Method GetCVKernelBody

compiler/lib/KernelGen/BareMetal/Resize.cpp:118–189

}

std::string ResizeKernel::GetCVKernelBody(TContext* context) const {
    auto kernel_sig = GetCVKernelSignature(context);
    std::string body_temp = R"(
        #include <math.h>
        #include <string.h>
        #include "tinycv_c.h"
        #define rep(i, n) for (int i = 0; i < (n); ++i)
        static inline uint8_t output_converter(float x){
            x = fmin(255.0f, fmax(0.0f, x));
            return (uint8_t) roundf(x);
        }
        static inline void get_nearest_linear_coord(float scale, int size, int idx, float* ah0, int* ih0, float* ah1, int* ih1){
            if (size == 1) {
                *ah0 = 1.f;
                *ih0 = 0;
                *ah1 = 0.f;
                *ih1 = 0;
            }

            float alpha = (idx + 0.5f) / scale - 0.5f;
            int origin_idx = (int)(floorf(alpha));
            alpha -= origin_idx;

            if (origin_idx < 0) {
                origin_idx = 0;
                alpha = 0;
            } else if (origin_idx + 1 >= size) {
                origin_idx = size - 2;
                alpha = 1;
            }

            *ah0 = 1 - alpha;
            *ih0 = origin_idx;
            *ah1 = alpha;
            *ih1 = origin_idx + 1;
        }
        void ${kernel_sig}{
            uint8_t * sptr = src->data;
            uint8_t * dptr = dst->data;
            int IH = src->rows;
            int IW = src->cols;
            int C = src->channels;
            int OH = dst->rows;
            int OW = dst->cols;

            float scale_h = (float)(OH) / IH;
            float scale_w = (float)(OW) / IW;

            rep(oh, OH) rep(ow, OW) {
                int ih0, ih1, iw0, iw1;
                float ah0, ah1, aw0, aw1;

                get_nearest_linear_coord(scale_h, IH, oh, &ah0, &ih0, &ah1, &ih1);
                
                get_nearest_linear_coord(scale_w, IW, ow, &aw0, &iw0, &aw1, &iw1);

                rep(c, C) {
                    dptr[(oh * OW + ow) * C + c] = output_converter(