Thu Nov 15 07:05:39 EDT 1990
Shar'd beta version for release to testers.

Sat Nov 17 17:06:17 EDT 1990
Fixed floor and ceil so that they address their local variables at -ve
offsets relative to the frame pointer.

Fixed hypot.s so that unnecessary register loads are avoided. I wasn't aware
of the correct assembler mnemonic to multiply %st(0) by the contents of a memory
location.

Sat Nov 17 18:49:31 EDT 1990
Copysign rewritten in assembler.

Sun Nov 18 11:21:24 EDT 1990
Sinh and cosh now in assembler.
Tanh now in assembler.
Fixed nasty bug in acos.s.
exp__E no longer needed - source deleted.

Tue Nov 20 21:24:26 EDT 1990
Sinh and cosh now multiply by 0.5 instead of dividing by 2. It's faster.
Added setinternal for convenience in some cases. Using this makes the
output of paranoia closer to that obtained on a Sun 4.
Added setcont.
Modified the Makefile so that test and paranoia are synonymous.
Modified atan2.c to avoid a floating point register load.
Fixed bug in atan2 - was giving incorrect result when y/x = +/-inf.

Thu Nov 22 06:52:47 EDT 1990
Fixed local variable handling in assembler files. Stack needs to be 
allocated for local variables.
All assembler routines are now .align 4.

Thu Nov 22 19:24:09 EDT 1990
Added MASK_ALL to fpumath.h

Fri Nov 23 21:11:33 EDT 1990
Inverse hyperbolics missing from fpumath.h - fixed.
Some definitions missing from atanh.c - didn't work on some systems.
Coded inverse hyperbolics in assembler.
Coded atan2 in assembler.

Sat Nov 24 17:12:34 EDT 1990
Modified hypot.s to avoid use of ffree.
atan2 was not working correctly - fixed
Ready for netwide release.

Mon Nov 26 19:25:56 EDT 1990
Put in explicit call to cc for compiling lgamma. This means that the
library can link to code not compiled with gcc (not quite true pow.s has to be
changed).

Wed Nov 28 22:57:43 EDT 1990
Coded the is??? functions required for IEEE conformance
Coded ieee_retrospective() which prints a summary of IEEE exceptions that are on
when the function is called.

Fri Nov 30 17:35:29 EDT 1990
Removed the last vestige of GNU specifics. You can now use the library with 'cc'
although you need gcc/gas to create it.

Sat Dec  1 15:48:19 EDT 1990
Found a bug in pow (pow(-x,x) shoudn't work for x non-integral) fixed.

Sat Dec  1 21:00:48 EST 1990
Rewrote copysign in a more efficient way.

Sun Dec  2 06:51:59 EDT 1990
Added infinity().
Amended README to show the current state.
Changed fpumath.h so as to define HUGE = infinity() if you compile (user
code) with IEEE defined.

Sun Dec  2 08:26:20 EDT 1990
Modified log functions and inverse hyperbolics for greater accuracy.

Sun Dec  2 10:08:47 EDT 1990
Added MASK_ALL1 to fpumath.h. This enables 64 bit precision significands.
Coded sqrtp to ensure that sqrt is rounded to 53 bits - used in
paranoia.c.
Put sqrtp in fpumath.h and in README.
Made very minor changes to pow.s, floor.s and ceil.s.
ieee_retrospective was printing its leading info for a LOS exception -
fixed.
Rewrote finite so that it does not use the floating point unit and so does
not raise any exceptions.
Produced d2dcomb.summ.

Sun Dec  2 18:40:47 EDT 1990
Fixed exp, exp10, expm1, exp2, sinh, cosh, tanh to gain more speed.

Wed Dec  5 17:25:50 EDT 1990
Coded max_..., min_..., etc.

Wed Dec  5 19:38:56 EDT 1990
Fixed some local parameter handling errors in a few of the assembler
files.

Wed Dec  5 20:54:37 EDT 1990
Define MAX(MIN)DOUBLE in terms of max(min)_(sub)normal in fpumath.h.

Sat Dec  8 06:09:01 EDT 1990
Fixed bug in isnan 8(%ebp) not $8(%ebp) - typo!
Fixed bug in isinf - not doing correct test.

Sat Dec  8 06:36:07 EDT 1990
Coded nextafter.

Sat Dec  8 06:46:44 EDT 1990
MAXDOUBLE is wrong in math.h it is correct in fpumath.h
Released to net. Posted to alt.sources.

Sun Dec  9 08:26:23 EDT 1990
Added a couple more special cases to nextafter.

Sun Dec  9 12:16:04 EDT 1990
Slight modification to Makefile

Sun Dec  9 14:33:13 EDT 1990
Coded float version - still testing. You must have an ANSI compiler (gcc is
ok) to use this version.

Mon Dec 10 17:21:16 EDT 1990
Float version done.
Fully prototyped all functions in fpumath.h.
Version 2.0 done.

Tue Dec 11 06:22:51 EDT 1990
Added a couple of extra cases to nextafter[f].
Made a couple of minor changes to fpumath.h.
Removed AT&T code (j0.c, j1.c, jn.c, erf.c, lgamma.c). These were
avaialble with BSD 4.3 so I had no way of knowing...

Wed Dec 12 05:50:31 EDT 1990
Finally got nextafter[f] working correctly.

Thu Dec 13 09:44:37 EDT 1990
sinh[f]() now gives good results for all argument ranges. Compute using a 
combination of exp() and a ratio of polynomials.

Thu Dec 13 18:48:55 EDT 1990
Compute expm1() as expm1() = exp(x)*(1-exp(-x)). This is much better than
before but still not perfect. I need Hart & Cheney...

Thu Dec 13 19:38:31 EDT 1990
cosh() was not producing good results for small x. Fixed.

Fri Dec 14 08:19:56 EDT 1990
Berkeley provided j0.c, j1.c, erf.c, gamma.c and lgamma.c (no jn.c yet) now
being used. This is *not* BSD code and is not covered by the BSD licence.
***No Berkeley code used.***
mathimpl.h deleted.
README updated.
Defined _getsw() (in ieee_retrospective.c) as a short.

Fri Dec 14 11:43:30 EDT 1990
Added shar to Makefile.
Version 2.0 ready for distribution.
