def get_test_lines(in_shape, permutation):
"""
Create a list of strings corresponding to the lines in a single test case.
Uses TensorFlow to compute the expected results for the given parameters,
and provides the code to call the test fixture to run the test.
"""
output, max_input_val = helpers.get_result_and_size(
get_transpose_result, in_shape=in_shape, permutation=permutation)
in_shape_str = list(map(str, in_shape))
perm_str = list(map(str, permutation))
test_case = TEST_CASE_TPL.format(n_dimensions=len(in_shape))
test_name = TEST_NAME_TPL.format(n_dimensions=len(in_shape)) + "x".join(
in_shape_str) + "_" + "x".join(perm_str)
test_lines = [
"TYPED_TEST({}, {}) {{".format(test_case, test_name),
" using DataType = typename TestFixture::DataType;",
" const std::vector<DataType> exp_out = {};".format(
helpers.format_tensor(output)),
" const std::vector<int> sizes = {{ {size_str} }};".format(
size_str=",".join(in_shape_str)),
" const std::vector<int> perm = {{ {perm_str} }};".format(
perm_str=",".join(perm_str)),
" const DataType max_input_val = {:.1f};".format(max_input_val),
" this->run(exp_out, sizes, perm, max_input_val, 0, 0);",
"}",
]
return test_lines
def get_test_lines(batch, m, k, n, beta, trans_lhs, trans_rhs):
"""
Create a list of strings corresponding to the lines in a single test case.
Uses TensorFlow to compute the expected results for the given parameters,
and provides the code to call the test fixture to run the test.
"""
output, max_input_val = helpers.get_result_and_size(get_matmul_result,
batch=batch,
m=m,
k=k,
n=n,
beta=beta,
trans_lhs=trans_lhs,
trans_rhs=trans_rhs)
test_case = TEST_CASE_TPL.format(batch=batch,
beta=beta,
trans_lhs=trans_lhs,
trans_rhs=trans_rhs)
test_name = TEST_NAME_TPL.format(m=m, k=k, n=n)
test_lines = [
"TYPED_TEST({}, {}) {{".format(test_case, test_name),
" using DataType = typename TestFixture::DataType;",
" const std::vector<DataType> exp_out = {};".format(
helpers.format_tensor(output)),
" const int batches = {};".format(batch),
" const int m = {};".format(m),
" const int k = {};".format(k),
" const int n = {};".format(n),
" const auto beta = static_cast<DataType>({});".format(beta),
" const DataType max_input_val = {:.1f};".format(max_input_val),
" this->run(exp_out, batches, m, k, n, beta, 0, 0, 0, max_input_val);",
"}",
]
return test_lines
def get_result(test_case, test_params):
output, _ = helpers.get_result_and_size(
get_result_function(test_case),
max_input_val=test_params.in_size,
floor_div=True,
pointwise_op=TENSORFLOW_OPS_MAP[test_case.test_type],
in_size=test_params.in_size)
return output
def get_result(test_case, test_params):
channel_idx = -1 if test_params.data_format == 'NHWC' else 1
output, max_input_val = helpers.get_result_and_size(
get_result_function(test_case),
max_input_val=test_params.in_shape[channel_idx],
floor_div=True,
softmax_op=TENSORFLOW_OPS_MAP[test_case.test_type],
input_shape=test_params.in_shape)
return output, max_input_val
def get_result_and_size(test_case, test_params):
"""
Get the result of the bias-add and max input value.
Ensures that the resulting values are less than the REQUIRED_MAX, and if
not will adjust the maximum value to allow in the input tensors.
"""
return helpers.get_result_and_size(get_result_function(test_case),
input_shape=test_params.in_shape)
def get_result_and_size(test_case, test_params):
"""
Get the result of the specified convolution and max input value.
Ensures that the resulting values are less than the REQUIRED_MAX, and if
not will adjust the maximum value to allow in the input tensors.
"""
window_shape = [test_case.window, test_case.window]
stride_shape = [test_case.stride, test_case.stride]
return helpers.get_result_and_size(get_result_function(test_case),
pool_op=test_case.test_type,
input_shape=test_params.in_shape,
window_shape=window_shape,
stride_shape=stride_shape,
padding=test_params.padding)
def get_result_and_size(test_case, test_params):
"""
Get the result of the specified convolution and max input value.
Ensures that the resulting values are less than the REQUIRED_MAX, and if
not will adjust the maximum value to allow in the input tensors.
"""
conv_fn = get_conv_fn(test_case.test_type)
filter_shape = [
test_case.window, test_case.window, test_params.in_shape[-1],
test_params.features
]
stride_shape = [1, test_case.stride, test_case.stride, 1]
return helpers.get_result_and_size(conv_fn,
input_shape=test_params.in_shape,
filter_shape=filter_shape,
stride_shape=stride_shape,
padding=test_params.padding)
