David Billinghurst wrote:
Tests gcc.dg/torture/fp-int-convert-float128-timode.c and
gcc.dg/torture/fp-int-convert-float128.c fail on a number of platforms
on gcc-4.3, 4.4 and trunk.
For some examples, see:
- http://gcc.gnu.org/ml/gcc-testresults/2009-05/msg00552.html
- http://gcc.gnu.org/ml/gcc-testresults/2009-05/msg00315.html
- http://gcc.gnu.org/ml/gcc-testresults/2009-04/msg02417.html
At present the tests are run for
{ dg-do run { target i?86-*-* x86_64-*-* ia64-*-* } }
128 bit floats aren't supported on cygwin, and I expect the same is true
elsewhere. This patch just skips the test for cygwin, and will be
simple to extend to other targets as required. The alternative approach
of modifying the dg-do run target became messy.
According to the manual:
" Not all targets support additional floating point types. `__float80'
is supported on i386, x86_64 and ia64 targets and target `__float128'
is supported on x86_64 and ia64 targets. "
... so shouldn't this be solely CPU dependent, rather than host platform? It
looks to me like it is simply incorrect to run float128 tests on i?86-*-* and
you should just remove that from the target selector, no?
cheers,
DaveK