🚀 go-pugleaf

In article <SCANNON.53.728064267@novell.com> SCANNON@novell.com (Steve Cannon) writes:
>>Does OS/2 2.0 take advantage of the fact that a math coprocessor is 
>>installed on systems running a CPU < 486?
>>If not, I'd like to see IBM make a version available that does.  Unless I'm 
>>way off base, it seems like a simple recompile of the code with the proper 
>>switches set could do the trick.

I, too, wondered about this, and I did an experiment.  I recently bought a
387 chip and decided to find out whether it would affect OS/2.  I'd heard that
it would be about a 25% speedup, but I wanted to be scientific about it.
Before installing the 387, I ran a PM benchmark program that does pretty
extensive tests.  After installing the 387, I ran the program again with all
other factors the same (i.e. no other progs running, same memory, etc.).
The net result is that a certain type of BitBLT was much faster but
EVERYTHING ELSE WAS THE SAME.  Not all BitBLTs speeded up, and I'm not sure
how often the one that speeded up is used, but that one factor was definitely
more than within the range of possible error.  I might still have results on
the test if anyone needs exact figures, and it would be neat for someone to
run the test on 2.0 GA vs. 2.0 w/32-bit GRE vs. 2.1b (maybe even have results
show up in a magazine).

I used 2.0 GA with some bug fixes but NO service pak (i.e. no 32-bit GRE);
it is possible that some things have changed since then.

Net result is that I have strong evidence that PM does use a math co-processor,
but I don't know how much practical affect it has nor do I know whether any
other part of OS/2 will use a 387.
+++++++++++++++++++++++++++++++++++
     David T. Chappell            --       North Carolina State University
    Stanford Univeristy           --   Computer And Technologies Theme Program
    davidc@leland.stanford.edu    --       david@catt.ncsu.edu