This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug tree-optimization/82133] [7/8 Regression] unroll-loops too aggressive
- From: "nickpapior at gmail dot com" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Tue, 12 Sep 2017 09:09:50 +0000
- Subject: [Bug tree-optimization/82133] [7/8 Regression] unroll-loops too aggressive
- Auto-submitted: auto-generated
- References: <bug-82133-4@http.gcc.gnu.org/bugzilla/>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82133
--- Comment #3 from Nick <nickpapior at gmail dot com> ---
According to
https://github.com/xianyi/OpenBLAS/issues/1292#issuecomment-327788223, it is in
the kernel/x86_64/dgemv_n_microk_haswell-4.c source where the loop-unrolling
has caused the parent function call which is kernel/x86_64/dgemv_n_4.c. I.e.
the unrolling is too aggressive in dgemv_n_4.c.
However, I am no expert in the OpenBLAS source code to be able to extract the
relevant sources and make a functioning C-program to try and reproduce it.
All I can see is that the use of the loop counter in dgemv_n_4.c is a
relatively complex method.
And yes, OpenBLAS works fine without any of the mentioned flags (it also works
for much more aggressive optimizations, but just not with the -funroll-loops
-fipa-ra flags.