Fix any NEON vs core regressions

Registered by Michael Hope on 2011-05-26

We tell people to enable -mfpu=neon by default. In some cases this code runs slower than the non-NEON code. Investigate and fix.

Blueprint information

Status:
Complete
Approver:
Michael Hope
Priority:
High
Drafter:
Michael Hope
Direction:
Needs approval
Assignee:
None
Definition:
Approved
Series goal:
Accepted for 4.6
Implementation:
Informational Informational
Milestone target:
milestone icon backlog
Started by
Matthew Gretton-Dann on 2013-05-21
Completed by
Matthew Gretton-Dann on 2013-05-21

Related branches

Sprints

Whiteboard

[2013-05-21 matthew-gretton-dann] This has been superseded by TCWG-63: http://cards.linaro.org/browse/TCWG-63

The stop point is 90 % of the cases from our standard SPEC and EEMBC suite are either as fast or faster at -O3 with NEON turned on.

Some of these will be covered by other items listed in https://blueprints.launchpad.net/gcc-linaro/+spec/linaro-toolchain-o-neon-performance

As they're decided, replace 'Fix round n' with the particular thing being fixed.

There's overlap with https://blueprints.launchpad.net/gcc-linaro/+spec/thumb-2-regressions-11.11

(?)

Work Items

Work items:
Benchmark NEON/vectorised code vs non: TODO
Investigate any regressions: TODO
Fix round 1: TODO
Fix round 2: TODO
Fix round 3: TODO
Upstream: TODO

Dependency tree

* Blueprints in grey have been implemented.

This blueprint contains Public information 
Everyone can see this information.