aclnnScatterNd
支持的产品型号
- Atlas 推理系列产品。
- Atlas 训练系列产品。
- Atlas A2训练系列产品/Atlas 800I A2推理产品。
接口原型
每个算子分为两段式接口,必须先调用“aclnnScatterNdGetWorkspaceSize”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“aclnnScatterNd”接口执行计算。
aclnnstatus aclnnScatterNdGetWorkspaceSize(const aclTensor *data,const aclTensor *indices,const aclTensor *updates, aclTensor *out, uint64_t *workspaceSize, aclOpExecutor **executor)
aclnnstatus aclnnScatterNd(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream)
功能描述
算子功能:拷贝data的数据至out,同时在指定indices处根据updates更新out中的数据。
aclnnScatterNdGetWorkspaceSize
参数说明:
data(aclTensor*,计算输入):Device侧的aclTensor, 且数据类型与updates,out一致,shape满足1<=rank(data)<=8。支持非连续的Tensor,数据格式支持ND。
- Atlas A2训练系列产品/Atlas 800I A2推理产品:数据类型支持FLOAT16, FLOAT, BOOL, BFLOAT16
- Atlas 训练系列产品、Atlas 推理系列产品:数据类型支持FLOAT16, FLOAT, BOOL
indices(aclTensor*,计算输入):Device侧的aclTensor,数据类型支持INT32, INT64。indices.shape[-1] <= rank(data),且1<=rank(indices)<=8。支持非连续的Tensor,数据格式支持ND。仅支持非负索引。indices中的索引数据不支持越界。
updates(aclTensor*,计算输入):Device侧的aclTensor, 且数据类型与data,out一致。shape要求rank(updates)=rank(data)+rank(indices)-indices.shape[-1] -1, 且满足1<=rank(updates)<=8。支持非连续的Tensor,数据格式支持ND。
- Atlas A2训练系列产品/Atlas 800I A2推理产品:数据类型支持FLOAT16, FLOAT, BOOL, BFLOAT16
- Atlas 训练系列产品、Atlas 推理系列产品:数据类型支持FLOAT16, FLOAT, BOOL
out(aclTensor*,计算输出):Device侧的aclTensor,数据类型支持FLOAT16, FLOAT, BOOL, BFLOAT16(仅Atlas A2训练系列产品/Atlas 800I A2推理产品支持), 且数据类型与data,out一致,shape与data一致。支持非连续的Tensor,数据格式支持ND。
- Atlas A2训练系列产品/Atlas 800I A2推理产品:数据类型支持FLOAT16, FLOAT, BOOL, BFLOAT16
- Atlas 训练系列产品、Atlas 推理系列产品:数据类型支持FLOAT16, FLOAT, BOOL
workspaceSize(uint64_t*,出参):返回需要在Device侧申请的workspace大小。
executor(aclOpExecutor**,出参):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。
返回 161001(ACLNN_ERR_PARAM_NULLPTR):1.传入的data、indices、updates、out中有空指针 返回 161002(ACLNN_ERR_PARAM_INVALID):1. 数据类型不在支持的范围之内; 2. shape不满足要求:1<=rank(data)<=8, 1<=rank(indices)<=8,rank(updates)=rank(data)+rank(indices)- indices.shape[-1] -1 3. shape不满足要求:1<=rank(indices)<=8, indices.shape[-1] <= rank(data) 4. shape不满足要求:1<=rank(updates)<=8, updates.shape == indices.shape[:-1] + data.shape[indices.shape[-1] :] 5. shape不满足要求:data.shape == out.shape
aclnnScatterNd
- 参数说明:
- workspace(void *, 入参):在Device侧申请的workspace内存地址。
- workspaceSize(uint64_t, 入参):在Device侧申请的workspace大小,由第一段接口aclnnScatterNdGetWorkspaceSize获取。
- executor(aclOpExecutor *, 入参):op执行器,包含了算子计算流程。
- stream(aclrtStream, 入参):指定执行任务的 AscendCL Stream流。
- 返回值: aclnnStatus:返回状态码,具体参见aclnn返回码。
约束与限制
无
调用示例
#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_scatter_nd.h"
#include "aclnn/aclnn_base.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::vector<int64_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,AscendCL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::vector<T>& hostData, const std::vector<int64_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::vector<int64_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对外接口列表
// 根据自己的实际device填写deviceId
int32_t deviceId = 0;
aclrtStream stream;
auto ret = Init(deviceId, &stream);
// check根据自己的需要处理
CHECK_RET(ret == 0, 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::vector<int64_t> dataShape = {8};
std::vector<int64_t> indicesShape = {4, 1};
std::vector<int64_t> updatesShape = {4};
std::vector<int64_t> outShape = {8};
void* dataDeviceAddr = nullptr;
void* indicesDeviceAddr = nullptr;
void* updatesDeviceAddr = nullptr;
void* outDeviceAddr = nullptr;
aclTensor* data = nullptr;
aclTensor* indices = nullptr;
aclTensor* updates = nullptr;
aclTensor* out = nullptr;
std::vector<float> selfHostData = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0};
std::vector<int32_t> indicesData = {4,3,1,7};
std::vector<float> updatesData = {9.0, 10.0, 11.0, 12.0};
std::vector<float> outData = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0};
ret = CreateAclTensor(selfHostData, dataShape, &dataDeviceAddr, aclDataType::ACL_FLOAT, &data);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(indicesData, indicesShape, &indicesDeviceAddr, aclDataType::ACL_INT32, &indices);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(updatesData, updatesShape, &updatesDeviceAddr, aclDataType::ACL_FLOAT, &updates);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(outData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建out aclTensor
// ret = CreateAclTensor(outData, outShape, &outDeviceAddr, aclDataType::ACL_INT32, &out);
// CHECK_RET(ret == ACL_SUCCESS, return ret);
// 3. 调用CANN算子库API,需要修改为具体的API
uint64_t workspaceSize = 0;
aclOpExecutor* executor;
// 调用aclnnAdd第一段接口
ret = aclnnScatterNdGetWorkspaceSize(data, indices, updates, out, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnScatterNdGetWorkspaceSize 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) {
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;);
}
ret = aclnnScatterNd(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnScatterNd 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(outShape);
std::vector<float> resultData(size, 0);
ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), outDeviceAddr, size * sizeof(float),
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和aclScalar,需要根据具体API的接口定义修改
aclDestroyTensor(data);
aclDestroyTensor(indices);
aclDestroyTensor(updates);
aclDestroyTensor(out);
// 7. 释放device资源,需要根据具体API的接口定义修改
aclrtFree(dataDeviceAddr);
aclrtFree(indicesDeviceAddr);
aclrtFree(updatesDeviceAddr);
aclrtFree(outDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}