(setup, tested_function, param_list, n_runs, warmup_runs)
| 18 | |
| 19 | |
| 20 | def exec_bench(setup, tested_function, param_list, n_runs, warmup_runs): |
| 21 | backend_list = get_backend_list() |
| 22 | for i, nx in enumerate(backend_list): |
| 23 | if nx.__name__ == "tf" and i < len(backend_list) - 1: |
| 24 | # Tensorflow should be the last one to be benchmarked because |
| 25 | # as far as I'm aware, there is no way to force it to release |
| 26 | # GPU memory. Hence, if any other backend is benchmarked after |
| 27 | # Tensorflow and requires the usage of a GPU, it will not have the |
| 28 | # full memory available and you may have a GPU Out Of Memory error |
| 29 | # even though your GPU can technically hold your tensors in memory. |
| 30 | backend_list.pop(i) |
| 31 | backend_list.append(nx) |
| 32 | break |
| 33 | |
| 34 | inputs = [setup(param) for param in param_list] |
| 35 | results = dict() |
| 36 | for nx in backend_list: |
| 37 | for i in range(len(param_list)): |
| 38 | print(nx, param_list[i]) |
| 39 | args = inputs[i] |
| 40 | results_nx = nx._bench( |
| 41 | tested_function, *args, n_runs=n_runs, warmup_runs=warmup_runs |
| 42 | ) |
| 43 | gc.collect() |
| 44 | results_nx_with_param_in_key = dict() |
| 45 | for key in results_nx: |
| 46 | new_key = (param_list[i], *key) |
| 47 | results_nx_with_param_in_key[new_key] = results_nx[key] |
| 48 | results.update(results_nx_with_param_in_key) |
| 49 | return results |
| 50 | |
| 51 | |
| 52 | def convert_to_html_table(results, param_name, main_title=None, comments=None): |
no test coverage detected