[llvm-commits] [PATCH] va_arg alignment type

[Rafael Espindola]

On 13 October 2010 22:21, David Meyer <pdox at google.com> wrote:
> Rafael,
> I'm afraid not. <2 x i64> is aligned to 8-bytes on X86-64, both ABI and
> stack.
> I'm not sure there's any way to "test" this change... it may have no
> external effects (except for Native Client). All major architectures have
> stack alignment the same as ABI alignment. In principle, however, I think
> this change is correct.

I tested

define <2 x i64> @test(i32 %a, i32 %b, ...) nounwind optsize {
entry:
  %ap = alloca i8*, align 4                       ; <i8**> [#uses=3]
  %ap1 = bitcast i8** %ap to i8*                  ; <i8*> [#uses=2]
  call void @llvm.va_start(i8* %ap1)
  %0 = va_arg i8** %ap, i32                       ; <i32> [#uses=0]
  %1 = va_arg i8** %ap,  <2 x i64>                    ; <double> [#uses=1]
  call void @llvm.va_end(i8* %ap1)
  ret  <2 x i64> %1
}
declare void @llvm.va_start(i8*) nounwind
declare void @llvm.va_end(i8*) nounwind

With and without your patch. Without your patch there is an extra

 	addq	$15, %rcx
 	andq	$-16, %rcx

in the output. So it is really reducing the alignment requirement of
<2 x i64>. I think but I am not sure that this is a fix and not a
regression :-)

So, is this what you were expecting?

> - David M

Cheers,
-- 
Rafael Ávila de Espíndola