[Piglit] non-concurrent piglit tests

[Eric Anholt]

Ilia Mirkin <imirkin at alum.mit.edu> writes:

> On Sun, Nov 22, 2015 at 7:12 PM, Eric Anholt <eric at anholt.net> wrote:
>> Ilia Mirkin <imirkin at alum.mit.edu> writes:
>>
>>> It looks like we're up to something like 1K non-concurrent piglit
>>> tests... maybe more. Can someone who actually understands the issues
>>> explain what makes a piglit test unreliable when run concurrently with
>>> another test? Then we can go and enable concurrency on probably 75% of
>>> the currently-marked-nonconcurrent tests.
>>
>> It's mostly us being conservative, historically.  We were scared to turn
>> it on for everything because of DRI1 buffer sharing , and just did it
>> for compiler tests first.  Then I moved shader_runner to do things to
>> fbos instead of windows and set a bunch of those to concurrent.  Then we
>> added more, etc.
>>
>> It turns out we haven't really had any problems with concurrent being
>> set, so we should probably just set it on for everything and see if
>> anything even breaks.
>
> It looks like we only add -auto to all piglits. Should we be adding
> -fbo -auto instead so that an fbo is bound to a drawbuffer? Presumably
> it falls back on regular winsys when EXT_framebuffer_object is not
> supported...

I do know you'll see failure with adding -fbo to everything.  You could
probably find a bunch of the cases with grepping for files with
piglit_present_results && glBindFramebuffer && !glBindFramebuffer(0).  I
also noticed just yesterday that -samples= doesn't work with -fbo.

I don't think adding -fbo is necessary to get everything concurrent,
though.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/piglit/attachments/20151123/37b12be1/attachment.sig>