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.

context.h 3.2 kB

5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112
  1. /**
  2. * Copyright 2019-2020 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 __CCE_RUNTIME_CONTEXT_H__
  17. #define __CCE_RUNTIME_CONTEXT_H__
  18. #include "base.h"
  19. #ifdef __cplusplus
  20. extern "C" {
  21. #endif // __cplusplus
  22. /**
  23. * @ingroup rt_context
  24. * @brief runtime context handle.
  25. */
  26. typedef void *rtContext_t;
  27. typedef enum tagDryRunFlag { RT_DRYRUN_FLAG_FALSE = 0, RT_DRYRUN_FLAG_TRUE = 1 } rtDryRunFlag_t;
  28. typedef enum tagCtxMode {
  29. RT_CTX_NORMAL_MODE = 0,
  30. RT_CTX_GEN_MODE = 1,
  31. } rtCtxMode_t;
  32. /**
  33. * @ingroup rt_context
  34. * @brief create context and associates it with the calling thread
  35. * @param [out] ctx created context
  36. * @param [in] flags context creation flag. set to 0.
  37. * @param [in] device device to create context on
  38. * @return RT_ERROR_NONE for ok
  39. */
  40. RTS_API rtError_t rtCtxCreate(rtContext_t *ctx, uint32_t flags, int32_t device);
  41. /**
  42. * @ingroup rt_context
  43. * @brief create context and associates it with the calling thread
  44. * @param [out] ctx created context
  45. * @param [in] flags context creation flag. set to 0.
  46. * @param [in] device device to create context on
  47. * @return RT_ERROR_NONE for ok
  48. */
  49. RTS_API rtError_t rtCtxCreateEx(rtContext_t *ctx, uint32_t flags, int32_t device);
  50. /**
  51. * @ingroup rt_context
  52. * @brief destroy context instance
  53. * @param [in] ctx context to destroy
  54. * @return RT_ERROR_NONE for ok
  55. */
  56. RTS_API rtError_t rtCtxDestroy(rtContext_t ctx);
  57. /**
  58. * @ingroup rt_context
  59. * @brief binds context to the calling CPU thread.
  60. * @param [in] ctx context to bind. if NULL, unbind current context.
  61. * @return RT_ERROR_NONE for ok
  62. */
  63. RTS_API rtError_t rtCtxSetCurrent(rtContext_t ctx);
  64. /**
  65. * @ingroup rt_context
  66. * @brief block for a context's tasks to complete
  67. * @return RT_ERROR_NONE for ok
  68. */
  69. RTS_API rtError_t rtCtxSynchronize(void);
  70. /**
  71. * @ingroup rt_context
  72. * @brief returns the context bound to the calling CPU thread.
  73. * @param [out] ctx returned context
  74. * @return RT_ERROR_NONE for ok
  75. */
  76. RTS_API rtError_t rtCtxGetCurrent(rtContext_t *ctx);
  77. /**
  78. * @ingroup rt_context
  79. * @brief returns the device ID for the current context
  80. * @param [out] device returned device id
  81. * @return RT_ERROR_NONE for ok
  82. */
  83. RTS_API rtError_t rtCtxGetDevice(int32_t *device);
  84. /**
  85. * @ingroup rt_context
  86. * @brief set ctx run mode: normal or dryrun
  87. * @param [in] ctx: context
  88. * @param [in] enable: set true means enable dryrun mode
  89. * @param [in] flag: reserved
  90. * @return RT_ERROR_NONE for ok
  91. */
  92. RTS_API rtError_t rtCtxSetDryRun(rtContext_t ctx, rtDryRunFlag_t enable, uint32_t flag);
  93. #ifdef __cplusplus
  94. }
  95. #endif
  96. #endif // __CCE_RUNTIME_CONTEXT_H__

图引擎模块(GE)是MindSpore的一个子模块,其代码由C++实现,位于前端模块ME和底层硬件之间,起到承接作用。图引擎模块以ME下发的图作为输入,然后进行一系列的深度图优化操作,最后输出一张可以在底层硬件上高效运行的图。GE针对昇腾AI处理器的硬件结构特点,做了特定的优化工作,以此来充分发挥出昇腾AI处理器的强大算力。在进行模型训练/推理时,GE会被自动调用而用户并不感知。GE主要由GE API和GE Core两部分组成,详细的架构图如下所示