RE: [Gcl-devel] RE: GCL bugs (decode-float, rationalize, expt)

[Mike Thomas]

Hi Stavros.

> Thanks for your confirmations of these bugs.  Just to clarify: I trust
> you agree that all these cases are buggy except for (expt 10 -300) and
> (integer-decode-float 1.0) (which both crash on my machine, but give the
> correct answers on Mike's and Camm's).

For the time being I trust you!!!  The logical processes sound sensible - I
balked at the arithmetic.

> I should have mentioned that I was using the most recent prebuilt binary
> (gcl_2.5.0.cvs20021220_mingw32), running on a 1GHz Athlon under Windows
> 2000.  Is it possible the Athlon is the issue in the cases where my
> result is different from yours?

Could be.  I'll try making an i486 binary later this week - can't do it
today sorry.

> Does GCL use GMP even for purely floating-point operations like
> decode-float?  Does GCL or GMP use architecture-specific instructions
> (gcc -march etc.?).

By way of example, "rationalize" and "rational" ("lsp/numlib.lsp") each
depend on "integer-decode-float" which in turn depends on
"Linteger_decode_float" ("o/num_co.c") and in turn "bignum2" from
"o/gmp_big.c".  I am fairly certain that if you trace the others you will
eventually tickle the GMP library.

GMP gets its speed through CPU specific hand coded assembler subroutines.

Cheers

Mike Thomas.