[Intel-gfx] Howto enable i830FALLBACK, regressions with xorg/intel from git.

Clemens Eisserer linuxhippy at gmail.com
Thu Oct 9 13:12:44 CEST 2008


Hi,

> In i830.h, enable I830DEBUG define will enable FALLBACK debug info.
Thanks a lot, good to know.

I wrote a simple test-case which I use as benchmark when executed in a
loop, the test basically only does fillrect(s) and composition.

With intel-2.1.1 on an 945GM I get the following profile:
116144   51.1772  no-vmlinux               no-vmlinux               (no symbols)
58445    25.7529  intel_drv.so             intel_drv.so             (no symbols)
22990    10.1302  [vdso] (tgid:2135 range:0xb7fc1000-0xb7fc2000) Xorg
                   (no symbols)
3428      1.5105  libexa.so                libexa.so
ExaOffscreenMarkUsed

However with xserver/intel git on an 945GM I get:
116343   78.2012  libpixman-1.so.0.13.1    libpixman-1.so.0.13.1
fbCompositeSolidMask_nx8x8888mmx
8811      5.9224  libpixman-1.so.0.13.1    libpixman-1.so.0.13.1
pixman_fill_mmx
6743      4.5324  no-vmlinux               no-vmlinux               (no symbols)
3767      2.5320  libpixman-1.so.0.13.1    libpixman-1.so.0.13.1    pixman_fill

So it seems this simple test is rendered software-only, however I
don't get any fallback-messages.
I also saw this behaviour with some other  tests I did, so it seems
not specific to that workload.
Could there be configuration issues which prevents the intel driver to
not allocate pixmaps in vram?

I uploaded the debug-output of the program running to:
http://pastebin.com/m719bb53a
The "benchmark" itself to: http://pastebin.com/f6030e18a
And Xorg.log to: http://pastebin.com/m2a59a061

Thanks, Clemens



More information about the Intel-gfx mailing list