diff options
author | Karsten Wiese <annabellesgarden@yahoo.de> | 2005-10-18 14:31:07 +0200 |
---|---|---|
committer | Jaroslav Kysela <perex@suse.cz> | 2005-11-04 13:19:00 +0100 |
commit | 4f550df58f4758ea023704b409830ad9c3b47771 (patch) | |
tree | 83f935c11032a2a4099e0996717482ddcce945ca /sound/pci/intel8x0.c | |
parent | f0597a416dc44e3afe25090e9af9d42bad62547d (diff) | |
download | op-kernel-dev-4f550df58f4758ea023704b409830ad9c3b47771.zip op-kernel-dev-4f550df58f4758ea023704b409830ad9c3b47771.tar.gz |
[ALSA] Reduce interrupt latency in sound/pci/via82xx.c
Modules: VIA82xx driver
The change only affects the via823x kind of chips.
Here the via8233_pcm_pointer_hw() function
(named snd_via8233_pcm_pointer() before)
needed to loop until a non zero position is red from the chip.
Measurements have shown that more than 200 loops are typically needed on
an Athlon64.
As io-reads cost many cycles, those loops sum up huge.
via8233_pcm_pointer_hw() runs either in interrupt or with interrupts
disabled. So it introduces significant interrupt latency.
The patch introduces a calculated position value hwptr_done,
that is updated by the interrupt routine when a period is completed.
It is only used, if the 823x chip returns a zero position, which can't
be interpreted reliably.
Further optimisation is applied on the 8233 chip's interrupt routine:
Only the SGD_SHADOW is read, as it contains all infos needed.
We ommit ~5 more register reads that way.
Signed-off-by: Karsten Wiese <annabellesgarden@yahoo.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Diffstat (limited to 'sound/pci/intel8x0.c')
0 files changed, 0 insertions, 0 deletions