aclnn_gelu_operation.cpp

#include "aclnn_gelu_operation.h"
#include "acl/acl.h"
#include "aclnnop/aclnn_gelu.h"
#include "aclnnop/aclnn_gelu_v2.h"
#include "utils/log.h"
#include "utils/utils.h"

const int DIM0 = 0;
const int DIM1 = 1;
const int DIM2 = 2;
const int DIM3 = 3;

GeluOperation::GeluOperation(const std::string &name, AclnnGeluParam param) : AclnnBaseOperation(name), param_(param)
{}

atb::Status GeluOperation::InferShape(
    const atb::SVector<atb::TensorDesc> &inTensorDesc, atb::SVector<atb::TensorDesc> &outTensorDesc) const
{
    LOG_INFO(opName_ + " InferShape start");
    outTensorDesc.at(0).format = inTensorDesc.at(0).format;
    outTensorDesc.at(0).dtype = inTensorDesc.at(0).dtype;
    outTensorDesc.at(0).shape.dimNum = inTensorDesc.at(0).shape.dimNum;

    if (inTensorDesc.at(0).shape.dimNum == DIM3) {
        LOG_INFO("[input0 dimNum = 3] CHECK " + opName_ + " input shape: [input0] " +
                 std::to_string(inTensorDesc.at(0).shape.dims[DIM0]) + ", " +
                 std::to_string(inTensorDesc.at(0).shape.dims[DIM1]) + ", " +
                 std::to_string(inTensorDesc.at(0).shape.dims[DIM2]));
        outTensorDesc.at(0).shape.dims[DIM0] = inTensorDesc.at(0).shape.dims[DIM0];
        outTensorDesc.at(0).shape.dims[DIM1] = inTensorDesc.at(0).shape.dims[DIM1];
        outTensorDesc.at(0).shape.dims[DIM2] = inTensorDesc.at(0).shape.dims[DIM2];
    } else if (inTensorDesc.at(0).shape.dimNum == DIM2) {
        LOG_INFO("[input0 dimNum = 2] CHECK " + opName_ + " input shape: [input0] " +
                 std::to_string(inTensorDesc.at(0).shape.dims[DIM0]) + ", " +
                 std::to_string(inTensorDesc.at(0).shape.dims[DIM1]));
        outTensorDesc.at(0).shape.dims[DIM0] = inTensorDesc.at(0).shape.dims[DIM0];
        outTensorDesc.at(0).shape.dims[DIM1] = inTensorDesc.at(0).shape.dims[DIM1];
    } else {
        LOG_ERROR(opName_ + " invalid dimNum = " + std::to_string(inTensorDesc.at(0).shape.dimNum));
    }

    LOG_INFO(opName_ + " InferShape end");
    return atb::NO_ERROR;
}

uint32_t GeluOperation::GetInputNum() const
{
    return 1;  // gelu入参个数
}

uint32_t GeluOperation::GetOutputNum() const
{
    return 1;  // gelu出参个数
}

// 重写父类方法, 创建输入输出tensor，并存入VariantPack
atb::Status GeluOperation::CreateAclnnVariantPack(const atb::VariantPack &variantPack)
{
    LOG_INFO(opName_ + " CreateAclnnVariantPack start");

    auto ret = CreateAclnnInTensor(variantPack);
    if (ret != 0) {
        LOG_ERROR(opName_ + " CreateAclnnInTensor fail");
        return atb::ERROR_INVALID_PARAM;
    }

    ret = CreateAclnnOutTensor(variantPack);
    if (ret != 0) {
        LOG_ERROR(opName_ + " CreateAclNNOutTensorVariantPack fail");
        return atb::ERROR_INVALID_PARAM;
    }

    LOG_INFO(opName_ + " CreateAclnnVariantPack end");
    return atb::NO_ERROR;
}

atb::Status GeluOperation::CreateAclnnInTensor(const atb::VariantPack &variantPack)
{
    aclInTensors_.resize(GetInputNum());
    for (size_t i = 0; i < aclInTensors_.size(); ++i) {
        auto aclnnTensor = CreateAclnnTensor(variantPack.inTensors.at(i), i);
        if (aclnnTensor->tensor == nullptr) {
            LOG_ERROR(opName_ + " InTensor aclCreateTensor index " + std::to_string(i) + " fail");
            return atb::ERROR_INTERNAL_ERROR;
        }
        aclInTensors_[i] = aclnnTensor;
    }
    return atb::NO_ERROR;
}

atb::Status GeluOperation::CreateAclnnOutTensor(const atb::VariantPack &variantPack)
{
    aclOutTensors_.resize(GetOutputNum());
    for (size_t i = 0; i < aclOutTensors_.size(); ++i) {
        auto aclnnTensor = CreateAclnnTensor(variantPack.outTensors.at(i), i);
        if (aclnnTensor->tensor == nullptr) {
            LOG_ERROR(opName_ + " outTensor aclCreateTensor index " + std::to_string(i) + " fail");
            return atb::ERROR_INTERNAL_ERROR;
        }
        LOG_INFO(opName_ + " input[" + std::to_string(i) + "] CreateAclnnTensor start");
        aclOutTensors_[i] = aclnnTensor;
    }
    return atb::NO_ERROR;
}

atb::SVector<int64_t> GetCopyTensorStride(atb::Dims &tensorDims)
{
    atb::SVector<int64_t> tmpStrides(tensorDims.dimNum, 1);
    if (tensorDims.dimNum > 8) {  // 8: tensor最大维度数量
        LOG_ERROR("tensor's dimNum is larger than 8, GetCopyTensorStride failed.");
        return tmpStrides;
    }
    for (int64_t i = static_cast<int64_t>(tensorDims.dimNum) - 2; i >= 0; i--) {
        tmpStrides[i] = (tensorDims.dims[i + 1] * tmpStrides[i + 1]);
    }
    return tmpStrides;
}

std::shared_ptr<AclnnTensor> GeluOperation::CreateAclnnTensor(atb::Tensor atbTensor, size_t tensorIdx)
{
    auto aclnnTensor = std::make_shared<AclnnTensor>();
    aclnnTensor->tensorIdx = static_cast<int>(tensorIdx);
    aclnnTensor->needUpdateTensorDataPtr = true;
    aclnnTensor->atbTensor = atbTensor;
    aclnnTensor->strides = GetCopyTensorStride(atbTensor.desc.shape);

    // 创建Aclnn tensor
    aclnnTensor->tensor = aclCreateTensor(atbTensor.desc.shape.dims,
        atbTensor.desc.shape.dimNum,
        atbTensor.desc.dtype,
        aclnnTensor->strides.data(),
        0,
        atbTensor.desc.format,
        atbTensor.desc.shape.dims,
        atbTensor.desc.shape.dimNum,
        atbTensor.deviceData);
    return aclnnTensor;
}

// 重写父类方法, 创建workspace和aclexecutor
atb::Status GeluOperation::SetAclnnWorkspaceExecutor()
{
    // 调用aclnn接口获取workspace大小
    LOG_INFO(opName_ + " SetAclnnWorkspaceExecutor start");
    if (param_.geluApproximate == -1) {
        auto ret = aclnnGeluGetWorkspaceSize(aclInTensors_.at(0)->tensor,  // self
            aclOutTensors_.at(0)->tensor,                                  // out
            &workspaceSize_,
            &aclExecutor_);
        CHECK_RET(ret, opName_ + " aclnnGeluGetWorkspaceSize failed, ret: " + std::to_string(ret));
        LOG_INFO(opName_ + " SetAclnnWorkspaceExecutor end, workspaceSize_: " + std::to_string(workspaceSize_));
        return ret;
    }
    auto ret = aclnnGeluV2GetWorkspaceSize(aclInTensors_.at(0)->tensor,  // x
        param_.geluApproximate,                                          // approximate
        aclOutTensors_.at(0)->tensor,                                    // y
        &workspaceSize_,
        &aclExecutor_);
    CHECK_RET(ret, opName_ + " aclnnGeluV2GetWorkspaceSize failed, ret: " + std::to_string(ret));
    LOG_INFO(opName_ + " SetAclnnWorkspaceExecutor end, workspaceSize_: " + std::to_string(workspaceSize_));
    return ret;
}

// 重写父类方法, 执行aclnn算子
atb::Status GeluOperation::ExecuteAclnnOp(uint8_t *workspace, aclrtStream &stream)
{
    // 调用aclnn算子进行算子下发
    LOG_INFO(opName_ + " ExecuteAclnnOp start");
    if (param_.geluApproximate == -1) {
        auto ret = aclnnGelu(workspace, workspaceSize_, aclExecutor_, stream);
        CHECK_RET(ret, opName_ + " ExecuteAclnnOp failed, ret: " + std::to_string(ret));
        LOG_INFO(opName_ + " ExecuteAclnnOp end");
        return ret;
    }
    auto ret = aclnnGeluV2(workspace, workspaceSize_, aclExecutor_, stream);
    CHECK_RET(ret, opName_ + " aclnnGeluV2 failed, ret: " + std::to_string(ret));
    LOG_INFO(opName_ + " ExecuteAclnnOp end");
    return ret;
}