[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [partial-reconfig] Another BM Question...



Roman, John...

Thank you very much, both of you, for your replies...

It turns out that in correcting an error, the Xilinx guys made
another... the previous version of xapp290 said for a left-most BM, bits
0 and 1 cannot go right-to-left... someone asked if this was correct,
and they replied that it wasn't, and that bits 2 and 3 cannot move
right-to-left... In the "corrected" version though, they fixed the bits
but messed up the direction...

The use of one output per CLB column (i.e. two horizontal slices)
remains a somewhat contentious issue, though... in August I had posted
about using two CLB columns (four horizontal slices) for a Spartan-IIE
BM, similar to the Virtex-II BM... 

http://www.itee.uq.edu.au/~listarch/partial-reconfig/archive/2005/08/msg00054.html

Apart from one reply, by Harish, who said he had built a similar BM for
the Virtex family, there weren't any others... I had therefore gone
ahead with the technique, building several BMs that I used for some
designs... none of them have made it to an FPGA yet, but the design flow
went ahead with no problems whatsoever, with the final PAR routing the
design perfectly with no errors...

However, Roman's post, and a look at the Spartan-IIE datasheets raised
several doubts in my mind about whether such BMs would actually work...
the datasheet provides a figure which seems to vindicate Roman about one
BM output being accessed by one CLB... using FPGA_Editor also seems to
suggest that the output from a single line goes to a CLB GRM, and
therefore four CLBs will be needed to read four lines... I'm uploading
the figure... 

A close look in FPGA_Editor at my designs, however, suggests that all is
well... all bits are apparently read at the destination module, although
some lines continue all the way to the edge of the device and travel
down vertically along the edge on to the CLBs that need them... the
outputs to the GRM don't really come into play... 

In the design I speak of, the BMs are all placed on the left-most
position... some move left-to-right and therefore should work... I can
fix the others by using the technique John suggests, but if the two-CLB
theory fails, my design, which is the way it is for maximum module
density, would fall flat on its face... I can unpload the final .ncd
file if anyone would like to have a look at it... 

Thanks in advance... really worried here... 

:)

Sincerely,
Umar

-- 
http://www.fastmail.fm - A fast, anti-spam email service.

Attachment: s2e_tbuf_bus.JPG
Description: image/pjpeg