aclnnCoalesceSparse【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 查看源码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×功能说明将相同坐标点的value进行累加求和进而减少Coo_Tensor的内存大小。函数原型每个算子分为两段式接口必须先调用“aclnnCoalesceSparseGetWorkspaceSize”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“aclnnCoalesceSparse”接口执行计算。aclnnStatus aclnnCoalesceSparseGetWorkspaceSize( const aclTensor *uniqueLen, const aclTensor *uniqueIndices, const aclTensor *indices, const aclTensor *values, const aclTensor *newIndicesOut, const aclTensor *newValuesOut, uint64_t *workspaceSize, aclOpExecutor **executor);aclnnStatus aclnnCoalesceSparse( void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream);aclnnCoalesceSparseGetWorkspaceSize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TensoruniqueLenaclTensor*输入去重后的索引数。不支持空Tensor。INT32、INT64ND1√uniqueIndicesaclTensor*输入去重后的索引数组。不支持空Tensor。INT32、INT64ND1√indicesaclTensor*输入索引数组。不支持空Tensor。重索引后的indices值不能超过int32上限。INT32、INT64ND2√valuesaclTensor*输入每个坐标对应的元素值。不支持空Tensor。INT32、FLOAT16、FLOAT32ND1-8√newIndicesOutaclTensor*输出合并后的索引数组。不支持空Tensor。INT32、INT64ND2√newValuesOutaclTensor*输出合并后的元素值。不支持空Tensor。INT32、FLOAT16、FLOAT32ND1-8√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。-----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码错误码描述ACLNN_ERR_PARAM_NULLPTR161001传入的uniqueLen、uniqueIndices、indices、values、newIndicesOut或newValuesOut是空指针。ACLNN_ERR_PARAM_INVALID161002uniqueLen、uniqueIndices、indices、values、newIndicesOut或newValuesOut的数据类型不在支持范围之内。values或newValuesOut的维度超过8维。重索引后的indices值不能超过int32上限。aclnnCoalesceSparse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aclnnCoalesceSparseGetWorkspaceSize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。约束说明无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 iostream #include vector #include acl/acl.h #include aclnnop/aclnn_coalesce_sparse.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; } int main() { // 1. 固定写法device/stream初始化参考acl API文档 // 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 uniqueLenShape {1}; std::vectorint64_t uniqueIndicesShape {2,2}; std::vectorint64_t indexShape {2,4}; std::vectorint64_t valueShape {4}; std::vectorint64_t newIndexShape {4}; std::vectorint64_t newValueShape {2}; void* uniqueLenDeviceAddr nullptr; void* uniqueIndicesDeviceAddr nullptr; void* indexDeviceAddr nullptr; void* valueDeviceAddr nullptr; void* newIndexDeviceAddr nullptr; void* newValueDeviceAddr nullptr; aclTensor* uniqueLen nullptr; aclTensor* uniqueIndices nullptr; aclTensor* index nullptr; aclTensor* value nullptr; aclTensor* newIndex nullptr; aclTensor* newValue nullptr; std::vectorint32_t uniqueLenData {2}; std::vectorint32_t uniqueIndicesData {0, 1, 0, 2}; std::vectorint32_t indexData {0, 0, 1, 1, 0, 0, 2, 2}; std::vectorfloat valueData {1, 2, 3, 4}; std::vectorint32_t newIndexData {0, 0, 0, 0}; std::vectorfloat newValueData {0, 0}; // 创建in aclTensor ret CreateAclTensor(uniqueLenData, uniqueLenShape, uniqueLenDeviceAddr, aclDataType::ACL_INT32, uniqueLen); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建in aclTensor ret CreateAclTensor(uniqueIndicesData, uniqueIndicesShape, uniqueIndicesDeviceAddr, aclDataType::ACL_INT32, uniqueIndices); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建in aclTensor ret CreateAclTensor(indexData, indexShape, indexDeviceAddr, aclDataType::ACL_INT32, index); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建in aclTensor ret CreateAclTensor(valueData, valueShape, valueDeviceAddr, aclDataType::ACL_FLOAT, value); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建out aclTensor ret CreateAclTensor(newIndexData, newIndexShape, newIndexDeviceAddr, aclDataType::ACL_INT32, newIndex); CHECK_RET(ret ACL_SUCCESS, return ret); // 创建out aclTensor ret CreateAclTensor(newValueData, newValueShape, newValueDeviceAddr, aclDataType::ACL_FLOAT, newValue); CHECK_RET(ret ACL_SUCCESS, return ret); // 3. 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aclnnCoalesceSparse第一段接口 ret aclnnCoalesceSparseGetWorkspaceSize(uniqueLen, uniqueIndices, index, value, newIndex, newValue, workspaceSize, executor); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnCoalesceSparseGetWorkspaceSize 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 static_castuint64_t(0)) { 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aclnnCoalesceSparse第二段接口 ret aclnnCoalesceSparse(workspaceAddr, workspaceSize, executor, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnCoalesceSparse failed. ERROR: %d\n, ret); return ret); // 4. 固定写法同步等待任务执行结束 ret aclrtSynchronizeStream(stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtSynchronizeStream failed. ERROR: %d\n, ret); return ret); // 5. 获取输出的值将device侧内存上的结果拷贝至host侧需要根据具体API的接口定义修改 auto size GetShapeSize(newValueShape); std::vectorfloat resultData(size, 0); ret aclrtMemcpy( resultData.data(), resultData.size() * sizeof(resultData[0]), newValueDeviceAddr, size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(copy result from device to host failed. ERROR: %d\n, ret); return ret); for (int64_t i 0; i size; i) { LOG_PRINT(result[%ld] is: %f\n, i, resultData[i]); } // 6. 释放aclTensor需要根据具体API的接口定义修改 aclDestroyTensor(uniqueLen); aclDestroyTensor(uniqueIndices); aclDestroyTensor(index); aclDestroyTensor(value); aclDestroyTensor(newIndex); aclDestroyTensor(newValue); // 7. 释放device资源 aclrtFree(uniqueLenDeviceAddr); aclrtFree(uniqueIndicesDeviceAddr); aclrtFree(indexDeviceAddr); aclrtFree(valueDeviceAddr); aclrtFree(newIndexDeviceAddr); aclrtFree(newValueDeviceAddr); if (workspaceSize static_castuint64_t(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