[alsa-devel] non-interleaved mode
Hi,
The driver for the sound engine I'm working on requires NONINTERLEAVED. Will ALSA core give my driver a full period of left-only samples, followed by a full period of right-only samples? Or are left/right samples split within a single period (ie. first half of period is left samples, second half is right samples)?
Thanks, Steve
At Tue, 10 Jul 2007 11:08:01 -0700, Steve Longerbeam wrote:
Hi,
The driver for the sound engine I'm working on requires NONINTERLEAVED. Will ALSA core give my driver a full period of left-only samples, followed by a full period of right-only samples? Or are left/right samples split within a single period (ie. first half of period is left samples, second half is right samples)?
The former case. The standard non-interleaved buffer consists of (the same buffer-size of) full-length of mono-streams.
Takashi
On Wed, 11 Jul 2007 10:47:56 +0200 Takashi Iwai tiwai@suse.de wrote:
At Tue, 10 Jul 2007 11:08:01 -0700, Steve Longerbeam wrote:
Hi,
The driver for the sound engine I'm working on requires NONINTERLEAVED. Will ALSA core give my driver a full period of left-only samples, followed by a full period of right-only samples? Or are left/right samples split within a single period (ie. first half of period is left samples, second half is right samples)?
The former case. The standard non-interleaved buffer consists of (the same buffer-size of) full-length of mono-streams.
Takashi
Doesn't this depend upon the driver implementation ? IIRC, I tried to request NON-INTERLEAVED on the PXA270 AC97 driver and it told me that wasn't a permitted operation (one of the reasons that I needed to use mmap access in SALSA-Lib).
At Wed, 11 Jul 2007 09:13:24 -0400, J. Scott Merritt wrote:
On Wed, 11 Jul 2007 10:47:56 +0200 Takashi Iwai tiwai@suse.de wrote:
At Tue, 10 Jul 2007 11:08:01 -0700, Steve Longerbeam wrote:
Hi,
The driver for the sound engine I'm working on requires NONINTERLEAVED. Will ALSA core give my driver a full period of left-only samples, followed by a full period of right-only samples? Or are left/right samples split within a single period (ie. first half of period is left samples, second half is right samples)?
The former case. The standard non-interleaved buffer consists of (the same buffer-size of) full-length of mono-streams.
Takashi
Doesn't this depend upon the driver implementation ? IIRC, I tried to request NON-INTERLEAVED on the PXA270 AC97 driver and it told me that wasn't a permitted operation (one of the reasons that I needed to use mmap access in SALSA-Lib).
The latter case, "multi-streams in each period" style, cannot be represented in ALSA API, more specifically, snd_pcm_channel_info struct. Thus, it has to be remapped virtually to the normal non-interleaved streams (each with full periods), or to avoid mmap and copy the data inside the driver.
There is another COMPLEX_MMAP access type, but it's also for linear frames with constant increments, where each channel can be represented with snd_pcm_channel_info record. The "multi-streams in each period" is non-linear.
Takashi
Takashi Iwai wrote:
At Wed, 11 Jul 2007 09:13:24 -0400, J. Scott Merritt wrote:
On Wed, 11 Jul 2007 10:47:56 +0200 Takashi Iwai tiwai@suse.de wrote:
At Tue, 10 Jul 2007 11:08:01 -0700, Steve Longerbeam wrote:
Hi,
The driver for the sound engine I'm working on requires NONINTERLEAVED. Will ALSA core give my driver a full period of left-only samples, followed by a full period of right-only samples? Or are left/right samples split within a single period (ie. first half of period is left samples, second half is right samples)?
The former case. The standard non-interleaved buffer consists of (the same buffer-size of) full-length of mono-streams.
Takashi
Doesn't this depend upon the driver implementation ? IIRC, I tried to request NON-INTERLEAVED on the PXA270 AC97 driver and it told me that wasn't a permitted operation (one of the reasons that I needed to use mmap access in SALSA-Lib).
The latter case, "multi-streams in each period" style, cannot be represented in ALSA API, more specifically, snd_pcm_channel_info struct. Thus, it has to be remapped virtually to the normal non-interleaved streams (each with full periods), or to avoid mmap and copy the data inside the driver.
There is another COMPLEX_MMAP access type, but it's also for linear frames with constant increments, where each channel can be represented with snd_pcm_channel_info record. The "multi-streams in each period" is non-linear.
Last night I ran an experiment. I created a 16-bit stereo .wav file with all left samples = 0x5555, and all right samples = 0xAAAA. I then played the .wav using aplay, and just before trigger start in my driver I searched the DMA buffer for the first occurrence of 0xAAAA. The right samples didn't start until exactly half-way through the entire DMA buffer!
So now I'm confused, you say there should be one full period of left, followed by one full period of right, but I'm observing that the whole buffer is split in half, with the first half left, the second half right. How did this happen? Is there any way for a driver to query ALSA core about the non-interleaved layout of samples in memory?
I'm running 2.6.21.1 which is alsa-driver 1.0.14rc3, and I'm using alsa-lib 1.0.14rc3 and alsa-utils 1.0.14rc2.
Steve
On Wed, 11 Jul 2007, Steve Longerbeam wrote:
Last night I ran an experiment. I created a 16-bit stereo .wav file with all left samples = 0x5555, and all right samples = 0xAAAA. I then played the .wav using aplay, and just before trigger start in my driver I searched the DMA buffer for the first occurrence of 0xAAAA. The right samples didn't start until exactly half-way through the entire DMA buffer!
Yes, it's the right non-interleaved behaviour. You can imagine this layout as separate "mono" ring buffers for each channels - see snd_pcm_lib_ioctl_channel_info() in pcm_lib.c . Driver might modify info->offset if required (see hdsp.c driver for example).
Jaroslav
----- Jaroslav Kysela perex@suse.cz Linux Kernel Sound Maintainer ALSA Project, SUSE Labs
Jaroslav Kysela wrote:
On Wed, 11 Jul 2007, Steve Longerbeam wrote:
Last night I ran an experiment. I created a 16-bit stereo .wav file with all left samples = 0x5555, and all right samples = 0xAAAA. I then played the .wav using aplay, and just before trigger start in my driver I searched the DMA buffer for the first occurrence of 0xAAAA. The right samples didn't start until exactly half-way through the entire DMA buffer!
Yes, it's the right non-interleaved behaviour. You can imagine this layout as separate "mono" ring buffers for each channels - see snd_pcm_lib_ioctl_channel_info() in pcm_lib.c . Driver might modify info->offset if required (see hdsp.c driver for example).
thanks, that cleared everything up for me.
Steve
participants (4)
-
J. Scott Merritt -
Jaroslav Kysela -
Steve Longerbeam -
Takashi Iwai