Re: [myhdl-list] It's a beautiful day (somewhat long response)

[Angel E. <ezq...@gm...>]

On Thursday, June 9, 2011, Christopher Felton <chr...@gm...> wrote:
> <snip>
>>
>>
>> Simulations
>> ===========
>> As discussed, the first couple simulations I tried failed because of the
>> unsupported dependencies.  Which required some investigation on the
>> failures and possible solutions.  At this time there is no solution for
>> the numpy and company other than rewriting the simulations, not to use
>> numpy.  The following is the list of my projects I attempted to run with
>> pypy.  Most of these are described on the myhdl wiki.
>>
>>     1. Recursive FFT (failed, numpy)
>>     2. GnuRadio Compatible USRP CIC Filter (failed, numpy)
>>     3. Fixed-Point SOS (failed, numpy)
>>     4. USBP (FX2 USB Interface) (failed in myhdl??)
>>
>> All in all, I could not get any of them to run with pypy?  Some were
>> expected but some were not.
>>
>>
>
> Ok, I *learnt* something since yesterday.  First, pypy will not work
> on my PPC Mac.  I should have realized this right away, the JIT is
> for x86 only (correct?).  But the surprising thing is that python
> is crazy slow on my PPC-Mac (circa 2003).  I am running py27 with
> the latest trunk myhdl.  py27 installed with macports (port install)
> To get another system in the mix I stole my wifes laptop which can
> run pypy (side note, how quickly things can be setup with MyHDL/Python).
>
> Also, my non-numpy simulations, yesterday, were failing.
> They failed, because I was using the traceSignals (?).  When I
> removed the traceSignals function I was able to run two of the
> simulations with pypy :)
>
> Below is a table of execution times for the different simulations.
>
> +---------+---------+----------+-----------+----------+----------+
> |         | OSX-CPy | OSXi-CPy | OSXi-pypy | Ubu-CPy  | Ubu-pypy |
> +---------+=========+==========+===========+==========+==========+
> | rFFT    | 10m.52s | 4m.6s    | failed*   | 4m.36s   | failed*  |
> +---------+---------+----------+-----------+----------+----------+
> | CIC     | tbr     | tbr      | failed*   | tbr      | failed*  |
> +---------+---------+----------+-----------+----------+----------+
> | USBP    | tbr     | 4m.30s   | 2m.20s    | 4m.50s   | 2m.28s   |
> +---------+---------+----------+-----------+----------+----------+
> | LED     | 30m.4s  | 15m.15s  | 2m.10s    | 16m.13s  | 1m.59s   |
> +---------+---------+----------+-----------+----------+----------+
> * failed due to the numpy dependency
>
> System configurations
> OSX  - Dual 2GHz PPC G5, 4.5GiB RAM, OS X 10.5.8
> OSXi - 2.26GHz Intel Core 2 Duo, 2GiB RAM, OS X 10.6.7
> Ubu  - AMD Athlon LE1620 1.7GiB RAM, Ubuntu (natty)
>
> The relevant information here is the ~2x increase and ~8x increase
> with pypy.
>
> Hope this is useful.
>
> Chris Felton

Awesome, thanks for the info. Sonit seems that when it works the
speedup is significant!

Jan, any idea why calling traceSignals would fail?

Angel