From:	khan AT xraylith DOT wisc DOT edu (Mumit Khan)
Subject:	Re: cygwin32-b18-g77-0.5.19.1: patch and binaries
16 May 1997 13:13:54 -0700 :
Approved:	cygnus DOT gnu-win32 AT cygnus DOT com
Distribution:	cygnus
Message-ID:	<9705161602.AA19643.cygnus.gnu-win32@modi.xraylith.wisc.edu>
Original-To:	"Masahiro Kan, Home" <m_kan AT ya2 DOT so-net DOT or DOT jp>
Original-Cc:	gnu-win32 AT cygnus DOT com
In-Reply-To:	Your message of "Fri, 16 May 1997 23:24:31 +0900."
	<199705161430 DOT XAA13844 AT mail DOT ya2 DOT so-net DOT or DOT jp>
Original-Sender:	owner-gnu-win32 AT cygnus DOT com

"Masahiro Kan, Home" <m_kan AT ya2 DOT so-net DOT or DOT jp> writes:
> 
>     Thanks for your effort.  I have downloaded g77-0.5.19.1 binary for 
> Cygwin32-b18 from your WWW page successfully.  I have tested briefly and 
> it worked fine with some problems.
> 
>    Test1: Electric field calculation program
>           Source code size    : 160kB   5.4klines
>           Executable file size: 178kB
>           Note :  Confirmed that bsearch bug is fixed.  Subroutine zwei
>                   was successfully compiled.  The execution speed is 
>                   comparable with that compiled by g77-0.5.20 binary for 
>                   Cygwin32-b17.1.

Well, not quite fixed. I just use a fixed version compiled into the binary
instead of using the buggy one from libc/cygwin.dll.

> 
>    Test2: Alternative Transients Program
>           Source code size    :3.3MB  110klines
>           Executable size     :2.6MB
>           Note :  The execution speed is about 20% slower than that
>                   compiled by g77-0.5.20 binary for Cygwin32-b17.1, 
>                   and has some problems when executed.
> 

Could you please send me the bug report on the exact problem you're
having? You can run it under gdb and send me the backtrace, which
will be a good starting point.

As for performance, g77-0.5.20 is quite a bit better, but a the cost
of the following:

    - new bugs! Complex support is broken with -finit-local-zero and
      many others. If you end up using the backend complex (fno-emulate-
      complex), then the performance goes down again. And there is the
      question of aliasing in 0.5.20 that is new from 0.5.19.1, which
      requires quite a bit of backend mods.
    - backend changes which I'm not comfortable with, since I've been
      out of touch with gcc backend for a few years and too much has
      changed since.

I'll probably wait for 0.5.21 before I do anything further.

btw, you should check out a few of the code-generation options to speed
up the code (-fast-math and a few others).

Regards,
Mumit -- khan AT xraylith DOT wisc DOT edu
http://www.xraylith.wisc.edu/~khan/

-
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 -