This wraps libdrm functionality to exec with contexts. This patch
shouldn't be applied until libdrm for contexts is updated.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Due to the way we calculate the workload by doubling it each time we
might end up with almost twice as much as we want. Hence increase our
fudge by 1.5 to account for that.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=50666
We need to hold forcewake lock in order to be able to read GT registers.
Otherwise, when the GPU is in RC6 mode, we'll read all zeros.
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
When looking at the pwrite/pread/wc performance, it is useful to judge
that against the performance of an ordinary CPU mmap.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Using BEGIN_BATCH can lead to a nice inf recursion through require_space
-> flush_batch -> BEGIN_BATCH.
Also fix things up to always require BATCH_RESERVED. We need 2 dwords
for the gen5 workaround and 2 dwords for MI_BB_END.
Assures that signals interrupting the wait works properly. Because of
the scheduling around signals, interrupted waits will *seem* faster as
the GPU continues to work while all the CPU scheduling stuff happens.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
We should get more kms tests soon, and not needing to copy-paste a
nice test pattern should be useful.
That establishes a firm depency of i-g-t on cairo over everything, but
I don't care so much about that.
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The path isn't properly autodetected and it yells stupid stuff into
stderr. Which causes spurious failures.
And really, these are automated tests. If you need to look at the output,
something is wrong.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Somehow nobody noticed this before, but we were missing blt and bsd6
initialization on Gen7.
Reported-by: Philippe Lecluse <philippe.lecluse@intel.com>
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
It optimized away the read loops ;-)
Also fix up the size to not be 1/4th of the one we want for tiled
buffer tests.
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
For now, only print their content for diffing, but also add the necessary
bits that can be used for more verbose output in the fugure.
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
