x11perf putimagexy10 extremly slow for glamor on i965

[Clemens Eisserer]

Hi,

I have observed extremly low x11perf-putimagexy10 results when using
glamor on top of the i965 driver - while shmput10 results are quite
ok.
First I thought this was limited to intel haswell, but my old
arrandale system is affected as well:

x11perf -putimagexy10: 1200 reps @   6.1682 msec (   162.0/sec):
PutImage XY 10x10 square
x11perf -shmput10: 1200000 reps @   0.0046 msec (217000.0/sec):
ShmPutImage 10x10 square

At least on arrandale, about 90% of cycles are spent in clflush_object
- so GEM memory domain overhead.
I wonder what could be the reason, last time I had a look at glamor
both code-paths used the same glTexSubImage2D routine to upload the
texture data,

Br, Clemens

PS: I've reported a bug a few months ago:
https://bugs.freedesktop.org/show_bug.cgi?id=98549