You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

graph_kernel_flags.h 7.6 kB

4 years ago
4 years ago
4 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220
  1. /**
  2. * Copyright 2021 Huawei Technologies Co., Ltd
  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. #ifndef MINDSPORE_CCSRC_UTILS_GRAPH_KERNEL_FLAGS_H
  17. #define MINDSPORE_CCSRC_UTILS_GRAPH_KERNEL_FLAGS_H
  18. #include <map>
  19. #include <memory>
  20. #include <string>
  21. #include <vector>
  22. #include <utility>
  23. #include "utils/ms_context.h"
  24. namespace mindspore {
  25. namespace context {
  26. constexpr unsigned int OptLevel_0 = 0; // Disabled
  27. constexpr unsigned int OptLevel_1 = 1; // Basic functions
  28. constexpr unsigned int OptLevel_2 = 2; // Default functions
  29. constexpr unsigned int OptLevel_3 = 3; // Experimental functions
  30. constexpr unsigned int OptLevel_MAX = 4;
  31. constexpr unsigned int OpLevel_0 = 0;
  32. constexpr unsigned int OpLevel_1 = 1;
  33. constexpr unsigned int OpLevel_MAX = 2;
  34. class GraphKernelFlags {
  35. public:
  36. static const GraphKernelFlags &GetInstance() {
  37. static std::unique_ptr<GraphKernelFlags> flags(nullptr);
  38. auto contexts = GetGraphKernelContext();
  39. if (flags == nullptr || contexts.first != flags->flags_cache_ || contexts.second != flags->enable_graph_kernel_) {
  40. flags.reset(new GraphKernelFlags(contexts.first, contexts.second));
  41. flags->Refresh();
  42. }
  43. return *flags;
  44. }
  45. // Dump all flags to json-format string
  46. std::string DumpAllFlags() const;
  47. // Check whether graph_kernel is enabled
  48. bool IsEnableGraphKernel() const { return opt_level > OptLevel_0; }
  49. GraphKernelFlags(const GraphKernelFlags &flags) = delete;
  50. ~GraphKernelFlags() = default;
  51. public:
  52. /**
  53. * Dump info as human-readable text.
  54. * A directory "graph_kernel_dump" will be created, and all information will be dumped in this directory.
  55. */
  56. bool dump_as_text{false};
  57. /**
  58. * Enable stitch fusion in graph kernel fusion strategy.
  59. *
  60. * Experimental feature, enabled by default when opt_level=3
  61. */
  62. bool enable_stitch_fusion{false};
  63. /**
  64. * Enable recompute fusion in graph kernel fusion strategy, enabled when op_level>=2.
  65. */
  66. bool enable_recompute_fusion{false};
  67. /**
  68. * Enable parallel fusion in graph kernel fusion strategy.
  69. *
  70. * Experimental feature, enabled by default when opt_level=3
  71. */
  72. bool enable_parallel_fusion{false};
  73. /**
  74. * Enable low precision in data transferring between graph kernel and computing in graph kernel
  75. * in graph kernel.
  76. * Experimental feature, enabled by the enable_low_precision flag
  77. */
  78. bool enable_low_precision{false};
  79. /**
  80. * Expand and cluster AKG's operators by level.
  81. */
  82. unsigned int fusion_ops_level{OpLevel_0};
  83. /**
  84. * Enable optimization for transform operators (Transpose/TransData)
  85. *
  86. * Experimental feature, enabled by default when opt_level=3.
  87. */
  88. bool enable_trans_op_optimize{false};
  89. /**
  90. * Optimization level, value from 0 to 3.
  91. * 0: Disable GraphKernel
  92. * 1: Enable GraphKernel with basic features only.
  93. * 2: Enable GraphKernel with all stable features.
  94. * 3: Enable GraphKernel with all experimental features.
  95. * The default value is OptLevel_2 when the context "enable_graph_kernel" is set,
  96. * but if it's also changed in "graph_kernel_flags", then the "graph_kernel_flags" will prevail.
  97. */
  98. unsigned int opt_level{0}; // defaults 0 or 2
  99. /**
  100. * Online tuning level, value from 0 to 3.
  101. * 0: Disable online tuning
  102. * 1-3: The higher level, the larger tuning space, and the more time it takes.
  103. */
  104. unsigned int online_tuning{0};
  105. /**
  106. * AKG's operator repository file path.
  107. */
  108. std::string repository_path;
  109. /**
  110. * Additional expanding operators (case sensitive).
  111. * The operators to be added into the default expanding operator list.
  112. */
  113. std::vector<std::string> enable_expand_ops;
  114. /**
  115. * Expanding operators to be enabled (case sensitive).
  116. * Unlike the "enable_expand_ops", the default list will be overwritten by this list.
  117. * Note that the "enable_expand_ops" and "disable_expand_ops" will be ignored if this flag is set.
  118. */
  119. std::vector<std::string> enable_expand_ops_only;
  120. /**
  121. * Expanding operators to be disabled (case sensitive).
  122. * The behavior is undefined when this list overlaps with "enable_expand_ops".
  123. */
  124. std::vector<std::string> disable_expand_ops;
  125. /**
  126. * Additional clustering operators (case sensitive).
  127. * The operators to be added into the default clustering operator list.
  128. */
  129. std::vector<std::string> enable_cluster_ops;
  130. /**
  131. * Clustering operators to be enabled (case sensitive).
  132. * Unlike the "enable_cluster_ops", the default list will be overwritten by this list.
  133. * Note that the "enable_cluster_ops" and "disable_cluster_ops" will be ignored if this flag is set.
  134. */
  135. std::vector<std::string> enable_cluster_ops_only;
  136. /**
  137. * Clustering operators to be disabled (case sensitive).
  138. * The behavior is undefined when this list overlaps with "enable_cluster_ops".
  139. */
  140. std::vector<std::string> disable_cluster_ops;
  141. /**
  142. * Arithmetic simplify expressions to be enabled (case sensitive).
  143. * The default list will be overwritten by this list.
  144. * Note that "disable_simplify_exprs" will be ignored if this flag is set.
  145. */
  146. std::vector<std::string> enable_simplify_exprs_only;
  147. /**
  148. * Arithmetic simplify expressions to be disabled (case sensitive).
  149. */
  150. std::vector<std::string> disable_simplify_exprs;
  151. /**
  152. * Passes to be enabled.
  153. * By default, the passes is controlled by "opt_level" and target device,
  154. * user can manually enable some passes by setting this flag.
  155. * The format is "stage_id.pass_id" or "stage_name.pass_name", which corresponds to the ir filename.
  156. */
  157. std::vector<std::string> enable_pass;
  158. /**
  159. * Passes to be disabled.
  160. * By default, the passes is controlled by "opt_level" and target device,
  161. * user can manually disable some passes by setting this flag.
  162. * The format is "stage_id.pass_id" or "stage_name.pass_name", which corresponds to the ir filename.
  163. */
  164. std::vector<std::string> disable_pass;
  165. private:
  166. GraphKernelFlags(const std::string &graph_kernel_flags, bool enable_graph_kernel)
  167. : flags_cache_(graph_kernel_flags), enable_graph_kernel_(enable_graph_kernel) {}
  168. // get the `graph_kernel_flags` and `enable_graph_kernel`
  169. static std::pair<std::string, bool> GetGraphKernelContext() {
  170. auto context = MsContext::GetInstance();
  171. MS_EXCEPTION_IF_NULL(context);
  172. // Use the environment variable in priority
  173. auto env_flags = std::getenv("MS_GRAPH_KERNEL_FLAGS");
  174. std::string flags = env_flags ? std::string(env_flags) : context->get_param<std::string>(MS_CTX_GRAPH_KERNEL_FLAGS);
  175. return std::make_pair(flags, context->get_param<bool>(MS_CTX_ENABLE_GRAPH_KERNEL));
  176. }
  177. // parse and refresh the flags
  178. void Refresh();
  179. // register the flags defined above
  180. void RegisterFlags(std::map<std::string, std::string> *flag_map);
  181. // cache the flag string to check whether the flags is changed.
  182. std::string flags_cache_;
  183. // cache the enable_graph_kernel value to check whether the context is changed.
  184. bool enable_graph_kernel_;
  185. };
  186. } // namespace context
  187. } // namespace mindspore
  188. #endif // MINDSPORE_CCSRC_UTILS_GRAPH_KERNEL_FLAGS_H