This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug libgomp/78468] [7 regression] libgomp.c/reduction-10.c and many more FAIL
- From: "jakub at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Thu, 24 Nov 2016 15:35:47 +0000
- Subject: [Bug libgomp/78468] [7 regression] libgomp.c/reduction-10.c and many more FAIL
- Auto-submitted: auto-generated
- References: <bug-78468-4@http.gcc.gnu.org/bugzilla/>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78468
--- Comment #13 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Yes, sure, there are 2 alloca uses in the parallel region, one is for y
variable, which uses __builtin_alloca_with_align ((p4 - 1 + 1) * 4, 32) and the
other is for a, which uses __builtin_alloca_with_align ((p5 - 1 + 1) * 8, 32).
p4 is 3 and p5 is 4, so the former should be 12 bytes and the latter 32 bytes.
But at least in the *.optimized dump I don't see anything that would attempt to
write more, there is at most y[2].t and for a at most a[127 / 32].t.
So, can you find out how much each alloca actually aligns and if what writes
are to those areas?
Does it fail also with OMP_NUM_THREADS=1?