Reversion of rL292621 caused about 7% performance regressions on Cortex-M

Project Infrastructure LLVM Dev List Archives
#1

Hi Wei,

Your reversion of rL292621 caused about 7% performance regressions in our benchmark on Cortex-M7/M4.

In your commit comment I see it causes build bot failures.

What kind are the failures? Compiler crashes or incorrect code generation? Will you fix them?

We are interested in the changes because of performance improvements they give.

Kind regards,

Evgeny Astigeevich

Senior Compiler Engineer
Compilation Tools
ARM

#2

Hi Evgeny,

The failure is a compiler crash on an out-of-tree target in apple.
Quentin is helping to find a reproducible but havn't succeeded yet.
Once I get a reproducible, I will fix it. Thanks for the interest on
the work. Details about the crash can be found here:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20170123/422853.html

Thanks,
Wei.

#3

Hi Wei,

Thank you for information.
Please let me know about any progress in fixing the failures.
I can help with checking that the final patch gives the same level of performance improvements.

Kind regards,
Evgeny Astigeevich
Senior Compiler Engineer
Compilation Tools
ARM

#4

Sure. Will keep you posted.

Thanks,
Wei.

#5

Hi Evgeny,

Quentin and Matthias found it was a problem about subreg live range
update and will push a fix soon
(http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20170123/424126.html).

Thanks,
Wei.

#6

Quentin recommited the patch with the fix at rL293353.

Thanks,
Wei.

#7

Hi Wei,

I checked the patch. It works. The regressions are fixed. I will have results of other benchmarks tomorrow.
Thank you very much for the quick resolution.

Kind regards,
Evgeny Astigeevich
Senior Compiler Engineer
Compilation Tools
ARM