Changes to existing ARMv7 routines

Submitted by Ben Avison on April 11, 2016, 12:26 p.m.

Details

Reviewer None
Submitted April 11, 2016, 12:26 p.m.
Last Updated April 11, 2016, 12:44 p.m.
Revision 1

Cover Letter(s)

Revision 1
      Since there are a few people around on the list at the moment who are
familiar with NEON, I'm hoping someone will be able to review my work so it
can make it into git.

To keep the number of patches manageable, here are a group which improve
incrementally upon existing ARMv7 routines, without adding any new ones yet.
Most of these are reposts which have had no review of the technical content.
The patch numbers have been reassigned within this series of 9 patches, and
won't match the numbers used when originally posted.

Ben Avison (9):
  armv7: Coalesce scalar accesses where possible
  armv7: Faster fill operations
  armv7: Use VLD-to-all-lanes
  armv7: Simplify constant load
  armv7: Use prefetch for small-width images too
  armv7: Use aligned memory writes in both copies of bilinear code
  armv7: Move common bilinear macro definitions to a new header file
  armv7: More use of fast paths with localized destination alpha
  armv7: More use of fast paths with localized source alpha

 pixman/Makefile.am                    |    3 +-
 pixman/pixman-arm-neon-asm-bilinear.S |  153 +-----------------
 pixman/pixman-arm-neon-asm-bilinear.h |  165 +++++++++++++++++++
 pixman/pixman-arm-neon-asm.S          |  280 +++++++--------------------------
 pixman/pixman-arm-neon-asm.h          |   20 +++
 pixman/pixman-arm-neon.c              |   21 +++
 6 files changed, 272 insertions(+), 370 deletions(-)
 create mode 100644 pixman/pixman-arm-neon-asm-bilinear.h
    

Revisions