[Mesa-dev] [PATCH] docs: describe the NVC0_COMPUTE envvar

[Samuel Pitoiset]

On 03/01/2015 11:06 PM, Ilia Mirkin wrote:
> On Sun, Mar 1, 2015 at 4:55 PM, Samuel Pitoiset
> <samuel.pitoiset at gmail.com> wrote:
>> On 03/01/2015 07:21 PM, Ilia Mirkin wrote:
>>> Why wouldn't they always just be enabled :) Also, do you have to set
>>> it to get MP counters on kepler+? And why is it called "compute" if it
>>> enables performance counters... probably does more than that, no?
>>
>> It is not enabled by default because compute support on Fermi has weird
>> effects on 3D. And, I didn't find the time to investigate about this, that's
>> why it is still a bit experimental. :)
>>
>> Actually, NVC0_COMPUTE allows compute support which is required for MP
>> counters in order to polls them. However, we only have to set it for
>> nvc0:nve4 chipsets, so the documentation of this envvar could be more
>> accurate.
>>
>> Should I have to submit a v2 and improve the documentation by something like
>> "NVC0_COMPUTE - enable compute support which is required for MP performance
>> counters on nvc0:nve4" ?
> It's unclear to me that this option should be advertised at all. First
> off it has weird side-effects, and secondly none of the other nouveau
> env vars are listed there (NV50_PROG_DEBUG/NV50_PROG_OPTIMIZE).

Okay, you have strong arguments against it.

And, if the other nouveau "debug" env vars are not described on that page,
NVC0_COMPUTE should not be there too.

Thank you for the review Ilia.

> However if you feel strongly about it, how about
>
> "Enable compute support on Fermi chips. This is required for ______ .
> However it has side-effects of ______ ."
>
> Note that "MP performance counters" only makes sense to people who are
> already fully aware of the NVC0_COMPUTE env var, and if they weren't,
> the first place they'd look would be in the code and not that envvars
> page.
>
>    -ilia