Hi, While investigating a build failure (ITK on Debian [1]), I discovered that fftw3 now uses the GCC extension __float128 for quad precision arithmetic. In a nutshell, the following code builds OK using Debian's GCC 4.6.2-9 on both my amd64 machine and a 32-bit chroot on the same machine. Gccxml, however, differs: it parses OK on 64-bits, but complains when used with the 32-bit compiler.
int main(int ac, char* av[])
{
/* Modeled on code in fftw3.h, from about line 370 */
/* __float128 (quad precision) is a gcc extension on i386, x86_64, and
ia64
for gcc >= 4.6 (compiled in FFTW with --enable-quad-precision) */
#if (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)) \
&& (defined(__i386__) || defined(__x86_64__) || defined(__ia64__))
typedef __float128 quadcomplex[2];
#endif
return 0;
}
The 32-bit gccxml invocation produces this output:
steve@riemann{Testing}gccxml float128.c
float128.c: In function 'int main(int, char**)':
float128.c:13: error: '__float128' does not name a type
Is there an easy solution to this? I can make gccxml accept this
input by doing:
steve@riemann{Testing}gccxml -D__float128=double float128.c
But I'm afraid this will mess up the API of fftw output by gccxml. Ideas?
Thanks,
-Steve
[1]
https://buildd.debian.org/status/fetch.php?pkg=wrapitk-python&arch=i386&ver=3.20.1.2&stamp=1323582679
signature.asc
Description: Digital signature
_______________________________________________ Powered by www.kitware.com Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Follow this link to subscribe/unsubscribe: http://www.gccxml.org/mailman/listinfo/gccxml
