Reapply r5364 (Update ARM and NEON optimizations for S32A_Opaque_BlitRow32)

Reapply r5364 (Update ARM and NEON optimizations for S32A_Opaque_BlitRow32)

This reverts r5378.

Committed: https://code.google.com/p/skia/source/detail?r=6727

[DerekS, 2012-10-31 13:26:08]

added the code back in and ran it through the following compile scenarios...

arm with thumb-1 instructions
armv7 with thumb-2 instructions
armv7 with full 32-bit instructions

All variations compiled and ran though a quick smoke test correctly.

[DerekS, 2012-10-31 13:42:45]

On 2012/10/31 13:26:08, DerekS wrote:
> added the code back in and ran it through the following compile scenarios...
> 
> arm with thumb-1 instructions
> armv7 with thumb-2 instructions
> armv7 with full 32-bit instructions
> 
> All variations compiled and ran though a quick smoke test correctly.

The error does come up when we run with runtime switchable neon instructions.  

/tmp/ccjeGji1.s:325: Error: thumb conditional instruction should be in IT block
-- `cmpeq r7,r5,LSR#24'
/tmp/ccjeGji1.s:326: Error: thumb conditional instruction should be in IT block
-- `cmpeq r7,r6,LSR#24'

I'll upload a patch with the correct changes.

[shikhas, 2012-11-05 20:47:51]

On 2012/10/31 13:42:45, DerekS wrote:

As I understand, the patch src/opts/SkBlitRow_opts_arm.cpp is having some build
issues. Can we upload the changes in 
src/opts/SkBlitRow_opts_arm_neon.cpp and bench/BitmapBench.cpp until the issues
with non-neon patch are resolved?

[shikhas, 2012-11-28 00:11:52]

Derek, Are we committing this change?
On 2012/11/05 20:47:51, shikhas wrote:
> On 2012/10/31 13:42:45, DerekS wrote:
> 
> As I understand, the patch src/opts/SkBlitRow_opts_arm.cpp is having some
build
> issues. Can we upload the changes in 
> src/opts/SkBlitRow_opts_arm_neon.cpp and bench/BitmapBench.cpp until the
issues
> with non-neon patch are resolved?

[DerekS, 2012-11-29 18:38:40]

Sorry for not addressing this sooner, but I haven't been able to fully test this
change within our new testing infrastructure in order to submit it.  I can bump
it up in my queue and get to it shortly, but if you have a plan to fix the
non-neon function I can defer and do both in one pass.

Our new testing infrastructure allows us to run your change over a set of known
problematic webpages and test the drawing performance.  As soon as I can verify
that there are no regressions there I can submit the neon version (and if you
have a patch the non-neon one as well).

[DerekS, 2012-12-10 14:04:52]

I'm not sure how gclient confused this CL with another I was writing, but the
last patch set uploaded was for an entirely different change and as such I'll
reopen this issue and continue tracking it.

[shikhas, 2013-01-11 20:51:00]

Derek, hope we upload the NEON optimization patch soon. Thanks!
On 2012/12/10 14:04:52, DerekS wrote:
> I'm not sure how gclient confused this CL with another I was writing, but the
> last patch set uploaded was for an entirely different change and as such I'll
> reopen this issue and continue tracking it.

[shikhas, 2013-01-30 18:31:44]

Derek,  gentle reminder that you could upload the skia patch for Neon
optimizations.
On 2013/01/11 20:51:00, shikhas wrote:
> Derek, hope we upload the NEON optimization patch soon. Thanks!
> On 2012/12/10 14:04:52, DerekS wrote:
> > I'm not sure how gclient confused this CL with another I was writing, but
the
> > last patch set uploaded was for an entirely different change and as such
I'll
> > reopen this issue and continue tracking it.