Re: [PATCH] Remove GMP in favor of MPFR

[Paul Brook <paul at codesourcery dot com>]

[ CCing gcc-patches. Original message can be found here: 
http://gcc.gnu.org/ml/fortran/2004-07/msg00005.html ]

On Sunday 04 July 2004 04:19, Steve Kargl wrote:
> Gang,
>
> Here is a reworking of gfortran to use MPFR in preference to
> GMP.  The diff (when uncompressed) is nearly 3800 lines and
> on the order of 100 kB.  The patch has been bootstrapped on
> FreeBSD.  The gfortran testsuite was run with no new regression.
> My private test suite also compiled and ran without a problem.

Looks ok.

+/* Andy may have used biased exponents e = E + bias where bias = 1.
+   Single precision model numbers.  */

I don't understand this comment, and the following defines are inconsistent 
(some differ by one, others are the same).

Is this a genuine error, or the difference between defining the valid 
exponents as <= E vs. < E ?

I'm fairly sure these numbers came from the IEEE-754. I remember changing them 
to not include denormalized numbers.

> There are a few FIXME comments where I simply converted the
> GMP code to MPFR code.  The reduced number of bits in the
> MPFR significand may lead to some potential overflows (see
> for example complex division in arith.c).  I clean the FIXME
> comments up with a followup patch.

Ok.

> I need someone who understands configure to write the
> tests for libmpfr.  For my testing, I hacked configure
> by replacing "-lgmp" everywhere with "-lmpfr -lgmp".

I don't claim to understand configure, but I'll see what I can come up with :)

> Finally, here's a mutant ChangeLog entry.  The complexity
> of patch has me somewhat stymied with submitting anything
> shorter.

I'd say one entry per file is sufficient for the "Convert GMP to MPFR" 
changes. New/removed functions should still be mentioned individually.

I'll give it a couple of days to see if anyone screams, then commit.

Paul