aclnnQuantMatmulAllReduceV3

支持的产品型号

• Atlas A2训练系列产品/Atlas 800I A2推理产品。

说明： 使用该接口时，请确保驱动固件包和CANN包都为配套的8.0.RC2版本或者配套的更高版本，否则将会引发报错，比如BUS ERROR等。

接口原型

每个算子分为两段式接口，必须先调用“aclnnQuantMatmulAllReduceV3GetWorkspaceSize”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“aclnnQuantMatmulAllReduceV3”接口执行计算。

• aclnnStatus aclnnQuantMatmulAllReduceV3GetWorkspaceSize(const aclTensor *x1, const aclTensor *x2, const aclTensor *biasOptional, const aclTensor *x3Optional, const aclTensor *dequantScale, const aclTensor *pertokenScaleOptional, const aclTensor *commQuantScale1Optional, const aclTensor *commQuantScale2Optional, const char* group, const char *reduceOp, int64_t commTurn, int64_t streamMode, const aclTensor *output, uint64_t *workspaceSize, aclOpExecutor **executor);
• aclnnStatus aclnnQuantMatmulAllReduceV3(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, aclrtStream stream);

功能描述

• 算子功能：对量化后的入参x1、x2进行matmul、dequant和pertoken计算，接着与x3进行add操作，再对输出进行perchannel量化，然后进行all_to_all通信，对第一次通讯结果进行reduceSum计算，接着进行all_gather通信，最后对第二次通信结果进行dequant，得到最终输出。
• 计算公式：$$matmulAddOutPut = (dequantScale * pertokenScaleOptional * (x1_{int8}@x2_{int8} + biasOptional_{int32}) + x3Optional);$$ $$alltoallOutPut_{int8} = alltoall(matmulAddOutPut / commQuantScale1Optional);$$ $$reduceSumOutPut_{int8} = (add(alltoallOutPut_{int8}) * (commQuantScale1Optional / commQuantScale2Optional));$$ $$outPut = (allgather(reduceSumOutPut_{int8}) * commQuantScale2Optional);$$

aclnnQuantMatmulAllReduceV3GetWorkspaceSize

• 参数说明：

  • x1（aclTensor*, 计算输入)：公式中的输入x1，数据类型支持INT8，数据格式支持ND。Device侧的aclTensor，mm左矩阵，不支持非连续输入。当前版本仅支持二维或者三维输入。
  • x2（aclTensor*, 计算输入)：公式中的输入x2，数据类型支持INT8，Atlas A2训练系列产品/Atlas 800I A2推理产品数据格式支持ND和FRACTAL_NZ格式。Device侧的aclTensor，mm右矩阵，输入的shape规则如下：
    • 当x2的Format为FRACTAL_NZ时，配合aclnnCalculateMatmulWeightSizeV2和aclnnTransMatmulWeight完成输入ND到NZ的转换，非连续的tensor仅支持transpose场景。
    • 当x2的Format为ND时，当前版本仅支持二维输入。
    • 当x2的Format为FRACTAL_NZ时，当前版本仅支持四维输入。
  • biasOptional（aclTensor*, 计算输入)：公式中的输入biasOptional，数据类型支持INT32，数据格式支持ND。Device侧的aclTensor，对应计算公式中bias偏移。可选，可为空。当前版本仅支持一维输入。
  • x3Optional（aclTensor*, 计算输入)：公式中的输入x3Optional，Atlas A2训练系列产品/Atlas 800I A2推理产品数据类型支持FLOAT16、BFLOAT16，数据格式支持ND。Device侧的aclTensor，mm计算后的add计算，维度与output一致。目前仅输出为BFLOAT16场景，且仅支持非空输入。
  • dequantScale（aclTensor*, 计算输入)：公式中的输入dequantScale，数据类型支持INT64、UINT64、FLOAT32、BFLOAT16，数据格式支持ND。mm计算后的去量化系数。shape在per-tensor场景为[1]，per-channel场景为[n]/[1, n]。
    • 输出为BFLOAT16时，直接将BFLOAT16类型的dequantScale传入本接口；
    • 输出为FLOAT16时，如果pertokenScale不为空，可直接将FLOAT32类型的dequantScale传入本接口，如果pertokenScale为空，则需提前调用TransQuantParamV2算子的aclnn接口来将dequantScale转成INT64/UINT64数据类型。
  • pertokenScaleOptional（aclTensor*, 计算输入)：公式中的输入pertokenScaleOptional，mm计算后的pertoken去量化系数。可选，可为空，数据类型支持FLOAT32，数据格式支持ND。x1为[b, s, k]时shape为[b*s]，x1为[m, k]时shape为[m]。
  • commQuantScale1Optional（aclTensor*, 计算输入)：公式中的输入commQuantScale1，matmulAdd计算后的perchannel量化系数。Device侧的aclTensor，可选，可为空，数据类型支持：BFLOAT16、FLOAT16，数据格式支持：ND, x2为[k, n]时, shape可为[n]或者[1,n]。
  • commQuantScale2Optional（aclTensor*, 计算输入)：公式中的输入commQuantScale2，allGather计算后的perchannel量化系数。Device侧的aclTensor，可选，可为空，数据类型支持：BFLOAT16、FLOAT16，数据格式支持：ND, x2为[k, n]时, shape可为[n]或者[1,n]。
  • group(char*, 计算输入)：通信域名称。数据类型支持String。通过Hccl提供的接口获取：extern HcclResult HcclGetCommName(HcclComm comm, char* commName); commName即为group。
  • reduceOp(char*, 计算输入)：reduce操作类型，数据类型支持String。目前仅支持"sum"。
  • commTurn(int64_t, 计算输入)：通信数据切分数，数据类型支持int64_t。即总数据量/单次通信量。当前版本仅支持输入为0。
  • streamMode(int64_t, 计算输入)：Host侧的整型，数据类型支持int64_t。AscendCL流模式的枚举，当前只支持枚举值1。
  • output(aclTensor *, 输出)：计算+通信的结果，数据类型支持FLOAT16、BFLOAT16，维度为(b, s, n)或者(m, n)。
  • 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。

aclnnQuantMatmulAllReduceV3

• 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aclnnQuantMatmulAllReduceV3GetWorkspaceSize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：

  返回aclnnStatus状态码，具体参见aclnn返回码。

约束与限制

• 增量场景不使能MC2，全量场景使能MC2。
• 输入x1可为2维或者3维，且不为空Tensor，其维度为(b, s, k)或者(m, k)。x2必须是2维，且不为空Tensor。其维度为(k, n)，k轴满足mm算子入参要求，k轴相等。
• m大小不超过2147483647，x1与x2的最后一维大小不超过65535，x1的最后一维指k，x2的最后一维指转置时的k或非转置时的n。
• 传入的x1、x2、dequantScale或者output不为空指针。
• x1和x2、dequantScale、output、bias（非空场景）、x3（非空场景）的数据类型和数据格式需要在支持的范围之内。
• 若输出output类型为FLOAT16，当pertokenSclale为空时，dequantScale的类型为INT64、UINT64，当pertokenSclale不为空时，dequantScale的类型为FLOAT32；若输出output类型为BFLOAT16,dequantScale的类型为BFLOAT16，x3的类型为BFLOAT16。
• 传入的commQuantScale1与commQuantScale2需要同时为空指针或同时不为空指针，若传入的commQuantScale1与commQuantScale2同时不为空指针，两个量化参数shape需保持一致，类型需与算子输出类型保持一致，且每张卡输入保持一致。
• x1维度为[b, s, k]时pertokenScaleOptional维度为[b*s]，x1维度为[m, k]时pertokenScaleOptional维度为[m]。
• streamMode的数据在可选范围内，目前仅支持1。
• 只支持x2矩阵转置/不转置，x1矩阵不支持转置场景。
• Atlas A2训练系列产品/Atlas 800I A2推理产品支持1、2、4、8卡，并且仅支持hccs链路all mesh组网。
• 不支持空tensor。
• 一个模型中的通算融合MC2算子，仅支持相同通信域。

调用示例

#include <iostream>
#include <vector>
#include <thread>
#include "aclnnop/aclnn_trans_matmul_weight.h"
#include "aclnnop/aclnn_quant_matmul_all_reduce_v3.h"

int ndev = 8;

#define ACL_CHECK(ret)                                                                                     \
    do {                                                                                                   \
        auto retcode = ret;                                                                                \
        if (retcode != ACL_SUCCESS) {                                                                      \
            printf("[ERROR] acl interface return err %s:%d, retcode: %d \n", __FILE__, __LINE__, retcode); \
            return retcode;                                                                                \
        }                                                                                                  \
    } while (0)

#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;
}

struct Args {
    uint32_t rankId;
    HcclComm hcclComm;
    aclrtStream stream;
    aclrtContext context;
    std::string format;
  };

template<typename T>
int CreateWeightNzAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,
                            aclDataType dataType, aclTensor **tensor, Args &args) {
    auto size = GetShapeSize(shape) * sizeof(T);
    const aclIntArray *mat2Size = aclCreateIntArray(shape.data(), shape.size());
    auto ret = aclnnCalculateMatmulWeightSizeV2(mat2Size, ACL_INT8, &size);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCalculateMatmulWeightSizeV2 failed. ERROR: %d\n", ret); return ret);
    auto tensorSize = size * sizeof(T);

    // 调用aclrtMalloc申请device内存
    ret = aclrtMalloc(deviceAddr, tensorSize, 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);
    
    uint64_t transWorkspaceSize;
    aclOpExecutor *executor;
    void *transWorkspaceAddr = nullptr;
    ret = aclnnTransMatmulWeightGetWorkspaceSize(*tensor, &transWorkspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS && transWorkspaceSize > 0, 
              printf("[ERROR] aclnnTransMatmulWeightGetWorkspaceSize failed. ret = %d \n", ret); return ret);
    ACL_CHECK(aclrtMalloc(&transWorkspaceAddr, transWorkspaceSize, ACL_MEM_MALLOC_HUGE_FIRST));
    ret = aclnnTransMatmulWeight(transWorkspaceAddr, transWorkspaceSize, executor, args.stream);
    CHECK_RET(ret == ACL_SUCCESS, printf("[ERROR] aclnnTransMatmulWeight failed. ret = %d \n", ret);return ret);
    ACL_CHECK(aclrtSynchronizeStreamWithTimeout(args.stream, 20000));

    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 launchOneThreadQuantMatmulAllReduce(Args &args) {
    int ret;
    ret = aclrtSetCurrentContext(args.context);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetCurrentContext failed. ERROR: %d\n", ret); return ret);
    char hcom_name[128];
    ret = HcclGetCommName(args.hcclComm, hcom_name);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("[ERROR] HcclGetCommName failed. ret = %d \n", ret); return -1);
    LOG_PRINT("[INFO] rank %d hcom: %s stream: %p, context : %p\n", args.rankId, hcom_name, args.stream,
              args.context);

    std::vector<int64_t> x1Shape = {32, 64};
    std::vector<int64_t> x2Shape = {64, 128};
    std::vector<int64_t> biasShape = {128};
    std::vector<int64_t> dequantScaleShape = {128};
    std::vector<int64_t> pertokenScaleShape = {32};
    std::vector<int64_t> commQuantScale1Shape = {128};
    std::vector<int64_t> commQuantScale2Shape = {128};
    std::vector<int64_t> x3Shape = {32, 128};
    std::vector<int64_t> outShape = {32, 128};
    void *x1DeviceAddr = nullptr;
    void *x2DeviceAddr = nullptr;
    void *biasDeviceAddr = nullptr;
    void *dequantScaleDeviceAddr = nullptr;
    void *pertokenScaleDeviceAddr = nullptr;
    void *commQuantScale1DeviceAddr = nullptr;
    void *commQuantScale2DeviceAddr = nullptr;
    void *x3DeviceAddr = nullptr;
    void *outDeviceAddr = nullptr;
    aclTensor *x1 = nullptr;
    aclTensor *x2 = nullptr;
    aclTensor *bias = nullptr;
    aclTensor *dequantScale = nullptr;
    aclTensor *pertokenScale = nullptr;
    aclTensor *commQuantScale1 = nullptr;
    aclTensor *commQuantScale2 = nullptr;
    aclTensor *x3 = nullptr;
    aclTensor *out = nullptr;

    int64_t commTurn = 0;
    int64_t streamMode = 1;
    uint64_t workspaceSize = 0;
    aclOpExecutor *executor;
    void *workspaceAddr = nullptr;

    long long x1ShapeSize = GetShapeSize(x1Shape);
    long long x2ShapeSize = GetShapeSize(x2Shape);
    long long biasShapeSize = GetShapeSize(biasShape);
    long long dequantScaleShapeSize = GetShapeSize(dequantScaleShape);
    long long pertokenScaleShapeSize = GetShapeSize(pertokenScaleShape);
    long long commQuantScale1ShapeSize = GetShapeSize(commQuantScale1Shape);
    long long commQuantScale2ShapeSize = GetShapeSize(commQuantScale2Shape);
    long long x3ShapeSize = GetShapeSize(x3Shape);
    long long outShapeSize = GetShapeSize(outShape);

    std::vector<int8_t> x1HostData(x1ShapeSize, 0);
    std::vector<int8_t> x2HostData(x2ShapeSize, 0);
    std::vector<int32_t> biasHostData(biasShapeSize, 0);
    std::vector<uint64_t> dequantScaleHostData(dequantScaleShapeSize, 0);
    std::vector<uint64_t> pertokenScaleHostData(pertokenScaleShapeSize, 0);
    std::vector<uint64_t> commQuantScale1HostData(commQuantScale1ShapeSize, 0);
    std::vector<uint64_t> commQuantScale2HostData(commQuantScale2ShapeSize, 0);
    std::vector<int16_t> x3HostData(x3ShapeSize, 0);
    std::vector<int16_t> outHostData(outShapeSize, 0);
    // 创建 tensor
    ret = CreateAclTensor(x1HostData, x1Shape, &x1DeviceAddr, aclDataType::ACL_INT8, &x1);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    if (args.format == "NZ") {
        ret = CreateWeightNzAclTensor(x2HostData, x2Shape, &x2DeviceAddr, aclDataType::ACL_INT8, &x2, args);
    } else {
        ret = CreateAclTensor(x2HostData, x2Shape, &x2DeviceAddr, aclDataType::ACL_INT8, &x2);
    }
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(biasHostData, biasShape, &biasDeviceAddr, aclDataType::ACL_INT32, &bias);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(dequantScaleHostData, dequantScaleShape, &dequantScaleDeviceAddr,
                          aclDataType::ACL_FLOAT, &dequantScale);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(pertokenScaleHostData, pertokenScaleShape, &pertokenScaleDeviceAddr,
                          aclDataType::ACL_FLOAT, &pertokenScale);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(commQuantScale1HostData, commQuantScale1Shape, &commQuantScale1DeviceAddr,
                          aclDataType::ACL_FLOAT16, &commQuantScale1);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(commQuantScale2HostData, commQuantScale2Shape, &commQuantScale2DeviceAddr,
                          aclDataType::ACL_FLOAT16, &commQuantScale2);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(x3HostData, x3Shape, &x3DeviceAddr, aclDataType::ACL_FLOAT16, &x3);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT16, &out);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // 调用第一段接口
    ret = aclnnQuantMatmulAllReduceV3GetWorkspaceSize(x1, x2, bias, x3, dequantScale, pertokenScale,
                                                      commQuantScale1, commQuantScale2, hcom_name,
                                                      "sum", commTurn, streamMode, out,
                                                      &workspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS,
              LOG_PRINT("aclnnQuantMatmulAllReduceV3GetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
    // 根据第一段接口计算出的workspaceSize申请device内存
    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);
    }
    // 调用第二段接口
    ret = aclnnQuantMatmulAllReduceV3(workspaceAddr, workspaceSize, executor, args.stream);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnQuantMatmulAllReduceV3 failed. ERROR: %d\n", ret); return ret);
    //（固定写法）同步等待任务执行结束
    ret = aclrtSynchronizeStreamWithTimeout(args.stream, 10000);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
    LOG_PRINT("device%d aclnnQuantMatmulAllReduceV3 execute success \n", args.rankId);
    // 释放device资源，需要根据具体API的接口定义修改
    if (x1 != nullptr) {
        aclDestroyTensor(x1);
    }
    if (x2 != nullptr) {
        aclDestroyTensor(x2);
    }
    if (bias != nullptr) {
        aclDestroyTensor(bias);
    }
    if (dequantScale != nullptr) {
        aclDestroyTensor(dequantScale);
    }
    if (pertokenScale != nullptr) {
        aclDestroyTensor(pertokenScale);
    }
    if (commQuantScale1 != nullptr) {
        aclDestroyTensor(commQuantScale1);
    }
    if (commQuantScale2 != nullptr) {
        aclDestroyTensor(commQuantScale2);
    }
    if (x3 != nullptr) {
        aclDestroyTensor(x3);
    }
    if (out != nullptr) {
        aclDestroyTensor(out);
    }
    if (x1DeviceAddr != nullptr) {
        aclrtFree(x1DeviceAddr);
    }
    if (x2DeviceAddr != nullptr) {
        aclrtFree(x2DeviceAddr);
    }
    if (biasDeviceAddr != nullptr) {
        aclrtFree(biasDeviceAddr);
    }
    if (dequantScaleDeviceAddr != nullptr) {
        aclrtFree(dequantScaleDeviceAddr);
    }
    if (pertokenScaleDeviceAddr != nullptr) {
        aclrtFree(pertokenScaleDeviceAddr);
    }
    if (commQuantScale1DeviceAddr != nullptr) {
        aclrtFree(commQuantScale1DeviceAddr);
    }
    if (commQuantScale2DeviceAddr != nullptr) {
        aclrtFree(commQuantScale2DeviceAddr);
    }
    if (x3DeviceAddr != nullptr) {
        aclrtFree(x3DeviceAddr);
    }
    if (outDeviceAddr != nullptr) {
        aclrtFree(outDeviceAddr);
    }
    if (workspaceSize > 0) {
        aclrtFree(workspaceAddr);
    }
    aclrtDestroyStream(args.stream);
    HcclCommDestroy(args.hcclComm);
    aclrtDestroyContext(args.context);
    aclrtResetDevice(args.rankId);
    return 0;
}

int main(int argc, char *argv[]) {
    int ret;
    int32_t devices[ndev];
    for (int i = 0; i < ndev; i++) {
        devices[i] = i;
    }
    HcclComm comms[128];
    ret = aclInit(nullptr);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret); return ret);
    // 初始化集合通信域
    for (int i = 0; i < ndev; i++) {
        ret = aclrtSetDevice(devices[i]);
        CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret);
    }
    ret = HcclCommInitAll(ndev, devices, comms);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("HcclCommInitAll failed. ERROR: %d\n", ret); return ret);
    Args args[ndev];
    aclrtStream stream[ndev];
    aclrtContext context[ndev];
    for (uint32_t rankId = 0; rankId < ndev; rankId++) {
        ret = aclrtSetDevice(rankId);
        CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret);
        ret = aclrtCreateContext(&context[rankId], rankId);
        CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateContext failed. ERROR: %d\n", ret); return ret);
        ret = aclrtCreateStream(&stream[rankId]);
        CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret); return ret);
    }
    // 启动多线程
    std::vector<std::unique_ptr<std::thread>> threads(ndev);
    for (uint32_t rankId = 0; rankId < ndev; rankId++) {
        args[rankId].rankId = rankId;
        args[rankId].hcclComm = comms[rankId];
        args[rankId].stream = stream[rankId];
        args[rankId].context = context[rankId];
        threads[rankId].reset(
                new(std::nothrow) std::thread(&launchOneThreadQuantMatmulAllReduce, std::ref(args[rankId])));
    }
    for (uint32_t rankId = 0; rankId < ndev; rankId++) {
        threads[rankId]->join();
    }
    aclFinalize();
    return 0;
}