|
From: | Omkar Wagh |
Subject: | [Libjit] libjit about 1000 times slower than native functions |
Date: | Tue, 17 Sep 2013 11:09:51 -0400 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130625 Thunderbird/17.0.7 |
HiI was just profiling the basic tutorial mul_add and it turns out that libjit is about 1000x slower than native function calls for the mul_add tutorial.
Profiling with callgrind suggests that most of the time is spent in :- jit_apply (17.5%) jit_apply_builder_add_arguments(14.25%) jit_type_normalize (9%) jit_thread_get_control (6.5%) jit_type_get_param (5%) And so on...So my basic question is, am I missing something here? Does libjit work well for small examples or is this some constant cost that will always be added for any kind of libjit compiled function and it's only when the compiled functions get "large enough" will libjit be worth it?
Regards, Omkar
[Prev in Thread] | Current Thread | [Next in Thread] |