[myhdl-list] myhdl with cython?

[David B. <da...@be...>]

Dear All,

i've been fiddling with myhdl for about a week or so. I've tried to do some real stuff I do at work using myhdl. As a part of the project involves FIR filtering, I have used the example found on Christopher Felton's page (https://bitbucket.org/cfelton/examples) and hacked away m_firfilt entity using my own coefficients. Now, my filter has 891 taps and I've tried to simulate using a testbench how it reacts on unit-amplitude (it is a bandpass) input signal.

It works great, at the same time it takes 'ages'. Simulation of such filter with roughly 5000 clock cycles takes almost 2 minutes. 4 times longer than I do with modelsim (from mentor).

So I wanted to check, whether it is feasible to use cython for these things. Following http://docs.cython.org/src/tutorial/cython_tutorial.html I hae created the setup and compiled in-the module using build_ext --inplace.

All that works OK, but then I tried to import the compiled module in python, and I get:

--------------------------------------------------------------------------------
>>> import FIR_tb
Working
Loaded  891  coefficients
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "FIR_tb.pyx", line 46, in init FIR_tb (FIR_tb.c:2580)
    tbm = traceSignals(tb)
  File "/usr/local/lib/python2.7/dist-packages/myhdl-1.0dev-py2.7.egg/myhdl/_traceSignals.py", line 85, in __call__
    h = _HierExtr(name, dut, *args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/myhdl-1.0dev-py2.7.egg/myhdl/_extractHierarchy.py", line 238, in __init__
    _top = dut(*args, **kwargs)
  File "FIR_tb.pyx", line 19, in FIR_tb.tb (FIR_tb.c:1720)
    @always(delay(10))
  File "/usr/local/lib/python2.7/dist-packages/myhdl-1.0dev-py2.7.egg/myhdl/_always.py", line 54, in _always_decorator
    raise AlwaysError(_error.ArgType)
myhdl.AlwaysError: decorated object should be a classic (non-generator) function
--------------------------------------------------------------------------------

for a testbench, which works OK when ran as a python module (non-compiled). For the sake of completeness here is the testbench (the entity itself is the same as in Christopher's examples)

--------------------------------------------------------------------------------
from myhdl import *
from FIR import *

print "Working"


def tb():

    # read FIR coeffs
    coeffs = map(int, open("/home/belohrad/git/didt/Python/firfilter/filter_coefficients_newline.txt", "rt").readlines())
    print "Loaded ", len(coeffs), " coefficients"

    DxD, QxD = [Signal(intbv(0)[32:0].signed()) for x in xrange(2)]
    ResetxRN = ResetSignal(0, active = 0, async=True)
    ClkxC = Signal(bool(0))

    i_fir = m_firfilt (ClkxC, ResetxRN, DxD, QxD, coeffs)

    @always(delay(10))
    def driveClk():
        ClkxC.next = not ClkxC

    @instance
    def tbi():
        ResetxRN.next = 0
        DxD.next = 0
        yield(delay(150))
        yield (ClkxC.posedge)
        ResetxRN.next = 1
        yield (ClkxC.posedge)
        # test unity but this is a band-pass filter
        for i in xrange(4906):
            DxD.next = 1
            yield (ClkxC.posedge)
            DxD.next = 0
            yield (ClkxC.posedge)
            DxD.next = -1
            yield (ClkxC.posedge)
            DxD.next = 0
            yield (ClkxC.posedge)

        raise StopSimulation()

    return tbi, driveClk, i_fir

tbm = traceSignals(tb)
sim = Simulation(tbm).run()
print "Done"
--------------------------------------------------------------------------------


Now, the error message is very cryptic to me. Is there anyone who tried to use cython for simulation? Is this error fatal and I cannot use cython at all due to some technical restrictions?


Thanks
.david.