Bl616 support

[marchingband]

Hello,
I am working on a port for BL616 (a RISC-V MCU with builtin audio ADC and DAC).
So far it seems straightforward, using Daisy as a template.
Would it be ok to ask for support here, should I need it?
thank you.

[dpwe]

This is a good place to ask.

The Daisy port was made much easier because we were able to take advantage of their existing I2S-feeding interrupts. The other Arduino platforms take different approaches: The ESP32 setup uses the RTOS multitasking to have separate threads for feeding the DAC, but the RP2040/RP2350 setup relies on polling to keep the DAC filled.

Good luck!

[marchingband]

Well, it went pretty smoothly, but I can only get 1 or 2 voices of DX7, and 4 voices of the Juno.
This is with MIDI USB Host running, and using the built-in audio DAC.
It's a great target for Amy in some ways, but only 300Mhz.

[dpwe]

That's surprising - on other platforms we usually get more DX7 voices than Juno (mainly, I think, because the DX7 doesn't run a VCF). The RP2040 gets 6 Juno patch 0 voices at 240 MHz, but that is across two cores.

One thing I found helpful is to think about where the MCU is waiting when it's idle. Generally this ends up being that it's blocking on the DAC write (since the output buffer is full), although on interrupt-serviced DACs maybe it's just zipping through its loop task very rapidly. To get maximum performance you want to ensure it can spend no time blocking when load is high.

I also used spare GPIOs and an oscilloscope to see exactly how long was being spent in rendering, copying to the DAC, etc, which helped convince me the cycles weren't being wasted.

[dpwe]

By the way, I'm interested in how you did it. It feels weird that we have three different approaches for three platforms - interrupts on Daisy, actual threads on ESP32, and polling on RP2040/2350. I'd like to have a genetic approach. I'm guessing you did polling (based on the RP2040 code).

[bwhitman]

I don't have a BL616 but it does look like it should handle much more than 2 DX7 voices using AMY fixed point.

The ESP32-P4 runs at 400MHz, just a little over, at RISC-V, and gets us 20+ Juno voices (and probably 30 DX7)

My uninformed guess the DAC (12-bit) could be slow and eating CPU cycles while writing (and blocking AMY compute). Does it have a DMA mode? (Edit, yes, it does: https://github.com/sipeed/M0S_BL616_example/blob/main/examples/peripherals/dac/dac_dma/main.c )

[marchingband]

I am running Amy in the DAC DMA interrupt, it’s a 16bit DAC. The USB host uses some ticks, without that I was getting more, but since my application requires the USB host I didn’t stress test that too hard. If P4 gets 40, I wonder if I should keep working on this to see if I can get more.

…

On Tue, Jul 1, 2025 at 8:18 AM Brian Whitman ***@***.***> wrote: *bwhitman* left a comment (shorepine/amy#466) <#466 (comment)> I don't have a BL616 but it does look like it should handle much more than 2 DX7 voices using AMY fixed point. The ESP32-P4 runs at 400MHz, just a little over, at RISC-V, and gets us 20+ Juno voices (and probably 30 DX7) My uninformed guess the DAC (12-bit) could be slow and eating CPU cycles while writing (and blocking AMY compute). Does it have a DMA mode? — Reply to this email directly, view it on GitHub <#466 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIE667DDZTNCUPG64BGYRIT3GKREBAVCNFSM6AAAAACANF7Y3OVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTAMRUGQ4DKNBQGA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[marchingband]

The DAC is mono, is it ok to have Amy output mono? There is a comment saying not to do that. Could that speed things up?

[bwhitman]

AMY now defaults to stereo and is not configurable to mono at run time. However, you can edit amy.h to have AMY run in mono at compile time: #define AMY_NCHANS 1. Or, you can only send every other sample of the rendered int16_t block to the DAC.

We love looking at devices that can run AMY and will try to get a BL616 to test on to help you out, but let us know if you learn any more!

[marchingband]

I recommend one of the AiThinker boards, because they have most pins broken out. The m0s dock has very few pins and is very expensive for some reason.

…

On Tue, Jul 1, 2025 at 6:43 PM Brian Whitman ***@***.***> wrote: *bwhitman* left a comment (shorepine/amy#466) <#466 (comment)> AMY now defaults to stereo and is not configurable to mono at run time. However, you can edit amy.h to have AMY run in mono at compile time: #define AMY_NCHANS 1. Or, you can only send every other sample of the rendered int16_t block to the DAC. We love looking at devices that can run AMY and will try to get a BL616 to test on to help you out, but let us know if you learn any more! — Reply to this email directly, view it on GitHub <#466 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIE667GBGKOSTPANY75X7XL3GM2K3AVCNFSM6AAAAACANF7Y3OVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTAMRWGA3DINJQGM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[marchingband]

I ran some benchmarks on amy_render, and this sometimes peaks to 10ms with 3 voices of DX7 playing, which seems an order of magnitude too high.
I am not sure why this would be the case, at 320Mhz.
I am running at 44100hz, 256byte buffer, stereo.
Possibly someone more knowledgable in risc-v isa could improve things? There must be a misaligned store, a memory cache issue, or something else that's a little outside my expertise.

If you decide to investigate, let me know how to best share my work so far with you.