aclnnSplitTensor【免费下载链接】ops-math本项目是CANN提供的数学类基础计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-math产品支持情况产品是否支持Ascend 950PR/Ascend 950DT×Atlas A3 训练系列产品/Atlas A3 推理系列产品×Atlas A2 训练系列产品/Atlas A2 推理系列产品√Atlas 200I/500 A2 推理产品×Atlas 推理系列产品×Atlas 训练系列产品×功能说明将输入self沿dim轴按照splitSections大小均匀切分。若dim轴无法被整除则非最后一块的大小等于splitSections最后一块小于splitSections。函数原型每个算子分为两段式接口必须先调用“aclnnSplitTensorGetWorkspaceSize”接口获取计算所需workspace大小以及包含了算子计算流程的执行器再调用“aclnnSplitTensor”接口执行计算。aclnnStatus aclnnSplitTensorGetWorkspaceSize( const aclTensor* self, uint64_t splitSections, int64_t dim, aclTensorList* out, uint64_t* workspaceSize, aclOpExecutor** executor)aclnnStatus aclnnSplitTensor( void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, aclrtStream stream)aclnnSplitTensorGetWorkspaceSize参数说明参数名输入/输出描述使用说明数据类型数据格式维度shape非连续TensorselfaclTensor*输入表示被split的输入tensor-FLOAT、FLOAT16、DOUBLE、INT32、UINT32、INT64、UINT64、INT16、UINT16、INT8、UINT8、BOOL、COMPLEX128、COMPLEX64、BFLOAT16ND1-8√splitSectionsuint64_t输入表示沿dim轴均匀切分后的块大小, 最后一块可以小于splitSections。-UINT64---dimint64_t输入表示输入self被split的维度取值范围在[-self.dim(), self.dim())-INT64---outaclTensorList*输出表示被split后的输出tensor的列表。每个输出的dtype需要保持一致。TensorList中每个输出tensor的维度与self的维度一致。FLOAT、FLOAT16、DOUBLE、INT32、UINT32、INT64、UINT64、INT16、UINT16、INT8、UINT8、BOOL、COMPLEX128、COMPLEX64、BFLOAT16ND-√workspaceSizeuint64_t*输出返回需要在Device侧申请的workspace大小。-----executoraclOpExecutor**输出返回op执行器包含了算子计算流程。-----Atlas 推理系列产品 、 Atlas 训练系列产品 数据类型不支持BFLOAT16。返回值aclnnStatus返回状态码具体参见aclnn返回码。第一段接口完成入参校验出现以下场景时报错返回值错误码描述ACLNN_ERR_PARAM_NULLPTR161001传入的self、out是空指针。ACLNN_ERR_PARAM_INVALID161002self和out的数据类型不在支持的范围之内。self的长度不在支持的范围之内。out中的tensor长度不在支持的范围之内。dim的取值越界。被split的维度shape为0且splitSections不为0。被split的维度shape不为0且splitSections为0。aclnnSplitTensor参数说明参数名输入/输出描述workspace输入在Device侧申请的workspace内存地址。workspaceSize输入在Device侧申请的workspace大小由第一段接口aclnnSplitTensorGetWorkspaceSize获取。executor输入op执行器包含了算子计算流程。stream输入指定执行任务的Stream。返回值aclnnStatus返回状态码具体参见aclnn返回码。约束说明确定性计算aclnnSplitTensor默认确定性实现。调用示例示例代码如下仅供参考具体编译和执行过程请参考编译与运行样例。#include chrono #include algorithm #include iostream #include vector #include acl/acl.h #include aclnnop/aclnn_split_tensor.h #define CHECK_RET(cond, return_expr) \ do { \ if (!(cond)) { \ return_expr; \ } \ } while (0) #define LOG_PRINT(message, ...) \ do { \ printf(message, ##__VA_ARGS__); \ } while (0) int64_t GetShapeSize(const std::vectorint64_t shape) { int64_t shapeSize 1; for (auto i : shape) { shapeSize * i; } return shapeSize; } int Init(int32_t deviceId, aclrtStream* stream) { // 固定写法资源初始化 auto ret aclInit(nullptr); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclInit failed. ERROR: %d\n, ret); return ret); ret aclrtSetDevice(deviceId); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtSetDevice failed. ERROR: %d\n, ret); return ret); ret aclrtCreateStream(stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtCreateStream failed. ERROR: %d\n, ret); return ret); return 0; } template typename T int CreateAclTensor(const std::vectorT hostData, const std::vectorint64_t shape, void** deviceAddr, aclDataType dataType, aclTensor** tensor) { auto size GetShapeSize(shape) * sizeof(T); // 调用aclrtMalloc申请device侧内存 auto ret aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtMalloc failed. ERROR: %d\n, ret); return ret); // 调用aclrtMemcpy将host侧数据拷贝到device侧内存上 ret aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtMemcpy failed. ERROR: %d\n, ret); return ret); // 计算连续tensor的strides std::vectorint64_t strides(shape.size(), 1); for (int64_t i shape.size() - 2; i 0; i--) { strides[i] shape[i 1] * strides[i 1]; } // 调用aclCreateTensor接口创建aclTensor *tensor aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), *deviceAddr); return 0; } void CheckResult(const std::vectorstd::vectorint64_t shapeList, const std::vectorvoid * addrList) { for (size_t i 0; i shapeList.size(); i) { auto size GetShapeSize(shapeList[i]); std::vectorfloat resultData(size, 0); auto ret aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), addrList[i], size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtMemcpy failed. ERROR: %d\n, ret); return); for (int64_t j 0; j size; j) { LOG_PRINT(result[%ld] is: %f\n, j, resultData[j]); } } } int main() { // 1.固定写法device/stream初始化参考acl API手册 // 根据自己的实际device填写deviceId int32_t deviceId 0; aclrtStream stream; auto ret Init(deviceId, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(Init acl failed. ERROR: %d\n, ret); return ret); // 2. 构造输入与输出需要根据API的接口自定义构造 std::vectorint64_t selfShape {4, 2}; std::vectorint64_t shape1 {2, 2}; std::vectorint64_t shape2 {2, 2}; uint64_t splitSections 2; int64_t dim 0; void* selfDeviceAddr nullptr; void* shape1DeviceAddr nullptr; void* shape2DeviceAddr nullptr; aclTensor* self nullptr; aclTensor* shape1Addr nullptr; aclTensor* shape2Addr nullptr; std::vectorfloat selfHostData {0, 1, 2, 3, 4, 5, 6, 7}; std::vectorfloat shape1HostData {0, 1, 4, 5}; std::vectorfloat shape2HostData {2, 3, 6, 7}; // 创建self aclTensor ret CreateAclTensor(selfHostData, selfShape, selfDeviceAddr, aclDataType::ACL_FLOAT, self); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(shape1HostData, shape1, shape1DeviceAddr, aclDataType::ACL_FLOAT, shape1Addr); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(shape2HostData, shape2, shape2DeviceAddr, aclDataType::ACL_FLOAT, shape2Addr); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建out aclTensorList std::vectoraclTensor* tmp {shape1Addr, shape2Addr}; aclTensorList* out aclCreateTensorList(tmp.data(), tmp.size()); CHECK_RET(out ! nullptr, return ret); // 调用CANN算子库API需要修改为具体的Api名称 uint64_t workspaceSize 0; aclOpExecutor *executor; // 调用aclnnSplitTensor第一段接口 ret aclnnSplitTensorGetWorkspaceSize(self, splitSections, dim, out, workspaceSize, executor); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnSplitTensorGetWorkspaceSize failed. ERROR: %d\n, ret); return ret); // 根据第一段接口计算出的workspaceSize申请device内存 void *workspaceAddr nullptr; if (workspaceSize 0) { auto ret aclrtMalloc(workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(allocate workspace failed. ERROR: %d\n, ret); return ret); } // 调用aclnnSplitTensor第二段接口 ret aclnnSplitTensor(workspaceAddr, workspaceSize, executor, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnSplitTensor failed. ERROR: %d\n, ret); return ret); ret aclrtSynchronizeStream(stream); CheckResult({shape1, shape2}, {shape1DeviceAddr, shape2DeviceAddr}); // 6. 释放aclTensor和aclScalar需要根据具体API的接口定义修改 aclDestroyTensor(self); aclDestroyTensorList(out); aclDestroyTensor(shape1Addr); aclDestroyTensor(shape2Addr); // 7. 释放device 资源 aclrtFree(selfDeviceAddr); aclrtFree(shape1DeviceAddr); aclrtFree(shape2DeviceAddr); if (workspaceSize 0) { aclrtFree(workspaceAddr); } aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }【免费下载链接】ops-math本项目是CANN提供的数学类基础计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-math创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
aclnnSplitTensor【免费下载链接】ops-math本项目是CANN提供的数学类基础计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-math产品支持情况产品是否支持Ascend 950PR/Ascend 950DT×Atlas A3 训练系列产品/Atlas A3 推理系列产品×Atlas A2 训练系列产品/Atlas A2 推理系列产品√Atlas 200I/500 A2 推理产品×Atlas 推理系列产品×Atlas 训练系列产品×功能说明将输入self沿dim轴按照splitSections大小均匀切分。若dim轴无法被整除则非最后一块的大小等于splitSections最后一块小于splitSections。函数原型每个算子分为两段式接口必须先调用“aclnnSplitTensorGetWorkspaceSize”接口获取计算所需workspace大小以及包含了算子计算流程的执行器再调用“aclnnSplitTensor”接口执行计算。aclnnStatus aclnnSplitTensorGetWorkspaceSize( const aclTensor* self, uint64_t splitSections, int64_t dim, aclTensorList* out, uint64_t* workspaceSize, aclOpExecutor** executor)aclnnStatus aclnnSplitTensor( void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, aclrtStream stream)aclnnSplitTensorGetWorkspaceSize参数说明参数名输入/输出描述使用说明数据类型数据格式维度shape非连续TensorselfaclTensor*输入表示被split的输入tensor-FLOAT、FLOAT16、DOUBLE、INT32、UINT32、INT64、UINT64、INT16、UINT16、INT8、UINT8、BOOL、COMPLEX128、COMPLEX64、BFLOAT16ND1-8√splitSectionsuint64_t输入表示沿dim轴均匀切分后的块大小, 最后一块可以小于splitSections。-UINT64---dimint64_t输入表示输入self被split的维度取值范围在[-self.dim(), self.dim())-INT64---outaclTensorList*输出表示被split后的输出tensor的列表。每个输出的dtype需要保持一致。TensorList中每个输出tensor的维度与self的维度一致。FLOAT、FLOAT16、DOUBLE、INT32、UINT32、INT64、UINT64、INT16、UINT16、INT8、UINT8、BOOL、COMPLEX128、COMPLEX64、BFLOAT16ND-√workspaceSizeuint64_t*输出返回需要在Device侧申请的workspace大小。-----executoraclOpExecutor**输出返回op执行器包含了算子计算流程。-----Atlas 推理系列产品 、 Atlas 训练系列产品 数据类型不支持BFLOAT16。返回值aclnnStatus返回状态码具体参见aclnn返回码。第一段接口完成入参校验出现以下场景时报错返回值错误码描述ACLNN_ERR_PARAM_NULLPTR161001传入的self、out是空指针。ACLNN_ERR_PARAM_INVALID161002self和out的数据类型不在支持的范围之内。self的长度不在支持的范围之内。out中的tensor长度不在支持的范围之内。dim的取值越界。被split的维度shape为0且splitSections不为0。被split的维度shape不为0且splitSections为0。aclnnSplitTensor参数说明参数名输入/输出描述workspace输入在Device侧申请的workspace内存地址。workspaceSize输入在Device侧申请的workspace大小由第一段接口aclnnSplitTensorGetWorkspaceSize获取。executor输入op执行器包含了算子计算流程。stream输入指定执行任务的Stream。返回值aclnnStatus返回状态码具体参见aclnn返回码。约束说明确定性计算aclnnSplitTensor默认确定性实现。调用示例示例代码如下仅供参考具体编译和执行过程请参考编译与运行样例。#include chrono #include algorithm #include iostream #include vector #include acl/acl.h #include aclnnop/aclnn_split_tensor.h #define CHECK_RET(cond, return_expr) \ do { \ if (!(cond)) { \ return_expr; \ } \ } while (0) #define LOG_PRINT(message, ...) \ do { \ printf(message, ##__VA_ARGS__); \ } while (0) int64_t GetShapeSize(const std::vectorint64_t shape) { int64_t shapeSize 1; for (auto i : shape) { shapeSize * i; } return shapeSize; } int Init(int32_t deviceId, aclrtStream* stream) { // 固定写法资源初始化 auto ret aclInit(nullptr); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclInit failed. ERROR: %d\n, ret); return ret); ret aclrtSetDevice(deviceId); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtSetDevice failed. ERROR: %d\n, ret); return ret); ret aclrtCreateStream(stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtCreateStream failed. ERROR: %d\n, ret); return ret); return 0; } template typename T int CreateAclTensor(const std::vectorT hostData, const std::vectorint64_t shape, void** deviceAddr, aclDataType dataType, aclTensor** tensor) { auto size GetShapeSize(shape) * sizeof(T); // 调用aclrtMalloc申请device侧内存 auto ret aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtMalloc failed. ERROR: %d\n, ret); return ret); // 调用aclrtMemcpy将host侧数据拷贝到device侧内存上 ret aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtMemcpy failed. ERROR: %d\n, ret); return ret); // 计算连续tensor的strides std::vectorint64_t strides(shape.size(), 1); for (int64_t i shape.size() - 2; i 0; i--) { strides[i] shape[i 1] * strides[i 1]; } // 调用aclCreateTensor接口创建aclTensor *tensor aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), *deviceAddr); return 0; } void CheckResult(const std::vectorstd::vectorint64_t shapeList, const std::vectorvoid * addrList) { for (size_t i 0; i shapeList.size(); i) { auto size GetShapeSize(shapeList[i]); std::vectorfloat resultData(size, 0); auto ret aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), addrList[i], size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtMemcpy failed. ERROR: %d\n, ret); return); for (int64_t j 0; j size; j) { LOG_PRINT(result[%ld] is: %f\n, j, resultData[j]); } } } int main() { // 1.固定写法device/stream初始化参考acl API手册 // 根据自己的实际device填写deviceId int32_t deviceId 0; aclrtStream stream; auto ret Init(deviceId, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(Init acl failed. ERROR: %d\n, ret); return ret); // 2. 构造输入与输出需要根据API的接口自定义构造 std::vectorint64_t selfShape {4, 2}; std::vectorint64_t shape1 {2, 2}; std::vectorint64_t shape2 {2, 2}; uint64_t splitSections 2; int64_t dim 0; void* selfDeviceAddr nullptr; void* shape1DeviceAddr nullptr; void* shape2DeviceAddr nullptr; aclTensor* self nullptr; aclTensor* shape1Addr nullptr; aclTensor* shape2Addr nullptr; std::vectorfloat selfHostData {0, 1, 2, 3, 4, 5, 6, 7}; std::vectorfloat shape1HostData {0, 1, 4, 5}; std::vectorfloat shape2HostData {2, 3, 6, 7}; // 创建self aclTensor ret CreateAclTensor(selfHostData, selfShape, selfDeviceAddr, aclDataType::ACL_FLOAT, self); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(shape1HostData, shape1, shape1DeviceAddr, aclDataType::ACL_FLOAT, shape1Addr); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(shape2HostData, shape2, shape2DeviceAddr, aclDataType::ACL_FLOAT, shape2Addr); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建out aclTensorList std::vectoraclTensor* tmp {shape1Addr, shape2Addr}; aclTensorList* out aclCreateTensorList(tmp.data(), tmp.size()); CHECK_RET(out ! nullptr, return ret); // 调用CANN算子库API需要修改为具体的Api名称 uint64_t workspaceSize 0; aclOpExecutor *executor; // 调用aclnnSplitTensor第一段接口 ret aclnnSplitTensorGetWorkspaceSize(self, splitSections, dim, out, workspaceSize, executor); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnSplitTensorGetWorkspaceSize failed. ERROR: %d\n, ret); return ret); // 根据第一段接口计算出的workspaceSize申请device内存 void *workspaceAddr nullptr; if (workspaceSize 0) { auto ret aclrtMalloc(workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(allocate workspace failed. ERROR: %d\n, ret); return ret); } // 调用aclnnSplitTensor第二段接口 ret aclnnSplitTensor(workspaceAddr, workspaceSize, executor, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnSplitTensor failed. ERROR: %d\n, ret); return ret); ret aclrtSynchronizeStream(stream); CheckResult({shape1, shape2}, {shape1DeviceAddr, shape2DeviceAddr}); // 6. 释放aclTensor和aclScalar需要根据具体API的接口定义修改 aclDestroyTensor(self); aclDestroyTensorList(out); aclDestroyTensor(shape1Addr); aclDestroyTensor(shape2Addr); // 7. 释放device 资源 aclrtFree(selfDeviceAddr); aclrtFree(shape1DeviceAddr); aclrtFree(shape2DeviceAddr); if (workspaceSize 0) { aclrtFree(workspaceAddr); } aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }【免费下载链接】ops-math本项目是CANN提供的数学类基础计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-math创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
平方复杂度无法被常规优化消除｜TaoToken 统一 Key 通道实测)
