ollama/llama/ggml-cuda/rope.cuh

#include "common.cuh"

#define CUDA_ROPE_BLOCK_SIZE 256

void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst);