Michael Paquier <michael.paquier@gmail.com> writes:
> On Tue, Nov 14, 2017 at 12:31 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Concretely, the attached patch fixes it for me. I've verified by
>> examining the assembly code that this stops gcc from using movdqa or
>> movaps in numeric.c, except for one place where it apparently can
>> prove that it's dealing with a sufficiently-aligned local variable.
> I am not seeing any difference in the assembly code generated by gcc
> -S with and without your patch. Perhaps I am missing something? What
> are you actually seeing?
Hm, double-check your procedure. I get this:
$ diff numeric.s.std numeric.s.patch
2746c2746
< movaps %xmm0, (%r14)
---
> movups %xmm0, (%r14)
13948,13951c13948,13951
< movdqa 16(%rbp), %xmm0
< movaps %xmm0, 16(%rax)
< movdqa 32(%rbp), %xmm0
< movaps %xmm0, 32(%rax)
---
> movdqu 16(%rbp), %xmm0
> movups %xmm0, 16(%rax)
> movdqu 32(%rbp), %xmm0
> movups %xmm0, 32(%rax)
14434,14435c14434,14435
< movdqa 16(%rbp), %xmm0
< movaps %xmm0, 16(%rax)
---
> movdqu 16(%rbp), %xmm0
> movups %xmm0, 16(%rax)
(If using -g, there are a bunch more diffs in the debug tables.)
regards, tom lane