On 09/21/2012 01:59 PM, Takashi Iwai wrote:
Hi,
recently Pierre questioned about the validity of COMBO positioning mode, and it turned out that the position reporting can be handled better if we take the difference between LPIB and DMAPOS as runtime->delay. (In short, LPIB points the position where the data fed to the codec, and DMAPOS points the data fetched to DMA.)
We cooked up a test patch below and it seems working well on a few machines I've tested. I think we might have still chance to put this into 3.7, as this would be a good improvement. The patch below is a combined work of two fixes: fixing the better stream start by always using SSYNC bit, and adjusting the position and runtime->delay.
Interesting. A quick question: is this affecting playback, recording, or both?
Of course, the biggest question is the test coverage. David, could you check whether this doesn't break so much on (some) machines you can test? Of course, if any other people can test it, it'll be a great help, too.
The patch should be applied to the latest sound.git tree master branch, but it should be applicable (and easily resolvable) for older versions.
Against sound master:
patching file sound/pci/hda/hda_intel.c Hunk #2 succeeded at 1987 (offset -17 lines). Hunk #3 succeeded at 2012 (offset -17 lines). Hunk #4 succeeded at 2044 (offset -17 lines). Hunk #5 succeeded at 2144 (offset -17 lines). Hunk #6 succeeded at 3426 (offset -104 lines). Hunk #7 succeeded at 3490 (offset -104 lines).
The patch extends this new mode only for recent Intel chips, just to be sure.
I don't think I have much such hardware here. But I've made DKMS package for easy testing by others. If you're running Ubuntu 12.04, you can just install this package, reboot, and test:
http://people.canonical.com/~diwic/temp/alsa-hda-dkms-ssync-delay_0.1_all.de...