新增OP¶

以下以添加argmax为例，详细说明新增op的方法。

1. 添加OpParam 结构体以传导 Op 的输入和输出¶

这里命名为 ArgmaxParam

在 paddlelite/lite/operators/op_params.h 中添加 ArgmaxParam 结构体，代码如下：

struct ArgmaxParam {
    lite::Tensor* X{};
    lite::Tensor* Out{};
    int Axis{0};
};

2. 添加 Argmax Op 并注册¶

在paddlelite/lite/operators/目录下新建argmax_op.h文件，主要代码如下：

class ArgmaxOpLite : public OpLite {
public:
    ArgmaxOpLite() {}
    explicit ArgmaxOpLite(const std::string &op_type) : OpLite(op_type) {}
    bool CheckShape() const override;
    bool InferShape() const override;
    bool AttachImpl(const cpp::OpDesc &opdesc, lite::Scope *scope) override;
    void AttachKernel(KernelBase *kernel) override { kernel->SetParam(param_); }
    std::string DebugString() const override { return "argmax"; }

#ifdef LITE_WITH_PROFILE
    void GetOpRuntimeInfo(paddle::lite::profile::OpCharacter *ch) {
        auto input_dims = param_.X->dims();
        auto output_dims = param_.Out->dims();
        ch->input_shape = ch->DimToStr(input_dims);
        ch->output_shape = ch->DimToStr(output_dims);
        ch->remark = "axis" + std::to_string(param_.Axis);

        auto axis = param_.Axis;
        if (axis < 0) {
            axis += input_dims.size();
        }
        int max_num = 1;
        for (int64_t i = axis + 1; i < input_dims.size(); i++)
            max_num *= input_dims[i];
        float gops = 1.0f;
        for (int i = 1; i <= max_num; i++) gops *= i;
        ch->macs = gops * output_dims.production();
    }
#endif

private:
    mutable ArgmaxParam param_;
};

ArgmaxOpLite 继承 OpLite ，成员变量包括 ArgmaxParam 结构体，需要实现的接口包括 CheckShape() 、InferShape() 、AttachImp() 、AttachKernel() 和 DebugString() 函数。AttachKernel() 和 DebugString() 函数较为简单，此处直接实现；

在 paddlelite/lite/operators/ 目录下新建argmax_op.cc文件，需要具体实现CheckShape()、InferShape()和AttachImp()函数。CheckShape()函数检查输入是否符合要求，InferShape()函数基于输入推断得到输出的维度，AttachImp()函数绑定Op的输入输出。然后在argmax_op.cc文件中注册argmax，核心代码如下：

bool ArgmaxOpLite::CheckShape() const {
    CHECK_OR_FALSE(param_.X);
    CHECK_OR_FALSE(param_.Out);
    CHECK_OR_FALSE(param_.Axis < (param_.X)->dims().size());
    return true;
}

bool ArgmaxOpLite::InferShape() const {
    auto x_dims = param_.X->dims();
    int x_rank = x_dims.size();
    int axis = param_.Axis;
    if (axis < 0) axis += x_rank;

std::vector<int64_t> out_dims;
    for (int64_t i = 0; i < axis; i++) {
        out_dims.push_back(x_dims[i]);
    }
    for (int64_t i = axis + 1; i < x_rank; i++) {
        out_dims.push_back(x_dims[i]);
    }

  // Set output dims
    param_.Out->Resize(lite::DDim(out_dims));
    return true;
}

bool ArgmaxOpLite::AttachImpl(const cpp::OpDesc &op_desc, lite::Scope *scope) {
    auto x = op_desc.Input("X").front();
    auto out = op_desc.Output("Out").front();

param_.X = scope->FindVar(x)->GetMutable<lite::Tensor>();
    param_.Out = scope->FindVar(out)->GetMutable<lite::Tensor>();
    param_.Axis = op_desc.GetAttr<int>("Axis");

return true;
}
REGISTER_LITE_OP(argmax, paddle::lite::operators::ArgmaxOpLite);

在paddlelite/lite/operators/CMakeLists.txt中添加add_operator(argmax_op basic SRCS argmax_op.cc DEPS ${op_DEPS})

3. 添加Argmax Kernel并绑定¶

以下以arm端argmax实现为例说明

在paddlelite/lite/kernels/arm/目录下新建argmax_compute.h文件，声明ArgmaxCompute类，并继承KernelLite，主要代码如下：

class ArgmaxCompute : public KernelLite<TARGET(kARM), PRECISION(kFloat)> {
public:
    using param_t = operators::ArgmaxParam;
    void Run() override;
    virtual ~ArgmaxCompute() = default;
#ifdef LITE_WITH_PROFILE
    virtual void SetProfileRuntimeKernelInfo(
        paddle::lite::profile::OpCharacter* ch) {
        ch->kernel_func_name = kernel_func_name_;
    }
    std::string kernel_func_name_{"NotImplForArgmax"};
#endif
};

在paddlelite/lite/kernels/arm/目录下新建argmax_compute.cc文件，主要实现Run函数。Run()函数调用paddlelite/lite/bachends/arm/math/argmax.h中的argmax_func()函数，根据输入计算输出。最后在argmax_compute.cc文件中，我们绑定argmax的输入输出（为tensor的输入参数都需要绑定），代码如下：

void ArgmaxCompute::Run() {
    auto& param = Param<operators::ArgmaxParam>();
    lite::Tensor* input = param.X;
    lite::Tensor* output = param.Out;
    int axis = param.Axis;
    lite::arm::math::argmax_func(input, axis, output);
#ifdef LITE_WITH_PROFILE
    kernel_func_name_ = "argmax_func";
#endif
    return;
}

REGISTER_LITE_KERNEL(
    argmax, kARM, kFloat, kNCHW, paddle::lite::kernels::arm::ArgmaxCompute, def)
    .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM))})
    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM))})
    .Finalize();

在paddlelite/lite/kernels/arm/CMakeLists.txt中添加

add_kernel(argmax_compute_arm ARM basic SRCS argmax_compute.cc DEPS ${lite_kernel_deps} math_arm)

4. 添加Argmax实现¶

在paddlelite/lite/backends/arm/math/目录下新建argmax.h文件，声明argmax_func()函数，代码如下：
```
void argmax_func(const lite::Tensor* input, const int axis, lite::Tensor* output);
```

在paddlelite/lite/backends/arm/math/目录下新建argmax.cc文件，具体实现argmax_func()函数，代码如下：

void argmax_func(const lite::Tensor *input,
                const int axis,
                lite::Tensor *output) {
auto input_ddim = input->dims();
auto output_ddim = output->dims();

const int size = input_ddim[axis];
const int in_channel = input_ddim.count(axis, input_ddim.size());
const int out_channel = output_ddim.count(axis, output_ddim.size());
const int in_stride = input_ddim.count(axis + 1, input_ddim.size());
const int out_stride = input_ddim.count(0, axis);

for (int n = 0; n < out_stride; n++) {
    for (int k = 0; k < in_stride; k++) {
    const float *in_ptr = input->data<float>() + n * in_channel + k;
    std::vector<std::pair<float, int>> vec;
    vec.resize(size);
    for (int i = 0; i < size; i++) {
        vec[i] = std::make_pair(in_ptr[i * in_stride], i);
    }
    // sort
    std::partial_sort(vec.begin(),
                        vec.begin() + 1,
                        vec.end(),
                        std::greater<std::pair<float, int>>());

    // out
    float *out_ptr = output->mutable_data<float>() + n * out_channel + k;
    *out_ptr = vec[0].second;
    }
}
}

在paddlelite/lite/backends/arm/math/CMakeFile.txt中的math_arm library中添加argmax.cc，在paddlelite/lite/backends/arm/math/funcs.h中添加#include "lite/arm/math/argmax.h"

5. 添加Argmax单测¶

在paddlelite/lite/tests/kernels目录下新建argmax_compute_test.cc文件，声明并实现ArgmaxComputeTester类；
ArgmaxComputeTester类中主要包括PrepareOpDesc、PrepareData和RunBaseline函数。PrepareOpDesc函数设定单测op的类型和输入输出参数，PrepareData函数对输入tensor进行初始化，RunBaseline是基于输入计算得到输出，用于和框架计算的输出进行对比；

使用gtest添加单测，代码如下：

TEST(Argmax, precision) {
    #ifdef LITE_WITH_ARM
    LOG(INFO) << "test argmax arm";
    Place place(TARGET(kARM));

    for (int axis : {0, 1, 2, 3}) {
        for (int n : {1, 3}) {
        for (int c : {3, 6}) {
            for (int h : {9, 18}) {
            for (int w : {9, 18}) {
                std::unique_ptr<arena::TestCase> tester(
                    new ArgmaxComputeTester(place, "def", axis, n, c, h, w));
                arena::Arena arena(std::move(tester), place, 2e-5);
                arena.TestPrecision();
            }
            }
        }
        }
    }
    #endif
}

在paddlelite/lite/tests/kernels/CMakeLists.txt中添加

lite_cc_test(test_kernel_argmax_compute SRCS argmax_compute_test.cc DEPS arena_framework ${x86_kernels} ${arm_kernels} ${lite_ops} ${host_kernels})

6. 编译运行¶

在paddlelite目录中，执行./lite/tools/ci_build.sh build_test_arm，该脚本会创建手机模拟器，并编译运行所有单测（花费时间较久）。如果运行无误，则表明添加argmax成功。