GCL/maxima/axiom/acl2, denormalized floats, and alpha
Greetings! Thank you so much! I've located the issue -- denormalized floats on alpha. Right now, GCL uses the -mieee gcc switch. How much worse is the -mieee-with-inexact? I hesitate to use this if it slows everything down, especially since it does not appear that maxima intends to use these numbers anyway. This having been said, a generic lisp system should support least-positive-double-float, et. al. Do you have an opinion? Take care, Paul Slootman <[EMAIL PROTECTED]> writes: > On Fri 13 Oct 2006, Camm Maguire wrote: > > > > Now down to the maxima build-deps: > > > > Build-Depends: gcl ( >= 2.6.7-22 ) , texinfo, libgmp3-dev, libreadline5-dev > > | libreadline-dev, locales, automake1.9, debhelper ( >=4.1.0 ) , > > libncurses5-dev, autoconf2.13, gawk | awk, tetex-extra | texlive, > > sharutils, libxmu-dev, libxaw7-dev > > > > All we really need are the libfoo-dev deps. > > I did most, but not gcl as I gather you're working on that now, right? > > > Paul > > > -- Camm Maguire[EMAIL PROTECTED] == "The earth is but one country, and mankind its citizens." -- Baha'u'llah -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
GCL/maxima/axiom/acl2, denormalized floats, and alpha
Greetings! Spoke too fast. May actually have a compiler bug here: Breakpoint 4, write_object (x=0x120413f70, level=0) at print.d:695 695 r = symbol_value(sLAread_default_float_formatA); (gdb) n 696 if (r == sLshort_float) (gdb) 699 write_double((double)sf(x), 'S', TRUE); (gdb) p x->SF.SFVAL $9 = 1.40129846e-45 (gdb) s write_double (d=2.6524947387065904e-315, e=83, shortp=1) at print.d:388 388 int n = FPRC+1; (gdb) s Casting the smallest denormalized shortfloat to a double in a function argument is destroying the number in the callee. Do you know of a workaround? This persists with -mieee-with-inexact, of course. Take care, Paul Slootman <[EMAIL PROTECTED]> writes: > On Fri 13 Oct 2006, Camm Maguire wrote: > > > > Now down to the maxima build-deps: > > > > Build-Depends: gcl ( >= 2.6.7-22 ) , texinfo, libgmp3-dev, libreadline5-dev > > | libreadline-dev, locales, automake1.9, debhelper ( >=4.1.0 ) , > > libncurses5-dev, autoconf2.13, gawk | awk, tetex-extra | texlive, > > sharutils, libxmu-dev, libxaw7-dev > > > > All we really need are the libfoo-dev deps. > > I did most, but not gcl as I gather you're working on that now, right? > > > Paul > > > -- Camm Maguire[EMAIL PROTECTED] == "The earth is but one country, and mankind its citizens." -- Baha'u'llah -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: GCL/maxima/axiom/acl2, denormalized floats, and alpha
Camm Maguire <[EMAIL PROTECTED]> writes: > Casting the smallest denormalized shortfloat to a double in a function > argument is destroying the number in the callee. Do you know of a > workaround? What kind of machine is this on? Might also be a kernel bug. Do you have a stand-alone testcase? -- Falk -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: GCL/maxima/axiom/acl2, denormalized floats, and alpha
On Wed October 18 2006 14:03, Falk Hueffner wrote: > Camm Maguire <[EMAIL PROTECTED]> writes: > > Casting the smallest denormalized shortfloat to a double in a function > > argument is destroying the number in the callee. Do you know of a > > workaround? > > What kind of machine is this on? Might also be a kernel bug. Do you > have a stand-alone testcase? Actually, that's a really good point. Upon reading it, I recalled discovering, and posting to this list a while ago, that the kernel's denormalized number fixup for EV56 and earlier CPUs had been broken in later kernels. (see the thread "Kernel IEEE Math Completion (everything is zeroed!?!)" at "http://lists.debian.org/debian-alpha/2006/06/";) That long and the short is that we figured out that it was busted somewhere between 2.4.27 and 2.6.16, but no solutions were purposed. Later! -T -- Tyson Whitehead (-twhitehe at uwo.ca -- MC-) Computer Engineer Dept. of Applied Mathematics, Graduate Student- Applied Mathematics University of Western Ontario, GnuPG Key ID# 0xF7666BFF London, Ontario, Canada pgp8zhTwqVENo.pgp Description: PGP signature
Re: GCL/maxima/axiom/acl2, denormalized floats, and alpha
Greetings, and thanks for the feedback!
#include
typedef union {
float f;
int i;
} U;
static void t(double d) {
printf("%e\n",d);
}
int main() {
U u;
u.i=1;
printf("%he\n",u.f);
t((double)u.f);
return 0;
}
./tt
Good:
1.401298e-45
1.401298e-45
Bad:
2.652495e-315
2.652495e-315
uname -a
Linux alf 2.6.17.11 #2 Wed Aug 30 10:01:12 CEST 2006 alpha GNU/Linux
cc -v
Using built-in specs.
Target: alpha-linux-gnu
Configured with: ../src/configure -v
--enable-languages=c,c++,fortran,objc,obj-c++,treelang --prefix=/usr
--enable-shared --with-system-zlib --libexecdir=/usr/lib
--without-included-gettext --enable-threads=posix --enable-nls
--program-suffix=-4.1 --enable-__cxa_atexit --enable-clocale=gnu
--enable-libstdcxx-debug --enable-mpfr --enable-checking=release alpha-linux-gnu
Thread model: posix
gcc version 4.1.2 20060901 (prerelease) (Debian 4.1.1-13)
-mieee does not help. In gdb, u.f is the right number, but gets
corrupted on function call.
Take care,
Falk Hueffner <[EMAIL PROTECTED]> writes:
> Camm Maguire <[EMAIL PROTECTED]> writes:
>
> > Casting the smallest denormalized shortfloat to a double in a function
> > argument is destroying the number in the callee. Do you know of a
> > workaround?
>
> What kind of machine is this on? Might also be a kernel bug. Do you
> have a stand-alone testcase?
>
> --
> Falk
>
>
>
--
Camm Maguire[EMAIL PROTECTED]
==
"The earth is but one country, and mankind its citizens." -- Baha'u'llah
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: GCL/maxima/axiom/acl2, denormalized floats, and alpha
Greetings! So should this be filed as a bug, and if so, against which package? Take care, Tyson Whitehead <[EMAIL PROTECTED]> writes: > On Wed October 18 2006 14:03, Falk Hueffner wrote: > > Camm Maguire <[EMAIL PROTECTED]> writes: > > > Casting the smallest denormalized shortfloat to a double in a function > > > argument is destroying the number in the callee. Do you know of a > > > workaround? > > > > What kind of machine is this on? Might also be a kernel bug. Do you > > have a stand-alone testcase? > > Actually, that's a really good point. > > Upon reading it, I recalled discovering, and posting to this list a while > ago, > that the kernel's denormalized number fixup for EV56 and earlier CPUs had > been broken in later kernels. > > (see the thread "Kernel IEEE Math Completion (everything is zeroed!?!)" > at "http://lists.debian.org/debian-alpha/2006/06/";) > > That long and the short is that we figured out that it was busted somewhere > between 2.4.27 and 2.6.16, but no solutions were purposed. > > Later! -T > > -- > Tyson Whitehead (-twhitehe at uwo.ca -- MC-) > Computer Engineer Dept. of Applied Mathematics, > Graduate Student- Applied Mathematics University of Western Ontario, > GnuPG Key ID# 0xF7666BFF London, Ontario, Canada -- Camm Maguire[EMAIL PROTECTED] == "The earth is but one country, and mankind its citizens." -- Baha'u'llah -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
