On Thu, 2013-08-29 at 12:20 -0400, David Malcolm wrote:
[...]
> Successfully bootstrapped and tested on x86_64-unknown-linux-gnu: all
> testcases show the same results as an unpatched build (relative to
> r202029).
I messed up the testing for this by accidentally configuring the builds
with --enable-checking=release, thus not running the run-time checks.
Upon rechecking with a plain "./configure" (and thus enabling the
run-time checks), the patches break 20 test cases involving OpenMP.
Sorry about this.
I'm working on a fix [1]; I'll also add gcc_unreachable to the
PCH-related hooks for gimple (and thus reduce the size of the
handwritten GTY code by a factor of 3; not sure if it will make it any
more popular...).
Dave
[1] I believe the issue is this overzealous generated is_a_helper test:
template <>
template <>
inline bool
is_a_helper <gimple_statement_omp_parallel>::test (gimple gs)
{
return gs->code == GIMPLE_OMP_PARALLEL;
}
which should clearly also permit gs->code to be GIMPLE_OMP_TASK; this
bug leads to various failures in omp-lowering for obvious reasons.
