[Pixman] Resurrecting: Added fast path for "pad" type repeats

[Pekka Paalanen]

On Mon, 7 Apr 2014 12:22:15 +0300
Pekka Paalanen <ppaalanen at gmail.com> wrote:

> Hi,
> 
> this thread started from
> http://lists.freedesktop.org/archives/pixman/2013-February/002619.html
> and continued in
> http://lists.freedesktop.org/archives/pixman/2013-March/002677.html
> 
> I'd like to hear what the thoughts of it are now, more questions below.

...

> In his original email, Ben showed a speedup of "3.86x" in
> t-firefox-chalkboard trimmed Cairo trace on ARMv6.

That was for the "pad" patch alone.

For the record, I rebased the over a year old patches 2-4 from
http://lists.freedesktop.org/archives/pixman/2013-February/002619.html
onto what I have so far posted on the pixman mailing list. The total
effect from those three patches on RPi is still, that the trimmed
t-firefox-chalkboard Cairo trace runs at 5x speed compared to without
the three patches.

No other Cairo trace from Siarhei's trimmed set shows reliable
improvements.

                                    Before          After
                                   Mean StdDev     Mean StdDev   Confidence   Change
t-firefox-chalkboard.trace         36.0   0.03      7.1   0.02    100.00%    +409.9%
t-chromium-tabs.trace               4.9   0.03      4.9   0.02    100.00%      +0.6%
t-grads-heat-map.trace              4.5   0.04      4.5   0.04     96.54%      +0.5%  (insignificant)
t-firefox-talos-svg.trace          20.6   0.05     20.5   0.04    100.00%      +0.5%
t-swfdec-youtube.trace              7.8   0.04      7.8   0.03     95.75%      +0.2%  (insignificant)
t-firefox-fishbowl.trace           21.2   0.04     21.1   0.03    100.00%      +0.2%
t-swfdec-giant-steps.trace         14.9   0.03     14.9   0.03     98.41%      +0.1%  (insignificant)
t-firefox-planet-gnome.trace       10.9   0.02     10.9   0.02     99.30%      +0.1%
t-firefox-fishtank.trace           13.2   0.01     13.2   0.01     99.82%      +0.1%
t-gnome-terminal-vim.trace         20.0   0.12     20.0   0.12     31.63%      +0.1%  (insignificant)
t-midori-zoomed.trace               8.0   0.01      8.0   0.01     83.50%      +0.0%  (insignificant)
t-evolution.trace                  13.1   0.06     13.1   0.06     19.74%      +0.0%  (insignificant)
t-firefox-canvas.trace             18.3   0.07     18.3   0.05     30.80%      -0.0%  (insignificant)
t-firefox-paintball.trace          14.1   0.01     14.1   0.01     99.70%      -0.0%
t-xfce4-terminal-a1.trace           4.8   0.01      4.8   0.01     98.62%      -0.2%  (insignificant)
t-firefox-canvas-alpha.trace       20.8   0.33     20.9   0.33     47.74%      -0.3%  (insignificant)
t-firefox-particles.trace          27.6   0.16     27.7   0.16     93.63%      -0.3%  (insignificant)
t-firefox-asteroids.trace          11.1   0.01     11.1   0.01    100.00%      -0.3%
t-gnome-system-monitor.trace       17.3   0.07     17.4   0.08     99.59%      -0.3%
t-firefox-scrolling.trace          24.9   0.09     24.9   0.12     99.74%      -0.4%
t-poppler-reseau.trace             22.5   0.11     22.6   0.12     99.99%      -0.5%
t-poppler.trace                     9.8   0.05      9.9   0.08    100.00%      -0.9%
t-gvim.trace                       33.3   0.20     33.6   0.22    100.00%      -0.9%
t-firefox-talos-gfx.trace          25.8   0.45     26.0   0.50     94.93%      -0.9%  (insignificant)
t-firefox-canvas-swscroll.trace    32.2   0.12     32.6   0.18    100.00%      -1.1%

30 iterations done per test per set.
At most 4 outliers rejected per test per set.


Thanks,
pq