llama.cpp/ggml/src/ggml-cuda/conv2d.cuh

#pragma once
#include "common.cuh"

#define CUDA_CONV2D_BLOCK_SIZE 256
void ggml_cuda_op_conv2d(ggml_backend_cuda_context & ctx, ggml_tensor * dst);