Function ggml_ensure_tensor_data_at_memory

ggml.c:14766–14782

/////////////////////////////////

static void ggml_ensure_tensor_data_at_memory(struct ggml_tensor * tensor) {
#if defined(GGML_USE_CUBLAS)
    if (tensor->backend == GGML_BACKEND_CPU) {
        // in this case, the data is already placed in the memory at compute time
        return;
    }

    if (tensor->buffer == NULL || tensor->data == NULL) {
        GGML_ASSERT(false && "not implemented: tensor has no buffer or data");
    }

    fprintf(stderr, "WARNING: transfering tensor %s to CPU at inference is safe but slow\n", ggml_get_name(tensor));
    ggml_cuda_copy_to_host(tensor);
#else
    UNUSED(tensor);
#endif
}

static void ggml_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor) {
    GGML_ASSERT(params);