def test_elemwise_mixed_dtype():
to_cpu = numpy.asarray
for dtype1 in ["int16", "float32", "int8"]:
for dtype2 in ["int16", "float32", "int8"]:
dtypeo = str((numpy.zeros(1, dtype=dtype1) +
numpy.zeros(1, dtype=dtype2)).dtype)
#print "dtypes", dtype1, dtype2, "o dtype", dtypeo
#print " Test inside a wrapping python object 2 inputs"
for shape in [(500,), (50, 5), (5, 6, 7)]:
input_vals = [rand(shape, dtype) for dtype in [dtype1, dtype2]]
del dtype
gpu_vals = [gpu_ndarray.GpuNdArrayObject(i)
for i in input_vals]
assert all([numpy.allclose(to_cpu(ig), i)
for ig, i in zip(gpu_vals, input_vals)])
gpu_vals = [MyGpuNdArray(x) for x in gpu_vals]
out = gpu_vals[0] + gpu_vals[1]
assert numpy.allclose(to_cpu(out),
input_vals[0] + input_vals[1])
out = gpu_vals[0] - gpu_vals[1]
assert numpy.allclose(to_cpu(out),
input_vals[0] - input_vals[1])
out = gpu_vals[0] * gpu_vals[1]
assert all_close(to_cpu(out),
input_vals[0] * input_vals[1])
if dtypeo.startswith("float"):
# TODO: execute for all dtype
out = gpu_vals[0] / gpu_vals[1]
assert numpy.allclose(to_cpu(out),
input_vals[0] / input_vals[1])
nb_in = 4
#print " Test inside a wrapping python object %d inputs"%nb_in
for shape in [(500,), (50, 5), (5, 6, 7)]:
input_vals = [rand(shape, dtype)
for dtype in [dtype1, dtype2, dtype1, dtype2]]
gpu_vals = [gpu_ndarray.GpuNdArrayObject(i)
for i in input_vals]
assert all([numpy.allclose(to_cpu(ig), i)
for ig, i in zip(gpu_vals, input_vals)])
gpu_vals = [MyGpuNdArray(x) for x in gpu_vals]
out = MyGpuNdArray.adds(*gpu_vals)
assert numpy.allclose(to_cpu(out),
reduce(numpy.add, input_vals))
out = MyGpuNdArray.multiplys(*gpu_vals)
assert all_close(to_cpu(out),
reduce(numpy.multiply, input_vals))
#print " Test broadcasting"
for shapes in [((1, 5), (4, 5)), ((33, 10), (33, 1)),
((33, 1, 5), (33, 10, 1)),
((33, 1, 5), (33, 10, 1), ((1, 10, 5))),
]:
input_vals = [rand(shape, dtype) for shape, dtype
in zip(shapes, [dtype1, dtype2])]
gpu_vals = [gpu_ndarray.GpuNdArrayObject(i)
for i in input_vals]
assert all([numpy.allclose(to_cpu(ig), i)
for ig, i in zip(gpu_vals, input_vals)])
gpu_vals = [MyGpuNdArray(x) for x in gpu_vals]
out = MyGpuNdArray.adds(*gpu_vals)
assert numpy.allclose(to_cpu(out),
reduce(numpy.add, input_vals))
out = MyGpuNdArray.multiplys(*gpu_vals)
assert all_close(to_cpu(out),
reduce(numpy.multiply, input_vals))
def test_elemwise_mixed_dtype():
to_cpu = numpy.asarray
for dtype1 in ["int16", "float32", "int8"]:
for dtype2 in ["int16", "float32", "int8"]:
dtypeo = str((numpy.zeros(1, dtype=dtype1) +
numpy.zeros(1, dtype=dtype2)).dtype)
#print "dtypes", dtype1, dtype2, "o dtype", dtypeo
#print " Test inside a wrapping python object 2 inputs"
for shape in [(500, ), (50, 5), (5, 6, 7)]:
input_vals = [rand(shape, dtype) for dtype in [dtype1, dtype2]]
del dtype
gpu_vals = [
gpu_ndarray.GpuNdArrayObject(i) for i in input_vals
]
assert all([
numpy.allclose(to_cpu(ig), i)
for ig, i in zip(gpu_vals, input_vals)
])
gpu_vals = [MyGpuNdArray(x) for x in gpu_vals]
out = gpu_vals[0] + gpu_vals[1]
assert numpy.allclose(to_cpu(out),
input_vals[0] + input_vals[1])
out = gpu_vals[0] - gpu_vals[1]
assert numpy.allclose(to_cpu(out),
input_vals[0] - input_vals[1])
out = gpu_vals[0] * gpu_vals[1]
assert all_close(to_cpu(out), input_vals[0] * input_vals[1])
if dtypeo.startswith("float"):
# TODO: execute for all dtype
out = gpu_vals[0] / gpu_vals[1]
assert numpy.allclose(to_cpu(out),
input_vals[0] / input_vals[1])
nb_in = 4
#print " Test inside a wrapping python object %d inputs"%nb_in
for shape in [(500, ), (50, 5), (5, 6, 7)]:
input_vals = [
rand(shape, dtype)
for dtype in [dtype1, dtype2, dtype1, dtype2]
]
gpu_vals = [
gpu_ndarray.GpuNdArrayObject(i) for i in input_vals
]
assert all([
numpy.allclose(to_cpu(ig), i)
for ig, i in zip(gpu_vals, input_vals)
])
gpu_vals = [MyGpuNdArray(x) for x in gpu_vals]
out = MyGpuNdArray.adds(*gpu_vals)
assert numpy.allclose(to_cpu(out),
reduce(numpy.add, input_vals))
out = MyGpuNdArray.multiplys(*gpu_vals)
assert all_close(to_cpu(out), reduce(numpy.multiply,
input_vals))
#print " Test broadcasting"
for shapes in [
((1, 5), (4, 5)),
((33, 10), (33, 1)),
((33, 1, 5), (33, 10, 1)),
((33, 1, 5), (33, 10, 1), ((1, 10, 5))),
]:
input_vals = [
rand(shape, dtype)
for shape, dtype in zip(shapes, [dtype1, dtype2])
]
gpu_vals = [
gpu_ndarray.GpuNdArrayObject(i) for i in input_vals
]
assert all([
numpy.allclose(to_cpu(ig), i)
for ig, i in zip(gpu_vals, input_vals)
])
gpu_vals = [MyGpuNdArray(x) for x in gpu_vals]
out = MyGpuNdArray.adds(*gpu_vals)
assert numpy.allclose(to_cpu(out),
reduce(numpy.add, input_vals))
out = MyGpuNdArray.multiplys(*gpu_vals)
assert all_close(to_cpu(out), reduce(numpy.multiply,
input_vals))
