[valgrind] [Bug 523790] s390: inefficient IR generated for branch insn

https://bugs.kde.org/show_bug.cgi?id=3D523790

--- Comment #2 from Florian Krohm <[email protected]> ---
Created attachment 194860
  --> https://bugs.kde.org/attachment.cgi?id=3D194860&action=3Dedit
memcheck performance measurements

So I was wondering what loop unrolling buys us and ran  memcheck via
perf/vg_perf --reps=3D5  perf
on s390 and x86-64
- as is
- with loop unrollling disabled
- with guest chasing disabled
- with loop unrolling and guest chasing disabled
On s390 loop unrolling does not make much of a difference. Disabling both l=
oop
unrolling and guest chasing seems to be
the sweet spot.
I did not look closely at x86-64. But disabling guest chasing does not seem
like a bad thing.

--=20
You are receiving this mail because:
You are watching all bug changes.=