Mail Archives: djgpp/1997/04/15/10:47:58
"Guido Bursch" <bursch AT uni-muenster DOT de> wrote:
>Hi all!
>I just ported my little vectorcube program from Turbo c
>to djgpp2. It took me about 2 hours for a 400 lines sourcecode
>(speaks for the quality of djgpp) without asm-optimizing.
>The result was a bit disappointing: it ran 4 times slower than with
>Turbo c which is surely my fault and not the of djgpp.
>gprof reported me that 72% (22% self) of the runtime were wasted by
>the function (or macro or whatever) "__djgpp_exception_processor"
>without giving me the number of calls to it.
>To my program is to say that FP- operations are heavyly used and I
>have to use an FP- emulator because I dont't own an 387 FPU.
>2 times I call an interrupt (set vga/text) , use
>1* __djgpp_nearptr_enable() + disable, clear the virtual screen with
>memset (400 times), copy it to vga - memory mith memcpy (400 times) and draw
>hlines with memset (44606 times).
>My question is now, which function may use "__djgpp_exception_processor"
>and how can I do the things without or with less use of it,
>and btw what is it ?
>I would be very happy, if someone could give me an answer to my
>question but please don't personally email me (we have problems with our
>mail_server).
On the odd chance that noone answers this who can give you a definite
answer, I would say it's the FPU emulation that is killing you. I have
ported code over to djgpp on my old 386 (similar to yours, it was a
bit of 3D program) and it ran about twice as fast; but I used fixed
precision.
Hmm... are you running djgpp on a 386? You are brave. So was I, but
eventually I decided to splurge 50$ and get a 486 mboard with 8 megs;
compile-times are much much less, now.
Oh, and let me also make a small suggestion if I may; Allegro.
Excellent fixed point routines, not to mention polygon stuff.
- Calvin -
- Raw text -