yding@unix.cis.pitt.edu (Yijun Ding ) (05/13/91)
Recently I have a program with small floatings. I find something works
on my TC but not on mainframe. The following code, log10() print
correctly on TC2.0. It is HUGE on unix, and 0 on VMS in University of
Pittsburgh. That is a surprise.
#include <math.h>
main()
{
double a=1.12345e-40;
printf("%lf %lf\n", a, log10(a));
}
Any one can explain why?bobmon@iuvax.cs.indiana.edu (RAMontante) (05/13/91)
yding@unix.cis.pitt.edu (Yijun Ding ) <127293@unix.cis.pitt.edu> : | | #include <math.h> | main() | { | double a=1.12345e-40; | printf("%lf %lf\n", a, log10(a)); | } The gcc compiler under unix warns that 1.12345e-40 exceeds the range of 'double'. It appears that the VAX compilers (gcc, cc, vcc) store doubles in 32 bits, just like floats (legal C, but...). Turbo C stores doubles in 64 bits.
a_rubin@dsg4.dse.beckman.com (05/13/91)
In <127293@unix.cis.pitt.edu> yding@unix.cis.pitt.edu (Yijun Ding ) writes: >Recently I have a program with small floatings. I find something works >on my TC but not on mainframe. The following code, log10() print >correctly on TC2.0. It is HUGE on unix, and 0 on VMS in University of >Pittsburgh. That is a surprise. >#include <math.h> >main() >{ > double a=1.12345e-40; > printf("%lf %lf\n", a, log10(a)); >} >Any one can explain why? I don't know about UNIX, but in VMS, 1.12345e-40 is calculated as a float, rather than a double; hence it is 0. (log10(0) is 0? THAT is a bug.) -- a_rubin@dsg4.dse.beckman.com My opinions are my own, and do not represent those of my employer.
wallyk@bicycle.WV.TEK.COM (Wally Kramer) (05/14/91)
yding@unix.cis.pitt.edu (Yijun Ding ) writes: >Recently I have a program with small floatings. I find something works >on my TC but not on mainframe. The following code, log10() print >correctly on TC2.0. It is HUGE on unix, and 0 on VMS in University of >Pittsburgh. That is a surprise. > >#include <math.h> >main() >{ > double a=1.12345e-40; > printf("%lf %lf\n", a, log10(a)); >} > >Any one can explain why? This might not be the appropriate place for this discussion, however, I myself frequently move software between VMS, Unix and MSDOS. For a Unix system, I used UTek (a BSD derivative) running on a 68020 and got the runtime result of 0.000000 -39.949446 which looks correct to me (and is the same as I got under TC++ 1.0). Naturally, for unix to link in a log10 function, you have to use cc small.c -lm On VMS, I also got the same result, but it was a little more trouble. $ cc /g_float small.c $ link small.obj, sys$input /option sys$share:vaxcrtlg /share ^Z $ run small 0.000000 -39.949446 The problem here is that the VAX floating point range used by default is D_floating: Mantissa Approx. Bits Range bits significant digits F_floating 32 .298e-38 to 1.7e+38 23 6-7 D_floating 64 .298e-38 to 1.7e+38 55 15 G_floating 64 .56e-308 to .9e+308 53 15 VAXC, by default, uses F_float for "float" (unless the /precision=double qualifier is used) and D_floats for "double". For the magnitude of value you want to use, you should use the /g_float qualifier in *all* your compilations. Observe also that you link against VAXCRTLG.EXE instead of VAXCRTL.EXE so that the library routines use G_floats. ----- Wally Kramer contracted from Step Technology, Portland, Oregon 503 244 1239 wallyk@orca.wv.tek.com +1 503 685 2658