def generate_outputs():
cwrap_files = filter_by_extension(options.files, '.cwrap')
nn_files = filter_by_extension(options.files, 'nn.yaml', '.h')
native_files = filter_by_extension(options.files, 'native_functions.yaml')
declarations = [
d for file in cwrap_files for d in cwrap_parser.parse(file)
]
declarations += nn_parse.run(nn_files)
declarations += native_parse.run(native_files)
declarations = preprocess_declarations.run(declarations)
for fname, env in generators.items():
fm = file_manager
if env['name'] == 'CUDA':
fm = cuda_file_manager
fm.write(fname, GENERATOR_DERIVED, env)
# note: this will fill in top_env['type/tensor_method_declarations/definitions']
# and modify the declarations to include any information that will all_backends
# be used by function_wrapper.create_derived
output_declarations = function_wrapper.create_generic(
top_env, declarations)
output_declarations = postprocess_output_declarations(output_declarations)
file_manager.write("Declarations.yaml", format_yaml(output_declarations))
# populated by generate_storage_type_and_tensor
all_types = []
for backend, density, scalar_type in iterate_types():
all_types.append(
generate_storage_type_and_tensor(backend, density, scalar_type,
declarations))
file_manager.write('Type.h', TYPE_H, top_env)
file_manager.write('TypeExtendedInterface.h', TYPE_EXTENDED_INTERFACE_H,
top_env)
file_manager.write('TypeDefault.h', TYPE_DEFAULT_H, top_env)
file_manager.write('TypeDefault.cpp', TYPE_DEFAULT_CPP, top_env)
file_manager.write('RegisterCPU.h', REGISTER_CPU_H, top_env)
file_manager.write('RegisterCPU.cpp', REGISTER_CPU_CPP, top_env)
cuda_file_manager.write('RegisterCUDA.h', REGISTER_CUDA_H, top_env)
cuda_file_manager.write('RegisterCUDA.cpp', REGISTER_CUDA_CPP, top_env)
file_manager.write('Tensor.h', TENSOR_H, top_env)
file_manager.write('TensorMethods.h', TENSOR_METHODS_H, top_env)
file_manager.write('Functions.h', FUNCTIONS_H, top_env)
file_manager.write('CPUCopy.cpp', copy_wrapper.create(all_types, 'CPU'))
cuda_file_manager.write('CUDACopy.cpp',
copy_wrapper.create(all_types, 'CUDA'))
file_manager.write('NativeFunctions.h', NATIVE_FUNCTIONS_H, top_env)
file_manager.check_all_files_written()
cuda_file_manager.check_all_files_written()
def generate_outputs():
cwrap_files = filter_by_extension(options.files, '.cwrap')
nn_files = filter_by_extension(options.files, 'nn.yaml', '.h')
native_files = filter_by_extension(options.files, 'native_functions.yaml')
declarations = [d
for file in cwrap_files
for d in cwrap_parser.parse(file)]
declarations += nn_parse.run(nn_files)
declarations += native_parse.run(native_files)
declarations = preprocess_declarations.run(declarations)
for fname, env in generators.items():
fm = file_manager
if env['name'] == 'CUDA':
fm = cuda_file_manager
fm.write(fname, GENERATOR_DERIVED, env)
# note: this will fill in top_env['type/tensor_method_declarations/definitions']
# and modify the declarations to include any information that will all_backends
# be used by function_wrapper.create_derived
output_declarations = function_wrapper.create_generic(top_env, declarations)
output_declarations = postprocess_output_declarations(output_declarations)
file_manager.write("Declarations.yaml", format_yaml(output_declarations))
# populated by generate_storage_type_and_tensor
all_types = []
for backend, density, scalar_type in iterate_types():
all_types.append(generate_storage_type_and_tensor(
backend, density, scalar_type, declarations))
file_manager.write('Type.h', TYPE_H, top_env)
file_manager.write('Type.cpp', TYPE_CPP, top_env)
cuda_file_manager.write('RegisterCUDA.h', REGISTER_CUDA_H, top_env)
cuda_file_manager.write('RegisterCUDA.cpp', REGISTER_CUDA_CPP, top_env)
file_manager.write('Tensor.h', TENSOR_H, top_env)
file_manager.write('TensorMethods.h', TENSOR_METHODS_H, top_env)
file_manager.write('Functions.h', FUNCTIONS_H, top_env)
file_manager.write('CPUCopy.cpp', copy_wrapper.create(all_types, 'CPU'))
cuda_file_manager.write('CUDACopy.cpp', copy_wrapper.create(all_types, 'CUDA'))
file_manager.write('NativeFunctions.h', NATIVE_FUNCTIONS_H, top_env)
file_manager.check_all_files_written()
cuda_file_manager.check_all_files_written()
def generate_outputs():
cwrap_files = [f for f in files if f.endswith('.cwrap')]
nn_files = [f for f in files if f.endswith('nn.yaml') or f.endswith('.h')]
native_files = [
f for f in files
if f.endswith('native_functions.yaml') or f.endswith('cuDNN.yaml')
]
declarations = [
d for file in cwrap_files for d in cwrap_parser.parse(file)
]
declarations += nn_parse.run(nn_files)
declarations += native_parse.run(native_files)
declarations = preprocess_declarations.run(declarations)
for fname, env in generators.items():
file_manager.write(fname, GENERATOR_DERIVED.substitute(env))
# note: this will fill in top_env['type/tensor_method_declarations/definitions']
# and modify the declarations to include any information that will all_backends
# be used by function_wrapper.create_derived
output_declarations = function_wrapper.create_generic(
top_env, declarations)
output_declarations = postprocess_output_declarations(output_declarations)
file_manager.write("Declarations.yaml", format_yaml(output_declarations))
# populated by generate_storage_type_and_tensor
all_types = []
for backend, density, scalar_type in iterate_types():
all_types.append(
generate_storage_type_and_tensor(backend, density, scalar_type,
declarations))
file_manager.write('Type.h', TYPE_H.substitute(top_env))
file_manager.write('Type.cpp', TYPE_CPP.substitute(top_env))
file_manager.write('Tensor.h', TENSOR_H.substitute(top_env))
file_manager.write('TensorMethods.h', TENSOR_METHODS_H.substitute(top_env))
file_manager.write('Functions.h', FUNCTIONS_H.substitute(top_env))
file_manager.write('Copy.cpp', copy_wrapper.create(all_types))
file_manager.write('NativeFunctions.h',
NATIVE_FUNCTIONS_H.substitute(top_env))
file_manager.check_all_files_written()
def generate_outputs():
cwrap_files = [f for f in files if f.endswith('.cwrap')]
nn_files = [f for f in files if f.endswith('nn.yaml') or f.endswith('.h')]
native_files = [f for f in files if f.endswith('native_functions.yaml')]
declarations = [d
for file in cwrap_files
for d in cwrap_parser.parse(file)]
declarations += nn_parse.run(nn_files)
declarations += native_parse.run(native_files)
declarations = preprocess_declarations.run(declarations)
for fname, env in generators.items():
file_manager.write(fname, GENERATOR_DERIVED.substitute(env))
# note: this will fill in top_env['type/tensor_method_declarations/definitions']
# and modify the declarations to include any information that will all_backends
# be used by function_wrapper.create_derived
output_declarations = function_wrapper.create_generic(top_env, declarations)
output_declarations = postprocess_output_declarations(output_declarations)
file_manager.write("Declarations.yaml", format_yaml(output_declarations))
# populated by generate_storage_type_and_tensor
all_types = []
for backend, density, scalar_type in iterate_types():
all_types.append(generate_storage_type_and_tensor(
backend, density, scalar_type, declarations))
file_manager.write('Type.h', TYPE_H.substitute(top_env))
file_manager.write('Type.cpp', TYPE_CPP.substitute(top_env))
file_manager.write('Tensor.h', TENSOR_H.substitute(top_env))
file_manager.write('TensorMethods.h', TENSOR_METHODS_H.substitute(top_env))
file_manager.write('Functions.h', FUNCTIONS_H.substitute(top_env))
file_manager.write('Copy.cpp', copy_wrapper.create(all_types))
file_manager.write('NativeFunctions.h', NATIVE_FUNCTIONS_H.substitute(top_env))
file_manager.check_all_files_written()
# note: this will fill in top_env['type/tensor_method_declarations/definitions']
# and modify the declarations to include any information that will all_backends
# be used by function_wrapper.create_derived
output_declarations = function_wrapper.create_generic(top_env, declarations)
write("Declarations.yaml", format_yaml(output_declarations))
# populated by generate_storage_type_and_tensor
all_types = []
for backend in backends:
for density in densities:
for scalar_type in scalar_types:
if density == 'Sparse' and scalar_type[0] == 'Half':
# THS does not do half type yet.
continue
all_types.append(
generate_storage_type_and_tensor(backend, density, scalar_type,
declarations))
write('Type.h', TYPE_H.substitute(top_env))
write('Type.cpp', TYPE_CPP.substitute(top_env))
write('Tensor.h', TENSOR_H.substitute(top_env))
write('TensorMethods.h', TENSOR_METHODS_H.substitute(top_env))
write('Functions.h', FUNCTIONS_H.substitute(top_env))
write('Dispatch.h', dispatch_macros.create(all_types))
write('Copy.cpp', copy_wrapper.create(all_types))
if options.output_dependencies is not None:
output_dependencies_file.close()
# and modify the declarations to include any information that will all_backends
# be used by function_wrapper.create_derived
output_declarations = function_wrapper.create_generic(top_env, declarations)
output_declarations = postprocess_output_declarations(output_declarations)
write("Declarations.yaml", format_yaml(output_declarations))
# populated by generate_storage_type_and_tensor
all_types = []
for backend in backends:
for density in densities:
for scalar_type in scalar_types:
if density == 'Sparse' and scalar_type[0] == 'Half':
# THS does not do half type yet.
continue
all_types.append(generate_storage_type_and_tensor(
backend, density, scalar_type, declarations))
write('Type.h', TYPE_H.substitute(top_env))
write('Type.cpp', TYPE_CPP.substitute(top_env))
write('Tensor.h', TENSOR_H.substitute(top_env))
write('TensorMethods.h', TENSOR_METHODS_H.substitute(top_env))
write('Functions.h', FUNCTIONS_H.substitute(top_env))
write('Dispatch.h', dispatch_macros.create(all_types))
write('Copy.cpp', copy_wrapper.create(all_types))
write('NativeFunctions.h', NATIVE_FUNCTIONS_H.substitute(top_env))
if options.output_dependencies is not None:
output_dependencies_file.close()
def generate_outputs():
cwrap_files = filter_by_extension(options.files, '.cwrap')
nn_files = filter_by_extension(options.files, 'nn.yaml', '.h')
native_files = filter_by_extension(options.files, 'native_functions.yaml')
declarations = [
d for file in cwrap_files for d in cwrap_parser.parse(file)
]
declarations += nn_parse.run(nn_files)
declarations += native_parse.run(native_files)
declarations = preprocess_declarations.run(declarations)
for fname, env in generators.items():
fm = file_manager
if env['name'] == 'CUDA':
fm = cuda_file_manager
fm.write(fname, GENERATOR_DERIVED, env)
# note: this will fill in top_env['type/tensor_method_declarations/definitions']
# and modify the declarations to include any information that will all_backends
# be used by function_wrapper.create_derived
output_declarations = function_wrapper.create_generic(
top_env, declarations)
output_declarations = postprocess_output_declarations(output_declarations)
file_manager.write("Declarations.yaml", format_yaml(output_declarations))
# populated by generate_storage_type_and_tensor
all_types = []
for backend, density, scalar_type in iterate_types():
all_types.append(
generate_storage_type_and_tensor(backend, density, scalar_type,
declarations))
core_files = {
'Type.h': TYPE_H,
'Tensor.h': TENSOR_H,
'TensorMethods.h': TENSOR_METHODS_H
}
for core_file, core_template_file in core_files.items():
core_file_manager.write(core_file, core_template_file, top_env)
file_manager.write('TypeExtendedInterface.h', TYPE_EXTENDED_INTERFACE_H,
top_env)
file_manager.write('TypeDefault.h', TYPE_DEFAULT_H, top_env)
file_manager.write('TypeDefault.cpp', TYPE_DEFAULT_CPP, top_env)
file_manager.write('RegisterCPU.h', REGISTER_CPU_H, top_env)
file_manager.write('RegisterCPU.cpp', REGISTER_CPU_CPP, top_env)
cuda_file_manager.write('RegisterCUDA.h', REGISTER_CUDA_H, top_env)
cuda_file_manager.write('RegisterCUDA.cpp', REGISTER_CUDA_CPP, top_env)
file_manager.write('Functions.h', FUNCTIONS_H, top_env)
file_manager.write('CPUCopy.cpp', copy_wrapper.create(all_types, 'CPU'))
cuda_file_manager.write('CUDACopy.cpp',
copy_wrapper.create(all_types, 'CUDA'))
file_manager.write('NativeFunctions.h', NATIVE_FUNCTIONS_H, top_env)
file_manager.check_all_files_written()
cuda_file_manager.check_all_files_written()
# check that generated files match source files
core_source_path = os.path.join(options.source_path, 'core')
match, mismatch, errors = filecmp.cmpfiles(core_install_dir,
core_source_path,
core_files.keys(),
shallow=False)
if errors:
raise RuntimeError(
"Error while trying to compare source and generated files for {}. "
"Source directory: {}. Generated directory: {}.".format(
errors, core_source_path, core_install_dir))
if mismatch:
file_component = '{}'.format(','.join(mismatch))
if len(mismatch) > 1:
file_component = '{' + file_component + '}'
update_cmd = "cp {}/{} {}".format(core_install_dir, file_component,
core_source_path)
raise RuntimeError(
"Source files: {} did not match generated files. To update the source files, "
"run \"{}\"".format(mismatch, update_cmd))
