[Mesa-dev] [PATCH 0/8] radeonsi: Implement AMD_performance_monitor

[Nicolai Hähnle]

On 24.11.2015 17:20, Marek Olšák wrote:
> On Tue, Nov 24, 2015 at 5:08 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>> On Tue, Nov 24, 2015 at 11:00 AM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
>>> One annoyance of the series as it currently stands is that exposing thousands
>>> of counters affects the context creation time due to the way the Mesa state
>>> tracker initializes itself, especially given that this is a feature that
>>> isn't used in production. Dealing with that is on my todo-list.
>>
>> Just a thought - only expose the (majority of the) counters in debug contexts?
>
> Deferring the initialization until the first use seems better.

I agree.

> If piglit takes longer to run now, the issue should be addressed
> before the patches land.

I haven't tested the running time of piglit, but with the patches I've 
just posted the overhead should be essentially zero now.

Cheers,
Nicolai

> Marek
>