ctk@ecsvax.UUCP (04/28/84)
I just got my copy of the new CI-C86 C compiler. This is v2.1 and was one
of the choices for the free update that came with v1.33. I managed to
recompile all my stuff with trouble only with a few assembler functions.
I writing this on a terminal emulator that I wrote using v1.33 and recompiled
with the new compiler.
The package comes in three double sided disks. Most of the files are
in a squeezed format and must be unsqueezed before use. You have your choice
of 8 libraries that use small or big model, dos2 or all versions of dos,
8087 or software floats. The documentation is much improved and as before
the source to all the library functions is included. The IO redirection
bug has been fixed.
I use the dos2, small model , 8087 library. All comments below are
about that library. The compiler is **big**. The library is over 100k; it's
a 4 pass compiler (the last pass being an 'optimizer') and each pass of the
compiler is 30-60 k. You use the standard dos linker that comes with dos.
Compiles and (especially) links are slower than with previous versions.
The package includes a library manager for .lib files that is compatible
with the dos linker.
I ran a few of the benchmarks from the 8/83 issue of byte.
program old time new time
intmath3 13.8sec 12.02sec
pointer1 75.2 49.37
sieve 17.4 13.51
float 45.3 14.39
As you can see the really dramatic improvement is with the floating point
support. This is most of my work and I'm very happy with this. The compiler
will do an LU factorization of a 50x50 matrix of doubles using partial
pivoting in 7 sec. This is faster than Microsoft fortran, a SageII running
either the P-system( big deal) or APL under cpm-68k.
I'm happy. My only complaint is that the thing is so big.
By the way, the exe files generated by the linker are a good deal smaller
for floating point intensive code than before. For non-floating point code
the exe files are roughly the same size.
C.T. Kelley
Dept. of Math.
N.C. State U.
Raleigh, N.C. 27650
decvax!mcnc!ecsvax!ctk