This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: ppc32: math failures - logb_downward (1) gives wrong result
On 06/21/2012 06:09 PM, Joseph S. Myers wrote:
> On Thu, 21 Jun 2012, Andreas Jaeger wrote:
>
>> Build current trunk (as of yesterday) on s390 with gcc 4.7, I see a couple of
>> failures in the math testsuite:
> ppc32 or s390? For ppc32 it's a known GCC bug (though there may not be an
> entry in GCC Bugzilla for it): the code sequence generated for converting
> integers to floating-point, on older processors, wrongly converts 0 to -0
> in round-downwards mode (if -frounding-math, a more complicated sequence
> should be used instead that temporarily sets round-to-nearest mode).
>
This issue is due on POWER4 PPC32 machines GCC does not generate the 'fcfid' instruction
for integer to FP conversion. This was fixed on revision 186392 in the FSF branch:
[gcc]
2012-04-12 Michael Meissner <meissner@linux.vnet.ibm.com <mailto:meissner%40linux.vnet.ibm.com>>
Backport from mainline
2012-04-12 Michael Meissner <meissner@linux.vnet.ibm.com <mailto:meissner%40linux.vnet.ibm.com>>
PR target/52775
* config/rs6000/rs6000.h (TARGET_FCFID): Add TARGET_PPC_GPOPT to
the list of options to enable the FCFID instruction.
(TARGET_EXTRA_BUILTINS): Adjust comment.
[gcc/testsuite]
012-04-12 Michael Meissner <meissner@linux.vnet.ibm.com <mailto:meissner%40linux.vnet.ibm.com>>
Backport from mainline
2012-04-12 Michael Meissner <meissner@linux.vnet.ibm.com <mailto:meissner%40linux.vnet.ibm.com>>
PR target/52775
* gcc.target/powerpc/pr52775.c: New file.
--
Adhemerval Zanella Netto
Software Engineer
Linux Technology Center Brazil
Toolchain / GLIBC on Power Architecture
azanella@linux.vnet.ibm.com / azanella@br.ibm.com
+55 61 8642-9890