[[!meta title="Eliminating glyph fallbacks"]]

[[!tag exa performance xorg intel i965]]

Sometimes things get worse before they get better.

A few days ago, I presented a patch for [storing glyphs as
pixmaps](http://cworth.org/exa/storing_glyphs_as_pixmaps/) which
improved performance, but not as dramatically as one would have hoped.

I profiled the result and found that there were still a lot of
software fallbacks going on. Tracking things down, (hints: enable
DEBUG\_TRACE\_FALL in xserver/exa/exa_priv.h and I830DEBUG in
xf86-video-intel/src/i830.h), I found a simple case statement that was
falling back to software for any compositing operation targeting an A8
buffer. Fortunately, it looks like this fallback was due to a
limitation in older graphics card that doesn't exist on the i965. So a
very simple [patch](Allow-i965-compositing-to-target-A8-buffers.patch)
eliminates the software fallback.

So lets take a look at before-and-after profiles:

<dl class="chart barchart">
    <dt><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//system.oprofile">
aa10text-fallbacks/</a> (<a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//
timing">144000 chars./sec.</a>) <a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fall
backs//system.symbols">symbols profile</a></dt>
    <dd style="width:65.9722%;">
        <ul>
            <li class="libexa" style="width:27.1577%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//libexa.oprofile">libexa</a><span>27%</span></li>
            <li class="libpixman" style="width:26.8338%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//libpixman.oprofile">libpixman</a><span>27%</span></li>
            <li class="vmlinux" style="width:24.6667%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//vmlinux.oprofile">vmlinux</a><span>25%</span></li>
            <li class="Xorg" style="width:4.9222%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//Xorg.oprofile">Xorg</a><span>5%</span></li>
            <li class="libc-2_6" style="width:4.4754%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//libc-2.6.oprofile">libc-2.6</a><span>4%</span></li>
            <li class="oprofiled" style="width:3.4928%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//oprofiled.oprofile">oprofiled</a><span>3%</span></li>
            <li class="intel_drv" style="width:2.5876%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//intel_drv.oprofile">intel_drv</a><span>3%</span></li>
            <li class="other" style="width:5.8638%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-fallbacks//other.oprofile">other</a><span>6%</span></li>
        </ul>
    </dd>
    <dt><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//system.oprofile">aa10text-no-fallbacks/</a> (<a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//timing">95000 chars./sec.</a>) <a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//system.symbols">symbols profile</a></dt>
    <dd style="width:100%;">
        <ul>
            <li class="vmlinux" style="width:42.1575%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//vmlinux.oprofile">vmlinux</a><span>42%</span></li>
            <li class="intel_drv" style="width:26.7106%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//intel_drv.oprofile">intel_drv</a><span>27%</span></li>
            <li class="libexa" style="width:7.9861%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//libexa.oprofile">libexa</a><span>8%</span></li>
            <li class="librt-2_6" style="width:7.7359%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//librt-2.6.oprofile">librt-2.6</a><span>8%</span></li>
            <li class="libc-2_6" style="width:5.3533%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//libc-2.6.oprofile">libc-2.6</a><span>5%</span></li>
            <li class="Xorg" style="width:3.9202%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//Xorg.oprofile">Xorg</a><span>4%</span></li>
            <li class="oprofiled" style="width:3.2670%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//oprofiled.oprofile">oprofiled</a><span>3%</span></li>
            <li class="other" style="width:2.8694%;"><a href="/exa/i965/eliminating_glyph_fallbacks/aa10text-no-fallbacks//other.oprofile">other</a><span>3%</span></li>
        </ul>
    </dd>
</dl>

Yikes! The patch takes us from 144k chars/sec. to only 95k
chars/sec. I'm regressing performance! But look again, and see that
the libexa time has been cut dramatically, and the libpixman time has
been eliminated altogether. That's exactly what we would hope to see
for eliminating software fallbacks. So I've finally gotten this
text-rendering benchmark to involve no software fallbacks. Hurrah!

Meanwhile, the intel_drv and vmlinux time have increased
dramatically. Take a look at how hot those hotspots are in their
profiles:

intel_drv:

	samples  %        symbol name
	29614    41.2170  i965_prepare_composite
	26641    37.0792  I830WaitLpRing
	9143     12.7253  i965_composite
	1618      2.2519  I830Sync

vmlinux:

	samples  %        symbol name
	28775    25.3748  delay_tsc
	21956    19.3616  system_call
	7535      6.6446  getnstimeofday
	5109      4.5053  schedule

So this is just the same, old [synchronous
compositing](http://cworth.org/exa/i965/synchronous_composite/) bug I
identified earlier. Performance has gotten worse since I'm stressing
out the driver and this bug more.

Dave Airlie has been doing some recent work that should let us fix
that bug once and for all. Hopefully it won't be too long before I can
actually post some positive progress here.

PS. I've also gotten one report that my patch for storing glyphs as
Pixmaps speeds glyph rendering up initially, but after the X server
has been running for about an hour or so, things get *really*
slow. Shame on me for not doing any testing more extensive than
starting the X server and then running a single client for a few
minutes, (either firefox or x11perf). The report is that most of the
time is disappearing into ExaOffscreenMarkUsed. Well the good news is
that Dave's work eliminates that function entirely, (along with lots
of migration code in EXA), so hopefully there's not any big problem to
fix there. I'll have to test more thoroughly after synching up with
Dave.