1 | /* Copyright 2015 The TensorFlow Authors. All Rights Reserved. |
2 | |
3 | Licensed under the Apache License, Version 2.0 (the "License"); |
4 | you may not use this file except in compliance with the License. |
5 | You may obtain a copy of the License at |
6 | |
7 | http://www.apache.org/licenses/LICENSE-2.0 |
8 | |
9 | Unless required by applicable law or agreed to in writing, software |
10 | distributed under the License is distributed on an "AS IS" BASIS, |
11 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
12 | See the License for the specific language governing permissions and |
13 | limitations under the License. |
14 | ==============================================================================*/ |
15 | |
16 | #ifndef TENSORFLOW_CORE_FRAMEWORK_RUN_HANDLER_UTIL_H_ |
17 | #define TENSORFLOW_CORE_FRAMEWORK_RUN_HANDLER_UTIL_H_ |
18 | |
19 | #include <cstdint> |
20 | #include <string> |
21 | #include <vector> |
22 | |
23 | namespace tensorflow { |
24 | |
25 | // Assign thread ranges to requests. |
26 | // Requests are numbered 0...num_active_requests-1, and |
27 | // threads are numbered 0...num_threads-1. |
28 | // On return, the range [start_vec->at(i), end_vec->at(i)) |
29 | // indicates the subrange of the threads available to request i. |
30 | // The ranges given to different requests may overlap. |
31 | // Lower numbered requests will tend to be assigned more threads. |
32 | // Thus, a client might associate older requests with lower |
33 | // array indices so they receive access to more threads. |
34 | // However, the routine ensures that each request is given access |
35 | // to at least min(min_threads_per_request, num_threads) threads. |
36 | // Every thread will be assigned to at least one request range, |
37 | // assuming there is at least one request. |
38 | void ComputeInterOpSchedulingRanges(int num_active_requests, int num_threads, |
39 | int min_threads_per_request, |
40 | std::vector<std::uint_fast32_t>* start_vec, |
41 | std::vector<std::uint_fast32_t>* end_vec); |
42 | |
43 | // Assign thread steal ranges to threads.Threads are numbered 0...num_threads-1. |
44 | // On return, the range [start_vec->at(i), end_vec->at(i)) indicates the steal |
45 | // range of the thread i. The ranges given to different threads may overlap. |
46 | void ComputeInterOpStealingRanges(int num_threads, int min_threads_per_domain, |
47 | std::vector<std::uint_fast32_t>* start_vec, |
48 | std::vector<std::uint_fast32_t>* end_vec); |
49 | |
50 | // For each of the num_threads determine the index of the active_request whose |
51 | // work queue should be attempted first by that the thread. Return a vector of |
52 | // size num_threads which represents how threads should be distributed across |
53 | // requests. |
54 | std::vector<int> ChooseRequestsWithExponentialDistribution( |
55 | int num_active_requests, int num_threads); |
56 | |
57 | // Look up environment variable named 'var_name' and return the value if it |
58 | // exist and can be parsed. Return 'default_value' otherwise. |
59 | double ParamFromEnvWithDefault(const char* var_name, double default_value); |
60 | |
61 | // Look up environment variable named 'var_name' and return the value if it |
62 | // exist and can be parsed. The value must be in format val1,val2... Return |
63 | // 'default_value' otherwise. |
64 | std::vector<double> ParamFromEnvWithDefault(const char* var_name, |
65 | std::vector<double> default_value); |
66 | |
67 | // Look up environment variable named 'var_name' and return the value if it |
68 | // exist and can be parsed. The value must be in format val1,val2... Return |
69 | // 'default_value' otherwise. |
70 | std::vector<int> ParamFromEnvWithDefault(const char* var_name, |
71 | std::vector<int> default_value); |
72 | |
73 | // Look up environment variable named 'var_name' and return the value if it |
74 | // exist and can be parsed. Return 'default_value' otherwise. |
75 | bool ParamFromEnvBoolWithDefault(const char* var_name, bool default_value); |
76 | |
77 | } // end namespace tensorflow |
78 | #endif // TENSORFLOW_CORE_FRAMEWORK_RUN_HANDLER_UTIL_H_ |
79 | |