diff options
author | Ulrich Drepper <drepper@redhat.com> | 1997-11-13 00:21:19 +0000 |
---|---|---|
committer | Ulrich Drepper <drepper@redhat.com> | 1997-11-13 00:21:19 +0000 |
commit | fe0ec73edb48eff3785f29ddbac75cae5f8b361b (patch) | |
tree | d862765d3e48bfde1ee82de94b4e4f039ee7f205 /math | |
parent | cc3fa755128c70f8afc2d604936f33e4d7d652d8 (diff) |
Update.cvs/libc-ud-971112
1997-11-13 01:07 Ulrich Drepper <drepper@cygnus.com>
* manual/arith.texi: Update documentation according to most recent
ISO C 9X draft.
Document fma, fdim, fmin, and fmax.
* manual/math.texi: Allow multiple defitino of mul etc.
* math/complex.h (I): Define using _Complex_U not _Imaginary_I.
* math/libm-test.c: Add tests for fma.
* math/math.h: Describe DECIMAL_DIG macro. Pretty print.
* sysdeps/alpha/fpu/bits/mathdef.h: Define INFINITY as of type float.
Define DECIMAL_DIG.
* sysdeps/generic/bits/mathdef.h: Likewise.
* sysdeps/i386/bits/mathdef.h: Likewise.
* sysdeps/m68k/fpu/bits/mathdef.h: Likewise.
* sysdeps/powerpc/bits/mathdef.h: Likewise.
* sysdeps/sparc/fpu/bits/mathdef.h: Likewise.
* sysdeps/ieee754/bits/nan.h: Define NAN as of type float.
* sysdeps/m68k/bits/nan.h. Likewise. Remove NANF and NANL.
1997-11-12 17:50 Ulrich Drepper <drepper@cygnus.com>
* sunrpc/xcrypt.c: Don't process #ident preprocessor instruction.
Reported by Philip Blundell <pb@nexus.co.uk>.
* string/strndup.c: Use K&R like definition.
* sysdeps/unix/sysv/linux/getcwd.c: New file. Use kernel information
instead of longish search for the name.
* sysdeps/posix/getcwd.c: Add support for use of the code as a
backup solution.
1997-11-12 15:31 Philip Blundell <pb@nexus.co.uk>
* sysdeps/unix/sysv/linux/arm/sysdep.h (SYS_ify): Don't add
SWI_BASE in twice.
* sysdeps/unix/sysv/linux/arm/profil-counter.h (profil_counter):
Use correct name to access PC.
* sysdeps/unix/arm/sysdep.S: Include <bits/errno.h> not <errnos.h>.
* sysdeps/generic/bits/types.h: Add __ino64_t and __off64_t.
* sysdeps/generic/bits/stat.h: Add struct stat64.
1997-11-12 16:08 Ulrich Drepper <drepper@cygnus.com>
* intl/loadmsgcat.c [_LIBC] (fstat): Don't define as __fstat since
now we have a definition as _fxstat.
* libio/fileops.c: Likewise.
* libio/oldfileops.c: Likewise.
Reported by Andreas Jaeger <aj@arthur.rhein-neckar.de>.
1997-11-12 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* sysdeps/wordsize-32/inttypes.h (SIG_ATOMIC_MAX): Correct value.
* sysdeps/wordsize-64/inttypes.h (SIG_ATOMIC_MAX): Likewise.
1997-11-11 Paul Eggert <eggert@twinsun.com>
Add overflow checking for 64-bit time_t and 32-bit int.
* time/time.h (__offtime): Now returns int.
* time/offtime.c (__offtime): Return nonzero if successful;
check for tm_year overflow.
(DIV): New macro.
(LEAPS_THRU_END_OF): Handle negative years correctly.
* time/tzset.c (__tz_convert): Return NULL if offtime cannot convert.
* time/mktime.c (ranged_convert): New function.
(ydhms_tm_diff): Return nonzero if TP is null.
(__mktime_internal): Handle cases correctly even if they are near or
past the limits of time_t values that can be broken down to struct tm.
(print_tm, check_result, main): Diagnose localtime failures.
* manual/time.texi: Document the fact that localtime returns 0
if the time can't be represented.
1997-11-12 06:03 Ulrich Drepper <drepper@cygnus.com>
* time/strftime.c (memset_space, memset_zero): Use MEMPCPY, not
mempcpy. Patch by Ken'ichi Handa <handa@etl.go.jp>.
* manual/time.texi: Document %F and %f format for strftime.
* manual/arith.texi: Document copysign, nextafter and nan.
1997-11-06 Andreas Jaeger <aj@arthur.rhein-neckar.de>
* test-installation.pl: New file. Tests for some installation
problems.
Diffstat (limited to 'math')
-rw-r--r-- | math/complex.h | 5 | ||||
-rw-r--r-- | math/libm-test.c | 45 | ||||
-rw-r--r-- | math/math.h | 69 |
3 files changed, 84 insertions, 35 deletions
diff --git a/math/complex.h b/math/complex.h index f4b1d4640c..72b1e597fc 100644 --- a/math/complex.h +++ b/math/complex.h @@ -40,9 +40,10 @@ __BEGIN_DECLS XXX This probably has to go into a gcc related file. */ #define _Complex_I (1.0iF) -/* Another more descriptive name is `I'. */ +/* Another more descriptive name is `I'. + XXX Once we have the imaginary support switch this to _Imaginary_I. */ #undef I -#define I _Imaginary_I +#define I _Complex_I /* Optimization aids. This is not yet implemented in gcc and once it diff --git a/math/libm-test.c b/math/libm-test.c index c77cbee855..5d68b4953b 100644 --- a/math/libm-test.c +++ b/math/libm-test.c @@ -4813,6 +4813,50 @@ llround_test (void) static void +fma_test (void) +{ + check ("fma(1.0, 2.0, 3.0) = 5.0", FUNC(fma) (1.0, 2.0, 3.0), 5.0); + check_isnan ("fma(NaN, 2.0, 3.0) = NaN", FUNC(fma) (nan_value, 2.0, 3.0)); + check_isnan ("fma(1.0, NaN, 3.0) = NaN", FUNC(fma) (1.0, nan_value, 3.0)); + check_isnan_maybe_exc ("fma(1.0, 2.0, NaN) = NaN", + FUNC(fma) (1.0, 2.0, nan_value), INVALID_EXCEPTION); + check_isnan_maybe_exc ("fma(+Inf, 0.0, NaN) = NaN", + FUNC(fma) (plus_infty, 0.0, nan_value), + INVALID_EXCEPTION); + check_isnan_maybe_exc ("fma(-Inf, 0.0, NaN) = NaN", + FUNC(fma) (minus_infty, 0.0, nan_value), + INVALID_EXCEPTION); + check_isnan_maybe_exc ("fma(0.0, +Inf, NaN) = NaN", + FUNC(fma) (0.0, plus_infty, nan_value), + INVALID_EXCEPTION); + check_isnan_maybe_exc ("fma(0.0, -Inf, NaN) = NaN", + FUNC(fma) (0.0, minus_infty, nan_value), + INVALID_EXCEPTION); + check_isnan_exc ("fma(+Inf, 0.0, 1.0) = NaN", + FUNC(fma) (plus_infty, 0.0, 1.0), INVALID_EXCEPTION); + check_isnan_exc ("fma(-Inf, 0.0, 1.0) = NaN", + FUNC(fma) (minus_infty, 0.0, 1.0), INVALID_EXCEPTION); + check_isnan_exc ("fma(0.0, +Inf, 1.0) = NaN", + FUNC(fma) (0.0, plus_infty, 1.0), INVALID_EXCEPTION); + check_isnan_exc ("fma(0.0, -Inf, 1.0) = NaN", + FUNC(fma) (0.0, minus_infty, 1.0), INVALID_EXCEPTION); + + check_isnan_exc ("fma(+Inf, +Inf, -Inf) = NaN", + FUNC(fma) (plus_infty, plus_infty, minus_infty), + INVALID_EXCEPTION); + check_isnan_exc ("fma(-Inf, +Inf, +Inf) = NaN", + FUNC(fma) (minus_infty, plus_infty, plus_infty), + INVALID_EXCEPTION); + check_isnan_exc ("fma(+Inf, -Inf, +Inf) = NaN", + FUNC(fma) (plus_infty, minus_infty, plus_infty), + INVALID_EXCEPTION); + check_isnan_exc ("fma(-Inf, -Inf, -Inf) = NaN", + FUNC(fma) (minus_infty, minus_infty, minus_infty), + INVALID_EXCEPTION); +} + + +static void inverse_func_pair_test (const char *test_name, mathfunc f1, mathfunc inverse, MATHTYPE x, MATHTYPE epsilon) @@ -5205,6 +5249,7 @@ main (int argc, char *argv[]) ctanh_test (); csqrt_test (); cpow_test (); + fma_test (); /* special tests */ identities (); diff --git a/math/math.h b/math/math.h index 51055849ce..0aedaf1d56 100644 --- a/math/math.h +++ b/math/math.h @@ -68,37 +68,37 @@ __BEGIN_DECLS /* Include the file of declarations again, this time using `float' instead of `double' and appending f to each function name. */ -#ifndef _Mfloat_ -# define _Mfloat_ float -#endif -#define _Mdouble_ _Mfloat_ -#ifdef __STDC__ -# define __MATH_PRECNAME(name,r) name##f##r -#else -# define __MATH_PRECNAME(name,r) name/**/f/**/r -#endif -#include <bits/mathcalls.h> -#undef _Mdouble_ -#undef __MATH_PRECNAME - -#if __STDC__ - 0 || __GNUC__ - 0 -/* Include the file of declarations again, this time using `long double' - instead of `double' and appending l to each function name. */ - -# ifndef _Mlong_double_ -# define _Mlong_double_ long double +# ifndef _Mfloat_ +# define _Mfloat_ float # endif -# define _Mdouble_ _Mlong_double_ +# define _Mdouble_ _Mfloat_ # ifdef __STDC__ -# define __MATH_PRECNAME(name,r) name##l##r +# define __MATH_PRECNAME(name,r) name##f##r # else -# define __MATH_PRECNAME(name,r) name/**/l/**/r +# define __MATH_PRECNAME(name,r) name/**/f/**/r # endif # include <bits/mathcalls.h> # undef _Mdouble_ # undef __MATH_PRECNAME -#endif /* __STDC__ || __GNUC__ */ +# if __STDC__ - 0 || __GNUC__ - 0 +/* Include the file of declarations again, this time using `long double' + instead of `double' and appending l to each function name. */ + +# ifndef _Mlong_double_ +# define _Mlong_double_ long double +# endif +# define _Mdouble_ _Mlong_double_ +# ifdef __STDC__ +# define __MATH_PRECNAME(name,r) name##l##r +# else +# define __MATH_PRECNAME(name,r) name/**/l/**/r +# endif +# include <bits/mathcalls.h> +# undef _Mdouble_ +# undef __MATH_PRECNAME + +# endif /* __STDC__ || __GNUC__ */ #endif /* Use misc or ISO C 9X. */ #undef __MATHDECL_1 @@ -118,23 +118,23 @@ extern int signgam; /* Get the architecture specific values describing the floating-point evaluation. The following symbols will get defined: - float_t floating-point type at least as wide as `float' used + float_t floating-point type at least as wide as `float' used to evaluate `float' expressions - double_t floating-point type at least as wide as `double' used + double_t floating-point type at least as wide as `double' used to evaluate `double' expressions - FLT_EVAL_METHOD + FLT_EVAL_METHOD Defined to 0 if `float_t' is `float' and `double_t' is `double' 1 if `float_t' and `double_t' are `double' 2 if `float_t' and `double_t' are `long double' else `float_t' and `double_t' are unspecified - INFINITY representation of the infinity value of type `float_t' + INFINITY representation of the infinity value of type `float' - FP_FAST_FMA - FP_FAST_FMAF - FP_FAST_FMAL + FP_FAST_FMA + FP_FAST_FMAF + FP_FAST_FMAL If defined it indicates that the the `fma' function generally executes about as fast as a multiply and an add. This macro is defined only iff the `fma' function is @@ -143,6 +143,9 @@ extern int signgam; FP_ILOGB0 Expands to a value returned by `ilogb (0.0)'. FP_ILOGBNAN Expands to a value returned by `ilogb (NAN)'. + DECIMAL_DIG Number of decimal digits supported by conversion between + decimal and all internal floating-point formats. + */ # include <bits/mathdef.h> @@ -211,8 +214,8 @@ extern _LIB_VERSION_TYPE _LIB_VERSION; /* In SVID error handling, `matherr' is called with this description of the exceptional condition. - We have a problem when using C++ since `exception' is reserved in - C++. */ + We have a problem when using C++ since `exception' is a reserved + name in C++. */ # ifdef __cplusplus struct __exception # else @@ -307,7 +310,7 @@ extern int matherr __P ((struct exception *__exc)); for unordered numbers. Since many FPUs provide special instructions to support these operations and these tests are defined in <bits/mathinline.h>, we define the generic macros at - this late point. */ + this late point and only if they are not defined yet. */ /* Return nonzero value if X is greater than Y. */ # ifndef isgreater |