* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
#include <iostream>
#include <string>
#include <vector>
#include "acl/acl.h"
#include "aclnn_scatter_reduce.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)
namespace {
constexpr int64_t kReduceNone = 0;
constexpr int64_t kReduceAdd = 1;
constexpr int64_t kReduceMul = 2;
constexpr int64_t kReduceMax = 3;
constexpr int64_t kReduceMin = 4;
constexpr int64_t kReduceMean = 5;
struct ReduceCase {
std::string name;
int64_t reduce;
bool includeSelf;
std::vector<float> selfHostData;
std::vector<int64_t> indexHostData;
std::vector<float> srcHostData;
};
const char* GetReduceName(int64_t reduce)
{
switch (reduce) {
case kReduceNone:
return "none";
case kReduceAdd:
return "add";
case kReduceMul:
return "mul";
case kReduceMax:
return "max";
case kReduceMin:
return "min";
case kReduceMean:
return "mean";
default:
return "unknown";
}
}
}
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)
{
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);
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);
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);
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];
}
*tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
shape.data(), shape.size(), *deviceAddr);
return 0;
}
void DestroyTensorAndBuffer(aclTensor* tensor, void* deviceAddr)
{
if (tensor != nullptr) {
aclDestroyTensor(tensor);
}
if (deviceAddr != nullptr) {
aclrtFree(deviceAddr);
}
}
int RunReduceCase(const ReduceCase& reduceCase, aclrtStream stream)
{
const std::vector<int64_t> selfShape = {4, 4};
const std::vector<int64_t> indexShape = {3, 4};
const std::vector<int64_t> srcShape = {4, 4};
const std::vector<int64_t> outShape = {4, 4};
const int64_t dim = 0;
void* selfDeviceAddr = nullptr;
void* indexDeviceAddr = nullptr;
void* srcDeviceAddr = nullptr;
void* outDeviceAddr = nullptr;
void* workspaceAddr = nullptr;
aclTensor* self = nullptr;
aclTensor* index = nullptr;
aclTensor* src = nullptr;
aclTensor* out = nullptr;
std::vector<float> outHostData(GetShapeSize(outShape), 0);
auto ret = CreateAclTensor(reduceCase.selfHostData, selfShape, &selfDeviceAddr, aclDataType::ACL_FLOAT, &self);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(reduceCase.indexHostData, indexShape, &indexDeviceAddr, aclDataType::ACL_INT64, &index);
CHECK_RET(ret == ACL_SUCCESS, DestroyTensorAndBuffer(self, selfDeviceAddr); return ret);
ret = CreateAclTensor(reduceCase.srcHostData, srcShape, &srcDeviceAddr, aclDataType::ACL_FLOAT, &src);
CHECK_RET(ret == ACL_SUCCESS, DestroyTensorAndBuffer(self, selfDeviceAddr);
DestroyTensorAndBuffer(index, indexDeviceAddr); return ret);
ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
CHECK_RET(ret == ACL_SUCCESS, DestroyTensorAndBuffer(self, selfDeviceAddr);
DestroyTensorAndBuffer(index, indexDeviceAddr); DestroyTensorAndBuffer(src, srcDeviceAddr); return ret);
uint64_t workspaceSize = 0;
aclOpExecutor* executor = nullptr;
ret = aclnnScatterReduceGetWorkspaceSize(self, dim, index, src, reduceCase.reduce, reduceCase.includeSelf, out,
&workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS,
LOG_PRINT("aclnnScatterReduceGetWorkspaceSize failed for %s. ERROR: %d\n", reduceCase.name.c_str(), ret);
DestroyTensorAndBuffer(self, selfDeviceAddr); DestroyTensorAndBuffer(index, indexDeviceAddr);
DestroyTensorAndBuffer(src, srcDeviceAddr); DestroyTensorAndBuffer(out, outDeviceAddr); return ret);
if (workspaceSize > 0) {
ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
CHECK_RET(ret == ACL_SUCCESS,
LOG_PRINT("allocate workspace failed for %s. ERROR: %d\n", reduceCase.name.c_str(), ret);
DestroyTensorAndBuffer(self, selfDeviceAddr); DestroyTensorAndBuffer(index, indexDeviceAddr);
DestroyTensorAndBuffer(src, srcDeviceAddr); DestroyTensorAndBuffer(out, outDeviceAddr); return ret);
}
ret = aclnnScatterReduce(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(
ret == ACL_SUCCESS, LOG_PRINT("aclnnScatterReduce failed for %s. ERROR: %d\n", reduceCase.name.c_str(), ret);
DestroyTensorAndBuffer(self, selfDeviceAddr); DestroyTensorAndBuffer(index, indexDeviceAddr);
DestroyTensorAndBuffer(src, srcDeviceAddr); DestroyTensorAndBuffer(out, outDeviceAddr);
if (workspaceAddr != nullptr) { aclrtFree(workspaceAddr); } return ret);
ret = aclrtSynchronizeStream(stream);
CHECK_RET(
ret == ACL_SUCCESS,
LOG_PRINT("aclrtSynchronizeStream failed for %s. ERROR: %d\n", reduceCase.name.c_str(), ret);
DestroyTensorAndBuffer(self, selfDeviceAddr); DestroyTensorAndBuffer(index, indexDeviceAddr);
DestroyTensorAndBuffer(src, srcDeviceAddr); DestroyTensorAndBuffer(out, outDeviceAddr);
if (workspaceAddr != nullptr) { aclrtFree(workspaceAddr); } return ret);
auto size = GetShapeSize(outShape);
std::vector<float> localResultData(size, 0);
ret = aclrtMemcpy(localResultData.data(), localResultData.size() * sizeof(localResultData[0]), outDeviceAddr,
size * sizeof(localResultData[0]), ACL_MEMCPY_DEVICE_TO_HOST);
CHECK_RET(
ret == ACL_SUCCESS,
LOG_PRINT("copy result from device to host failed for %s. ERROR: %d\n", reduceCase.name.c_str(), ret);
DestroyTensorAndBuffer(self, selfDeviceAddr); DestroyTensorAndBuffer(index, indexDeviceAddr);
DestroyTensorAndBuffer(src, srcDeviceAddr); DestroyTensorAndBuffer(out, outDeviceAddr);
if (workspaceAddr != nullptr) { aclrtFree(workspaceAddr); } return ret);
LOG_PRINT("[%s] reduce=%s includeSelf=%d\n", reduceCase.name.c_str(), GetReduceName(reduceCase.reduce),
reduceCase.includeSelf);
for (int64_t i = 0; i < size; i++) {
LOG_PRINT("result[%ld] is: %f\n", i, localResultData[i]);
}
DestroyTensorAndBuffer(self, selfDeviceAddr);
DestroyTensorAndBuffer(index, indexDeviceAddr);
DestroyTensorAndBuffer(src, srcDeviceAddr);
DestroyTensorAndBuffer(out, outDeviceAddr);
if (workspaceAddr != nullptr) {
aclrtFree(workspaceAddr);
}
return 0;
}
int main()
{
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);
const std::vector<int64_t> indexHostData = {0, 1, 2, 1, 0, 1, 2, 0, 2, 2, 1, 0};
const std::vector<ReduceCase> cases = {
{"scatter_reduce_none",
kReduceNone,
true,
std::vector<float>(16, 3),
indexHostData,
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}},
{"scatter_reduce_add",
kReduceAdd,
true,
std::vector<float>(16, 3),
indexHostData,
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}},
{"scatter_reduce_max",
kReduceMax,
true,
std::vector<float>(16, 3),
indexHostData,
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}},
{"scatter_reduce_min",
kReduceMin,
true,
std::vector<float>(16, 3),
indexHostData,
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}},
{"scatter_reduce_mean",
kReduceMean,
true,
std::vector<float>(16, 3),
indexHostData,
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}},
{"scatter_reduce_mul",
kReduceMul,
true,
std::vector<float>(16, 2),
indexHostData,
{1, 2, 3, 4, 2, 3, 4, 5, 3, 4, 5, 6, 4, 5, 6, 7}},
};
for (const auto& reduceCase : cases) {
ret = RunReduceCase(reduceCase, stream);
if (ret != ACL_SUCCESS) {
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return ret;
}
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}
