aclnnNonMaxSuppression
支持的产品型号
- Atlas 推理系列产品。
接口原型
每个算子分为两段式接口,必须先调用“aclnnNonMaxSuppressionGetWorkspaceSize”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“aclnnNonMaxSuppression”接口执行计算。
aclnnStatus aclnnNonMaxSuppressionGetWorkspaceSize(const aclTensor *boxes, const aclTensor *scores, aclIntArray *maxOutputBoxesPerClass, aclFloatArray *iouThreshold, aclFloatArray *scoreThreshold, int32_t centerPointBox, aclTensor *selectedIndices, uint64_t *workspaceSize, aclOpExecutor **executor)
aclnnStatus aclnnNonMaxSuppression(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream)
功能描述
算子功能:删除分数小于 scoreThreshold 的边界框, 筛选出 与之前被选中 重叠较高(IOU较高)的框。
aclnnNonMaxSuppressionGetWorkspaceSize
参数说明:
boxes(aclTensor*, 计算输入):Device侧的aclTensor,数据类型支持FLOAT、FLOAT16。支持非连续的Tensor,支持数据格式ND。3维shape:[num_batches, spatial_dimension, 4]。
scores(aclTensor*, 计算输入):Device侧的aclTensor,数据类型支持FLOAT、FLOAT16。支持非连续的Tensor,支持数据格式ND。3维shape:[num_batches, num_classes, spatial_dimension]。
maxOutputBoxesPerClass(aclIntArray*, 计算输入):Host侧的aclIntArray,数据类型支持INT32。支持连续的Tensor,支持数据格式ND。标量,默认值为0,表示无输出,数值上限为700。表示每个类别每batch要选择的最大框数。
iouThreshold(aclFloatArray*, 计算输入):Host侧的aclFloatArray,数据类型支持FLOAT。支持连续的Tensor,支持数据格式ND。标量,取值范围[0, 1]。表示判断框相对于IOU是否重叠过多的阈值。
scoreThreshold(aclFloatArray*, 计算输入):Host侧的aclFloatArray,数据类型支持FLOAT。支持连续的Tensor,支持数据格式ND。标量,取值范围[0, 1]。表示根据得分决定何时移除框的阈值。
centerPointBox(int, 计算输入):Host侧的整数,数据类型支持INT32。取值范围0或1。默认值为0。作为属性 它决定了边界框格式。等于0时,主要用于TF模型, 数据以(y1, x1, y2, x2)形式提供,其中(y1, x1) (y2, x2)是对角线框角坐标,需要用户自行保证x1<x2, y1<y2; 等于1时, 主要用于pytorch模型。框数据以(x_center, y_center, width, height)形式提供。
selectedIndices(aclTensor*, 计算输出):Device侧的aclTensor,数据类型支持INT32。支持非连续的Tensor,支持数据格式ND。输出的是一组整数,索引到表示所选框的边界框的输入集合中
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。
返回码:
aclnnStatus:返回状态码,具体参见aclnn返回码。
返回361001(ACLNN_ERR_RUNTIME_ERROR):1. 当前产品不支持
返回161001(ACLNN_ERR_PARAM_NULLPTR):1. 传入的boxes、scores、out是空指针
返回161002(ACLNN_ERR_PARAM_INVALID):1. boxes、scores和maxOutputBoxesPerClass的数据类型不在支持的范围内:
2. boxes、scores和 selectedIndices 的数据格式不在支持的范围内:
3. boxes、scores 的shape不在支持的范围内:
1) boxes、scores需为3维;
2) boxes第0维 必须等于 scores第0维度;
3) boxes第1维 必须等于 scores第2维度;
4) boxes第2维 必须等于 4
4. iouThreshold,scoreThreshold,centerPointBox,maxOutputBoxesPerClass数值不在支持的范围内
aclnnNonMaxSuppression
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aclnnNonMaxSuppressionGetWorkspaceSize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。
返回码:
aclnnStatus:返回状态码,具体参见aclnn返回码。
约束与限制
maxOutputBoxesPerClass 参数上限为700。输入参数boxes和scores的数据类型要求保持一致。
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_non_max_suppression.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)
template <typename T>
int64_t GetShapeSize(const std::vector<T>& shape) {
int64_t shape_size = 1;
for (auto i : shape) {
shape_size *= i;
}
return shape_size;
}
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;
}
template <typename T>
int CreateAclIntArray(const std::vector<T>& hostData, void** deviceAddr, aclIntArray** intArray) {
auto size = GetShapeSize(hostData) * 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);
// 调用aclCreateIntArray接口创建aclIntArray
*intArray = aclCreateIntArray(hostData.data(), hostData.size());
return 0;
}
int main() {
// 1. (固定写法)device/stream初始化,参考AscendCL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_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::vector<int64_t> boxesShape = {1, 7, 4};
std::vector<int64_t> scoresShape = {1, 1, 7};
std::vector<int64_t> maxSizePerClassShape = {3};
std::vector<int64_t> selectedIndicesShape = {3, 3};
void* boxesDeviceAddr = nullptr;
void* scoresDeviceAddr = nullptr;
void* maxSizePerClassDeviceAddr = nullptr;
void* outDeviceAddr = nullptr;
aclTensor* boxes = nullptr;
aclTensor* scores = nullptr;
aclIntArray* maxOutputBoxesPerClass = nullptr;
aclFloatArray* iouThd = nullptr;
aclFloatArray* scoresThd = nullptr;
aclTensor* selectedIndices = nullptr;
std::vector<float> boxesHostData = {
49.1, 32.4, 51.0, 35.9,
49.3, 32.9, 51.0, 35.3,
49.2, 31.8, 51.0, 35.4,
35.1, 11.5, 39.1, 15.7,
35.6, 11.8, 39.3, 14.2,
35.3, 11.5, 39.9, 14.5,
35.2, 11.7, 39.7, 15.7,
};
std::vector<float> scoresHostData = {0.9, 0.9, 0.5, 0.5, 0.5, 0.4, 0.3};
std::vector<int64_t> maxOutputBoxesPerClassHostData = {3};
std::vector<float> iouThresholdHostData = {0.6};
std::vector<float> scoreThresholdHostData = {0};
std::vector<int32_t> outHostData = {0, 0, 0, 0, 0, 0, 0, 0, 0};
// 创建aclTensor: boxes
ret = CreateAclTensor(boxesHostData, boxesShape, &boxesDeviceAddr, aclDataType::ACL_FLOAT, &boxes);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建aclTensor: scores
ret = CreateAclTensor(scoresHostData, scoresShape, &scoresDeviceAddr, aclDataType::ACL_FLOAT, &scores);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建AclIntArray: maxOutputBoxesPerClass
ret = CreateAclIntArray(maxOutputBoxesPerClassHostData, &maxSizePerClassDeviceAddr, &maxOutputBoxesPerClass);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建AclFloatArray: iouThreshold
iouThd = aclCreateFloatArray(iouThresholdHostData.data(), iouThresholdHostData.size());
CHECK_RET(iouThd != nullptr, return 0);
// 创建AclFloatArray: scoresThreshold
scoresThd = aclCreateFloatArray(scoreThresholdHostData.data(), scoreThresholdHostData.size());
CHECK_RET(scoresThd != nullptr, return 0);
// 创建aclTensor: selectedIndices
ret = CreateAclTensor(outHostData, selectedIndicesShape, &outDeviceAddr, aclDataType::ACL_INT32, &selectedIndices);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建attr int: centerPointBox
int64_t centerPointBox = 0;
// 3. 调用CANN算子库API
uint64_t workspaceSize = 0;
aclOpExecutor* executor;
// 调用aclnnNonMaxSuppression第一段接口
ret = aclnnNonMaxSuppressionGetWorkspaceSize(boxes, scores, maxOutputBoxesPerClass, iouThd, scoresThd, centerPointBox, selectedIndices, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnNonMaxSuppressionGetWorkspaceSize 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;);
}
// 调用aclnnNonMaxSuppression第二段接口
ret = aclnnNonMaxSuppression(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnNonMaxSuppression 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(selectedIndicesShape);
std::vector<int32_t> resultData(size, 0);
ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]),
outDeviceAddr, size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy resultData from device to host failed. ERROR: %d\n", ret); return ret);
for (int64_t i = 0; i < size; i++) {
LOG_PRINT("result[%ld] is: %d\n", i, resultData[i]);
}
// 6. 释放aclTensor,需要根据具体API的接口定义修改
aclDestroyTensor(boxes);
aclDestroyTensor(scores);
aclDestroyIntArray(maxOutputBoxesPerClass);
aclDestroyFloatArray(iouThd);
aclDestroyFloatArray(scoresThd);
aclDestroyTensor(selectedIndices);
// 7. 释放device资源,需要根据具体API的接口定义修改
aclrtFree(boxesDeviceAddr);
aclrtFree(scoresDeviceAddr);
aclrtFree(maxSizePerClassDeviceAddr);
aclrtFree(outDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}