From: Paul Brannan <pbrannan@...>
Date: 2008-04-02T00:11:29+09:00
Subject: Re: some Enumerable methods slower in 1.9 on OS X after revision 15124


--6e7ZaeXHKrTJCxdu
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

On Tue, Apr 01, 2008 at 03:27:51AM +0900, Nobuyoshi Nakada wrote:
> Hi,
> 
> At Mon, 31 Mar 2008 22:07:03 +0900,
> Paul Brannan wrote in [ruby-core:16086]:
> > Also, I wonder if there is any benefit to __builtin_setjmp instead of 
> > _setjmp?
> 
> Unfortunately,
> 
>   error: '__builtin_longjmp' second argument must be 1

See attached for a workaround, but I was not able to measure any performance
benefit on gcc 3.3. :(

(but I'm also not sure how to construct a good benchmark where the
method call overhead doesn't dwarf what I'm trying to measure)

> it seems not be able to pass a variable.  And combination with
> _longjmp caused SEGV.
> 
> > Also, just thinking, I wonder how hard it would be to make use of dwarf2 
> > exceptions on gcc via unwind.h?
> 
> How faster is it?

I don't know; we'd have to try it and find out.

No one seems to have benchmarks, just blanket statements like "dwarf2
exceptions are faster than sjlj for non-exceptional cases".

Paul


--6e7ZaeXHKrTJCxdu
Content-Type: text/plain; charset=us-ascii
Content-Disposition: attachment; filename="builtin_setjmp.patch"


--6e7ZaeXHKrTJCxdu--