Michel Dänzer michel at daenzer.net
Wed Feb 4 22:29:34 PST 2015

On 05.02.2015 13:54, Ilia Mirkin wrote:
> On Wed, Feb 4, 2015 at 11:45 PM, Michel Dänzer <michel at daenzer.net> wrote:
>> On 05.02.2015 12:48, Ilia Mirkin wrote:
>>> Is there a benchmark that demonstrates this? I'd like to test it out
>>> with nouveau.
>> Mesa demos src/tests/streaming_rect on Kaveri (radeonsi):
>> Unpatched:  42 frames in  1.023 seconds = 41.056 FPS
>> Patched:   615 frames in  1.000 seconds = 615.000 FPS
> Hm, 260fps for me on a GF108 (nvc1) with or without the patch (with
> vblank_mode=0 LIBGL_DRI3_DISABLE=1). I guess with nouveau you don't
> get that uncacheable nonsense?

I guess so, but note that the 'uncacheable nonsense' (i.e.
write-combining for CPU mappings of BOs in GART) accounts for about half
of the extra boost I'm getting with Fredrik's patches.

Anyway, if you look at the description of PIPE_USAGE_* in
src/gallium/docs/source/screen.rst, PIPE_USAGE_STAGING is the only one
which can be expected to be fast for CPU reads.

Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer

More information about the mesa-dev mailing list