Re: [microblaze-uclinux] Microblaze Performance.

[Paul Hartke <phartke@xxxxxxxxxxxx>]

The Microblaze FPU unit is single-precision and won't affect this
application which is double precision based.

In terms of hardware acceleration of computationally complex blocks--such as
IDCTs and FFTs, check out
http://direct.xilinx.com/bvdocs/appnotes/xapp529.pdf

Finally, folks implement algorithms differently in FPGAs than in DSPs since
there are different tradeoffs and ways of doing thing.  Trying to use an
FPGA exactly as you would a DSP isn't likely to get you the results you are
looking for.

Paul

Quoting Abot Botbot <dejanigma@hotmail.com>:
> Hey Folks,
>
>    This is my first post so try not to let my inexperience offend. My
> question relates to the hardware options one uses in setting up the
> microblaze. With EDK and BSB I quickly got a microblaze working. With
> some
> finagling and an XMD stub I got large, complicated programs running out
> of
> the external SDRAM (im using the Digilent S3 starter board, or an
> equivalent). But I see a major problem with the performance. Here are the
> numbers and comparissons I am using: The code runs a 64-point FFT on an
> array of data with real and imaginary parts. It is a lot of calculation,
> so
> it is easy to time. I run 1000 of these in a loop and it takes
> approximately
> 4 minutes, an unfathomably long amount of time in the world of DSP. The
> comparison is vs a Motorola DSP board which is also doing emulated
> floating
> point arithmetic, which performs 1000 FFTs in under a second. My theory
> at
> this point is that I am missing a major design flaw which is inhibiting
> my
> performance. Technically, the microblaze has a core clock of 100 MHz as
> well
> as the SDRAM. They are interfaced using the Cache-Link FSL bus and a BRAM
> cache local to the microblaze. With these statistics, there is no real
> reason for it to behave so poorly. Even adding a hardware FPU doesn't
> change
> the result in the least.
>    Here is what I would like, if anyone would be so kind: A basic rundown
> of
> the hardware configuration options necissary to optomize microblaze
> performance for running from external memory. There must be a way to make
> the microblaze run in an equivalent way to this outdated 1998 Motorola
> 24-bit DSP chip. If not then Xilinx and IBM have a lot of explaining to
> do.
> For refference I will supply the relevant design files I have available.
>
> Thanks So Much,
> Andy
>
> _________________________________________________________________
> Don’t just search. Find. Check out the new MSN Search!
> http://search.msn.click-url.com/go/onm00200636ave/direct/01/
>
___________________________
microblaze-uclinux mailing list
microblaze-uclinux@itee.uq.edu.au
Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
Mailing List Archive : http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/