def bench_ifft():
name = 'ifft'
print("computing {}".format(name), end="", flush=True)
fix_num = 1
for dim in shapes :
for shape in dim :
for module in modules:
m = importlib.import_module(module)
data = gen_data(numpy, numpy.random.random, shape)
data = m.asarray(data)
if module is 'scipy':
res, rt = run_benchmark(m, m.fft.ifft, data)
else:
res, rt = run_benchmark(m, m.fft.ifft, data)
if module is 'numpy':
res_np = res
elif module is 'nlcpy':
res_vp = res
elif module is 'scipy':
res_sp = res
set_dict(module, name, rt, shape)
print(".", end="", flush=True)
numpy.testing.assert_allclose(res_np, res_vp)
numpy.testing.assert_allclose(res_np, res_sp)
write_runtime(name, str(fix_num))
fix_num += 1
cleanup_dict()
print("done", flush=True)
def bench_solve():
print("computing solve", end="", flush=True)
for shape in shapes:
for module in modules:
m = importlib.import_module(module)
data1 = gen_data(numpy, numpy.random.random, shape)
data2 = gen_data(numpy, numpy.random.random, shape)
data1 = m.asarray(data1)
data2 = m.asarray(data2)
res, rt = run_benchmark(m, m.linalg.solve, (data1, data2))
if module is 'numpy':
res_np = res
elif module is 'nlcpy':
res_vp = res
set_dict(module, 'solve\nlinear equation', data1.nbytes, rt)
print(".", end="", flush=True)
numpy.testing.assert_allclose(res_np, res_vp, atol=1e-6)
print("done", flush=True)
def bench_std():
print("computing std", end="", flush=True)
for shape in shapes:
for module in modules:
m = importlib.import_module(module)
data = gen_data(numpy, numpy.random.random, shape)
data = m.asarray(data)
res, rt = run_benchmark(m, m.std, data)
if module is 'numpy':
res_np = res
elif module is 'nlcpy':
res_vp = res
set_dict(module, 'standard\ndeviation', data.nbytes, rt)
print(".", end="", flush=True)
numpy.testing.assert_allclose(res_np, res_vp)
print("done", flush=True)
def bench_fft():
print("computing fft", end="", flush=True)
for shape in shapes:
for module in modules:
m = importlib.import_module(module)
data = gen_data(numpy, numpy.random.random, shape)
data = m.asarray(data)
res, rt = run_benchmark(m, m.fft.fft, data)
if module is 'numpy':
res_np = res
elif module is 'nlcpy':
res_vp = res
set_dict(module, 'multiple\n1-D fft', data.nbytes, rt)
print(".", end="", flush=True)
numpy.testing.assert_allclose(res_np, res_vp)
print("done", flush=True)
def bench_matmul():
compute_func = lambda data: data @ data
print("computing matmul", end="", flush=True)
for shape in shapes:
for module in modules:
m = importlib.import_module(module)
data = gen_data(numpy, numpy.random.random, shape)
data = m.asarray(data)
res, rt = run_benchmark(m, compute_func, data)
if module is 'numpy':
res_np = res
elif module is 'nlcpy':
res_vp = res
set_dict(module, 'matrix\nmultiplication', data.nbytes, rt)
print(".", end="", flush=True)
numpy.testing.assert_allclose(res_np, res_vp)
print("done", flush=True)
def bench_unary():
for op in unary_ops:
print("computing {}".format(op), end="", flush=True)
for shape in shapes:
for module in modules:
m = importlib.import_module(module)
compute_func = getattr(m, op)
data = gen_data(numpy, numpy.random.random, shape)
data = m.asarray(data)
if op in not_float_ops:
data = data.astype(dtype=int)
res, rt = run_benchmark(m, compute_func, data, is_binary=False)
if module is 'numpy':
res_np = res
elif module is 'nlcpy':
res_vp = res
set_dict(module, op, data.nbytes, rt)
print(".", end="", flush=True)
numpy.testing.assert_allclose(res_np, res_vp)
print("done", flush=True)