aclnnMoeFinalizeRouting
支持的产品型号
- 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“aclnnMoeFinalizeRoutingGetWorkspaceSize”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“aclnnMoeFinalizeRouting”接口执行计算。
aclnnStatus aclnnMoeFinalizeRoutingGetWorkspaceSize(const aclTensor* expandedX, const aclTensor* x1, const aclTensor* x2Optional, const aclTensor* bias, const aclTensor* scales, const aclTensor* expandedRowIdx, const aclTensor* expandedExpertIdx, const aclTensor* out, uint64_t* workspaceSize, aclOpExecutor** executor)
aclnnStatus aclnnMoeFinalizeRouting(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, const aclrtStream stream)
功能描述
- 算子功能:MoE计算中,最后处理合并MoE FFN的输出结果。
- 计算公式:
aclnnMoeFinalizeRoutingGetWorkspaceSize
参数说明:
- expandedX (aclTensor*,计算输入):Device侧的aclTensor,公式中的expandedX ,MoE的FFN输出,要求是一个2D的Tensor,数据类型支持FLOAT16、BFLOAT16、FLOAT32,数据格式要求为ND。限制:其shape支持(NUM_ROWS * K, H),NUM_ROWS为行数,K为从总的专家E中选出K个专家,H为列数。
- x1(aclTensor*,计算输入):Device侧的aclTensor,公式中的x1,要求是一个2D的Tensor,数据类型要求与expandedX一致 ,shape要求与out的shape一致。
- x2Optional(aclTensor*,计算输入):Device侧的aclTensor,公式中的x2Optional,要求是一个2D的Tensor,数据类型要求与expandedX一致 ,shape要求与out的shape一致。
- bias(aclTensor*,计算输入):Device侧的aclTensor,公式中的bias,要求是一个2D的Tensor,数据类型要求与expandedX一致。限制:其shape支持(E,H),E为总的专家个数,H为列数。
- scales(aclTensor*,计算输入):Device侧的aclTensor,公式中的scales,要求是一个2D的Tensor,数据类型要求与expandedX一致 ,限制:其shape支持(NUM_ROWS,K)。
- expandedRowIdx(aclTensor*,计算输入):Device侧的aclTensor,公式中的expandedRowIdx,要求是一个1D的Tensor,数据类型支持INT32。限制:其shape支持(NUM_ROWS * K),Tensor中的值取值范围是[0,NUM_ROWS * K-1]。
- expandedExpertIdx(aclTensor*,计算输入):Device侧的aclTensor,公式中的expandedExpertIdx,要求是一个2D的Tensor,数据类型支持INT32。限制:其shape支持(NUM_ROWS,K),Tensor中的值取值范围是[0, E-1],E为总的专家个数。
- out(aclTensor*,计算输出):Device侧的aclTensor,公式中的输出,要求是一个2D的Tensor,数据类型与expandedX 需要保持一致。限制:其shape支持(NUM_ROWS, H)。
- 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:
返回aclnnStatus状态码,具体参见aclnn返回码。
161001 (ACLNN_ERR_PARAM_NULLPTR): 1. 传入的expandedX、x1、x2Optional、bias、scales、expandedRowIdx和expandedExpertIdx是空指针时。 161002 (ACLNN_ERR_PARAM_INVALID): 1. expandedX、x1、x2Optional、bias、scales、expandedRowIdx和expandedExpertIdx的数据类型不在支持的范围之内。 2. expandedX、x1、x2Optional、bias、scales、expandedRowIdx和expandedExpertIdx的shape不在支持的范围之内。
aclnnMoeFinalizeRouting
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aclnnMoeFinalizeRoutingGetWorkspaceSize获取。
- executor(aclOpExecutor*,入参):op执行器,包含了算子计算流程。
- stream(const aclrtStream,入参):指定执行任务的AscendCL stream流。
返回值:
返回aclnnStatus状态码,具体参见aclnn返回码。
约束与限制
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 "acl/acl.h"
#include "aclnnop/aclnn_moe_finalize_routing.h"
#include <iostream>
#include <vector>
#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 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;
}
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> expandedXShape = {3 * 2, 4};
std::vector<int64_t> x1Shape = {3, 4};
std::vector<int64_t> x2OptionalShape = {3, 4};
std::vector<int64_t> biasShape = {2, 4};
std::vector<int64_t> scalesShape = {3, 2};
std::vector<int64_t> expandedExpertIdxShape = {3, 2};
std::vector<int64_t> expandedRowIdxShape = {3 * 2};
std::vector<int64_t> outShape = {3, 4};
void* expandedXAddr = nullptr;
void* x1Addr = nullptr;
void* x2OptionalAddr = nullptr;
void* biasAddr = nullptr;
void* scalesDeviceAddr = nullptr;
void* expandedExpertIdxAddr = nullptr;
void* expandedRowIdxAddr = nullptr;
void* outDeviceAddr = nullptr;
aclTensor* expandedX = nullptr;
aclTensor* x1 = nullptr;
aclTensor* x2Optional = nullptr;
aclTensor* bias = nullptr;
aclTensor* scales = nullptr;
aclTensor* expandedExpertIdx = nullptr;
aclTensor* expandedRowIdx = nullptr;
aclTensor* out = nullptr;
std::vector<float> expandedXHostData = {0.1, 1.1, 2.1, 3.1, 4.1, 5.1, 6.1, 7.1, 8.1, 9.1, 10.1, 11.1,
0.1, 1.1, 2.1, 3.1, 4.1, 5.1, 6.1, 7.1, 8.1, 9.1, 10.1, 11.1};
std::vector<float> x1HostData = {0.2, 1.2, 2.2, 3.2, 4.2, 5.2, 6.2, 7.2, 8.2, 9.2, 10.2, 11.2};
std::vector<float> x2OptionalHostData = {0.2, 1.2, 2.2, 3.2, 4.2, 5.2, 6.2, 7.2, 8.2, 9.2, 10.2, 11.2};
std::vector<float> biasHostData = {0.2, 0.4, 0.2, 0.4, 0.2, 0.4, 0.2, 0.4};
std::vector<float> scalesHostData = {1.3, 1.6, 1.2, 1.8, 1.2, 2.3};
std::vector<int32_t> expandedExpertIdxHostData = {0, 1, 0, 1, 0, 1};
std::vector<int32_t> expandedRowIdxHostData = {2, 1, 4, 3, 0, 5};
std::vector<float> outHostData(12, 0.0f);
// 创建expandedX aclTensor
ret = CreateAclTensor(expandedXHostData, expandedXShape, &expandedXAddr,
aclDataType::ACL_FLOAT, &expandedX);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建x1 aclTensor
ret = CreateAclTensor(x1HostData, x1Shape, &x1Addr, aclDataType::ACL_FLOAT, &x1);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建x2Optional aclScalar
ret = CreateAclTensor(x2OptionalHostData, x2OptionalShape, &x2OptionalAddr, aclDataType::ACL_FLOAT, &x2Optional);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建bias aclTensor
ret = CreateAclTensor(biasHostData, biasShape, &biasAddr, aclDataType::ACL_FLOAT, &bias);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建totalWeightOut aclTensor
ret = CreateAclTensor(scalesHostData, scalesShape, &scalesDeviceAddr, aclDataType::ACL_FLOAT, &scales);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建expandedExpertIdx aclTensor
ret = CreateAclTensor(expandedExpertIdxHostData, expandedExpertIdxShape, &expandedExpertIdxAddr,
aclDataType::ACL_INT32, &expandedExpertIdx);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建expandedRowIdx aclTensor
ret = CreateAclTensor(expandedRowIdxHostData, expandedRowIdxShape, &expandedRowIdxAddr,
aclDataType::ACL_INT32, &expandedRowIdx);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建Out aclTensor
ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 3.调用CANN算子库API,需要修改为具体的算子接口
uint64_t workspaceSize = 0;
aclOpExecutor* executor;
// 调用aclnnMoeFinalizeRouting第一段接口
ret = aclnnMoeFinalizeRoutingGetWorkspaceSize(expandedX, x1, x2Optional, bias, scales,
expandedRowIdx, expandedExpertIdx, out, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMoeFinalizeRoutingGetWorkspaceSize 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aclnnMoeFinalizeRouting第二段接口
ret = aclnnMoeFinalizeRouting(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMoeFinalizeRouting 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, 0.0f);
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(expandedX);
aclDestroyTensor(x1);
aclDestroyTensor(x2Optional);
aclDestroyTensor(bias);
aclDestroyTensor(scales);
aclDestroyTensor(expandedExpertIdx);
aclDestroyTensor(expandedRowIdx);
aclDestroyTensor(out);
// 7. 释放device资源,需要根据具体API的接口定义修改
aclrtFree(expandedXAddr);
aclrtFree(x1Addr);
aclrtFree(x2OptionalAddr);
aclrtFree(biasAddr);
aclrtFree(scalesDeviceAddr);
aclrtFree(expandedExpertIdxAddr);
aclrtFree(expandedRowIdxAddr);
aclrtFree(outDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}