Date:	Wed, 20 May 1998 12:29:10 +0300 (IDT)
From:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To:	Lionel Portmann <Lionel DOT Portmann AT epfl DOT ch>
cc:	djgpp AT delorie DOT com
Subject:	Re: DJGPP vs. mingwin32 I/O performances
In-Reply-To:	<3561B91D.702A@epfl.ch>
Message-ID:	<Pine.SUN.3.91.980520122851.24974R-100000@is>
MIME-Version:	1.0

On Tue, 19 May 1998, Lionel Portmann wrote:

> I tried both ports of gcc v2.8.1 for PC, DJGPP and MINGWIN32.
> I compiled the very same source code with both environnement,
> and found that the mingwin32 executable was a least 1.5 times slower 
> than djgpp one. I read in a FAQ that djgpp has to switch between
> protected mode and real mode every time it has to do I/O access,
> so I expected mingwin32 to be faster than djgpp.

It is a common misconception that since a mode switch is required, I/O
performance in DJGPP should suck.  I think the FAQ says that any
real-world program has much more to do than just read files, and so
the effect of the mode switch will be negligible in many cases.

IMHO, the quality of the runtime library is much more important here
than the mode switch issue.  I don't know which libraries are used by
mingwin32, but the solution to this puzzle is probably in those
libraries.

> I have to add that at compilation time, mingwin32 gave me more warnings
> than djgpp, which might indicate that the settings are not equivalent.
> On the other hand, -O3 switch was ON in both versions. 

Add -v to the switches, and you will see all the options used in both
cases.

- Raw text -