[Bug target/66867] Suboptimal code generation for atomic_compare_exchange

[sebastian.huber@embedded-brains.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66867

--- Comment #3 from Sebastian Huber <sebastian.huber@embedded-brains.de> ---
clang 3.7 generates optimal code on x86 in both cases:

        .text
        .file   "test.c"
        .globl  f
        .align  16, 0x90
        .type   f,@function
f:                                      # @f
        .cfi_startproc
# BB#0:
        movl    $1, %ecx
        xorl    %eax, %eax
        lock            cmpxchgl        %ecx, (%rdi)
        retq
.Lfunc_end0:
        .size   f, .Lfunc_end0-f
        .cfi_endproc

        .globl  g
        .align  16, 0x90
        .type   g,@function
g:                                      # @g
        .cfi_startproc
# BB#0:
        movl    $1, %ecx
        xorl    %eax, %eax
        lock            cmpxchgl        %ecx, (%rdi)
        retq
.Lfunc_end1:
        .size   g, .Lfunc_end1-g
        .cfi_endproc


        .ident  "clang version 3.7.0 (tags/RELEASE_370/final)"
        .section        ".note.GNU-stack","",@progbits