[Bug tree-optimization/78348] [7 REGRESSION] 15% performance drop for coremark-pro/nnet-test after r242038
rguenth at gcc dot gnu.org
gcc-bugzilla@gcc.gnu.org
Mon Feb 6 14:14:00 GMT 2017
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78348
--- Comment #12 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Dominik Vogt from comment #11)
> Fails if configured with "--with-arch=zEC12", passes without that.
The innermost loop is fully unrolled. Looks like somebody bumped
--param max-unroll-times for zEC12:
if (opts->x_s390_tune >= PROCESSOR_2097_Z10)
{
maybe_set_param_value (PARAM_MAX_UNROLLED_INSNS, 100,
opts->x_param_values,
opts_set->x_param_values);
maybe_set_param_value (PARAM_MAX_UNROLL_TIMES, 32,
opts->x_param_values,
opts_set->x_param_values);
you can test/commit a patch adjusting dg-options to specify sth
like --param max-unroll-times=8
More information about the Gcc-bugs
mailing list