I'm experiencing troubles with Full Speed isochronous USB (at just above 6 Mbps) where the USB DMA activity seems to be the cause of dropped sync events on my other DMA channels (which are critical to my function).
An idea I just had is that I could skip the USB DMA transfer from DSP memory to USB Buffer RAM by directly writing my 8-bit isochronous packet data into the proper I/O space address for the endpoint USB Buffer RAM. Normally, it would seem like a resource waste to replace a DMA transfer with a software loop copying individual bytes. However, in my application I already need to loop through the 16-bit data and pack it into 12-bit pseudo words to stay within the Full Speed bandwidth, so I was thinking that I could just modify my 12-bit packing routine to output to I/O space in the USB Buffer RAM instead. This might actually take fewer cycles, and could easily alleviate the sync drop events on the other DMA channels.
I would have to manage the double-buffer ping pong myself, and properly calculate the Buffer RAM address so that I can continue to use the CSL USB API for the other endpoints. In general I was thinking that if I had the source for USB_postTransaction() then I could probably very quickly write a solution. Source for the full CSL USB API would be best, but perhaps I would only need the one function, depending upon whether it would compile on its own using just the public headers.
Is this source available? Would I need to make a special arrangement?
P.S. This request is related to my ongoing problem with isochronous USB transactions where a non-USB DMA channel is dropping events, but only when USB is active. I'm assuming that avoiding USB DMA could help with the other problem, thus this request. See "C5506 DMA to EMIF causes many Sync drop events, but not DMA to SARAM."
