Adding support for Pounder DDS QSPI profile stream #186
Conversation
src/main.rs
Outdated
| ); | ||
|
|
||
| // IO_Update should be latched for 50ns after the QSPI profile write. Profile writes | ||
| // are always 16 bytes, with 2 cycles required per byte, coming out to a total of 32 |
There was a problem hiding this comment.
This is now configurable, but the limit is 16 bytes. This is only limited by the QSPI FIFO size and the size used in the ad9959 serializer, so it could theoretically change in the future and invalidate these numbers
|
Above is a capture of SCK going to the DDS and the IO_Update signal. As can be seen, IO_Update is properly toggling on for 50nS after the DDS profile has been written over SPI |
| _ => panic!("Invalid"), | ||
| } | ||
| unsafe { | ||
| core::slice::from_raw_parts::<'a, u32>( |
There was a problem hiding this comment.
I think for this you'd have to consume self here (and not indirectly in write_profile), no?
There was a problem hiding this comment.
The issue here is that we're returning a slice and not an array. The slice has to be owned by something to exist properly. If we consumed self to generate a [u32; 4] array here and then loaned a slice ref to that array, the array itself would actually go out of scope too soon (in this function return). Hence, we have to keep self alive for the entire lifetime of the u32 slice.
I've played around with it and haven't found a nicer way to move ownership around, but if you have an idea, feel free to suggest it
There was a problem hiding this comment.
Instead of the slice, maybe return (move) the array and a length.
For precision and metrology applications we'll need to be able to set and get frequencies without rounding errors. Setting requires that the user can pass in (ultimately) the desired frequency as a fraction |
|
bors r+ |
|
Bors isn't behaving. |
This is an initial draft PR of the proposed updates to the high-speed QSPI data stream to Pounder.
Current measurements indicate that each profile serialization + staging process currently takes 300ns for a profile of 12 bytes. The actual QSPI data write is only 75ns for 3 32-bit words, so there is room for future improvement when the layout of each profile is known in advance.
TODO:
Reserved for future PR:
Define a "static" layout for the pounder profile to avoid the overhead of the serialization process. This would allow us to copy bytes into an array at known offsets and do it really quickly.
This fixes #172 and addresses part of the IO architecture in #147