Function test_reuse_and_free

subprojects/llama.cpp/tests/test-alloc.cpp:391–412

}

static void test_reuse_and_free() {
    dummy_backend backend      = dummy_backend_init(40);
    auto [ctx, graph, ctx_ptr] = make_context();

    ggml_tensor * x[9];
    x[0] = make_input_with_size(ctx, 24);
    x[1] = make_input_with_size(ctx, 8);
    x[2] = make_input_with_size(ctx, 8);
    x[3] = ggml_add(ctx, x[1], x[2]);        // reuse, free x2
    x[4] = ggml_pad(ctx, x[0], 2, 0, 0, 0);  // alloc new buffer, free x0
    x[5] = ggml_scale(ctx, x[4], 2.0f);      // alloc from free block
    x[6] = ggml_add(ctx, x[4], x[5]);        // reuse, free x5
    x[7] = ggml_view_1d(ctx, x[6], 2, 8);    // view
    x[8] = ggml_add(ctx, x[3], x[7]);        // reuse
    assign_names(ctx);

    ggml_gallocr_ptr galloc = allocate_graph(graph, x[8], &backend.buffer_type);
    check_all_allocated(graph);
    check_no_overlap(graph);
    check_max_size(ctx);
    GGML_ASSERT(backend.context->allocated_total() <= 40 + 32 + 32);
}

static void test_merge_free_block(size_t max_buffer_size) {
    dummy_backend backend      = dummy_backend_init(max_buffer_size);