Commit Graph

274 Commits

Author SHA1 Message Date
David Schultz
3d266bde6d Replace strong references with weak references. There's no
particularly good reason to do this, except that __strong_reference
does type checking, whereas __weak_reference does not.
On Alpha, the compiler won't accept a 'long double' parameter in
place of a 'double' parameter even thought the two types are
identical.
2005-03-07 21:27:37 +00:00
Stefan Farfeleder
3ddc6e9440 Remove an obsolete sentence from a comment. 2005-03-07 20:28:26 +00:00
David Schultz
c8642491d5 - If z is 0, one of x or y is 0, and the other is infinite, raise
an invalid exception and return an NaN.
- If a long double has 113 bits of precision, implement fma in terms
  of simple long double arithmetic instead of complicated double arithmetic.
- If a long double is the same as a double, alias fma as fmal.
2005-03-07 05:02:09 +00:00
David Schultz
388bf3b630 Document scalbnl and scalblnl. 2005-03-07 05:00:44 +00:00
David Schultz
6af2c5a60c Document nextafterl and nexttoward{,f,l}. 2005-03-07 05:00:29 +00:00
David Schultz
15a53f77fd Add nexttoward to the list of implemented functions, and explicitly
list the four that are still missing.
2005-03-07 04:59:53 +00:00
David Schultz
66d672d8cb Document fmal. 2005-03-07 04:59:43 +00:00
David Schultz
94e03502dc Remove ldexp and ldexpf. The former is in libc, and the latter is
identical to scalbnf, which is now aliased as ldexpf.  Note that the
old implementations made the mistake of setting errno and were the
only libm routines to do so.
2005-03-07 04:59:30 +00:00
David Schultz
aeb5e711f3 - Remove s_ldexpf.c (now aliased to scalbn.)
- Add nexttoward{,f,l} and nextafterl.  On all platforms,
  nexttowardl is an alias for nextafterl.
- Add fmal.
- Add man pages for new routines: fmal, nextafterl,
  nexttoward{,f,l}, scalb{,l}nl.

Note that on platforms where long double is the same as double, we
generally just alias the double versions of the routines, since doing
so avoids extra work on the source code level and redundant code in
the binary.  In particular:

		ldbl53		ldbl64/113
fmal       	s_fma.c		s_fmal.c
ldexpl     	s_scalbn.c	s_scalbnl.c
nextafterl 	s_nextafter.c	s_nextafterl.c
nexttoward 	s_nextafter.c	s_nexttoward.c
nexttowardf	s_nexttowardf.c	s_nexttowardf.c
nexttowardl	s_nextafter.c	s_nextafterl.c
scalbnl    	s_scalbn.c	s_scalbnl.c
2005-03-07 04:59:11 +00:00
David Schultz
228ad57d05 - Define FP_FAST_FMA for sparc64, since fma() is now implemented using
sparc64's 128-bit long doubles.
- Define FP_FAST_FMAL for ia64.
- Prototypes for fmal, frexpl, ldexpl, nextafterl, nexttoward{,f,l},
  scalblnl, and scalbnl.
2005-03-07 04:58:43 +00:00
David Schultz
beed720c37 Alias scalbn as ldexpl and scalbnl on platforms where long double is
the same as double.
2005-03-07 04:58:03 +00:00
David Schultz
7b6a19039d - Implement scalblnl.
- In scalbln and scalblnf, check the bounds of the second argument.
  This is probably unnecessary, but strictly speaking, we should
  report an error if someone tries to compute scalbln(x, INT_MAX + 1ll).
2005-03-07 04:57:50 +00:00
David Schultz
caacab9b5f Implement nexttowardf. This is used on both platforms with 11-bit
exponents and platforms with 15-bit exponents for long doubles.
2005-03-07 04:57:38 +00:00
David Schultz
ef94de735a Implement nexttoward and nextafterl; the latter is also known as
nexttowardl.  These are not needed on machines where long doubles
look like IEEE-754 doubles, so the implementation only supports
the usual long double formats with 15-bit exponents.

Anything bizarre, such as machines where floating-point and integer
data have different endianness, will cause problems.  This is the case
with big endian ia64 according to libc/ia64/_fpmath.h.  Please contact
me if you managed to get a machine running this way.
2005-03-07 04:56:46 +00:00
David Schultz
a506506a1c - Try harder to trick gcc into not optimizing away statements
that are intended to raise underflow and inexact exceptions.
- On systems where long double is the same as double, nextafter
  should be aliased as nexttoward, nexttowardl, and nextafterl.
2005-03-07 04:55:58 +00:00
David Schultz
e0fe8e4440 Implement frexpl. 2005-03-07 04:54:51 +00:00
David Schultz
f8a40fca14 Alias frexp as frexpl on platforms where a long double is the same as
a double.
2005-03-07 04:54:39 +00:00
David Schultz
65e60ab108 Implement fmal. 2005-03-07 04:54:20 +00:00
David Schultz
b1f37dcef4 - Define the LDBL_PREC to be the number of significant bits in a long
double's mantissa.
- Add an assembly version of fmal.
2005-03-07 04:54:02 +00:00
David Schultz
99401fa2e9 - Define the LDBL_PREC to be the number of significant bits in a long
double's mantissa.
- Add an assembly version of scalbnl.
2005-03-07 04:53:48 +00:00
David Schultz
4be31f0664 Define the LDBL_PREC to be the number of significant bits in a long
double's mantissa.
2005-03-07 04:53:36 +00:00
David Schultz
4442891961 Add an assembly version of fmal. 2005-03-07 04:53:11 +00:00
David Schultz
cd7d05b5a2 Add scalbnl, also known as as ldexpl. 2005-03-07 04:52:58 +00:00
David Schultz
4b2011300b Alias scalbnf as ldexpf. The two are identical in binary
floating-point formats.
2005-03-07 04:52:43 +00:00
David Schultz
1b32579f23 Fix a mistake in the exponent range. 2005-03-06 19:08:18 +00:00
David Schultz
f4a5643005 Work around a gcc bug. This fixes feholdexcept() et al. at -O1.
Symptoms of the problem included assembler warnings and
nondeterministic runtime behavior when a fe*() call that affects the
fpsr is closely followed by a float point op.

The bug (at least, I think it's a bug) is that gcc does not insert a
break between a volatile asm and a dependent instruction if the
volatile asm came from an inlined function.  Volatile asms seem to be
fine in other circumstances, even without -mvolatile-asm-stop, so
perhaps the compiler adds the stop bits before inlining takes place.
The problem does not occur at -O0 because inlining is disabled, and it
doesn't happen at -O2 because -fschedule-insns2 knows better.
2005-03-05 20:34:45 +00:00
David Schultz
57276bb6ea Un-document the non-extant exp10() and exp10f() functions.
exp10() was a casualty of the transition away from the VAX.
2005-02-26 08:54:45 +00:00
David Schultz
aa28340df9 Revert rev 1.8, which causes small (e.g. 2 ulp) errors for some
inputs.  The trouble with replacing two floats with a double is that
the latter has 6 extra bits of precision, which actually hurts
accuracy in many cases.  All of the constants are optimal when float
arithmetic is used, and would need to be recomputed to do this right.

Noticed by:	bde (ucbtest)
2005-02-24 06:32:13 +00:00
David Schultz
adec44c08b Use hardware instructions for sqrt() and sqrtf(). 2005-02-21 18:27:57 +00:00
David Schultz
96efaf6c36 Use double arithmetic instead of simulating it with two floats. This
results in a performance gain on the order of 10% for amd64 (sledge),
ia64 (pluto1), i386+SSE (Pentium 4), and sparc64 (panther), and a
negligible improvement for i386 without SSE.  (The i386 port still
uses the hardware instruction, though.)
2005-02-21 17:44:57 +00:00
David Schultz
f674c13c78 Remove the i387 versions of atan(), atan2(), and atan2f().
They are slower than the MI routines on modern hardware,
except for degenerate cases such as the Pentium 4.

PR:		67469
2005-02-21 16:04:23 +00:00
David Schultz
c4691a5da9 Remove i387 versions of asin() and acos(). Although the hardware
instruction was faster on the 486, it's slower than our MD version on
modern processors.

Determined by:	bde
PR:		67469
2005-02-20 22:51:08 +00:00
David Schultz
dab1571b90 Remove the float versions of the i387 trig functions obtained from
NetBSD.  They're buggy, giving particularly for inputs larger in
magnitude than 2**63.

Noticed by:	bde
PR:		67469
2005-02-20 22:50:40 +00:00
David Schultz
e02846ce13 Fix a small scripting snafu in the previous revision. 2005-02-04 20:05:39 +00:00
David Schultz
b21154f677 Remove another vestige of support for a non-IEEE libm. 2005-02-04 18:32:13 +00:00
David Schultz
3f70824172 Reduce diffs against vendor source (Sun fdlibm 5.3). 2005-02-04 18:26:06 +00:00
David Schultz
79b990338f Move machine-dependent crud to its own makefile. 2005-02-04 14:33:39 +00:00
David Schultz
e1b61b5b93 Remove wrappers and other cruft intended to support SVID, mistakes in
C90, and other arcana.  Most of these features were never fully
supported or enabled by default.

Ok:	bde, stefanf
2005-02-04 14:08:32 +00:00
Ruslan Ermilov
1f8ee0e102 Typo. 2005-01-28 21:14:16 +00:00
Ruslan Ermilov
d7a604cc33 Properly terminate sentence. 2005-01-28 21:13:34 +00:00
David Schultz
29bf6af890 - Move the functions presently described in in ieee(3) to their own
manpages.  They are not very related, so separating them makes it
  easier to add meaningful cross-references and extend some of the
  descriptions.
- Move the part of math(3) that discusses IEEE 754 to the ieee(3)
  manpage.
2005-01-27 05:46:17 +00:00
Olivier Houchard
15d3b4db61 Define FE_TONEAREST, FE_TOWARDZERO, FE_UPWARD, FE_DOWNWARD and _ROUND_MASK to
unbreak the build for arm.
2005-01-24 00:35:02 +00:00
David Schultz
cb2d2321cd Update comment to reflect the code change in the previous revision.
Noticed by:	ceri
2005-01-23 22:56:08 +00:00
David Schultz
52611c608e Many changes, including the following major ones:
- Rearrange the list of functions into categories.
- Remove the ulps column.  It was appropriate for only some
  of the functions in the list, and correct for even fewer
  of them.
- Add some new paragraphs, and remove some old ones about
  NaNs that may do more harm than good.
- Document precisions other than double-precision.
2005-01-23 22:05:33 +00:00
David Schultz
3c4d0a0973 If x == y, return y, not x. C99 (though not IEEE 754) requires that
nextafter(+0.0, -0.0) returns -0.0 and nextafter(-0.0, +0.0) returns +0.0.
2005-01-23 15:46:22 +00:00
David Schultz
d5580d091a Add fma() and fmaf(), which implement a fused multiply-add operation. 2005-01-22 09:53:18 +00:00
Ruslan Ermilov
24a0682c64 Sort sections. 2005-01-20 09:17:07 +00:00
Ruslan Ermilov
5391441c05 Use the \*(If string provided by mdoc(7), to represent infinity. 2005-01-16 16:49:10 +00:00
Ruslan Ermilov
1fbb01b7f0 Removed redundant .br call. 2005-01-16 16:46:14 +00:00
David Schultz
cd3cc47033 amd64 assembly versions of sqrt(), lrint(), and llrint() using SSE2. 2005-01-15 03:32:28 +00:00