feat(dnn/atlas): add atlas stub

GitOrigin-RevId: c63294378e49732a7a41b0b6672a070a518c7882

CMakeLists.txt

@@ -507,8 +507,8 @@ endif ()
 
 
 if(MGE_WITH_ATLAS)
-    include(cmake/aclrt.cmake)
-    list(APPEND MGE_ATLAS_LIBS libascendcl)
+    add_subdirectory(dnn/atlas-stub)
+    list(APPEND MGE_ATLAS_LIBS atlas-stub)
     set(MGE_ATLAS_LIBS "${MGE_ATLAS_LIBS}")
     set(MGB_ATLAS ${MGE_WITH_ATLAS})
 endif()

dnn/atlas-stub/CMakeLists.txt

+add_library(atlas-stub STATIC src/libatlas-wrap.cpp)
+target_include_directories(atlas-stub PUBLIC $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>)
+install(TARGETS atlas-stub EXPORT ${MGE_EXPORT_TARGETS})
+
+add_library(acl-cblas STATIC src/libacl_cblas-wrap.cpp)
+target_include_directories(acl-cblas PUBLIC $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>)

dnn/atlas-stub/include/acl/acl.h

+/**
+* @file acl.h
+*
+* Copyright (C) Huawei Technologies Co., Ltd. 2019-2020. All Rights Reserved.
+*
+* This program is distributed in the hope that it will be useful,
+* but WITHOUT ANY WARRANTY; without even the implied warranty of
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+*/
+
+#ifndef INC_EXTERNAL_ACL_ACL_H_
+#define INC_EXTERNAL_ACL_ACL_H_
+
+#include "acl_rt.h"
+#include "acl_op.h"
+#include "acl_mdl.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @ingroup AscendCL
+ * @brief acl initialize
+ *
+ * @par Restriction
+ * The aclInit interface can be called only once in a process
+ * @param configPath [IN]    the config path,it can be NULL
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclInit(const char *configPath);
+
+/**
+ * @ingroup AscendCL
+ * @brief acl finalize
+ *
+ * @par Restriction
+ * Need to call aclFinalize before the process exits.
+ * After calling aclFinalize,the services cannot continue to be used normally.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclFinalize();
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // INC_EXTERNAL_ACL_ACL_H_

dnn/atlas-stub/include/acl/acl_base.h

+/**
+* @file acl_base.h
+*
+* Copyright (C) Huawei Technologies Co., Ltd. 2019-2020. All Rights Reserved.
+*
+* This program is distributed in the hope that it will be useful,
+* but WITHOUT ANY WARRANTY; without even the implied warranty of
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+*/
+
+#ifndef INC_EXTERNAL_ACL_ACL_BASE_H_
+#define INC_EXTERNAL_ACL_ACL_BASE_H_
+
+#include <stdint.h>
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef FUNC_VISIBILITY
+#define ACL_FUNC_VISIBILITY __attribute__((visibility("default")))
+#else
+#define ACL_FUNC_VISIBILITY
+#endif
+
+typedef void *aclrtStream;
+typedef void *aclrtEvent;
+typedef void *aclrtContext;
+typedef int aclError;
+typedef uint16_t aclFloat16;
+typedef struct aclDataBuffer aclDataBuffer;
+typedef struct aclTensorDesc aclTensorDesc;
+
+const int ACL_ERROR_NONE = 0;
+
+const int ACL_ERROR_INVALID_PARAM = 100000;
+const int ACL_ERROR_UNINITIALIZE = 100001;
+const int ACL_ERROR_REPEAT_INITIALIZE = 100002;
+const int ACL_ERROR_INVALID_FILE = 100003;
+const int ACL_ERROR_WRITE_FILE = 100004;
+const int ACL_ERROR_INVALID_FILE_SIZE = 100005;
+const int ACL_ERROR_PARSE_FILE = 100006;
+const int ACL_ERROR_FILE_MISSING_ATTR = 100007;
+const int ACL_ERROR_FILE_ATTR_INVALID = 100008;
+const int ACL_ERROR_INVALID_DUMP_CONFIG = 100009;
+const int ACL_ERROR_INVALID_PROFILING_CONFIG = 100010;
+const int ACL_ERROR_INVALID_MODEL_ID = 100011;
+const int ACL_ERROR_DESERIALIZE_MODEL = 100012;
+const int ACL_ERROR_PARSE_MODEL = 100013;
+const int ACL_ERROR_READ_MODEL_FAILURE = 100014;
+const int ACL_ERROR_MODEL_SIZE_INVALID = 100015;
+const int ACL_ERROR_MODEL_MISSING_ATTR = 100016;
+const int ACL_ERROR_MODEL_INPUT_NOT_MATCH = 100017;
+const int ACL_ERROR_MODEL_OUTPUT_NOT_MATCH = 100018;
+const int ACL_ERROR_MODEL_NOT_DYNAMIC = 100019;
+const int ACL_ERROR_OP_TYPE_NOT_MATCH = 100020;
+const int ACL_ERROR_OP_INPUT_NOT_MATCH = 100021;
+const int ACL_ERROR_OP_OUTPUT_NOT_MATCH = 100022;
+const int ACL_ERROR_OP_ATTR_NOT_MATCH = 100023;
+const int ACL_ERROR_OP_NOT_FOUND = 100024;
+const int ACL_ERROR_OP_LOAD_FAILED = 100025;
+const int ACL_ERROR_UNSUPPORTED_DATA_TYPE = 100026;
+const int ACL_ERROR_FORMAT_NOT_MATCH = 100027;
+const int ACL_ERROR_BIN_SELECTOR_NOT_REGISTERED = 100028;
+const int ACL_ERROR_KERNEL_NOT_FOUND = 100029;
+const int ACL_ERROR_BIN_SELECTOR_ALREADY_REGISTERED = 100030;
+const int ACL_ERROR_KERNEL_ALREADY_REGISTERED = 100031;
+const int ACL_ERROR_INVALID_QUEUE_ID = 100032;
+const int ACL_ERROR_REPEAT_SUBSCRIBE = 100033;
+const int ACL_ERROR_STREAM_NOT_SUBSCRIBE = 100034;
+const int ACL_ERROR_THREAD_NOT_SUBSCRIBE = 100035;
+const int ACL_ERROR_WAIT_CALLBACK_TIMEOUT = 100036;
+const int ACL_ERROR_REPEAT_FINALIZE = 100037;
+const int ACL_ERROR_NOT_STATIC_AIPP = 100038;
+
+const int ACL_ERROR_BAD_ALLOC = 200000;
+const int ACL_ERROR_API_NOT_SUPPORT = 200001;
+const int ACL_ERROR_INVALID_DEVICE = 200002;
+const int ACL_ERROR_MEMORY_ADDRESS_UNALIGNED = 200003;
+const int ACL_ERROR_RESOURCE_NOT_MATCH = 200004;
+const int ACL_ERROR_INVALID_RESOURCE_HANDLE = 200005;
+const int ACL_ERROR_FEATURE_UNSUPPORTED = 200006;
+
+const int ACL_ERROR_STORAGE_OVER_LIMIT = 300000;
+
+const int ACL_ERROR_INTERNAL_ERROR = 500000;
+const int ACL_ERROR_FAILURE = 500001;
+const int ACL_ERROR_GE_FAILURE = 500002;
+const int ACL_ERROR_RT_FAILURE = 500003;
+const int ACL_ERROR_DRV_FAILURE = 500004;
+const int ACL_ERROR_PROFILING_FAILURE = 500005;
+
+typedef enum {
+    ACL_DT_UNDEFINED = -1,
+    ACL_FLOAT = 0,
+    ACL_FLOAT16 = 1,
+    ACL_INT8 = 2,
+    ACL_INT32 = 3,
+    ACL_UINT8 = 4,
+    ACL_INT16 = 6,
+    ACL_UINT16 = 7,
+    ACL_UINT32 = 8,
+    ACL_INT64 = 9,
+    ACL_UINT64 = 10,
+    ACL_DOUBLE = 11,
+    ACL_BOOL = 12,
+} aclDataType;
+
+typedef enum {
+    ACL_FORMAT_UNDEFINED = -1,
+    ACL_FORMAT_NCHW = 0,
+    ACL_FORMAT_NHWC = 1,
+    ACL_FORMAT_ND = 2,
+    ACL_FORMAT_NC1HWC0 = 3,
+    ACL_FORMAT_FRACTAL_Z = 4,
+    ACL_FORMAT_FRACTAL_NZ = 29,
+} aclFormat;
+
+typedef enum {
+    ACL_DEBUG = 0,
+    ACL_INFO = 1,
+    ACL_WARNING = 2,
+    ACL_ERROR = 3,
+} aclLogLevel;
+
+/**
+ * @ingroup AscendCL
+ * @brief Converts data of type aclFloat16 to data of type float
+ *
+ * @param value [IN]   Data to be converted
+ * @retval Transformed data
+ */
+ACL_FUNC_VISIBILITY float aclFloat16ToFloat(aclFloat16 value);
+
+/**
+ * @ingroup AscendCL
+ * @brief Converts data of type float to data of type aclFloat16
+ *
+ * @param value [IN]   Data to be converted
+ * @retval Transformed data
+ */
+ACL_FUNC_VISIBILITY aclFloat16 aclFloatToFloat16(float value);
+
+/**
+ * @ingroup AscendCL
+ * @brief create data of aclDataBuffer
+ *
+ * @param data [IN]    pointer to data
+ * @li Need to be managed by the user,
+ *  call aclrtMalloc interface to apply for memory,
+ *  call aclrtFree interface to release memory
+ * @param size [IN]    size of data in bytes
+ * @retval pointer to created instance. nullptr if run out of memory
+ *
+ * @see aclrtMalloc | aclrtFree
+ */
+ACL_FUNC_VISIBILITY aclDataBuffer *aclCreateDataBuffer(void *data, size_t size);
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy data of aclDataBuffer
+ *
+ * @par Function
+ *  Only the aclDataBuffer type data is destroyed here.
+ *  The memory of the data passed in when the aclDataDataBuffer interface
+ *  is called to create aclDataBuffer type data must be released by the user
+ * @param  dataBuffer [IN]   pointer to the aclDataBuffer
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclCreateDataBuffer
+ */
+ACL_FUNC_VISIBILITY aclError aclDestroyDataBuffer(const aclDataBuffer *dataBuffer);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data address from aclDataBuffer
+ *
+ * @param dataBuffer [IN]    pointer to the data of aclDataBuffer
+ * @retval data address
+ */
+ACL_FUNC_VISIBILITY void *aclGetDataBufferAddr(const aclDataBuffer *dataBuffer);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data size of aclDataBuffer
+ *
+ * @param  dataBuffer [IN]    pointer to the data of aclDataBuffer
+ * @retval data size
+ */
+ACL_FUNC_VISIBILITY uint32_t aclGetDataBufferSize(const aclDataBuffer *dataBuffer);
+
+/**
+ * @ingroup AscendCL
+ * @brief get size of aclDataType
+ *
+ * @param  dataType [IN]    aclDataType data the size to get
+ * @retval size of the aclDataType
+ */
+ACL_FUNC_VISIBILITY size_t aclDataTypeSize(aclDataType dataType);
+
+// interfaces of tensor desc
+/**
+ * @ingroup AscendCL
+ * @brief create data aclTensorDesc
+ *
+ * @param  dataType [IN]    Data types described by tensor
+ * @param  numDims [IN]     the number of dimensions of the shape
+ * @param  dims [IN]        the size of the specified dimension
+ * @param  format [IN]      tensor format
+ * @retval aclTensorDesc pointer.
+ * @retval nullptr if param is invalid or run out of memory
+ */
+ACL_FUNC_VISIBILITY aclTensorDesc *aclCreateTensorDesc(aclDataType dataType,
+                                                       int numDims,
+                                                       const int64_t *dims,
+                                                       aclFormat format);
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy data aclTensorDesc
+ *
+ * @param desc [IN]     pointer to the data of aclTensorDesc to destroy
+ */
+ACL_FUNC_VISIBILITY void aclDestroyTensorDesc(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data type specified by the tensor description
+ *
+ * @param desc [IN]        pointer to the instance of aclTensorDesc
+ * @retval data type specified by the tensor description.
+ * @retval ACL_DT_UNDEFINED if description is null
+ */
+ACL_FUNC_VISIBILITY aclDataType aclGetTensorDescType(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get data format specified by the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @retval data format specified by the tensor description.
+ * @retval ACL_FORMAT_UNDEFINED if description is null
+ */
+ACL_FUNC_VISIBILITY aclFormat aclGetTensorDescFormat(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get tensor size specified by the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @retval data size specified by the tensor description.
+ * @retval 0 if description is null
+ */
+ACL_FUNC_VISIBILITY size_t aclGetTensorDescSize(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get element count specified by the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @retval element count specified by the tensor description.
+ * @retval 0 if description is null
+ */
+ACL_FUNC_VISIBILITY size_t aclGetTensorDescElementCount(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief get number of dims specified by the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @retval number of dims specified by the tensor description.
+ * @retval 0 if description is null
+ */
+ACL_FUNC_VISIBILITY size_t aclGetTensorDescNumDims(const aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief Get the size of the specified dim in the tensor description
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @param  index [IN]       index of dims, start from 0.
+ * @retval dim specified by the tensor description and index.
+ * @retval -1 if description or index is invalid
+ */
+ACL_FUNC_VISIBILITY int64_t aclGetTensorDescDim(const aclTensorDesc *desc, size_t index);
+
+/**
+ * @ingroup AscendCL
+ * @brief set tensor description name
+ *
+ * @param desc [IN]        pointer to the instance of aclTensorDesc
+ * @param name [IN]        tensor description name
+ */
+ACL_FUNC_VISIBILITY void aclSetTensorDescName(aclTensorDesc *desc, const char *name);
+
+/**
+ * @ingroup AscendCL
+ * @brief get tensor description name
+ *
+ * @param  desc [IN]        pointer to the instance of aclTensorDesc
+ * @retval tensor description name.
+ * @retval empty string if description is null
+ */
+ACL_FUNC_VISIBILITY const char *aclGetTensorDescName(aclTensorDesc *desc);
+
+/**
+ * @ingroup AscendCL
+ * @brief Convert the format in the source aclTensorDesc according to
+ * the specified dstFormat to generate a new target aclTensorDesc.
+ * The format in the source aclTensorDesc remains unchanged.
+ *
+ * @param  srcDesc [IN]     pointer to the source tensor desc
+ * @param  dstFormat [IN]   destination format
+ * @param  dstDesc [OUT] pointer to the pointer to the destination tensor desc
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclTransTensorDescFormat(const aclTensorDesc *srcDesc, aclFormat dstFormat,
+    aclTensorDesc **dstDesc);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the storage format specified by the tensor description
+ *
+ * @param  desc [IN|OUT]     pointer to the instance of aclTensorDesc
+ * @param  format [IN]       the storage format
+ * @retval ACL_ERROR_NONE    The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorStorageFormat(aclTensorDesc *desc, aclFormat format);
+
+/**
+ * @ingroup AscendCL
+ * @brief Set the storage shape specified by the tensor description
+ *
+ * @param  desc [IN|OUT]      pointer to the instance of aclTensorDesc
+ * @param  numDims [IN]       the number of dimensions of the shape
+ * @param  dims [IN]          the size of the specified dimension
+ * @retval ACL_ERROR_NONE     The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclSetTensorStorageShape(aclTensorDesc *desc, int numDims, const int64_t *dims);
+
+/**
+ * @ingroup AscendCL
+ * @brief an interface for users to output  APP logs
+ *
+ * @param logLevel [IN]    the level of current log
+ * @param func [IN]        the function where the log is located
+ * @param file [IN]        the file where the log is located
+ * @param line [IN]        Number of source lines where the log is located
+ * @param fmt [IN]         the format of current log
+ * @param ... [IN]         the value of current log
+ */
+ACL_FUNC_VISIBILITY void aclAppLog(aclLogLevel logLevel, const char *func, const char *file, uint32_t line,
+    const char *fmt, ...);
+
+#define ACL_APP_LOG(level, fmt, ...) \
+    aclAppLog(level, __FUNCTION__, __FILE__, __LINE__, fmt, ##__VA_ARGS__)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // INC_EXTERNAL_ACL_ACL_BASE_H_

dnn/atlas-stub/include/acl/acl_op.h

+/**
+* @file acl_op.h
+*
+* Copyright (C) Huawei Technologies Co., Ltd. 2019-2020. All Rights Reserved.
+*
+* This program is distributed in the hope that it will be useful,
+* but WITHOUT ANY WARRANTY; without even the implied warranty of
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+*/
+#ifndef INC_EXTERNAL_ACL_ACL_OP_H_
+#define INC_EXTERNAL_ACL_ACL_OP_H_
+
+#include "acl_base.h"
+#include "acl_rt.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct aclopHandle aclopHandle;
+typedef struct aclopAttr aclopAttr;
+typedef struct aclopKernelDesc aclopKernelDesc;
+
+typedef void (*aclDataDeallocator)(void *data, size_t length);
+
+const int ACL_COMPILE_FLAG_BIN_SELECTOR = 1;
+
+typedef enum aclEngineType {
+    ACL_ENGINE_SYS,
+    ACL_ENGINE_AICORE,
+    ACL_ENGINE_VECTOR,
+} aclopEngineType;
+
+/**
+ * @ingroup AscendCL
+ * @brief Set base directory that contains single op models
+ *
+ * @par Restriction
+ * The aclopSetModelDir interface can be called only once in a process.
+ * @param  modelDir [IN]   path of the directory
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetModelDir(const char *modelDir);
+
+/**
+ * @ingroup AscendCL
+ * @brief load single op models from memory
+ *
+ * @par Restriction
+ * The aclopLoad interface can be called more than one times in a process.
+ * @param model [IN]        address of single op models
+ * @param modelSize [IN]    size of single op models
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopLoad(const void *model,  size_t modelSize);
+
+/**
+ * @ingroup AscendCL
+ * @brief create data of type aclopAttr
+ *
+ * @retval pointer to created instance.
+ * @retval nullptr if run out of memory
+ */
+ACL_FUNC_VISIBILITY aclopAttr *aclopCreateAttr();
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy data of typ aclopAttr
+ *
+ * @param attr [IN]   pointer to the instance of aclopAttr
+ */
+ACL_FUNC_VISIBILITY void aclopDestroyAttr(const aclopAttr *attr);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is bool
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ *                         false if attrValue is 0, true otherwise.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrBool(aclopAttr *attr, const char *attrName, uint8_t attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is int64_t
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrInt(aclopAttr *attr, const char *attrName, int64_t attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is float
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrFloat(aclopAttr *attr, const char *attrName, float attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is string
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param attrValue [IN]   attribute value
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrString(aclopAttr *attr, const char *attrName, const char *attrValue);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of bools
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values. false if attrValue is 0, true otherwise.
+ * @param values [IN]      pointer to values
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListBool(aclopAttr *attr, const char *attrName, int numValues,
+    const uint8_t *values);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of ints
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values
+ * @param values [IN]      pointer to values
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListInt(aclopAttr *attr, const char *attrName, int numValues,
+    const int64_t *values);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of floats
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values
+ * @param values [IN]      pointer to values
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListFloat(aclopAttr *attr, const char *attrName, int numValues,
+    const float *values);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of strings
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numValues [IN]   number of values
+ * @param values [IN]      pointer to values
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListString(aclopAttr *attr, const char *attrName, int numValues,
+    const char **values);
+
+/**
+ * @ingroup AscendCL
+ * @brief set an attribute. the type of the attribute is list of list of ints
+ *
+ * @param attr [IN]        pointer to the instance of aclopAttr
+ * @param attrName [IN]    attribute name
+ * @param numLists [IN]    number of lists
+ * @param numValues [IN]   pointer to number of values of each list
+ * @param values [IN]      pointer to values
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetAttrListListInt(aclopAttr *attr,
+                                                     const char *attrName,
+                                                     int numLists,
+                                                     const int *numValues,
+                                                     const int64_t *const values[]);
+
+/**
+ * @ingroup AscendCL
+ * @brief Load and execute the specified operator asynchronously
+ *
+ * @par Restriction
+ * @li The input and output organization of each operator is different,
+ * and the application needs to organize the operator strictly
+ * according to the operator input and output parameters when calling.
+ * @li When the user calls aclopExecute,
+ * the ACL finds the corresponding task according to the optype,
+ * the description of the input tesnsor,
+ * the description of the output tesnsor, and attr, and issues the execution.
+ * @param opType [IN]      type of op
+ * @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 [OUT]    pointer to array of output buffers
+ * @param attr [IN]        pointer to instance of aclopAttr.
+ *                         may pass nullptr if the op has no attribute
+ * @param stream [IN]      stream
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopExecute(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,
+                                          aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a instance of aclopHandle.
+ *
+ * @param opType [IN]      type of op
+ * @param numInputs [IN]   number of inputs
+ * @param inputDesc [IN]   pointer to array of input tensor descriptions
+ * @param numOutputs [IN]  number of outputs
+ * @param outputDesc [IN]  pointer to array of output tensor descriptions
+ * @param opAttr [IN]      pointer to instance of aclopAttr.
+ *                         may pass nullptr if the op has no attribute
+ * @param handle [OUT]     pointer to the pointer to the handle
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCreateHandle(const char *opType,
+                                               int numInputs,
+                                               const aclTensorDesc *const inputDesc[],
+                                               int numOutputs,
+                                               const aclTensorDesc *const outputDesc[],
+                                               const aclopAttr *opAttr,
+                                               aclopHandle **handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief destroy aclopHandle instance
+ *
+ * @param handle [IN]   pointer to the instance of aclopHandle
+ */
+ACL_FUNC_VISIBILITY void aclopDestroyHandle(aclopHandle *handle);
+
+/**
+ * @ingroup AscendCL
+ * @brief execute an op with the handle.
+ *        can save op model matching cost compared with aclopExecute
+ *
+ * @param handle [IN]      pointer to the instance of aclopHandle.
+ *                         The aclopCreateHandle interface has been called
+ *                         in advance to create aclopHandle type data.
+ * @param numInputs [IN]   number of inputs
+ * @param inputs [IN]      pointer to array of input buffers.
+ *                         The aclCreateDataBuffer interface has been called
+ *                         in advance to create aclDataBuffer type data.
+ * @param numOutputs [IN]  number of outputs
+ * @param outputs [IN]     pointer to array of output buffers
+ * @param stream [IN]      stream
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclopCreateHandle | aclCreateDataBuffer
+ */
+ACL_FUNC_VISIBILITY aclError aclopExecWithHandle(aclopHandle *handle,
+                                                 int numInputs,
+                                                 const aclDataBuffer *const inputs[],
+                                                 int numOutputs,
+                                                 aclDataBuffer *const outputs[],
+                                                 aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief cast data type
+ *
+ * @param srcDesc [IN]     source tensor desc
+ * @param srcBuffer [IN]   source tensor buffer
+ * @param dstDesc [IN]     destination tensor desc
+ * @param dstBuffer [OUT]  destination tensor buffer
+ * @param truncate [IN]    do not truncate if value is 0, truncate otherwise
+ * @param stream [IN]      stream
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCast(const aclTensorDesc *srcDesc,
+                                       const aclDataBuffer *srcBuffer,
+                                       const aclTensorDesc *dstDesc,
+                                       aclDataBuffer *dstBuffer,
+                                       uint8_t truncate,
+                                       aclrtStream stream);
+
+/**
+ * @ingroup AscendCL
+ * @brief create a handle for casting datatype
+ *
+ * @param srcDesc [IN]     source tensor desc
+ * @param dstDesc [IN]     destination tensor desc
+ * @param truncate [IN]    do not truncate if value is 0, truncate otherwise
+ * @param handle [IN]     pointer to the pointer to the handle
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopCreateHandleForCast(aclTensorDesc *srcDesc,
+                                                      aclTensorDesc *dstDesc,
+                                                      uint8_t truncate,
+                                                      aclopHandle **handle);
+
+
+/**
+ * @ingroup AscendCL
+ * @brief create kernel
+ *
+ * @param opType [IN]           op type
+ * @param kernelId [IN]         kernel id
+ * @param kernelName [IN]       kernel name
+ * @param binData [IN]          kernel bin data
+ * @param binSize [IN]          kernel bin size
+ * @param enginetype [IN]       enigne type
+ * @param deallocator [IN]      callback function for deallocating bin data,
+ *                              null if bin data to be deallocated by caller
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclopCompile
+ */
+ACL_FUNC_VISIBILITY aclError aclopCreateKernel(const char *opType,
+                                               const char *kernelId,
+                                               const char *kernelName,
+                                               void *binData,
+                                               int binSize,
+                                               aclopEngineType enginetype,
+                                               aclDataDeallocator deallocator);
+
+
+/**
+ * @ingroup AscendCL
+ * @brief create kernel
+ *
+ * @param numInputs [IN]            number of inputs
+ * @param inputDesc [IN]            pointer to array of input tensor descriptions
+ * @param numOutputs [IN]           number of outputs
+ * @param outputDesc [IN]           pointer to array of output tensor descriptions
+ * @param opAttr [IN]               pointer to instance of aclopAttr
+ * @param aclopKernelDesc [IN]      pointer to instance of aclopKernelDesc
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+typedef aclError (*aclopCompileFunc)(int numInputs,
+                                     const aclTensorDesc *const inputDesc[],
+                                     int numOutputs,
+                                     const aclTensorDesc *const outputDesc[],
+                                     const aclopAttr *opAttr,
+                                     aclopKernelDesc *aclopKernelDesc);
+
+/**
+ * @ingroup AscendCL
+ * @brief register compile function
+ *
+ * @param opType [IN]         op type
+ * @param func [IN]           compile function
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclopUnregisterCompileFunc
+ */
+ACL_FUNC_VISIBILITY aclError aclopRegisterCompileFunc(const char *opType, aclopCompileFunc func);
+
+/**
+ * @ingroup AscendCL
+ * @brief unregister compile function
+ *
+ * @param opType [IN]         op type
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopUnregisterCompileFunc(const char *opType);
+
+/**
+ * @ingroup AscendCL
+ * @brief set kernel args
+ *
+ * @param kernelDesc [IN]               pointer to instance of aclopKernelDesc
+ * @param kernelId [IN]                 kernel id
+ * @param blockDim [IN]                 block dim
+ * @param args [IN]                     args
+ * @param argSize [IN]                  size in bytes of args
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetKernelArgs(aclopKernelDesc *kernelDesc,
+                                                const char *kernelId,
+                                                uint32_t blockDim,
+                                                const void *args,
+                                                uint32_t argSize);
+
+/**
+ * @ingroup AscendCL
+ * @brief set workspace sizes
+ *
+ * @param kernelDesc [IN]               pointer to instance of aclopKernelDesc
+ * @param numWorkspaces [IN]            number of workspaces
+ * @param workspaceSizes [IN]           pointer to array of sizes of workspaces
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopSetKernelWorkspaceSizes(aclopKernelDesc *kernelDesc, int numWorkspaces,
+    size_t *workspaceSizes);
+
+/**
+ * @ingroup AscendCL
+ * @brief compile op with dynamic shape
+ *
+ * @param opType [IN]           op type
+ * @param numInputs [IN]        number of inputs
+ * @param inputDesc [IN]        pointer to array of input tensor descriptions
+ * @param numOutputs [IN]       number of outputs
+ * @param outputDesc [IN]       pointer to array of output tensor descriptions
+ * @param attr [IN]             pointer to instance of aclopAttr.
+ *                              may pass nullptr if the op has no attribute
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclopUpdateParams(const char *opType,
+                                               int numInputs,
+                                               const aclTensorDesc *const inputDesc[],
+                                               int numOutputs,
+                                               const aclTensorDesc *const outputDesc[],
+                                               const aclopAttr *attr);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // INC_EXTERNAL_ACL_ACL_OP_H_

dnn/atlas-stub/include/acl/ops/acl_fv.h

+/* *
+ * @file acl_fv.h
+ *
+ * Copyright (C) Huawei Technologies Co., Ltd. 2019-2020. All Rights Reserved.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ */
+#ifndef INC_EXTERNAL_ACL_OPS_ACL_RETR_H_
+#define INC_EXTERNAL_ACL_OPS_ACL_RETR_H_
+
+#include "../acl.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct aclfvFeatureInfo aclfvFeatureInfo;
+typedef struct aclfvRepoRange aclfvRepoRange;
+typedef struct aclfvQueryTable aclfvQueryTable;
+typedef struct aclfvSearchInput aclfvSearchInput;
+typedef struct aclfvSearchResult aclfvSearchResult;
+
+// search operation type
+enum aclfvSearchType {
+    SEARCH_1_N, // 1:N operation type
+    SEARCH_N_M  // N:M operation type
+};
+
+/* *
+ * @ingroup AscendCL
+ * @brief Create fv feature info.
+ * @param id0 [IN]: The first level library id0
+ * @param id1 [IN]: Secondary library id1
+ * @param offset [IN]: The offset of the first feature in the library
+ * @param featureLen [IN]: Single feature length
+ * @param featureCount [IN]: Single feature count
+ * @param featureData [IN/OUT]: Feature value list
+ * @param featureDataLen [IN]: Feature value list length
+ * @retval null for failed.
+ * @retval OtherValues success.
+ */
+ACL_FUNC_VISIBILITY aclfvFeatureInfo *aclfvCreateFeatureInfo(uint32_t id0, uint32_t id1, uint32_t offset,
+    uint32_t featureLen, uint32_t featureCount, uint8_t *featureData, uint32_t featureDataLen);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Destroy fv feature info.
+ *
+ * @par Function
+ * Can only destroy fv feature info information created
+ * through aclfvCreateFeatureInfo interface.
+ * @param featureInfo [IN/OUT]     fv feature info.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateFeatureInfo
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroyFeatureInfo(aclfvFeatureInfo *featureInfo);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Create fv repo range.
+ * @param id0Min [IN]: id0 start value
+ * @param id0Min [IN]: id0 max
+ * @param id1Min [IN]: id0 start value
+ * @param id1Max [IN]: id1 max
+ * @retval null for failed. OtherValues success
+ */
+ACL_FUNC_VISIBILITY aclfvRepoRange *aclfvCreateRepoRange(uint32_t id0Min, uint32_t id0Max, uint32_t id1Min,
+    uint32_t id1Max);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Destroy fv repo range.
+ *
+ * @par Function
+ * Can only destroy fv repo range information created
+ * through aclfvCreateRepoRange interface.
+ * @param repoRange [IN/OUT]     fv repo range.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateRepoRange
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroyRepoRange(aclfvRepoRange *repoRange);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Create query table.
+ * @param queryCnt [IN]: Number of tables, the maximum number is 6
+ * @param tableLen [IN]: Single table length, table length is 32KB
+ * @param tableData [IN/OUT]: Feature value list
+ * @param tableDataLen [IN]: The length of memory requested by the featureData
+ * pointer
+ * @retval null for failed. OtherValues success
+ */
+ACL_FUNC_VISIBILITY aclfvQueryTable *aclfvCreateQueryTable(uint32_t queryCnt, uint32_t tableLen, uint8_t *tableData,
+    uint32_t tableDataLen);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Destroy query table.
+ *
+ * @par Function
+ * Can only destroy query table information created
+ * through aclfvCreateQueryTable interface.
+ * @param queryTable [IN/OUT]     query table.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateQueryTable
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroyQueryTable(aclfvQueryTable *queryTable);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Create search input.
+ * @param queryTable [IN/OUT]: query table
+ * @param repoRange [IN/OUT]: query repo range
+ * @param topk [IN]: query topk
+ * @retval null for failed. OtherValues success
+ */
+ACL_FUNC_VISIBILITY aclfvSearchInput *aclfvCreateSearchInput(aclfvQueryTable *queryTable, aclfvRepoRange *repoRange,
+    uint32_t topk);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Destroy search input.
+ *
+ * @par Function
+ * Can only destroy search input information created
+ * through aclfvCreateSearchInput interface.
+ * @param searchInput [IN/OUT]     search input.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateSearchInput
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroySearchInput(aclfvSearchInput *searchInput);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Create search result.
+ * @param queryCnt [IN]: Retrieve the number of features
+ * @param resultNum [IN/OUT]: The number of search results for each feature, the
+ * number is queryCnt
+ * @param resultNumDataLen [IN]: resultNum memory length
+ * @param id0 [IN/OUT]: Level 1 library id0
+ * @param id1 [IN/OUT]: Secondary library id1
+ * @param resultOffset [IN/OUT]: The offset of the bottom library corresponding
+ * to each feature retrieval result, total length topK * queryCnt
+ * @param resultDistance [IN/OUT]: Distance, total length topK * queryCnt
+ * @param dataLen [IN]: The memory size requested by
+ * id0\id1\reslutOffset\resultDistance
+ * @retval null for failed. OtherValues success
+ */
+ACL_FUNC_VISIBILITY aclfvSearchResult *aclfvCreateSearchResult(uint32_t queryCnt, uint32_t *resultNum,
+    uint32_t resultNumDataLen, uint32_t *id0, uint32_t *id1, uint32_t *resultOffset, float *resultDistance,
+    uint32_t dataLen);
+
+/* *
+ * @ingroup AscendCL
+ * @brief Destroy search result.
+ *
+ * @par Function
+ * Can only destroy search result information created
+ * through aclfvCreateSearchResult interface.
+ * @param searchResult [IN/OUT]     search result.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure
+ *
+ * @see aclfvCreateSearchResult
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDestroySearchResult(aclfvSearchResult *searchResult);
+
+/* *
+ * @ingroup AscendCL
+ * @brief fv IP initialize.
+ *
+ * @param fsNum [IN]     max repo num, used to apply for memory.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvInit(uint64_t fsNum);
+
+/* *
+ * @ingroup AscendCL
+ * @brief release fv resources.
+ *
+ * @par Function
+ * Can only release fv resources created
+ * through aclfvInit interface.
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure.
+ *
+ * @see aclfvInit
+ */
+ACL_FUNC_VISIBILITY aclError aclfvRelease();
+
+/* *
+ * @ingroup AscendCL
+ * @brief fv repo add.
+ * @param type [IN]: repo add type
+ * @param featureInfo [IN/OUT]: add feature information
+ * @param stream [IN]: stream of task execute
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvRepoAdd(aclfvSearchType type, aclfvFeatureInfo *featureInfo, aclrtStream stream);
+
+/* *
+ * @ingroup AscendCL
+ * @brief fv repo del.
+ * @param type [IN]: repo delete type
+ * @param repoRange [IN/OUT]: repo range information
+ * @param stream [IN]: stream of task execute
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvRepoDel(aclfvSearchType type, aclfvRepoRange *repoRange, aclrtStream stream);
+
+/* *
+ * @ingroup AscendCL
+ * @brief fv accurate del.
+ * @param featureInfo [IN/OUT]: accurate delete feature information
+ * @param stream [IN]: stream of task execute
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvDel(aclfvFeatureInfo *featureInfo, aclrtStream stream);
+
+/* *
+ * @ingroup AscendCL
+ * @brief fv accurate modify.
+ * @param featureInfo [IN/OUT]: accurate modify feature information
+ * @param stream [IN]: stream of task execute
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvModify(aclfvFeatureInfo *featureInfo, aclrtStream stream);
+
+/* *
+ * @ingroup AscendCL
+ * @brief fv search.
+ * @param type [IN]: search type
+ * @param searchInput [IN/OUT]: search input
+ * @param searchRst [IN/OUT]: search result
+ * @param stream [IN]: stream of task execute
+ * @retval ACL_ERROR_NONE The function is successfully executed.
+ * @retval OtherValues Failure.
+ */
+ACL_FUNC_VISIBILITY aclError aclfvSearch(aclfvSearchType type, aclfvSearchInput *searchInput,
+    aclfvSearchResult *searchRst, aclrtStream stream);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // INC_EXTERNAL_ACL_OPS_ACL_RETR_H_

dnn/atlas-stub/src/libacl_cblas-wrap.cpp

+/**
+ * \file dnn/atlas-stub/src/libacl_cblas-wrap.cpp
+ * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+ *
+ * Copyright (c) 2014-2020 Megvii Inc. All rights reserved.
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ * implied.
+ */
+
+#pragma GCC visibility push(default)
+
+#include <cstdio>
+#define LOGE(fmt, v...) fprintf(stderr, "err: " fmt "\n", ##v)
+
+#include "acl/acl.h"
+#include "acl/ops/acl_cblas.h"
+
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
+
+#if defined(_WIN32)
+#include <windows.h>
+#define RTLD_LAZY 0
+
+static void* dlopen(const char* file, int) {
+    return static_cast<void*>(LoadLibraryA(file));
+}
+
+static void* dlerror() {
+    const char* errmsg = "dlerror not aviable in windows";
+    return const_cast<char*>(errmsg);
+}
+
+static void* dlsym(void* handle, const char* name) {
+    FARPROC symbol = GetProcAddress((HMODULE)handle, name);
+    return reinterpret_cast<void*>(symbol);
+}
+
+#else
+#include <dlfcn.h>
+#include <unistd.h>
+#endif
+
+static void log_failed_load(int func_idx);
+namespace {
+template <typename T>
+T on_init_failed(int func_idx);
+template <>
+aclError on_init_failed(int func_idx) {
+    log_failed_load(func_idx);
+    return ACL_ERROR_INTERNAL_ERROR;
+}
+}  // namespace
+
+#include "./libacl_cblas-wrap.h"
+
+static const char* default_so_paths[] = {
+        "/usr/local/Ascend/acllib/lib64/libacl_cblas.so",
+        "libacl_cblas.so",
+};
+
+static void* get_library_handle() {
+    void* handle = nullptr;
+    for (size_t i = 0; i < (sizeof(default_so_paths) / sizeof(char*)); i++) {
+        handle = dlopen(default_so_paths[i], RTLD_LAZY);
+        if (handle) {
+            break;
+        }
+    }
+
+    if (!handle) {
+        LOGE("Failed to load atlas library");
+        return nullptr;
+    }
+    return handle;
+}
+
+static void log_failed_load(int func_idx) {
+    LOGE("failed to load atlas func: %s", g_func_name[func_idx]);
+}
+
+static void* resolve_library_func(void* handle, const char* func) {
+    if (!handle) {
+        LOGE("handle should not be nullptr!");
+        return nullptr;
+    }
+    auto ret = dlsym(handle, func);
+    if (!ret) {
+        LOGE("failed to load atlas func: %s", func);
+    }
+    return ret;
+}

scripts/whl/manylinux2010/do_build.sh

@@ -57,6 +57,7 @@ do
     export EXTRA_CMAKE_ARGS="${EXTRA_CMAKE_ARGS} -DPYTHON_EXECUTABLE=${PYTHON_DIR}/bin/python3"
     export EXTRA_CMAKE_ARGS="${EXTRA_CMAKE_ARGS} -DPYTHON_LIBRARY=${PYTHON_DIR}lib/"
     export EXTRA_CMAKE_ARGS="${EXTRA_CMAKE_ARGS} -DPYTHON_INCLUDE_DIR=${PYTHON_DIR}include/python${MAJOR}.${MINOR}"
+    export EXTRA_CMAKE_ARGS="${EXTRA_CMAKE_ARGS} -DMGE_WITH_ATLAS=ON"
 
     if [ ${BUILD_WHL_CPU_ONLY} = "OFF" ]; then
         ${SRC_DIR}/scripts/cmake-build/host_build.sh -c -t -r