Re: svn/HEAD has regressions

I think that we've gone back and forth with respect to the
optimization level on libgdtoa.a.  Dropping back to -O1 would be fine.

It would probably also be worth checking the regressions with
'-codegen c', to see if there are any other problems with gcc 4.4.3-9.
 We do some casts from/to double/float to/from uint64_t/uint32_t (via
a memcpy, which is supposed to be kosher with respect to the C
aliasing rules).

On Sun, May 2, 2010 at 9:39 AM, Wesley W. Terpstra <[email protected]> wrote:
> On Sun, May 2, 2010 at 2:40 PM, Wesley W. Terpstra <[email protected]>
> wrote:
>>
>> My conclusion? gcc 4.4.3-9 is miscompiling our runtime, while 4.3.2-1.1
>> (lenny) and 4.4.3-2 (last sid upload) do not. I'll try to narrow down which
>> object file in particular causes the problem.
>
> It's runtime/gdtoa/gdtoa.o where the problem appears... and only on -O2.
> Without debugging further, one solution would be to compile gdtoa.a with
> -O1. :)
>