565 support for SIMD xfermodes

565 support for SIMD xfermodes

This uses the most basic approach possible:
  - to load an Sk4px from 565, convert to SkPMColors on the stack serially then load those SkPMColors.
  - to store an Sk4px to 565, store to SkPMColors on the stack then convert to 565 serially.

Clearly, we can optimize these loads and stores.  That's a TODO.

The code using SkPMFloat is the same idea but a little more long-term viable, as we're only operating on one pixel at a time anyway.  We could probably write 565 <-> SkPMFloat methods, but I'd rather not until it's really compelling.

The speedups are varied but similar across SSE and NEON: a few uninteresting, many 50% faster, some 2x faster, and SoftLight ~4x faster.

This will cause minor GM diffs, but I don't think any layout test changes.

BUG=skia:

Committed: https://skia.googlesource.com/skia/+/942930dcaa51f66d82cdaf46ae62efebd16c8cd0

Committed: https://skia.googlesource.com/skia/+/860dcaa2ddfdadc050af4f943a84a9d499315066

Committed: https://skia.googlesource.com/skia/+/ced1585149d4ac8a68efd80e11dbc23bca6620d4

[mtklein_C, 2015-07-20 16:52:13]

The CQ bit was checked by mtklein@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-07-20 16:52:23]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1245673002/40001

[commit-bot: I haz the power, 2015-07-20 17:08:09]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-07-20 17:08:10]

Dry run: This issue passed the CQ dry run.

[mtklein_C, 2015-07-20 17:15:12]

mtklein@chromium.org changed reviewers:
+ msarett@google.com

[mtklein_C, 2015-07-20 17:15:12]

I was planning on doing this all in one go, hooking in 565 with fully
specialized loads and stores, but it turns out even just this change is a big
speedup.  Will follow up with specialized loads and stores once I figure them
out.  :)

[msarett, 2015-07-20 17:34:00]

lgtm

I don't want to hold you up.  I plan to take a closer look later.

[mtklein, 2015-07-20 17:34:40]

The CQ bit was checked by mtklein@google.com

[commit-bot: I haz the power, 2015-07-20 17:34:54]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1245673002/40001

[commit-bot: I haz the power, 2015-07-20 17:35:33]

Committed patchset #3 (id:40001) as
https://skia.googlesource.com/skia/+/942930dcaa51f66d82cdaf46ae62efebd16c8cd0

[mtklein, 2015-07-21 12:02:29]

A revert of this CL (patchset #3 id:40001) has been created in
https://codereview.chromium.org/1242973004/ by mtklein@google.com.

The reason for reverting is: 942930d (included in this roll) introduced a 140 kB
sizes regression in
libskia.so. Please investigate and reland if this regression is necessary.
.

[msarett, 2015-07-21 13:00:24]

This looks cool.  As you said, it's the setup step.

Does a sizes regression just mean that the skia library increased in size?  What
might cause that?

[mtklein, 2015-07-21 13:09:43]

On 2015/07/21 13:00:24, msarett wrote:
> This looks cool.  As you said, it's the setup step.
> 
> Does a sizes regression just mean that the skia library increased in size? 
What
> might cause that?

Yeah.  It landed optimized 565 code for every xfermode at once.  That'll at
least double the size of SkXfermode.o.  I think  it's mostly due to everything
being templated, inlined, and then fairly aggressively unrolled by things like
Sk4px::MapDstSrc.  I'm looking through calling convention ABI docs now to see if
I can factor this into separate function calls without hurting performance too
much (looking good for x86-64, probably not good for x86, untested but
optimistic for ARMv8, untested and pessimistic for ARMv7).  I may also back off
the unrolling a step or two, again provided it doesn't hurt perf too badly.

So, my plan is:
  1) leave this reverted for now and leave the other CL floating
  2) hack on existing code to see if I can decrease the binary size by better
factoring without harming perf too much
  3) circle back to the 565 CLs

I don't think it'd be wasted time to look at the other CL now though.  It's
almost all new SIMD code that'll live on exactly as-is, except perhaps behind a
function call instead of inlined.

[msarett, 2015-07-21 13:22:48]

Gotcha agreed.

I'm looking through it now :).

[mtklein, 2015-07-21 21:00:41]

Going to try this again Matt.  I've merged the two 565 CLs together here and
rebased them onto head.  This should now be ~26K new code, and it does still
make 565 modes faster across the board.

[msarett, 2015-07-21 21:06:35]

Looks familiar :)
lgtm

[mtklein, 2015-07-21 21:15:14]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-07-21 21:15:27]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1245673002/60001

[commit-bot: I haz the power, 2015-07-21 21:48:07]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-07-21 21:48:08]

Dry run: This issue passed the CQ dry run.

[mtklein, 2015-07-22 00:22:57]

The CQ bit was checked by mtklein@google.com

[commit-bot: I haz the power, 2015-07-22 00:23:09]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1245673002/60001

[commit-bot: I haz the power, 2015-07-22 00:23:42]

Committed patchset #4 (id:60001) as
https://skia.googlesource.com/skia/+/860dcaa2ddfdadc050af4f943a84a9d499315066

[mtklein, 2015-07-22 01:34:22]

A revert of this CL (patchset #4 id:60001) has been created in
https://codereview.chromium.org/1248893004/ by mtklein@google.com.

The reason for reverting is: NEON 565 gold images have gone ugly.  This is what
I get for writing and testing SSE and just writing NEON.

E.g. colortype_xfermodes, dstreadshuffle, bigbitmaprect, pictures,
textbloblooper, aaxfermodes (only Plus).

[mtklein_C, 2015-07-22 17:25:14]

The CQ bit was checked by mtklein@chromium.org

[mtklein_C, 2015-07-22 17:25:14]

The patchset sent to the CQ was uploaded after l-g-t-m from msarett@google.com
Link to the patchset: https://codereview.chromium.org/1245673002/#ps80001
(title: "fix typo")

[commit-bot: I haz the power, 2015-07-22 17:25:27]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1245673002/80001

[commit-bot: I haz the power, 2015-07-22 17:52:56]

Committed patchset #5 (id:80001) as
https://skia.googlesource.com/skia/+/ced1585149d4ac8a68efd80e11dbc23bca6620d4