[Dspforum] FFT QDMA help

[O'Connor, Sean R.]

That was it!  Thanks, Doug.

One question about resetting the CIPR.  waitfordma() waits for bit 0 in CIPR to be set to 1 (using a mask = 1), but at the end of waitfordma() CIPR is 'cleared' and set to mask, or 1.  I guess I was expecting CIPR bit 0 to be reset to 0?

________________________________
From: dspforum-bounces at echelonembedded.com [mailto:dspforum-bounces at echelonembedded.com] On Behalf Of Wenstrand, Doug S.
Sent: Thursday, October 22, 2009 12:33 PM
To: dspforum at echelonembedded.com
Subject: Re: [Dspforum] FFT QDMA help

When its works at non-real time, but doesn't at real-time, its almost certain that you aren't waiting for the DMA completion correctly.
Take a look at the wait_for_dma() function in the notes.  It is doing something that you aren't doing - namely it looks for the CIPR bit to be set, but then makes sure to clear it for the next time.  If that bit in the CIPR remains high, you will think your next DMA transfer is complete right away - even though it has barely started
Doug



From: dspforum-bounces at echelonembedded.com [mailto:dspforum-bounces at echelonembedded.com] On Behalf Of O'Connor, Sean R.
Sent: Thursday, October 22, 2009 11:41 AM
To: dspforum at echelonembedded.com
Subject: [Dspforum] FFT QDMA help

Doug/Joseph,

I am having issues with this assignment, so I have uploaded my project directory to the website.

The QDMA portion appears to work fine.  I verified this by loading increasing integers (cast as floats) into the RAM starting at address 0x80000000, QDMAing the data to the internal memory and then QDMAing it out starting at 0x803A9800 without doing any FFT functions.  I exported this file and analyzed it in matlab, and it was perfectly intact.

Performing the STFT on the data produces QNANs and INFs in the data output, however, a single FFT works fine.  I can QDMA any chunk of 8192 complex samples, take the FFT, QDMA it out to 0x803A9800, export it to matlab and it looks fine.  It seems multiple FFTs cause issues... I will continue to work at it, but any input you might have would be greatly appreciated.

Thanks,
Sean
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://echelonembedded.com/pipermail/dspforum_echelonembedded.com/attachments/20091022/37c69018/attachment.html>