ip@me.utoronto.ca (Bevis Ip) (12/15/89)
I think this applies to MIPS in general as well.
-----------------------
From: gwyn@BRL.MIL (Doug Gwyn, VLD/VMB)
Newsgroups: comp.sys.sgi
Subject: Irix 3.2 (4D) ldexp() function broken
Message-ID: <8912140834.aa03045@VGR.BRL.MIL>
Date: 14 Dec 89 13:34:13 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 16
$ cat > foo.c
#include <math.h>
#include <stdio.h>
main() {
printf( "%g\n", ldexp( 0.0, 0 ) );
return 0;
}
^D
$ cc -o foo foo.c
./foo
$ ./foo
1.11254e-308
$ # answer should be "0"; works properly on all other BRL UNIX systems
$ # I think the problem is that they're treating true-0 as a denormalized
$ # number and "rounding" it. This is RONG.mac@rhea.ardent.com (Mike McNamara) (12/19/89)
> In article <1989Dec14.135452.16496@me.toronto.edu> ip@me.utoronto.ca (Bevis Ip) writes: > > I think this applies to MIPS in general as well. > > ----------------------- > From: gwyn@BRL.MIL (Doug Gwyn, VLD/VMB) > Newsgroups: comp.sys.sgi > > $ cat > foo.c > #include <math.h> > #include <stdio.h> > > main() { > printf( "%g\n", ldexp( 0.0, 0 ) ); > return 0; > } > ^D > $ cc -o foo foo.c > ./foo > $ ./foo > 1.11254e-308 > $ # answer should be "0"; works properly on all other BRL UNIX systems > $ # I think the problem is that they're treating true-0 as a denormalized > $ # number and "rounding" it. This is RONG. One should note that these and other fortran compiler problems do NOT apply to Ardent's MIPS based systems. The distribution of MIPS compilers (and MIPS compiler bugs) is roughly as follows: +--> PRIME MIPS ---> SGI--+--> CDC +--> ??? I.E., when SGI used the MIPS chip, they also licensed the MIPS OS and MIPS compilers. Hence they got a quite nice compiler and OS. They also get any errors that might lurk therein. (I use "error" instead of "bug" in the spirit of Dijkstra's column in December 1989 CACM.) PRIME and CDC (among others) re-sell the SGI boxes, with compiler, OS and so on. Hence the get the features that SGI has added. SGI has introduced new libraries and features to the software they received from MIPS, and has fixed so errors; but the stuff is definately derivitive of MIPS Co. PRIME and CDC to a large extent simply re-sell the SGI boxes. I am not aware whether DEC has retargeted Ultrix f77 to the MIPS, sells the MIPS Fortran compiler, or perhaps retargeted the VMS Fortran compiler. Ardent uses neither the MIPS OS nor any of the MIPS compilers. Hence the line "MIPS in general" in the above context applies to MIPS Co. machines, SGI's machines, PRIME's PXL machines, CDC's and perhaps DEC's; it does not apply to Ardent's (nor Stardent's, our new name) -- Michael McNamara (St)ardent, Inc. mac@ardent.com
vjs@rhyolite.wpd.sgi.com (Vernon Schryver) (12/19/89)
In article <MAC.89Dec18161619@rhea.ardent.com>, mac@rhea.ardent.com (Mike McNamara) writes: > ... > I.E., when SGI used the MIPS chip, they also licensed the MIPS > OS and MIPS compilers. Hence they got a quite nice compiler and OS. > ... > -- > Michael McNamara (St)ardent, Inc. mac@ardent.com Ahem. This is not what I understand to be true. The impression I have from outside of the SGI languages group(s), is that we did and might still buy compiler stuff from MIPS. However, my recollection of the work done during the SVR3 port to MIPS chips was that both companies contributed lots of engineering time and money. I am only a drudge in the OS and network trenches, but it seemed at the time as inaccurate to say MIPS got the OS from SGI as vis versa. Vernon Schryver Silicon Graphics vjs@sgi.com