def test_reduction_axis(inspecs, reduction, axis, nnabla_opts):
func = getattr(F, reduction)
fb = FunctionBenchmark(
func, inspecs, [], dict(axis=axis),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_binary_classification_loss(inspecs, loss, nnabla_opts):
func = getattr(F, loss)
fb = FunctionBenchmark(
func, inspecs, [], {},
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_pairwise_logical(inspecs, op, nnabla_opts):
func = getattr(F, op)
fb = FunctionBenchmark(
func, inspecs, [], {},
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_embed(inspecs, n_inputs, n_features, nnabla_opts):
fb = FunctionBenchmark(
PF.embed, inspecs, [],
dict(n_inputs=n_inputs, n_features=n_features),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_reduction_axis(inspecs, reduction, axis, nnabla_opts):
func = getattr(F, reduction)
fb = FunctionBenchmark(
func, inspecs, [], dict(axis=axis),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_pairwise_arithmetic(inspecs, op, nnabla_opts):
func = getattr(F, op)
fb = FunctionBenchmark(
func, inspecs, [], {},
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_activation(inspecs, activation, nnabla_opts):
func = getattr(F, activation)
fb = FunctionBenchmark(
func, inspecs, [], {},
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_embed(inspecs, n_inputs, n_features, nnabla_opts):
fb = FunctionBenchmark(
PF.embed, inspecs, [],
dict(n_inputs=n_inputs, n_features=n_features),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_instance_normalization(inspec_and_axis, nnabla_opts):
inspec, axis = inspec_and_axis
fb = FunctionBenchmark(PF.instance_normalization, inspec, [],
dict(channel_axis=axis), nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_logical_not(inspecs, nnabla_opts):
func = F.logical_not
fb = FunctionBenchmark(
func, inspecs, [], {},
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_categorical_classification_loss(inspecs, loss, nnabla_opts):
func = getattr(F, loss)
fb = FunctionBenchmark(
func, inspecs, [], dict(axis=1),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_pooling(inspecs, pool, nnabla_opts):
if pool == 'average':
func = F.average_pooling
elif pool == 'max':
func = F.max_pooling
fb = FunctionBenchmark(
func, inspecs, [], dict(kernel=(2, 2), stride=(2, 2)),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_pooling(inspecs, pool, nnabla_opts):
if pool == 'average':
func = F.average_pooling
elif pool == 'max':
func = F.max_pooling
fb = FunctionBenchmark(
func, inspecs, [], dict(kernel=(2, 2), stride=(2, 2)),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_cumprod(seed, test_case, exclusive, reverse, with_mask, nnabla_opts):
x_shape = test_case.shape
axis = test_case.axis
def init(shape):
rng = np.random.RandomState(seed)
return create_cumprod_input(rng, shape, axis, with_mask)
need_grad = True
inputs = [Inspec(x_shape, init, need_grad)]
func_kwargs = dict(
axis=axis,
exclusive=exclusive,
reverse=reverse,
)
fb = FunctionBenchmark(F.cumprod, inputs, [], func_kwargs, nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_softmax(inspecs, axis, nnabla_opts):
fb = FunctionBenchmark(F.softmax, inspecs, [], dict(axis=axis),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_convolution(inputs, func_kwargs, nnabla_opts):
fb = FunctionBenchmark(
PF.convolution, inputs, [], func_kwargs,
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_logical_not(inspecs, nnabla_opts):
func = F.logical_not
fb = FunctionBenchmark(func, inspecs, [], {}, nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_scalar_logical(inspecs, op, nnabla_opts):
func = getattr(F, op)
fb = FunctionBenchmark(func, inspecs, [1], {}, nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_bn(inspecs, batch_stat, nnabla_opts):
fb = FunctionBenchmark(PF.batch_normalization, inspecs, [],
dict(batch_stat=batch_stat), nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_convolution(inputs, func_kwargs, nnabla_opts):
fb = FunctionBenchmark(
PF.convolution, inputs, [], func_kwargs,
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_categorical_classification_loss(inspecs, loss, nnabla_opts):
func = getattr(F, loss)
fb = FunctionBenchmark(func, inspecs, [], dict(axis=1), nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_affine(inspecs, n_outmaps, nnabla_opts):
fb = FunctionBenchmark(PF.affine, inspecs, [], dict(n_outmaps=n_outmaps),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_bn(inspecs, batch_stat, nnabla_opts):
fb = FunctionBenchmark(
PF.batch_normalization, inspecs, [], dict(batch_stat=batch_stat),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_layer_normalization(inspec_and_axis, nnabla_opts):
inspec, axis = inspec_and_axis
fb = FunctionBenchmark(PF.layer_normalization, inspec, [], dict(),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_activation(inspecs, shared, nnabla_opts):
fb = FunctionBenchmark(
PF.prelu, inspecs, [1], {},
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_activation(inspecs, activation, nnabla_opts):
func = getattr(F, activation)
fb = FunctionBenchmark(func, inspecs, [], {}, nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_binary_classification_loss(inspecs, loss, nnabla_opts):
func = getattr(F, loss)
fb = FunctionBenchmark(func, inspecs, [], {}, nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_pad(inspecs, nnabla_opts):
fb = FunctionBenchmark(F.pad, inspecs, [(10, 10, 10, 10), 'constant', 0.0],
{}, nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_activation(inspecs, shared, nnabla_opts):
fb = FunctionBenchmark(PF.prelu, inspecs, [1], {}, nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_mul2_with_broadcast(inspecs, op, nnabla_opts):
func = getattr(F, op)
fb = FunctionBenchmark(func, inspecs, [], {}, nnabla_opts.ext,
nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
def test_transpose(inspecs, axis, nnabla_opts):
fb = FunctionBenchmark(F.transpose, inspecs, [axis], dict(),
nnabla_opts.ext, nnabla_opts.ext_kwargs)
fb.benchmark()
fb.write(writer=nnabla_opts.function_benchmark_writer)
