On Sun, 08 Dec 2019 18:31:27 +0100, Klaus Ethgen wrote:
Hi Takashi,
Am So den 8. Dez 2019 um 9:45 schrieb Takashi Iwai:
On Sat, 07 Dec 2019 21:06:43 +0100,
Find attached the perf output of the relevant event handler.
Thanks. So it seems wasting in the busy loop for waiting for the response from the codec.
How about a patch below? It's untested, might break things, but at least avoiding a busy loop.
I'll give it a try. Kernel is compiling right now. It might be hard to see if the patch is working but I suppose that it do when I don't have that trouble within a week or so.
BTW, if your machine is Coffeelake or Cannonlake, this patch wont' work on the old branch because of forced polling mode. It was removed in 5.5-rc1, so please test with Linus tree at best.
I hardly have such new Architectures on that laptop. It is a lenovo x61s so it is from latest 2009. It has a Intel Core 2 Duo LV (Merom). :-)
OK, then could you give alsa-info.sh output, at least? Run the script with --no-upload option and attach the output.
Basically my patch should improve a lot for Nvidia codecs, not only about the CPU time reduction, and it'd be worth for merging to upstream. OTOH, it's a change at the very core part, so it needs more testing, so it's likely for 5.6.
thanks,
Takashi