From:	stefano AT tirreno DOT it ("Stefano Gragnani")
Subject:	Runtime speed with GNU compilers
14 Apr 1997 16:50:59 -0700 :
Approved:	cygnus DOT gnu-win32 AT cygnus DOT com
Distribution:	cygnus
Message-ID:	<199704142148.XAA27137.cygnus.gnu-win32@mail.tirreno.it>
Mime-Version:	1.0
Original-To:	<gnu-win32 AT cygnus DOT com>
X-MSMail-Priority:	Normal
X-Priority:	3
X-Mailer:	Microsoft Internet Mail 4.70.1155
Original-Sender:	owner-gnu-win32 AT cygnus DOT com

Thanks for the replies. I have compiled and runned another time the
code-example with various options, the results are:

1) without options ==> 2063 seconds
2) only with the '-O3' options (with the capital "O")  ==>  2362 seconds
!!!!!
3) with '-O3' + '-ffast-math'  ==> 113 seconds
4) with only '-ffast-math'  ==> about 2000 seconds.

Why with the '-O3' options the time is grater that without optimizations ? 
In the online documentation that I have read the '-O6' option is not
documented (only the '-O1', '-O2' and 'O3' are mentioned) but it is
accepted from the compiler: wich are "its optimizations"?
The '-ffast-math' is not aderent to the IEEE 754 arithmetic and therefore,
in general, it use is not recommended and is not good that g++ required it
(plus the '-O3' or the '-O6' option) to obtain the best result (113
seconds). (The Metrowerk compiler is aderent to the IEEE 754 and is result
is 125 seconds ...)
If the problems are the Cygnus libraries under NT, there are other
libraries for NT?

Thanks
Stefano Gragnani

-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request AT cygnus DOT com" with one line of text: "help".

- Raw text -