[PATCH] fix wrong code for push of long double

Uros Bizjak ubizjak@gmail.com
Thu Aug 6 23:55:00 GMT 2009


On 08/06/2009 09:34 PM, Kaveh R. GHAZI wrote:
> On Thu, 6 Aug 2009, Uros Bizjak wrote:
>
>    
>> On 08/06/2009 05:09 AM, Mikulas Patocka wrote:
>>
>>      
>>> My tests are valid on all architectures, they contain portable code (is
>>> there any arch without long double?). On other architectures they don't
>>> test any known bug, but they could be run anyway to protect the bug from
>>> reappearing. I don't see why you want to run them only on i386.
>>>
>>>        
>> x86_64 simply does not use these code paths, since you can't use pushq
>> together with pushl to move 3x 32bit decomposed XFmode operand.
>>      
>
> Uros,
>
> I see a new error on the 4.3 branch from this patch on x86_64 using -m32:
>
> http://gcc.gnu.org/ml/gcc-testresults/2009-08/msg00688.html
>
> FAIL: gcc.target/i386/pr40906-3.c (test for excess errors)
>
> The error is:
> 	.../gcc.target/i386/pr40906-3.c:8: error: expected ')' before 'a'
> 	.../gcc.target/i386/pr40906-3.c:15: error: expected ')' before 'b'
>
> ---
>    

I have removed wrong test, __float128 is not supported for i686 for 4.3.

> Unrelated to this patch, I also see errors on 4.3 and 4.4 on ssp-1.c when
> using -m32 and -fpic/-fPIC.  You may wish to take a look as well:
>
> FAIL: gcc.dg/ssp-1.c execution test
>
> 	*** stack smashing detected ***: ./ssp-1.exe terminated
>
> This one started around 7/23/2009 and may have been caused by one of your
> x86 checkins e.g. for PR40832.
>    

No, I don't think any of these patches caused this failure. 40832 only 
substitutes a .word with is equivalent define, and is not triggered in 
ssp-1.exe. The other one of my patches deals with vectorisation and is 
also not triggered for this case.

(Other patches are Alpha specific).

Uros.



More information about the Gcc-patches mailing list