home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.lang.fortran
- Path: sparky!uunet!sun-barr!ames!news.hawaii.edu!galileo!tholen
- From: tholen@galileo.ifa.hawaii.edu (Dave Tholen)
- Subject: Re: MS Fortran 5.1 upgrade. Is it worth it?
- Message-ID: <1992Jul25.220659.2257@news.Hawaii.Edu>
- Sender: root@news.Hawaii.Edu (News Service)
- Nntp-Posting-Host: galileo.ifa.hawaii.edu
- Organization: University of Hawaii
- Date: Sat, 25 Jul 1992 22:06:59 GMT
- Lines: 46
-
- Luc Bauwens writes:
-
- > The 32 bit apps run about 35% faster than the 16 bits MS apps.
-
- I have found that the speedup depends on the relative amounts of I/O and
- number crunching. WATCOM's I/O performance would appear to be slightly
- worse than Microsoft's, but in the number-crunching department, the
- 32-bit code wins handily, and by more than a factor of two.
-
- I ran a simple benchmark that consisted of the following: a DO loop that
- executes 1,000,000 times; inside the loop is a random number generator
- that provides a number between -1.0 and 1.0, and a statement that takes
- the arcsine of the random number and adds the result to an accumulator.
- Outside the loop, the value of the accumulator is written to standard
- output. I discovered that the accumulation step was necessary to prevent
- WATCOM's optimizer from eliminating the arcsine step, which I put in there
- to make sure the coprocessor was getting used. By the way, everything was
- in DOUBLE PRECISION.
-
- Here are some timings on a 25 MHz 386/387 combination along with the compiler
- options used to generate the executable. Note that WFL refers to the WATCOM
- product, while FL refers to the Microsoft product:
-
- 74.36 sec WFL386 /3 /OX
- 74.48 WFL386
- 80.67 WFL386 /OD
- 198.20 WFL386 /FPC
-
- 162.67 FL -G2 -FPi87
- 1,532.35 FL -G2 -FPa
- 15,561.15 FL -G2 -FPi with NO87=Coprocessor disabled
-
- With the WATCOM product, it appears that OS/2 takes over the coprocessor
- emulation task, but OS/2 doesn't recognize the NO87 environment variable,
- so I never could get it to run the test without the coprocessor. The
- factor of 100 speedup using the coprocessor with the Microsoft product
- is ridiculous. Something must be going on that I don't understand. I
- wish I had a system with 16-bit OS/2 version 1.3 installed to determine
- if the 32-bit operating system is somehow handicapping 16-bit programs
- that emulate the 80-bit arithmetic of the coprocessor.
-
- Anyway, I am keeping the Microsoft compiler on my hard disk, simply
- because WATCOM does not (yet?) support the 16-bit Kbd, Mou, and Vio API
- calls. It does permit full PM development using the 32-bit APIs, but
- they are only thinking about providing support for the old 16-bit APIs;
- no guarantee that it will happen.
-