Re: [PATCH v2 19/24] ASoC: SOF: ipc4-topology: Set FAST_MODE for host copier in compr mode

[Pierre-Louis Bossart <pierre-louis.bossart@linux.dev>]

>>> FAST_MODE allows the host DMA to work in opportunistic, free running mode,
>>> which matches with the bitstream nature of compressed devices.
>>
>> Does it?
> 
> Yes, it does.
> 
>> My memory of the 'opportunistic' behavior was that the DMA would try to
>> use every possible window to send to/from the DSP, so that the bus would
>> be idle for longer periods of time overall.
> 
> The FAST_MODE detaches the host copier form the sampling rate completely
> and will 'allow' prolonged periods w/o new data to be available. The
> IPC4 definition of the feature bit says:
> /* If FAST_MODE bit is set in CopierModuleCfg::copier_feature_mask then
>    copier is able to transfer more than ibs. This bit shall be set only
>    if all sinks are connected to data processing queue.
> */
>> That would make sense as well for PCM, no?
> 
> We have similar mode for PCMs with Deep Bufferr, but that is a different
> mechanism: it only makes the DMA move in bigger chunks, expressed in ms
> (SOF_TKN_INTEL_COPIER_DEEP_BUFFER_DMA_MS -> dma_buffer_size), so it is
> still bound by the sampling rate checks. It is not opportunistic.
>> Maybe this is the right thing to do for compressed data, I just wonder
>> why it is NOT the right thing to do for PCM...
> 
> Basically FAST_MODE is a free run mode of the host copier. A PCM stream
> has a sampling rate the host side can and should be paced by, a
> compressed stream is a bitstream where the host side has no such reference.

For the host-dsp transfers, there is no reason at all for the DMAs to be
paced by the PCM rate...It's only for the DSP-interface DMAs that it
makes sense.