No longer seems to show that "every packet is corrupt" error. I suspect
it was caused by the "filling the transfer contents" loop in
usbIsoInit() taking place during two 1ms periods. This would, in
theory, cause it to skip over 1ms of data every ISO_PACKETS_PER_CTX *
NUM_FUTURE_CTX frames.
Bugs found recently:
- Trigger doesn't work on negative voltages (at least in MM mode).
- Double Sample Rate button won't de-trigger when CH1 scope is disabled.
The error where every packet can become corrupt when you get a "bad
launch" seems more pronounced here. But it still succeeds enough times
to show that correct synchronisation could fix it.
And found the bug from before; or at least the trigger. It seems to
only occur when the rear PCI USB port is used. Front hub or rear USB2
ports don't show issues at all.
But will not respond until device has been connected at least once! On
a side note I found out how silly the saveState/loadState method in
genericUsbDriver was. It was always a better idea to just "poke" the
signals to ensure that it is reset in the same way that would happen
from GUI interactions.
From the looks of things, anyway.
This fix came with a cost:
- All modes apart from 0 are broken.
- Randomly, the DMA write and USB read may access the same half of
isoBuf. This is determined on boot, and doesn't seem to desynchronise
and resynchronise over time.
The good news is that both of these have a fix. Give it another day of
coding...
Added triple endpoint support for the Desktop interface.
Should be possible to fix the one-packet-per-sample glitch easily, now.
Even if it isn't, the device now only needs to reserve 768 bytes/frame,
not 1023!!!!
Have moved from one big (1023 byte) iso endpoint to 3x 256 byte ones.
This will fix the error where Labrador is picky about which port it's
connected to, and *should* fix the one-sample-per-packet error.
Swapped an append to a replace with an ADT designed to return fast
pointers to char[] blocks.
Worked out that the one-corrupt-sample-per-packet bug was definitely not
fixed. I must have been looking at a 1kHz wave before.
functionGenControl and espoComboBox now use QCoreApplication::
applicationDirPath(). This is to fix deploys on Mac. It’s possible
but unlikely that it broke Windows/Linux.
These #defines are located in desktop_settings.h and allow you to edit
how many samples are discarded per packet. The idea was to use this to
test the MCU-side code and ensure that it doesn't corrupt a packet per
frame, but this seems to have magically disappeared (???).
I remember queueing USB transfers a long time ago, but can't remember
this having any success. It's more likely to just be a random clock
skew issue that only appears on some boards - and it's only "fixed" on
the one particular test board I have in hand now. I think.
It's a bit dense; some would say unnecessarily so. Might see what I can
do with it in the morning but there's a reasonable chance it'll become
unworkable.
- Reverted the dual transistor IC back to a single transistor (TSM
thingo)
- Added PSU indicator LED
- Added PSU control/expansion header.
Note that this is only in eschema, not pcbnew!
That's it. Literally a single #define that determines how many
milliseconds the OS waits before reconnecting. Fixed an error with
Windows not reading it on new mobo, hopefully fixes it under OS X as
well.
May have accidentally broken Linux (untested, I wasn’t being shotgunny
but equally wasn’t being super careful). Seems to randomly fail on
launch. Probably a race condition on the mutex?
EspoTek