1 | /******************************************************************************* |
2 | * Copyright 2019-2022 Intel Corporation |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. |
6 | * You may obtain a copy of the License at |
7 | * |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * |
10 | * Unless required by applicable law or agreed to in writing, software |
11 | * distributed under the License is distributed on an "AS IS" BASIS, |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
13 | * See the License for the specific language governing permissions and |
14 | * limitations under the License. |
15 | *******************************************************************************/ |
16 | |
17 | #include <stdio.h> |
18 | #include <stdlib.h> |
19 | |
20 | #include <sstream> |
21 | |
22 | #include "dnnl_common.hpp" |
23 | #include "utils/parser.hpp" |
24 | |
25 | #include "sum/sum.hpp" |
26 | |
27 | namespace sum { |
28 | |
29 | void check_correctness(const settings_t &s) { |
30 | for_(const auto &i_sdt : s.sdt) |
31 | for_(const auto &i_ddt : s.ddt) |
32 | for_(const auto &i_stag : s.stag) |
33 | for_(const auto &i_dtag : s.dtag) |
34 | for_(const auto &i_input_scales : s.input_scales) |
35 | for_(const auto &i_scratchpad_mode : s.scratchpad_mode) |
36 | for_(const auto &i_ctx_init : s.ctx_init) |
37 | for_(const auto &i_ctx_exe : s.ctx_exe) |
38 | for (auto i_inplace : s.inplace) { |
39 | const int n_inputs = static_cast<int>(i_sdt.size()); |
40 | const int n_stags = static_cast<int>(i_stag.size()); |
41 | if (n_stags != n_inputs && n_stags != 1) { |
42 | BENCHDNN_PRINT(0, |
43 | "ERROR: Expected number of stag arguments is `1` or `%d`, " |
44 | "provided `%d`.\n" , |
45 | n_inputs, n_stags); |
46 | SAFE_V(FAIL); |
47 | } |
48 | |
49 | const int n_input_scales = static_cast<int>(i_input_scales.size()); |
50 | if (n_input_scales != n_inputs && n_input_scales != 1) { |
51 | BENCHDNN_PRINT(0, |
52 | "ERROR: Expected number of scales arguments is `1` or " |
53 | "`%d`, provided `%d`.\n" , |
54 | n_inputs, n_input_scales); |
55 | SAFE_V(FAIL); |
56 | } |
57 | |
58 | auto attr = settings_t::get_attr(i_scratchpad_mode); |
59 | |
60 | const prb_t prb(s.prb_dims, i_sdt, i_ddt, i_stag, i_dtag, |
61 | i_input_scales, i_inplace, attr, i_ctx_init, i_ctx_exe); |
62 | std::stringstream ss; |
63 | ss << prb; |
64 | const std::string cpp_pstr = ss.str(); |
65 | const char *pstr = cpp_pstr.c_str(); |
66 | BENCHDNN_PRINT(1, "run: %s\n" , pstr); |
67 | |
68 | res_t res {}; |
69 | doit(&prb, &res); |
70 | |
71 | parse_result(res, pstr); |
72 | |
73 | if (is_bench_mode(PERF)) { |
74 | perf_report_t pr(&prb, s.perf_template); |
75 | pr.report(&res, pstr); |
76 | } |
77 | } |
78 | } |
79 | |
80 | static const std::string help_scales |
81 | = "FLOAT[:FLOAT...] (Default: `1.f`)\n Input scales for source " |
82 | "values.\n If a single value is specified, will be broadcasted " |
83 | "for all sources, otherwise number of scales should match number of " |
84 | "inputs.\n" ; |
85 | |
86 | int bench(int argc, char **argv) { |
87 | driver_name = "sum" ; |
88 | using namespace parser; |
89 | static settings_t s; |
90 | static const settings_t def {}; |
91 | for (; argc > 0; --argc, ++argv) { |
92 | const bool parsed_options = parse_bench_settings(argv[0]) |
93 | || parse_batch(bench, argv[0]) |
94 | || parse_multi_dt(s.sdt, def.sdt, argv[0]) |
95 | || parse_dt(s.ddt, def.ddt, argv[0], "ddt" ) |
96 | || parse_multi_tag(s.stag, def.stag, argv[0]) |
97 | || parse_tag(s.dtag, def.dtag, argv[0], "dtag" ) |
98 | || parse_multivector_option(s.input_scales, def.input_scales, |
99 | atof, argv[0], "scales" , help_scales) |
100 | || parse_inplace(s.inplace, def.inplace, argv[0]) |
101 | || parse_attr_scratchpad_mode( |
102 | s.scratchpad_mode, def.scratchpad_mode, argv[0]) |
103 | || parse_ctx_init(s.ctx_init, def.ctx_init, argv[0]) |
104 | || parse_ctx_exe(s.ctx_exe, def.ctx_exe, argv[0]) |
105 | || parse_perf_template(s.perf_template, s.perf_template_def, |
106 | s.perf_template_csv(), argv[0]) |
107 | || parse_reset(s, argv[0]) || parse_help(argv[0]); |
108 | if (!parsed_options) { |
109 | catch_unknown_options(argv[0]); |
110 | |
111 | parse_prb_dims(s.prb_dims, argv[0]); |
112 | check_correctness(s); |
113 | } |
114 | } |
115 | |
116 | return parse_last_argument(); |
117 | } |
118 | |
119 | } // namespace sum |
120 | |