llama : use ggml_backend_dev_get_extra_bufts

[danbev]

This commit updates the name argument for the ggml_backend_reg_get_proc_address function call in make_cpu_buft_list to be ggml_backend_dev_get_extra_bufts instead of ggml_backend_cpu_get_extra_bufts.

The motivation for this change is that in ggml_backend_cpu_get_proc_address, the name parameter is matched against ggml_backend_dev_get_extra_bufts and not ggml_backend_cpu_get_extra_bufts:

static void * ggml_backend_cpu_get_proc_address(ggml_backend_reg_t reg,
    const char * name) {
    if (strcmp(name, "ggml_backend_set_n_threads") == 0) {
        return (void *)ggml_backend_cpu_set_n_threads;
    }
    if (strcmp(name, "ggml_backend_dev_get_extra_bufts") == 0) {
        return (void *)ggml_backend_cpu_get_extra_bufts;
    }

    return NULL;

    GGML_UNUSED(reg);
}

• I have read the contributing guidelines
• Self-reported review complexity:
  • Low

[slaren]

This is already fixed in #9921.