This is difficult to test without hardware mocks or some
form of phase readback, but the symptom was that e.g.
`self.dds.set(…, ref_time=now_mu() - 1)` would fail
periodically, that is, whenever bit 32 of the timestamp
would be set (which would be turned into the sign bit).
This is a fairly sinister issue, and is probably a compiler
bug of some sort (either accepts-invalid or wrong type inference).
Previously the TSC was truncated to an even coarse RTIO periods before doing
the setting SPI xfer. Afterwards the the IO update pulse would introduce
at least one but less than two RTIO cycles. Ultimately the RTIO TSC was
truncated again to even. If the SPI xfer takes an odd number of RTIO
periods, then a subsequent xfer would collide.
close#1229
Signed-off-by: Robert Jördens <rj@quartiq.de>
This now reliably locates the SYNC_CLK-IO_UPDATE edge by doing two
scans at different delays between start and stop IO_UPDATE.
It also works well when one delay is very close to the edge.
And it correctly identifies which (start or stop) pulse hit or crossed
the SYNC_CLK edge.
for #1143
Signed-off-by: Robert Jördens <rj@quartiq.de>
* simplified and cross-referenced the explanation of the different
phase modes.
* semantically and functionally merged absolute and tracking/coherent
phase modes.
* simplified numerics to calculate phase correction
* added warning about possible inconsistency with DMA and default
phase mode
* restricted __all__ imports
* moved continuous/relative phase offset tracking from an instance
variable to a "handle" returned by set()/set_mu() in order to avoid
state inconsistency with DMA (#1113#1115)
for #1143
Signed-off-by: Robert Jördens <rj@quartiq.de>
* expose multi device sync functionality
* sync delay configuration interface
* auto-tuning of sync delay from device_db seed
for #1143
Signed-off-by: Robert Jördens <rj@quartiq.de>
