Re: [Mingw-users] GCC 4.4.0 failed to compile GSL 1.12

[Massimo G. <mas...@al...>]

Roumen Petrov wrote:
> Massimo Gaspari wrote:
>   
>> Aaron W. LaFramboise wrote
>>     
>>>> Can you try adding -ffloat-store to the CFLAGS and rebuilding from 
>>>> scratch to see if this helps?  This often solves rounding issues such as 
>>>> these.
>>>>
>>>>     
>>>>         
>> I know.
>> With -ffloat-store the issue disappear. Moreover the well known bug in 
>> vegas (montecarlo) is still present, it can be avoided with -ffloat-store
>>
>> Max
>>     
>
> May be issue is related to precision supported by x86 FPU (long double 
> vs double) and cache of results in registers. Explicit conversion (cast) 
> to double in code may resolve issue for any compiler.
>
> I'm not sure that this is a GCC bug.
>
> Roumen
>   

May be you are right. But GSL is developed using GCC/Linux  x86 and the 
bugs are not present there ( at least using Gcc 4.2.x). Moreover, as far 
as I remember, the bugs are absent in cygwin. But I'm no longer using 
cygwin and so I cannot test this statement.
The list of supported platform (mostly gcc based) is available at 
http://www.gnu.org/software/gsl/, MinGW was present but then removed 
because of the reported bug (the montecarlo one).
Brian Gough, the GSL maintainer, analyzed the montecarlo bugs and he is 
confident it's a compiler bug.
There was also a similar bug in poly and so you can read in the 
installation instruction

|With gcc-4.3.2 the tests in the poly/ directory fail due to a
compiler optimization bug which is fixed in later versions of GCC



Max

|