ARM Skia NEON patches - 27 - S32A_D565_Blend

ARM Skia NEON patches - 27 - S32A_D565_Blend


BlitRow565: new intrinsics version of S32A_D565_Blend

This new version is basically a rewrite of the existing code with
a few speed and accuracy improvements. There is a switch to enable
pixel perfect results at the cost of a (quite big) decrease of
performances (disabled in this patch).

Here are the benchmark results (speedup vs. existing code):

+-------+------------+------------+
| count | Cortex -A9 | Cortex-A15 |
+-------+------------+------------+
| 1     | +103.6%    | +12%       |
+-------+------------+------------+
| 2     | +3.6%      | +21.6%     |
+-------+------------+------------+
| 4     | +0.8%      | -0.8%      |
+-------+------------+------------+
| 8     | +3.9%      | -1%        |
+-------+------------+------------+
| 16    | +14.7%     | +5.7%      |
+-------+------------+------------+
| 64    | +18.1%     | +13.2%     |
+-------+------------+------------+
| 256   | +16.3%     | +27.4%     |
+-------+------------+------------+
| 1024  | +78.2%     | +17.4%     |
+-------+------------+------------+

Signed-off-by: Kévin PETIT <kevin.petit@arm.com>;

BUG=skia:

Committed: http://code.google.com/p/skia/source/detail?r=13438

[kevin.petit, 2014-02-10 16:26:23]

This one will allow to share more code with aarch64 whilst enjoying a nice
speedup on ARMv7.

[djsollen, 2014-02-13 17:37:27]

lgtm

[djsollen, 2014-02-13 17:38:50]

don't forgot to rebaseline and remove the ignored tests after submitting this.

[kevin.petit, 2014-02-13 17:41:21]

The CQ bit was checked by kevin.petit@arm.com

[commit-bot: I haz the power, 2014-02-13 17:41:31]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/kevin.petit@arm.com/156113005/10001

[kevin.petit, 2014-02-13 17:41:56]

On 2014/02/13 17:38:50, djsollen wrote:
> don't forgot to rebaseline and remove the ignored tests after submitting this.

I won't :-).

[commit-bot: I haz the power, 2014-02-13 18:37:44]

Change committed as 13438