[X86] Don't select anyext GR32->GR64 to SUBREG_TO_REG. Use INSERT_SUBREG instead.

[craig.topper]

Summary

As far as I know SUBREG_TO_REG is stating that the upper bits are 0. But if we are just converting the GR32 with no checks, then we have no reason to say the upper bits are 0.

Seems to me like we should be selecting this to INSERT_SUBREG.

[DavidKreitzer]

This change seems obviously correct. The diffs in the tests all look like tweaks in the register allocation behavior and aren't testing the behavior you are fixing. Is it possible to write a test that miscompiles due to the existing bug?

[craig.topper]

I havent' found any piece of code that analyzes SUB_REG_TO_REG that closely. Everything I found pretty much treats it much like a copy. Does anyone know of anything that cares today?

[DavidKreitzer]

I'd recommend going ahead with this, Craig. Even if it isn't possible for this bug to trigger a failure today, it looks like a latent bug that should be fixed.

llvm/trunk/lib/Target/X86/X86InstrCompiler.td

 def : Pat<(i32 (anyext GR16:$src)),
           (INSERT_SUBREG (i32 (IMPLICIT_DEF)), GR16:$src, sub_16bit)>;
 
 def : Pat<(i64 (anyext GR8 :$src)),
           (SUBREG_TO_REG (i64 0), (MOVZX32rr8  GR8  :$src), sub_32bit)>;
 def : Pat<(i64 (anyext GR16:$src)),
           (SUBREG_TO_REG (i64 0), (MOVZX32rr16 GR16 :$src), sub_32bit)>;
 def : Pat<(i64 (anyext GR32:$src)),
-          (SUBREG_TO_REG (i64 0), GR32:$src, sub_32bit)>;
+          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, sub_32bit)>;
 
 
 // Any instruction that defines a 32-bit result leaves the high half of the
 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
 // be copying from a truncate. Any other 32-bit operation will zero-extend
 // up to 64 bits. AssertSext/AssertZext aren't saying anything about the upper
 // 32 bits, they're probably just qualifying a CopyFromReg.
 def def32 : PatLeaf<(i32 GR32:$src), [{

llvm/trunk/test/CodeGen/X86/vector-shuffle-variable-128.ll

 ; SSE2-NEXT:    andl $7, %ecx
 ; SSE2-NEXT:    andl $7, %r8d
 ; SSE2-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
 ; SSE2-NEXT:    andl $7, %r9d
 ; SSE2-NEXT:    movzwl {{[0-9]+}}(%rsp), %r10d
 ; SSE2-NEXT:    andl $7, %r10d
 ; SSE2-NEXT:    movzwl {{[0-9]+}}(%rsp), %eax
 ; SSE2-NEXT:    andl $7, %eax
-; SSE2-NEXT:    movzwl -24(%rsp,%rax,2), %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    movzwl -24(%rsp,%r10,2), %eax
-; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    movzwl -24(%rsp,%rcx,2), %ecx
+; SSE2-NEXT:    movd %ecx, %xmm0
+; SSE2-NEXT:    movzwl -24(%rsp,%rdx,2), %ecx
+; SSE2-NEXT:    movd %ecx, %xmm1
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    movzwl -24(%rsp,%r9,2), %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    movzwl -24(%rsp,%r8,2), %eax
-; SSE2-NEXT:    movd %eax, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE2-NEXT:    movzwl -24(%rsp,%rcx,2), %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    movzwl -24(%rsp,%rdx,2), %eax
+; SSE2-NEXT:    movzwl -24(%rsp,%rsi,2), %ecx
+; SSE2-NEXT:    movd %ecx, %xmm2
+; SSE2-NEXT:    movzwl -24(%rsp,%rdi,2), %ecx
+; SSE2-NEXT:    movd %ecx, %xmm0
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    movzwl -24(%rsp,%r9,2), %ecx
+; SSE2-NEXT:    movd %ecx, %xmm1
+; SSE2-NEXT:    movzwl -24(%rsp,%r8,2), %ecx
+; SSE2-NEXT:    movd %ecx, %xmm2
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE2-NEXT:    movzwl -24(%rsp,%rax,2), %eax
 ; SSE2-NEXT:    movd %eax, %xmm1
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    movzwl -24(%rsp,%rsi,2), %eax
+; SSE2-NEXT:    movzwl -24(%rsp,%r10,2), %eax
 ; SSE2-NEXT:    movd %eax, %xmm3
-; SSE2-NEXT:    movzwl -24(%rsp,%rdi,2), %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
 ; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: var_shuffle_v8i16_v8i16_xxxxxxxx_i16:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    # kill: %R9D<def> %R9D<kill> %R9<def>
 ; SSSE3-NEXT:    # kill: %R8D<def> %R8D<kill> %R8<def>
 ; SSSE3-NEXT:    # kill: %ECX<def> %ECX<kill> %RCX<def>
 ; SSSE3-NEXT:    # kill: %EDX<def> %EDX<kill> %RDX<def>
 ; SSSE3-NEXT:    # kill: %ESI<def> %ESI<kill> %RSI<def>
 ; SSSE3-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
 ; SSSE3-NEXT:    andl $7, %edi
 ; SSSE3-NEXT:    andl $7, %esi
 ; SSSE3-NEXT:    andl $7, %edx
 ; SSSE3-NEXT:    andl $7, %ecx
 ; SSSE3-NEXT:    andl $7, %r8d
 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
 ; SSSE3-NEXT:    andl $7, %r9d
 ; SSSE3-NEXT:    movzwl {{[0-9]+}}(%rsp), %r10d
 ; SSSE3-NEXT:    andl $7, %r10d
 ; SSSE3-NEXT:    movzwl {{[0-9]+}}(%rsp), %eax
 ; SSSE3-NEXT:    andl $7, %eax
-; SSSE3-NEXT:    movzwl -24(%rsp,%rax,2), %eax
-; SSSE3-NEXT:    movd %eax, %xmm0
-; SSSE3-NEXT:    movzwl -24(%rsp,%r10,2), %eax
-; SSSE3-NEXT:    movd %eax, %xmm1
+; SSSE3-NEXT:    movzwl -24(%rsp,%rcx,2), %ecx
+; SSSE3-NEXT:    movd %ecx, %xmm0
+; SSSE3-NEXT:    movzwl -24(%rsp,%rdx,2), %ecx
+; SSSE3-NEXT:    movd %ecx, %xmm1
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSSE3-NEXT:    movzwl -24(%rsp,%r9,2), %eax
-; SSSE3-NEXT:    movd %eax, %xmm0
-; SSSE3-NEXT:    movzwl -24(%rsp,%r8,2), %eax
-; SSSE3-NEXT:    movd %eax, %xmm2
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSSE3-NEXT:    movzwl -24(%rsp,%rcx,2), %eax
-; SSSE3-NEXT:    movd %eax, %xmm0
-; SSSE3-NEXT:    movzwl -24(%rsp,%rdx,2), %eax
+; SSSE3-NEXT:    movzwl -24(%rsp,%rsi,2), %ecx
+; SSSE3-NEXT:    movd %ecx, %xmm2
+; SSSE3-NEXT:    movzwl -24(%rsp,%rdi,2), %ecx
+; SSSE3-NEXT:    movd %ecx, %xmm0
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSSE3-NEXT:    movzwl -24(%rsp,%r9,2), %ecx
+; SSSE3-NEXT:    movd %ecx, %xmm1
+; SSSE3-NEXT:    movzwl -24(%rsp,%r8,2), %ecx
+; SSSE3-NEXT:    movd %ecx, %xmm2
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSSE3-NEXT:    movzwl -24(%rsp,%rax,2), %eax
 ; SSSE3-NEXT:    movd %eax, %xmm1
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSSE3-NEXT:    movzwl -24(%rsp,%rsi,2), %eax
+; SSSE3-NEXT:    movzwl -24(%rsp,%r10,2), %eax
 ; SSSE3-NEXT:    movd %eax, %xmm3
-; SSSE3-NEXT:    movzwl -24(%rsp,%rdi,2), %eax
-; SSSE3-NEXT:    movd %eax, %xmm0
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
 ; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: var_shuffle_v8i16_v8i16_xxxxxxxx_i16:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    # kill: %R9D<def> %R9D<kill> %R9<def>
 ; SSE41-NEXT:    # kill: %R8D<def> %R8D<kill> %R8<def>
 ; SSE41-NEXT:    # kill: %ECX<def> %ECX<kill> %RCX<def>

llvm/trunk/test/CodeGen/X86/vector-shuffle-variable-256.ll

 ; AVX1-NEXT:    andq $-32, %rsp
 ; AVX1-NEXT:    subq $64, %rsp
 ; AVX1-NEXT:    # kill: %R9D<def> %R9D<kill> %R9<def>
 ; AVX1-NEXT:    # kill: %R8D<def> %R8D<kill> %R8<def>
 ; AVX1-NEXT:    # kill: %ECX<def> %ECX<kill> %RCX<def>
 ; AVX1-NEXT:    # kill: %EDX<def> %EDX<kill> %RDX<def>
 ; AVX1-NEXT:    # kill: %ESI<def> %ESI<kill> %RSI<def>
 ; AVX1-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; AVX1-NEXT:    andl $15, %edi
 ; AVX1-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1-NEXT:    movzwl (%rsp,%rdi,2), %eax
+; AVX1-NEXT:    vmovd %eax, %xmm0
+; AVX1-NEXT:    andl $15, %esi
+; AVX1-NEXT:    vpinsrw $1, (%rsp,%rsi,2), %xmm0, %xmm0
+; AVX1-NEXT:    andl $15, %edx
+; AVX1-NEXT:    vpinsrw $2, (%rsp,%rdx,2), %xmm0, %xmm0
+; AVX1-NEXT:    andl $15, %ecx
+; AVX1-NEXT:    vpinsrw $3, (%rsp,%rcx,2), %xmm0, %xmm0
+; AVX1-NEXT:    andl $15, %r8d
+; AVX1-NEXT:    vpinsrw $4, (%rsp,%r8,2), %xmm0, %xmm0
+; AVX1-NEXT:    andl $15, %r9d
+; AVX1-NEXT:    vpinsrw $5, (%rsp,%r9,2), %xmm0, %xmm0
+; AVX1-NEXT:    movl 16(%rbp), %eax
+; AVX1-NEXT:    andl $15, %eax
+; AVX1-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    movl 24(%rbp), %eax
+; AVX1-NEXT:    andl $15, %eax
+; AVX1-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm0, %xmm0
 ; AVX1-NEXT:    movl 32(%rbp), %eax
 ; AVX1-NEXT:    andl $15, %eax
 ; AVX1-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX1-NEXT:    vmovd %eax, %xmm0
+; AVX1-NEXT:    vmovd %eax, %xmm1
 ; AVX1-NEXT:    movl 40(%rbp), %eax
 ; AVX1-NEXT:    andl $15, %eax
-; AVX1-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl 48(%rbp), %eax
 ; AVX1-NEXT:    andl $15, %eax
-; AVX1-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl 56(%rbp), %eax
 ; AVX1-NEXT:    andl $15, %eax
-; AVX1-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl 64(%rbp), %eax
 ; AVX1-NEXT:    andl $15, %eax
-; AVX1-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl 72(%rbp), %eax
 ; AVX1-NEXT:    andl $15, %eax
-; AVX1-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl 80(%rbp), %eax
 ; AVX1-NEXT:    andl $15, %eax
-; AVX1-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm0, %xmm0
-; AVX1-NEXT:    movl 88(%rbp), %eax
-; AVX1-NEXT:    andl $15, %eax
-; AVX1-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm0, %xmm0
-; AVX1-NEXT:    andl $15, %edi
-; AVX1-NEXT:    movzwl (%rsp,%rdi,2), %eax
-; AVX1-NEXT:    vmovd %eax, %xmm1
-; AVX1-NEXT:    andl $15, %esi
-; AVX1-NEXT:    vpinsrw $1, (%rsp,%rsi,2), %xmm1, %xmm1
-; AVX1-NEXT:    andl $15, %edx
-; AVX1-NEXT:    vpinsrw $2, (%rsp,%rdx,2), %xmm1, %xmm1
-; AVX1-NEXT:    andl $15, %ecx
-; AVX1-NEXT:    vpinsrw $3, (%rsp,%rcx,2), %xmm1, %xmm1
-; AVX1-NEXT:    andl $15, %r8d
-; AVX1-NEXT:    vpinsrw $4, (%rsp,%r8,2), %xmm1, %xmm1
-; AVX1-NEXT:    andl $15, %r9d
-; AVX1-NEXT:    vpinsrw $5, (%rsp,%r9,2), %xmm1, %xmm1
-; AVX1-NEXT:    movl 16(%rbp), %eax
-; AVX1-NEXT:    andl $15, %eax
 ; AVX1-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm1, %xmm1
-; AVX1-NEXT:    movl 24(%rbp), %eax
+; AVX1-NEXT:    movl 88(%rbp), %eax
 ; AVX1-NEXT:    andl $15, %eax
 ; AVX1-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    movq %rbp, %rsp
 ; AVX1-NEXT:    popq %rbp
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shuffle_v16i16_v16i16_xxxxxxxxxxxxxxxx_i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    pushq %rbp
 ; AVX2-NEXT:    movq %rsp, %rbp
 ; AVX2-NEXT:    andq $-32, %rsp
 ; AVX2-NEXT:    subq $64, %rsp
 ; AVX2-NEXT:    # kill: %R9D<def> %R9D<kill> %R9<def>
 ; AVX2-NEXT:    # kill: %R8D<def> %R8D<kill> %R8<def>
 ; AVX2-NEXT:    # kill: %ECX<def> %ECX<kill> %RCX<def>
 ; AVX2-NEXT:    # kill: %EDX<def> %EDX<kill> %RDX<def>
 ; AVX2-NEXT:    # kill: %ESI<def> %ESI<kill> %RSI<def>
 ; AVX2-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; AVX2-NEXT:    andl $15, %edi
 ; AVX2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX2-NEXT:    movzwl (%rsp,%rdi,2), %eax
+; AVX2-NEXT:    vmovd %eax, %xmm0
+; AVX2-NEXT:    andl $15, %esi
+; AVX2-NEXT:    vpinsrw $1, (%rsp,%rsi,2), %xmm0, %xmm0
+; AVX2-NEXT:    andl $15, %edx
+; AVX2-NEXT:    vpinsrw $2, (%rsp,%rdx,2), %xmm0, %xmm0
+; AVX2-NEXT:    andl $15, %ecx
+; AVX2-NEXT:    vpinsrw $3, (%rsp,%rcx,2), %xmm0, %xmm0
+; AVX2-NEXT:    andl $15, %r8d
+; AVX2-NEXT:    vpinsrw $4, (%rsp,%r8,2), %xmm0, %xmm0
+; AVX2-NEXT:    andl $15, %r9d
+; AVX2-NEXT:    vpinsrw $5, (%rsp,%r9,2), %xmm0, %xmm0
+; AVX2-NEXT:    movl 16(%rbp), %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    movl 24(%rbp), %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm0, %xmm0
 ; AVX2-NEXT:    movl 32(%rbp), %eax
 ; AVX2-NEXT:    andl $15, %eax
 ; AVX2-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX2-NEXT:    vmovd %eax, %xmm0
+; AVX2-NEXT:    vmovd %eax, %xmm1
 ; AVX2-NEXT:    movl 40(%rbp), %eax
 ; AVX2-NEXT:    andl $15, %eax
-; AVX2-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl 48(%rbp), %eax
 ; AVX2-NEXT:    andl $15, %eax
-; AVX2-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl 56(%rbp), %eax
 ; AVX2-NEXT:    andl $15, %eax
-; AVX2-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl 64(%rbp), %eax
 ; AVX2-NEXT:    andl $15, %eax
-; AVX2-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl 72(%rbp), %eax
 ; AVX2-NEXT:    andl $15, %eax
-; AVX2-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl 80(%rbp), %eax
 ; AVX2-NEXT:    andl $15, %eax
-; AVX2-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm0, %xmm0
-; AVX2-NEXT:    movl 88(%rbp), %eax
-; AVX2-NEXT:    andl $15, %eax
-; AVX2-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm0, %xmm0
-; AVX2-NEXT:    andl $15, %edi
-; AVX2-NEXT:    movzwl (%rsp,%rdi,2), %eax
-; AVX2-NEXT:    vmovd %eax, %xmm1
-; AVX2-NEXT:    andl $15, %esi
-; AVX2-NEXT:    vpinsrw $1, (%rsp,%rsi,2), %xmm1, %xmm1
-; AVX2-NEXT:    andl $15, %edx
-; AVX2-NEXT:    vpinsrw $2, (%rsp,%rdx,2), %xmm1, %xmm1
-; AVX2-NEXT:    andl $15, %ecx
-; AVX2-NEXT:    vpinsrw $3, (%rsp,%rcx,2), %xmm1, %xmm1
-; AVX2-NEXT:    andl $15, %r8d
-; AVX2-NEXT:    vpinsrw $4, (%rsp,%r8,2), %xmm1, %xmm1
-; AVX2-NEXT:    andl $15, %r9d
-; AVX2-NEXT:    vpinsrw $5, (%rsp,%r9,2), %xmm1, %xmm1
-; AVX2-NEXT:    movl 16(%rbp), %eax
-; AVX2-NEXT:    andl $15, %eax
 ; AVX2-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm1, %xmm1
-; AVX2-NEXT:    movl 24(%rbp), %eax
+; AVX2-NEXT:    movl 88(%rbp), %eax
 ; AVX2-NEXT:    andl $15, %eax
 ; AVX2-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm1, %xmm1
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    movq %rbp, %rsp
 ; AVX2-NEXT:    popq %rbp
 ; AVX2-NEXT:    retq
   %x0  = extractelement <16 x i16> %x, i32 %i0
   %x1  = extractelement <16 x i16> %x, i32 %i1
   %x2  = extractelement <16 x i16> %x, i32 %i2
   %x3  = extractelement <16 x i16> %x, i32 %i3
   %x4  = extractelement <16 x i16> %x, i32 %i4

 ; AVX1-LABEL: var_shuffle_v16i16_v8i16_xxxxxxxxxxxxxxxx_i16:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    # kill: %R9D<def> %R9D<kill> %R9<def>
 ; AVX1-NEXT:    # kill: %R8D<def> %R8D<kill> %R8<def>
 ; AVX1-NEXT:    # kill: %ECX<def> %ECX<kill> %RCX<def>
 ; AVX1-NEXT:    # kill: %EDX<def> %EDX<kill> %RDX<def>
 ; AVX1-NEXT:    # kill: %ESI<def> %ESI<kill> %RSI<def>
 ; AVX1-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; AVX1-NEXT:    andl $7, %edi
 ; AVX1-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1-NEXT:    movzwl -24(%rsp,%rdi,2), %eax
+; AVX1-NEXT:    vmovd %eax, %xmm0
+; AVX1-NEXT:    andl $7, %esi
+; AVX1-NEXT:    vpinsrw $1, -24(%rsp,%rsi,2), %xmm0, %xmm0
+; AVX1-NEXT:    andl $7, %edx
+; AVX1-NEXT:    vpinsrw $2, -24(%rsp,%rdx,2), %xmm0, %xmm0
+; AVX1-NEXT:    andl $7, %ecx
+; AVX1-NEXT:    vpinsrw $3, -24(%rsp,%rcx,2), %xmm0, %xmm0
+; AVX1-NEXT:    andl $7, %r8d
+; AVX1-NEXT:    vpinsrw $4, -24(%rsp,%r8,2), %xmm0, %xmm0
+; AVX1-NEXT:    andl $7, %r9d
+; AVX1-NEXT:    vpinsrw $5, -24(%rsp,%r9,2), %xmm0, %xmm0
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
-; AVX1-NEXT:    movzwl -24(%rsp,%rax,2), %eax
-; AVX1-NEXT:    vmovd %eax, %xmm0
+; AVX1-NEXT:    vpinsrw $6, -24(%rsp,%rax,2), %xmm0, %xmm0
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
-; AVX1-NEXT:    vpinsrw $1, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $7, -24(%rsp,%rax,2), %xmm0, %xmm0
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
-; AVX1-NEXT:    vpinsrw $2, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    movzwl -24(%rsp,%rax,2), %eax
+; AVX1-NEXT:    vmovd %eax, %xmm1
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
-; AVX1-NEXT:    vpinsrw $3, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $1, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
-; AVX1-NEXT:    vpinsrw $4, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $2, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
-; AVX1-NEXT:    vpinsrw $5, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $3, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
-; AVX1-NEXT:    vpinsrw $6, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX1-NEXT:    vpinsrw $4, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
-; AVX1-NEXT:    vpinsrw $7, -24(%rsp,%rax,2), %xmm0, %xmm0
-; AVX1-NEXT:    andl $7, %edi
-; AVX1-NEXT:    movzwl -24(%rsp,%rdi,2), %eax
-; AVX1-NEXT:    vmovd %eax, %xmm1
-; AVX1-NEXT:    andl $7, %esi
-; AVX1-NEXT:    vpinsrw $1, -24(%rsp,%rsi,2), %xmm1, %xmm1
-; AVX1-NEXT:    andl $7, %edx
-; AVX1-NEXT:    vpinsrw $2, -24(%rsp,%rdx,2), %xmm1, %xmm1
-; AVX1-NEXT:    andl $7, %ecx
-; AVX1-NEXT:    vpinsrw $3, -24(%rsp,%rcx,2), %xmm1, %xmm1
-; AVX1-NEXT:    andl $7, %r8d
-; AVX1-NEXT:    vpinsrw $4, -24(%rsp,%r8,2), %xmm1, %xmm1
-; AVX1-NEXT:    andl $7, %r9d
-; AVX1-NEXT:    vpinsrw $5, -24(%rsp,%r9,2), %xmm1, %xmm1
+; AVX1-NEXT:    vpinsrw $5, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
 ; AVX1-NEXT:    vpinsrw $6, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX1-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX1-NEXT:    andl $7, %eax
 ; AVX1-NEXT:    vpinsrw $7, -24(%rsp,%rax,2), %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shuffle_v16i16_v8i16_xxxxxxxxxxxxxxxx_i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    # kill: %R9D<def> %R9D<kill> %R9<def>
 ; AVX2-NEXT:    # kill: %R8D<def> %R8D<kill> %R8<def>
 ; AVX2-NEXT:    # kill: %ECX<def> %ECX<kill> %RCX<def>
 ; AVX2-NEXT:    # kill: %EDX<def> %EDX<kill> %RDX<def>
 ; AVX2-NEXT:    # kill: %ESI<def> %ESI<kill> %RSI<def>
 ; AVX2-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; AVX2-NEXT:    andl $7, %edi
 ; AVX2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX2-NEXT:    movzwl -24(%rsp,%rdi,2), %eax
+; AVX2-NEXT:    vmovd %eax, %xmm0
+; AVX2-NEXT:    andl $7, %esi
+; AVX2-NEXT:    vpinsrw $1, -24(%rsp,%rsi,2), %xmm0, %xmm0
+; AVX2-NEXT:    andl $7, %edx
+; AVX2-NEXT:    vpinsrw $2, -24(%rsp,%rdx,2), %xmm0, %xmm0
+; AVX2-NEXT:    andl $7, %ecx
+; AVX2-NEXT:    vpinsrw $3, -24(%rsp,%rcx,2), %xmm0, %xmm0
+; AVX2-NEXT:    andl $7, %r8d
+; AVX2-NEXT:    vpinsrw $4, -24(%rsp,%r8,2), %xmm0, %xmm0
+; AVX2-NEXT:    andl $7, %r9d
+; AVX2-NEXT:    vpinsrw $5, -24(%rsp,%r9,2), %xmm0, %xmm0
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
-; AVX2-NEXT:    movzwl -24(%rsp,%rax,2), %eax
-; AVX2-NEXT:    vmovd %eax, %xmm0
+; AVX2-NEXT:    vpinsrw $6, -24(%rsp,%rax,2), %xmm0, %xmm0
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
-; AVX2-NEXT:    vpinsrw $1, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $7, -24(%rsp,%rax,2), %xmm0, %xmm0
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
-; AVX2-NEXT:    vpinsrw $2, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    movzwl -24(%rsp,%rax,2), %eax
+; AVX2-NEXT:    vmovd %eax, %xmm1
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
-; AVX2-NEXT:    vpinsrw $3, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $1, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
-; AVX2-NEXT:    vpinsrw $4, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $2, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
-; AVX2-NEXT:    vpinsrw $5, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $3, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
-; AVX2-NEXT:    vpinsrw $6, -24(%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpinsrw $4, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
-; AVX2-NEXT:    vpinsrw $7, -24(%rsp,%rax,2), %xmm0, %xmm0
-; AVX2-NEXT:    andl $7, %edi
-; AVX2-NEXT:    movzwl -24(%rsp,%rdi,2), %eax
-; AVX2-NEXT:    vmovd %eax, %xmm1
-; AVX2-NEXT:    andl $7, %esi
-; AVX2-NEXT:    vpinsrw $1, -24(%rsp,%rsi,2), %xmm1, %xmm1
-; AVX2-NEXT:    andl $7, %edx
-; AVX2-NEXT:    vpinsrw $2, -24(%rsp,%rdx,2), %xmm1, %xmm1
-; AVX2-NEXT:    andl $7, %ecx
-; AVX2-NEXT:    vpinsrw $3, -24(%rsp,%rcx,2), %xmm1, %xmm1
-; AVX2-NEXT:    andl $7, %r8d
-; AVX2-NEXT:    vpinsrw $4, -24(%rsp,%r8,2), %xmm1, %xmm1
-; AVX2-NEXT:    andl $7, %r9d
-; AVX2-NEXT:    vpinsrw $5, -24(%rsp,%r9,2), %xmm1, %xmm1
+; AVX2-NEXT:    vpinsrw $5, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
 ; AVX2-NEXT:    vpinsrw $6, -24(%rsp,%rax,2), %xmm1, %xmm1
 ; AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
 ; AVX2-NEXT:    andl $7, %eax
 ; AVX2-NEXT:    vpinsrw $7, -24(%rsp,%rax,2), %xmm1, %xmm1
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %x0  = extractelement <8 x i16> %x, i32 %i0
   %x1  = extractelement <8 x i16> %x, i32 %i1
   %x2  = extractelement <8 x i16> %x, i32 %i2
   %x3  = extractelement <8 x i16> %x, i32 %i3
   %x4  = extractelement <8 x i16> %x, i32 %i4
   %x5  = extractelement <8 x i16> %x, i32 %i5
   %x6  = extractelement <8 x i16> %x, i32 %i6