# FFT and convolution test (Python), "generators" version
#
# Copyright (c) 2020 Project Nayuki. (MIT License)
# https://www.nayuki.io/page/free-small-fft-in-multiple-languages
#
# Copyright (C) 2021 Luke Kenneth Casson Leighton <lkcl@lkcl.net>
# https://libre-soc.org/openpower/sv/remap/
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
# - The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
# - The Software is provided "as is", without warranty of any kind, express or
# implied, including but not limited to the warranties of merchantability,
# fitness for a particular purpose and noninfringement. In no event shall the
# authors or copyright holders be liable for any claim, damages or other
# liability, whether in an action of contract, tort or otherwise, arising
# from, out of or in connection with the Software or the use or other
# dealings in the Software.
#
import cmath, math, random
from copy import deepcopy
from remap_fft_yield import iterate_indices
#
# Computes the discrete Fourier transform (DFT) or inverse transform of the
# given complex vector, returning the result as a new vector.
# The vector can have any length. This is a wrapper function. The inverse
# transform does not perform scaling, so it is not a true inverse.
#
def transform(vec, inverse, generators=False):
n = len(vec)
if n == 0:
return []
elif n & (n - 1) == 0: # Is power of 2
return transform_radix2(vec, inverse, generators)
else: # More complicated algorithm for arbitrary sizes
assert False
#
# Computes the discrete Fourier transform (DFT) of the given complex vector,
# returning the result as a new vector.
# The vector's length must be a power of 2. Uses the Cooley-Tukey
# decimation-in-time radix-2 algorithm.
#
def transform_radix2(vec, inverse, generators_mode):
# Returns the integer whose value is the reverse of the lowest 'width'
# bits of the integer 'val'.
def reverse_bits(val, width):
result = 0
for _ in range(width):
result = (result << 1) | (val & 1)
val >>= 1
return result
# Initialization
n = len(vec)
levels = n.bit_length() - 1
if 2**levels != n:
raise ValueError("Length is not a power of 2")
# Now, levels = log2(n)
coef = (2 if inverse else -2) * cmath.pi / n
exptable = [cmath.rect(1, i * coef) for i in range(n // 2)]
# Copy with bit-reversed permutation
vec = [vec[reverse_bits(i, levels)] for i in range(n)]
#
# Radix-2 decimation-in-time FFT
#
if generators_mode:
# loop using SVP64 REMAP "generators"
# set the dimension sizes here
# set total. err don't know how to calculate how many there are...
# do it manually for now
VL = 0
size = 2
while size <= n:
halfsize = size // 2
tablestep = n // size
for i in range(0, n, size):
for j in range(i, i + halfsize):
VL += 1
size *= 2
# set up an SVSHAPE
class SVSHAPE:
pass
# j schedule
SVSHAPE0 = SVSHAPE()
SVSHAPE0.lims = [n, 0, 0]
SVSHAPE0.order = [0,1,2]
SVSHAPE0.mode = 0b01 # FFT mode
SVSHAPE0.skip = 0b00
SVSHAPE0.offset = 0
SVSHAPE0.invxyz = [0,0,0] # inversion if desired
# j+halfstep schedule
SVSHAPE1 = deepcopy(SVSHAPE0)
SVSHAPE1.skip = 0b01
# k schedule
SVSHAPE2 = deepcopy(SVSHAPE0)
SVSHAPE2.skip = 0b10
# enumerate over the iterator function, getting 3 *different* indices
for idx, (jl, jh, k) in enumerate(zip(iterate_indices(SVSHAPE0),
iterate_indices(SVSHAPE1),
iterate_indices(SVSHAPE2))):
if idx >= VL:
break
# exact same actual computation, just embedded in a single
# for-loop but using triple generators to create the schedule
temp1 = vec[jh] * exptable[k]
temp2 = vec[jl]
vec[jh] = temp2 - temp1
vec[jl] = temp2 + temp1
else:
# loop using standard python nested for-loops
size = 2
while size <= n:
halfsize = size // 2
tablestep = n // size
for i in range(0, n, size):
k = 0
for j in range(i, i + halfsize):
# exact same actual computation, just embedded in
# triple-nested for-loops
jl, jh = j, j+halfsize
temp1 = vec[jh] * exptable[k]
temp2 = vec[jl]
vec[jh] = temp2 - temp1
vec[jl] = temp2 + temp1
k += tablestep
size *= 2
return vec
#
# Computes the circular convolution of the given real or complex vectors,
# returning the result as a new vector. Each vector's length must be the same.
# realoutput=True: Extract the real part of the convolution, so that the
# output is a list of floats. This is useful if both inputs are real.
# realoutput=False: The output is always a list of complex numbers
# (even if both inputs are real).
#
def convolve(xvec, yvec, realoutput=True):
assert len(xvec) == len(yvec)
n = len(xvec)
xvec = transform(xvec, False)
yvec = transform(yvec, False)
for i in range(n):
xvec[i] *= yvec[i]
xvec = transform(xvec, True)
# Scaling (because this FFT implementation omits it) and postprocessing
if realoutput:
return [(val.real / n) for val in xvec]
else:
return [(val / n) for val in xvec]
###################################
# ---- Main and test functions ----
###################################
def main():
global _maxlogerr
# Test power-of-2 size FFTs
for i in range(0, 12 + 1):
_test_fft(1 << i)
# Test power-of-2 size convolutions
for i in range(0, 12 + 1):
_test_convolution(1 << i)
print()
print(f"Max log err = {_maxlogerr:.1f}")
print(f"Test {'passed' if _maxlogerr < -10 else 'failed'}")
def _test_fft(size):
input = _random_vector(size)
expect = _naive_dft(input, False)
actual = transform(input, False, False)
actual_generated = transform(input, False, True)
assert actual == actual_generated # check generator-version is identical
err_gen = _log10_rms_err(actual, actual_generated) # superfluous but hey
err = _log10_rms_err(expect, actual)
actual = [(x / size) for x in expect]
actual = transform(actual, True)
err = max(_log10_rms_err(input, actual), err)
print(f"fftsize={size:4d} logerr={err:5.1f} generr={err_gen:5.1f}")
def _test_convolution(size):
input0 = _random_vector(size)
input1 = _random_vector(size)
expect = _naive_convolution(input0, input1)
actual = convolve(input0, input1, False)
print(f"convsize={size:4d} logerr={_log10_rms_err(expect, actual):5.1f}")
# ---- Naive reference computation functions ----
def _naive_dft(input, inverse):
n = len(input)
output = []
if n == 0:
return output
coef = (2 if inverse else -2) * math.pi / n
for k in range(n): # For each output element
s = 0
for t in range(n): # For each input element
s += input[t] * cmath.rect(1, (t * k % n) * coef)
output.append(s)
return output
def _naive_convolution(xvec, yvec):
assert len(xvec) == len(yvec)
n = len(xvec)
result = [0] * n
for i in range(n):
for j in range(n):
result[(i + j) % n] += xvec[i] * yvec[j]
return result
# ---- Utility functions ----
_maxlogerr = -math.inf
def _log10_rms_err(xvec, yvec):
global _maxlogerr
assert len(xvec) == len(yvec)
err = 10.0**(-99 * 2)
for (x, y) in zip(xvec, yvec):
err += abs(x - y) ** 2
err = math.sqrt(err / max(len(xvec), 1)) # a root mean square (RMS) error
err = math.log10(err)
_maxlogerr = max(err, _maxlogerr)
return err
def _random_vector(n):
return [complex(random.uniform(-1.0, 1.0),
random.uniform(-1.0, 1.0)) for _ in range(n)]
if __name__ == "__main__":
main()