diff options
author | Alexandre Oliva <aoliva@redhat.com> | 2016-12-31 07:46:01 -0200 |
---|---|---|
committer | Alexandre Oliva <aoliva@redhat.com> | 2017-01-06 01:31:00 -0200 |
commit | 172df0c0ad97307f3d28b9a070c341a74dad233a (patch) | |
tree | 9a25a24357e94d8f12fa7ca3a31013c977341209 /liboffloadmic | |
parent | 2dccc7c49bef10d058260151cef37bababc47d20 (diff) | |
download | gcc-aoliva/non-SFN-fixes-for-GCC7.tar.gz |
[-fcompare-debug] var tracking options are not optimization optionsaoliva/non-SFN-fixes-for-GCC7
If we include them in the ICF hash, they may cause congruence_groups to
be processed in a different order due to different hashes, which in turn
causes different funcdef_nos to be assigned to functions. Since these
numbers are included in -fcompare-debug dumps, they cause failures.
Since these options are not optimization options, in that they do not
(or should not, save for bugs like this) affect the executable code
output by the compiler, they should not be marked as such.
This patch replaces the Optimization flag in the var-tracking options
with the newly-introduced PerFunction flag, so that it can still be
controlled on a per-function basis, but that disregards it in the hash
computation used by ICF.
This fixes -fcompare-debug failures in numerous LTO testcases.
for gcc/ChangeLog
* doc/options.texi (PerFunction): New.
* opt-functions.awk (switch_flags): Map both Optimization and
PerFunction to CL_OPTIMIZATION.
* opth-gen.awk: Test for PerFunction flag along with
Optimization.
* optc-save-gen.awk: Likewise. Introduce var_opt_hash and set
it only when the latter is present. Skip those that don't in
the hash function generator.
* common.opt (fvar-tracking): Mark as PerFunction instead of
Optimization.
(fvar-tracking-assignments): Likewise.
(fvar-tracking-assignments-toggle): Likewise.
(fvar-tracking-uninit): Likewise.
Diffstat (limited to 'liboffloadmic')
0 files changed, 0 insertions, 0 deletions