diff --git a/inc/external/acl/acl_base.h b/inc/external/acl/acl_base.h index 1f81b15a..c938d4fb 100644 --- a/inc/external/acl/acl_base.h +++ b/inc/external/acl/acl_base.h @@ -134,6 +134,7 @@ static const int ACL_ERROR_DRV_FAILURE = 500004; static const int ACL_ERROR_PROFILING_FAILURE = 500005; #define ACL_TENSOR_SHAPE_RANGE_NUM 2 +#define ACL_TENSOR_VALUE_RANGE_NUM 2 #define ACL_UNKNOWN_RANK 0xFFFFFFFFFFFFFFFE typedef enum { @@ -338,6 +339,19 @@ ACL_FUNC_VISIBILITY aclError aclSetTensorShapeRange(aclTensorDesc *desc, size_t /** * @ingroup AscendCL + * @brief set value range for aclTensorDesc + * + * @param desc [OUT] pointer to the data of aclTensorDesc + * @param valueCount [IN] the number of value + * @param valueRange [IN] the range of value + * + * @retval ACL_SUCCESS The function is successfully executed. + * @retval OtherValues Failure + */ +ACL_FUNC_VISIBILITY aclError aclSetTensorValueRange(aclTensorDesc *desc, size_t valueCount, + int64_t valueRange[][ACL_TENSOR_VALUE_RANGE_NUM]); +/** + * @ingroup AscendCL * @brief get data type specified by the tensor description * * @param desc [IN] pointer to the instance of aclTensorDesc diff --git a/inc/external/acl/acl_op_compiler.h b/inc/external/acl/acl_op_compiler.h index b64b2bad..8480b89c 100644 --- a/inc/external/acl/acl_op_compiler.h +++ b/inc/external/acl/acl_op_compiler.h @@ -41,6 +41,8 @@ typedef enum { typedef enum aclCompileFlag { ACL_OP_COMPILE_DEFAULT, ACL_OP_COMPILE_FUZZ } aclOpCompileFlag; +typedef struct aclGraphDumpOption aclGraphDumpOption; + /** * @ingroup AscendCL * @brief compile op @@ -114,6 +116,55 @@ ACL_FUNC_VISIBILITY aclError aclSetCompileopt(aclCompileOpt opt, const char *val */ ACL_FUNC_VISIBILITY aclError aclopSetCompileFlag(aclOpCompileFlag flag); +/** + * @ingroup AscendCL + * @brief generate graph and dump + * + * @param opType [IN] op type + * @param numInputs [IN] number of inputs + * @param inputDesc [IN] pointer to array of input tensor descriptions + * @param inputs [IN] pointer to array of input buffers + * @param numOutputs [IN] number of outputs + * @param outputDesc [IN] pointer to array of output tensor descriptions + * @param outputs [IN] pointer to array of outputs buffers + * @param attr [IN] pointer to instance of aclopAttr. + * may pass nullptr if the op has no attribute + * @param engineType [IN] engine type + * @param graphDumpPath [IN] dump path, if the suffix is ".txt", it means file path, else it means directory path + * @param graphDumpOpt [IN] dump option, nullptr is supported + * + * @retval ACL_SUCCESS The function is successfully executed. + * @retval OtherValues Failure + */ +ACL_FUNC_VISIBILITY aclError aclGenGraphAndDumpForOp( + const char *opType, int numInputs, const aclTensorDesc *const inputDesc[], const aclDataBuffer *const inputs[], + int numOutputs, const aclTensorDesc *const outputDesc[], aclDataBuffer *const outputs[], const aclopAttr *attr, + aclopEngineType engineType, const char *graphDumpPath, const aclGraphDumpOption *graphDumpOpt); + +/** + * @ingroup AscendCL + * @brief Create the graph dump option + * + * @retval null for failed + * @retval OtherValues success + * + * @see aclDestroyGraphDumpOpt + */ +ACL_FUNC_VISIBILITY aclGraphDumpOption *aclCreateGraphDumpOpt(); + +/** + * @ingroup AscendCL + * @brief Destroy graph dump option + * + * @param graphDumpOpt [IN] pointer to the graph dump option + * + * @retval ACL_SUCCESS The function is successfully executed. + * @retval OtherValues Failure + * + * @see aclCreateGraphDumpOpt + */ +ACL_FUNC_VISIBILITY aclError aclDestroyGraphDumpOpt(const aclGraphDumpOption *graphDumpOpt); + #ifdef __cplusplus } #endif diff --git a/inc/external/acl/acl_prof.h b/inc/external/acl/acl_prof.h index 2ad13063..1f1e38dc 100644 --- a/inc/external/acl/acl_prof.h +++ b/inc/external/acl/acl_prof.h @@ -367,6 +367,97 @@ MSVP_PROF_API aclprofStepInfo *aclprofCreateStepInfo(); */ MSVP_PROF_API void aclprofDestroyStepInfo(aclprofStepInfo *stepinfo); +/** + * @ingroup AscendCL + * @brief create pointer to aclprofstamp + * + * + * @retval aclprofStamp pointer + */ +MSVP_PROF_API void *aclprofCreateStamp(); + +/** + * @ingroup AscendCL + * @brief destory stamp pointer + * + * + * @retval void + */ +MSVP_PROF_API void aclprofDestroyStamp(void *stamp); + +/** + * @ingroup AscendCL + * @brief set category and name + * + * + * @retval void + */ +MSVP_PROF_API aclError aclprofSetCategoryName(uint32_t category, const char *categoryName); + +/** + * @ingroup AscendCL + * @brief set category to stamp + * + * + * @retval void + */ +MSVP_PROF_API aclError aclprofSetStampCategory(void *stamp, uint32_t category); + +/** + * @ingroup AscendCL + * @brief set message to stamp + * + * + * @retval void + */ +MSVP_PROF_API aclError aclprofSetStampTraceMessage(void *stamp, const char *msg, uint32_t msgLen); + +/** + * @ingroup AscendCL + * @brief Record mark timestamp + * + * @retval ACL_SUCCESS The function is successfully executed. + * @retval OtherValues Failure + */ +MSVP_PROF_API aclError aclprofMark(void *stamp); + +/** + * @ingroup AscendCL + * @brief Record push timestamp + * + * @retval ACL_SUCCESS The function is successfully executed. + * @retval OtherValues Failure + */ +MSVP_PROF_API aclError aclprofPush(void *stamp); + +/** + * @ingroup AscendCL + * @brief Record pop timestamp + * + * + * @retval ACL_SUCCESS The function is successfully executed. + * @retval OtherValues Failure + */ +MSVP_PROF_API aclError aclprofPop(); + +/** + * @ingroup AscendCL + * @brief Record range start timestamp + * + * @retval ACL_SUCCESS The function is successfully executed. + * @retval OtherValues Failure + */ +MSVP_PROF_API aclError aclprofRangeStart(void *stamp, uint32_t *rangeId); + +/** + * @ingroup AscendCL + * @brief Record range end timestamp + * + * @retval ACL_SUCCESS The function is successfully executed. + * @retval OtherValues Failure + */ +MSVP_PROF_API aclError aclprofRangeStop(uint32_t rangeId); + #ifdef __cplusplus } #endif diff --git a/inc/external/acl/error_codes/ge_error_codes.h b/inc/external/acl/error_codes/ge_error_codes.h index cafc5a64..027c83ea 100644 --- a/inc/external/acl/error_codes/ge_error_codes.h +++ b/inc/external/acl/error_codes/ge_error_codes.h @@ -32,42 +32,43 @@ #endif #include +#include #ifdef __cplusplus extern "C" { #endif -static const uint32_t ACL_ERROR_GE_PARAM_INVALID = 145000; -static const uint32_t ACL_ERROR_GE_EXEC_NOT_INIT = 145001; -static const uint32_t ACL_ERROR_GE_EXEC_MODEL_PATH_INVALID = 145002; -static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ID_INVALID = 145003; -static const uint32_t ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID = 145006; -static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ADDR_INVALID = 145007; -static const uint32_t ACL_ERROR_GE_EXEC_MODEL_QUEUE_ID_INVALID = 145008; -static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED = 145009; -static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_ADDR_INVALID = 145011; -static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_LENGTH_INVALID = 145012; -static const uint32_t ACL_ERROR_GE_DYNAMIC_BATCH_SIZE_INVALID = 145013; -static const uint32_t ACL_ERROR_GE_AIPP_BATCH_EMPTY = 145014; -static const uint32_t ACL_ERROR_GE_AIPP_NOT_EXIST = 145015; -static const uint32_t ACL_ERROR_GE_AIPP_MODE_INVALID = 145016; -static const uint32_t ACL_ERROR_GE_OP_TASK_TYPE_INVALID = 145017; -static const uint32_t ACL_ERROR_GE_OP_KERNEL_TYPE_INVALID = 145018; -static const uint32_t ACL_ERROR_GE_PLGMGR_PATH_INVALID = 145019; -static const uint32_t ACL_ERROR_GE_FORMAT_INVALID = 145020; -static const uint32_t ACL_ERROR_GE_SHAPE_INVALID = 145021; -static const uint32_t ACL_ERROR_GE_DATATYPE_INVALID = 145022; -static const uint32_t ACL_ERROR_GE_MEMORY_ALLOCATION = 245000; -static const uint32_t ACL_ERROR_GE_MEMORY_OPERATE_FAILED = 245001; -static const uint32_t ACL_ERROR_GE_INTERNAL_ERROR = 545000; -static const uint32_t ACL_ERROR_GE_LOAD_MODEL = 545001; -static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED = 545002; -static const uint32_t ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED = 545003; -static const uint32_t ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED = 545004; -static const uint32_t ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED = 545005; -static const uint32_t ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA = 545006; -static const uint32_t ACL_ERROR_GE_COMMAND_HANDLE = 545007; -static const uint32_t ACL_ERROR_GE_GET_TENSOR_INFO = 545008; -static const uint32_t ACL_ERROR_GE_UNLOAD_MODEL = 545009; +static const uint32_t ACL_ERROR_GE_PARAM_INVALID = 145000U; +static const uint32_t ACL_ERROR_GE_EXEC_NOT_INIT = 145001U; +static const uint32_t ACL_ERROR_GE_EXEC_MODEL_PATH_INVALID = 145002U; +static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ID_INVALID = 145003U; +static const uint32_t ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID = 145006U; +static const uint32_t ACL_ERROR_GE_EXEC_MODEL_ADDR_INVALID = 145007U; +static const uint32_t ACL_ERROR_GE_EXEC_MODEL_QUEUE_ID_INVALID = 145008U; +static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED = 145009U; +static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_ADDR_INVALID = 145011U; +static const uint32_t ACL_ERROR_GE_DYNAMIC_INPUT_LENGTH_INVALID = 145012U; +static const uint32_t ACL_ERROR_GE_DYNAMIC_BATCH_SIZE_INVALID = 145013U; +static const uint32_t ACL_ERROR_GE_AIPP_BATCH_EMPTY = 145014U; +static const uint32_t ACL_ERROR_GE_AIPP_NOT_EXIST = 145015U; +static const uint32_t ACL_ERROR_GE_AIPP_MODE_INVALID = 145016U; +static const uint32_t ACL_ERROR_GE_OP_TASK_TYPE_INVALID = 145017U; +static const uint32_t ACL_ERROR_GE_OP_KERNEL_TYPE_INVALID = 145018U; +static const uint32_t ACL_ERROR_GE_PLGMGR_PATH_INVALID = 145019U; +static const uint32_t ACL_ERROR_GE_FORMAT_INVALID = 145020U; +static const uint32_t ACL_ERROR_GE_SHAPE_INVALID = 145021U; +static const uint32_t ACL_ERROR_GE_DATATYPE_INVALID = 145022U; +static const uint32_t ACL_ERROR_GE_MEMORY_ALLOCATION = 245000U; +static const uint32_t ACL_ERROR_GE_MEMORY_OPERATE_FAILED = 245001U; +static const uint32_t ACL_ERROR_GE_INTERNAL_ERROR = 545000U; +static const uint32_t ACL_ERROR_GE_LOAD_MODEL = 545001U; +static const uint32_t ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED = 545002U; +static const uint32_t ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED = 545003U; +static const uint32_t ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED = 545004U; +static const uint32_t ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED = 545005U; +static const uint32_t ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA = 545006U; +static const uint32_t ACL_ERROR_GE_COMMAND_HANDLE = 545007U; +static const uint32_t ACL_ERROR_GE_GET_TENSOR_INFO = 545008U; +static const uint32_t ACL_ERROR_GE_UNLOAD_MODEL = 545009U; #ifdef __cplusplus } // namespace ge diff --git a/inc/external/acl/error_codes/rt_error_codes.h b/inc/external/acl/error_codes/rt_error_codes.h index abfa30db..d16af89f 100644 --- a/inc/external/acl/error_codes/rt_error_codes.h +++ b/inc/external/acl/error_codes/rt_error_codes.h @@ -44,6 +44,7 @@ static const int32_t ACL_ERROR_RT_STREAM_NO_CB_REG = 107015; // callbac static const int32_t ACL_ERROR_RT_INVALID_MEMORY_TYPE = 107016; // invalid memory type static const int32_t ACL_ERROR_RT_INVALID_HANDLE = 107017; // invalid handle static const int32_t ACL_ERROR_RT_INVALID_MALLOC_TYPE = 107018; // invalid malloc type +static const int32_t ACL_ERROR_RT_WAIT_TIMEOUT = 107019; // wait timeout static const int32_t ACL_ERROR_RT_FEATURE_NOT_SUPPORT = 207000; // feature not support static const int32_t ACL_ERROR_RT_MEMORY_ALLOCATION = 207001; // memory allocation error @@ -61,6 +62,7 @@ static const int32_t ACL_ERROR_RT_OVER_LIMIT = 207012; // over limit static const int32_t ACL_ERROR_RT_QUEUE_EMPTY = 207013; // queue is empty static const int32_t ACL_ERROR_RT_QUEUE_FULL = 207014; // queue is full static const int32_t ACL_ERROR_RT_REPEATED_INIT = 207015; // repeated init +static const int32_t ACL_ERROR_RT_AIVEC_OVER_FLOW = 207016; // aivec over flow static const int32_t ACL_ERROR_RT_INTERNAL_ERROR = 507000; // runtime internal error static const int32_t ACL_ERROR_RT_TS_ERROR = 507001; // ts internel error @@ -99,6 +101,11 @@ static const int32_t ACL_ERROR_RT_DEV_SETUP_ERROR = 507033; // devic static const int32_t ACL_ERROR_RT_VECTOR_CORE_TIMEOUT = 507034; // vector core timeout static const int32_t ACL_ERROR_RT_VECTOR_CORE_EXCEPTION = 507035; // vector core exception static const int32_t ACL_ERROR_RT_VECTOR_CORE_TRAP_EXCEPTION = 507036; // vector core trap exception +static const int32_t ACL_ERROR_RT_CDQ_BATCH_ABNORMAL = 507037; // cdq alloc batch abnormal +static const int32_t ACL_ERROR_RT_DIE_MODE_CHANGE_ERROR = 507038; // can not change die mode +static const int32_t ACL_ERROR_RT_DIE_SET_ERROR = 507039; // single die mode can not set die +static const int32_t ACL_ERROR_RT_INVALID_DIEID = 507040; // invalid die id +static const int32_t ACL_ERROR_RT_DIE_MODE_NOT_SET = 507041; // die mode not set static const int32_t ACL_ERROR_RT_DRV_INTERNAL_ERROR = 507899; // drv internal error static const int32_t ACL_ERROR_RT_AICPU_INTERNAL_ERROR = 507900; // aicpu internal error @@ -107,5 +114,4 @@ static const int32_t ACL_ERROR_RT_SOCKET_CLOSE = 507901; // hdc disconn #ifdef __cplusplus } #endif - #endif // __INC_EXTERNEL_RT_ERROR_CODES_H__ diff --git a/inc/external/ge/ge_api_error_codes.h b/inc/external/ge/ge_api_error_codes.h index 5d2d6963..8aee0a77 100644 --- a/inc/external/ge/ge_api_error_codes.h +++ b/inc/external/ge/ge_api_error_codes.h @@ -83,16 +83,15 @@ class GE_FUNC_VISIBILITY ErrorNoRegisterar { }; // Code compose(4 byte), runtime: 2 bit, type: 2 bit, level: 3 bit, sysid: 8 bit, modid: 5 bit, value: 12 bit -#define GE_ERRORNO(runtime, type, level, sysid, modid, name, value, desc) \ - constexpr ge::Status name = (static_cast(0xFFU & (static_cast(runtime))) << 30U) | \ - (static_cast(0xFFU & (static_cast(type))) << 28U) | \ - (static_cast(0xFFU & (static_cast(level))) << 25U) | \ - (static_cast(0xFFU & (static_cast(sysid))) << 17U) | \ - (static_cast(0xFFU & (static_cast(modid))) << 12U) | \ - (static_cast(0x0FFFU) & (static_cast(value))); \ - const ErrorNoRegisterar g_##name##_errorno(name, desc); +#define GE_ERRORNO(runtime, type, level, sysid, modid, name, value, desc) \ + constexpr ge::Status(name) = (static_cast(0xFFU & (static_cast(runtime))) << 30U) | \ + (static_cast(0xFFU & (static_cast(type))) << 28U) | \ + (static_cast(0xFFU & (static_cast(level))) << 25U) | \ + (static_cast(0xFFU & (static_cast(sysid))) << 17U) | \ + (static_cast(0xFFU & (static_cast(modid))) << 12U) | \ + (static_cast(0x0FFFU) & (static_cast(value))); -#define GE_ERRORNO_EXTERNAL(name, desc) const ErrorNoRegisterar g_##name##_errorno(name, desc); +#define GE_ERRORNO_EXTERNAL(name, desc) constexpr ErrorNoRegisterar g_errorno_##name(name, desc); using Status = uint32_t; @@ -100,41 +99,6 @@ using Status = uint32_t; GE_ERRORNO(0, 0, 0, 0, 0, SUCCESS, 0, "success"); GE_ERRORNO(0b11, 0b11, 0b111, 0xFFU, 0b11111, FAILED, 0xFFFU, "failed"); /*lint !e401*/ -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_PARAM_INVALID, "Parameter invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_NOT_INIT, "GE executor not initialized yet."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_PATH_INVALID, "Model file path invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_ID_INVALID, "Model id invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID, "Data size of model invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_ADDR_INVALID, "Model addr invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_MODEL_QUEUE_ID_INVALID, "Queue id of model invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED, "The model loaded repeatedly."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_DYNAMIC_INPUT_ADDR_INVALID, "Dynamic input addr invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_DYNAMIC_INPUT_LENGTH_INVALID, "Dynamic input size invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_DYNAMIC_BATCH_SIZE_INVALID, "Dynamic batch size invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_AIPP_BATCH_EMPTY, "AIPP batch parameter empty."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_AIPP_NOT_EXIST, "AIPP parameter not exist."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_AIPP_MODE_INVALID, "AIPP mode invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_OP_TASK_TYPE_INVALID, "Task type invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_OP_KERNEL_TYPE_INVALID, "Kernel type invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_PLGMGR_PATH_INVALID, "Plugin path is invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_FORMAT_INVALID, "Format is invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_SHAPE_INVALID, "Shape is invalid."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_DATATYPE_INVALID, "Datatype is invalid."); - -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_MEMORY_ALLOCATION, "Memory allocation error."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_MEMORY_OPERATE_FAILED, "Failed to operate memory."); - -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_INTERNAL_ERROR, "Internal error."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_LOAD_MODEL, "Load model error."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED, "Failed to load model partition."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED, "Failed to load weight partition."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED, "Failed to load task partition."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED, "Failed to load op kernel partition."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA, "Failed to release the model data."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_COMMAND_HANDLE, "Command handle error."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_GET_TENSOR_INFO, "Get tensor info error."); -GE_ERRORNO_EXTERNAL(ACL_ERROR_GE_UNLOAD_MODEL, "Load model error."); - } // namespace ge #endif // INC_EXTERNAL_GE_GE_API_ERROR_CODES_H_ diff --git a/inc/external/runtime/rt_error_codes.h b/inc/external/runtime/rt_error_codes.h index 8966f738..d16af89f 100644 --- a/inc/external/runtime/rt_error_codes.h +++ b/inc/external/runtime/rt_error_codes.h @@ -62,6 +62,7 @@ static const int32_t ACL_ERROR_RT_OVER_LIMIT = 207012; // over limit static const int32_t ACL_ERROR_RT_QUEUE_EMPTY = 207013; // queue is empty static const int32_t ACL_ERROR_RT_QUEUE_FULL = 207014; // queue is full static const int32_t ACL_ERROR_RT_REPEATED_INIT = 207015; // repeated init +static const int32_t ACL_ERROR_RT_AIVEC_OVER_FLOW = 207016; // aivec over flow static const int32_t ACL_ERROR_RT_INTERNAL_ERROR = 507000; // runtime internal error static const int32_t ACL_ERROR_RT_TS_ERROR = 507001; // ts internel error @@ -113,5 +114,4 @@ static const int32_t ACL_ERROR_RT_SOCKET_CLOSE = 507901; // hdc disconn #ifdef __cplusplus } #endif - #endif // __INC_EXTERNEL_RT_ERROR_CODES_H__ diff --git a/inc/framework/common/debug/ge_log.h b/inc/framework/common/debug/ge_log.h index b1bf929f..2cb56222 100644 --- a/inc/framework/common/debug/ge_log.h +++ b/inc/framework/common/debug/ge_log.h @@ -40,7 +40,7 @@ enum TraceStatus { TRACE_INIT = 0, TRACE_RUNNING, TRACE_WAITING, TRACE_STOP }; class GE_FUNC_VISIBILITY GeLog { public: - static const uint64_t GetTid() { + static uint64_t GetTid() { #ifdef __GNUC__ const uint64_t tid = static_cast(syscall(__NR_gettid)); #else @@ -56,11 +56,11 @@ inline bool IsLogEnable(const int32_t module_name, const int32_t log_level) { return (enable == 1); } -#define GELOGE(ERROR_CODE, fmt, ...) \ - do { \ - dlog_error(GE_MODULE_NAME, "%lu %s: ErrorNo: %u(%s) %s" fmt, GeLog::GetTid(), &__FUNCTION__[0], ERROR_CODE, \ - ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()), ErrorManager::GetInstance().GetLogHeader().c_str(), \ - ##__VA_ARGS__); \ +#define GELOGE(ERROR_CODE, fmt, ...) \ + do { \ + dlog_error(GE_MODULE_NAME, "%lu %s: ErrorNo: %u(%s) %s" fmt, GeLog::GetTid(), &__FUNCTION__[0], (ERROR_CODE), \ + ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()), ErrorManager::GetInstance().GetLogHeader().c_str(), \ + ##__VA_ARGS__); \ } while (false) #define GELOGW(fmt, ...) \ @@ -91,7 +91,7 @@ inline bool IsLogEnable(const int32_t module_name, const int32_t log_level) { #define GELOGT(VALUE, fmt, ...) \ do { \ - TraceStatus stat = VALUE; \ + TraceStatus stat = (VALUE); \ const char_t *const TraceStatStr[] = {"INIT", "RUNNING", "WAITING", "STOP"}; \ const int32_t idx = static_cast(stat); \ char_t *k = const_cast("status"); \ @@ -102,7 +102,7 @@ inline bool IsLogEnable(const int32_t module_name, const int32_t log_level) { #define GE_LOG_ERROR(MOD_NAME, ERROR_CODE, fmt, ...) \ do { \ - dlog_error(MOD_NAME, "%lu %s: ErrorNo: %u(%s) %s" fmt, GeLog::GetTid(), &__FUNCTION__[0], ERROR_CODE, \ + dlog_error((MOD_NAME), "%lu %s: ErrorNo: %u(%s) %s" fmt, GeLog::GetTid(), &__FUNCTION__[0], (ERROR_CODE), \ ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()), ErrorManager::GetInstance().GetLogHeader().c_str(), \ ##__VA_ARGS__); \ } while (false) diff --git a/inc/framework/common/fmk_error_codes.h b/inc/framework/common/fmk_error_codes.h index 9f1719ac..e244502d 100644 --- a/inc/framework/common/fmk_error_codes.h +++ b/inc/framework/common/fmk_error_codes.h @@ -74,7 +74,7 @@ class GE_FUNC_VISIBILITY StatusFactory { class GE_FUNC_VISIBILITY ErrorNoRegisterar { public: - ErrorNoRegisterar(uint32_t err, const std::string &desc) { + ErrorNoRegisterar(const uint32_t err, const std::string &desc) { StatusFactory::Instance()->RegisterErrorNo(err, desc); } ~ErrorNoRegisterar() {} diff --git a/inc/framework/common/ge_types.h b/inc/framework/common/ge_types.h index 868debe7..2e374831 100644 --- a/inc/framework/common/ge_types.h +++ b/inc/framework/common/ge_types.h @@ -40,8 +40,8 @@ enum FrameworkType { CAFFE = 0, MINDSPORE = 1, TENSORFLOW = 3, - ANDROID_NN, - ONNX, + ANDROID_NN = 4, + ONNX = 5, }; enum class GraphStage : int64_t { GRAPH_STAGE_FUZZ = 0, GRAPH_STAGE_RESERVED }; @@ -81,7 +81,7 @@ struct DataBuffer { uint64_t length; // Data length bool isDataSupportMemShare = false; uint32_t placement = 0U; - DataBuffer(void *data_in, uint64_t data_len, bool is_support_mem_share, uint32_t placement = 0U) + DataBuffer(void *data_in, const uint64_t data_len, const bool is_support_mem_share, const uint32_t placement = 0U) : data(data_in), length(data_len), isDataSupportMemShare(is_support_mem_share), placement(placement) {} DataBuffer() : data(nullptr), length(0UL), isDataSupportMemShare(false) {} diff --git a/inc/framework/common/helper/model_helper.h b/inc/framework/common/helper/model_helper.h index 8da856e1..53881ccd 100644 --- a/inc/framework/common/helper/model_helper.h +++ b/inc/framework/common/helper/model_helper.h @@ -40,6 +40,7 @@ class GE_FUNC_VISIBILITY ModelHelper { Status SaveOriginalGraphToOmModel(const ge::Graph &graph, const std::string &output_file); Status LoadModel(const ge::ModelData &model_data); Status LoadRootModel(const ge::ModelData &model_data); + static void SetModelToGeModel(GeModelPtr &ge_model, Model &model); GeModelPtr GetGeModel(); GeRootModelPtr GetGeRootModel(); @@ -67,7 +68,6 @@ class GE_FUNC_VISIBILITY ModelHelper { Status GenerateGeModel(OmFileLoadHelper &om_load_helper); Status GenerateGeRootModel(OmFileLoadHelper &om_load_helper); Status LoadModelData(OmFileLoadHelper &om_load_helper); - void SetModelToGeModel(GeModelPtr &ge_model, Model &model) const; Status LoadModelData(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, const size_t mode_index) const; Status LoadWeights(OmFileLoadHelper &om_load_helper); Status LoadWeights(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, const size_t mode_index) const; diff --git a/inc/framework/common/helper/om_file_helper.h b/inc/framework/common/helper/om_file_helper.h index 45d61395..80232bae 100644 --- a/inc/framework/common/helper/om_file_helper.h +++ b/inc/framework/common/helper/om_file_helper.h @@ -21,25 +21,22 @@ #include #include "external/ge/ge_ir_build.h" -#include "framework/common/fmk_types.h" #include "framework/common/types.h" #include "framework/common/ge_types.h" -using ProcParam = struct PROC_PARAM; -using std::string; using std::vector; namespace ge { struct ModelPartition { ModelPartitionType type; - uint8_t *data = 0; - uint32_t size = 0; + uint8_t *data = nullptr; + uint32_t size = 0U; }; struct OmFileContext { std::vector partition_datas_; - std::vector partition_table_; - uint32_t model_data_len_ = 0; + std::vector partition_table_; + uint32_t model_data_len_ = 0U; }; struct SaveParam { @@ -55,13 +52,13 @@ class GE_FUNC_VISIBILITY OmFileLoadHelper { public: Status Init(const ge::ModelData &model); - Status Init(uint8_t *model_data, const uint32_t model_data_size); + Status Init(uint8_t *const model_data, const uint32_t model_data_size); - Status Init(uint8_t *model_data, const uint32_t model_data_size, uint32_t model_num); + Status Init(uint8_t *const model_data, const uint32_t model_data_size, const uint32_t model_num); - Status GetModelPartition(ModelPartitionType type, ModelPartition &partition); + Status GetModelPartition(const ModelPartitionType type, ModelPartition &partition); - Status GetModelPartition(ModelPartitionType type, ModelPartition &partition, size_t model_index); + Status GetModelPartition(const ModelPartitionType type, ModelPartition &partition, const size_t model_index); OmFileContext context_; @@ -70,9 +67,9 @@ class GE_FUNC_VISIBILITY OmFileLoadHelper { private: Status CheckModelValid(const ge::ModelData &model) const; - Status LoadModelPartitionTable(uint8_t *model_data, const uint32_t model_data_size); + Status LoadModelPartitionTable(uint8_t *const model_data, const uint32_t model_data_size); - Status LoadModelPartitionTable(uint8_t *model_data, const uint32_t model_data_size, uint32_t model_num); + Status LoadModelPartitionTable(uint8_t *const model_data, const uint32_t model_data_size, const uint32_t model_num); bool is_inited_{false}; }; @@ -89,16 +86,16 @@ class GE_FUNC_VISIBILITY OmFileSaveHelper { ModelPartitionTable *GetPartitionTable(); - Status AddPartition(ModelPartition &partition); + Status AddPartition(const ModelPartition &partition); - Status AddPartition(ModelPartition &partition, size_t cur_index); + Status AddPartition(const ModelPartition &partition, const size_t cur_index); const std::vector &GetModelPartitions() const; - Status SaveModel(const SaveParam &save_param, const char *target_file, ge::ModelBufferData &model, - bool is_offline = true); + Status SaveModel(const SaveParam &save_param, const char_t *const output_file, ge::ModelBufferData &model, + const bool is_offline = true); - Status SaveModelToFile(const char *output_file, ge::ModelBufferData &model, bool is_offline = true); + Status SaveModelToFile(const char_t *const output_file, ge::ModelBufferData &model, const bool is_offline = true); std::vector model_contexts_; diff --git a/inc/framework/common/op/attr_value_util.h b/inc/framework/common/op/attr_value_util.h index 27415df8..8d31f9da 100644 --- a/inc/framework/common/op/attr_value_util.h +++ b/inc/framework/common/op/attr_value_util.h @@ -34,143 +34,18 @@ #include #include #include - +#include "external/graph/types.h" #include "graph/debug/ge_attr_define.h" #include "proto/om.pb.h" -using domi::AttrDef; -using domi::AttrDef_ListValue; -using domi::ModelDef; -using domi::NamedAttrs; -using domi::OpDef; - namespace ge { -using AttrDefMap = ::google::protobuf::Map<::std::string, ::domi::AttrDef>; -using AttrDefPair = ::google::protobuf::MapPair; - -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, AttrDef &attr, OpDef *opdef); -// DEFINE_ADD_ATTR_VALUE -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const std::string &value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const char *value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttr(const char *key, const char *value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const uint32_t value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const int32_t value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const int64_t value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const float value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const double value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const bool value, AttrDefMap *attrs); - -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const AttrDef_ListValue &value, AttrDefMap *attrs); - -// DEFINE_ADD_ATTR_VALUE -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const std::string &value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const char *value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttr(const char *key, const char *value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const uint32_t value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const int32_t value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const int64_t value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const float value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const double value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const bool value, OpDef *opdef); - -GE_FUNC_VISIBILITY void AddOpAttr(const std::string &key, const AttrDef_ListValue &value, OpDef *opdef); - -GE_FUNC_VISIBILITY void AddOpBytesAttr(const std::string &key, const void *value, size_t size, OpDef *opdef); - -// DEFINE_ADD_ATTR_VALUE_LIST -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const double value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const float value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const uint32_t value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const int32_t value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const std::string value, AttrDefMap *attrs); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const double value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const float value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const uint32_t value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const int32_t value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const bool value, OpDef *opdef); -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const int64_t value, OpDef *opdef); - -GE_FUNC_VISIBILITY void AddOpAttrList(const std::string &key, const std::string &value, OpDef *opdef); - -GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, std::string *value, const OpDef *opdef); -GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, int32_t *value, const OpDef *opdef); -GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, int64_t *value, const OpDef *opdef); -GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, uint32_t *value, const OpDef *opdef); -GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, float *value, const OpDef *opdef); -GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, double *value, const OpDef *opdef); -GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, bool *value, const OpDef *opdef); -GE_FUNC_VISIBILITY bool GetOpAttr(const std::string &key, AttrDef_ListValue *value, const OpDef *opdef); - -GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, std::string value, const OpDef *opdef); -GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, int32_t value, const OpDef *opdef); -GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, int64_t value, const OpDef *opdef); -GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, uint32_t value, const OpDef *opdef); -GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, float value, const OpDef *opdef); -GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, double value, const OpDef *opdef); -GE_FUNC_VISIBILITY uint32_t GetOpAttrListSize(const std::string &key, bool value, const OpDef *opdef); - -GE_FUNC_VISIBILITY bool GetBytesAttr(const std::string &key, std::string *value, const OpDef *opdef); -GE_FUNC_VISIBILITY bool GetBytesAttr(const std::string &key, std::string *value, const ModelDef *model_def); - -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const std::string &value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const char *value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const char *key, const char *value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const uint32_t value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const int32_t value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const int64_t value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const float value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const double value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const bool value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const void *value, size_t size, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttr(const std::string &key, const AttrDef_ListValue &value, ModelDef *model_def); - -GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const double value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const float value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const uint32_t value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const int32_t value, ModelDef *model_def); -GE_FUNC_VISIBILITY void AddModelAttrList(const std::string &key, const std::string &value, ModelDef *model_def); - -GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, std::string *value, const ModelDef *model_def); -GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, int32_t *value, const ModelDef *model_def); -GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, int64_t *value, const ModelDef *model_def); -GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, uint32_t *value, const ModelDef *model_def); -GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, float *value, const ModelDef *model_def); -GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, double *value, const ModelDef *model_def); -GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, bool *value, const ModelDef *model_def); -GE_FUNC_VISIBILITY bool GetModelAttr(const std::string &key, AttrDef_ListValue *value, const ModelDef *model_def); - -GE_FUNC_VISIBILITY bool HasOpAttr(const OpDef *opdef, const std::string &attr_name); - -GE_FUNC_VISIBILITY void SetAttrDef(const std::string &value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrDef(const char *value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrDef(const uint32_t value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrDef(const int32_t value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrDef(const float value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrDef(const double value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrDef(const bool value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrList(const std::string &value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrList(const bool value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrList(const float value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrList(const double value, AttrDef *out); -GE_FUNC_VISIBILITY void SetAttrList(const uint32_t value, AttrDef *out); - -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, std::string *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, int32_t *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, int64_t *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, uint32_t *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, float *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, double *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, bool *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, AttrDef_ListValue *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, NamedAttrs *&value, AttrDefMap *attr); -GE_FUNC_VISIBILITY bool GetAttrDefValue(const std::string &key, const NamedAttrs *&value, const AttrDefMap &attr); - -GE_FUNC_VISIBILITY bool GetAttrDefListValue(const std::string &key, int32_t idx, int32_t *value, - const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefListValue(const std::string &key, int32_t idx, uint32_t *value, - const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefListValue(const std::string &key, int32_t idx, float *value, const AttrDefMap &attr); -GE_FUNC_VISIBILITY bool GetAttrDefListValue(const std::string &key, int32_t idx, double *value, const AttrDefMap &attr); +GE_FUNC_VISIBILITY void SetAttrDef(const std::string &value, domi::AttrDef *const out); +GE_FUNC_VISIBILITY void SetAttrDef(const char *value, domi::AttrDef *const out); +GE_FUNC_VISIBILITY void SetAttrDef(const uint32_t value, domi::AttrDef *const out); +GE_FUNC_VISIBILITY void SetAttrDef(const int32_t value, domi::AttrDef *const out); +GE_FUNC_VISIBILITY void SetAttrDef(const int64_t value, domi::AttrDef *const out); +GE_FUNC_VISIBILITY void SetAttrDef(const float32_t value, domi::AttrDef *const out); +GE_FUNC_VISIBILITY void SetAttrDef(const float64_t value, domi::AttrDef *const out); +GE_FUNC_VISIBILITY void SetAttrDef(const bool value, domi::AttrDef *const out); } // namespace ge - -#endif // INC_FRAMEWORK_COMMON_OP_ATTR_VALUE_UTIL_H_ +#endif // INC_FRAMEWORK_COMMON_OP_ATTR_VALUE_UTIL_H_ \ No newline at end of file diff --git a/inc/framework/common/op/ge_op_utils.h b/inc/framework/common/op/ge_op_utils.h index 8b28258a..63b75208 100644 --- a/inc/framework/common/op/ge_op_utils.h +++ b/inc/framework/common/op/ge_op_utils.h @@ -31,7 +31,6 @@ #include "proto/insert_op.pb.h" namespace ge { -using domi::Status; // Add Sub Mul GE_FUNC_VISIBILITY extern const uint32_t ADD_INPUT_NUM; @@ -55,8 +54,8 @@ GE_FUNC_VISIBILITY extern const uint32_t SWITCH_DATA_INPUT; GE_FUNC_VISIBILITY extern const uint32_t SWITCH_PRED_INPUT; // Merge -GE_FUNC_VISIBILITY extern const uint32_t MERGE_DATA_OUTPUT; -GE_FUNC_VISIBILITY extern const uint32_t MERGE_INDEX_OUTPUT; +GE_FUNC_VISIBILITY extern const int32_t MERGE_DATA_OUTPUT; +GE_FUNC_VISIBILITY extern const int32_t MERGE_INDEX_OUTPUT; // FunctionOp GE_FUNC_VISIBILITY extern const uint32_t IF_COND_INPUT; @@ -66,7 +65,7 @@ GE_FUNC_VISIBILITY extern const uint32_t FOR_DELTA_INPUT; GE_FUNC_VISIBILITY extern const uint32_t FOR_DATA_INPUT; GE_FUNC_VISIBILITY extern const int32_t NORMAL_TENSOR_SIZE; - +/*lint -e148*/ class GE_FUNC_VISIBILITY OpUtils { public: /// @@ -95,8 +94,8 @@ class GE_FUNC_VISIBILITY OpUtils { /// @param [out] aipp_params aipp parameters /// @return enum of tagCCAippInputFormat /// - static Status ConvertAippParams(const NamedAttrs &aipp_attr, domi::AippOpParams *aipp_params); - static Status TransferDim(const std::vector &dim, std::vector &dim_vector); + + static Status ConvertAippParams(const NamedAttrs &aipp_attr, domi::AippOpParams &aipp_params); template static void SliceData(const std::vector &input, int64_t chunk_size, std::vector &output, int64_t begin, int64_t out_dim, int64_t stride); @@ -107,45 +106,13 @@ class GE_FUNC_VISIBILITY OpUtils { static Status SetOutputSliceDataByDataType(void *data, int64_t data_size, const std::vector &input_dims, const std::vector &begin, const std::vector &output_dims, ge::GeTensor *output, const std::vector &stride); - static Status SetOutputSliceData(void *data, int64_t data_size, int32_t data_type, + static Status SetOutputSliceData(void *const data, const int64_t data_size, const int32_t data_type, const std::vector &input_dims, const std::vector &begin, - const std::vector &output_dims, ge::GeTensor *const output, + const std::vector &output_dims, GeTensor *const output, const std::vector &stride); - - /// - /// @ingroup domi_omg - /// @brief Convert the convolutional weight data from [h, w, c, k] to [k, c, h, w] - /// @param [in] input Weight data in HWCK format - /// @param [in] H value of H dimension - /// @param [in] W value of W dimension - /// @param [in] C value of C dimension - /// @param [in] K value of K dimension - /// @param [out] output Data pointer after conversion. The format is KCHW. - /// - static void TransDataHWCK2KCHW(const void *input, int64_t h, int64_t w, int64_t c, int64_t k, void **output); - /// - /// @ingroup domi_omg - /// @brief Converts the convolutional weight data from [k, c, h, w] to [h, w, c, k]. - /// @param [in] input Weight data in HWCK format - /// @param [in] K value of K dimension - /// @param [in] C value of C dimension - /// @param [in] H value of H dimension - /// @param [in] W value of W dimension - /// @param [out] output Data pointer after conversion. The format is HWCK - /// - static void TransDataKCHW2HWCK(const void *input, int64_t k, int64_t c, int64_t h, int64_t w, void *output); - - static std::vector GetWeights(const ge::Node &node); - static std::vector GetWeights(ge::ConstNodePtr node); - static std::vector MutableWeights(const ge::Node &node); - static std::vector MutableWeights(const ge::NodePtr node); - static Status SetWeights(ge::Node &node, const std::vector &weights); - static Status SetWeights(const ge::NodePtr node, const std::vector &weights); static Status GetShapeDataFromConstTensor(const ConstGeTensorPtr &tensor, const DataType type, std::vector &dims); - - private: - static uint32_t GetRealDimCnt(const GeTensorDesc &tensor_desc); }; +/*lint +e148*/ } // namespace ge #endif // INC_FRAMEWORK_COMMON_OP_GE_OP_UTILS_H_ diff --git a/inc/framework/common/op_types.h b/inc/framework/common/op_types.h index 4e4d4dda..76f5110a 100644 --- a/inc/framework/common/op_types.h +++ b/inc/framework/common/op_types.h @@ -20,6 +20,8 @@ #include #include +#include "graph/types.h" + namespace ge { class GE_FUNC_VISIBILITY OpTypeContainer { public: @@ -47,17 +49,17 @@ class GE_FUNC_VISIBILITY OpTypeContainer { class GE_FUNC_VISIBILITY OpTypeRegistrar { public: - explicit OpTypeRegistrar(const std::string &op_type) { + explicit OpTypeRegistrar(const std::string &op_type) noexcept { OpTypeContainer::Instance()->Register(op_type); } ~OpTypeRegistrar() {} }; #define REGISTER_OPTYPE_DECLARE(var_name, str_name) \ - FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char *var_name; + FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const char_t *var_name; #define REGISTER_OPTYPE_DEFINE(var_name, str_name) \ - const char *var_name = str_name; \ + const char_t *var_name = str_name; \ const OpTypeRegistrar g_##var_name##_reg(str_name); #define IS_OPTYPE_EXISTING(str_name) (OpTypeContainer::Instance()->IsExisting(str_name)) diff --git a/inc/framework/common/profiling_definitions.h b/inc/framework/common/profiling_definitions.h new file mode 100644 index 00000000..e6da4ca8 --- /dev/null +++ b/inc/framework/common/profiling_definitions.h @@ -0,0 +1,136 @@ +/** + * Copyright 2021 Huawei Technologies Co., Ltd + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef AIR_CXX_PROFILING_DEFINITIONS_H +#define AIR_CXX_PROFILING_DEFINITIONS_H +#include +#include +#include +#include +#include +#include "graph/profiler.h" +namespace ge { +namespace profiling { +enum { + kInferShape, + kTiling, + kUpdateShape, + kConstPrepare, + kInitHybridExecuteArgs, + kInitInferShapeContext, + kDestroyInferShapeContext, + kResetSubgraphExecutor, + kCommitInferShapeTask, + kDeviceToHost, + kPrepareTask, + kLaunchTask, + kCommitTilingTask, + kAtomic, + kKernelLaunchPrepare, + kRtKernelLaunch, + kOpExecute, + kAllocMem, + kCopyH2D, + kProfilingIndexEnd +}; + +class ProfilingContext { + public: + static bool IsDumpToStdEnabled(); + static ProfilingContext &GetInstance(); + ProfilingContext(); + ~ProfilingContext(); + + /* + * 还有一种思路是`IsEnabled`只判断profiler_是否为空指针，不再设置单独的enabled标记位，这样可以少一个标记位。 + * 但是这么做就意味着，profiler_实例在未使能profiling时，必须是空指针状态。 + * 为了性能考虑，profiling机制在编译和加载时，就会调用`RegisterString`，向profiler_注册字符串，后续执行时，只会使用注册好的index了。 + * 因此存在一种场景：编译时并未使能profiling（因为编译时间很长，使能profiling也无法真实反应执行时的耗时状态）， + * 因此编译时注册字符串的动作并没有生效。在执行时，动态的打开了profiling，这种场景下，执行时无法拿到注册后字符串 + */ + bool IsEnabled() const noexcept { + return enabled_ && profiler_ != nullptr; + } + void SetEnable() noexcept { + enabled_ = true; + } + void SetDisable() noexcept { + enabled_ = false; + } + + void RecordCurrentThread(int64_t element, int64_t event, EventType et) { + if (IsEnabled()) { + profiler_->RecordCurrentThread(element, event, et); + } + } + + const Profiler *GetProfiler() const { + return profiler_.get(); + } + + void Dump(std::ostream &out_stream) const { + if (IsEnabled()) { + profiler_->Dump(out_stream); + } else { + out_stream << "Profiling not enable, skip to dump" << std::endl; + } + } + + void DumpToStdOut() const { + Dump(std::cout); + } + + void Reset() { + if (IsEnabled()) { + profiler_->Reset(); + } + } + + int64_t RegisterString(const std::string &str); + + private: + void RegisterString(int64_t index, const std::string &str); + void Init(); + + private: + bool enabled_; + int64_t str_index_; + std::unordered_map strings_to_index_; + std::mutex strings_to_index_mutex_; + std::unique_ptr profiler_; +}; + +class ScopeProfiler { + public: + ScopeProfiler(int64_t element, int64_t event) : element_(element), event_(event) { + ProfilingContext::GetInstance().RecordCurrentThread(element_, event, kEventStart); + } + ~ScopeProfiler() { + ProfilingContext::GetInstance().RecordCurrentThread(element_, event_, kEventEnd); + } + + private: + int64_t element_; + int64_t event_; +}; +} // namespace profiling +} // namespace ge +#define PROFILING_START(element, event) \ + profiling::ProfilingContext::GetInstance().RecordCurrentThread(element, event, profiling::kEventStart) +#define PROFILING_END(element, event) \ + profiling::ProfilingContext::GetInstance().RecordCurrentThread(element, event, profiling::kEventEnd) +#define PROFILING_SCOPE(element, event) profiling::ScopeProfiler profiler(element, event) +#endif // AIR_CXX_PROFILING_DEFINITIONS_H diff --git a/inc/framework/common/scope_guard.h b/inc/framework/common/scope_guard.h index 4ec4cfa7..2ac6408c 100644 --- a/inc/framework/common/scope_guard.h +++ b/inc/framework/common/scope_guard.h @@ -25,9 +25,9 @@ /// MAKE_GUARD([&] { Release Resource 1 }) /// Acquire Resource 2 // MAKE_GUARD([&] { Release Resource 2 }) -#define GE_MAKE_GUARD(var, callback) const ScopeGuard const_guard_##var(callback) +#define GE_MAKE_GUARD(var, callback) const ::ge::ScopeGuard const_guard_##var(callback) -#define GE_DISMISSABLE_GUARD(var, callback) ScopeGuard make_guard_##var(callback) +#define GE_DISMISSABLE_GUARD(var, callback) ::ge::ScopeGuard make_guard_##var(callback) #define GE_DISMISS_GUARD(var) make_guard_##var.Dismiss() namespace ge { diff --git a/inc/framework/common/types.h b/inc/framework/common/types.h index e5681ba6..22a3a174 100644 --- a/inc/framework/common/types.h +++ b/inc/framework/common/types.h @@ -53,23 +53,11 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_TASK_GEN_BASE_ADDR; FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_TASK_GEN_WEIGHT_ADDR; FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string MODEL_ATTR_FUSION_MODEL_DEF; - FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint64_t ALLOC_MEMORY_MAX_SIZE; // Max size of 8 GB. -template -static std::pair flip_pair(const std::pair &p) { - return std::pair(p.second, p.first); -} - -template -static std::map flip_map(std::map src) { - std::map dst; - std::transform(src.begin(), src.end(), std::inserter(dst, dst.begin()), flip_pair); - return dst; -} - REGISTER_OPTYPE_DECLARE(DATA, "Data"); REGISTER_OPTYPE_DECLARE(AIPPDATA, "AippData"); +REGISTER_OPTYPE_DECLARE(QUEUE_DATA, "QueueData"); REGISTER_OPTYPE_DECLARE(CONVOLUTION, "Convolution"); REGISTER_OPTYPE_DECLARE(CORRELATION, "Correlation"); REGISTER_OPTYPE_DECLARE(CORRELATIONV2, "Correlation_V2"); @@ -516,30 +504,6 @@ REGISTER_OPTYPE_DECLARE(GETDYNAMICDIMS, "GetDynamicDims"); // profiling training trace node REGISTER_OPTYPE_DECLARE(PROFILINGTRAININGTRACE, "ProfilingTrainingTrace"); -enum InputMode { INPUT = 0, CONST_INPUT }; - -// Definition of the processing status enum of the process module -enum ModelProcessState { - INIT_STATE = 0, // init status - WAIT_EVENT_STATE, // Wait for the event status - IND_RSLT_STATE, // The model execution result is being output to the high level - STOPPED_STATE, // Model execution completed. The model enters this state after Model Manager::Stop - RESERVED_STATE, // reserved -}; - -// Indicates the enun definition of the execution mode of the access module -enum SysMode { - INFERENCE = 0, // Normal, that is, Inference mode - DEBUG, // Debug mode - TIME, // Model execution time mode, including the execution time of each OP - STOP, // STOP mode - RESET, // RESET mode - PERFORMANCE, // Impact of enabling the performance model: 1. The input data of the model is considered ready and does - // not need to be converted - ANDROID_DEBUG, // Exports Android platform computing data - RESERVED, // reserved -}; - // @brief encryption type of the model file enum ModelEncryptType { UNENCRYPTED, // not encrypted @@ -577,22 +541,22 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t MODEL_FIL /// /// @brief model name length /// -static constexpr uint32_t MODEL_NAME_LENGTH = 32; +constexpr uint32_t MODEL_NAME_LENGTH = 32U; /// /// @brief length of user-defined information /// -static constexpr uint32_t USER_DEFINE_INFO_LENGTH = 32; +constexpr uint32_t USER_DEFINE_INFO_LENGTH = 32U; /// /// @brief length of the model file signature /// -static constexpr uint32_t MODEL_FILE_CHECKSUM_LENGTH = 64; +constexpr uint32_t MODEL_FILE_CHECKSUM_LENGTH = 64U; /// /// @brief length of the reserved field in the model file header /// -static constexpr uint32_t MODEL_FILE_RESERVED_LENGTH = 75; +constexpr uint32_t MODEL_FILE_RESERVED_LENGTH = 75U; // DATA node type FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string DATA_TYPE; @@ -617,7 +581,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OP_TYP FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string OP_CONF_DELIMITER; // dim default size value -static const int32_t DIM_DEFAULT_SIZE = 4; +constexpr int32_t DIM_DEFAULT_SIZE = 4; // dim extension default value FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const int32_t DIM_DEFAULT_VALUE; @@ -650,34 +614,35 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const uint32_t STREAM_SW FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY extern const std::string NODE_NAME_GLOBAL_STEP; -static const uint32_t PLATFORM_VERSION_LEN = 20; +constexpr uint32_t PLATFORM_VERSION_LEN = 20U; // Definition of the file header of the model file struct ModelFileHeader { - uint32_t magic = MODEL_FILE_MAGIC_NUM; // magic number of DOMI - uint32_t headsize = MODEL_FILE_HEAD_LEN; // length of the model header. The value is fixed at 256 - uint32_t version = MODEL_VERSION; // version 1.0 - uint8_t checksum[MODEL_FILE_CHECKSUM_LENGTH] = {0}; // signature - uint32_t length = 0; // Ciphertext length. In the non-encryption model, the length is the plaintext length. - uint8_t is_encrypt = ModelEncryptType::UNENCRYPTED; // whether encrypted 0:not encrypt, 1:encrypt - uint8_t is_checksum = ModelCheckType::CHECK; // whether to check the checksum - uint8_t modeltype = 0; // 0：IR model 1：standard model 2: OM Tiny model - uint8_t genmode = 0; // 0：offline generate 1：online generate - uint8_t name[MODEL_NAME_LENGTH] = {0}; // Model name, which contains 32 characters - uint32_t ops = 0; // Computing power (Kops) - uint8_t userdefineinfo[USER_DEFINE_INFO_LENGTH] = {0}; // User-defined information. The value contains 32 characters - uint32_t om_ir_version = 0; - uint32_t model_num = 0; - uint8_t platform_version[PLATFORM_VERSION_LEN] = {0}; - uint8_t platform_type = {0}; - uint8_t reserved[MODEL_FILE_RESERVED_LENGTH] = {0}; // Reserved field 75 + uint32_t magic = MODEL_FILE_MAGIC_NUM; // magic number of DOMI + uint32_t headsize = MODEL_FILE_HEAD_LEN; // length of the model header. The value is fixed at 256 + uint32_t version = MODEL_VERSION; // version 1.0 + uint8_t checksum[MODEL_FILE_CHECKSUM_LENGTH] = {0U}; // signature + uint32_t length = 0U; // Ciphertext length. In the non-encryption model, the length is the plaintext length. + uint8_t is_encrypt = + static_cast(ModelEncryptType::UNENCRYPTED); // whether encrypted 0:not encrypt, 1:encrypt + uint8_t is_checksum = static_cast(ModelCheckType::CHECK); // whether to check the checksum + uint8_t modeltype = 0U; // 0：IR model 1：standard model 2: OM Tiny model + uint8_t genmode = 0U; // 0：offline generate 1：online generate + uint8_t name[MODEL_NAME_LENGTH] = {0U}; // Model name, which contains 32 characters + uint32_t ops = 0U; // Computing power (Kops) + uint8_t userdefineinfo[USER_DEFINE_INFO_LENGTH] = {0U}; // User-defined information. The value contains 32 characters + uint32_t om_ir_version = 0U; + uint32_t model_num = 0U; + uint8_t platform_version[PLATFORM_VERSION_LEN] = {0U}; + uint8_t platform_type = {0U}; + uint8_t reserved[MODEL_FILE_RESERVED_LENGTH] = {0U}; // Reserved field 75 }; -static constexpr uint8_t TARGET_TYPE_LTTE_8BIT = 0; -static constexpr uint8_t TARGET_TYPE_MINI_8BIT = 1; +constexpr uint8_t TARGET_TYPE_LTTE_8BIT = 0U; +constexpr uint8_t TARGET_TYPE_MINI_8BIT = 1U; // number of partitions in the current model -static constexpr uint32_t PARTITION_SIZE = 5; +constexpr uint32_t PARTITION_SIZE = 5U; enum ModelPartitionType { MODEL_DEF = 0, WEIGHTS_DATA, TASK_INFO, TBE_KERNELS, CUST_AICPU_KERNELS }; @@ -694,20 +659,6 @@ struct ModelPartitionTable { #define SIZE_OF_MODEL_PARTITION_TABLE(table) (sizeof(ModelPartitionTable) + sizeof(ModelPartitionMemInfo) * (table).num) -// Filter format -typedef enum tagDomiFilterFormat { - DOMI_FILTER_KCHW, // KCHW - DOMI_FILTER_HWCK, // HWCK - DOMI_FILTER_RESERVED -} domiFilterFormat_t; - -// Const data trans type -typedef enum tagDomiConstDataTransType { - DOMI_CONST_DATA_NOT_CHANGE = 0, // No action is required - DOMI_CONST_DATA_TRANS_MATMUL, // The const input to MatMul and needs to be transposed - DOMI_CONST_DATA_RESERVED -} domiConstDataTransType_t; - // mode of activation typedef enum tagDomiActivationMode { DOMI_ACTIVATION_SIGMOID = 0, // sigmoid @@ -727,170 +678,6 @@ typedef enum tagDomiActivationMode { DOMI_ACTIVATION_RESERVED } domiActivationMode_t; -// mode of batchnorm -typedef enum tagDomiBatchNormMode { - DOMI_BATCHNORM_PER_ACTIVATION = 0, // bnScale, bnBias tensor dims are 1xCxHxW - DOMI_BATCHNORM_SPATIAL, // bnScale, bnBias tensor dims are 1xCx1x1 - DOMI_BATCHNORM_RESERVED -} domiBatchNormMode_t; - -// eltwise mode -typedef enum tagDomiEltwiseMode { - DOMI_ELTWISE_PROD = 0, // prod - DOMI_ELTWISE_SUM, // sum - DOMI_ELTWISE_MAX, // max - DOMI_ELTWISE_RESERVED -} domiEltwiseMode_t; - -// mode of padding -typedef enum tagDomiPaddingMode { - DOMI_PADDING_CEIL = 0, // Default padding mode - DOMI_PADDING_DIRECTASSIGN, // Default padding mode: NOTSET - DOMI_PADDING_VALID, // VALID padding mode - DOMI_PADDING_SAME, // Padding values of 0 are always used - DOMI_PADDING_CEIL_NEW, // Padding values of 0 are always used - DOMI_PADDING_VALID_NEW, // Padding values of 0 are always used - DOMI_PADDING_SAME_NEW, // Padding values of 0 are always used - DOMI_PADDING_RESERVED -} domiPaddingMode_t; - -// algorithm of convolution forward -typedef enum tagDomiConvolutionFwdAlgo { - DOMI_CONVOLUTION_FWD_ALGO_GEMM = 0, // matrix gemm algo - DOMI_CONVOLUTION_FWD_ALGO_WINOGRAD, // Winograd Transform algo - DOMI_CONVOLUTION_FWD_ALGO_GEMM_ACCU_FLOAT32, // accumulate in L0c with FP32 - DOMI_CONVOLUTION_FWD_ALGO_RESERVED -} domiConvolutionFwdAlgo_t; - -typedef enum tagDomiFullConnectFwdAlgo { - DOMI_FULLCONNECT_FWD_ALGO_HALF = 0, // accumulate in L0c with FP16 - DOMI_FULLCONNECT_FWD_ALGO_FLOAT32 // accumulate in L0c with FP32 -} domiFullConnectFwdAlgo_t; - -typedef enum tagDomiPooingFwdAlgo { - DOMI_POOLING_FWD_ALGO_HALF = 0, // accumulate in L0c with FP16 - DOMI_POOLING_FWD_ALGO_FLOAT32 // accumulate in L0c with FP32 -} domiPooingFwdAlgo_t; - -// mode of convolution -typedef enum tagDomiConvolutionMode { - DOMI_CONV_CONVOLUTION = 0, // math convolution - DOMI_CONV_CROSS_CORRELATION, // cross-correlation convolution - DOMI_CONV_DECONVOLUTION, // deconvolution, also named transposed convolution - DOMI_CONV_MODE_DEPTHWISE, // depthwise convolution - DOMI_CONV_MODE_RESERVED -} domiConvolutionMode_t; - -// softmax mode -typedef enum tagDomiSoftmaxMode { - DOMI_SOFTMAX_MODE_INSTANCE = 0, // compute the softmax over all C, H, W for each N - DOMI_SOFTMAX_MODE_CHANNEL, // compute the softmax over all C for each H, W, N - DOMI_SOFTMAX_MODE_HEIGHT, // compute the softmax over all H for each N, C, W - DOMI_SOFTMAX_MODE_WIDTH, // compute the softmax over all W for each N, C, H - DOMI_SOFTMAX_MODE_RESERVED -} domiSoftmaxMode_t; - -// softmax algorithm -typedef enum tagDomiSoftmaxAlgo { - DOMI_SOFTMAX_FAST = 0, // straightforward implementation - DOMI_SOFTMAX_ACCURATE, // subtract max from every point to avoid overflow - DOMI_SOFTMAX_LOG, // perform the Log softmax operation to avoid overflow - DOMI_SOFTMAX_ACCURATE_FP32, - DOMI_SOFTMAX_RESERVED -} domiSoftmaxAlgo_t; - -// algorithm of convolution backward -typedef enum tagDomiConvolutionBwdAlgo { - DOMI_CONVOLUTION_BWD_ALGO_GEMM = 0, // matrix gemm algo - DOMI_CONVOLUTION_BWD_ALGO_WINOGRAD, // Winograd Transform algo - DOMI_CONVOLUTION_BWD_ALGO_RESERVED -} domiConvolutionBwdAlgo_t; - -// mode of pooling -typedef enum tagDomiPoolingMode { - DOMI_POOLING_MAX = 0, // max pooling - DOMI_POOLING_AVG, // average pooling - DOMI_POOLING_L2, // L2 pooling - DOMI_POOLING_RESERVED -} domiPoolingMode_t; - -// propagate Nan -typedef enum tagDomiNanPropagation { - DOMI_NAN_NOT_PROPAGATE = 0, // Nan numbers are not propagated - DOMI_NAN_PROPAGATE, // Nan numbers are propagated - DOMI_NAN_PROPAGATE_RESERVED -} domiNanPropagation_t; - -// mode of cropandresize -typedef enum tagDomiCropAndResizeMode { - DOMI_RESIZE_METHOD_BILINEAR = 0, // resize bilinear - DOMI_RESIZE_METHOD_NEAREST, // resize nearest - DOMI_RESIZE_RESERVED -} domiCropAndResizeMode_t; - -// yolo version -typedef enum tagDomiYoloVersion { DOMI_YOLO_V2 = 1, DOMI_YOLO_V3, DOMI_YOLO_TRSERVED } domiYoloVersion_t; - -typedef enum tagDomiRNNScopePassType { - DOMI_STATIC_BIDIRECTIONAL_RNN_GENERAL_PASS = 0, - DOMI_DYNAMIC_BIDIRECTIONAL_RNN_GENERAL_PASS, - DOMI_DYNAMIC_BIDIRECTIONAL_RNN_BIDAF_PASS -} domiRNNScopePassType; - -// RNNDataLayout -typedef enum tagDomiRNNDataLayout { - DOMI_RNN_ND_TBX = 0, // data[max_time,batch_size,Xt] - DOMI_RNN_ND_BTX, // data[batch_size,max_time,Xt] - DOMI_RNN_5D_TX1BX, // data[max_time,Xt,1,batch_size,Xt] - DOMI_RNN_5D_BX1TX, // dataa[batch_size,Xt,1,max_time,Xt] - DOMI_RNN_4DTBX1, - DOMI_ENN_DL_RESERVED -} domiRNNDataLayout_t; - -// RNNInputMode -typedef enum tagDomiRNNInputMode { DOMI_RNN_LINEAR_INPUT = 0, DOMI_RNN_SKIP_INPUT } domiRNNInputMode_t; - -// RNNDirectionMode -typedef enum tagDomiRNNDirectionMode { DOMI_RNN_UNIDIRECTIONAL = 0, DOMI_RNN_BIDIRECTIONAL } domiDirectionMode_t; - -typedef enum tagDomiPoolingCeilMode { DOMI_POOLING_FLOOR = 0, DOMI_POOLING_CEIL } domiPoolingCeilMode_t; - -// RNNMode -typedef enum tagDomiRNNActivationMode { - DOMI_RNN_ACTIVATION_SIGMOID = 0, // sigmoid - DOMI_RNN_ACTIVATION_TANH, // tanh - DOMI_RNN_ACTIVATION_RELU, // ReLU - DOMI_RNN_ACTIVATION_RELU1, // ReLU1 - DOMI_RNN_ACTIVATION_RELU6, // ReLU6 - DOMI_RNN_ACTIVATION_RESERVED -} domiRNNActivationMode_t; - -typedef enum tagDomiRNNLSTMOutMode { - DOMI_RNN_LSTM_OUT_SEPARATE = 0, - DOMI_RNN_LSTM_OUT_CONCAT, - DOMI_RNN_LSTM_OUT_RESERVED -} domiRNNLSTMOutPutMode_t; -typedef enum tagDomiRNNLSTMStateOutMode { - DOMI_RNN_LSTM_STATE_OUT_SEPARATE = 0, - DOMI_RNN_LSTM_STATE_OUT_CONCAT_ALL, - DOMI_RNN_LSTM_STATE_OUT_RESERVED -} domiRNNLSTMStateOutMode_t; - -typedef enum tagDomiRNNMode { - DOMI_RNN_RELU = 0, - DOMI_RNN_TANH, - DOMI_LSTM, - DOMI_GRU, - DOMI_RNN_MODE_RESERVED -} domiRNNMode_t; - -typedef enum tagDomiResizeBilinearMode { - DOMI_RESIZE_OUTPUT_DIM_BY_ZOOM_FACTOR = 0, // Output dimension specified by zoom factor - DOMI_RESIZE_OUTPUT_DIM_BY_SHRINK_FACTOR, // specified by shrink factor - DOMI_RESIZE_OUTPUT_DIM_EXPLICIT, // specified explicitly - DOMI_RESIZE_OUTPUT_DIM_RESERVED -} domiResizeOutputDimMode_t; - #pragma pack(1) // single-byte alignment // DUMP file struct struct FileHeader { diff --git a/inc/framework/common/util.h b/inc/framework/common/util.h index fac7e4ca..aa05997e 100644 --- a/inc/framework/common/util.h +++ b/inc/framework/common/util.h @@ -14,8 +14,8 @@ * limitations under the License. */ -#ifndef INC_FRAMEWORK_COMMON_UTIL_H_ -#define INC_FRAMEWORK_COMMON_UTIL_H_ +#ifndef AIR_INC_FRAMEWORK_COMMON_UTIL_H_ +#define AIR_INC_FRAMEWORK_COMMON_UTIL_H_ #include #include @@ -24,13 +24,14 @@ #include #include +#include "external/graph/types.h" #include "framework/common/debug/log.h" #include "framework/common/scope_guard.h" #include "framework/common/ge_inner_error_codes.h" #define GE_CHECK_POSITIVE_SIZE_RANGE(size) \ do { \ - if (size <= 0) { \ + if ((size) <= 0) { \ GELOGE(ge::FAILED, "param[%s] is not a positive number", #size); \ return PARAM_INVALID; \ } \ @@ -115,7 +116,7 @@ // Check if the parameter is null. If yes, return PARAM_INVALID and record the error #define GE_CHECK_NOTNULL(val) \ do { \ - if (val == nullptr) { \ + if ((val) == nullptr) { \ REPORT_INNER_ERROR("E19999", "Param:%s is nullptr, check invalid", #val); \ GELOGE(ge::FAILED, "[Check][Param:%s]null is invalid.", #val); \ return ge::PARAM_INVALID; \ @@ -125,7 +126,7 @@ // Check if the parameter is null. If yes, just return and record the error #define GE_CHECK_NOTNULL_JUST_RETURN(val) \ do { \ - if (val == nullptr) { \ + if ((val) == nullptr) { \ GELOGE(ge::FAILED, "param[%s] must not be null.", #val); \ return; \ } \ @@ -134,7 +135,7 @@ // Check whether the parameter is null. If so, execute the exec_expr expression and record the error log #define GE_CHECK_NOTNULL_EXEC(val, exec_expr) \ do { \ - if (val == nullptr) { \ + if ((val) == nullptr) { \ GELOGE(ge::FAILED, "param[%s] must not be null.", #val); \ exec_expr; \ } \ @@ -143,7 +144,7 @@ // Check whether the parameter is null. If yes, return directly and record the error log #define GE_RT_VOID_CHECK_NOTNULL(val) \ do { \ - if (val == nullptr) { \ + if ((val) == nullptr) { \ GELOGE(ge::FAILED, "param[%s] must not be null.", #val); \ return; \ } \ @@ -152,7 +153,7 @@ // Check if the parameter is null. If yes, return false and record the error log #define GE_RT_FALSE_CHECK_NOTNULL(val) \ do { \ - if (val == nullptr) { \ + if ((val) == nullptr) { \ GELOGE(ge::FAILED, "param[%s] must not be null.", #val); \ return false; \ } \ @@ -161,7 +162,7 @@ // Check if the parameter is out of bounds #define GE_CHECK_SIZE(size) \ do { \ - if (size == 0) { \ + if ((size) == 0) { \ GELOGE(ge::FAILED, "param[%s] is out of range", #size); \ return ge::PARAM_INVALID; \ } \ @@ -170,7 +171,7 @@ // Check if the value on the left is greater than or equal to the value on the right #define GE_CHECK_GE(lhs, rhs) \ do { \ - if (lhs < rhs) { \ + if ((lhs) < (rhs)) { \ GELOGE(ge::FAILED, "param[%s] is less than[%s]", #lhs, #rhs); \ return ge::PARAM_INVALID; \ } \ @@ -179,7 +180,7 @@ // Check if the value on the left is less than or equal to the value on the right #define GE_CHECK_LE(lhs, rhs) \ do { \ - if (lhs > rhs) { \ + if ((lhs) > (rhs)) { \ GELOGE(ge::FAILED, "param[%s] is greater than[%s]", #lhs, #rhs); \ return ge::PARAM_INVALID; \ } \ @@ -187,103 +188,36 @@ #define GE_DELETE_NEW_SINGLE(var) \ do { \ - if (var != nullptr) { \ - delete var; \ - var = nullptr; \ + if ((var) != nullptr) { \ + delete (var); \ + (var) = nullptr; \ } \ } while (false) #define GE_DELETE_NEW_ARRAY(var) \ do { \ - if (var != nullptr) { \ - delete[] var; \ - var = nullptr; \ + if ((var) != nullptr) { \ + delete[](var); \ + (var) = nullptr; \ } \ } while (false) #define GE_FREE_RT_LOG(addr) \ do { \ - if (addr != nullptr) { \ + if ((addr) != nullptr) { \ const rtError_t error = rtFree(addr); \ if (error != RT_ERROR_NONE) { \ GELOGE(RT_FAILED, "Call rtFree failed, error: %#x", error); \ } \ - addr = nullptr; \ + (addr) = nullptr; \ } \ } while (false) /** - * @ingroup domi_common - * @brief version of om.proto file - */ -static constexpr int32_t OM_PROTO_VERSION = 2; - -/** * Finding an Integer Ceiling Value Without Precision Loss */ #define CEIL(N, n) (((N) + (n)-1) / (n)) -namespace ge { -using google::protobuf::Message; - -/// -/// @ingroup domi_common -/// @brief Reads the proto structure from an array. -/// @param [in] data proto data to be read -/// @param [in] size proto data size -/// @param [out] proto Memory for storing the proto file -/// @return true success -/// @return false fail -/// -GE_FUNC_VISIBILITY bool ReadProtoFromArray(const void *data, int32_t size, Message *proto); - -/// -/// @ingroup domi_proto -/// @brief Reads the proto file in the text format. -/// @param [in] file path of proto file -/// @param [out] message Memory for storing the proto file -/// @return true success -/// @return false fail -/// -GE_FUNC_VISIBILITY bool ReadProtoFromText(const char *file, google::protobuf::Message *message); - -/// -/// @ingroup: domi_common -/// @brief: get length of file -/// @param [in] input_file: path of file -/// @return long： File length. If the file length fails to be obtained, the value -1 is returned. -/// -GE_FUNC_VISIBILITY extern long GetFileLength(const std::string &input_file); - -/// -/// @ingroup domi_common -/// @brief Reads all data from a binary file. -/// @param [in] file_name path of file -/// @param [out] buffer Output memory address, which needs to be released by the caller. -/// @param [out] length Output memory size -/// @return false fail -/// @return true success -/// -GE_FUNC_VISIBILITY bool ReadBytesFromBinaryFile(const char *const file_name, char **buffer, int32_t &length); - -GE_FUNC_VISIBILITY bool ReadBytesFromBinaryFile(const char *file_name, std::vector &buffer); - -/// -/// @ingroup domi_common -/// @brief Recursively Creating a Directory -/// @param [in] directory_path Path, which can be a multi-level directory. -/// @return 0 success -/// @return -1 fail -/// -GE_FUNC_VISIBILITY extern int32_t CreateDirectory(const std::string &directory_path); - -/// -/// @ingroup domi_common -/// @brief Obtains the current time string. -/// @return Time character string in the format ： %Y%m%d%H%M%S, eg: 20171011083555 -/// -GE_FUNC_VISIBILITY std::string CurrentTimeInStr(); - /// /// @ingroup domi_common /// @brief onverts Vector of a number to a string. @@ -294,7 +228,7 @@ template GE_FUNC_VISIBILITY std::string ToString(std::vector &v) { std::stringstream ss; ss << "["; - for (T x : v) { + for (const T x : v) { ss << x; ss << ", "; } @@ -314,7 +248,7 @@ template GE_FUNC_VISIBILITY std::string ToString(const google::protobuf::RepeatedField &rpd_field) { std::stringstream ss; ss << "["; - for (T x : rpd_field) { + for (const T x : rpd_field) { ss << x; ss << ", "; } @@ -343,6 +277,72 @@ GE_FUNC_VISIBILITY std::string ToString(const google::protobuf::RepeatedPtrField return str_ret; } +namespace ge { +/** + * @ingroup domi_common + * @brief version of om.proto file + */ +constexpr int32_t OM_PROTO_VERSION = 2; + +/// +/// @ingroup domi_common +/// @brief Reads the proto structure from an array. +/// @param [in] data proto data to be read +/// @param [in] size proto data size +/// @param [out] proto Memory for storing the proto file +/// @return true success +/// @return false fail +/// +GE_FUNC_VISIBILITY bool ReadProtoFromArray(const void *const data, const int32_t size, + google::protobuf::Message *const proto); + +/// +/// @ingroup domi_proto +/// @brief Reads the proto file in the text format. +/// @param [in] file path of proto file +/// @param [out] message Memory for storing the proto file +/// @return true success +/// @return false fail +/// +GE_FUNC_VISIBILITY bool ReadProtoFromText(const char_t *const file, google::protobuf::Message *const message); + +/// +/// @ingroup: domi_common +/// @brief: get length of file +/// @param [in] input_file: path of file +/// @return int64_t： File length. If the file length fails to be obtained, the value -1 is returned. +/// +GE_FUNC_VISIBILITY extern int64_t GetFileLength(const std::string &input_file); + +/// +/// @ingroup domi_common +/// @brief Reads all data from a binary file. +/// @param [in] file_name path of file +/// @param [out] buffer Output memory address, which needs to be released by the caller. +/// @param [out] length Output memory size +/// @return false fail +/// @return true success +/// +GE_FUNC_VISIBILITY bool ReadBytesFromBinaryFile(const char_t *const file_name, char_t **const buffer, int32_t &length); + +GE_FUNC_VISIBILITY bool ReadBytesFromBinaryFile(const char_t *file_name, std::vector &buffer); + +/// +/// @ingroup domi_common +/// @brief Recursively Creating a Directory +/// @param [in] directory_path Path, which can be a multi-level directory. +/// @return 0 success +/// @return -1 fail +/// +GE_FUNC_VISIBILITY extern int32_t CreateDirectory(const std::string &directory_path); + +/// +/// @ingroup domi_common +/// @brief Obtains the current time string. +/// @return Time character string in the format ： %Y%m%d%H%M%S, eg: 20171011083555 +/// +GE_FUNC_VISIBILITY std::string CurrentTimeInStr(); + /// /// @ingroup domi_common /// @brief Obtains the absolute time (timestamp) of the current system. @@ -366,7 +366,7 @@ GE_FUNC_VISIBILITY uint32_t GetCurrentSecondTimestap(); /// @param [in] b /// @return false: true: The result is within the normal int64 range. /// -GE_FUNC_VISIBILITY bool CheckInt64MulOverflow(int64_t a, int64_t b); +GE_FUNC_VISIBILITY bool CheckInt64MulOverflow(const int64_t a, const int64_t b); /// /// @ingroup domi_common @@ -374,7 +374,7 @@ GE_FUNC_VISIBILITY bool CheckInt64MulOverflow(int64_t a, int64_t b); /// @param [in] path of input file /// @param [out] Absolute path of a file. If the absolute path cannot be obtained, an empty string is returned /// -GE_FUNC_VISIBILITY std::string RealPath(const char *path); +GE_FUNC_VISIBILITY std::string RealPath(const char_t *path); /// /// @ingroup domi_common @@ -401,7 +401,7 @@ GE_FUNC_VISIBILITY bool CheckOutputPathValid(const std::string &file_path, const /// @param [in] str file path /// @param [out] result /// -GE_FUNC_VISIBILITY bool ValidateStr(const std::string &str, const std::string &mode); +GE_FUNC_VISIBILITY bool ValidateStr(const std::string &file_path, const std::string &mode); /// /// @ingroup domi_common @@ -411,7 +411,7 @@ GE_FUNC_VISIBILITY bool ValidateStr(const std::string &str, const std::string &m /// @return 0 success /// @return -1 fail /// -GE_FUNC_VISIBILITY Status CheckPath(const char *path, size_t length); +GE_FUNC_VISIBILITY Status CheckPath(const char_t *path, size_t length); } // namespace ge -#endif // INC_FRAMEWORK_COMMON_UTIL_H_ +#endif // AIR_INC_FRAMEWORK_COMMON_UTIL_H_ diff --git a/inc/framework/executor/ge_executor.h b/inc/framework/executor/ge_executor.h index c5df77df..51fcaeb3 100644 --- a/inc/framework/executor/ge_executor.h +++ b/inc/framework/executor/ge_executor.h @@ -32,6 +32,7 @@ namespace ge { class SingleOp; class DynamicSingleOp; +class GeRootModel; struct RunModelData { uint32_t index; // Data index @@ -227,6 +228,18 @@ class GE_FUNC_VISIBILITY GeExecutor { /// /// @ingroup ge + /// @brief Load task list from ModelData with queue. + /// @param [out] model_id: model id allocate from manager. + /// @param [in] root_model: Instance of GeRootModel. + /// @param [in] input_queue_ids: input queue ids create from user. + /// @param [in] output_queue_ids: input queue ids create from user. + /// @return: 0 for success / others for fail + /// + Status LoadModelWithQ(uint32_t &model_id, const std::shared_ptr &root_model, + const std::vector &input_queue_ids, const std::vector &output_queue_ids); + + /// + /// @ingroup ge /// @brief Synchronous execution of offline model(Do not create thread) /// @param [in] uint32_t model_id: Model ID to execute /// @param [in] void* stream: stream to execute diff --git a/inc/framework/generator/ge_generator.h b/inc/framework/generator/ge_generator.h index b49fa53b..9d85a40a 100644 --- a/inc/framework/generator/ge_generator.h +++ b/inc/framework/generator/ge_generator.h @@ -120,9 +120,9 @@ class GE_FUNC_VISIBILITY GeGenerator { GraphStage graph_stage = GraphStage::GRAPH_STAGE_RESERVED); bool CheckNoAicore(const ComputeGraphPtr &graph); void RemoveConst(const std::vector &inputs, std::vector &outputs); - Status CheckForSingleOp(OpDescPtr &op_desc, const std::vector &inputs, + Status CheckForSingleOp(const OpDescPtr &op_desc, const std::vector &inputs, const std::vector &outputs); - Status InferFormatForSingleOp(OpDescPtr &op_desc, Graph &graph); + Status InferFormatForSingleOp(const OpDescPtr &op_desc, const Graph &graph); using GeRootModelPtr = std::shared_ptr; Status SetModelNameForDump(const GeRootModelPtr &ge_root_model); diff --git a/inc/framework/memory/memory_api.h b/inc/framework/memory/memory_api.h index a04a1ebe..30ade3b7 100644 --- a/inc/framework/memory/memory_api.h +++ b/inc/framework/memory/memory_api.h @@ -17,11 +17,7 @@ #ifndef INC_FRAMEWORK_MEMORY_MEMORY_API_H_ #define INC_FRAMEWORK_MEMORY_MEMORY_API_H_ -#include -#include - #include "external/ge/ge_api_error_codes.h" -#include "graph/types.h" #include "runtime/mem.h" namespace ge { diff --git a/inc/framework/memory/memory_assigner.h b/inc/framework/memory/memory_assigner.h index 1ccbd785..be615426 100644 --- a/inc/framework/memory/memory_assigner.h +++ b/inc/framework/memory/memory_assigner.h @@ -33,7 +33,7 @@ class GE_FUNC_VISIBILITY MemoryAssigner { MemoryAssigner &operator=(const MemoryAssigner &) = delete; - Status AssignMemory(bool is_loop_graph, std::map &mem_offset, size_t &zero_copy_mem_size); + Status AssignMemory(std::map &mem_offset, size_t &zero_copy_mem_size); private: ge::ComputeGraphPtr compute_graph_; diff --git a/inc/framework/omg/model_tool.h b/inc/framework/omg/model_tool.h deleted file mode 100644 index 24554e65..00000000 --- a/inc/framework/omg/model_tool.h +++ /dev/null @@ -1,35 +0,0 @@ -/** - * Copyright (c) Huawei Technologies Co., Ltd. 2021. All rights reserved. - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -#ifndef INC_FRAMEWORK_OMG_MODEL_TOOL_H_ -#define INC_FRAMEWORK_OMG_MODEL_TOOL_H_ - -#include -#include - -#include "framework/common/debug/ge_log.h" -#include "proto/ge_ir.pb.h" - -namespace ge { -class GE_FUNC_VISIBILITY ModelTool { - public: - static Status GetModelInfoFromOm(const char *model_file, ge::proto::ModelDef &model_def, uint32_t &modeldef_size); - - static Status GetModelInfoFromPbtxt(const char *model_file, ge::proto::ModelDef &model_def); -}; -} // namespace ge - -#endif // INC_FRAMEWORK_OMG_MODEL_TOOL_H_ diff --git a/inc/framework/omg/omg.h b/inc/framework/omg/omg.h index 029b7a24..a0d6865e 100644 --- a/inc/framework/omg/omg.h +++ b/inc/framework/omg/omg.h @@ -64,7 +64,7 @@ GE_FUNC_VISIBILITY Status InitDomiOmgContext(const std::string &input_shape, con GE_FUNC_VISIBILITY Status ParseGraph(ge::Graph &graph, const std::map &atc_params, const char *model_file, const char *weights_file, domi::FrameworkType type, const char *op_conf = nullptr, const char *target = nullptr, - RunMode run_mode = GEN_OM_MODEL, bool is_dynamic_input = false); + RunMode run_mode = RunMode::GEN_OM_MODEL, bool is_dynamic_input = false); /** * @ingroup domi_omg diff --git a/inc/framework/omg/omg_inner_types.h b/inc/framework/omg/omg_inner_types.h index 3f7b5db1..1addd326 100644 --- a/inc/framework/omg/omg_inner_types.h +++ b/inc/framework/omg/omg_inner_types.h @@ -31,12 +31,7 @@ using domi::DOMI_TENSOR_ND; using domi::DOMI_TENSOR_RESERVED; using domi::domiTensorFormat_t; -using domi::FRAMEWORK_RESERVED; -using domi::FrameworkType; -using std::map; -using std::string; using std::unordered_map; -using std::vector; namespace ge { /** @@ -51,36 +46,13 @@ enum RunMode { DISPLAY_OM_INFO = 6 // display model info }; -/// -/// @ingroup domi_omg -/// @brief high-precision mode -/// -enum HighPrecisionMode { - // the FP16 high-precision function is disabled in common mode - HIGH_PRECISION_DEFAULT = 0, - - // high-precision mode, enabling FP16 high-precision mode (Convolution/FullConnect/AvgPooling are involved) - HIGH_PRECISION_FP16 = 1 -}; - -/// -/// @ingroup domi_omg -/// @brief description buffer data -/// -struct OMGBufferData { - void *data; - uint32_t length; -}; - struct OmgContext { - OmgContext() { - format = DOMI_TENSOR_ND; - } - domiTensorFormat_t format; + OmgContext() : format(domi::DOMI_TENSOR_ND) {} + domi::domiTensorFormat_t format; // format of the input specified by the command line - std::unordered_map input_nodes_format_map; - std::vector output_formats; + std::unordered_map input_nodes_format_map; + std::vector output_formats; // user-designate input dims std::vector>> user_input_dims; @@ -107,9 +79,9 @@ struct OmgContext { // net data nodes tensor names(caffe or onnx) std::vector data_tensor_names; // preferential format used by the entire network - domiTensorFormat_t net_format = DOMI_TENSOR_RESERVED; + domi::domiTensorFormat_t net_format = domi::DOMI_TENSOR_RESERVED; domi::FrameworkType type = domi::FRAMEWORK_RESERVED; - RunMode run_mode = ONLY_PRE_CHECK; + RunMode run_mode = RunMode::ONLY_PRE_CHECK; bool train_flag = false; std::string output_type; diff --git a/inc/framework/omg/parser/op_parser.h b/inc/framework/omg/parser/op_parser.h index 04731ff3..b511a8b0 100644 --- a/inc/framework/omg/parser/op_parser.h +++ b/inc/framework/omg/parser/op_parser.h @@ -50,7 +50,7 @@ class GE_FUNC_VISIBILITY OpParser { * @return SUCCESS * @return FAILED */ - virtual domi::Status ParseParams(const Message *op_src, ge::OpDescPtr &op_desc) = 0; + virtual domi::Status ParseParams(const google::protobuf::Message *op_src, ge::OpDescPtr &op_desc) = 0; /** * @ingroup domi_omg @@ -60,7 +60,7 @@ class GE_FUNC_VISIBILITY OpParser { * @return SUCCESS * @return FAILED */ - virtual domi::Status ParseParams(const Message *op_src, ge::Operator &op_dest) = 0; + virtual domi::Status ParseParams(const google::protobuf::Message *op_src, ge::Operator &op_dest) = 0; /** * @ingroup domi_omg @@ -70,7 +70,7 @@ class GE_FUNC_VISIBILITY OpParser { * @return SUCCESS * @return FAILED */ - virtual domi::Status ParseWeights(const Message *op_src, ge::NodePtr &node) = 0; + virtual domi::Status ParseWeights(const google::protobuf::Message *op_src, ge::NodePtr &node) = 0; /** * @ingroup domi_omg @@ -80,7 +80,7 @@ class GE_FUNC_VISIBILITY OpParser { * @return SUCCESS * @return FAILED */ - virtual domi::Status GetFormat(const Message *op_src, domi::domiTensorFormat_t &format) { + virtual domi::Status GetFormat(const google::protobuf::Message *op_src, domi::domiTensorFormat_t &format) { (void)op_src; // Indicates that the op does not provide a value for format format = domi::DOMI_TENSOR_RESERVED; diff --git a/inc/framework/omg/parser/parser_inner_ctx.h b/inc/framework/omg/parser/parser_inner_ctx.h index 969a94f8..5cf0d00b 100644 --- a/inc/framework/omg/parser/parser_inner_ctx.h +++ b/inc/framework/omg/parser/parser_inner_ctx.h @@ -29,8 +29,8 @@ namespace ge { struct ParserContext { // format of the input specified by the command line - std::unordered_map input_nodes_format_map; - std::vector output_formats; + std::unordered_map input_nodes_format_map; + std::vector output_formats; // user-designate input dims std::vector>> user_input_dims; std::map> input_dims; @@ -58,7 +58,7 @@ struct ParserContext { bool train_flag = false; domi::domiTensorFormat_t format = domi::DOMI_TENSOR_ND; domi::FrameworkType type = domi::FRAMEWORK_RESERVED; - RunMode run_mode = GEN_OM_MODEL; + RunMode run_mode = RunMode::GEN_OM_MODEL; // save caffe custom proto path, used by caffe parse std::string custom_proto_path; // save caffe proto path, used by caffe parse diff --git a/inc/framework/omg/version.h b/inc/framework/omg/version.h index 156fbc5c..53f155fa 100644 --- a/inc/framework/omg/version.h +++ b/inc/framework/omg/version.h @@ -19,8 +19,6 @@ #include #include -#include -#include #include "framework/common/debug/log.h" #include "framework/common/string_util.h" @@ -34,7 +32,7 @@ class GE_FUNC_VISIBILITY PlatformVersionManager { static Status GetPlatformVersion(std::string &ver) { ver = "1.11.z"; const std::vector version_splits = StringUtils::Split(ver, '.'); - GE_IF_BOOL_EXEC(version_splits.size() < 3, GELOGW("Read platform version error!"); return FAILED;); + GE_IF_BOOL_EXEC(version_splits.size() < 3U, GELOGW("Read platform version error!"); return FAILED;); GELOGI("Read current platform version: %s.", ver.c_str()); return SUCCESS; diff --git a/metadef b/metadef index 1d99928b..238e5308 160000 --- a/metadef +++ b/metadef @@ -1 +1 @@ -Subproject commit 1d99928bfcb02e45acc7db73e3ee57304ff1131a +Subproject commit 238e5308bf0191ac4598984ad6de80fa4a0cff4d diff --git a/third_party/fwkacllib/inc/cce/fwk_adpt_struct.h b/third_party/fwkacllib/inc/cce/fwk_adpt_struct.h index 5733d68f..ec92a036 100644 --- a/third_party/fwkacllib/inc/cce/fwk_adpt_struct.h +++ b/third_party/fwkacllib/inc/cce/fwk_adpt_struct.h @@ -21,7 +21,7 @@ namespace aicpu { namespace FWKAdapter { - +using char_t = char; // API RETURN CODE enum FWKAdptAPIRetCode { FWK_ADPT_SUCCESS = 0, // success @@ -63,6 +63,8 @@ enum FWKTaskExtInfoType { FWK_ADPT_EXT_BITMAP, FWK_ADPT_EXT_TOPIC_TYPE, FWK_ADPT_EXT_ASYNCWAIT, + FWK_ADPT_EXT_UNKNOWN_SHAPE_INPUT_INDEX, + FWK_ADPT_EXT_UNKNOWN_SHAPE_OUTPUT_INDEX, FWK_ADPT_EXT_INVALID }; @@ -113,7 +115,7 @@ struct StrFWKKernel { typedef StrFWKKernel FWKOperateParam; // Extent info ShapeAndType -const uint32_t kMaxShapeDims = 8; +const uint32_t kMaxShapeDims = 8U; #pragma pack(push, 1) struct ShapeAndType { int32_t type; @@ -122,13 +124,13 @@ struct ShapeAndType { #pragma pack(pop) // Extend info structure for extInfoAddr -const uint32_t kExtInfoHeadSize = 8; +const uint32_t kExtInfoHeadSize = 8U; #pragma pack(push, 1) struct ExtInfo { int32_t infoType; // extend type uint32_t infoLen; // length for infoMsg - char infoMsg[0]; // extend value + char_t infoMsg[0]; // extend value }; #pragma pack(pop) @@ -143,9 +145,9 @@ struct ResultSummary { #pragma pack(push, 1) struct AsyncWait { - uint8_t waitType; // wait type, FWK_ADPT_WAIT_TYPE_EVENT: event wait - uint32_t waitId; // wait id, GE refresh - uint32_t timeOut; // reserved + uint8_t waitType; // wait type, FWk_ADPT_WAIT_TPYE_EVENT: event wait + uint32_t waitId; // wait id, GE refresh + uint32_t timeOut; // reserved uint64_t reserved; }; #pragma pack(pop) diff --git a/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_typedef_linux.h b/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_typedef_linux.h index 9df5b9ce..9c6f6499 100644 --- a/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_typedef_linux.h +++ b/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_typedef_linux.h @@ -79,6 +79,9 @@ typedef long LONG; #define MMPA_THREAD_SCHED_OTHER SCHED_OTHER #define MMPA_THREAD_MIN_STACK_SIZE PTHREAD_STACK_MIN +#define MMPA_PATH_SEPARATOR_STR "/" +#define MMPA_PATH_SEPARATOR_CHAR '/' + #define MM_MUTEX_INITIALIZER PTHREAD_MUTEX_INITIALIZER #define MMPA_MAX_NI 19 diff --git a/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_typedef_win.h b/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_typedef_win.h index 1627d7a9..9f8a72cd 100644 --- a/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_typedef_win.h +++ b/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_typedef_win.h @@ -1,83 +1,86 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -#ifndef MMPA_TYPEDEF_WIN_H -#define MMPA_TYPEDEF_WIN_H - -#ifdef __cplusplus -#if __cplusplus -extern "C" { -#endif // __cpluscplus -#endif // __cpluscplus - -#ifndef FALSE -#define FALSE 0 -#endif - -#ifndef TRUE -#define TRUE 1 -#endif - -#define EN_OK 0 -#define EN_ERR 1 -#define EN_ERROR (-1) -#define EN_INVALID_PARAM (-2) -#define EN_TIMEOUT (-3) - -#define HANDLE_INVALID_VALUE (-1) -#define INVALID_SOCKET_HANDLE INVALID_SOCKET -#define MMPA_MEM_MAX_LEN (0x7fffffff) -#define MMPA_PROCESS_ERROR (0x7fffffff) - -#define MMPA_ONE_THOUSAND 1000 -#define MMPA_COMPUTER_BEGIN_YEAR 1900 -#define SUMMER_TIME_OR_NOT (-1) -#define MMPA_ZERO 0 -#define MMPA_VALUE_ONE 1 -#define MMPA_SOCKET_MAIN_EDITION 2 -#define MMPA_SOCKET_SECOND_EDITION 0 -#define MMPA_PIPE_BUF_SIZE 1024 -#define MMPA_MAX_SCANDIR_COUNT 1024 -#define MAX_IOVEC_SIZE 32 -#define MMPA_PIPE_COUNT 2 -#define MMPA_THREADNAME_SIZE 16 -#define MMPA_MIN_OS_NAME_SIZE (MAX_COMPUTERNAME_LENGTH + 1) -#define MMPA_MIN_OS_VERSION_SIZE 64 - -#define MMPA_MAX_NI 19 -#define MMPA_MIDDLE_NI 5 -#define MMPA_LOW_NI (-5) -#define MMPA_MIN_NI (-20) -#define MMPA_MAX_FILE 128 - -#define MMPA_MAX_THREAD_PIO 99 -#define MMPA_MIDDLE_THREAD_PIO 66 -#define MMPA_LOW_THREAD_PIO 33 -#define MMPA_MIN_THREAD_PIO 1 - -#define MMPA_THREAD_SCHED_RR 0 -#define MMPA_THREAD_SCHED_FIFO 0 -#define MMPA_THREAD_SCHED_OTHER 0 -#define MMPA_THREAD_MIN_STACK_SIZE 0 - -#define MM_MUTEX_INITIALIZER NULL - -#ifdef __cplusplus -#if __cplusplus -} -#endif // __cpluscplus -#endif // __cpluscplus -#endif // _MMPA_TYPEDEF_WIN_H_ +/** + * Copyright 2019-2020 Huawei Technologies Co., Ltd + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef MMPA_TYPEDEF_WIN_H +#define MMPA_TYPEDEF_WIN_H + +#ifdef __cplusplus +#if __cplusplus +extern "C" { +#endif // __cpluscplus +#endif // __cpluscplus + +#ifndef FALSE +#define FALSE 0 +#endif + +#ifndef TRUE +#define TRUE 1 +#endif + +#define EN_OK 0 +#define EN_ERR 1 +#define EN_ERROR (-1) +#define EN_INVALID_PARAM (-2) +#define EN_TIMEOUT (-3) + +#define HANDLE_INVALID_VALUE (-1) +#define INVALID_SOCKET_HANDLE INVALID_SOCKET +#define MMPA_MEM_MAX_LEN (0x7fffffff) +#define MMPA_PROCESS_ERROR (0x7fffffff) + +#define MMPA_ONE_THOUSAND 1000 +#define MMPA_COMPUTER_BEGIN_YEAR 1900 +#define SUMMER_TIME_OR_NOT (-1) +#define MMPA_ZERO 0 +#define MMPA_VALUE_ONE 1 +#define MMPA_SOCKET_MAIN_EDITION 2 +#define MMPA_SOCKET_SECOND_EDITION 0 +#define MMPA_PIPE_BUF_SIZE 1024 +#define MMPA_MAX_SCANDIR_COUNT 1024 +#define MAX_IOVEC_SIZE 32 +#define MMPA_PIPE_COUNT 2 +#define MMPA_THREADNAME_SIZE 16 +#define MMPA_MIN_OS_NAME_SIZE (MAX_COMPUTERNAME_LENGTH + 1) +#define MMPA_MIN_OS_VERSION_SIZE 64 + +#define MMPA_MAX_NI 19 +#define MMPA_MIDDLE_NI 5 +#define MMPA_LOW_NI (-5) +#define MMPA_MIN_NI (-20) +#define MMPA_MAX_FILE 128 + +#define MMPA_PATH_SEPARATOR_STR "\\" +#define MMPA_PATH_SEPARATOR_CHAR '\\' + +#define MMPA_MAX_THREAD_PIO 99 +#define MMPA_MIDDLE_THREAD_PIO 66 +#define MMPA_LOW_THREAD_PIO 33 +#define MMPA_MIN_THREAD_PIO 1 + +#define MMPA_THREAD_SCHED_RR 0 +#define MMPA_THREAD_SCHED_FIFO 0 +#define MMPA_THREAD_SCHED_OTHER 0 +#define MMPA_THREAD_MIN_STACK_SIZE 0 + +#define MM_MUTEX_INITIALIZER NULL + +#ifdef __cplusplus +#if __cplusplus +} +#endif // __cpluscplus +#endif // __cpluscplus +#endif // _MMPA_TYPEDEF_WIN_H_ diff --git a/third_party/fwkacllib/inc/ops/array_ops.h b/third_party/fwkacllib/inc/ops/array_ops.h index d56ac5bb..886d5382 100644 --- a/third_party/fwkacllib/inc/ops/array_ops.h +++ b/third_party/fwkacllib/inc/ops/array_ops.h @@ -201,6 +201,48 @@ REG_OP(Unique) .OP_END_FACTORY_REG(Unique) /** +*@brief Finds unique elements in a N-D tensor. \n + +*@par Inputs: +*Inputs "x" and "axis" are N-D vectors. +* @li x: A N-D tensor. \n + +*@par Attributes: +*sorted: An optional attr of type int, default to 1. +*axis: An optional attr of type int, default to -1000. +*return_idx: An optional attr of type bool, default to false. +*return_inverse: An optional attr of type bool, default to false. +*return_counts: An optional attr of type bool, default to false.\n + +*@par Outputs: +*@li y: "x" in the unique output "y". +*@li idx: A tensor the same size as "x". The index of each value of "x". +*@li inverse_idx: A tensor the same size as "x". The index of each value of "y". +*@li count: A tensor the same size as "x". The index of each value of "y". \n + +*@attention Constraints: +*UniqueV2 runs on the Ascend AI CPU, which delivers poor performance. \n + +*@par Third-party framework compatibility +*Compatible with the Pytorch operator unique. +*/ + +REG_OP(UniqueV2) + .INPUT(x, TensorType({ DT_INT8, DT_UINT8, DT_INT16, DT_UINT16, \ + DT_INT32, DT_INT64, DT_FLOAT16, DT_FLOAT, DT_DOUBLE, DT_STRING })) + .OUTPUT(y, TensorType({ DT_INT8, DT_UINT8, DT_INT16, DT_UINT16, \ + DT_INT32, DT_INT64, DT_FLOAT16, DT_FLOAT, DT_DOUBLE, DT_STRING })) + .OUTPUT(idx, TensorType({ DT_INT64 })) + .OUTPUT(inverse_idx, TensorType({ DT_INT64 })) + .OUTPUT(count, TensorType({ DT_INT64 })) + .ATTR(sorted, Int, 1) + .ATTR(axis, Int, -1000) + .ATTR(return_idx, Bool, false) + .ATTR(return_inverse, Bool, false) + .ATTR(return_counts, Bool, false) + .OP_END_FACTORY_REG(UniqueV2) + +/** *@brief Finds unique elements in a 1D tensor. \n *@par Inputs: diff --git a/third_party/fwkacllib/inc/ops/elewise_calculation_ops.h b/third_party/fwkacllib/inc/ops/elewise_calculation_ops.h index be201579..ae5b7831 100644 --- a/third_party/fwkacllib/inc/ops/elewise_calculation_ops.h +++ b/third_party/fwkacllib/inc/ops/elewise_calculation_ops.h @@ -3821,6 +3821,10 @@ REG_OP(CosineSimilarity) * @li max_grad_norm: A Tensor. Datatype is same as exp_avg. Shape (1, ).\n * @li global_grad_norm: A Tensor. Datatype is same as exp_avg. Shape (1, ).\n * @li weight_decay: A Tensor. Datatype is same as exp_avg. Shape (1, ).\n +* @li step_size: A Optional Tensor. Datatype is same as exp_avg. Shape (1, ).\n + +* @par Attributes: +* @li adam_mode: An optional bool. Defaults to "adam". \n *@par Outputs: *three inputs, including: @@ -3840,9 +3844,11 @@ REG_OP(ApplyAdamV2) .INPUT(max_grad_norm, TensorType({ DT_FLOAT, DT_FLOAT16 })) .INPUT(global_grad_norm, TensorType({ DT_FLOAT, DT_FLOAT16 })) .INPUT(weight_decay, TensorType({ DT_FLOAT, DT_FLOAT16 })) + .OPTIONAL_INPUT(step_size, TensorType({ DT_FLOAT, DT_FLOAT16 })) .OUTPUT(var, TensorType({ DT_FLOAT, DT_FLOAT16 })) .OUTPUT(m, TensorType({ DT_FLOAT, DT_FLOAT16 })) .OUTPUT(v, TensorType({ DT_FLOAT, DT_FLOAT16 })) + .ATTR(adam_mode, String, "adam") .OP_END_FACTORY_REG(ApplyAdamV2) } // namespace ge diff --git a/third_party/fwkacllib/inc/ops/nn_detect_ops.h b/third_party/fwkacllib/inc/ops/nn_detect_ops.h index 108a20a7..442d86a1 100644 --- a/third_party/fwkacllib/inc/ops/nn_detect_ops.h +++ b/third_party/fwkacllib/inc/ops/nn_detect_ops.h @@ -2076,7 +2076,7 @@ REG_OP(GIoUGrad) * trans: An optional attr, true for 'xyxyt', false for 'xywht'. *@par Outputs: -* overlaps: A 3D Tensor of type float16 or float32 with shape [B, N, K]. +* overlaps: A 3D Tensor of type float32 with shape [B, N, K]. *@attention Constraints: * In each batch, the invalid box cannot appear before the valid box. @@ -2087,6 +2087,38 @@ REG_OP(RotatedOverlaps) .OUTPUT(overlaps, TensorType({DT_FLOAT})) .ATTR(trans, Bool, false) .OP_END_FACTORY_REG(RotatedOverlaps) + +/** +*@brief RotatedIou . \n + +*@par Inputs: +*@li boxes : data of grad increment, a 3D Tensor of type float32 with +* shape (B, 5, N). "N" indicates the number of boxes, and the value +* "5" refers to [x1, y1, x2, y2, theta] or [x, y, w, h, theta]. +*@li query_boxes: Bounding boxes, a 3D Tensor of type float32 with +* shape (B, 5, K). "K" indicates the number of boxes, and the value +* "5" refers to [x1, y1, x2, y2, theta] or [x, y, w, h, theta]. + +*@par Attributes: +*@li trans: An optional attr, true for 'xyxyt', false for 'xywht'. +*@li mode: An optional attr, a character string with the value range of ['iou', 'iof'], +* only support 'iou' now. +*@li is_cross: Cross calculation when it is True, and one-to-one calculation when it is False. + +*@par Outputs: +* iou: A 3D Tensor of float32 with shape [B, N, K]. + +*@attention Constraints: +* In each batch, the invalid box cannot appear before the valid box. +*/ +REG_OP(RotatedIou) + .INPUT(boxes, TensorType({DT_FLOAT})) + .INPUT(query_boxes, TensorType({DT_FLOAT})) + .OUTPUT(iou, TensorType({DT_FLOAT})) + .ATTR(trans, Bool, false) + .ATTR(mode, String, "iou") + .ATTR(is_cross, Bool, true) + .OP_END_FACTORY_REG(RotatedIou) } // namespace ge #endif // OPS_BUILT_IN_OP_PROTO_INC_NN_DETECT_OPS_H_ diff --git a/third_party/fwkacllib/inc/ops/nn_norm_ops.h b/third_party/fwkacllib/inc/ops/nn_norm_ops.h index 9ce7abfd..562c17f8 100644 --- a/third_party/fwkacllib/inc/ops/nn_norm_ops.h +++ b/third_party/fwkacllib/inc/ops/nn_norm_ops.h @@ -1644,6 +1644,36 @@ REG_OP(NormalizeBatch) .REQUIRED_ATTR(normalize_type, String) .ATTR(epsilon, Float, 0.00001) .OP_END_FACTORY_REG(NormalizeBatch) + +/** +*@brief GroupNorm and Reul operator +* calculating: x, gamma, beta +* y = relu(gamma*((x - mean) / np.sqrt(variance + 0.001)) + beta) + +*@par Inputs: +*Three inputs, including: +* @li x: A Tensor. Must be one of the following types: float16, float32. +* @li gamma: A Tensor. Must be one of the following types: float16, float32. +* @li beta: A Tensor. Must be one of the following types: float16, float32 . \n + +*@par Attributes: +* @li num_groups: A require attribute, the type is int32. +* @li eps: A optional attribute, the type is float32. Defaults to 0.00001. \n + +*@par Outputs: +*One outputs, including: +* @li y: A Tensor. Must be one of the following types: float16, float32. +*@par Restrictions: +*Warning: THIS FUNCTION IS EXPERIMENTAL. Please do not use/ +*/ +REG_OP(GroupNormRelu) + .INPUT(x, TensorType({DT_FLOAT, DT_FLOAT16})) + .INPUT(gamma, TensorType({DT_FLOAT, DT_FLOAT16})) + .INPUT(beta, TensorType({DT_FLOAT, DT_FLOAT16})) + .OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16})) + .REQUIRED_ATTR(num_groups, Int) + .ATTR(eps, Float, 0.00001) + .OP_END_FACTORY_REG(GroupNormRelu) } // namespace ge #endif // OPS_BUILT_IN_OP_PROTO_INC_NN_NORM_OPS_H_ diff --git a/third_party/fwkacllib/inc/ops/nonlinear_fuc_ops.h b/third_party/fwkacllib/inc/ops/nonlinear_fuc_ops.h index d5960395..b4f39f2c 100644 --- a/third_party/fwkacllib/inc/ops/nonlinear_fuc_ops.h +++ b/third_party/fwkacllib/inc/ops/nonlinear_fuc_ops.h @@ -159,7 +159,22 @@ REG_OP(FastGelu) .INPUT(x, TensorType({DT_FLOAT16, DT_FLOAT})) .OUTPUT(y, TensorType({DT_FLOAT16, DT_FLOAT})) .OP_END_FACTORY_REG(FastGelu) +/** +*@brief Computes the for the fast_gelu_v2 of "x" . \n + +*@par Inputs: +*One input, including: +*x: A Tensor. Must be one of the following types: float16, float32 +*@par Outputs: +*y: A Tensor. Has the same type as "x". +*@par Third-party framework compatibility +*Compatible with the TensorFlow operator FastGeluV2 +*/ +REG_OP(FastGeluV2) + .INPUT(x, TensorType({DT_FLOAT16, DT_FLOAT})) + .OUTPUT(y, TensorType({DT_FLOAT16, DT_FLOAT})) + .OP_END_FACTORY_REG(FastGeluV2) /** *@brief Computes the gradient for the fast_gelu of "x" . \n diff --git a/third_party/fwkacllib/inc/ops/ocr_ops.h b/third_party/fwkacllib/inc/ops/ocr_ops.h index a5755659..efaa7aa7 100644 --- a/third_party/fwkacllib/inc/ops/ocr_ops.h +++ b/third_party/fwkacllib/inc/ops/ocr_ops.h @@ -81,6 +81,7 @@ REG_OP(OCRRecognitionPreHandle) .OUTPUT(imgs, TensorType({DT_UINT8})) .OUTPUT(imgs_relation, TensorType({DT_INT32})) .OUTPUT(imgs_lang, TensorType({DT_INT32})) + .OUTPUT(imgs_piece_fillers, TensorType({DT_INT32})) .ATTR(batch_size, Int, 8) .ATTR(data_format, String, "NHWC") .ATTR(pad_mode, String, "REPLICATE") diff --git a/third_party/fwkacllib/inc/ops/reduce_ops.h b/third_party/fwkacllib/inc/ops/reduce_ops.h index fa572b66..a273953c 100644 --- a/third_party/fwkacllib/inc/ops/reduce_ops.h +++ b/third_party/fwkacllib/inc/ops/reduce_ops.h @@ -1326,6 +1326,43 @@ REG_OP(ReduceMeanVariance) .ATTR(axes, ListInt, {}) .ATTR(keep_dims, Bool, true) .OP_END_FACTORY_REG(ReduceMeanVariance) + +/** +* @brief Calculates the standard deviation or the variance of Tensors with the average value. + +* @par Inputs: +* Two inputs, including: +* @li x: A Tensor. Must be one of the following types: float16, float32. \n +* @li mean: A Tensor. It's the mean of X. Has the same shape and type as "x" \n + +* @par Attributes: +* Four Attributes, including: +* @li dim: An listint. \n +* @li if_std: An optional bool. Defaults to "False" +* If "True", Calculate the standard deviation +* If "False", Calculate the variance +* @li unbiased: An optional bool. Defaults to "True". +* If "True", Use Bessel Correction. +* If "False", Do not use Bessel Correction. \n +* @li keepdim: An optional bool. Defaults to "False". +* If "True", Keep the original tensor dimension. +* If "False", Do not keep the original tensor dimension. \n + +* @par Outputs: +* @li output_var: A Tensor. It's the standard deviation or the variance of X. Has the same type as "x". + +* @par Third-party framework compatibility +* Compatible with the Pytorch operator Var_mean. +*/ +REG_OP(ReduceStdV2Update) + .INPUT(x, TensorType({DT_FLOAT,DT_FLOAT16})) + .INPUT(mean, TensorType({DT_FLOAT,DT_FLOAT16})) + .OUTPUT(output_var, TensorType({DT_FLOAT,DT_FLOAT16})) + .REQUIRED_ATTR(dim, ListInt) + .ATTR(if_std, Bool, false) + .ATTR(unbiased, Bool, true) + .ATTR(keepdim, Bool, false) + .OP_END_FACTORY_REG(ReduceStdV2Update) } //namespace ge #endif // OPS_BUILT_IN_OP_PROTO_INC_REDUCE_OPS_H_ diff --git a/third_party/fwkacllib/inc/runtime/base.h b/third_party/fwkacllib/inc/runtime/base.h index aa98ed9a..3cc73a2c 100644 --- a/third_party/fwkacllib/inc/runtime/base.h +++ b/third_party/fwkacllib/inc/runtime/base.h @@ -98,11 +98,11 @@ typedef struct rtExceptionInfo { uint32_t tid; uint32_t deviceid; uint32_t retcode; -} rtExceptionInfo; +} rtExceptionInfo_t; typedef void (*rtErrorCallback)(rtExceptionType); -typedef void (*rtTaskFailCallback)(rtExceptionInfo *exceptionInfo); +typedef void (*rtTaskFailCallback)(rtExceptionInfo_t *exceptionInfo); typedef void (*rtDeviceStateCallback)(uint32_t devId, bool isOpen); diff --git a/third_party/fwkacllib/inc/runtime/context.h b/third_party/fwkacllib/inc/runtime/context.h index e920a939..378470c7 100644 --- a/third_party/fwkacllib/inc/runtime/context.h +++ b/third_party/fwkacllib/inc/runtime/context.h @@ -140,7 +140,7 @@ RTS_API rtError_t rtSetGroup(int32_t groupId); * @param [in] groupid count * @return RT_ERROR_NONE for ok, errno for failed */ -RTS_API rtError_t rtGetGroupInfo(int32_t groupId, rtGroupInfo_t *groupInfo, uint32_t count); +RTS_API rtError_t rtGetGroupInfo(int32_t groupId, rtGroupInfo_t *groupInfo, uint32_t cnt); /** * @ingroup diff --git a/third_party/fwkacllib/inc/runtime/dev.h b/third_party/fwkacllib/inc/runtime/dev.h index e5b741c7..bce6625b 100644 --- a/third_party/fwkacllib/inc/runtime/dev.h +++ b/third_party/fwkacllib/inc/runtime/dev.h @@ -94,11 +94,11 @@ typedef enum tagGetDevMsgType { /** * @ingroup dvrt_dev * @brief get total device number. - * @param [in|out] count the device number + * @param [in|out] cnt the device number * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtGetDeviceCount(int32_t *count); +RTS_API rtError_t rtGetDeviceCount(int32_t *cnt); /** * @ingroup dvrt_dev * @brief get device ids @@ -338,7 +338,7 @@ RTS_API rtError_t rtSetTSDevice(uint32_t tsId); * @return RT_ERROR_NONE for ok * @return RT_ERROR_DRV_ERR for can not get run mode */ -RTS_API rtError_t rtGetRunMode(rtRunMode *mode); +RTS_API rtError_t rtGetRunMode(rtRunMode *runMode); /** * @ingroup dvrt_dev diff --git a/third_party/fwkacllib/inc/runtime/dvfsprofile.h b/third_party/fwkacllib/inc/runtime/dvfsprofile.h index 7cb8c8a6..2813d73b 100644 --- a/third_party/fwkacllib/inc/runtime/dvfsprofile.h +++ b/third_party/fwkacllib/inc/runtime/dvfsprofile.h @@ -23,11 +23,11 @@ typedef enum dvfsProfileMode { /** * @ingroup dvrt_dvfsprofile * @brief Set the performance mode of the device - * @param [in] mode dvfsProfileMode + * @param [in] profMode dvfsProfileMode * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtSetDvfsProfile(DvfsProfileMode mode); +RTS_API rtError_t rtSetDvfsProfile(DvfsProfileMode profMode); /** * @ingroup dvrt_dvfsprofile diff --git a/third_party/fwkacllib/inc/runtime/kernel.h b/third_party/fwkacllib/inc/runtime/kernel.h index 08860ccd..a68bf9f6 100644 --- a/third_party/fwkacllib/inc/runtime/kernel.h +++ b/third_party/fwkacllib/inc/runtime/kernel.h @@ -519,13 +519,13 @@ RTS_API rtError_t rtConfigureCall(uint32_t numBlocks, rtSmDesc_t *smDesc, rtStre /** * @ingroup rt_kernel * @brief setup argment for next rtLaunch in current thread - * @param [in] arg argment address for kernel function + * @param [in] args argment address for kernel function * @param [in] size argment size * @param [in] offset argment table offset * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtSetupArgument(const void *arg, uint32_t size, uint32_t offset); +RTS_API rtError_t rtSetupArgument(const void *args, uint32_t size, uint32_t offset); /** * @ingroup rt_kernel @@ -544,11 +544,11 @@ RTS_API rtError_t rtLaunch(const void *stubFunc); * @param [in] ptr host memory * @param [in] size host memory size * @param [in] flag reserved. set to 0 - * @param [out] arg returned arg. used for next kernel's arg. + * @param [out] args returned arg. used for next kernel's arg. * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtKernelConfigTransArg(const void *ptr, uint64_t size, uint32_t flag, void **arg); +RTS_API rtError_t rtKernelConfigTransArg(const void *ptr, uint64_t size, uint32_t flag, void **args); /** * @ingroup rt_kernel diff --git a/third_party/fwkacllib/inc/runtime/rt_mem_queue.h b/third_party/fwkacllib/inc/runtime/rt_mem_queue.h index 49a272f0..8cb39535 100644 --- a/third_party/fwkacllib/inc/runtime/rt_mem_queue.h +++ b/third_party/fwkacllib/inc/runtime/rt_mem_queue.h @@ -222,24 +222,24 @@ RTS_API rtError_t rtMemQueueInit(int32_t devId); /** * @ingroup rt_mem_queue - * @brief enqueu mbuf + * @brief enqueue memBuf * @param [in] devId the logical device id * @param [in] qid queue id - * @param [in] mbuf enqueue mbuf + * @param [in] memBuf enqueue memBuf * @return RT_ERROR_NONE for ok */ -RTS_API rtError_t rtMemQueueEnQueue(int32_t devId, uint32_t qid, void *mbuf); +RTS_API rtError_t rtMemQueueEnQueue(int32_t devId, uint32_t qid, void *memBuf); /** * @ingroup rt_mem_queue - * @brief enqueu mbuf + * @brief dequeue memBuf * @param [in] devId the logical device id * @param [in] qid queue id - * @param [out] mbuf dequeue mbuf + * @param [out] memBuf dequeue memBuf * @return RT_ERROR_NONE for ok */ -RTS_API rtError_t rtMemQueueDeQueue(int32_t devId, uint32_t qid, void **mbuf); +RTS_API rtError_t rtMemQueueDeQueue(int32_t devId, uint32_t qid, void **memBuf); /** * @ingroup rt_mem_queue @@ -350,47 +350,47 @@ RTS_API rtError_t rtMbufInit(rtMemBuffCfg_t *cfg); /** * @ingroup rt_mem_queue * @brief alloc buff -* @param [out] buff: buff addr alloced +* @param [out] memBuf: buff addr alloced * @param [in] size: The amount of memory space requested * @return RT_ERROR_NONE for ok */ -RTS_API rtError_t rtMbufAlloc(rtMbufPtr_t *mbuf, uint64_t size); +RTS_API rtError_t rtMbufAlloc(rtMbufPtr_t *memBuf, uint64_t size); /** * @ingroup rt_mem_queue * @brief free buff -* @param [in] buff: buff addr to be freed +* @param [in] memBuf: buff addr to be freed * @return RT_ERROR_NONE for ok */ -RTS_API rtError_t rtMbufFree(rtMbufPtr_t mbuf); +RTS_API rtError_t rtMbufFree(rtMbufPtr_t memBuf); /** * @ingroup rt_mem_queue * @brief get Data addr of Mbuf -* @param [in] mbuf: Mbuf addr +* @param [in] memBuf: Mbuf addr * @param [out] buf: Mbuf data addr * @return RT_ERROR_NONE for ok */ -RTS_API rtError_t rtMbufGetBuffAddr(rtMbufPtr_t mbuf, void **buf); +RTS_API rtError_t rtMbufGetBuffAddr(rtMbufPtr_t memBuf, void **buf); /** * @ingroup rt_mem_queue * @brief get total Buffer size of Mbuf -* @param [in] mbuf: Mbuf addr +* @param [in] memBuf: Mbuf addr * @param [out] totalSize: total buffer size of Mbuf * @return RT_ERROR_NONE for ok */ -RTS_API rtError_t rtMbufGetBuffSize(rtMbufPtr_t mbuf, uint64_t *totalSize); +RTS_API rtError_t rtMbufGetBuffSize(rtMbufPtr_t memBuf, uint64_t *totalSize); /** * @ingroup rt_mem_queue * @brief Get the address and length of its user_data from the specified Mbuf -* @param [in] mbuf: Mbuf addr +* @param [in] memBuf: Mbuf addr * @param [out] priv: address of its user_data * @param [out] size: length of its user_data * @return RT_ERROR_NONE for ok */ -RTS_API rtError_t rtMbufGetPrivInfo (rtMbufPtr_t mbuf, void **priv, uint64_t *size); +RTS_API rtError_t rtMbufGetPrivInfo (rtMbufPtr_t memBuf, void **priv, uint64_t *size); // mem group typedef struct { diff --git a/third_party/fwkacllib/inc/runtime/rt_model.h b/third_party/fwkacllib/inc/runtime/rt_model.h index 2807a705..ef9da8ee 100644 --- a/third_party/fwkacllib/inc/runtime/rt_model.h +++ b/third_party/fwkacllib/inc/runtime/rt_model.h @@ -44,6 +44,7 @@ typedef enum tagModelTaskType { RT_MODEL_TASK_PROFILER_TRACE_EX, RT_MODEL_TASK_FFTS_TASK, RT_MODEL_TASK_FFTS_PLUS_TASK, + RT_MODEL_TASK_DSA_TASK, } rtModelTaskType_t; typedef enum tagModelStreamType { diff --git a/third_party/fwkacllib/inc/runtime/rt_stars_define.h b/third_party/fwkacllib/inc/runtime/rt_stars_define.h index 260809af..8447bb72 100644 --- a/third_party/fwkacllib/inc/runtime/rt_stars_define.h +++ b/third_party/fwkacllib/inc/runtime/rt_stars_define.h @@ -32,6 +32,37 @@ typedef struct tagStarsSqeHeader { uint16_t taskId; } rtStarsSqeHeader_t; +typedef struct tagStarsDsaSqe { + // 0-7 bytes + rtStarsSqeHeader_t sqeHeader; + // 8-11 bytes + uint32_t start : 1; + uint32_t functionType : 3; + uint32_t dataType : 3; + uint32_t algoType : 3; + uint32_t paramVldBitmap : 5; + uint32_t paramAddrValBitmap : 7; + uint32_t reserved0 : 10; + // 12-15 bytes + uint16_t sqeIndex; + uint8_t kernelCredit; + uint8_t reserved1; + // 16-31 bytes + uint32_t dsaCfgResultAddrLow; + uint32_t dsaCfgResultAddrHigh; + uint32_t dsaCfgStateAddrLow; + uint32_t dsaCfgStateAddrHigh; + // 32-47 bytes + uint32_t dsaCfgParamAddrLow; + uint32_t dsaCfgParamAddrHigh; + uint32_t dsaCfgSeedLow; + uint32_t dsaCfgSeedHigh; + // 48-63 bytes + uint32_t dsaCfgNumberLow; + uint32_t dsaCfgNumberHigh; + uint32_t reserved2[2]; +} rtStarsDsaSqe_t; + // ffts+ type typedef enum tagFftsPlusType { RT_FFTS_PLUS_TYPE_RES1 = 2, // Reserved diff --git a/third_party/fwkacllib/inc/toolchain/prof_acl_api.h b/third_party/fwkacllib/inc/toolchain/prof_acl_api.h index 8ecd5f14..16aaf26b 100644 --- a/third_party/fwkacllib/inc/toolchain/prof_acl_api.h +++ b/third_party/fwkacllib/inc/toolchain/prof_acl_api.h @@ -1,17 +1,8 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. +/* + * Copyright (c) Huawei Technologies Co., Ltd. 2019-2021. All rights reserved. + * Description: handle perf data + * Author: xp + * Create: 2019-10-13 */ #ifndef MSPROFILER_API_PROF_ACL_API_H_ @@ -25,6 +16,7 @@ #define PROF_L2CACHE 0x00000010ULL #define PROF_HCCL_TRACE 0x00000020ULL #define PROF_TRAINING_TRACE 0x00000040ULL +#define PROF_MSPROFTX 0x00000080ULL // system profilinig switch #define PROF_CPU 0x00010000ULL @@ -42,11 +34,13 @@ #define PROF_SCHEDULE_TRACE 0x0000010000000ULL #define PROF_AIVECTORCORE_METRICS 0x0000020000000ULL #define PROF_SUBTASK_TIME 0x0000040000000ULL - -#define PROF_TASK_TRACE 0x0000005000062ULL +#define PROF_OP_DETAIL 0x0000080000000ULL #define PROF_MODEL_LOAD 0x8000000000000000ULL +#define PROF_TASK_TRACE (PROF_MODEL_EXECUTE | PROF_RUNTIME_TRACE | PROF_TRAINING_TRACE | \ + PROF_HCCL_TRACE | PROF_TASK_TIME) + // DataTypeConfig MASK #define PROF_ACL_API_MASK 0x00000001ULL #define PROF_TASK_TIME_MASK 0x00000002ULL @@ -55,6 +49,7 @@ #define PROF_L2CACHE_MASK 0x00000010ULL #define PROF_HCCL_TRACE_MASK 0x00000020ULL #define PROF_TRAINING_TRACE_MASK 0x00000040ULL +#define PROF_MSPROFTX_MASK 0x00000080ULL // system profilinig mask #define PROF_CPU_MASK 0x00010000ULL @@ -72,6 +67,7 @@ #define PROF_SCHEDULE_TRACE_MASK 0x0000010000000ULL #define PROF_AIVECTORCORE_METRICS_MASK 0x0000020000000ULL #define PROF_SUBTASK_TIME_MASK 0x0000040000000ULL +#define PROF_OP_DETAIL_MASK 0x0000080000000ULL #define PROF_MODEL_LOAD_MASK 0x8000000000000000ULL @@ -135,6 +131,15 @@ MSVP_PROF_API Status aclgrphProfGraphUnSubscribe(const uint32_t graphId); * @retval 0 for failed */ MSVP_PROF_API size_t aclprofGetGraphId(const void *opInfo, size_t opInfoLen, uint32_t index); + +/** +* @ingroup AscendCL +* @brief set stamp pay load +* +* +* @retval void +*/ +MSVP_PROF_API int aclprofSetStampPayload(void *stamp, const int32_t type, void *value); #ifdef __cplusplus } #endif diff --git a/third_party/fwkacllib/inc/toolchain/prof_callback.h b/third_party/fwkacllib/inc/toolchain/prof_callback.h index 5f5a151b..6398d075 100644 --- a/third_party/fwkacllib/inc/toolchain/prof_callback.h +++ b/third_party/fwkacllib/inc/toolchain/prof_callback.h @@ -1,17 +1,8 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. +/* + * Copyright (c) Huawei Technologies Co., Ltd. 2019-2021. All rights reserved. + * Description: handle perf data + * Author: xp + * Create: 2019-10-13 */ #ifndef MSPROFILER_PROF_CALLBACK_H_ @@ -76,7 +67,8 @@ enum MsprofReporterModuleId { MSPROF_MODULE_HCCL, // HCCL MSPROF_MODULE_ACL, // AclModule MSPROF_MODULE_FRAMEWORK, // Framework - MSPROF_MODULE_RUNTIME // runtime + MSPROF_MODULE_RUNTIME, // runtime + MSPROF_MODULE_MSPROF // msprofTx }; /**