[testsuite] Re: PowerPC long double stdarg breakage

Alan Modra amodra@bigpond.net.au
Mon Apr 24 13:02:00 GMT 2006


On Fri, Apr 21, 2006 at 05:56:26AM -0400, Jakub Jelinek wrote:
> On Fri, Apr 21, 2006 at 05:36:36AM -0400, Jakub Jelinek wrote:
> > FYI, the scalar-by-value-6 test fails on x86_64-linux and ia64-linux.
> 
> But in that case it is testcase bug, on both x86_64-linux and ia64-linux
> there are many padding bits in sizeof (long double), and those may contain
> random values.  This means the USE_MEMCMP method is highly unportable.
> Is there any reason not to use normal comparisons always (i.e. kill
> the #define USE_MEMCMP 1 line in scalar-by-value-6_y.c and cut
> scalar-by-value-y.h into approx. a half)?  On ppc32, the testcase without
> your fix clearly fails even with plain comparisons.

I wrote the testcase .h harness a long time ago..   I think the reason I
used memcmp was to avoid a function call to a possibly unimplemented
comparison function.  There isn't any reason to do this for powerpc or
powerpc64 these days.  So, yes, kill the #define USE_MEMCMP.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre



More information about the Gcc-patches mailing list