llvm dev - Feb 2017 - An option switching LSR to choose solution with minimum instructions

Hi All,

A new option for LSR "-lsr-insns-cost" was introduced in the patch
https://reviews.llvm.org/D28307.
Basically it set instruction count as priority number one for
solutions with low register use.
This lead to smaller loops in most cases.

The goal is to enable the option by default for x86.
Could you please check your benchmarks to see if there are gains/loses
and report this here?

That I have for now is:
177.mesa on -O2 +3%
256.bzip2 on -Ofast -flto +1.5%
and overall good improvements for x86.
The only regression is on linpack test 3% on HSW, however it differs
on other CPUs (say for Atoms it is 3% gain).

Thanks,
Evgeny

We did a quick check with and without the patch for cpu2006 INT bms only. Looks
like overall impact is neutral. Most of the ups/downs are in the noise range
except probably sjeng.


cpu2006

diff

400.perlbench

1.001288

401.bzip2

0.996352

445.gobmk

0.998473

456.hmmer

1.009015

458.sjeng

1.011942

462.libquantum

1.003234

464.h264ref

0.995798

471.omnetpp

1.004245

473.astar

1.002221

483.xalancbmk

0.993139

403.gcc

1.001213

429.mcf

1.004345

Geomean

1.00176






-----Original Message-----
From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of Evgeny
Stupachenko via llvm-dev
Sent: Wednesday, February 15, 2017 2:15 AM
To: llvm-dev at lists.llvm.org
Subject: [llvm-dev] An option switching LSR to choose solution with minimum
instructions



Hi All,



A new option for LSR "-lsr-insns-cost" was introduced in the patch
https://reviews.llvm.org/D28307.

Basically it set instruction count as priority number one for solutions with low
register use.

This lead to smaller loops in most cases.



The goal is to enable the option by default for x86.

Could you please check your benchmarks to see if there are gains/loses and
report this here?



That I have for now is:

177.mesa on -O2 +3%

256.bzip2 on -Ofast -flto +1.5%

and overall good improvements for x86.

The only regression is on linpack test 3% on HSW, however it differs on other
CPUs (say for Atoms it is 3% gain).



Thanks,

Evgeny

_______________________________________________

LLVM Developers mailing list

llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>

http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170216/70d03bd1/attachment.html>