Use vld1 for quad loads
Registered by
Michael Hope
The backend currently generates vldmia instructions when loading a quad word such as those from the vectoriser.
Change to use a vld1 which is as fast and supports unaligned accesses, then test and fix the alignment assertions which give around a 20 % speed boost.
Blueprint information
- Status:
- Complete
- Approver:
- Michael Hope
- Priority:
- Medium
- Drafter:
- Michael Hope
- Direction:
- Approved
- Assignee:
- Ulrich Weigand
- Definition:
- Approved
- Series goal:
- Accepted for 4.6
- Implementation:
-
Implemented
- Milestone target:
-
4.7-2012.10
- Started by
- Ulrich Weigand
- Completed by
- Ulrich Weigand
Whiteboard
Meta:
Headline: TBD
Acceptance: TBD
Roadmap id: TCWG2011-GCC-O3
Now unblocked.
(?)
