[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Libjit] libjit about 1000 times slower than native functions

From: Omkar Wagh
Subject: Re: [Libjit] libjit about 1000 times slower than native functions
Date: Thu, 19 Sep 2013 14:51:14 -0400
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130625 Thunderbird/17.0.7

That project seems to be hibernating. The last release was on 2012/04. Is it stable?

On 09/18/2013 04:09 PM, Ben Vanik wrote:
I'd recommend looking at asmjit, Omkar. It can be made to have much lower per-call overhead by avoiding the apply step.

Ben Vanik

On Tue, Sep 17, 2013 at 2:18 PM, Omkar Wagh <address@hidden> wrote:

On 09/17/2013 01:06 PM, Aleksey Demakov wrote:
Yes, jit_apply mechanism is slow. When calling small JIT-ted functions from C almost all time is going to be spent switching from C to JIT runtime rather then executing the function itself. So it's going to perform poorly. For a more accurate idea how fast would be JITed code it is better to benchmark a function that performs some computation many times in a loop or recursively.
So our use case is actually a few small expressions (dynamically given) that we'd like to compile to native code. Response time is very critical but start-up time (one time compilation) not so much...

I can see from your answer that libjit probably won't be useful for us. Can you recommend something that might help us (better than building an _expression_ tree evaluator which we already have)? I was thinking of compiling it into a shared library (.so) and loading it immediately but are there some standard tools to do that?


reply via email to

[Prev in Thread] Current Thread [Next in Thread]