Re-investigate Graphite Optimizations

Registered by Rob Beane

Graphite options can give a significant boost in performance. We should re-enable graphite optimizations in our platform builds either all over or for places where they make most sense. We use to build arm flags with a couple graphite options, but was later removed due to testing stability . So lets re-add them along with a couple new options.

See: http://gcc.gnu.org/wiki/Graphite
CFLAGS: -floop-interchange -floop-strip-mine -floop-block -ffast-math -funsafe-loop-optimizations

Blueprint information

Status:
Started
Approver:
Rob Beane
Priority:
Medium
Drafter:
None
Direction:
Approved
Assignee:
Rob Beane
Definition:
Approved
Series goal:
Accepted for v1.0.0
Implementation:
Good progress
Milestone target:
milestone icon mr2
Started by
Rob Beane

Related branches

Sprints

Whiteboard

INFO: 2D/skia was known to be improved

[2012/09/13 robbeane] After benching with and without graphite optimizations, it only shows a small increase in performance. But we have to take stability in account, and since the performance gain is so small, we will only add graphite to certain packages that make most sense (ie. libskia). We need to put together a list of packages that should use graphite optimizations and show before/after benchmarking results before pushing them to git.

(?)

Work Items

Work items:
Add graphite flags to build: DONE
Compile and test new flags: DONE
Benchmark and if faster push to git: TODO

This blueprint contains Public information 
Everyone can see this information.

Subscribers

No subscribers.