Tune Swing Modulo Scheduling or Selective Scheduling for ARM

Bug #663793 reported by Yao Qi
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Linaro GCC
Won't Fix
Undecided
Unassigned

Bug Description

GCC has two scheduling algorithms that support fancy loop optimizations: -fmodulo-sched and -fselective-scheduling / -fsel-sched-pipelining. When I tried them on FSF GCC mainline (2010-10-17), neither was super helpful for ARM.
Performance number can be found here https://wiki.linaro.org/YaoQi/Sandbox/Thumb2SpeedOptimization

They should be helpful on A-family processors where you have a lot of cache and pipeline to work with.

[codesourcery #6873]

Tags: speed task
Revision history for this message
Yao Qi (yao-codesourcery) wrote :

It is hard. If we can do it, we could improve a lot. The goal is to get the performance we get from loop unrolling with smaller code size.

Revision history for this message
Michael Hope (michaelh1) wrote :
Revision history for this message
Michael Hope (michaelh1) wrote :

Closed out as Revital is tracking this via blueprints.

Changed in gcc-linaro:
status: New → Won't Fix
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.