use new faster/vector impl for chopping conics

use new faster/vector impl for chopping conics

BUG=skia:

Committed: https://skia.googlesource.com/skia/+/55011038816a3fc7f0c0a39d482fb85347cc2e78

[reed1, 2015-03-26 15:15:00]

reed@google.com changed reviewers:
+ caryclark@google.com, mtklein@google.com

[reed1, 2015-03-26 15:15:01]

no gm changes on my mac

[reed1, 2015-03-26 15:15:05]

The CQ bit was checked by reed@google.com

[commit-bot: I haz the power, 2015-03-26 15:15:15]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1035943002/1

[commit-bot: I haz the power, 2015-03-26 15:15:16]

Note for Reviewers:
The CQ is waiting for an approval.
If you believe that the CL is not ready yet, or if you would like to L-G-T-M
with comments then please uncheck the CQ checkbox.

Waiting for LGTM from valid reviewer(s) till 2015-03-26 21:15 UTC

[commit-bot: I haz the power, 2015-03-26 15:16:34]

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

[commit-bot: I haz the power, 2015-03-26 15:16:35]

Try jobs failed on following builders:
  Build-Ubuntu-Clang-x86_64-Debug-Trybot on client.skia.compile (JOB_FAILED,
http://build.chromium.org/p/client.skia.compile/builders/Build-Ubuntu-Clang-x...)

[mtklein, 2015-03-26 15:21:33]

https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp
File src/core/SkGeometry.cpp (right):

https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1 +
fW));
Is SkScalarInvert faster than .invert()?

We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f below.

[reed1, 2015-03-26 15:24:41]

https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp
File src/core/SkGeometry.cpp (right):

https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1 +
fW));
On 2015/03/26 15:21:33, mtklein wrote:
> Is SkScalarInvert faster than .invert()?
> 
> We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f below.

I need scale later w/o the 1/2 (line 1305 and 1307)

https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1 +
fW));
On 2015/03/26 15:21:33, mtklein wrote:
> Is SkScalarInvert faster than .invert()?

I can retry it, but I was tweaking this to be pixel-exact w/ the old version.

> 
> We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f below.

[mtklein, 2015-03-26 15:29:04]

lgtm

This makes me we want to compare the cost of twice() then twice again later vs.
mul by 0.5f.  The twices can both be destructive, but the mul 0.5f needs another
register.

[reed1, 2015-03-26 15:29:17]

On 2015/03/26 15:24:41, reed1 wrote:
> https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp
> File src/core/SkGeometry.cpp (right):
> 
>
https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
> src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1 +
> fW));
> On 2015/03/26 15:21:33, mtklein wrote:
> > Is SkScalarInvert faster than .invert()?
> > 
> > We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f below.
> 
> I need scale later w/o the 1/2 (line 1305 and 1307)
> 
>
https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
> src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1 +
> fW));
> On 2015/03/26 15:21:33, mtklein wrote:
> > Is SkScalarInvert faster than .invert()?
> 
> I can retry it, but I was tweaking this to be pixel-exact w/ the old version.
> 
> > 
> > We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f below.

Is invert() perfect (matches scalar) in NEON?

[mtklein, 2015-03-26 15:30:19]

On 2015/03/26 15:29:17, reed1 wrote:
> On 2015/03/26 15:24:41, reed1 wrote:
> > https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp
> > File src/core/SkGeometry.cpp (right):
> > 
> >
>
https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
> > src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1 +
> > fW));
> > On 2015/03/26 15:21:33, mtklein wrote:
> > > Is SkScalarInvert faster than .invert()?
> > > 
> > > We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f below.
> > 
> > I need scale later w/o the 1/2 (line 1305 and 1307)
> > 
> >
>
https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
> > src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1 +
> > fW));
> > On 2015/03/26 15:21:33, mtklein wrote:
> > > Is SkScalarInvert faster than .invert()?
> > 
> > I can retry it, but I was tweaking this to be pixel-exact w/ the old
version.
> > 
> > > 
> > > We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f below.
> 
> Is invert() perfect (matches scalar) in NEON?

So far tested, yes.  If it's intended to be.  We can flesh out the single test
in tests/Sk2xTest.cpp.

[mtklein, 2015-03-26 15:32:20]

On 2015/03/26 15:30:19, mtklein wrote:
> On 2015/03/26 15:29:17, reed1 wrote:
> > On 2015/03/26 15:24:41, reed1 wrote:
> > > https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp
> > > File src/core/SkGeometry.cpp (right):
> > > 
> > >
> >
>
https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
> > > src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1
+
> > > fW));
> > > On 2015/03/26 15:21:33, mtklein wrote:
> > > > Is SkScalarInvert faster than .invert()?
> > > > 
> > > > We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f
below.
> > > 
> > > I need scale later w/o the 1/2 (line 1305 and 1307)
> > > 
> > >
> >
>
https://codereview.chromium.org/1035943002/diff/1/src/core/SkGeometry.cpp#new...
> > > src/core/SkGeometry.cpp:1293: Sk2s scale = Sk2s(SkScalarInvert(SK_Scalar1
+
> > > fW));
> > > On 2015/03/26 15:21:33, mtklein wrote:
> > > > Is SkScalarInvert faster than .invert()?
> > > 
> > > I can retry it, but I was tweaking this to be pixel-exact w/ the old
> version.
> > > 
> > > > 
> > > > We might try twice(Sk2s(SKScalar1+fW)).invert() to avoid the * 0.5f
below.
> > 
> > Is invert() perfect (matches scalar) in NEON?
> 
> So far tested, yes.  If it's intended to be.  We can flesh out the single test
> in tests/Sk2xTest.cpp.

(approxInvert() is the faster, non accurate one.)

[reed1, 2015-03-26 15:37:02]

using inver() seems the same or a tiny bit slower. If it requires loading 1.0 in
a vector, that might explain it.

[reed1, 2015-03-26 15:37:10]

The CQ bit was checked by reed@google.com

[reed1, 2015-03-26 15:37:11]

The patchset sent to the CQ was uploaded after l-g-t-m from mtklein@google.com
Link to the patchset: https://codereview.chromium.org/1035943002/#ps20001
(title: "fix bench")

[commit-bot: I haz the power, 2015-03-26 15:37:26]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1035943002/20001

[mtklein, 2015-03-26 15:46:04]

On 2015/03/26 15:37:02, reed1 wrote:
> using inver() seems the same or a tiny bit slower. If it requires loading 1.0
in
> a vector, that might explain it.

Yeah, it does on your desktop.  Let's table that idea.

[commit-bot: I haz the power, 2015-03-26 16:10:26]

Committed patchset #2 (id:20001) as
https://skia.googlesource.com/skia/+/55011038816a3fc7f0c0a39d482fb85347cc2e78

[mtklein, 2015-03-26 19:13:31]

> This makes me we want to compare the cost of twice() then twice again later
vs.
> mul by 0.5f.  The twices can both be destructive, but the mul 0.5f needs
another
> register.

Just took a quick look at this.  Very minor if any effect.