2[1,2,3]
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX1-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
 ; AVX1-NEXT:  .LBB0_2: # %else
-; AVX1-NEXT:    vpextrb $4, %xmm3, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $2, %al
 ; AVX1-NEXT:    je .LBB0_4
 ; AVX1-NEXT:  # %bb.3: # %cond.load1
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
 ; AVX1-NEXT:  .LBB0_4: # %else2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $4, %al
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    je .LBB0_6
-; AVX1-NEXT:  # %bb.5: # %cond.load4
-; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    jne .LBB0_5
+; AVX1-NEXT:  # %bb.6: # %else5
+; AVX1-NEXT:    testb $8, %al
+; AVX1-NEXT:    jne .LBB0_7
+; AVX1-NEXT:  .LBB0_8: # %else8
+; AVX1-NEXT:    vmovaps %xmm2, %xmm0
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+; AVX1-NEXT:  .LBB0_5: # %cond.load4
+; AVX1-NEXT:    vmovq %xmm0, %rcx
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; AVX1-NEXT:  .LBB0_6: # %else5
-; AVX1-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $8, %al
 ; AVX1-NEXT:    je .LBB0_8
-; AVX1-NEXT:  # %bb.7: # %cond.load7
+; AVX1-NEXT:  .LBB0_7: # %cond.load7
 ; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; AVX1-NEXT:  .LBB0_8: # %else8
 ; AVX1-NEXT:    vmovaps %xmm2, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: gather_v4f32_ptr_v4i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm3
-; AVX2-NEXT:    vpextrb $0, %xmm3, %eax
+; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
+; AVX2-NEXT:    vmovmskps %xmm1, %eax
 ; AVX2-NEXT:    testb $1, %al
 ; AVX2-NEXT:    je .LBB0_2
 ; AVX2-NEXT:  # %bb.1: # %cond.load
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0],xmm2[1,2,3]
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
 ; AVX2-NEXT:  .LBB0_2: # %else
-; AVX2-NEXT:    vpextrb $4, %xmm3, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $2, %al
 ; AVX2-NEXT:    je .LBB0_4
 ; AVX2-NEXT:  # %bb.3: # %cond.load1
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
 ; AVX2-NEXT:  .LBB0_4: # %else2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $4, %al
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT:    je .LBB0_6
-; AVX2-NEXT:  # %bb.5: # %cond.load4
-; AVX2-NEXT:    vmovq %xmm0, %rax
+; AVX2-NEXT:    jne .LBB0_5
+; AVX2-NEXT:  # %bb.6: # %else5
+; AVX2-NEXT:    testb $8, %al
+; AVX2-NEXT:    jne .LBB0_7
+; AVX2-NEXT:  .LBB0_8: # %else8
+; AVX2-NEXT:    vmovaps %xmm2, %xmm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+; AVX2-NEXT:  .LBB0_5: # %cond.load4
+; AVX2-NEXT:    vmovq %xmm0, %rcx
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; AVX2-NEXT:  .LBB0_6: # %else5
-; AVX2-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $8, %al
 ; AVX2-NEXT:    je .LBB0_8
-; AVX2-NEXT:  # %bb.7: # %cond.load7
+; AVX2-NEXT:  .LBB0_7: # %cond.load7
 ; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; AVX2-NEXT:  .LBB0_8: # %else8
 ; AVX2-NEXT:    vmovaps %xmm2, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: gather_v4f32_ptr_v4i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    # kill: def $xmm2 killed $xmm2 def $ymm2
 ; AVX512-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1

 ; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,0,1]
 ; SSE-NEXT:    pmovsxdq %xmm0, %xmm4
 ; SSE-NEXT:    psllq $2, %xmm4
 ; SSE-NEXT:    paddq %xmm3, %xmm4
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; SSE-NEXT:    pmovsxdq %xmm0, %xmm0
 ; SSE-NEXT:    pxor %xmm5, %xmm5
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm5
-; SSE-NEXT:    pextrb $0, %xmm5, %eax
+; SSE-NEXT:    movmskps %xmm5, %eax
 ; SSE-NEXT:    testb $1, %al
 ; SSE-NEXT:    je .LBB1_2
 ; SSE-NEXT:  # %bb.1: # %cond.load
-; SSE-NEXT:    movq %xmm4, %rax
-; SSE-NEXT:    movss {{.*#+}} xmm6 = mem[0],zero,zero,zero
-; SSE-NEXT:    blendps {{.*#+}} xmm2 = xmm6[0],xmm2[1,2,3]
+; SSE-NEXT:    movq %xmm4, %rcx
+; SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; SSE-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3,4,5,6,7]
 ; SSE-NEXT:  .LBB1_2: # %else
 ; SSE-NEXT:    psllq $2, %xmm0
-; SSE-NEXT:    pextrb $4, %xmm5, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $2, %al
 ; SSE-NEXT:    je .LBB1_4
 ; SSE-NEXT:  # %bb.3: # %cond.load1
-; SSE-NEXT:    pextrq $1, %xmm4, %rax
+; SSE-NEXT:    pextrq $1, %xmm4, %rcx
 ; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
 ; SSE-NEXT:  .LBB1_4: # %else2
 ; SSE-NEXT:    paddq %xmm0, %xmm3
-; SSE-NEXT:    pxor %xmm0, %xmm0
-; SSE-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSE-NEXT:    pextrb $8, %xmm1, %eax
-; SSE-NEXT:    testb $1, %al
-; SSE-NEXT:    je .LBB1_6
-; SSE-NEXT:  # %bb.5: # %cond.load4
-; SSE-NEXT:    movq %xmm3, %rax
+; SSE-NEXT:    testb $4, %al
+; SSE-NEXT:    jne .LBB1_5
+; SSE-NEXT:  # %bb.6: # %else5
+; SSE-NEXT:    testb $8, %al
+; SSE-NEXT:    jne .LBB1_7
+; SSE-NEXT:  .LBB1_8: # %else8
+; SSE-NEXT:    movaps %xmm2, %xmm0
+; SSE-NEXT:    retq
+; SSE-NEXT:  .LBB1_5: # %cond.load4
+; SSE-NEXT:    movq %xmm3, %rcx
 ; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; SSE-NEXT:  .LBB1_6: # %else5
-; SSE-NEXT:    pextrb $12, %xmm1, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $8, %al
 ; SSE-NEXT:    je .LBB1_8
-; SSE-NEXT:  # %bb.7: # %cond.load7
+; SSE-NEXT:  .LBB1_7: # %cond.load7
 ; SSE-NEXT:    pextrq $1, %xmm3, %rax
 ; SSE-NEXT:    insertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; SSE-NEXT:  .LBB1_8: # %else8
 ; SSE-NEXT:    movaps %xmm2, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: gather_v4f32_v4i32_v4i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vmovq %rdi, %xmm3
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,1,0,1]
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovsxdq %xmm4, %xmm4
 ; AVX1-NEXT:    vpsllq $2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpaddq %xmm4, %xmm3, %xmm4
 ; AVX1-NEXT:    vpmovsxdq %xmm0, %xmm0
 ; AVX1-NEXT:    vpsllq $2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpaddq %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm3
-; AVX1-NEXT:    vpextrb $0, %xmm3, %eax
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vmovmskps %xmm1, %eax
 ; AVX1-NEXT:    testb $1, %al
 ; AVX1-NEXT:    je .LBB1_2
 ; AVX1-NEXT:  # %bb.1: # %cond.load
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vmovd {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3,4,5,6,7]
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX1-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
 ; AVX1-NEXT:  .LBB1_2: # %else
-; AVX1-NEXT:    vpextrb $4, %xmm3, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $2, %al
 ; AVX1-NEXT:    je .LBB1_4
 ; AVX1-NEXT:  # %bb.3: # %cond.load1
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
 ; AVX1-NEXT:  .LBB1_4: # %else2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $4, %al
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    je .LBB1_6
-; AVX1-NEXT:  # %bb.5: # %cond.load4
-; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    jne .LBB1_5
+; AVX1-NEXT:  # %bb.6: # %else5
+; AVX1-NEXT:    testb $8, %al
+; AVX1-NEXT:    jne .LBB1_7
+; AVX1-NEXT:  .LBB1_8: # %else8
+; AVX1-NEXT:    vmovaps %xmm2, %xmm0
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+; AVX1-NEXT:  .LBB1_5: # %cond.load4
+; AVX1-NEXT:    vmovq %xmm0, %rcx
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; AVX1-NEXT:  .LBB1_6: # %else5
-; AVX1-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $8, %al
 ; AVX1-NEXT:    je .LBB1_8
-; AVX1-NEXT:  # %bb.7: # %cond.load7
+; AVX1-NEXT:  .LBB1_7: # %cond.load7
 ; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; AVX1-NEXT:  .LBB1_8: # %else8
 ; AVX1-NEXT:    vmovaps %xmm2, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: gather_v4f32_v4i32_v4i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vmovq %rdi, %xmm3
 ; AVX2-NEXT:    vpbroadcastq %xmm3, %ymm3
 ; AVX2-NEXT:    vpmovsxdq %xmm0, %ymm0
 ; AVX2-NEXT:    vpsllq $2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpaddq %ymm0, %ymm3, %ymm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm3
-; AVX2-NEXT:    vpextrb $0, %xmm3, %eax
+; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
+; AVX2-NEXT:    vmovmskps %xmm1, %eax
 ; AVX2-NEXT:    testb $1, %al
 ; AVX2-NEXT:    je .LBB1_2
 ; AVX2-NEXT:  # %bb.1: # %cond.load
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0],xmm2[1,2,3]
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
 ; AVX2-NEXT:  .LBB1_2: # %else
-; AVX2-NEXT:    vpextrb $4, %xmm3, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $2, %al
 ; AVX2-NEXT:    je .LBB1_4
 ; AVX2-NEXT:  # %bb.3: # %cond.load1
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
 ; AVX2-NEXT:  .LBB1_4: # %else2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $4, %al
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT:    je .LBB1_6
-; AVX2-NEXT:  # %bb.5: # %cond.load4
-; AVX2-NEXT:    vmovq %xmm0, %rax
+; AVX2-NEXT:    jne .LBB1_5
+; AVX2-NEXT:  # %bb.6: # %else5
+; AVX2-NEXT:    testb $8, %al
+; AVX2-NEXT:    jne .LBB1_7
+; AVX2-NEXT:  .LBB1_8: # %else8
+; AVX2-NEXT:    vmovaps %xmm2, %xmm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+; AVX2-NEXT:  .LBB1_5: # %cond.load4
+; AVX2-NEXT:    vmovq %xmm0, %rcx
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; AVX2-NEXT:  .LBB1_6: # %else5
-; AVX2-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $8, %al
 ; AVX2-NEXT:    je .LBB1_8
-; AVX2-NEXT:  # %bb.7: # %cond.load7
+; AVX2-NEXT:  .LBB1_7: # %cond.load7
 ; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; AVX2-NEXT:  .LBB1_8: # %else8
 ; AVX2-NEXT:    vmovaps %xmm2, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: gather_v4f32_v4i32_v4i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    # kill: def $xmm2 killed $xmm2 def $zmm2
 ; AVX512-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1

 ; SSE-LABEL: gather_v4f32_v4i64_v4i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movq %rdi, %xmm4
 ; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,0,1]
 ; SSE-NEXT:    psllq $2, %xmm0
 ; SSE-NEXT:    paddq %xmm4, %xmm0
 ; SSE-NEXT:    pxor %xmm5, %xmm5
 ; SSE-NEXT:    pcmpeqd %xmm2, %xmm5
-; SSE-NEXT:    pextrb $0, %xmm5, %eax
+; SSE-NEXT:    movmskps %xmm5, %eax
 ; SSE-NEXT:    testb $1, %al
 ; SSE-NEXT:    je .LBB2_2
 ; SSE-NEXT:  # %bb.1: # %cond.load
-; SSE-NEXT:    movq %xmm0, %rax
-; SSE-NEXT:    movss {{.*#+}} xmm6 = mem[0],zero,zero,zero
-; SSE-NEXT:    blendps {{.*#+}} xmm3 = xmm6[0],xmm3[1,2,3]
+; SSE-NEXT:    movq %xmm0, %rcx
+; SSE-NEXT:    movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
+; SSE-NEXT:    pblendw {{.*#+}} xmm3 = xmm2[0,1],xmm3[2,3,4,5,6,7]
 ; SSE-NEXT:  .LBB2_2: # %else
 ; SSE-NEXT:    psllq $2, %xmm1
-; SSE-NEXT:    pextrb $4, %xmm5, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $2, %al
 ; SSE-NEXT:    je .LBB2_4
 ; SSE-NEXT:  # %bb.3: # %cond.load1
-; SSE-NEXT:    pextrq $1, %xmm0, %rax
+; SSE-NEXT:    pextrq $1, %xmm0, %rcx
 ; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0],mem[0],xmm3[2,3]
 ; SSE-NEXT:  .LBB2_4: # %else2
 ; SSE-NEXT:    paddq %xmm1, %xmm4
-; SSE-NEXT:    pxor %xmm0, %xmm0
-; SSE-NEXT:    pcmpeqd %xmm0, %xmm2
-; SSE-NEXT:    pextrb $8, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
-; SSE-NEXT:    je .LBB2_6
-; SSE-NEXT:  # %bb.5: # %cond.load4
-; SSE-NEXT:    movq %xmm4, %rax
+; SSE-NEXT:    testb $4, %al
+; SSE-NEXT:    jne .LBB2_5
+; SSE-NEXT:  # %bb.6: # %else5
+; SSE-NEXT:    testb $8, %al
+; SSE-NEXT:    jne .LBB2_7
+; SSE-NEXT:  .LBB2_8: # %else8
+; SSE-NEXT:    movaps %xmm3, %xmm0
+; SSE-NEXT:    retq
+; SSE-NEXT:  .LBB2_5: # %cond.load4
+; SSE-NEXT:    movq %xmm4, %rcx
 ; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3]
-; SSE-NEXT:  .LBB2_6: # %else5
-; SSE-NEXT:    pextrb $12, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $8, %al
 ; SSE-NEXT:    je .LBB2_8
-; SSE-NEXT:  # %bb.7: # %cond.load7
+; SSE-NEXT:  .LBB2_7: # %cond.load7
 ; SSE-NEXT:    pextrq $1, %xmm4, %rax
 ; SSE-NEXT:    insertps {{.*#+}} xmm3 = xmm3[0,1,2],mem[0]
-; SSE-NEXT:  .LBB2_8: # %else8
 ; SSE-NEXT:    movaps %xmm3, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: gather_v4f32_v4i64_v4i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpsllq $2, %xmm3, %xmm3
 ; AVX1-NEXT:    vmovq %rdi, %xmm4
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,1,0,1]
 ; AVX1-NEXT:    vpaddq %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpsllq $2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm3
-; AVX1-NEXT:    vpextrb $0, %xmm3, %eax
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vmovmskps %xmm1, %eax
 ; AVX1-NEXT:    testb $1, %al
 ; AVX1-NEXT:    je .LBB2_2
 ; AVX1-NEXT:  # %bb.1: # %cond.load
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vmovd {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3,4,5,6,7]
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX1-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
 ; AVX1-NEXT:  .LBB2_2: # %else
-; AVX1-NEXT:    vpextrb $4, %xmm3, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $2, %al
 ; AVX1-NEXT:    je .LBB2_4
 ; AVX1-NEXT:  # %bb.3: # %cond.load1
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
 ; AVX1-NEXT:  .LBB2_4: # %else2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $4, %al
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    je .LBB2_6
-; AVX1-NEXT:  # %bb.5: # %cond.load4
-; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    jne .LBB2_5
+; AVX1-NEXT:  # %bb.6: # %else5
+; AVX1-NEXT:    testb $8, %al
+; AVX1-NEXT:    jne .LBB2_7
+; AVX1-NEXT:  .LBB2_8: # %else8
+; AVX1-NEXT:    vmovaps %xmm2, %xmm0
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+; AVX1-NEXT:  .LBB2_5: # %cond.load4
+; AVX1-NEXT:    vmovq %xmm0, %rcx
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; AVX1-NEXT:  .LBB2_6: # %else5
-; AVX1-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testb $8, %al
 ; AVX1-NEXT:    je .LBB2_8
-; AVX1-NEXT:  # %bb.7: # %cond.load7
+; AVX1-NEXT:  .LBB2_7: # %cond.load7
 ; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; AVX1-NEXT:  .LBB2_8: # %else8
 ; AVX1-NEXT:    vmovaps %xmm2, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: gather_v4f32_v4i64_v4i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vmovq %rdi, %xmm3
 ; AVX2-NEXT:    vpbroadcastq %xmm3, %ymm3
 ; AVX2-NEXT:    vpsllq $2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpaddq %ymm0, %ymm3, %ymm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm3
-; AVX2-NEXT:    vpextrb $0, %xmm3, %eax
+; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
+; AVX2-NEXT:    vmovmskps %xmm1, %eax
 ; AVX2-NEXT:    testb $1, %al
 ; AVX2-NEXT:    je .LBB2_2
 ; AVX2-NEXT:  # %bb.1: # %cond.load
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0],xmm2[1,2,3]
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
 ; AVX2-NEXT:  .LBB2_2: # %else
-; AVX2-NEXT:    vpextrb $4, %xmm3, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $2, %al
 ; AVX2-NEXT:    je .LBB2_4
 ; AVX2-NEXT:  # %bb.3: # %cond.load1
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3]
 ; AVX2-NEXT:  .LBB2_4: # %else2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $4, %al
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT:    je .LBB2_6
-; AVX2-NEXT:  # %bb.5: # %cond.load4
-; AVX2-NEXT:    vmovq %xmm0, %rax
+; AVX2-NEXT:    jne .LBB2_5
+; AVX2-NEXT:  # %bb.6: # %else5
+; AVX2-NEXT:    testb $8, %al
+; AVX2-NEXT:    jne .LBB2_7
+; AVX2-NEXT:  .LBB2_8: # %else8
+; AVX2-NEXT:    vmovaps %xmm2, %xmm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+; AVX2-NEXT:  .LBB2_5: # %cond.load4
+; AVX2-NEXT:    vmovq %xmm0, %rcx
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; AVX2-NEXT:  .LBB2_6: # %else5
-; AVX2-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $8, %al
 ; AVX2-NEXT:    je .LBB2_8
-; AVX2-NEXT:  # %bb.7: # %cond.load7
+; AVX2-NEXT:  .LBB2_7: # %cond.load7
 ; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX2-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; AVX2-NEXT:  .LBB2_8: # %else8
 ; AVX2-NEXT:    vmovaps %xmm2, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: gather_v4f32_v4i64_v4i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    # kill: def $xmm2 killed $xmm2 def $ymm2
 ; AVX512-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1

 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movq %rdi, %xmm6
 ; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm6[0,1,0,1]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm0[2,3,0,1]
 ; SSE-NEXT:    pmovsxdq %xmm0, %xmm0
 ; SSE-NEXT:    paddq %xmm8, %xmm0
 ; SSE-NEXT:    pxor %xmm6, %xmm6
 ; SSE-NEXT:    pcmpeqb %xmm4, %xmm6
-; SSE-NEXT:    pextrb $0, %xmm6, %eax
+; SSE-NEXT:    pmovmskb %xmm6, %eax
 ; SSE-NEXT:    testb $1, %al
 ; SSE-NEXT:    je .LBB3_2
 ; SSE-NEXT:  # %bb.1: # %cond.load
-; SSE-NEXT:    movq %xmm0, %rax
-; SSE-NEXT:    pinsrb $0, (%rax), %xmm5
+; SSE-NEXT:    movq %xmm0, %rcx
+; SSE-NEXT:    pinsrb $0, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_2: # %else
-; SSE-NEXT:    pmovsxdq %xmm7, %xmm7
-; SSE-NEXT:    pextrb $1, %xmm6, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    pmovsxdq %xmm7, %xmm4
+; SSE-NEXT:    testb $2, %al
 ; SSE-NEXT:    je .LBB3_4
 ; SSE-NEXT:  # %bb.3: # %cond.load1
-; SSE-NEXT:    pextrq $1, %xmm0, %rax
-; SSE-NEXT:    pinsrb $1, (%rax), %xmm5
+; SSE-NEXT:    pextrq $1, %xmm0, %rcx
+; SSE-NEXT:    pinsrb $1, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_4: # %else2
-; SSE-NEXT:    paddq %xmm8, %xmm7
-; SSE-NEXT:    pxor %xmm6, %xmm6
-; SSE-NEXT:    pcmpeqb %xmm4, %xmm6
-; SSE-NEXT:    pextrb $2, %xmm6, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    paddq %xmm8, %xmm4
+; SSE-NEXT:    testb $4, %al
 ; SSE-NEXT:    je .LBB3_6
 ; SSE-NEXT:  # %bb.5: # %cond.load4
-; SSE-NEXT:    movq %xmm7, %rax
-; SSE-NEXT:    pinsrb $2, (%rax), %xmm5
+; SSE-NEXT:    movq %xmm4, %rcx
+; SSE-NEXT:    pinsrb $2, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_6: # %else5
 ; SSE-NEXT:    pmovsxdq %xmm1, %xmm0
-; SSE-NEXT:    pextrb $3, %xmm6, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $8, %al
 ; SSE-NEXT:    je .LBB3_8
 ; SSE-NEXT:  # %bb.7: # %cond.load7
-; SSE-NEXT:    pextrq $1, %xmm7, %rax
-; SSE-NEXT:    pinsrb $3, (%rax), %xmm5
+; SSE-NEXT:    pextrq $1, %xmm4, %rcx
+; SSE-NEXT:    pinsrb $3, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_8: # %else8
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; SSE-NEXT:    paddq %xmm8, %xmm0
-; SSE-NEXT:    pxor %xmm6, %xmm6
-; SSE-NEXT:    pcmpeqb %xmm4, %xmm6
-; SSE-NEXT:    pextrb $4, %xmm6, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $16, %al
 ; SSE-NEXT:    je .LBB3_10
 ; SSE-NEXT:  # %bb.9: # %cond.load10
-; SSE-NEXT:    movq %xmm0, %rax
-; SSE-NEXT:    pinsrb $4, (%rax), %xmm5
+; SSE-NEXT:    movq %xmm0, %rcx
+; SSE-NEXT:    pinsrb $4, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_10: # %else11
 ; SSE-NEXT:    pmovsxdq %xmm1, %xmm1
-; SSE-NEXT:    pextrb $5, %xmm6, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $32, %al
 ; SSE-NEXT:    je .LBB3_12
 ; SSE-NEXT:  # %bb.11: # %cond.load13
-; SSE-NEXT:    pextrq $1, %xmm0, %rax
-; SSE-NEXT:    pinsrb $5, (%rax), %xmm5
+; SSE-NEXT:    pextrq $1, %xmm0, %rcx
+; SSE-NEXT:    pinsrb $5, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_12: # %else14
 ; SSE-NEXT:    paddq %xmm8, %xmm1
-; SSE-NEXT:    pxor %xmm6, %xmm6
-; SSE-NEXT:    pcmpeqb %xmm4, %xmm6
-; SSE-NEXT:    pextrb $6, %xmm6, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $64, %al
 ; SSE-NEXT:    je .LBB3_14
 ; SSE-NEXT:  # %bb.13: # %cond.load16
-; SSE-NEXT:    movq %xmm1, %rax
-; SSE-NEXT:    pinsrb $6, (%rax), %xmm5
+; SSE-NEXT:    movq %xmm1, %rcx
+; SSE-NEXT:    pinsrb $6, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_14: # %else17
 ; SSE-NEXT:    pmovsxdq %xmm2, %xmm0
-; SSE-NEXT:    pextrb $7, %xmm6, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testb $-128, %al
 ; SSE-NEXT:    je .LBB3_16
 ; SSE-NEXT:  # %bb.15: # %cond.load19
-; SSE-NEXT:    pextrq $1, %xmm1, %rax
-; SSE-NEXT:    pinsrb $7, (%rax), %xmm5
+; SSE-NEXT:    pextrq $1, %xmm1, %rcx
+; SSE-NEXT:    pinsrb $7, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_16: # %else20
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,3,0,1]
 ; SSE-NEXT:    paddq %xmm8, %xmm0
-; SSE-NEXT:    pxor %xmm2, %xmm2
-; SSE-NEXT:    pcmpeqb %xmm4, %xmm2
-; SSE-NEXT:    pextrb $8, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testl $256, %eax # imm = 0x100
 ; SSE-NEXT:    je .LBB3_18
 ; SSE-NEXT:  # %bb.17: # %cond.load22
-; SSE-NEXT:    movq %xmm0, %rax
-; SSE-NEXT:    pinsrb $8, (%rax), %xmm5
+; SSE-NEXT:    movq %xmm0, %rcx
+; SSE-NEXT:    pinsrb $8, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_18: # %else23
 ; SSE-NEXT:    pmovsxdq %xmm1, %xmm1
-; SSE-NEXT:    pextrb $9, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testl $512, %eax # imm = 0x200
 ; SSE-NEXT:    je .LBB3_20
 ; SSE-NEXT:  # %bb.19: # %cond.load25
-; SSE-NEXT:    pextrq $1, %xmm0, %rax
-; SSE-NEXT:    pinsrb $9, (%rax), %xmm5
+; SSE-NEXT:    pextrq $1, %xmm0, %rcx
+; SSE-NEXT:    pinsrb $9, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_20: # %else26
 ; SSE-NEXT:    paddq %xmm8, %xmm1
-; SSE-NEXT:    pxor %xmm2, %xmm2
-; SSE-NEXT:    pcmpeqb %xmm4, %xmm2
-; SSE-NEXT:    pextrb $10, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testl $1024, %eax # imm = 0x400
 ; SSE-NEXT:    je .LBB3_22
 ; SSE-NEXT:  # %bb.21: # %cond.load28
-; SSE-NEXT:    movq %xmm1, %rax
-; SSE-NEXT:    pinsrb $10, (%rax), %xmm5
+; SSE-NEXT:    movq %xmm1, %rcx
+; SSE-NEXT:    pinsrb $10, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_22: # %else29
 ; SSE-NEXT:    pmovsxdq %xmm3, %xmm0
-; SSE-NEXT:    pextrb $11, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testl $2048, %eax # imm = 0x800
 ; SSE-NEXT:    je .LBB3_24
 ; SSE-NEXT:  # %bb.23: # %cond.load31
-; SSE-NEXT:    pextrq $1, %xmm1, %rax
-; SSE-NEXT:    pinsrb $11, (%rax), %xmm5
+; SSE-NEXT:    pextrq $1, %xmm1, %rcx
+; SSE-NEXT:    pinsrb $11, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_24: # %else32
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[2,3,0,1]
 ; SSE-NEXT:    paddq %xmm8, %xmm0
-; SSE-NEXT:    pxor %xmm2, %xmm2
-; SSE-NEXT:    pcmpeqb %xmm4, %xmm2
-; SSE-NEXT:    pextrb $12, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testl $4096, %eax # imm = 0x1000
 ; SSE-NEXT:    je .LBB3_26
 ; SSE-NEXT:  # %bb.25: # %cond.load34
-; SSE-NEXT:    movq %xmm0, %rax
-; SSE-NEXT:    pinsrb $12, (%rax), %xmm5
+; SSE-NEXT:    movq %xmm0, %rcx
+; SSE-NEXT:    pinsrb $12, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_26: # %else35
 ; SSE-NEXT:    pmovsxdq %xmm1, %xmm1
-; SSE-NEXT:    pextrb $13, %xmm2, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testl $8192, %eax # imm = 0x2000
 ; SSE-NEXT:    je .LBB3_28
 ; SSE-NEXT:  # %bb.27: # %cond.load37
-; SSE-NEXT:    pextrq $1, %xmm0, %rax
-; SSE-NEXT:    pinsrb $13, (%rax), %xmm5
+; SSE-NEXT:    pextrq $1, %xmm0, %rcx
+; SSE-NEXT:    pinsrb $13, (%rcx), %xmm5
 ; SSE-NEXT:  .LBB3_28: # %else38
 ; SSE-NEXT:    paddq %xmm1, %xmm8
-; SSE-NEXT:    pxor %xmm0, %xmm0
-; SSE-NEXT:    pcmpeqb %xmm0, %xmm4
-; SSE-NEXT:    pextrb $14, %xmm4, %eax
-; SSE-NEXT:    testb $1, %al
-; SSE-NEXT:    je .LBB3_30
-; SSE-NEXT:  # %bb.29: # %cond.load40
-; SSE-NEXT:    movq %xmm8, %rax
-; SSE-NEXT:    pinsrb $14, (%rax), %xmm5
-; SSE-NEXT:  .LBB3_30: # %else41
-; SSE-NEXT:    pextrb $15, %xmm4, %eax
-; SSE-NEXT:    testb $1, %al
+; SSE-NEXT:    testl $16384, %eax # imm = 0x4000
+; SSE-NEXT:    jne .LBB3_29
+; SSE-NEXT:  # %bb.30: # %else41
+; SSE-NEXT:    testl $32768, %eax # imm = 0x8000
+; SSE-NEXT:    jne .LBB3_31
+; SSE-NEXT:  .LBB3_32: # %else44
+; SSE-NEXT:    movdqa %xmm5, %xmm0
+; SSE-NEXT:    retq
+; SSE-NEXT:  .LBB3_29: # %cond.load40
+; SSE-NEXT:    movq %xmm8, %rcx
+; SSE-NEXT:    pinsrb $14, (%rcx), %xmm5
+; SSE-NEXT:    testl $32768, %eax # imm = 0x8000
 ; SSE-NEXT:    je .LBB3_32
-; SSE-NEXT:  # %bb.31: # %cond.load43
+; SSE-NEXT:  .LBB3_31: # %cond.load43
 ; SSE-NEXT:    pextrq $1, %xmm8, %rax
 ; SSE-NEXT:    pinsrb $15, (%rax), %xmm5
-; SSE-NEXT:  .LBB3_32: # %else44
 ; SSE-NEXT:    movdqa %xmm5, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: gather_v16i8_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vmovq %rdi, %xmm4
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,1,0,1]
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovsxdq %xmm6, %xmm6
 ; AVX1-NEXT:    vpaddq %xmm6, %xmm4, %xmm6
 ; AVX1-NEXT:    vpmovsxdq %xmm0, %xmm0
 ; AVX1-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
 ; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm6
-; AVX1-NEXT:    vpextrb $0, %xmm6, %eax
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovmskb %xmm2, %eax
 ; AVX1-NEXT:    testb $1, %al
 ; AVX1-NEXT:    je .LBB3_2
 ; AVX1-NEXT:  # %bb.1: # %cond.load
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $0, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $0, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_2: # %else
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm7 = xmm5[2,3,0,1]
-; AVX1-NEXT:    vpmovsxdq %xmm5, %xmm5
-; AVX1-NEXT:    vpextrb $1, %xmm6, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[2,3,0,1]
+; AVX1-NEXT:    vpmovsxdq %xmm5, %xmm6
+; AVX1-NEXT:    testb $2, %al
 ; AVX1-NEXT:    je .LBB3_4
 ; AVX1-NEXT:  # %bb.3: # %cond.load1
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $1, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $1, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_4: # %else2
-; AVX1-NEXT:    vpmovsxdq %xmm7, %xmm6
-; AVX1-NEXT:    vpaddq %xmm5, %xmm4, %xmm5
-; AVX1-NEXT:    vpxor %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpcmpeqb %xmm7, %xmm2, %xmm7
-; AVX1-NEXT:    vpextrb $2, %xmm7, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vpmovsxdq %xmm2, %xmm5
+; AVX1-NEXT:    vpaddq %xmm6, %xmm4, %xmm2
+; AVX1-NEXT:    testb $4, %al
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    je .LBB3_6
 ; AVX1-NEXT:  # %bb.5: # %cond.load4
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $2, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $2, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_6: # %else5
-; AVX1-NEXT:    vpaddq %xmm6, %xmm4, %xmm6
-; AVX1-NEXT:    vpextrb $3, %xmm7, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vpaddq %xmm5, %xmm4, %xmm5
+; AVX1-NEXT:    testb $8, %al
 ; AVX1-NEXT:    je .LBB3_8
 ; AVX1-NEXT:  # %bb.7: # %cond.load7
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $3, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $3, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_8: # %else8
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm5, %ymm0
-; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm2, %xmm5
-; AVX1-NEXT:    vpextrb $4, %xmm5, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm2, %ymm0
+; AVX1-NEXT:    testb $16, %al
 ; AVX1-NEXT:    je .LBB3_10
 ; AVX1-NEXT:  # %bb.9: # %cond.load10
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $4, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $4, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_10: # %else11
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm1[2,3,0,1]
-; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm7
-; AVX1-NEXT:    vpextrb $5, %xmm5, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
+; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm6
+; AVX1-NEXT:    testb $32, %al
 ; AVX1-NEXT:    je .LBB3_12
 ; AVX1-NEXT:  # %bb.11: # %cond.load13
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $5, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $5, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_12: # %else14
-; AVX1-NEXT:    vpmovsxdq %xmm6, %xmm6
-; AVX1-NEXT:    vpaddq %xmm7, %xmm4, %xmm5
-; AVX1-NEXT:    vpxor %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpcmpeqb %xmm7, %xmm2, %xmm7
-; AVX1-NEXT:    vpextrb $6, %xmm7, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vpmovsxdq %xmm2, %xmm5
+; AVX1-NEXT:    vpaddq %xmm6, %xmm4, %xmm2
+; AVX1-NEXT:    testb $64, %al
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    je .LBB3_14
 ; AVX1-NEXT:  # %bb.13: # %cond.load16
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $6, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $6, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_14: # %else17
-; AVX1-NEXT:    vpaddq %xmm6, %xmm4, %xmm6
-; AVX1-NEXT:    vpextrb $7, %xmm7, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vpaddq %xmm5, %xmm4, %xmm5
+; AVX1-NEXT:    testb $-128, %al
 ; AVX1-NEXT:    je .LBB3_16
 ; AVX1-NEXT:  # %bb.15: # %cond.load19
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $7, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $7, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_16: # %else20
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm5, %ymm0
-; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm2, %xmm5
-; AVX1-NEXT:    vpextrb $8, %xmm5, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm2, %ymm0
+; AVX1-NEXT:    testl $256, %eax # imm = 0x100
 ; AVX1-NEXT:    je .LBB3_18
 ; AVX1-NEXT:  # %bb.17: # %cond.load22
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $8, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $8, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_18: # %else23
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm1[2,3,0,1]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
 ; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm1
-; AVX1-NEXT:    vpextrb $9, %xmm5, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testl $512, %eax # imm = 0x200
 ; AVX1-NEXT:    je .LBB3_20
 ; AVX1-NEXT:  # %bb.19: # %cond.load25
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $9, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $9, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_20: # %else26
-; AVX1-NEXT:    vpmovsxdq %xmm6, %xmm5
+; AVX1-NEXT:    vpmovsxdq %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
-; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm6
-; AVX1-NEXT:    vpextrb $10, %xmm6, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testl $1024, %eax # imm = 0x400
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    je .LBB3_22
 ; AVX1-NEXT:  # %bb.21: # %cond.load28
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $10, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $10, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_22: # %else29
-; AVX1-NEXT:    vpaddq %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpextrb $11, %xmm6, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vpaddq %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    testl $2048, %eax # imm = 0x800
 ; AVX1-NEXT:    je .LBB3_24
 ; AVX1-NEXT:  # %bb.23: # %cond.load31
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $11, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $11, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_24: # %else32
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm0
+; AVX1-NEXT:    testl $4096, %eax # imm = 0x1000
 ; AVX1-NEXT:    je .LBB3_26
 ; AVX1-NEXT:  # %bb.25: # %cond.load34
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $12, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $12, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_26: # %else35
-; AVX1-NEXT:    vpextrb $13, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testl $8192, %eax # imm = 0x2000
 ; AVX1-NEXT:    je .LBB3_28
 ; AVX1-NEXT:  # %bb.27: # %cond.load37
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $13, (%rax), %xmm3, %xmm3
+; AVX1-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $13, (%rcx), %xmm3, %xmm3
 ; AVX1-NEXT:  .LBB3_28: # %else38
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    testl $16384, %eax # imm = 0x4000
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    je .LBB3_30
-; AVX1-NEXT:  # %bb.29: # %cond.load40
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vpinsrb $14, (%rax), %xmm3, %xmm3
-; AVX1-NEXT:  .LBB3_30: # %else41
-; AVX1-NEXT:    vpextrb $15, %xmm1, %eax
-; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    jne .LBB3_29
+; AVX1-NEXT:  # %bb.30: # %else41
+; AVX1-NEXT:    testl $32768, %eax # imm = 0x8000
+; AVX1-NEXT:    jne .LBB3_31
+; AVX1-NEXT:  .LBB3_32: # %else44
+; AVX1-NEXT:    vmovdqa %xmm3, %xmm0
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+; AVX1-NEXT:  .LBB3_29: # %cond.load40
+; AVX1-NEXT:    vmovq %xmm0, %rcx
+; AVX1-NEXT:    vpinsrb $14, (%rcx), %xmm3, %xmm3
+; AVX1-NEXT:    testl $32768, %eax # imm = 0x8000
 ; AVX1-NEXT:    je .LBB3_32
-; AVX1-NEXT:  # %bb.31: # %cond.load43
+; AVX1-NEXT:  .LBB3_31: # %cond.load43
 ; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX1-NEXT:    vpinsrb $15, (%rax), %xmm3, %xmm3
-; AVX1-NEXT:  .LBB3_32: # %else44
 ; AVX1-NEXT:    vmovdqa %xmm3, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: gather_v16i8_v16i32_v16i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vmovq %rdi, %xmm4
 ; AVX2-NEXT:    vpbroadcastq %xmm4, %ymm4
 ; AVX2-NEXT:    vpmovsxdq %xmm0, %ymm5
 ; AVX2-NEXT:    vpaddq %ymm5, %ymm4, %ymm5
 ; AVX2-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX2-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm6
-; AVX2-NEXT:    vpextrb $0, %xmm6, %eax
+; AVX2-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovmskb %xmm2, %eax
 ; AVX2-NEXT:    testb $1, %al
 ; AVX2-NEXT:    je .LBB3_2
 ; AVX2-NEXT:  # %bb.1: # %cond.load
-; AVX2-NEXT:    vmovq %xmm5, %rax
-; AVX2-NEXT:    vpinsrb $0, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vmovq %xmm5, %rcx
+; AVX2-NEXT:    vpinsrb $0, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_2: # %else
-; AVX2-NEXT:    vpextrb $1, %xmm6, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $2, %al
 ; AVX2-NEXT:    je .LBB3_4
 ; AVX2-NEXT:  # %bb.3: # %cond.load1
-; AVX2-NEXT:    vpextrq $1, %xmm5, %rax
-; AVX2-NEXT:    vpinsrb $1, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vpextrq $1, %xmm5, %rcx
+; AVX2-NEXT:    vpinsrb $1, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_4: # %else2
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX2-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm6
-; AVX2-NEXT:    vpextrb $2, %xmm6, %eax
-; AVX2-NEXT:    testb $1, %al
-; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX2-NEXT:    testb $4, %al
+; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm0
 ; AVX2-NEXT:    je .LBB3_6
 ; AVX2-NEXT:  # %bb.5: # %cond.load4
-; AVX2-NEXT:    vmovq %xmm5, %rax
-; AVX2-NEXT:    vpinsrb $2, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $2, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_6: # %else5
-; AVX2-NEXT:    vpmovsxdq %xmm0, %ymm0
-; AVX2-NEXT:    vpextrb $3, %xmm6, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    vpmovsxdq %xmm2, %ymm2
+; AVX2-NEXT:    testb $8, %al
 ; AVX2-NEXT:    je .LBB3_8
 ; AVX2-NEXT:  # %bb.7: # %cond.load7
-; AVX2-NEXT:    vpextrq $1, %xmm5, %rax
-; AVX2-NEXT:    vpinsrb $3, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $3, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_8: # %else8
-; AVX2-NEXT:    vpaddq %ymm0, %ymm4, %ymm0
-; AVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpcmpeqb %xmm5, %xmm2, %xmm5
-; AVX2-NEXT:    vpextrb $4, %xmm5, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    vpaddq %ymm2, %ymm4, %ymm0
+; AVX2-NEXT:    testb $16, %al
 ; AVX2-NEXT:    je .LBB3_10
 ; AVX2-NEXT:  # %bb.9: # %cond.load10
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $4, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $4, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_10: # %else11
-; AVX2-NEXT:    vpextrb $5, %xmm5, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $32, %al
 ; AVX2-NEXT:    je .LBB3_12
 ; AVX2-NEXT:  # %bb.11: # %cond.load13
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $5, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $5, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_12: # %else14
-; AVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpcmpeqb %xmm5, %xmm2, %xmm5
-; AVX2-NEXT:    vpextrb $6, %xmm5, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testb $64, %al
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX2-NEXT:    je .LBB3_14
 ; AVX2-NEXT:  # %bb.13: # %cond.load16
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $6, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $6, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_14: # %else17
-; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm6
-; AVX2-NEXT:    vpextrb $7, %xmm5, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm2
+; AVX2-NEXT:    testb $-128, %al
 ; AVX2-NEXT:    je .LBB3_16
 ; AVX2-NEXT:  # %bb.15: # %cond.load19
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $7, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $7, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_16: # %else20
-; AVX2-NEXT:    vpaddq %ymm6, %ymm4, %ymm0
-; AVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpcmpeqb %xmm5, %xmm2, %xmm5
-; AVX2-NEXT:    vpextrb $8, %xmm5, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    vpaddq %ymm2, %ymm4, %ymm0
+; AVX2-NEXT:    testl $256, %eax # imm = 0x100
 ; AVX2-NEXT:    je .LBB3_18
 ; AVX2-NEXT:  # %bb.17: # %cond.load22
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $8, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $8, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_18: # %else23
-; AVX2-NEXT:    vpextrb $9, %xmm5, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testl $512, %eax # imm = 0x200
 ; AVX2-NEXT:    je .LBB3_20
 ; AVX2-NEXT:  # %bb.19: # %cond.load25
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $9, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $9, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_20: # %else26
 ; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpcmpeqb %xmm5, %xmm2, %xmm5
-; AVX2-NEXT:    vpextrb $10, %xmm5, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testl $1024, %eax # imm = 0x400
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; AVX2-NEXT:    je .LBB3_22
 ; AVX2-NEXT:  # %bb.21: # %cond.load28
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $10, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $10, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_22: # %else29
 ; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm1
-; AVX2-NEXT:    vpextrb $11, %xmm5, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testl $2048, %eax # imm = 0x800
 ; AVX2-NEXT:    je .LBB3_24
 ; AVX2-NEXT:  # %bb.23: # %cond.load31
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $11, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $11, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_24: # %else32
 ; AVX2-NEXT:    vpaddq %ymm1, %ymm4, %ymm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpcmpeqb %xmm1, %xmm2, %xmm1
-; AVX2-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testl $4096, %eax # imm = 0x1000
 ; AVX2-NEXT:    je .LBB3_26
 ; AVX2-NEXT:  # %bb.25: # %cond.load34
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $12, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $12, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_26: # %else35
-; AVX2-NEXT:    vpextrb $13, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testl $8192, %eax # imm = 0x2000
 ; AVX2-NEXT:    je .LBB3_28
 ; AVX2-NEXT:  # %bb.27: # %cond.load37
-; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $13, (%rax), %xmm3, %xmm3
+; AVX2-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $13, (%rcx), %xmm3, %xmm3
 ; AVX2-NEXT:  .LBB3_28: # %else38
-; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpcmpeqb %xmm1, %xmm2, %xmm1
-; AVX2-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    testl $16384, %eax # imm = 0x4000
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT:    je .LBB3_30
-; AVX2-NEXT:  # %bb.29: # %cond.load40
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vpinsrb $14, (%rax), %xmm3, %xmm3
-; AVX2-NEXT:  .LBB3_30: # %else41
-; AVX2-NEXT:    vpextrb $15, %xmm1, %eax
-; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    jne .LBB3_29
+; AVX2-NEXT:  # %bb.30: # %else41
+; AVX2-NEXT:    testl $32768, %eax # imm = 0x8000
+; AVX2-NEXT:    jne .LBB3_31
+; AVX2-NEXT:  .LBB3_32: # %else44
+; AVX2-NEXT:    vmovdqa %xmm3, %xmm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+; AVX2-NEXT:  .LBB3_29: # %cond.load40
+; AVX2-NEXT:    vmovq %xmm0, %rcx
+; AVX2-NEXT:    vpinsrb $14, (%rcx), %xmm3, %xmm3
+; AVX2-NEXT:    testl $32768, %eax # imm = 0x8000
 ; AVX2-NEXT:    je .LBB3_32
-; AVX2-NEXT:  # %bb.31: # %cond.load43
+; AVX2-NEXT:  .LBB3_31: # %cond.load43
 ; AVX2-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX2-NEXT:    vpinsrb $15, (%rax), %xmm3, %xmm3
-; AVX2-NEXT:  .LBB3_32: # %else44
 ; AVX2-NEXT:    vmovdqa %xmm3, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: gather_v16i8_v16i32_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastq %rdi, %zmm3
 ; AVX512-NEXT:    vpmovsxdq %ymm0, %zmm4
 ; AVX512-NEXT:    vpaddq %zmm4, %zmm3, %zmm4
 ; AVX512-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX512-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm5
-; AVX512-NEXT:    vpmovsxbd %xmm5, %zmm5
-; AVX512-NEXT:    vptestmd %zmm5, %zmm5, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
+; AVX512-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm1
+; AVX512-NEXT:    vpmovmskb %xmm1, %eax
 ; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    je .LBB3_2
-; AVX512-NEXT:  # %bb.1: # %cond.load
-; AVX512-NEXT:    vmovq %xmm4, %rax
-; AVX512-NEXT:    vpinsrb $0, (%rax), %xmm2, %xmm2
-; AVX512-NEXT:  .LBB3_2: # %else
-; AVX512-NEXT:    kshiftrw $1, %k0, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
-; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    je .LBB3_4
-; AVX512-NEXT:  # %bb.3: # %cond.load1
-; AVX512-NEXT:    vpextrq $1, %xmm4, %rax
-; AVX512-NEXT:    vpinsrb $1, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    jne .LBB3_1
+; AVX512-NEXT:  # %bb.2: # %else
+; AVX512-NEXT:    testb $2, %al
+; AVX512-NEXT:    jne .LBB3_3
 ; AVX512-NEXT:  .LBB3_4: # %else2
-; AVX512-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX512-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm5
-; AVX512-NEXT:    vpmovsxbd %xmm5, %zmm5
-; AVX512-NEXT:    vptestmd %zmm5, %zmm5, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    kmovw %k1, %eax
-; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    je .LBB3_6
-; AVX512-NEXT:  # %bb.5: # %cond.load4
-; AVX512-NEXT:    vextracti128 $1, %ymm4, %xmm5
-; AVX512-NEXT:    vmovq %xmm5, %rax
-; AVX512-NEXT:    vpinsrb $2, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    testb $4, %al
+; AVX512-NEXT:    jne .LBB3_5
 ; AVX512-NEXT:  .LBB3_6: # %else5
-; AVX512-NEXT:    kshiftrw $3, %k0, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
-; AVX512-NEXT:    testb $1, %al
+; AVX512-NEXT:    testb $8, %al
 ; AVX512-NEXT:    je .LBB3_8
-; AVX512-NEXT:  # %bb.7: # %cond.load7
-; AVX512-NEXT:    vextracti128 $1, %ymm4, %xmm5
-; AVX512-NEXT:    vpextrq $1, %xmm5, %rax
-; AVX512-NEXT:    vpinsrb $3, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:  .LBB3_7: # %cond.load7
+; AVX512-NEXT:    vextracti128 $1, %ymm4, %xmm1
+; AVX512-NEXT:    vpextrq $1, %xmm1, %rcx
+; AVX512-NEXT:    vpinsrb $3, (%rcx), %xmm2, %xmm2
 ; AVX512-NEXT:  .LBB3_8: # %else8
-; AVX512-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX512-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm5
-; AVX512-NEXT:    vpmovsxbd %xmm5, %zmm5
-; AVX512-NEXT:    vptestmd %zmm5, %zmm5, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    kmovw %k1, %eax
-; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    vextracti32x4 $2, %zmm4, %xmm5
+; AVX512-NEXT:    testb $16, %al
+; AVX512-NEXT:    vextracti32x4 $2, %zmm4, %xmm1
 ; AVX512-NEXT:    je .LBB3_10
 ; AVX512-NEXT:  # %bb.9: # %cond.load10
-; AVX512-NEXT:    vmovq %xmm5, %rax
-; AVX512-NEXT:    vpinsrb $4, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    vmovq %xmm1, %rcx
+; AVX512-NEXT:    vpinsrb $4, (%rcx), %xmm2, %xmm2
 ; AVX512-NEXT:  .LBB3_10: # %else11
-; AVX512-NEXT:    kshiftrw $5, %k0, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
-; AVX512-NEXT:    testb $1, %al
+; AVX512-NEXT:    testb $32, %al
 ; AVX512-NEXT:    je .LBB3_12
 ; AVX512-NEXT:  # %bb.11: # %cond.load13
-; AVX512-NEXT:    vpextrq $1, %xmm5, %rax
-; AVX512-NEXT:    vpinsrb $5, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    vpextrq $1, %xmm1, %rcx
+; AVX512-NEXT:    vpinsrb $5, (%rcx), %xmm2, %xmm2
 ; AVX512-NEXT:  .LBB3_12: # %else14
-; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX512-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm5
-; AVX512-NEXT:    vpmovsxbd %xmm5, %zmm5
-; AVX512-NEXT:    vptestmd %zmm5, %zmm5, %k0
-; AVX512-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512-NEXT:    kmovw %k1, %eax
-; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    vextracti32x4 $3, %zmm4, %xmm4
+; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512-NEXT:    testb $64, %al
+; AVX512-NEXT:    vextracti32x4 $3, %zmm4, %xmm0
 ; AVX512-NEXT:    je .LBB3_14
 ; AVX512-NEXT:  # %bb.13: # %cond.load16
-; AVX512-NEXT:    vmovq %xmm4, %rax
-; AVX512-NEXT:    vpinsrb $6, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    vmovq %xmm0, %rcx
+; AVX512-NEXT:    vpinsrb $6, (%rcx), %xmm2, %xmm2
 ; AVX512-NEXT:  .LBB3_14: # %else17
-; AVX512-NEXT:    vpmovsxdq %ymm0, %zmm0
-; AVX512-NEXT:    kshiftrw $7, %k0, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
-; AVX512-NEXT:    testb $1, %al
+; AVX512-NEXT:    vpmovsxdq %ymm1, %zmm1
+; AVX512-NEXT:    testb $-128, %al
 ; AVX512-NEXT:    je .LBB3_16
 ; AVX512-NEXT:  # %bb.15: # %cond.load19
-; AVX512-NEXT:    vpextrq $1, %xmm4, %rax
-; AVX512-NEXT:    vpinsrb $7, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX512-NEXT:    vpinsrb $7, (%rcx), %xmm2, %xmm2
 ; AVX512-NEXT:  .LBB3_16: # %else20
-; AVX512-NEXT:    vpaddq %zmm0, %zmm3, %zmm0
-; AVX512-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512-NEXT:    vpcmpeqb %xmm3, %xmm1, %xmm3
-; AVX512-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    kmovw %k1, %eax
-; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    je .LBB3_18
-; AVX512-NEXT:  # %bb.17: # %cond.load22
-; AVX512-NEXT:    vmovq %xmm0, %rax
-; AVX512-NEXT:    vpinsrb $8, (%rax), %xmm2, %xmm2
-; AVX512-NEXT:  .LBB3_18: # %else23
-; AVX512-NEXT:    kshiftrw $9, %k0, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
-; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    je .LBB3_20
-; AVX512-NEXT:  # %bb.19: # %cond.load25
-; AVX512-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX512-NEXT:    vpinsrb $9, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    vpaddq %zmm1, %zmm3, %zmm0
+; AVX512-NEXT:    testl $256, %eax # imm = 0x100
+; AVX512-NEXT:    jne .LBB3_17
+; AVX512-NEXT:  # %bb.18: # %else23
+; AVX512-NEXT:    testl $512, %eax # imm = 0x200
+; AVX512-NEXT:    jne .LBB3_19
 ; AVX512-NEXT:  .LBB3_20: # %else26
-; AVX512-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512-NEXT:    vpcmpeqb %xmm3, %xmm1, %xmm3
-; AVX512-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512-NEXT:    kshiftrw $10, %k0, %k1
-; AVX512-NEXT:    kmovw %k1, %eax
-; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    je .LBB3_22
-; AVX512-NEXT:  # %bb.21: # %cond.load28
-; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; AVX512-NEXT:    vmovq %xmm3, %rax
-; AVX512-NEXT:    vpinsrb $10, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    testl $1024, %eax # imm = 0x400
+; AVX512-NEXT:    jne .LBB3_21
 ; AVX512-NEXT:  .LBB3_22: # %else29
-; AVX512-NEXT:    kshiftrw $11, %k0, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
-; AVX512-NEXT:    testb $1, %al
+; AVX512-NEXT:    testl $2048, %eax # imm = 0x800
 ; AVX512-NEXT:    je .LBB3_24
-; AVX512-NEXT:  # %bb.23: # %cond.load31
-; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; AVX512-NEXT:    vpextrq $1, %xmm3, %rax
-; AVX512-NEXT:    vpinsrb $11, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:  .LBB3_23: # %cond.load31
+; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX512-NEXT:    vpextrq $1, %xmm1, %rcx
+; AVX512-NEXT:    vpinsrb $11, (%rcx), %xmm2, %xmm2
 ; AVX512-NEXT:  .LBB3_24: # %else32
-; AVX512-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512-NEXT:    vpcmpeqb %xmm3, %xmm1, %xmm3
-; AVX512-NEXT:    vpmovsxbd %xmm3, %zmm3
-; AVX512-NEXT:    vptestmd %zmm3, %zmm3, %k0
-; AVX512-NEXT:    kshiftrw $12, %k0, %k1
-; AVX512-NEXT:    kmovw %k1, %eax
-; AVX512-NEXT:    testb $1, %al
-; AVX512-NEXT:    vextracti32x4 $2, %zmm0, %xmm3
+; AVX512-NEXT:    testl $4096, %eax # imm = 0x1000
+; AVX512-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
 ; AVX512-NEXT:    je .LBB3_26
 ; AVX512-NEXT:  # %bb.25: # %cond.load34
-; AVX512-NEXT:    vmovq %xmm3, %rax
-; AVX512-NEXT:    vpinsrb $12, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    vmovq %xmm1, %rcx
+; AVX512-NEXT:    vpinsrb $12, (%rcx), %xmm2, %xmm2
 ; AVX512-NEXT:  .LBB3_26: # %else35
-; AVX512-NEXT:    kshiftrw $13, %k0, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
-; AVX512-NEXT:    testb $1, %al
+; AVX512-NEXT:    testl $8192, %eax # imm = 0x2000
 ; AVX512-NEXT:    je .LBB3_28
 ; AVX512-NEXT:  # %bb.27: # %cond.load37
-; AVX512-NEXT:    vpextrq $1, %xmm3, %rax
-; AVX512-NEXT:    vpinsrb $13, (%rax), %xmm2, %xmm2
+; AVX512-NEXT:    vpextrq $1, %xmm1, %rcx
+; AVX512-NEXT:    vpinsrb $13, (%rcx), %xmm2, %xmm2
 ; AVX512-NEXT:  .LBB3_28: # %else38
-; AVX512-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512-NEXT:    vpcmpeqb %xmm3, %xmm1, %xmm1
-; AVX512-NEXT:    vpmovsxbd %xmm1, %zmm1
-; AVX512-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512-NEXT:    kshiftrw $14, %k0, %k1
-; AVX512-NEXT:    kmovw %k1, %eax
-; AVX512-NEXT:    testb $1, %al
+; AVX512-NEXT:    testl $16384, %eax # imm = 0x4000
 ; AVX512-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512-NEXT:    je .LBB3_30
-; AVX512-NEXT:  # %bb.29: # %cond.load40
-; AVX512-NEXT:    vmovq %xmm0, %rax
-; AVX512-NEXT:    vpinsrb $14, (%rax), %xmm2, %xmm2
-; AVX512-NEXT:  .LBB3_30: # %else41
-; AVX512-NEXT:    kshiftrw $15, %k0, %k0
-; AVX512-NEXT:    kmovw %k0, %eax
-; AVX512-NEXT:    testb $1, %al
+; AVX512-NEXT:    jne .LBB3_29
+; AVX512-NEXT:  # %bb.30: # %else41
+; AVX512-NEXT:    testl $32768, %eax # imm = 0x8000
+; AVX512-NEXT:    jne .LBB3_31
+; AVX512-NEXT:  .LBB3_32: # %else44
+; AVX512-NEXT:    vmovdqa %xmm2, %xmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
+; AVX512-NEXT:  .LBB3_1: # %cond.load
+; AVX512-NEXT:    vmovq %xmm4, %rcx
+; AVX512-NEXT:    vpinsrb $0, (%rcx), %xmm2, %xmm2
+; AVX512-NEXT:    testb $2, %al
+; AVX512-NEXT:    je .LBB3_4
+; AVX512-NEXT:  .LBB3_3: # %cond.load1
+; AVX512-NEXT:    vpextrq $1, %xmm4, %rcx
+; AVX512-NEXT:    vpinsrb $1, (%rcx), %xmm2, %xmm2
+; AVX512-NEXT:    testb $4, %al
+; AVX512-NEXT:    je .LBB3_6
+; AVX512-NEXT:  .LBB3_5: # %cond.load4
+; AVX512-NEXT:    vextracti128 $1, %ymm4, %xmm1
+; AVX512-NEXT:    vmovq %xmm1, %rcx
+; AVX512-NEXT:    vpinsrb $2, (%rcx), %xmm2, %xmm2
+; AVX512-NEXT:    testb $8, %al
+; AVX512-NEXT:    jne .LBB3_7
+; AVX512-NEXT:    jmp .LBB3_8
+; AVX512-NEXT:  .LBB3_17: # %cond.load22
+; AVX512-NEXT:    vmovq %xmm0, %rcx
+; AVX512-NEXT:    vpinsrb $8, (%rcx), %xmm2, %xmm2
+; AVX512-NEXT:    testl $512, %eax # imm = 0x200
+; AVX512-NEXT:    je .LBB3_20
+; AVX512-NEXT:  .LBB3_19: # %cond.load25
+; AVX512-NEXT:    vpextrq $1, %xmm0, %rcx
+; AVX512-NEXT:    vpinsrb $9, (%rcx), %xmm2, %xmm2
+; AVX512-NEXT:    testl $1024, %eax # imm = 0x400
+; AVX512-NEXT:    je .LBB3_22
+; AVX512-NEXT:  .LBB3_21: # %cond.load28
+; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX512-NEXT:    vmovq %xmm1, %rcx
+; AVX512-NEXT:    vpinsrb $10, (%rcx), %xmm2, %xmm2
+; AVX512-NEXT:    testl $2048, %eax # imm = 0x800
+; AVX512-NEXT:    jne .LBB3_23
+; AVX512-NEXT:    jmp .LBB3_24
+; AVX512-NEXT:  .LBB3_29: # %cond.load40
+; AVX512-NEXT:    vmovq %xmm0, %rcx
+; AVX512-NEXT:    vpinsrb $14, (%rcx), %xmm2, %xmm2
+; AVX512-NEXT:    testl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    je .LBB3_32
-; AVX512-NEXT:  # %bb.31: # %cond.load43
+; AVX512-NEXT:  .LBB3_31: # %cond.load43
 ; AVX512-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX512-NEXT:    vpinsrb $15, (%rax), %xmm2, %xmm2
-; AVX512-NEXT:  .LBB3_32: # %else44
 ; AVX512-NEXT:    vmovdqa %xmm2, %xmm0
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %vptr0 = insertelement <16 x i8*> undef, i8* %base, i32 0
   %vptr1 = shufflevector <16 x i8*> %vptr0, <16 x i8*> undef, <16 x i32> zeroinitializer
   %vptr2 = getelementptr i8, <16 x i8*> %vptr1, <16 x i32> %idx
 
   %mask = icmp eq <16 x i8> %trigger, zeroinitializer

llvm/trunk/test/CodeGen/X86/masked_gather_scatter.ll

 
 ; SCALAR-LABEL: test2
 ; SCALAR:      extractelement <16 x float*>
 ; SCALAR-NEXT: load float
 ; SCALAR-NEXT: insertelement <16 x float>
 ; SCALAR-NEXT: br label %else
 ; SCALAR: else:
 ; SCALAR-NEXT:  %res.phi.else = phi
-; SCALAR-NEXT:  %Mask1 = extractelement <16 x i1> %imask, i64 1
-; SCALAR-NEXT:  br i1 %Mask1, label %cond.load1, label %else2
+; SCALAR-NEXT:  and i16 %{{.*}}, 2
+; SCALAR-NEXT:  icmp ne i16 %{{.*}}, 0
+; SCALAR-NEXT:  br i1 %{{.*}}, label %cond.load1, label %else2
 
 define <16 x float> @test2(float* %base, <16 x i32> %ind, i16 %mask) {
 ; KNL_64-LABEL: test2:
 ; KNL_64:       # %bb.0:
 ; KNL_64-NEXT:    kmovw %esi, %k1
 ; KNL_64-NEXT:    vgatherdps (%rdi,%zmm0,4), %zmm1 {%k1}
 ; KNL_64-NEXT:    vmovaps %zmm1, %zmm0
 ; KNL_64-NEXT:    retq
; SKX_32-NEXT:    retl
   %gt1 = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> %gep.random, i32 4, <16 x i1> %imask, <16 x i32>undef)
   %gt2 = call <16 x i32> @llvm.masked.gather.v16i32.v16p0i32(<16 x i32*> %gep.random, i32 4, <16 x i1> %imask, <16 x i32>%gt1)
   %res = add <16 x i32> %gt1, %gt2
   ret <16 x i32> %res
 }
 
 
 ; SCALAR-LABEL: test5
-; SCALAR:        %Mask0 = extractelement <16 x i1> %imask, i64 0
-; SCALAR-NEXT:   br i1 %Mask0, label %cond.store, label %else
+; SCALAR:        and i16 %scalar_mask, 1
+; SCALAR-NEXT:   icmp ne i16 %{{.*}}, 0
+; SCALAR-NEXT:   br i1 %{{.*}}, label %cond.store, label %else
 ; SCALAR: cond.store:
 ; SCALAR-NEXT:  %Elt0 = extractelement <16 x i32> %val, i64 0
 ; SCALAR-NEXT:  %Ptr0 = extractelement <16 x i32*> %gep.random, i64 0
 ; SCALAR-NEXT:  store i32 %Elt0, i32* %Ptr0, align 4
 ; SCALAR-NEXT:  br label %else
 ; SCALAR: else:
-; SCALAR-NEXT: %Mask1 = extractelement <16 x i1> %imask, i64 1
-; SCALAR-NEXT:  br i1 %Mask1, label %cond.store1, label %else2
+; SCALAR-NEXT:   and i16 %scalar_mask, 2
+; SCALAR-NEXT:   icmp ne i16 %{{.*}}, 0
+; SCALAR-NEXT:  br i1 %{{.*}}, label %cond.store1, label %else2
 
 define void @test5(i32* %base, <16 x i32> %ind, i16 %mask, <16 x i32>%val) {
 ; KNL_64-LABEL: test5:
 ; KNL_64:       # %bb.0:
 ; KNL_64-NEXT:    kmovw %esi, %k1
 ; KNL_64-NEXT:    kmovw %k1, %k2
 ; KNL_64-NEXT:    vpscatterdd %zmm1, (%rdi,%zmm0,4) {%k2}
 ; KNL_64-NEXT:    vpscatterdd %zmm1, (%rdi,%zmm0,4) {%k1}

 
 ; Check non-power-of-2 case. It should be scalarized.
 declare <3 x i32> @llvm.masked.gather.v3i32.v3p0i32(<3 x i32*>, i32, <3 x i1>, <3 x i32>)
 define <3 x i32> @test30(<3 x i32*> %base, <3 x i32> %ind, <3 x i1> %mask, <3 x i32> %src0) {
 ; KNL_64-LABEL: test30:
 ; KNL_64:       # %bb.0:
 ; KNL_64-NEXT:    vpslld $31, %xmm2, %xmm2
 ; KNL_64-NEXT:    vptestmd %zmm2, %zmm2, %k0
+; KNL_64-NEXT:    kshiftrw $1, %k0, %k1
+; KNL_64-NEXT:    kshiftrw $2, %k0, %k2
 ; KNL_64-NEXT:    kmovw %k0, %eax
+; KNL_64-NEXT:    andb $1, %al
+; KNL_64-NEXT:    kmovw %k1, %ecx
+; KNL_64-NEXT:    andb $1, %cl
+; KNL_64-NEXT:    addb %cl, %cl
+; KNL_64-NEXT:    orb %al, %cl
+; KNL_64-NEXT:    kmovw %k2, %eax
+; KNL_64-NEXT:    andb $1, %al
+; KNL_64-NEXT:    shlb $2, %al
+; KNL_64-NEXT:    orb %cl, %al
 ; KNL_64-NEXT:    vpmovsxdq %xmm1, %ymm1
 ; KNL_64-NEXT:    vpsllq $2, %ymm1, %ymm1
 ; KNL_64-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
 ; KNL_64-NEXT:    testb $1, %al
-; KNL_64-NEXT:    je .LBB31_2
-; KNL_64-NEXT:  # %bb.1: # %cond.load
-; KNL_64-NEXT:    vmovq %xmm0, %rax
-; KNL_64-NEXT:    vpinsrd $0, (%rax), %xmm3, %xmm3
-; KNL_64-NEXT:  .LBB31_2: # %else
-; KNL_64-NEXT:    kshiftrw $1, %k0, %k1
-; KNL_64-NEXT:    kmovw %k1, %eax
-; KNL_64-NEXT:    testb $1, %al
-; KNL_64-NEXT:    je .LBB31_4
-; KNL_64-NEXT:  # %bb.3: # %cond.load1
-; KNL_64-NEXT:    vpextrq $1, %xmm0, %rax
-; KNL_64-NEXT:    vpinsrd $1, (%rax), %xmm3, %xmm3
+; KNL_64-NEXT:    jne .LBB31_1
+; KNL_64-NEXT:  # %bb.2: # %else
+; KNL_64-NEXT:    testb $2, %al
+; KNL_64-NEXT:    jne .LBB31_3
 ; KNL_64-NEXT:  .LBB31_4: # %else2
-; KNL_64-NEXT:    kshiftrw $2, %k0, %k0
-; KNL_64-NEXT:    kmovw %k0, %eax
-; KNL_64-NEXT:    testb $1, %al
+; KNL_64-NEXT:    testb $4, %al
+; KNL_64-NEXT:    jne .LBB31_5
+; KNL_64-NEXT:  .LBB31_6: # %else5
+; KNL_64-NEXT:    vmovdqa %xmm3, %xmm0
+; KNL_64-NEXT:    vzeroupper
+; KNL_64-NEXT:    retq
+; KNL_64-NEXT:  .LBB31_1: # %cond.load
+; KNL_64-NEXT:    vmovq %xmm0, %rcx
+; KNL_64-NEXT:    vpinsrd $0, (%rcx), %xmm3, %xmm3
+; KNL_64-NEXT:    testb $2, %al
+; KNL_64-NEXT:    je .LBB31_4
+; KNL_64-NEXT:  .LBB31_3: # %cond.load1
+; KNL_64-NEXT:    vpextrq $1, %xmm0, %rcx
+; KNL_64-NEXT:    vpinsrd $1, (%rcx), %xmm3, %xmm3
+; KNL_64-NEXT:    testb $4, %al
 ; KNL_64-NEXT:    je .LBB31_6
-; KNL_64-NEXT:  # %bb.5: # %cond.load4
+; KNL_64-NEXT:  .LBB31_5: # %cond.load4
 ; KNL_64-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL_64-NEXT:    vmovq %xmm0, %rax
 ; KNL_64-NEXT:    vpinsrd $2, (%rax), %xmm3, %xmm3
-; KNL_64-NEXT:  .LBB31_6: # %else5
 ; KNL_64-NEXT:    vmovdqa %xmm3, %xmm0
 ; KNL_64-NEXT:    vzeroupper
 ; KNL_64-NEXT:    retq
 ;
 ; KNL_32-LABEL: test30:
 ; KNL_32:       # %bb.0:
 ; KNL_32-NEXT:    subl $12, %esp
 ; KNL_32-NEXT:    .cfi_def_cfa_offset 16
 ; KNL_32-NEXT:    vmovdqa %xmm0, %xmm3
 ; KNL_32-NEXT:    vpslld $31, %xmm2, %xmm0
 ; KNL_32-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; KNL_32-NEXT:    kshiftrw $1, %k0, %k1
+; KNL_32-NEXT:    kshiftrw $2, %k0, %k2
 ; KNL_32-NEXT:    kmovw %k0, %eax
+; KNL_32-NEXT:    andb $1, %al
+; KNL_32-NEXT:    kmovw %k1, %ecx
+; KNL_32-NEXT:    andb $1, %cl
+; KNL_32-NEXT:    addb %cl, %cl
+; KNL_32-NEXT:    orb %al, %cl
+; KNL_32-NEXT:    kmovw %k2, %eax
+; KNL_32-NEXT:    andb $1, %al
+; KNL_32-NEXT:    shlb $2, %al
+; KNL_32-NEXT:    orb %cl, %al
 ; KNL_32-NEXT:    vmovdqa {{[0-9]+}}(%esp), %xmm0
 ; KNL_32-NEXT:    vpslld $2, %xmm1, %xmm1
 ; KNL_32-NEXT:    vpaddd %xmm1, %xmm3, %xmm1
 ; KNL_32-NEXT:    testb $1, %al
-; KNL_32-NEXT:    je .LBB31_2
-; KNL_32-NEXT:  # %bb.1: # %cond.load
-; KNL_32-NEXT:    vmovd %xmm1, %eax
-; KNL_32-NEXT:    vpinsrd $0, (%eax), %xmm0, %xmm0
-; KNL_32-NEXT:  .LBB31_2: # %else
-; KNL_32-NEXT:    kshiftrw $1, %k0, %k1
-; KNL_32-NEXT:    kmovw %k1, %eax
-; KNL_32-NEXT:    testb $1, %al
-; KNL_32-NEXT:    je .LBB31_4
-; KNL_32-NEXT:  # %bb.3: # %cond.load1
-; KNL_32-NEXT:    vpextrd $1, %xmm1, %eax
-; KNL_32-NEXT:    vpinsrd $1, (%eax), %xmm0, %xmm0
+; KNL_32-NEXT:    jne .LBB31_1
+; KNL_32-NEXT:  # %bb.2: # %else
+; KNL_32-NEXT:    testb $2, %al
+; KNL_32-NEXT:    jne .LBB31_3
 ; KNL_32-NEXT:  .LBB31_4: # %else2
-; KNL_32-NEXT:    kshiftrw $2, %k0, %k0
-; KNL_32-NEXT:    kmovw %k0, %eax
-; KNL_32-NEXT:    testb $1, %al
+; KNL_32-NEXT:    testb $4, %al
+; KNL_32-NEXT:    jne .LBB31_5
+; KNL_32-NEXT:  .LBB31_6: # %else5
+; KNL_32-NEXT:    addl $12, %esp
+; KNL_32-NEXT:    .cfi_def_cfa_offset 4
+; KNL_32-NEXT:    vzeroupper
+; KNL_32-NEXT:    retl
+; KNL_32-NEXT:  .LBB31_1: # %cond.load
+; KNL_32-NEXT:    .cfi_def_cfa_offset 16
+; KNL_32-NEXT:    vmovd %xmm1, %ecx
+; KNL_32-NEXT:    vpinsrd $0, (%ecx), %xmm0, %xmm0
+; KNL_32-NEXT:    testb $2, %al
+; KNL_32-NEXT:    je .LBB31_4
+; KNL_32-NEXT:  .LBB31_3: # %cond.load1
+; KNL_32-NEXT:    vpextrd $1, %xmm1, %ecx
+; KNL_32-NEXT:    vpinsrd $1, (%ecx), %xmm0, %xmm0
+; KNL_32-NEXT:    testb $4, %al
 ; KNL_32-NEXT:    je .LBB31_6
-; KNL_32-NEXT:  # %bb.5: # %cond.load4
+; KNL_32-NEXT:  .LBB31_5: # %cond.load4
 ; KNL_32-NEXT:    vpextrd $2, %xmm1, %eax
 ; KNL_32-NEXT:    vpinsrd $2, (%eax), %xmm0, %xmm0
-; KNL_32-NEXT:  .LBB31_6: # %else5
 ; KNL_32-NEXT:    addl $12, %esp
 ; KNL_32-NEXT:    .cfi_def_cfa_offset 4
 ; KNL_32-NEXT:    vzeroupper
 ; KNL_32-NEXT:    retl
 ;
 ; SKX-LABEL: test30:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpslld $31, %xmm2, %xmm2
 ; SKX-NEXT:    vpmovd2m %xmm2, %k0
+; SKX-NEXT:    kshiftrb $1, %k0, %k1
+; SKX-NEXT:    kshiftrb $2, %k0, %k2
 ; SKX-NEXT:    kmovw %k0, %eax
+; SKX-NEXT:    andb $1, %al
+; SKX-NEXT:    kmovw %k1, %ecx
+; SKX-NEXT:    andb $1, %cl
+; SKX-NEXT:    addb %cl, %cl
+; SKX-NEXT:    orb %al, %cl
+; SKX-NEXT:    kmovw %k2, %eax
+; SKX-NEXT:    andb $1, %al
+; SKX-NEXT:    shlb $2, %al
+; SKX-NEXT:    orb %cl, %al
 ; SKX-NEXT:    vpmovsxdq %xmm1, %ymm1
 ; SKX-NEXT:    vpsllq $2, %ymm1, %ymm1
 ; SKX-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
 ; SKX-NEXT:    testb $1, %al
-; SKX-NEXT:    je .LBB31_2
-; SKX-NEXT:  # %bb.1: # %cond.load
-; SKX-NEXT:    vmovq %xmm0, %rax
-; SKX-NEXT:    vpinsrd $0, (%rax), %xmm3, %xmm3
-; SKX-NEXT:  .LBB31_2: # %else
-; SKX-NEXT:    kshiftrb $1, %k0, %k1
-; SKX-NEXT:    kmovw %k1, %eax
-; SKX-NEXT:    testb $1, %al
-; SKX-NEXT:    je .LBB31_4
-; SKX-NEXT:  # %bb.3: # %cond.load1
-; SKX-NEXT:    vpextrq $1, %xmm0, %rax
-; SKX-NEXT:    vpinsrd $1, (%rax), %xmm3, %xmm3
+; SKX-NEXT:    jne .LBB31_1
+; SKX-NEXT:  # %bb.2: # %else
+; SKX-NEXT:    testb $2, %al
+; SKX-NEXT:    jne .LBB31_3
 ; SKX-NEXT:  .LBB31_4: # %else2
-; SKX-NEXT:    kshiftrb $2, %k0, %k0
-; SKX-NEXT:    kmovw %k0, %eax
-; SKX-NEXT:    testb $1, %al
+; SKX-NEXT:    testb $4, %al
+; SKX-NEXT:    jne .LBB31_5
+; SKX-NEXT:  .LBB31_6: # %else5
+; SKX-NEXT:    vmovdqa %xmm3, %xmm0
+; SKX-NEXT:    vzeroupper
+; SKX-NEXT:    retq
+; SKX-NEXT:  .LBB31_1: # %cond.load
+; SKX-NEXT:    vmovq %xmm0, %rcx
+; SKX-NEXT:    vpinsrd $0, (%rcx), %xmm3, %xmm3
+; SKX-NEXT:    testb $2, %al
+; SKX-NEXT:    je .LBB31_4
+; SKX-NEXT:  .LBB31_3: # %cond.load1
+; SKX-NEXT:    vpextrq $1, %xmm0, %rcx
+; SKX-NEXT:    vpinsrd $1, (%rcx), %xmm3, %xmm3
+; SKX-NEXT:    testb $4, %al
 ; SKX-NEXT:    je .LBB31_6
-; SKX-NEXT:  # %bb.5: # %cond.load4
+; SKX-NEXT:  .LBB31_5: # %cond.load4
 ; SKX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; SKX-NEXT:    vmovq %xmm0, %rax
 ; SKX-NEXT:    vpinsrd $2, (%rax), %xmm3, %xmm3
-; SKX-NEXT:  .LBB31_6: # %else5
 ; SKX-NEXT:    vmovdqa %xmm3, %xmm0
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
 ;
 ; SKX_32-LABEL: test30:
 ; SKX_32:       # %bb.0:
 ; SKX_32-NEXT:    subl $12, %esp
 ; SKX_32-NEXT:    .cfi_def_cfa_offset 16
 ; SKX_32-NEXT:    vmovdqa %xmm0, %xmm3
 ; SKX_32-NEXT:    vpslld $31, %xmm2, %xmm0
 ; SKX_32-NEXT:    vpmovd2m %xmm0, %k0
+; SKX_32-NEXT:    kshiftrb $1, %k0, %k1
+; SKX_32-NEXT:    kshiftrb $2, %k0, %k2
 ; SKX_32-NEXT:    kmovw %k0, %eax
+; SKX_32-NEXT:    andb $1, %al
+; SKX_32-NEXT:    kmovw %k1, %ecx
+; SKX_32-NEXT:    andb $1, %cl
+; SKX_32-NEXT:    addb %cl, %cl
+; SKX_32-NEXT:    orb %al, %cl
+; SKX_32-NEXT:    kmovw %k2, %eax
+; SKX_32-NEXT:    andb $1, %al
+; SKX_32-NEXT:    shlb $2, %al
+; SKX_32-NEXT:    orb %cl, %al
 ; SKX_32-NEXT:    vmovdqa {{[0-9]+}}(%esp), %xmm0
 ; SKX_32-NEXT:    vpslld $2, %xmm1, %xmm1
 ; SKX_32-NEXT:    vpaddd %xmm1, %xmm3, %xmm1
 ; SKX_32-NEXT:    testb $1, %al
-; SKX_32-NEXT:    je .LBB31_2
-; SKX_32-NEXT:  # %bb.1: # %cond.load
-; SKX_32-NEXT:    vmovd %xmm1, %eax
-; SKX_32-NEXT:    vpinsrd $0, (%eax), %xmm0, %xmm0
-; SKX_32-NEXT:  .LBB31_2: # %else
-; SKX_32-NEXT:    kshiftrb $1, %k0, %k1
-; SKX_32-NEXT:    kmovw %k1, %eax
-; SKX_32-NEXT:    testb $1, %al
-; SKX_32-NEXT:    je .LBB31_4
-; SKX_32-NEXT:  # %bb.3: # %cond.load1
-; SKX_32-NEXT:    vpextrd $1, %xmm1, %eax
-; SKX_32-NEXT:    vpinsrd $1, (%eax), %xmm0, %xmm0
+; SKX_32-NEXT:    jne .LBB31_1
+; SKX_32-NEXT:  # %bb.2: # %else
+; SKX_32-NEXT:    testb $2, %al
+; SKX_32-NEXT:    jne .LBB31_3
 ; SKX_32-NEXT:  .LBB31_4: # %else2
-; SKX_32-NEXT:    kshiftrb $2, %k0, %k0
-; SKX_32-NEXT:    kmovw %k0, %eax
-; SKX_32-NEXT:    testb $1, %al
+; SKX_32-NEXT:    testb $4, %al
+; SKX_32-NEXT:    jne .LBB31_5
+; SKX_32-NEXT:  .LBB31_6: # %else5
+; SKX_32-NEXT:    addl $12, %esp
+; SKX_32-NEXT:    .cfi_def_cfa_offset 4
+; SKX_32-NEXT:    retl
+; SKX_32-NEXT:  .LBB31_1: # %cond.load
+; SKX_32-NEXT:    .cfi_def_cfa_offset 16
+; SKX_32-NEXT:    vmovd %xmm1, %ecx
+; SKX_32-NEXT:    vpinsrd $0, (%ecx), %xmm0, %xmm0
+; SKX_32-NEXT:    testb $2, %al
+; SKX_32-NEXT:    je .LBB31_4
+; SKX_32-NEXT:  .LBB31_3: # %cond.load1
+; SKX_32-NEXT:    vpextrd $1, %xmm1, %ecx
+; SKX_32-NEXT:    vpinsrd $1, (%ecx), %xmm0, %xmm0
+; SKX_32-NEXT:    testb $4, %al
 ; SKX_32-NEXT:    je .LBB31_6
-; SKX_32-NEXT:  # %bb.5: # %cond.load4
+; SKX_32-NEXT:  .LBB31_5: # %cond.load4
 ; SKX_32-NEXT:    vpextrd $2, %xmm1, %eax
 ; SKX_32-NEXT:    vpinsrd $2, (%eax), %xmm0, %xmm0
-; SKX_32-NEXT:  .LBB31_6: # %else5
 ; SKX_32-NEXT:    addl $12, %esp
 ; SKX_32-NEXT:    .cfi_def_cfa_offset 4
 ; SKX_32-NEXT:    retl
 
   %sext_ind = sext <3 x i32> %ind to <3 x i64>
   %gep.random = getelementptr i32, <3 x i32*> %base, <3 x i64> %sext_ind
   %res = call <3 x i32> @llvm.masked.gather.v3i32.v3p0i32(<3 x i32*> %gep.random, i32 4, <3 x i1> %mask, <3 x i32> %src0)
   ret <3 x i32>%res

llvm/trunk/test/CodeGen/X86/masked_gather_scatter_widen.ll

 ;
 ; WIDEN_AVX2-LABEL: test_scatter_v2i32_index:
 ; WIDEN_AVX2:       # %bb.0:
 ; WIDEN_AVX2-NEXT:    vpmovsxdq %xmm1, %xmm1
 ; WIDEN_AVX2-NEXT:    vpsllq $3, %xmm1, %xmm1
 ; WIDEN_AVX2-NEXT:    vmovq %rdi, %xmm3
 ; WIDEN_AVX2-NEXT:    vpbroadcastq %xmm3, %xmm3
 ; WIDEN_AVX2-NEXT:    vpaddq %xmm1, %xmm3, %xmm1
-; WIDEN_AVX2-NEXT:    vpextrb $0, %xmm2, %eax
-; WIDEN_AVX2-NEXT:    testb $1, %al
-; WIDEN_AVX2-NEXT:    je .LBB1_2
-; WIDEN_AVX2-NEXT:  # %bb.1: # %cond.store
-; WIDEN_AVX2-NEXT:    vmovq %xmm1, %rax
-; WIDEN_AVX2-NEXT:    vmovlps %xmm0, (%rax)
-; WIDEN_AVX2-NEXT:  .LBB1_2: # %else
-; WIDEN_AVX2-NEXT:    vpextrb $8, %xmm2, %eax
+; WIDEN_AVX2-NEXT:    vpsllq $63, %xmm2, %xmm2
+; WIDEN_AVX2-NEXT:    vmovmskpd %xmm2, %eax
 ; WIDEN_AVX2-NEXT:    testb $1, %al
+; WIDEN_AVX2-NEXT:    jne .LBB1_1
+; WIDEN_AVX2-NEXT:  # %bb.2: # %else
+; WIDEN_AVX2-NEXT:    testb $2, %al
+; WIDEN_AVX2-NEXT:    jne .LBB1_3
+; WIDEN_AVX2-NEXT:  .LBB1_4: # %else2
+; WIDEN_AVX2-NEXT:    retq
+; WIDEN_AVX2-NEXT:  .LBB1_1: # %cond.store
+; WIDEN_AVX2-NEXT:    vmovq %xmm1, %rcx
+; WIDEN_AVX2-NEXT:    vmovlps %xmm0, (%rcx)
+; WIDEN_AVX2-NEXT:    testb $2, %al
 ; WIDEN_AVX2-NEXT:    je .LBB1_4
-; WIDEN_AVX2-NEXT:  # %bb.3: # %cond.store1
+; WIDEN_AVX2-NEXT:  .LBB1_3: # %cond.store1
 ; WIDEN_AVX2-NEXT:    vpextrq $1, %xmm1, %rax
 ; WIDEN_AVX2-NEXT:    vmovhps %xmm0, (%rax)
-; WIDEN_AVX2-NEXT:  .LBB1_4: # %else2
 ; WIDEN_AVX2-NEXT:    retq
 ;
 ; PROMOTE_AVX2-LABEL: test_scatter_v2i32_index:
 ; PROMOTE_AVX2:       # %bb.0:
 ; PROMOTE_AVX2-NEXT:    vpsllq $32, %xmm1, %xmm3
 ; PROMOTE_AVX2-NEXT:    vpsrad $31, %xmm3, %xmm3
 ; PROMOTE_AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm3[1],xmm1[2],xmm3[3]
 ; PROMOTE_AVX2-NEXT:    vpsllq $3, %xmm1, %xmm1
 ; PROMOTE_AVX2-NEXT:    vmovq %rdi, %xmm3
 ; PROMOTE_AVX2-NEXT:    vpbroadcastq %xmm3, %xmm3
 ; PROMOTE_AVX2-NEXT:    vpaddq %xmm1, %xmm3, %xmm1
-; PROMOTE_AVX2-NEXT:    vpextrb $0, %xmm2, %eax
-; PROMOTE_AVX2-NEXT:    testb $1, %al
-; PROMOTE_AVX2-NEXT:    je .LBB1_2
-; PROMOTE_AVX2-NEXT:  # %bb.1: # %cond.store
-; PROMOTE_AVX2-NEXT:    vmovq %xmm1, %rax
-; PROMOTE_AVX2-NEXT:    vmovlps %xmm0, (%rax)
-; PROMOTE_AVX2-NEXT:  .LBB1_2: # %else
-; PROMOTE_AVX2-NEXT:    vpextrb $8, %xmm2, %eax
+; PROMOTE_AVX2-NEXT:    vpsllq $63, %xmm2, %xmm2
+; PROMOTE_AVX2-NEXT:    vmovmskpd %xmm2, %eax
 ; PROMOTE_AVX2-NEXT:    testb $1, %al
+; PROMOTE_AVX2-NEXT:    jne .LBB1_1
+; PROMOTE_AVX2-NEXT:  # %bb.2: # %else
+; PROMOTE_AVX2-NEXT:    testb $2, %al
+; PROMOTE_AVX2-NEXT:    jne .LBB1_3
+; PROMOTE_AVX2-NEXT:  .LBB1_4: # %else2
+; PROMOTE_AVX2-NEXT:    retq
+; PROMOTE_AVX2-NEXT:  .LBB1_1: # %cond.store
+; PROMOTE_AVX2-NEXT:    vmovq %xmm1, %rcx
+; PROMOTE_AVX2-NEXT:    vmovlps %xmm0, (%rcx)
+; PROMOTE_AVX2-NEXT:    testb $2, %al
 ; PROMOTE_AVX2-NEXT:    je .LBB1_4
-; PROMOTE_AVX2-NEXT:  # %bb.3: # %cond.store1
+; PROMOTE_AVX2-NEXT:  .LBB1_3: # %cond.store1
 ; PROMOTE_AVX2-NEXT:    vpextrq $1, %xmm1, %rax
 ; PROMOTE_AVX2-NEXT:    vmovhps %xmm0, (%rax)
-; PROMOTE_AVX2-NEXT:  .LBB1_4: # %else2
 ; PROMOTE_AVX2-NEXT:    retq
   %gep = getelementptr double, double *%base, <2 x i32> %ind
   call void @llvm.masked.scatter.v2f64.v2p0f64(<2 x double> %a1, <2 x double*> %gep, i32 4, <2 x i1> %mask)
   ret void
 }
 
 define <2 x i32> @test_gather_v2i32_data(<2 x i32*> %ptr, <2 x i1> %mask, <2 x i32> %src0) {
 ; WIDEN_SKX-LABEL: test_gather_v2i32_data:

 ; PROMOTE_KNL-NEXT:    kshiftlw $14, %k0, %k0
 ; PROMOTE_KNL-NEXT:    kshiftrw $14, %k0, %k1
 ; PROMOTE_KNL-NEXT:    vpscatterqd %ymm0, (,%zmm1) {%k1}
 ; PROMOTE_KNL-NEXT:    vzeroupper
 ; PROMOTE_KNL-NEXT:    retq
 ;
 ; WIDEN_AVX2-LABEL: test_scatter_v2i32_data:
 ; WIDEN_AVX2:       # %bb.0:
-; WIDEN_AVX2-NEXT:    vpextrb $0, %xmm2, %eax
-; WIDEN_AVX2-NEXT:    testb $1, %al
-; WIDEN_AVX2-NEXT:    je .LBB3_2
-; WIDEN_AVX2-NEXT:  # %bb.1: # %cond.store
-; WIDEN_AVX2-NEXT:    vmovq %xmm1, %rax
-; WIDEN_AVX2-NEXT:    vmovss %xmm0, (%rax)
-; WIDEN_AVX2-NEXT:  .LBB3_2: # %else
-; WIDEN_AVX2-NEXT:    vpextrb $8, %xmm2, %eax
+; WIDEN_AVX2-NEXT:    vpsllq $63, %xmm2, %xmm2
+; WIDEN_AVX2-NEXT:    vmovmskpd %xmm2, %eax
 ; WIDEN_AVX2-NEXT:    testb $1, %al
+; WIDEN_AVX2-NEXT:    jne .LBB3_1
+; WIDEN_AVX2-NEXT:  # %bb.2: # %else
+; WIDEN_AVX2-NEXT:    testb $2, %al
+; WIDEN_AVX2-NEXT:    jne .LBB3_3
+; WIDEN_AVX2-NEXT:  .LBB3_4: # %else2
+; WIDEN_AVX2-NEXT:    retq
+; WIDEN_AVX2-NEXT:  .LBB3_1: # %cond.store
+; WIDEN_AVX2-NEXT:    vmovq %xmm1, %rcx
+; WIDEN_AVX2-NEXT:    vmovss %xmm0, (%rcx)
+; WIDEN_AVX2-NEXT:    testb $2, %al
 ; WIDEN_AVX2-NEXT:    je .LBB3_4
-; WIDEN_AVX2-NEXT:  # %bb.3: # %cond.store1
+; WIDEN_AVX2-NEXT:  .LBB3_3: # %cond.store1
 ; WIDEN_AVX2-NEXT:    vpextrq $1, %xmm1, %rax
 ; WIDEN_AVX2-NEXT:    vextractps $1, %xmm0, (%rax)
-; WIDEN_AVX2-NEXT:  .LBB3_4: # %else2
 ; WIDEN_AVX2-NEXT:    retq
 ;
 ; PROMOTE_AVX2-LABEL: test_scatter_v2i32_data:
 ; PROMOTE_AVX2:       # %bb.0:
-; PROMOTE_AVX2-NEXT:    vpextrb $0, %xmm2, %eax
-; PROMOTE_AVX2-NEXT:    testb $1, %al
-; PROMOTE_AVX2-NEXT:    je .LBB3_2
-; PROMOTE_AVX2-NEXT:  # %bb.1: # %cond.store
-; PROMOTE_AVX2-NEXT:    vmovq %xmm1, %rax
-; PROMOTE_AVX2-NEXT:    vmovss %xmm0, (%rax)
-; PROMOTE_AVX2-NEXT:  .LBB3_2: # %else
-; PROMOTE_AVX2-NEXT:    vpextrb $8, %xmm2, %eax
+; PROMOTE_AVX2-NEXT:    vpsllq $63, %xmm2, %xmm2
+; PROMOTE_AVX2-NEXT:    vmovmskpd %xmm2, %eax
 ; PROMOTE_AVX2-NEXT:    testb $1, %al
+; PROMOTE_AVX2-NEXT:    jne .LBB3_1
+; PROMOTE_AVX2-NEXT:  # %bb.2: # %else
+; PROMOTE_AVX2-NEXT:    testb $2, %al
+; PROMOTE_AVX2-NEXT:    jne .LBB3_3
+; PROMOTE_AVX2-NEXT:  .LBB3_4: # %else2
+; PROMOTE_AVX2-NEXT:    retq
+; PROMOTE_AVX2-NEXT:  .LBB3_1: # %cond.store
+; PROMOTE_AVX2-NEXT:    vmovq %xmm1, %rcx
+; PROMOTE_AVX2-NEXT:    vmovss %xmm0, (%rcx)
+; PROMOTE_AVX2-NEXT:    testb $2, %al
 ; PROMOTE_AVX2-NEXT:    je .LBB3_4
-; PROMOTE_AVX2-NEXT:  # %bb.3: # %cond.store1
+; PROMOTE_AVX2-NEXT:  .LBB3_3: # %cond.store1
 ; PROMOTE_AVX2-NEXT:    vpextrq $1, %xmm1, %rax
 ; PROMOTE_AVX2-NEXT:    vextractps $2, %xmm0, (%rax)
-; PROMOTE_AVX2-NEXT:  .LBB3_4: # %else2
 ; PROMOTE_AVX2-NEXT:    retq
   call void @llvm.masked.scatter.v2i32.v2p0i32(<2 x i32> %a1, <2 x i32*> %ptr, i32 4, <2 x i1> %mask)
   ret void
 }
 
 define <2 x i32> @test_gather_v2i32_data_index(i32* %base, <2 x i32> %ind, <2 x i1> %mask, <2 x i32> %src0) {
 ; WIDEN_SKX-LABEL: test_gather_v2i32_data_index:
 ; WIDEN_SKX:       # %bb.0:

 ;
 ; WIDEN_AVX2-LABEL: test_scatter_v2i32_data_index:
 ; WIDEN_AVX2:       # %bb.0:
 ; WIDEN_AVX2-NEXT:    vpmovsxdq %xmm1, %xmm1
 ; WIDEN_AVX2-NEXT:    vpsllq $2, %xmm1, %xmm1
 ; WIDEN_AVX2-NEXT:    vmovq %rdi, %xmm3
 ; WIDEN_AVX2-NEXT:    vpbroadcastq %xmm3, %xmm3
 ; WIDEN_AVX2-NEXT:    vpaddq %xmm1, %xmm3, %xmm1
-; WIDEN_AVX2-NEXT:    vpextrb $0, %xmm2, %eax
-; WIDEN_AVX2-NEXT:    testb $1, %al
-; WIDEN_AVX2-NEXT:    je .LBB5_2
-; WIDEN_AVX2-NEXT:  # %bb.1: # %cond.store
-; WIDEN_AVX2-NEXT:    vmovq %xmm1, %rax
-; WIDEN_AVX2-NEXT:    vmovss %xmm0, (%rax)
-; WIDEN_AVX2-NEXT:  .LBB5_2: # %else
-; WIDEN_AVX2-NEXT:    vpextrb $8, %xmm2, %eax
+; WIDEN_AVX2-NEXT:    vpsllq $63, %xmm2, %xmm2
+; WIDEN_AVX2-NEXT:    vmovmskpd %xmm2, %eax
 ; WIDEN_AVX2-NEXT:    testb $1, %al
+; WIDEN_AVX2-NEXT:    jne .LBB5_1
+; WIDEN_AVX2-NEXT:  # %bb.2: # %else
+; WIDEN_AVX2-NEXT:    testb $2, %al
+; WIDEN_AVX2-NEXT:    jne .LBB5_3
+; WIDEN_AVX2-NEXT:  .LBB5_4: # %else2
+; WIDEN_AVX2-NEXT:    retq
+; WIDEN_AVX2-NEXT:  .LBB5_1: # %cond.store
+; WIDEN_AVX2-NEXT:    vmovq %xmm1, %rcx
+; WIDEN_AVX2-NEXT:    vmovss %xmm0, (%rcx)
+; WIDEN_AVX2-NEXT:    testb $2, %al
 ; WIDEN_AVX2-NEXT:    je .LBB5_4
-; WIDEN_AVX2-NEXT:  # %bb.3: # %cond.store1
+; WIDEN_AVX2-NEXT:  .LBB5_3: # %cond.store1
 ; WIDEN_AVX2-NEXT:    vpextrq $1, %xmm1, %rax
 ; WIDEN_AVX2-NEXT:    vextractps $1, %xmm0, (%rax)
-; WIDEN_AVX2-NEXT:  .LBB5_4: # %else2
 ; WIDEN_AVX2-NEXT:    retq
 ;
 ; PROMOTE_AVX2-LABEL: test_scatter_v2i32_data_index:
 ; PROMOTE_AVX2:       # %bb.0:
 ; PROMOTE_AVX2-NEXT:    vpsllq $32, %xmm1, %xmm3
 ; PROMOTE_AVX2-NEXT:    vpsrad $31, %xmm3, %xmm3
 ; PROMOTE_AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm3[1],xmm1[2],xmm3[3]
 ; PROMOTE_AVX2-NEXT:    vpsllq $2, %xmm1, %xmm1
 ; PROMOTE_AVX2-NEXT:    vmovq %rdi, %xmm3
 ; PROMOTE_AVX2-NEXT:    vpbroadcastq %xmm3, %xmm3
 ; PROMOTE_AVX2-NEXT:    vpaddq %xmm1, %xmm3, %xmm1
-; PROMOTE_AVX2-NEXT:    vpextrb $0, %xmm2, %eax
-; PROMOTE_AVX2-NEXT:    testb $1, %al
-; PROMOTE_AVX2-NEXT:    je .LBB5_2
-; PROMOTE_AVX2-NEXT:  # %bb.1: # %cond.store
-; PROMOTE_AVX2-NEXT:    vmovq %xmm1, %rax
-; PROMOTE_AVX2-NEXT:    vmovss %xmm0, (%rax)
-; PROMOTE_AVX2-NEXT:  .LBB5_2: # %else
-; PROMOTE_AVX2-NEXT:    vpextrb $8, %xmm2, %eax
+; PROMOTE_AVX2-NEXT:    vpsllq $63, %xmm2, %xmm2
+; PROMOTE_AVX2-NEXT:    vmovmskpd %xmm2, %eax
 ; PROMOTE_AVX2-NEXT:    testb $1, %al
+; PROMOTE_AVX2-NEXT:    jne .LBB5_1
+; PROMOTE_AVX2-NEXT:  # %bb.2: # %else
+; PROMOTE_AVX2-NEXT:    testb $2, %al
+; PROMOTE_AVX2-NEXT:    jne .LBB5_3
+; PROMOTE_AVX2-NEXT:  .LBB5_4: # %else2
+; PROMOTE_AVX2-NEXT:    retq
+; PROMOTE_AVX2-NEXT:  .LBB5_1: # %cond.store
+; PROMOTE_AVX2-NEXT:    vmovq %xmm1, %rcx
+; PROMOTE_AVX2-NEXT:    vmovss %xmm0, (%rcx)
+; PROMOTE_AVX2-NEXT:    testb $2, %al
 ; PROMOTE_AVX2-NEXT:    je .LBB5_4
-; PROMOTE_AVX2-NEXT:  # %bb.3: # %cond.store1
+; PROMOTE_AVX2-NEXT:  .LBB5_3: # %cond.store1
 ; PROMOTE_AVX2-NEXT:    vpextrq $1, %xmm1, %rax
 ; PROMOTE_AVX2-NEXT:    vextractps $2, %xmm0, (%rax)
-; PROMOTE_AVX2-NEXT:  .LBB5_4: # %else2
 ; PROMOTE_AVX2-NEXT:    retq
   %gep = getelementptr i32, i32 *%base, <2 x i32> %ind
   call void @llvm.masked.scatter.v2i32.v2p0i32(<2 x i32> %a1, <2 x i32*> %gep, i32 4, <2 x i1> %mask)
   ret void
 }
 
 declare <2 x double> @llvm.masked.gather.v2f64.v2p0f64(<2 x double*>, i32, <2 x i1>, <2 x double>)
 declare void @llvm.masked.scatter.v2f64.v2p0f64(<2 x double>, <2 x double*>, i32, <2 x i1>)
 declare <2 x i32> @llvm.masked.gather.v2i32.v2p0i32(<2 x i32*>, i32, <2 x i1>, <2 x i32>)
 declare void @llvm.masked.scatter.v2i32.v2p0i32(<2 x i32> , <2 x i32*> , i32 , <2 x i1>)

llvm/trunk/test/Transforms/ScalarizeMaskedMemIntrin/X86/expand-masked-gather.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S %s -scalarize-masked-mem-intrin -mtriple=x86_64-linux-gnu | FileCheck %s
 
 define <2 x i64> @scalarize_v2i64(<2 x i64*> %p, <2 x i1> %mask, <2 x i64> %passthru) {
 ; CHECK-LABEL: @scalarize_v2i64(
-; CHECK-NEXT:    [[MASK0:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
-; CHECK-NEXT:    br i1 [[MASK0]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
+; CHECK-NEXT:    [[SCALAR_MASK:%.*]] = bitcast <2 x i1> [[MASK:%.*]] to i2
+; CHECK-NEXT:    [[TMP1:%.*]] = and i2 [[SCALAR_MASK]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ne i2 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
 ; CHECK:       cond.load:
 ; CHECK-NEXT:    [[PTR0:%.*]] = extractelement <2 x i64*> [[P:%.*]], i64 0
 ; CHECK-NEXT:    [[LOAD0:%.*]] = load i64, i64* [[PTR0]], align 8
 ; CHECK-NEXT:    [[RES0:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[LOAD0]], i64 0
 ; CHECK-NEXT:    br label [[ELSE]]
 ; CHECK:       else:
 ; CHECK-NEXT:    [[RES_PHI_ELSE:%.*]] = phi <2 x i64> [ [[RES0]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
-; CHECK-NEXT:    [[MASK1:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
-; CHECK-NEXT:    br i1 [[MASK1]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
+; CHECK-NEXT:    [[TMP3:%.*]] = and i2 [[SCALAR_MASK]], -2
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne i2 [[TMP3]], 0
+; CHECK-NEXT:    br i1 [[TMP4]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
 ; CHECK:       cond.load1:
 ; CHECK-NEXT:    [[PTR1:%.*]] = extractelement <2 x i64*> [[P]], i64 1
 ; CHECK-NEXT:    [[LOAD1:%.*]] = load i64, i64* [[PTR1]], align 8
 ; CHECK-NEXT:    [[RES1:%.*]] = insertelement <2 x i64> [[RES_PHI_ELSE]], i64 [[LOAD1]], i64 1
 ; CHECK-NEXT:    br label [[ELSE2]]
 ; CHECK:       else2:
 ; CHECK-NEXT:    [[RES_PHI_ELSE3:%.*]] = phi <2 x i64> [ [[RES1]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
 ; CHECK-NEXT:    ret <2 x i64> [[RES_PHI_ELSE3]]

llvm/trunk/test/Transforms/ScalarizeMaskedMemIntrin/X86/expand-masked-load.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S %s -scalarize-masked-mem-intrin -mtriple=x86_64-linux-gnu | FileCheck %s
 
 define <2 x i64> @scalarize_v2i64(<2 x i64>* %p, <2 x i1> %mask, <2 x i64> %passthru) {
 ; CHECK-LABEL: @scalarize_v2i64(
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64*
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
-; CHECK-NEXT:    br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
+; CHECK-NEXT:    [[SCALAR_MASK:%.*]] = bitcast <2 x i1> [[MASK:%.*]] to i2
+; CHECK-NEXT:    [[TMP2:%.*]] = and i2 [[SCALAR_MASK]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne i2 [[TMP2]], 0
+; CHECK-NEXT:    br i1 [[TMP3]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
 ; CHECK:       cond.load:
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = load i64, i64* [[TMP3]], align 8
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[TMP4]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = load i64, i64* [[TMP4]], align 8
+; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[TMP5]], i64 0
 ; CHECK-NEXT:    br label [[ELSE]]
 ; CHECK:       else:
-; CHECK-NEXT:    [[RES_PHI_ELSE:%.*]] = phi <2 x i64> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
-; CHECK-NEXT:    br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
+; CHECK-NEXT:    [[RES_PHI_ELSE:%.*]] = phi <2 x i64> [ [[TMP6]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
+; CHECK-NEXT:    [[TMP7:%.*]] = and i2 [[SCALAR_MASK]], -2
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne i2 [[TMP7]], 0
+; CHECK-NEXT:    br i1 [[TMP8]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
 ; CHECK:       cond.load1:
-; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1
-; CHECK-NEXT:    [[TMP8:%.*]] = load i64, i64* [[TMP7]], align 8
-; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x i64> [[RES_PHI_ELSE]], i64 [[TMP8]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1
+; CHECK-NEXT:    [[TMP10:%.*]] = load i64, i64* [[TMP9]], align 8
+; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <2 x i64> [[RES_PHI_ELSE]], i64 [[TMP10]], i64 1
 ; CHECK-NEXT:    br label [[ELSE2]]
 ; CHECK:       else2:
-; CHECK-NEXT:    [[RES_PHI_ELSE3:%.*]] = phi <2 x i64> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
+; CHECK-NEXT:    [[RES_PHI_ELSE3:%.*]] = phi <2 x i64> [ [[TMP11]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
 ; CHECK-NEXT:    ret <2 x i64> [[RES_PHI_ELSE3]]
 ;
   %ret = call <2 x i64> @llvm.masked.load.v2i64.p0v2i64(<2 x i64>* %p, i32 128, <2 x i1> %mask, <2 x i64> %passthru)
   ret <2 x i64> %ret
 }
 
 define <2 x i64> @scalarize_v2i64_ones_mask(<2 x i64>* %p, <2 x i64> %passthru) {
 ; CHECK-LABEL: @scalarize_v2i64_ones_mask(
;
   %ret = call <2 x i64> @llvm.masked.load.v2i64.p0v2i64(<2 x i64>* %p, i32 8, <2 x i1> <i1 false, i1 true>, <2 x i64> %passthru)
   ret <2 x i64> %ret
 }
 
 ; This use a byte sized but non power of 2 element size. This used to crash due to bad alignment calculation.
 define <2 x i24> @scalarize_v2i24(<2 x i24>* %p, <2 x i1> %mask, <2 x i24> %passthru) {
 ; CHECK-LABEL: @scalarize_v2i24(
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i24>* [[P:%.*]] to i24*
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
-; CHECK-NEXT:    br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
+; CHECK-NEXT:    [[SCALAR_MASK:%.*]] = bitcast <2 x i1> [[MASK:%.*]] to i2
+; CHECK-NEXT:    [[TMP2:%.*]] = and i2 [[SCALAR_MASK]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne i2 [[TMP2]], 0
+; CHECK-NEXT:    br i1 [[TMP3]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
 ; CHECK:       cond.load:
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i24, i24* [[TMP1]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = load i24, i24* [[TMP3]], align 1
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x i24> [[PASSTHRU:%.*]], i24 [[TMP4]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i24, i24* [[TMP1]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = load i24, i24* [[TMP4]], align 1
+; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x i24> [[PASSTHRU:%.*]], i24 [[TMP5]], i64 0
 ; CHECK-NEXT:    br label [[ELSE]]
 ; CHECK:       else:
-; CHECK-NEXT:    [[RES_PHI_ELSE:%.*]] = phi <2 x i24> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
-; CHECK-NEXT:    br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
+; CHECK-NEXT:    [[RES_PHI_ELSE:%.*]] = phi <2 x i24> [ [[TMP6]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
+; CHECK-NEXT:    [[TMP7:%.*]] = and i2 [[SCALAR_MASK]], -2
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne i2 [[TMP7]], 0
+; CHECK-NEXT:    br i1 [[TMP8]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
 ; CHECK:       cond.load1:
-; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i24, i24* [[TMP1]], i32 1
-; CHECK-NEXT:    [[TMP8:%.*]] = load i24, i24* [[TMP7]], align 1
-; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x i24> [[RES_PHI_ELSE]], i24 [[TMP8]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i24, i24* [[TMP1]], i32 1
+; CHECK-NEXT:    [[TMP10:%.*]] = load i24, i24* [[TMP9]], align 1
+; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <2 x i24> [[RES_PHI_ELSE]], i24 [[TMP10]], i64 1
 ; CHECK-NEXT:    br label [[ELSE2]]
 ; CHECK:       else2:
-; CHECK-NEXT:    [[RES_PHI_ELSE3:%.*]] = phi <2 x i24> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
+; CHECK-NEXT:    [[RES_PHI_ELSE3:%.*]] = phi <2 x i24> [ [[TMP11]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
 ; CHECK-NEXT:    ret <2 x i24> [[RES_PHI_ELSE3]]
 ;
   %ret = call <2 x i24> @llvm.masked.load.v2i24.p0v2i24(<2 x i24>* %p, i32 8, <2 x i1> %mask, <2 x i24> %passthru)
   ret <2 x i24> %ret
 }
 
 ; This use a byte sized but non power of 2 element size. This used to crash due to bad alignment calculation.
 define <2 x i48> @scalarize_v2i48(<2 x i48>* %p, <2 x i1> %mask, <2 x i48> %passthru) {
 ; CHECK-LABEL: @scalarize_v2i48(
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i48>* [[P:%.*]] to i48*
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
-; CHECK-NEXT:    br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
+; CHECK-NEXT:    [[SCALAR_MASK:%.*]] = bitcast <2 x i1> [[MASK:%.*]] to i2
+; CHECK-NEXT:    [[TMP2:%.*]] = and i2 [[SCALAR_MASK]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne i2 [[TMP2]], 0
+; CHECK-NEXT:    br i1 [[TMP3]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
 ; CHECK:       cond.load:
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i48, i48* [[TMP1]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = load i48, i48* [[TMP3]], align 2
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x i48> [[PASSTHRU:%.*]], i48 [[TMP4]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i48, i48* [[TMP1]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = load i48, i48* [[TMP4]], align 2
+; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x i48> [[PASSTHRU:%.*]], i48 [[TMP5]], i64 0
 ; CHECK-NEXT:    br label [[ELSE]]
 ; CHECK:       else:
-; CHECK-NEXT:    [[RES_PHI_ELSE:%.*]] = phi <2 x i48> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
-; CHECK-NEXT:    br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
+; CHECK-NEXT:    [[RES_PHI_ELSE:%.*]] = phi <2 x i48> [ [[TMP6]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
+; CHECK-NEXT:    [[TMP7:%.*]] = and i2 [[SCALAR_MASK]], -2
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne i2 [[TMP7]], 0
+; CHECK-NEXT:    br i1 [[TMP8]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
 ; CHECK:       cond.load1:
-; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i48, i48* [[TMP1]], i32 1
-; CHECK-NEXT:    [[TMP8:%.*]] = load i48, i48* [[TMP7]], align 2
-; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x i48> [[RES_PHI_ELSE]], i48 [[TMP8]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i48, i48* [[TMP1]], i32 1
+; CHECK-NEXT:    [[TMP10:%.*]] = load i48, i48* [[TMP9]], align 2
+; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <2 x i48> [[RES_PHI_ELSE]], i48 [[TMP10]], i64 1
 ; CHECK-NEXT:    br label [[ELSE2]]
 ; CHECK:       else2:
-; CHECK-NEXT:    [[RES_PHI_ELSE3:%.*]] = phi <2 x i48> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
+; CHECK-NEXT:    [[RES_PHI_ELSE3:%.*]] = phi <2 x i48> [ [[TMP11]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
 ; CHECK-NEXT:    ret <2 x i48> [[RES_PHI_ELSE3]]
 ;
   %ret = call <2 x i48> @llvm.masked.load.v2i48.p0v2i48(<2 x i48>* %p, i32 16, <2 x i1> %mask, <2 x i48> %passthru)
   ret <2 x i48> %ret
 }
 
 declare <2 x i24> @llvm.masked.load.v2i24.p0v2i24(<2 x i24>*, i32, <2 x i1>, <2 x i24>)
 declare <2 x i48> @llvm.masked.load.v2i48.p0v2i48(<2 x i48>*, i32, <2 x i1>, <2 x i48>)
 declare <2 x i64> @llvm.masked.load.v2i64.p0v2i64(<2 x i64>*, i32, <2 x i1>, <2 x i64>)

llvm/trunk/test/Transforms/ScalarizeMaskedMemIntrin/X86/expand-masked-store.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S %s -scalarize-masked-mem-intrin -mtriple=x86_64-linux-gnu | FileCheck %s
 
 define void @scalarize_v2i64(<2 x i64>* %p, <2 x i1> %mask, <2 x i64> %data) {
 ; CHECK-LABEL: @scalarize_v2i64(
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64*
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
-; CHECK-NEXT:    br i1 [[TMP2]], label [[COND_STORE:%.*]], label [[ELSE:%.*]]
+; CHECK-NEXT:    [[SCALAR_MASK:%.*]] = bitcast <2 x i1> [[MASK:%.*]] to i2
+; CHECK-NEXT:    [[TMP2:%.*]] = and i2 [[SCALAR_MASK]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne i2 [[TMP2]], 0
+; CHECK-NEXT:    br i1 [[TMP3]], label [[COND_STORE:%.*]], label [[ELSE:%.*]]
 ; CHECK:       cond.store:
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i64> [[DATA:%.*]], i64 0
-; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0
-; CHECK-NEXT:    store i64 [[TMP3]], i64* [[TMP4]], align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i64> [[DATA:%.*]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0
+; CHECK-NEXT:    store i64 [[TMP4]], i64* [[TMP5]], align 8
 ; CHECK-NEXT:    br label [[ELSE]]
 ; CHECK:       else:
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
-; CHECK-NEXT:    br i1 [[TMP5]], label [[COND_STORE1:%.*]], label [[ELSE2:%.*]]
+; CHECK-NEXT:    [[TMP6:%.*]] = and i2 [[SCALAR_MASK]], -2
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne i2 [[TMP6]], 0
+; CHECK-NEXT:    br i1 [[TMP7]], label [[COND_STORE1:%.*]], label [[ELSE2:%.*]]
 ; CHECK:       cond.store1:
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i64> [[DATA]], i64 1
-; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1
-; CHECK-NEXT:    store i64 [[TMP6]], i64* [[TMP7]], align 8
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x i64> [[DATA]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1
+; CHECK-NEXT:    store i64 [[TMP8]], i64* [[TMP9]], align 8
 ; CHECK-NEXT:    br label [[ELSE2]]
 ; CHECK:       else2:
 ; CHECK-NEXT:    ret void
 ;
   call void @llvm.masked.store.v2i64.p0v2i64(<2 x i64> %data, <2 x i64>* %p, i32 128, <2 x i1> %mask)
   ret void
 }