DSP Resources

Maximum Number of Audio Voices

Q: What is the maximum number of audio voices?

A: This is a complex issue. The number of voices available is determined by the sum of all resources used by all the active voices. These include:

EI, I, and EO memory
The 512 word instruction, `N', memory
The thirteen input FIFOs and four output FIFOs
The 500+ ticks of DSP cycles per frame

Typically the N memory and the number of ticks are the limiting factors. Instrument prioritization and voice stealing provide a good dynamic real-time system that optimizes DSP usage.

The real limits are not known yet, but you can play at least 8 variable rate, 16-bit samples with loudness control. Fixed rate samples are much less expensive than variable rate samples and are limited by the 13 input FIFOs. Resonant filters are cheaper than a variable rate sampler.

Allocating DSP Resources

Q: After eight loads of halfmono8.dsp, the ninth load fails and I receive the error message: "Couldn't allocate DSP resources." Can you explain why?

A: Only eight fit. Halfmono8.dsp requires 49 DSP code words, and 38 DSP tick(cycles). For code words, the total is 512. The system takes 70, and a mixer4x2.dsp takes 32. That leaves 410 available. Divide that by 49 yields 8.36 halfmono8 instruments worth of code maximum.

For ticks, the total is 565. The system takes 45, and a mixer4x2.dsp takes 32, that leaves 488 available. Dividing that by 38 yields 12.84 halfmono8 instruments worth of ticks maximum. The code words are the limiting resource.

DSP DMA

Q: How many DMA channels are used for each DSP program such as sampler.dsp? Does the number of channels vary with the instrument?

A: Synthesis instruments such as svfilter.dsp and triangle.dsp use 0 DMA channels. Mono and stereo sample playing instruments each use one DMA channel. The only instrument that uses two input DMA channels is sampler3d.dsp. The delay1tap.dsp instrument uses one output DMA channel and one input DMA channel.

Q: How many DMA channels can be allocated for the DSP?

A: There are 13 INPUT DMA channels used for playing samples. There are four OUTPUT DMA channels used for echo and delay effects.

Q: Could you briefly tell me about the relationship between the DSP and DMA controller?

A: The DSP plays audio samples by reading the sample data from a register in the DSP address space. There is a register for each DMA channel. The register connects to a FIFO which can store up to eight 16 bit sample words. When the FIFO is down to two sample words, a DMA request is issued that fills the FIFO from main memory.

DSP Usage Monitor

Q: It seems to me that I saw a tool somewhere that would give a readout of the current used DSP resources. Can anyone tell me where this is?

A: The tool is called audiomon and it resides in $c (an alias set to $boot/System/Programs). audiomon displays a bar graph of DSP resources on the graphics screen. There is also a tool called DumpDSPP.c that prints to the terminal so it can be used when the graphics screen is busy.

Dynamic Voice Allocation

Q: Is dynamic voice allocation supported?

A: The score playing code in the Music library supports dynamic voice allocation. When a voice is needed, the code attempts the following:

Adopt an unused instrument of the same type.
If that fails create a new instrument.
If that fails adopt another instrument of a different type, kill it and use its resources.
If that fails scavenge instruments from itself or others.

The code uses the Audio folio calls AbandonInstrument(), AdoptInstrument(), ScavengeInstrument(), and GetAudioItemInfo() to implement these functions.

For an example of using the score player as a sound effects manager, see the example program tsc_soundfx.c.