Mail Archives: djgpp/1994/11/23/10:53:54
A number of people suggested testing without memory managers
installed. I've done this: no real effect.
Others suggested disabling the (secondary) cache just to see
if BC-compiled program fits in, while GCC-compiled does not.
This slowed down both programs, but the performance ratio stayed
the same.
I omitted the -O3 and -funroll-loops and put -O2 and -fomit-frame-pointer
instead, as some of you suggested. This improved the performance
to some degree: it is now about 2/3 of the speed of BC-compiled program.
I've also seen that some of the screen functions cause mode switch
(e.g., wherex(), wherey(), clreol() etc.). I will next comment them out
to see what happens.
A heretic thought started crawling into my mind: can it be that GCC
indeed *does* produce slow code for C++ programs? I don't really
work in C++, so all my experience about GCC code efficiency comes
from C programs. What about those constructors/destructors behind
the scenes: could it be they are much slower in G++ than they are
in BC++? Opinions, anyone? Maybe I will re-write the program in
pure C one of these days...
Thanks again to all people who wrote to me on this.
- Raw text -