https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120632
Bug ID: 120632
Summary: AArch64 SVE build of GROMACS measurably slower with
GCC than LLVM
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Keywords: aarch64-sve, missed-optimization
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: ktkachov at gcc dot gnu.org
Target Milestone: ---
Target: aarch64
Building and running GROMACS in the same way as in PR119351 for AArch64 SVE
shows worse performance with GCC than built with Clang.
I've reproduced a ~15% slowdown on a Grace CPU with GCC trunk vs Clang trunk, I
expect this to reproduce on other SVE CPUs. I see the worse performance at
least as far back as GCC 13, I haven't tried earlier so I don't know if it's a
regression.
I don't have any analysis on why this may be, but recording here for wider
input and in case anyone else wants to have a look
