Re: [gobo-eiffel-develop] Speed of Gestalt compiled with GEC versus ISE

[Eric B. <er...@go...>]

You should note the gec always checks for call-on-void-target.
I think that this is a requirement from ECMA when the attached
mechanism is not enabled.

Also, gec checks for CAT-calls, but when it found no CAT-call
at compilation time then there is no check needed at execution
time. So, no penalty here.

One thing to remember is that there has been not that much effort
spent yet on optimizing the C code generated by gec. So I guess
that when that is done it should be faster than 40 seconds.
For example, as far as I remember, SE optimizes the generated
C code better than gec. One reason why I'm interested in still
supporting SE 1.2 is to be able to benchmark the C code generated
by gec against the one generated by SE.

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com


Colin Adams wrote:
> No I didn't.
> 
> Doing so does improve the speed dramatically, as you say - now 72 seconds.
> Still significantly slower than GECs 40seconds.
> 
> Thanks for the tip, anyway.
> 
> On 04/02/2008, Jann Röder <roe...@et...> wrote:
>> Did you turn of exception tracing (this makes a huge difference if you
>> have many feature calls) and void call exceptions for ISE ?
>>
>> Colin Adams wrote:
>>> That timing information is rather interesting - on my 5-year old
>>> 32-bit intel machine, I get timings of 2:15 for ISE (comparable with
>>> yours) and 40 seconds with GEC (at roughly 3.5 times quicker, that is
>>> in line with other measurements I have seen).
>