Optimize highQualityFilter

Optimize highQualityFilter

portable version:
before:
  10M   1       806µs   807µs   810µs   821µs   1%      █▂▁▁▃▁▁▁█▁ 8888    bitmap_BGRA_8888_A_scale_rotate_bicubic
after:
  10M   1       566µs   568µs   569µs   579µs   1%      ▄▂▂█▂▁▁▁▃▁ 8888    bitmap_BGRA_8888_A_scale_rotate_bicubic

SSE version:
before:
  10M   1       485µs   486µs   487µs   494µs   1%      ▇▂▁▁▁▁█▂▁▁ 8888    bitmap_BGRA_8888_A_scale_rotate_bicubic
after:
  10M   1       419µs   420µs   421µs   430µs   1%      ▅▃▂▁▁█▂▁▁▁ 8888    bitmap_BGRA_8888_A_scale_rotate_bicubic

BUG=skia:

Committed: https://skia.googlesource.com/skia/+/6ff4acedb58ab5ef83b8e598637976d592572630

[qiankun, 2014-11-25 10:39:23]

qiankun.miao@intel.com changed reviewers:
+ humper@google.com, mtklein@google.com

[qiankun, 2014-11-25 10:40:14]

PTAL.

[mtklein, 2014-11-25 13:57:56]

Looks great.

https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp
File src/core/SkBitmapFilter.cpp (right):

https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp#...
src/core/SkBitmapFilter.cpp:32: SkBitmapFilter* filter = s.getBitmapFilter();
Can this be 'const SkBitmapFilter*'?

https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp#...
src/core/SkBitmapFilter.cpp:65: continue;
I take it this is where the speed win comes from?

[humper, 2014-11-25 14:00:07]

On 2014/11/25 13:57:56, mtklein wrote:
> Looks great.
> 
> https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp
> File src/core/SkBitmapFilter.cpp (right):
> 
>
https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp#...
> src/core/SkBitmapFilter.cpp:32: SkBitmapFilter* filter = s.getBitmapFilter();
> Can this be 'const SkBitmapFilter*'?
> 
>
https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp#...
> src/core/SkBitmapFilter.cpp:65: continue;
> I take it this is where the speed win comes from?

lgtm modulo mike's const request

[qiankun, 2014-11-25 14:22:35]

https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp
File src/core/SkBitmapFilter.cpp (right):

https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp#...
src/core/SkBitmapFilter.cpp:32: SkBitmapFilter* filter = s.getBitmapFilter();
On 2014/11/25 13:57:56, mtklein wrote:
> Can this be 'const SkBitmapFilter*'?

Done.

https://codereview.chromium.org/759603002/diff/1/src/core/SkBitmapFilter.cpp#...
src/core/SkBitmapFilter.cpp:65: continue;
On 2014/11/25 13:57:56, mtklein wrote:
> I take it this is where the speed win comes from?

Yes. Also, hoist filter calculation outside the loop contributes a little
(~10us).

[reed1, 2014-11-25 14:34:32]

reed@google.com changed reviewers:
+ reed@google.com

[reed1, 2014-11-25 14:34:33]

https://codereview.chromium.org/759603002/diff/20001/src/core/SkBitmapFilter.cpp
File src/core/SkBitmapFilter.cpp (right):

https://codereview.chromium.org/759603002/diff/20001/src/core/SkBitmapFilter....
src/core/SkBitmapFilter.cpp:67: fr += combined_weight * SkGetPackedR32(c);
boy oh boy, this seems to beg for Sk4f ...

[mtklein, 2014-11-25 15:02:21]

The CQ bit was checked by mtklein@google.com

[mtklein, 2014-11-25 15:02:22]

lgtm

https://codereview.chromium.org/759603002/diff/20001/src/core/SkBitmapFilter.cpp
File src/core/SkBitmapFilter.cpp (right):

https://codereview.chromium.org/759603002/diff/20001/src/core/SkBitmapFilter....
src/core/SkBitmapFilter.cpp:67: fr += combined_weight * SkGetPackedR32(c);
On 2014/11/25 14:34:33, reed1 wrote:
> boy oh boy, this seems to beg for Sk4f ...

Yep, I will certainly try this out.

[commit-bot: I haz the power, 2014-11-25 15:02:28]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/759603002/20001

[commit-bot: I haz the power, 2014-11-25 15:12:32]

Committed patchset #2 (id:20001) as
https://skia.googlesource.com/skia/+/6ff4acedb58ab5ef83b8e598637976d592572630