Cypress FX2 slave FIFO issue

Do you have a question? Post it now! No Registration Necessary

Translate This Thread From English to

Threaded View
Hi all,

I am trying to temporarily disable my EP6 slave FIFO which is set to
auto-in during data stream, and  I am running into trouble no matter
which approach I take.
A little background:
I'm grabbing data from a camera whose Line Valid (LV) signal clearly
indicates when there is valid data, since it's only active when Frame
Valid (FV) is also active. So, my current approach is to directly
connect LV to the SLWR pin, and set SLWR polarity to active high.

If I forget about synchronizing the frame, I can acquire a whole image
from the camera, but the data overlaps from one frame to the next, and
the image is broken up cleanly into four sections, unless I happen to
luckily make the image request at exactly the right time.

To synchronize to the frame, I would like to poll to see when FV has
gone from low to high (it's active high) using a simple input pin. So
the idea is to ignore all data until a request has come in and the
next frame starts. I have tried two ways.

Approach 1:
Normally leave EP6FIFOCFG set to 0x00, and change it temporarily to
0x0C (enable AUTO IN) on command.
This doesn't seem to play nicely at all, I don't get any data out when
I do this. I think you're not supposed to change this after setting it
once, for some reason? The documentation doesn't really say much about
that though.
Alternately I tried changing the VALID bit of EP6CFG, but that also
did not seem to work at all. Once again, I think you're supposed to
set that and leave it.

Approach 2:
Leave E6 FIFO set to AUTO IN all the time, and enable the SLCS which
is controlled by a free IO port on the FX2. This doesn't seem to work
either; I think it messes with EP2 which I use as an OUT endpoint but
with AUTO OUT off so I can just check the input data in my firmware.

I could externally AND the SLWR signal with an I/O port pin that I
control myself, but that's not the cleanest solution.
Is there something else I can do in software to make this work?

Cheers,

Steve


Re: Cypress FX2 slave FIFO issue
Quoted text here. Click to load it

You could do an all software, have your firmware always transfer the
data from the FIFO to the USB bus.  This way you can ignore (remove from
FIFO but do not send to USB) the data you do not want.  The downside,
this approach will be slower.

And don't forget that if the FIFO has less than the fifo size it will
not be auto committed, it will need a PKTEND.  You need a way to flush
the last little bit that might be in the FIFO.  Otherwise you will have
residue data when the FV is active.

Best of luck,
.chris

Re: Cypress FX2 slave FIFO issue
Quoted text here. Click to load it

Chris,

Thanks for the suggestion.
Are you implying that I move each data byte with the CPU, or just
check to see when there is enough data and arm the endpoint manually?

As for the PKTEND thing, the camera is configured to 2048 pixels per
line, which is a multiple of 512, so in theory if I could just disable
the endpoint FIFO during the time that I don't want to transfer data,
it would work out just perfectly.

Steve

Re: Cypress FX2 slave FIFO issue
Alright, I've found a solution that seems to have worked!
This is a bit of a kludge, which is perfectly OK since this is
basically a one-off thing so I can do testing and calibration on my
desk without needing an FPGA frame grabber (which admittedly would
have been easier to handle synchronization with).

At any rate, I was previously running with both FIFOADR 1 and 0 tied
high and low respectively, fixing the input to EP6. This time I took
PA1 and wired it up to FIFOADR0.
It occurred to me that I'm not using EP8 for anything at all, so the
trick is to select EP8 by setting PA1 high to effectively deselect EP6
until the next Frame Valid low-to-high transition after a frame is
requested. Works like a charm!
I'm really not sure why this should work better than using SLCS,
unless SLCS prevents the chip from receiving data from the host PC on
an OUT endpoint which isn't even hooked up to the AUTO OUT on the
slave fifo interface. The TRM is a bit unclear about that, but I guess
that must have been the problem.

Steve

Site Timeline