def test_convolve(num_samps, num_taps, mode='same'):
cpu_sig = np.random.rand(num_samps)
cpu_win = signal.windows.hann(num_taps)
gpu_sig = cp.asarray(cpu_sig)
gpu_win = cusignal.windows.hann(num_taps)
cpu_conv = signal.convolve(cpu_sig, cpu_win, mode=mode)
gpu_conv = cp.asnumpy(cusignal.convolve(gpu_sig, gpu_win, mode=mode))
assert array_equal(cpu_conv, gpu_conv)
def conv(h,sample,device):
if device =='cuda':
from cusignal import convolve
from cupy import array,vstack
if device =='cpu':
from numpy import convolve,vstack
from numpy import array,atleast_2d
res = []
for row_cnt,row in enumerate(sample):
res.append(convolve(h[0],row))
if len(res) == 2:
return vstack(res)
else:
return atleast_2d(array(res))
def gpu_version(self, sig, win, mode, method):
with cp.cuda.Stream.null:
out = cusignal.convolve(sig, win, mode=mode, method=method)
cp.cuda.Stream.null.synchronize()
return out