[Pixman] [PATCH] ARMv6: Add fast path for over_reverse_n_8888

[Siarhei Siamashka]

On Fri, 28 Mar 2014 11:13:21 +0200
Pekka Paalanen <ppaalanen at gmail.com> wrote:

> From: Ben Avison <bavison at riscosopen.org>
> 
> Benchmark results, "before" is upstream commit
> c343846 lowlevel-blt-bench: add in_reverse_8888_8888 test
> and "after" is with this patch only added on top.
> 
> lowlevel-blt-bench, over_reverse_n_8888, 100 iterations:
> 
>        Before          After
>       Mean StdDev     Mean StdDev   Confidence   Change
> L1    15.1    0.1    274.5    2.3    100.00%   +1718.9%
> L2    12.8    0.3    181.8    0.7    100.00%   +1315.5%
> M     10.8    0.0     77.9    0.0    100.00%    +621.2%
> HT     9.7    0.0     29.4    0.2    100.00%    +204.9%
> VT     9.5    0.0     26.7    0.1    100.00%    +179.3%
> R      9.3    0.0     25.3    0.1    100.00%    +173.6%
> RT     6.0    0.1     11.0    0.2    100.00%     +82.9%
> 
> At most 16 outliers rejected per case per set.
> 
> cairo-perf-trace with trimmed traces, 30 iterations:
> 
>                                     Before          After
>                                    Mean StdDev     Mean StdDev   Confidence   Change
> t-poppler.trace                    12.9    0.1      9.7    0.0    100.00%     +32.6%
> t-firefox-talos-gfx.trace          33.2    0.7     32.9    0.4     95.23%      +0.9%  (insignificant)
> t-firefox-particles.trace          27.4    0.1     27.3    0.2     99.65%      +0.4%
> t-firefox-canvas-alpha.trace       20.5    0.3     20.5    0.3     57.51%      +0.3%  (insignificant)
> t-poppler-reseau.trace             22.4    0.1     22.4    0.1     95.69%      +0.3%  (insignificant)
> t-firefox-fishtank.trace           13.2    0.0     13.2    0.0     99.84%      +0.1%
> t-swfdec-giant-steps.trace         14.9    0.0     14.9    0.0     87.68%      +0.1%  (insignificant)
> t-swfdec-youtube.trace              7.8    0.0      7.8    0.0     35.22%      +0.1%  (insignificant)
> t-firefox-planet-gnome.trace       11.5    0.0     11.5    0.0     29.37%      +0.0%  (insignificant)
> t-firefox-fishbowl.trace           21.2    0.0     21.2    0.0     18.09%      +0.0%  (insignificant)
> t-grads-heat-map.trace              4.4    0.0      4.4    0.0      1.84%      +0.0%  (insignificant)
> t-firefox-paintball.trace          18.0    0.0     18.0    0.0     33.43%      -0.0%  (insignificant)
> t-firefox-talos-svg.trace          20.5    0.0     20.5    0.1     68.56%      -0.1%  (insignificant)
> t-midori-zoomed.trace               8.0    0.0      8.0    0.0     99.98%      -0.1%
> t-firefox-canvas-swscroll.trace    32.1    0.1     32.1    0.1     85.27%      -0.1%  (insignificant)
> t-gnome-system-monitor.trace       17.2    0.0     17.2    0.0     99.97%      -0.2%
> t-firefox-chalkboard.trace         36.5    0.0     36.6    0.0    100.00%      -0.2%
> t-firefox-asteroids.trace          11.1    0.0     11.1    0.0    100.00%      -0.2%
> t-firefox-canvas.trace             17.9    0.0     18.0    0.0    100.00%      -0.3%
> t-chromium-tabs.trace               4.9    0.0      4.9    0.0     97.95%      -0.3%  (insignificant)
> t-xfce4-terminal-a1.trace           4.8    0.0      4.8    0.0    100.00%      -0.4%
> t-firefox-scrolling.trace          31.1    0.1     31.2    0.1    100.00%      -0.5%
> t-evolution.trace                  13.7    0.1     13.8    0.1     99.99%      -0.6%
> t-gnome-terminal-vim.trace         22.0    0.2     22.2    0.1     99.99%      -0.7%
> t-gvim.trace                       33.2    0.2     33.5    0.2    100.00%      -0.8%
> 
> At most 6 outliers rejected per case per set.
> 
> Cairo perf reports the running time, but the change is computed for
> operations per second instead (inverse of running time).
> 
> Changes in the order of +/- 1% can be accounted for measurement errors,
> even if they are deemed to be statistically significant. This claim is
> based on comparing two 30-iteration identical "before" runs using the
> exact same binaries, and observing changes from -0.4% to +0.5% with
> >=99% confidence.
> 
> Confidence is based on Welch's t-test.
> 
> v4, Pekka Paalanen <pekka.paalanen at collabora.co.uk> :
> 	Rebased, re-benchmarked on Raspberry Pi, commit message.

Thanks, pushed this patch to pixman git.

It looks really good. And the other patches too, except for one
problem (more about it in another e-mail).

-- 
Best regards,
Siarhei Siamashka