Mesa 19.2.6 Released Due To POWER Fallout

[phoronix]

Phoronix: Mesa 19.2.6 Released Due To POWER Fallout

Mesa 19.2.5 was just released earlier this week but now v19.2.6 has already been released due to the previous point release breaking IBM POWER builds...

http://www.phoronix.com/scan.php?page=news_item&px=Mesa-19.2.6-Released

[-MacNuke-]

Four releases in two weeks... They really should ramp up the CI efforts.

[rene]

Four releases in two weeks... They really should ramp up the CI efforts.

yeah, and QA

[skeevy420]

I think I took a wrong turn at Albuquerque.

[Calinou]

Four releases in two weeks... They really should ramp up the CI efforts.

Cross-compiling for uncommon architectures isn't easy, and would require valuable CI minutes that could be spent testing more common use cases.

[-MacNuke-]

Cross-compiling for uncommon architectures isn't easy, and would require valuable CI minutes that could be spent testing more common use cases.

What "CI minutes"? It's not like someone is sitting in front of some monitor monitoring CI activity. It is a fully automated service that does not need any action at all. It just reports if everything is OK based on the rules.

[robclark]

Four releases in two weeks... They really should ramp up the CI efforts.

Our CI has been ramping up considerably in last couple months, from basically only intel's post-merge CI (admittedly they do have a pretty nice setup) prior to the gitlab migration, to today where we have in addition to pre-merge build-tests, we have pre-merge deqp running on actual hardware (two, soon to be three generations of freedreno, two generations of panfrost, and folks are working on adding other drivers), in addition to softpipe and llvmpipe in the cloud.

Ofc non of this would have helped this particular POWER issue, because that isn't something in CI. But the decentralized nature of the gitlab CI setup means that if someone had some POWER hw sitting around (enough that a CI run could finish in ~5min and not become a bottleneck) we could add that to the mix.

UPDATE: I'll add to that, POWER has the disadvantage of not too many users, and the folks who use POWER aren't likely to be running bleeding edge mesa master. So bugs specific to POWER don't get noticed/reported until after the release.

[madscientist159]

Our CI has been ramping up considerably in last couple months, from basically only intel's post-merge CI (admittedly they do have a pretty nice setup) prior to the gitlab migration, to today where we have in addition to pre-merge build-tests, we have pre-merge deqp running on actual hardware (two, soon to be three generations of freedreno, two generations of panfrost, and folks are working on adding other drivers), in addition to softpipe and llvmpipe in the cloud.

Ofc non of this would have helped this particular POWER issue, because that isn't something in CI. But the decentralized nature of the gitlab CI setup means that if someone had some POWER hw sitting around (enough that a CI run could finish in ~5min and not become a bottleneck) we could add that to the mix.

UPDATE: I'll add to that, POWER has the disadvantage of not too many users, and the folks who use POWER aren't likely to be running bleeding edge mesa master. So bugs specific to POWER don't get noticed/reported until after the release.

We could spin a CI instance up on Integricloud -- what would we need to get it added to Mesa CI?

[nuetzel]

yeah, and QA

Some bit slower release adoption and 'you' were save...;-)

[robclark]

We could spin a CI instance up on Integricloud -- what would we need to get it added to Mesa CI?

https://gitlab.freedesktop.org/mesa/...b-ci/README.md is a reasonable starting point (although some of the points about GPU reset wouldn't apply to llvmpipe/softpipe on POWER testing). I guess first step is that you'd need to have a POWER builder to product mesa binary artifacts for the test job to test. The toplevel setup of what jobs depend on what is in $mesa/.gitlab-ci.yml

anholt and tomeu and eric_engestrom (and probably some others) on #dri-devel are probably worth chatting with, as they set up a lot of the current CI infrastructure.