aclnnMoeTokenPermuteWithEp
支持的产品型号
Atlas A2 训练系列产品/Atlas 800I A2 推理产品/A200I A2 Box 异构组件 Atlas A3 训练系列产品/Atlas A3 推理系列产品
功能说明
- 算子功能:MoE的permute计算,根据索引indices将tokens和可选probs广播后排序并按照rangeOptional中范围切片。
- 计算公式:
paddedModeOptional
false
时当rangeOptional[0] <= sortedIndices[i] < rangeOptional[1]时
paddedModeOptional为
true
时
函数原型
每个算子分为两段式接口,必须先调用 “aclnnMoeTokenPermuteWithEpGetWorkspaceSize”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“aclnnMoeTokenPermuteWithEp”接口执行计算。
aclnnStatus aclnnMoeTokenPermuteWithEpGetWorkspaceSize(const aclTensor *tokens, const aclTensor *indices, const aclTensor *probsOptional, const aclIntArray *rangeOptional, int64_t numOutTokensOptional, bool paddedModeOptional, aclTensor *permuteTokensOut, aclTensor *sortedIndicesOut, aclTensor *permuteProbsOut, uint64_t *workspaceSize, aclOpExecutor **executor)
aclnnStatus aclnnMoeTokenPermutewithEp(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream)
aclnnMoeTokenPermuteWithEpGetWorkspaceSize
参数说明:
- tokens(aclTensor *,计算输入):表示permute中的输入tokens,公式中的
tokens
,Device侧的aclTensor。shape支持2D维度,shape为(num_tokens,hidden_size),num_tokens为tokens的数目,hidden_size为每个tokens的长度。数据类型支持BFLOAT16、FLOAT16、FLOAT32,数据格式支持ND。支持非连续的Tensor,不支持空Tensor。 - indices(aclTensor *,计算输入):表述输入tokens对应的专家索引,公式中的
indices
,Device侧的aclTensor。shape支持1D或2D维度。paddedMode为false时表示每一个输入token对应的topK个处理专家索引,shape为(num_tokens,topK_num)或(num_tokens),paddedModeOptional为true时表示每个专家选中的token索引(暂不支持)。要求元素个数小于2134372523,值大于等于0小于2134372523,topK_num小于等于512。数据类型支持INT32、INT64,数据格式支持ND。支持非连续的Tensor,不支持空Tensor。 - probsOptional(aclTensor *,计算输入):表示输入tokens对应的专家概率,公式中的
probsOptional
,Device侧的aclTensor。可选计算输入,与计算输出permuteProbsOut对应,传入空则不输出permuteProbsOut。shape支持元素个数与indices相同。数据类型支持BFLOAT16、FLOAT16、FLOAT32,数据格式支持ND。支持非连续的Tensor。 - rangeOptional(aclIntArray *,计算输入):ep切分的有效范围,size为2。为空时,忽略probsOptional和permuteTokensOut,执行逻辑回退到aclnnMoeTokenPermute。
- numOutTokens(int64_t,计算输入):有效输出token数,在rangeOptional为空时生效。设置为0时,表示不会删除任何token。不为0时,会按照numOutTokens进行切片丢弃按照indices排序好的token中超过numOutTokens的部分,为负数时按照切片索引为负数时处理。
- paddedModeOptional(bool,计算输入):paddedModeOptional为true时表示indices已被填充为代表每个专家选中的token索引,此时不对indices进行排序。目前仅支持paddedModeOptional为false。
- permuteTokensOut(aclTensor *,计算输出):表示根据indices进行扩展并排序过的tokens,公式中的
permuteTokensOut
,Device侧的aclTensor。shape支持2D维度,shape为(rangeOptional[1] - rangeOptional[0],hidden_size)。数据类型同tokens,数据格式支持ND。 - sortedIndicesOut(aclTensor *,计算输出):表示permuteTokensOut和tokens的映射关系,公式中的
sortedIndicesOut
,Device侧的aclTensor。shape支持1D维度,Shape为(num_tokens * topK_num)。数据类型支持INT32,数据格式要求为ND。 - permuteProbsOut(aclTensor *,计算输出):表示根据indices进行扩展并排序过的probs,公式中的
permuteTokensOut
,Device侧的aclTensor。shape支持1D维度,shape为(rangeOptional[1] - rangeOptional[0])。数据类型同tokens,数据格式要求为ND。 - 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。
- tokens(aclTensor *,计算输入):表示permute中的输入tokens,公式中的
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: 161001(ACLNN_ERR_PARAM_NULLPTR):1. 输入和输出的Tensor是空指针。 161002(ACLNN_ERR_PARAM_INVALID):1. 输入和输出的数据类型不在支持的范围内。
aclnnMoeTokenPermuteWithEp
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aclnnMoeTokenPermuteWithEpGetWorkspaceSize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。
约束说明
- indices 要求元素个数小于
2134372523
,值大于等于0
小于2134372523
(值不在范围内排序结果不正确),第二维小于512
。 - 不支持paddedModeOptional为
True
。 - 当rangeOptional为空时,忽略probsOptional和permuteTokensOut,执行逻辑回退到aclnnMoeTokenPermute。
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_token_permute_with_ep.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 CreateAclIntArray(const std::vector<T>& hostData, void** deviceAddr, aclIntArray** intArray) {
auto size = GetShapeSize(hostData) * sizeof(T);
// Call aclrtMalloc to allocate memory on the device.
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);
// Call aclrtMemcpy to copy the data on the host to the memory on the device.
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);
// Call aclCreateIntArray to create an aclIntArray.
*intArray = aclCreateIntArray(hostData.data(), hostData.size());
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的接口定义构造
std::vector<int64_t> xShape = {3, 4};
std::vector<int64_t> idxShape = {3, 2};
std::vector<int64_t> probsShape = {3, 2};
std::vector<int64_t> expandedXOutShape = {4, 4};
std::vector<int64_t> idxOutShape = {6};
std::vector<int64_t> expandedProbsOutShape = {4};
void* xDeviceAddr = nullptr;
void* indicesDeviceAddr = nullptr;
void* probsDeviceAddr = nullptr;
void* expandedXOutDeviceAddr = nullptr;
void* sortedIndicesOutDeviceAddr = nullptr;
void* expandedProbsOutDeviceAddr = nullptr;
void* rangeDeviceAddr = nullptr;
aclTensor* x = nullptr;
aclTensor* indices = nullptr;
aclTensor* probs = nullptr;
aclIntArray* range = nullptr;
int64_t numTokenOut = 6;
bool padMode = false;
aclTensor* expandedXOut = nullptr;
aclTensor* sortedIndicesOut = nullptr;
aclTensor* expandedProbsOut = nullptr;
std::vector<float> xHostData = {0.1, 0.1, 0.1, 0.1, 0.2, 0.2, 0.2, 0.2, 0.3, 0.3, 0.3, 0.3};
std::vector<int> indicesHostData = {1, 2, 3, 1, 2, 3};
std::vector<float> probsHostData = {0.5, 0.3, 0.4, 0.2, 0.5, 0.4};
std::vector<float> expandedXOutHostData = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
std::vector<int> sortedIndicesOutHostData = {0, 0, 0, 0, 0, 0};
std::vector<float> expandedProbsOutHostData = {0, 0, 0, 0};
std::vector<int64_t> rangeHostData = {1, 5};
// 创建self aclTensor
ret = CreateAclTensor(xHostData, xShape, &xDeviceAddr, aclDataType::ACL_BF16, &x);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(indicesHostData, idxShape, &indicesDeviceAddr, aclDataType::ACL_INT32, &indices);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(probsHostData, probsShape, &probsDeviceAddr, aclDataType::ACL_BF16, &probs);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建out aclTensor
ret = CreateAclTensor(expandedXOutHostData, expandedXOutShape, &expandedXOutDeviceAddr, aclDataType::ACL_BF16, &expandedXOut);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(sortedIndicesOutHostData, idxOutShape, &sortedIndicesOutDeviceAddr, aclDataType::ACL_INT32, &sortedIndicesOut);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(expandedProbsOutHostData, expandedProbsOutShape, &expandedProbsOutDeviceAddr, aclDataType::ACL_BF16, &expandedProbsOut);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 创建相关attr
ret = CreateAclIntArray(rangeHostData, &rangeDeviceAddr, &range);
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aclnnMoeTokenPermute第一段接口
ret = aclnnMoeTokenPermuteWithEpGetWorkspaceSize(x, indices, probs, range, numTokenOut, padMode, expandedXOut, sortedIndicesOut, expandedProbsOut, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMoeTokenPermuteGetWorkspaceSize 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aclnnMoeTokenPermute第二段接口
ret = aclnnMoeTokenPermuteWithEp(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnMoeTokenPermute 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 expandedXSize = GetShapeSize(expandedXOutShape);
std::vector<float> expandedXData(expandedXSize, 0);
ret = aclrtMemcpy(expandedXData.data(), expandedXData.size() * sizeof(expandedXData[0]), expandedXOutDeviceAddr, expandedXSize * 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 < expandedXSize; i++) {
LOG_PRINT("expandedXData[%ld] is: %f\n", i, expandedXData[i]);
}
auto sortedIndicesSize = GetShapeSize(idxOutShape);
std::vector<int> sortedIndicesData(sortedIndicesSize, 0);
ret = aclrtMemcpy(sortedIndicesData.data(), sortedIndicesData.size() * sizeof(sortedIndicesData[0]), sortedIndicesOutDeviceAddr, sortedIndicesSize * sizeof(int32_t),
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 < sortedIndicesSize; i++) {
LOG_PRINT("sortedIndicesData[%ld] is: %d\n", i, sortedIndicesData[i]);
}
auto expandedProbsSize = GetShapeSize(expandedProbsOutShape);
std::vector<float> expandedProbsData(expandedProbsSize, 0);
ret = aclrtMemcpy(expandedProbsData.data(), expandedProbsData.size() * sizeof(expandedProbsData[0]), expandedProbsOutDeviceAddr, expandedProbsSize * 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 < expandedProbsSize; i++) {
LOG_PRINT("expandedProbsData[%ld] is: %f\n", i, expandedProbsData[i]);
}
// 6. 释放aclTensor和aclScalar,需要根据具体API的接口定义修改
aclDestroyTensor(x);
aclDestroyTensor(indices);
aclDestroyTensor(probs);
aclDestroyTensor(expandedXOut);
aclDestroyTensor(sortedIndicesOut);
aclDestroyTensor(expandedProbsOut);
// 7. 释放device资源,需要根据具体API的接口定义修改
aclrtFree(xDeviceAddr);
aclrtFree(indicesDeviceAddr);
aclrtFree(probsDeviceAddr);
aclrtFree(expandedXOutDeviceAddr);
aclrtFree(sortedIndicesOutDeviceAddr);
aclrtFree(expandedProbsOutDeviceAddr);
aclrtFree(rangeDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}