home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
3D GFX
/
3D GFX.iso
/
amiutils
/
m_p
/
povray040
/
compiling
next >
Wrap
Text File
|
1995-12-30
|
1KB
|
27 lines
The unix config and makefile were used to generate an executable. The only
changes to the sourcecode of PoV-Ray version 2.2 were:
- frame.h: included "math-68881.h" instead of "math.h" for slight speed
improvement and reduction of executable size
changed welcoming message to reflect the nature of this version
of PoV-Ray
uncommented a #define to set default output to dump format
- unix.c: included "math-68881.h" instead of "math.h" for sake of
completeness
- makefile: changed compiler and compiler options. Just take a look at it!
Setting optimization level to 3 actually slowed the executable
down, maybe the unrolled loops blew the cache. If you are willing
to compile every part of the source with all combinations of
optimizations, please mail me your results :-)
- povray: removed the symbol hunks from the executable to save space
I tweaked other bits of the source to replace divides by multiplications (e.g.
the VNormalize and VInverseScale macros in vector.h). Theoretically one fdiv
and three fmul instructions should execute double as fast as three fdivs.
Practically that version was quite a bit slower... Maybe the timing table I
found in an old "Kickstart" magazine is wrong. At least I don't need to include
sourcecode now...