def gpu_version(self, sig, filt, boundary, mode):
with cp.cuda.Stream.null:
out = cusignal.correlate2d(sig,
filt,
boundary=boundary,
mode=mode)
cp.cuda.Stream.null.synchronize()
return out
def test_correlate2d(num_samps, num_taps, boundary, mode):
cpu_sig = np.random.rand(num_samps, num_samps)
cpu_filt = np.random.rand(num_taps, num_taps)
gpu_sig = cp.asarray(cpu_sig)
gpu_filt = cp.asarray(cpu_filt)
cpu_correlate2d = signal.correlate2d(
cpu_sig, cpu_filt, boundary=boundary, mode=mode
)
gpu_correlate2d = cp.asnumpy(
cusignal.correlate2d(gpu_sig, gpu_filt, boundary=boundary, mode=mode)
)
assert array_equal(cpu_correlate2d, gpu_correlate2d)
def test_correlate2d(self, rand_2d_data_gen, num_samps, num_taps, boundary,
mode, use_numba):
cpu_sig, gpu_sig = rand_2d_data_gen(num_samps)
cpu_filt, gpu_filt = rand_2d_data_gen(num_taps)
cpu_correlate2d = signal.correlate2d(cpu_sig,
cpu_filt,
boundary=boundary,
mode=mode)
gpu_correlate2d = cp.asnumpy(
cusignal.correlate2d(
gpu_sig,
gpu_filt,
boundary=boundary,
mode=mode,
use_numba=use_numba,
))
assert array_equal(cpu_correlate2d, gpu_correlate2d)