[SelectionDAG] Enable lowering unordered atomics loads w/LoadSDNode (and stores w/StoreSDNode) by default

[reames]

Summary

At this point, the experimental lowering introduced w/r371441 appears to be solidly correct - it's survived a good amount of fuzzing and a manual audit of differences in a fairly large corpus - and generated code which is mostly better than the current default. (There are a few cases which are worse in the test diffs, but nothing which looks concerning.)

As a reminder, the new lowering changes the representation of an unordered atomic load from an AtomicSDNode - which is essentially a black box which gets passed through without combines messing with it - to a LoadSDNode w/a atomic marker on the MMO. The later parallels the way we handle volatiles, and I've audited the code to ensure that every location which checks one checks the other.

Once this patch lands, the next steps are:

1. Wait a week or two to ensure nothing falls out.
2. Remove the experimental flag, consolidate some of the code in SelectionDAG as noted in existing todos, delete now stale td rules - i.e. prepare for this being the default long term
3. Go through and implement a few of the individual TODOs (performance opportunities to improve ISEL) previously noted

[reames]

ping

[arsenm]

LGTM

[reames]

It looks like I screwed up somewhere. I went to land the approached patch, ran make check and saw a bunch of unexpected failures.

Looking into the diffs thereof, it appears the new implementation is still buggy. We appear to be splitting an atomic load/store in some cases rather than emitting a library call. This is definitely incorrect.

As such, I'm considering the LGTMs here revoked, and will refresh this patch once issue is fixed.

[qcolombet]

Thanks Philip for double checking.

[reames]

Add the missing type legalization, and asserts. I had failed to account for the fact that we don't expand atomic loads/stores which aren't natively legal, but can be represented as cmpxchg in IR, despite having a mechanism to do so.

This version looks to correct the bug in the test cases spotted, but I'm still nervous. There's code in LegalizeDAG.cpp (as opposed to type legalization) for ATOMIC_LOAD and ATOMIC_STORE which I don't have a parallel for, and which appears to be dead. (I added assert(false) and no x86 tests failed.) I don't really know what to do beyond that though.

[arsenm]

LGTM

[reames]

This landed. For some reason the normal auto-close didn't trigger.

Author: Philip Reames
Date: 2019-10-29T12:46:24-07:00
New Revision: 2460989eabb8adca4b7973aa23997b5ec832219b

URL: https://github.com/llvm/llvm-project/commit/2460989eabb8adca4b7973aa23997b5ec832219b
DIFF: https://github.com/llvm/llvm-project/commit/2460989eabb8adca4b7973aa23997b5ec832219b.diff

LOG: [SelectionDAG] Enable lowering unordered atomics loads w/LoadSDNode (and stores w/StoreSDNode) by default

Enable the new SelectionDAG representation for unordered loads and stores introduced in r371441 by default. As a reminder, the new lowering changes the representation of an unordered atomic load from an AtomicSDNode - which is essentially a black box which gets passed through without combines messing with it - to a LoadSDNode w/a atomic marker on the MMO. The later parallels the way we handle volatiles, and I've audited the code to ensure that every location which checks one checks the other.

This has been fairly heavily fuzzed, and I examined diffs in a reasonable large corpus of assembly by hand, so I'm reasonable sure this is correct for the common case. Late in the review for this, it was discovered that I hadn't correctly handled cases which could be legalized into CAS operations. This points out that there's a strong bias in the IR of the frontend I'm working with towards only legal atomics. If there are problems with this patch, the most likely area will be legalization.

Differential Revision: https://reviews.llvm.org/D69219

llvm/lib/Target/X86/X86ISelLowering.cpp

 
 static cl::opt<bool> MulConstantOptimization(
     "mul-constant-optimization", cl::init(true),
     cl::desc("Replace 'mul x, Const' with more effective instructions like "
              "SHIFT, LEA, etc."),
     cl::Hidden);
 
 static cl::opt<bool> ExperimentalUnorderedISEL(
-    "x86-experimental-unordered-atomic-isel", cl::init(false),
+    "x86-experimental-unordered-atomic-isel", cl::init(true),
     cl::desc("Use LoadSDNode and StoreSDNode instead of "
              "AtomicSDNode for unordered atomic loads and "
              "stores respectively."),
     cl::Hidden);
 
 /// Call this when the user attempts to do something unsupported, like
 /// returning a double without SSE2 enabled on x86_64. This is not fatal, unlike
 /// report_fatal_error, so calling code should attempt to recover without

llvm/test/CodeGen/X86/atomic-non-integer-fp128.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mattr=-sse | FileCheck %s --check-prefix=X64-NOSSE
 ; RUN: llc < %s -mtriple=x86_64-linux-generic -verify-machineinstrs | FileCheck %s --check-prefix=X64-SSE
 
 ; Note: This test is testing that the lowering for atomics matches what we
 ; currently emit for non-atomics + the atomic restriction.  The presence of
 ; particular lowering detail in these tests should not be read as requiring
 ; that detail for correctness unless it's related to the atomicity itself.
 ; (Specifically, there were reviewer questions about the lowering for halfs
 ;  and their calling convention which remain unresolved.)
 
 define void @store_fp128(fp128* %fptr, fp128 %v) {
 ; X64-NOSSE-LABEL: store_fp128:
 ; X64-NOSSE:       # %bb.0:
-; X64-NOSSE-NEXT:    pushq %rax
-; X64-NOSSE-NEXT:    .cfi_def_cfa_offset 16
-; X64-NOSSE-NEXT:    callq __sync_lock_test_and_set_16
-; X64-NOSSE-NEXT:    popq %rax
-; X64-NOSSE-NEXT:    .cfi_def_cfa_offset 8
+; X64-NOSSE-NEXT:    movq %rdx, 8(%rdi)
+; X64-NOSSE-NEXT:    movq %rsi, (%rdi)
 ; X64-NOSSE-NEXT:    retq
 ;
 ; X64-SSE-LABEL: store_fp128:
 ; X64-SSE:       # %bb.0:
-; X64-SSE-NEXT:    subq $24, %rsp
-; X64-SSE-NEXT:    .cfi_def_cfa_offset 32
-; X64-SSE-NEXT:    movaps %xmm0, (%rsp)
-; X64-SSE-NEXT:    movq (%rsp), %rsi
-; X64-SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rdx
-; X64-SSE-NEXT:    callq __sync_lock_test_and_set_16
-; X64-SSE-NEXT:    addq $24, %rsp
-; X64-SSE-NEXT:    .cfi_def_cfa_offset 8
+; X64-SSE-NEXT:    movaps %xmm0, (%rdi)
 ; X64-SSE-NEXT:    retq
   store atomic fp128 %v, fp128* %fptr unordered, align 16
   ret void
 }

llvm/test/CodeGen/X86/atomic-non-integer.ll

 ; X64-AVX512-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX512-NEXT:    movw %ax, (%rdi)
 ; X64-AVX512-NEXT:    retq
   store atomic half %v, half* %fptr unordered, align 2
   ret void
 }
 
 define void @store_float(float* %fptr, float %v) {
-; X86-LABEL: store_float:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl %ecx, (%eax)
-; X86-NEXT:    retl
+; X86-SSE-LABEL: store_float:
+; X86-SSE:       # %bb.0:
+; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE-NEXT:    movss %xmm0, (%eax)
+; X86-SSE-NEXT:    retl
+;
+; X86-AVX-LABEL: store_float:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX-NEXT:    vmovss %xmm0, (%eax)
+; X86-AVX-NEXT:    retl
+;
+; X86-NOSSE-LABEL: store_float:
+; X86-NOSSE:       # %bb.0:
+; X86-NOSSE-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOSSE-NEXT:    fstps (%eax)
+; X86-NOSSE-NEXT:    retl
 ;
 ; X64-SSE-LABEL: store_float:
 ; X64-SSE:       # %bb.0:
 ; X64-SSE-NEXT:    movss %xmm0, (%rdi)
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX-LABEL: store_float:
 ; X64-AVX:       # %bb.0:

 ; X86-SSE1-NEXT:    popl %esi
 ; X86-SSE1-NEXT:    .cfi_def_cfa_offset 8
 ; X86-SSE1-NEXT:    popl %ebx
 ; X86-SSE1-NEXT:    .cfi_def_cfa_offset 4
 ; X86-SSE1-NEXT:    retl
 ;
 ; X86-SSE2-LABEL: store_double:
 ; X86-SSE2:       # %bb.0:
-; X86-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SSE2-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
-; X86-SSE2-NEXT:    movlps %xmm0, (%eax)
+; X86-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE2-NEXT:    movsd %xmm0, (%eax)
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-AVX-LABEL: store_double:
 ; X86-AVX:       # %bb.0:
-; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-AVX-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; X86-AVX-NEXT:    vmovlps %xmm0, (%eax)
+; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX-NEXT:    vmovsd %xmm0, (%eax)
 ; X86-AVX-NEXT:    retl
 ;
 ; X86-NOSSE-LABEL: store_double:
 ; X86-NOSSE:       # %bb.0:
 ; X86-NOSSE-NEXT:    pushl %ebx
 ; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 8
 ; X86-NOSSE-NEXT:    pushl %esi
 ; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 12

 ; X64-AVX-NEXT:    retq
   store atomic double %v, double* %fptr unordered, align 8
   ret void
 }
 
 define void @store_fp128(fp128* %fptr, fp128 %v) {
 ; X86-SSE-LABEL: store_fp128:
 ; X86-SSE:       # %bb.0:
-; X86-SSE-NEXT:    subl $36, %esp
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 36
-; X86-SSE-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X86-SSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl %eax
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    calll __sync_lock_test_and_set_16
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset -4
-; X86-SSE-NEXT:    addl $56, %esp
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset -56
+; X86-SSE-NEXT:    pushl %edi
+; X86-SSE-NEXT:    .cfi_def_cfa_offset 8
+; X86-SSE-NEXT:    pushl %esi
+; X86-SSE-NEXT:    .cfi_def_cfa_offset 12
+; X86-SSE-NEXT:    .cfi_offset %esi, -12
+; X86-SSE-NEXT:    .cfi_offset %edi, -8
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-SSE-NEXT:    movl %esi, 12(%edi)
+; X86-SSE-NEXT:    movl %edx, 8(%edi)
+; X86-SSE-NEXT:    movl %ecx, 4(%edi)
+; X86-SSE-NEXT:    movl %eax, (%edi)
+; X86-SSE-NEXT:    popl %esi
+; X86-SSE-NEXT:    .cfi_def_cfa_offset 8
+; X86-SSE-NEXT:    popl %edi
+; X86-SSE-NEXT:    .cfi_def_cfa_offset 4
 ; X86-SSE-NEXT:    retl
 ;
 ; X86-AVX-LABEL: store_fp128:
 ; X86-AVX:       # %bb.0:
-; X86-AVX-NEXT:    subl $44, %esp
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 48
-; X86-AVX-NEXT:    vmovaps {{[0-9]+}}(%esp), %xmm0
+; X86-AVX-NEXT:    vmovups {{[0-9]+}}(%esp), %xmm0
 ; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-AVX-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-AVX-NEXT:    vmovups %xmm0, {{[0-9]+}}(%esp)
-; X86-AVX-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X86-AVX-NEXT:    movl %eax, (%esp)
-; X86-AVX-NEXT:    calll __sync_lock_test_and_set_16
-; X86-AVX-NEXT:    addl $40, %esp
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 4
+; X86-AVX-NEXT:    vmovaps %xmm0, (%eax)
 ; X86-AVX-NEXT:    retl
 ;
 ; X86-NOSSE-LABEL: store_fp128:
 ; X86-NOSSE:       # %bb.0:
-; X86-NOSSE-NEXT:    subl $36, %esp
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 36
-; X86-NOSSE-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl %eax
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    calll __sync_lock_test_and_set_16
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset -4
-; X86-NOSSE-NEXT:    addl $56, %esp
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset -56
+; X86-NOSSE-NEXT:    pushl %edi
+; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 8
+; X86-NOSSE-NEXT:    pushl %esi
+; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 12
+; X86-NOSSE-NEXT:    .cfi_offset %esi, -12
+; X86-NOSSE-NEXT:    .cfi_offset %edi, -8
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NOSSE-NEXT:    movl %esi, 12(%edi)
+; X86-NOSSE-NEXT:    movl %edx, 8(%edi)
+; X86-NOSSE-NEXT:    movl %ecx, 4(%edi)
+; X86-NOSSE-NEXT:    movl %eax, (%edi)
+; X86-NOSSE-NEXT:    popl %esi
+; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 8
+; X86-NOSSE-NEXT:    popl %edi
+; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 4
 ; X86-NOSSE-NEXT:    retl
 ;
 ; X64-SSE-LABEL: store_fp128:
 ; X64-SSE:       # %bb.0:
-; X64-SSE-NEXT:    subq $24, %rsp
-; X64-SSE-NEXT:    .cfi_def_cfa_offset 32
-; X64-SSE-NEXT:    movaps %xmm0, (%rsp)
-; X64-SSE-NEXT:    movq (%rsp), %rsi
-; X64-SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rdx
-; X64-SSE-NEXT:    callq __sync_lock_test_and_set_16
-; X64-SSE-NEXT:    addq $24, %rsp
-; X64-SSE-NEXT:    .cfi_def_cfa_offset 8
+; X64-SSE-NEXT:    movaps %xmm0, (%rdi)
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX-LABEL: store_fp128:
 ; X64-AVX:       # %bb.0:
-; X64-AVX-NEXT:    subq $24, %rsp
-; X64-AVX-NEXT:    .cfi_def_cfa_offset 32
-; X64-AVX-NEXT:    vmovaps %xmm0, (%rsp)
-; X64-AVX-NEXT:    movq (%rsp), %rsi
-; X64-AVX-NEXT:    movq {{[0-9]+}}(%rsp), %rdx
-; X64-AVX-NEXT:    callq __sync_lock_test_and_set_16
-; X64-AVX-NEXT:    addq $24, %rsp
-; X64-AVX-NEXT:    .cfi_def_cfa_offset 8
+; X64-AVX-NEXT:    vmovaps %xmm0, (%rdi)
 ; X64-AVX-NEXT:    retq
   store atomic fp128 %v, fp128* %fptr unordered, align 16
   ret void
 }
 
 define half @load_half(half* %fptr) {
 ; X86-SSE-LABEL: load_half:
 ; X86-SSE:       # %bb.0:

 ; X64-AVX512-NEXT:    vmovd %eax, %xmm0
 ; X64-AVX512-NEXT:    vcvtph2ps %xmm0, %xmm0
 ; X64-AVX512-NEXT:    retq
   %v = load atomic half, half* %fptr unordered, align 2
   ret half %v
 }
 
 define float @load_float(float* %fptr) {
-; X86-SSE1-LABEL: load_float:
-; X86-SSE1:       # %bb.0:
-; X86-SSE1-NEXT:    pushl %eax
-; X86-SSE1-NEXT:    .cfi_def_cfa_offset 8
-; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SSE1-NEXT:    movl (%eax), %eax
-; X86-SSE1-NEXT:    movl %eax, (%esp)
-; X86-SSE1-NEXT:    flds (%esp)
-; X86-SSE1-NEXT:    popl %eax
-; X86-SSE1-NEXT:    .cfi_def_cfa_offset 4
-; X86-SSE1-NEXT:    retl
-;
-; X86-SSE2-LABEL: load_float:
-; X86-SSE2:       # %bb.0:
-; X86-SSE2-NEXT:    pushl %eax
-; X86-SSE2-NEXT:    .cfi_def_cfa_offset 8
-; X86-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SSE2-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE2-NEXT:    movss %xmm0, (%esp)
-; X86-SSE2-NEXT:    flds (%esp)
-; X86-SSE2-NEXT:    popl %eax
-; X86-SSE2-NEXT:    .cfi_def_cfa_offset 4
-; X86-SSE2-NEXT:    retl
-;
-; X86-AVX-LABEL: load_float:
-; X86-AVX:       # %bb.0:
-; X86-AVX-NEXT:    pushl %eax
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 8
-; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-AVX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-AVX-NEXT:    vmovss %xmm0, (%esp)
-; X86-AVX-NEXT:    flds (%esp)
-; X86-AVX-NEXT:    popl %eax
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 4
-; X86-AVX-NEXT:    retl
-;
-; X86-NOSSE-LABEL: load_float:
-; X86-NOSSE:       # %bb.0:
-; X86-NOSSE-NEXT:    pushl %eax
-; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 8
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl (%eax), %eax
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    flds (%esp)
-; X86-NOSSE-NEXT:    popl %eax
-; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 4
-; X86-NOSSE-NEXT:    retl
+; X86-LABEL: load_float:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    flds (%eax)
+; X86-NEXT:    retl
 ;
 ; X64-SSE-LABEL: load_float:
 ; X64-SSE:       # %bb.0:
 ; X64-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX-LABEL: load_float:
 ; X64-AVX:       # %bb.0:
 ; X64-AVX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X64-AVX-NEXT:    retq
   %v = load atomic float, float* %fptr unordered, align 4
   ret float %v
 }
 
 define double @load_double(double* %fptr) {
-; X86-SSE1-LABEL: load_double:
-; X86-SSE1:       # %bb.0:
-; X86-SSE1-NEXT:    subl $20, %esp
-; X86-SSE1-NEXT:    .cfi_def_cfa_offset 24
-; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SSE1-NEXT:    fildll (%eax)
-; X86-SSE1-NEXT:    fistpll {{[0-9]+}}(%esp)
-; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-SSE1-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-SSE1-NEXT:    movl %eax, (%esp)
-; X86-SSE1-NEXT:    fldl (%esp)
-; X86-SSE1-NEXT:    addl $20, %esp
-; X86-SSE1-NEXT:    .cfi_def_cfa_offset 4
-; X86-SSE1-NEXT:    retl
-;
-; X86-SSE2-LABEL: load_double:
-; X86-SSE2:       # %bb.0:
-; X86-SSE2-NEXT:    subl $12, %esp
-; X86-SSE2-NEXT:    .cfi_def_cfa_offset 16
-; X86-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SSE2-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
-; X86-SSE2-NEXT:    movlps %xmm0, (%esp)
-; X86-SSE2-NEXT:    fldl (%esp)
-; X86-SSE2-NEXT:    addl $12, %esp
-; X86-SSE2-NEXT:    .cfi_def_cfa_offset 4
-; X86-SSE2-NEXT:    retl
-;
-; X86-AVX-LABEL: load_double:
-; X86-AVX:       # %bb.0:
-; X86-AVX-NEXT:    subl $12, %esp
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 16
-; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-AVX-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; X86-AVX-NEXT:    vmovlps %xmm0, (%esp)
-; X86-AVX-NEXT:    fldl (%esp)
-; X86-AVX-NEXT:    addl $12, %esp
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 4
-; X86-AVX-NEXT:    retl
-;
-; X86-NOSSE-LABEL: load_double:
-; X86-NOSSE:       # %bb.0:
-; X86-NOSSE-NEXT:    subl $20, %esp
-; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 24
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    fildll (%eax)
-; X86-NOSSE-NEXT:    fistpll {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl %ecx, {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    movl %eax, (%esp)
-; X86-NOSSE-NEXT:    fldl (%esp)
-; X86-NOSSE-NEXT:    addl $20, %esp
-; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 4
-; X86-NOSSE-NEXT:    retl
+; X86-LABEL: load_double:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    fldl (%eax)
+; X86-NEXT:    retl
 ;
 ; X64-SSE-LABEL: load_double:
 ; X64-SSE:       # %bb.0:
 ; X64-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX-LABEL: load_double:
 ; X64-AVX:       # %bb.0:
 ; X64-AVX-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
 ; X64-AVX-NEXT:    retq
   %v = load atomic double, double* %fptr unordered, align 8
   ret double %v
 }
 
 define fp128 @load_fp128(fp128* %fptr) {
-; X86-SSE-LABEL: load_fp128:
-; X86-SSE:       # %bb.0:
-; X86-SSE-NEXT:    pushl %edi
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 8
-; X86-SSE-NEXT:    pushl %esi
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 12
-; X86-SSE-NEXT:    subl $20, %esp
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 32
-; X86-SSE-NEXT:    .cfi_offset %esi, -12
-; X86-SSE-NEXT:    .cfi_offset %edi, -8
-; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SSE-NEXT:    subl $8, %esp
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 8
-; X86-SSE-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X86-SSE-NEXT:    pushl $0
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl $0
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl $0
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl $0
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl $0
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl $0
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl $0
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl $0
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    pushl %eax
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-SSE-NEXT:    calll __sync_val_compare_and_swap_16
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset -4
-; X86-SSE-NEXT:    addl $44, %esp
-; X86-SSE-NEXT:    .cfi_adjust_cfa_offset -44
-; X86-SSE-NEXT:    movl (%esp), %eax
-; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SSE-NEXT:    movl %edi, 8(%esi)
-; X86-SSE-NEXT:    movl %edx, 12(%esi)
-; X86-SSE-NEXT:    movl %eax, (%esi)
-; X86-SSE-NEXT:    movl %ecx, 4(%esi)
-; X86-SSE-NEXT:    movl %esi, %eax
-; X86-SSE-NEXT:    addl $20, %esp
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 12
-; X86-SSE-NEXT:    popl %esi
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 8
-; X86-SSE-NEXT:    popl %edi
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 4
-; X86-SSE-NEXT:    retl $4
+; X86-SSE1-LABEL: load_fp128:
+; X86-SSE1:       # %bb.0:
+; X86-SSE1-NEXT:    pushl %edi
+; X86-SSE1-NEXT:    .cfi_def_cfa_offset 8
+; X86-SSE1-NEXT:    pushl %esi
+; X86-SSE1-NEXT:    .cfi_def_cfa_offset 12
+; X86-SSE1-NEXT:    .cfi_offset %esi, -12
+; X86-SSE1-NEXT:    .cfi_offset %edi, -8
+; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SSE1-NEXT:    movl (%ecx), %edx
+; X86-SSE1-NEXT:    movl 4(%ecx), %esi
+; X86-SSE1-NEXT:    movl 8(%ecx), %edi
+; X86-SSE1-NEXT:    movl 12(%ecx), %ecx
+; X86-SSE1-NEXT:    movl %ecx, 12(%eax)
+; X86-SSE1-NEXT:    movl %edi, 8(%eax)
+; X86-SSE1-NEXT:    movl %esi, 4(%eax)
+; X86-SSE1-NEXT:    movl %edx, (%eax)
+; X86-SSE1-NEXT:    popl %esi
+; X86-SSE1-NEXT:    .cfi_def_cfa_offset 8
+; X86-SSE1-NEXT:    popl %edi
+; X86-SSE1-NEXT:    .cfi_def_cfa_offset 4
+; X86-SSE1-NEXT:    retl $4
+;
+; X86-SSE2-LABEL: load_fp128:
+; X86-SSE2:       # %bb.0:
+; X86-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SSE2-NEXT:    movaps (%ecx), %xmm0
+; X86-SSE2-NEXT:    movaps %xmm0, (%eax)
+; X86-SSE2-NEXT:    retl $4
 ;
 ; X86-AVX-LABEL: load_fp128:
 ; X86-AVX:       # %bb.0:
-; X86-AVX-NEXT:    pushl %esi
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 8
-; X86-AVX-NEXT:    subl $56, %esp
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 64
-; X86-AVX-NEXT:    .cfi_offset %esi, -8
-; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-AVX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; X86-AVX-NEXT:    vmovups %ymm0, {{[0-9]+}}(%esp)
-; X86-AVX-NEXT:    movl %eax, {{[0-9]+}}(%esp)
-; X86-AVX-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X86-AVX-NEXT:    movl %eax, (%esp)
-; X86-AVX-NEXT:    vzeroupper
-; X86-AVX-NEXT:    calll __sync_val_compare_and_swap_16
-; X86-AVX-NEXT:    subl $4, %esp
-; X86-AVX-NEXT:    vmovups {{[0-9]+}}(%esp), %xmm0
-; X86-AVX-NEXT:    vmovaps %xmm0, (%esi)
-; X86-AVX-NEXT:    movl %esi, %eax
-; X86-AVX-NEXT:    addl $56, %esp
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 8
-; X86-AVX-NEXT:    popl %esi
-; X86-AVX-NEXT:    .cfi_def_cfa_offset 4
+; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-AVX-NEXT:    vmovaps (%ecx), %xmm0
+; X86-AVX-NEXT:    vmovaps %xmm0, (%eax)
 ; X86-AVX-NEXT:    retl $4
 ;
 ; X86-NOSSE-LABEL: load_fp128:
 ; X86-NOSSE:       # %bb.0:
 ; X86-NOSSE-NEXT:    pushl %edi
 ; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 8
 ; X86-NOSSE-NEXT:    pushl %esi
 ; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 12
-; X86-NOSSE-NEXT:    subl $20, %esp
-; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 32
 ; X86-NOSSE-NEXT:    .cfi_offset %esi, -12
 ; X86-NOSSE-NEXT:    .cfi_offset %edi, -8
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NOSSE-NEXT:    subl $8, %esp
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 8
-; X86-NOSSE-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X86-NOSSE-NEXT:    pushl $0
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl $0
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl $0
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl $0
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl $0
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl $0
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl $0
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl $0
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl {{[0-9]+}}(%esp)
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    pushl %eax
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset 4
-; X86-NOSSE-NEXT:    calll __sync_val_compare_and_swap_16
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset -4
-; X86-NOSSE-NEXT:    addl $44, %esp
-; X86-NOSSE-NEXT:    .cfi_adjust_cfa_offset -44
-; X86-NOSSE-NEXT:    movl (%esp), %eax
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-NOSSE-NEXT:    movl %edi, 8(%esi)
-; X86-NOSSE-NEXT:    movl %edx, 12(%esi)
-; X86-NOSSE-NEXT:    movl %eax, (%esi)
-; X86-NOSSE-NEXT:    movl %ecx, 4(%esi)
-; X86-NOSSE-NEXT:    movl %esi, %eax
-; X86-NOSSE-NEXT:    addl $20, %esp
-; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 12
+; X86-NOSSE-NEXT:    movl (%ecx), %edx
+; X86-NOSSE-NEXT:    movl 4(%ecx), %esi
+; X86-NOSSE-NEXT:    movl 8(%ecx), %edi
+; X86-NOSSE-NEXT:    movl 12(%ecx), %ecx
+; X86-NOSSE-NEXT:    movl %ecx, 12(%eax)
+; X86-NOSSE-NEXT:    movl %edi, 8(%eax)
+; X86-NOSSE-NEXT:    movl %esi, 4(%eax)
+; X86-NOSSE-NEXT:    movl %edx, (%eax)
 ; X86-NOSSE-NEXT:    popl %esi
 ; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 8
 ; X86-NOSSE-NEXT:    popl %edi
 ; X86-NOSSE-NEXT:    .cfi_def_cfa_offset 4
 ; X86-NOSSE-NEXT:    retl $4
 ;
 ; X64-SSE-LABEL: load_fp128:
 ; X64-SSE:       # %bb.0:
-; X64-SSE-NEXT:    subq $24, %rsp
-; X64-SSE-NEXT:    .cfi_def_cfa_offset 32
-; X64-SSE-NEXT:    xorl %esi, %esi
-; X64-SSE-NEXT:    xorl %edx, %edx
-; X64-SSE-NEXT:    xorl %ecx, %ecx
-; X64-SSE-NEXT:    xorl %r8d, %r8d
-; X64-SSE-NEXT:    callq __sync_val_compare_and_swap_16
-; X64-SSE-NEXT:    movq %rdx, {{[0-9]+}}(%rsp)
-; X64-SSE-NEXT:    movq %rax, (%rsp)
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0
-; X64-SSE-NEXT:    addq $24, %rsp
-; X64-SSE-NEXT:    .cfi_def_cfa_offset 8
+; X64-SSE-NEXT:    movaps (%rdi), %xmm0
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX-LABEL: load_fp128:
 ; X64-AVX:       # %bb.0:
-; X64-AVX-NEXT:    subq $24, %rsp
-; X64-AVX-NEXT:    .cfi_def_cfa_offset 32
-; X64-AVX-NEXT:    xorl %esi, %esi
-; X64-AVX-NEXT:    xorl %edx, %edx
-; X64-AVX-NEXT:    xorl %ecx, %ecx
-; X64-AVX-NEXT:    xorl %r8d, %r8d
-; X64-AVX-NEXT:    callq __sync_val_compare_and_swap_16
-; X64-AVX-NEXT:    movq %rdx, {{[0-9]+}}(%rsp)
-; X64-AVX-NEXT:    movq %rax, (%rsp)
-; X64-AVX-NEXT:    vmovaps (%rsp), %xmm0
-; X64-AVX-NEXT:    addq $24, %rsp
-; X64-AVX-NEXT:    .cfi_def_cfa_offset 8
+; X64-AVX-NEXT:    vmovaps (%rdi), %xmm0
 ; X64-AVX-NEXT:    retq
   %v = load atomic fp128, fp128* %fptr unordered, align 16
   ret fp128 %v
 }
 
 
 ; sanity check the seq_cst lowering since that's the
 ; interesting one from an ordering perspective on x86.

llvm/test/CodeGen/X86/atomic-unordered.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -O0 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake | FileCheck --check-prefixes=CHECK,CHECK-O0,CHECK-O0-CUR %s
-; RUN: llc -O3 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake | FileCheck --check-prefixes=CHECK,CHECK-O3,CHECK-O3-CUR %s
-; RUN: llc -O0 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake -x86-experimental-unordered-atomic-isel | FileCheck --check-prefixes=CHECK,CHECK-O0,CHECK-O0-EX %s
-; RUN: llc -O3 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake -x86-experimental-unordered-atomic-isel | FileCheck --check-prefixes=CHECK,CHECK-O3,CHECK-O3-EX %s
+; RUN: llc -O0 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake -x86-experimental-unordered-atomic-isel=0 | FileCheck --check-prefixes=CHECK,CHECK-O0,CHECK-O0-CUR %s
+; RUN: llc -O3 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake  -x86-experimental-unordered-atomic-isel=0 | FileCheck --check-prefixes=CHECK,CHECK-O3,CHECK-O3-CUR %s
+; RUN: llc -O0 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake -x86-experimental-unordered-atomic-isel=1 | FileCheck --check-prefixes=CHECK,CHECK-O0,CHECK-O0-EX %s
+; RUN: llc -O3 < %s -mtriple=x86_64-linux-generic -verify-machineinstrs -mcpu=skylake -x86-experimental-unordered-atomic-isel=1 | FileCheck --check-prefixes=CHECK,CHECK-O3,CHECK-O3-EX %s
 
 define i8 @load_i8(i8* %ptr) {
 ; CHECK-LABEL: load_i8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movb (%rdi), %al
 ; CHECK-NEXT:    retq
   %v = load atomic i8, i8* %ptr unordered, align 1
   ret i8 %v

llvm/test/CodeGen/X86/atomic128.ll

 ; CHECK-NEXT:    lock cmpxchg16b (%rdi)
 ; CHECK-NEXT:    jne LBB13_1
 ; CHECK-NEXT:  ## %bb.2: ## %atomicrmw.end
 ; CHECK-NEXT:    popq %rbx
 ; CHECK-NEXT:    retq
 ;
 ; CHECK32-LABEL: atomic_store_relaxed:
 ; CHECK32:       # %bb.0:
-; CHECK32-NEXT:    subl $36, %esp
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset 36
-; CHECK32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; CHECK32-NEXT:    pushl {{[0-9]+}}(%esp)
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset 4
-; CHECK32-NEXT:    pushl {{[0-9]+}}(%esp)
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset 4
-; CHECK32-NEXT:    pushl {{[0-9]+}}(%esp)
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset 4
-; CHECK32-NEXT:    pushl {{[0-9]+}}(%esp)
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset 4
-; CHECK32-NEXT:    pushl {{[0-9]+}}(%esp)
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset 4
-; CHECK32-NEXT:    pushl %eax
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset 4
-; CHECK32-NEXT:    calll __sync_lock_test_and_set_16
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset -4
-; CHECK32-NEXT:    addl $56, %esp
-; CHECK32-NEXT:    .cfi_adjust_cfa_offset -56
+; CHECK32-NEXT:    pushl %edi
+; CHECK32-NEXT:    .cfi_def_cfa_offset 8
+; CHECK32-NEXT:    pushl %esi
+; CHECK32-NEXT:    .cfi_def_cfa_offset 12
+; CHECK32-NEXT:    .cfi_offset %esi, -12
+; CHECK32-NEXT:    .cfi_offset %edi, -8
+; CHECK32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; CHECK32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; CHECK32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; CHECK32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; CHECK32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; CHECK32-NEXT:    movl %esi, 12(%edi)
+; CHECK32-NEXT:    movl %edx, 8(%edi)
+; CHECK32-NEXT:    movl %ecx, 4(%edi)
+; CHECK32-NEXT:    movl %eax, (%edi)
+; CHECK32-NEXT:    popl %esi
+; CHECK32-NEXT:    .cfi_def_cfa_offset 8
+; CHECK32-NEXT:    popl %edi
+; CHECK32-NEXT:    .cfi_def_cfa_offset 4
 ; CHECK32-NEXT:    retl
    store atomic i128 %in, i128* %p unordered, align 16
    ret void
 }
 
 

llvm/test/CodeGen/X86/combineIncDecVector-crash.ll

 ; CHECK-NEXT:    movq %rsi, {{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    movl %edi, {{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    movl $10, %esi
 ; CHECK-NEXT:    movl $10, %edx
 ; CHECK-NEXT:    movl $400, %ecx # imm = 0x190
 ; CHECK-NEXT:    callq newarray
 ; CHECK-NEXT:  .Ltmp0:
 ; CHECK-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; CHECK-NEXT:    addss (%rax), %xmm0
 ; CHECK-NEXT:    movdqu (%rax), %xmm1
 ; CHECK-NEXT:    pcmpeqd %xmm2, %xmm2
 ; CHECK-NEXT:    psubd %xmm2, %xmm1
 ; CHECK-NEXT:    movdqu %xmm1, (%rax)
+; CHECK-NEXT:    addss {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    movss %xmm0, (%rax)
 bci_0:
    %token418 = call token (i64, i32, i8 * (i64, i32, i32, i32)*, i32,
 i32, ...) @llvm.experimental.gc.statepoint.p0f_p1i8i64i32i32i32f(i64
 2882400000, i32 0, i8 * (i64, i32, i32, i32)* nonnull @newarray, i32 4,
 i32 0, i64 undef, i32 10, i32 10, i32 400, i32 0, i32 35, i32 0, i32 1,
 i32 0, i32 43, i32 1, i32 13, i32 0, i32 3, i32 400, i32 3, i32 %0, i32
 4, i64 %1, i32 7, i8* null, i32 3, i32 -11464, i32 7, i8* null, i32 3,