move more geometry to simd

move more geometry to simd

Remove duplicate quad and cubic code around
computing the polynomial coefficients, and
use common SIMD-based code instead.

R=reed@google.com
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1633143002

Committed: https://skia.googlesource.com/skia/+/5ba2b9612ae4bc3a244bf45f1ec55c3a5a41e181

[caryclark, 2016-01-26 15:06:07]

Description was changed from

==========
move more geometry to simd

Remove duplicate quad and cubic code around
computing the polynomial coefficients, and
use common SIMD-based code instead.

R=reed@google.com
BUG=skia:
==========

to

==========
move more geometry to simd

Remove duplicate quad and cubic code around
computing the polynomial coefficients, and
use common SIMD-based code instead.

R=reed@google.com
BUG=skia:
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

[caryclark, 2016-01-26 15:06:47]

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

[commit-bot: I haz the power, 2016-01-26 15:06:56]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1633143002/1
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1633143002/1

[commit-bot: I haz the power, 2016-01-26 15:18:31]

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

[commit-bot: I haz the power, 2016-01-26 15:18:34]

Dry run: This issue passed the CQ dry run.

[caryclark, 2016-01-26 16:37:30]

Please take a look. This changes 3 GMs, 14 instances on the try.

I doubt there is so much of a performance gain, but it does delete about 150
lines of code.

If you think this is worth pursuing, I'll create a CL with legacy guards.

[reed1, 2016-01-26 18:11:14]

Definitely seems worth-while.

Some of the changes go from scaler->vector
Some of the changes just seem to call a helper to create the coefficients.

Do both types of changes need a guard, or only the first? Just curious if we can
reduce the scope of the guard.

[caryclark, 2016-01-26 18:59:18]

On 2016/01/26 18:11:14, reed1 wrote:
> Definitely seems worth-while.
> 
> Some of the changes go from scaler->vector
> Some of the changes just seem to call a helper to create the coefficients.
> 
> Do both types of changes need a guard, or only the first? Just curious if we
can
> reduce the scope of the guard.

I'm a bit surprised that any of them need the guard, since they should be
calling the same math that was there before.

I'll code up the guards as a bunch of separate switches so I can play with them
and understand which cause the
GMs to change.

[caryclark, 2016-01-26 19:25:13]

I may have gotten lucky guesing that only the cubic eval() yields different
results.
At least, that's the case locally. Gold is giving it a go right now.

[caryclark, 2016-01-26 19:29:31]

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

[commit-bot: I haz the power, 2016-01-26 19:29:42]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1633143002/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1633143002/20001

[commit-bot: I haz the power, 2016-01-26 19:41:46]

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

[commit-bot: I haz the power, 2016-01-26 19:41:46]

Dry run: This issue passed the CQ dry run.

[caryclark, 2016-01-26 20:50:21]

I confirmed that adding a legacy guard around the old cubic eval was enough to
eliminate all Gold changes. If it looks good to you, I'll wait until the Chrome
guard lands (in flight) before landing this.

[reed1, 2016-01-26 20:53:17]

sweet

lgtm

[caryclark, 2016-01-27 00:50:32]

The CQ bit was checked by caryclark@google.com

[commit-bot: I haz the power, 2016-01-27 00:50:33]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1633143002/40001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1633143002/40001

[commit-bot: I haz the power, 2016-01-27 01:02:33]

Description was changed from

==========
move more geometry to simd

Remove duplicate quad and cubic code around
computing the polynomial coefficients, and
use common SIMD-based code instead.

R=reed@google.com
BUG=skia:
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

to

==========
move more geometry to simd

Remove duplicate quad and cubic code around
computing the polynomial coefficients, and
use common SIMD-based code instead.

R=reed@google.com
BUG=skia:
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...

Committed:
https://skia.googlesource.com/skia/+/5ba2b9612ae4bc3a244bf45f1ec55c3a5a41e181
==========

[commit-bot: I haz the power, 2016-01-27 01:02:34]

Committed patchset #3 (id:40001) as
https://skia.googlesource.com/skia/+/5ba2b9612ae4bc3a244bf45f1ec55c3a5a41e181

[reed1, 2016-01-27 10:39:31]

We'll need to rebaseline these to have the DEPS roll pass:

svg/batik/text/textOnPath.svg
svg/hixie/perf/001.xml
svg/hixie/perf/002.xml