glibc/sysdeps/ieee754/ldbl-128ibm/math_ldbl.h

233 lines
6.1 KiB
C
Raw Normal View History

2006-01-27 Dwayne Grant McConnell <decimal@us.ibm.com> Jakub Jelinek <jakub@redhat.com> Roland McGrath <roland@redhat.com> Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/powerpc64/fpu/s_truncf.S: Comment fix. * sysdeps/powerpc/powerpc32/fpu/s_truncf.S: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_llroundf.S: Likewise. * sysdeps/powerpc/fpu/libm-test-ulps: Update. * math/libm-test.inc (check_float_internal): Allow ulp <= 0.5. (erfc_test): Don't run erfcl (27.0L) test if erfcl (27.0L) is denormal. [TEST_LDOUBLE] (ceil_test, floor_test, llrint_test, llround_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/powerpc/powerpc32/fpu/s_copysignl.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fabs.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fabsl.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fdim.c: New file. * sysdeps/powerpc/powerpc32/fpu/s_fmax.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fmin.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_isnan.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_ceill.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_copysignl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fabs.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fabsl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fdim.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_floorl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fmax.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fmin.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_isnan.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_llrintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_llroundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_lrintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_lroundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_nearbyintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_roundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_truncl.S: New file. * sysdeps/unix/sysv/linux/powerpc/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/powerpc64/fpu/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/configure.in: New file. * sysdeps/unix/sysv/linux/powerpc/configure: New file. * sysdeps/unix/sysv/linux/powerpc/bits/wordsize.h (__LONG_DOUBLE_MATH_OPTIONAL): Define. (__NO_LONG_DOUBLE_MATH): Define. * sysdeps/unix/sysv/linux/powerpc/nldbl-abi.h: New file. * sysdeps/powerpc/fpu/s_isnan.c: Include math_ldbl_opt.h. * sysdeps/powerpc/powerpc64/fpu/s_ceil.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (ceill): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_copysign.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (copysignl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_floor.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (floorl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_llrint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llrintl, lrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_llround.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llroundl, lroundl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_rint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (rintl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_round.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (roundl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_trunc.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (truncl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_ceil.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (ceill): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_copysign.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (copysignl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_floor.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (floorl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_lrint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (lrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_llrint.c: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_lround.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (lroundl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_rint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (rintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_round.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (roundl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_trunc.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (truncl): Add compatibility symbols. * misc/qefgcvt_r.c [LDBL_MIN_10_EXP == -291] (FLOAT_MIN_10_NORM): New. * sysdeps/powerpc/fpu/bits/mathdef.h (__NO_LONG_DOUBLE_MATH): Remove. * sysdeps/powerpc/Implies: Add ieee754/ldbl-128ibm. * sysdeps/powerpc/powerpc32/Implies: Remove powerpc/soft-fp. * sysdeps/ieee754/ldbl-128ibm/Makefile: New file. * sysdeps/ieee754/ldbl-128ibm/e_acoshl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_acosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_asinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_atan2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_atanhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_coshl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_expl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_gammal_r.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_hypotl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_j0l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_j1l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_lgammal_r.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_log10l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_log2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_logl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_powl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_remainderl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_sinhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_sqrtl.c: New file. * sysdeps/ieee754/ldbl-128ibm/ieee754.h: New file. * sysdeps/ieee754/ldbl-128ibm/k_cosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_sinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_tanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c: New file. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h: New file. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_asinhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_atanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_cbrtl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_cosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_erfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_expm1l.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_fabsl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_finitel.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_frexpl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_ilogbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_isinfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_isnanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_log1pl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_logbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_modfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_remquol.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_scalblnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_scalbnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_signbitl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_sinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_tanhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_tanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: New file. * sysdeps/ieee754/ldbl-128ibm/strtold_l.c: New file. * sysdeps/ieee754/ldbl-128ibm/t_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/w_expl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_copysignl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_llrintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_llroundl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nearbyintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_lrintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_lroundl.c: New file. * sysdeps/ieee754/ldbl-128/e_powl.c: Fix old comment.
2006-01-28 00:15:15 +00:00
#ifndef _MATH_PRIVATE_H_
#error "Never use <math_ldbl.h> directly; include <math_private.h> instead."
#endif
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
#include <ieee754.h>
#include <stdint.h>
2013-06-05 20:44:03 +00:00
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
/* To suit our callers we return *hi64 and *lo64 as if they came from
an ieee854 112 bit mantissa, that is, 48 bits in *hi64 (plus one
implicit bit) and 64 bits in *lo64. */
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
static inline void
2012-06-05 13:13:41 +00:00
ldbl_extract_mantissa (int64_t *hi64, uint64_t *lo64, int *exp, long double x)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
/* We have 105 bits of mantissa plus one implicit digit. Since
106 bits are representable we use the first implicit digit for
the number before the decimal point and the second implicit bit
as bit 53 of the mantissa. */
2012-06-05 13:13:41 +00:00
uint64_t hi, lo;
union ibm_extended_long_double u;
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
u.ld = x;
*exp = u.d[0].ieee.exponent - IEEE754_DOUBLE_BIAS;
2006-01-27 Dwayne Grant McConnell <decimal@us.ibm.com> Jakub Jelinek <jakub@redhat.com> Roland McGrath <roland@redhat.com> Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/powerpc64/fpu/s_truncf.S: Comment fix. * sysdeps/powerpc/powerpc32/fpu/s_truncf.S: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_llroundf.S: Likewise. * sysdeps/powerpc/fpu/libm-test-ulps: Update. * math/libm-test.inc (check_float_internal): Allow ulp <= 0.5. (erfc_test): Don't run erfcl (27.0L) test if erfcl (27.0L) is denormal. [TEST_LDOUBLE] (ceil_test, floor_test, llrint_test, llround_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/powerpc/powerpc32/fpu/s_copysignl.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fabs.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fabsl.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fdim.c: New file. * sysdeps/powerpc/powerpc32/fpu/s_fmax.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fmin.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_isnan.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_ceill.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_copysignl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fabs.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fabsl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fdim.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_floorl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fmax.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fmin.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_isnan.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_llrintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_llroundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_lrintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_lroundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_nearbyintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_roundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_truncl.S: New file. * sysdeps/unix/sysv/linux/powerpc/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/powerpc64/fpu/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/configure.in: New file. * sysdeps/unix/sysv/linux/powerpc/configure: New file. * sysdeps/unix/sysv/linux/powerpc/bits/wordsize.h (__LONG_DOUBLE_MATH_OPTIONAL): Define. (__NO_LONG_DOUBLE_MATH): Define. * sysdeps/unix/sysv/linux/powerpc/nldbl-abi.h: New file. * sysdeps/powerpc/fpu/s_isnan.c: Include math_ldbl_opt.h. * sysdeps/powerpc/powerpc64/fpu/s_ceil.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (ceill): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_copysign.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (copysignl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_floor.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (floorl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_llrint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llrintl, lrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_llround.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llroundl, lroundl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_rint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (rintl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_round.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (roundl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_trunc.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (truncl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_ceil.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (ceill): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_copysign.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (copysignl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_floor.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (floorl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_lrint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (lrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_llrint.c: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_lround.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (lroundl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_rint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (rintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_round.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (roundl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_trunc.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (truncl): Add compatibility symbols. * misc/qefgcvt_r.c [LDBL_MIN_10_EXP == -291] (FLOAT_MIN_10_NORM): New. * sysdeps/powerpc/fpu/bits/mathdef.h (__NO_LONG_DOUBLE_MATH): Remove. * sysdeps/powerpc/Implies: Add ieee754/ldbl-128ibm. * sysdeps/powerpc/powerpc32/Implies: Remove powerpc/soft-fp. * sysdeps/ieee754/ldbl-128ibm/Makefile: New file. * sysdeps/ieee754/ldbl-128ibm/e_acoshl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_acosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_asinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_atan2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_atanhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_coshl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_expl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_gammal_r.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_hypotl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_j0l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_j1l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_lgammal_r.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_log10l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_log2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_logl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_powl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_remainderl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_sinhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_sqrtl.c: New file. * sysdeps/ieee754/ldbl-128ibm/ieee754.h: New file. * sysdeps/ieee754/ldbl-128ibm/k_cosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_sinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_tanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c: New file. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h: New file. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_asinhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_atanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_cbrtl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_cosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_erfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_expm1l.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_fabsl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_finitel.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_frexpl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_ilogbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_isinfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_isnanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_log1pl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_logbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_modfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_remquol.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_scalblnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_scalbnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_signbitl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_sinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_tanhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_tanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: New file. * sysdeps/ieee754/ldbl-128ibm/strtold_l.c: New file. * sysdeps/ieee754/ldbl-128ibm/t_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/w_expl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_copysignl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_llrintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_llroundl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nearbyintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_lrintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_lroundl.c: New file. * sysdeps/ieee754/ldbl-128/e_powl.c: Fix old comment.
2006-01-28 00:15:15 +00:00
lo = ((uint64_t) u.d[1].ieee.mantissa0 << 32) | u.d[1].ieee.mantissa1;
hi = ((uint64_t) u.d[0].ieee.mantissa0 << 32) | u.d[0].ieee.mantissa1;
2013-06-05 20:44:03 +00:00
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
if (u.d[0].ieee.exponent != 0)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
int ediff;
/* If not a denormal or zero then we have an implicit 53rd bit. */
hi |= (uint64_t) 1 << 52;
if (u.d[1].ieee.exponent != 0)
lo |= (uint64_t) 1 << 52;
else
/* A denormal is to be interpreted as having a biased exponent
of 1. */
lo = lo << 1;
/* We are going to shift 4 bits out of hi later, because we only
want 48 bits in *hi64. That means we want 60 bits in lo, but
we currently only have 53. Shift the value up. */
lo = lo << 7;
/* The lower double is normalized separately from the upper.
We may need to adjust the lower mantissa to reflect this.
The difference between the exponents can be larger than 53
when the low double is much less than 1ULP of the upper
(in which case there are significant bits, all 0's or all
1's, between the two significands). The difference between
the exponents can be less than 53 when the upper double
exponent is nearing its minimum value (in which case the low
double is denormal ie. has an exponent of zero). */
ediff = u.d[0].ieee.exponent - u.d[1].ieee.exponent - 53;
if (ediff > 0)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
if (ediff < 64)
lo = lo >> ediff;
else
lo = 0;
}
else if (ediff < 0)
lo = lo << -ediff;
if (u.d[0].ieee.negative != u.d[1].ieee.negative
&& lo != 0)
{
hi--;
lo = ((uint64_t) 1 << 60) - lo;
if (hi < (uint64_t) 1 << 52)
{
/* We have a borrow from the hidden bit, so shift left 1. */
hi = (hi << 1) | (lo >> 59);
lo = (((uint64_t) 1 << 60) - 1) & (lo << 1);
*exp = *exp - 1;
}
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
}
}
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
else
/* If the larger magnitude double is denormal then the smaller
one must be zero. */
hi = hi << 1;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
*lo64 = (hi << 60) | lo;
*hi64 = hi >> 4;
}
2006-01-27 Dwayne Grant McConnell <decimal@us.ibm.com> Jakub Jelinek <jakub@redhat.com> Roland McGrath <roland@redhat.com> Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/powerpc64/fpu/s_truncf.S: Comment fix. * sysdeps/powerpc/powerpc32/fpu/s_truncf.S: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_llroundf.S: Likewise. * sysdeps/powerpc/fpu/libm-test-ulps: Update. * math/libm-test.inc (check_float_internal): Allow ulp <= 0.5. (erfc_test): Don't run erfcl (27.0L) test if erfcl (27.0L) is denormal. [TEST_LDOUBLE] (ceil_test, floor_test, llrint_test, llround_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/powerpc/powerpc32/fpu/s_copysignl.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fabs.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fabsl.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fdim.c: New file. * sysdeps/powerpc/powerpc32/fpu/s_fmax.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_fmin.S: New file. * sysdeps/powerpc/powerpc32/fpu/s_isnan.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_ceill.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_copysignl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fabs.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fabsl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fdim.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_floorl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fmax.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_fmin.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_isnan.c: New file. * sysdeps/powerpc/powerpc64/fpu/s_llrintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_llroundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_lrintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_lroundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_nearbyintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_roundl.S: New file. * sysdeps/powerpc/powerpc64/fpu/s_truncl.S: New file. * sysdeps/unix/sysv/linux/powerpc/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/powerpc64/fpu/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/Implies: New file. * sysdeps/unix/sysv/linux/powerpc/configure.in: New file. * sysdeps/unix/sysv/linux/powerpc/configure: New file. * sysdeps/unix/sysv/linux/powerpc/bits/wordsize.h (__LONG_DOUBLE_MATH_OPTIONAL): Define. (__NO_LONG_DOUBLE_MATH): Define. * sysdeps/unix/sysv/linux/powerpc/nldbl-abi.h: New file. * sysdeps/powerpc/fpu/s_isnan.c: Include math_ldbl_opt.h. * sysdeps/powerpc/powerpc64/fpu/s_ceil.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (ceill): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_copysign.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (copysignl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_floor.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (floorl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_llrint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llrintl, lrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_llround.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llroundl, lroundl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_rint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (rintl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_round.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (roundl): Add compatibility symbols. * sysdeps/powerpc/powerpc64/fpu/s_trunc.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (truncl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_ceil.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (ceill): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_copysign.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (copysignl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_floor.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (floorl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_lrint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (lrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_llrint.c: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (llrintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_lround.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (lroundl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_rint.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (rintl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_round.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (roundl): Add compatibility symbols. * sysdeps/powerpc/powerpc32/fpu/s_trunc.S: Include math_ldbl_opt.h. [LONG_DOUBLE_COMPAT] (truncl): Add compatibility symbols. * misc/qefgcvt_r.c [LDBL_MIN_10_EXP == -291] (FLOAT_MIN_10_NORM): New. * sysdeps/powerpc/fpu/bits/mathdef.h (__NO_LONG_DOUBLE_MATH): Remove. * sysdeps/powerpc/Implies: Add ieee754/ldbl-128ibm. * sysdeps/powerpc/powerpc32/Implies: Remove powerpc/soft-fp. * sysdeps/ieee754/ldbl-128ibm/Makefile: New file. * sysdeps/ieee754/ldbl-128ibm/e_acoshl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_acosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_asinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_atan2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_atanhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_coshl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_expl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_gammal_r.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_hypotl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_j0l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_j1l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_jnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_lgammal_r.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_log10l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_log2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_logl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_powl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_remainderl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_sinhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/e_sqrtl.c: New file. * sysdeps/ieee754/ldbl-128ibm/ieee754.h: New file. * sysdeps/ieee754/ldbl-128ibm/k_cosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_sinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/k_tanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c: New file. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h: New file. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_asinhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_atanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_cbrtl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_cosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_erfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_expm1l.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_fabsl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_finitel.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_frexpl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_ilogbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_isinfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_isnanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_log1pl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_logbl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_modfl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nextafterl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_remquol.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_scalblnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_scalbnl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_signbitl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_sinl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_tanhl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_tanl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: New file. * sysdeps/ieee754/ldbl-128ibm/strtold_l.c: New file. * sysdeps/ieee754/ldbl-128ibm/t_sincosl.c: New file. * sysdeps/ieee754/ldbl-128ibm/w_expl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_copysignl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_llrintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_llroundl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_nearbyintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_lrintl.c: New file. * sysdeps/ieee754/ldbl-128ibm/s_lroundl.c: New file. * sysdeps/ieee754/ldbl-128/e_powl.c: Fix old comment.
2006-01-28 00:15:15 +00:00
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
static inline long double
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
ldbl_insert_mantissa (int sign, int exp, int64_t hi64, uint64_t lo64)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
union ibm_extended_long_double u;
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
int expnt2;
uint64_t hi, lo;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
u.d[0].ieee.negative = sign;
u.d[1].ieee.negative = sign;
u.d[0].ieee.exponent = exp + IEEE754_DOUBLE_BIAS;
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
u.d[1].ieee.exponent = 0;
expnt2 = exp - 53 + IEEE754_DOUBLE_BIAS;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
/* Expect 113 bits (112 bits + hidden) right justified in two longs.
2013-06-05 20:44:03 +00:00
The low order 53 bits (52 + hidden) go into the lower double */
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
lo = (lo64 >> 7) & (((uint64_t) 1 << 53) - 1);
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
/* The high order 53 bits (52 + hidden) go into the upper double */
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
hi = lo64 >> 60;
hi |= hi64 << 4;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
if (lo != 0)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
int lzcount;
/* hidden bit of low double controls rounding of the high double.
If hidden is '1' and either the explicit mantissa is non-zero
or hi is odd, then round up hi and adjust lo (2nd mantissa)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
plus change the sign of the low double to compensate. */
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
if ((lo & ((uint64_t) 1 << 52)) != 0
&& ((hi & 1) != 0 || (lo & (((uint64_t) 1 << 52) - 1)) != 0))
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
hi++;
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
if ((hi & ((uint64_t) 1 << 53)) != 0)
{
hi = hi >> 1;
u.d[0].ieee.exponent++;
}
u.d[1].ieee.negative = !sign;
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
lo = ((uint64_t) 1 << 53) - lo;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
}
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
/* Normalize the low double. Shift the mantissa left until
the hidden bit is '1' and adjust the exponent accordingly. */
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
if (sizeof (lo) == sizeof (long))
lzcount = __builtin_clzl (lo);
else if ((lo >> 32) != 0)
lzcount = __builtin_clzl ((long) (lo >> 32));
else
lzcount = __builtin_clzl ((long) lo) + 32;
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
lzcount = lzcount - (64 - 53);
lo <<= lzcount;
expnt2 -= lzcount;
if (expnt2 >= 1)
/* Not denormal. */
u.d[1].ieee.exponent = expnt2;
else
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
/* Is denormal. Note that biased exponent of 0 is treated
as if it was 1, hence the extra shift. */
if (expnt2 > -53)
lo >>= 1 - expnt2;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
else
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
lo = 0;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
}
}
else
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
u.d[1].ieee.negative = 0;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
u.d[1].ieee.mantissa1 = lo;
u.d[1].ieee.mantissa0 = lo >> 32;
u.d[0].ieee.mantissa1 = hi;
u.d[0].ieee.mantissa0 = hi >> 32;
return u.ld;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
}
2013-06-05 20:44:03 +00:00
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
/* Handy utility functions to pack/unpack/cononicalize and find the nearbyint
of long double implemented as double double. */
static inline long double
default_ldbl_pack (double a, double aa)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
union ibm_extended_long_double u;
u.d[0].d = a;
u.d[1].d = aa;
return u.ld;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
}
static inline void
default_ldbl_unpack (long double l, double *a, double *aa)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
{
union ibm_extended_long_double u;
u.ld = l;
*a = u.d[0].d;
*aa = u.d[1].d;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
}
#ifndef ldbl_pack
# define ldbl_pack default_ldbl_pack
#endif
#ifndef ldbl_unpack
# define ldbl_unpack default_ldbl_unpack
#endif
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
PowerPC floating point little-endian [2 of 15] http://sourceware.org/ml/libc-alpha/2013-08/msg00082.html This patch replaces occurrences of GET_LDOUBLE_* and SET_LDOUBLE_* macros, and union ieee854_long_double_shape_type in ldbl-128ibm/, and a stray one in the 32-bit fpu support. These files have no significant changes apart from rewriting the long double bit access. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_high): Define. * sysdeps/ieee754/ldbl-128ibm/e_acoshl.c (__ieee754_acoshl): Rewrite all uses of ieee854 long double macros and unions. * sysdeps/ieee754/ldbl-128ibm/e_acosl.c (__ieee754_acosl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_asinl.c (__ieee754_asinl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_atanhl.c (__ieee754_atanhl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_coshl.c (__ieee754_coshl): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_log2l.c (__ieee754_log2l): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Likewise. * sysdeps/ieee754/ldbl-128ibm/e_sinhl.c (__ieee754_sinhl): Likewise. * sysdeps/ieee754/ldbl-128ibm/k_cosl.c (__kernel_cosl): Likewise. * sysdeps/ieee754/ldbl-128ibm/k_sincosl.c (__kernel_sincosl): Likewise. * sysdeps/ieee754/ldbl-128ibm/k_sinl.c (__kernel_sinl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_asinhl.c (__asinhl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_atanl.c (__atanl): Likewise. Simplify sign and nan test too. * sysdeps/ieee754/ldbl-128ibm/s_cosl.c (__cosl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_fabsl.c (__fabsl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_finitel.c (___finitel): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_fpclassifyl.c (___fpclassifyl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_isnanl.c (___isnanl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_issignalingl.c (__issignalingl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_logbl.c (__logbl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_signbitl.c (___signbitl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_sincosl.c (__sincosl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_sinl.c (__sinl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_tanl.c (__tanl): Likewise. * sysdeps/powerpc/powerpc32/power7/fpu/s_logbl.c (__logbl): Likewise.
2013-08-17 08:54:05 +00:00
/* Extract high double. */
#define ldbl_high(x) ((double) x)
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
/* Convert a finite long double to canonical form.
Does not handle +/-Inf properly. */
static inline void
ldbl_canonicalize (double *a, double *aa)
{
double xh, xl;
xh = *a + *aa;
xl = (*a - xh) + *aa;
*a = xh;
*aa = xl;
}
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
/* Simple inline nearbyint (double) function.
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
Only works in the default rounding mode
but is useful in long double rounding functions. */
static inline double
ldbl_nearbyint (double a)
{
Fix for [BZ #15680] IBM long double inaccuracy http://sourceware.org/ml/libc-alpha/2013-06/msg00919.html I discovered a number of places where denormals and other corner cases were being handled wrongly. - printf_fphex.c: Testing for the low double exponent being zero is unnecessary. If the difference in exponents is less than 53 then the high double exponent must be nearing the low end of its range, and the low double exponent hit rock bottom. - ldbl2mpn.c: A denormal (ie. exponent of zero) value is treated as if the exponent was one, so shift mantissa left by one. Code handling normalisation of the low double mantissa lacked a test for shift count greater than bits in type being shifted, and lacked anything to handle the case where the difference in exponents is less than 53 as in printf_fphex.c. - math_ldbl.h (ldbl_extract_mantissa): Same as above, but worse, with code testing for exponent > 1 for some reason, probably a typo for >= 1. - math_ldbl.h (ldbl_insert_mantissa): Round the high double as per mpn2ldbl.c (hi is odd or explicit mantissas non-zero) so that the number we return won't change when applying ldbl_canonicalize(). Add missing overflow checks and normalisation of high mantissa. Correct misleading comment: "The hidden bit of the lo mantissa is zero" is not always true as can be seen from the code rounding the hi mantissa. Also by inspection, lzcount can never be less than zero so remove that test. Lastly, masking bitfields to their widths can be left to the compiler. - mpn2ldbl.c: The overflow checks here on rounding of high double were just plain wrong. Incrementing the exponent must be accompanied by a shift right of the mantissa to keep the value unchanged. Above notes for ldbl_insert_mantissa are also relevant. [BZ #15680] * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c: Comment fix. * sysdeps/ieee754/ldbl-128ibm/printf_fphex.c (PRINT_FPHEX_LONG_DOUBLE): Tidy code by moving -53 into ediff calculation. Remove unnecessary test for denormal exponent. * sysdeps/ieee754/ldbl-128ibm/ldbl2mpn.c (__mpn_extract_long_double): Correct handling of denormals. Avoid undefined shift behaviour. Correct normalisation of low mantissa when low double is denormal. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (ldbl_extract_mantissa): Likewise. Comment. Use uint64_t* for hi64. (ldbl_insert_mantissa): Make both hi64 and lo64 parms uint64_t. Correct normalisation of low mantissa. Test for overflow of high mantissa and normalise. (ldbl_nearbyint): Use more readable constant for two52. * sysdeps/ieee754/ldbl-128ibm/mpn2ldbl.c (__mpn_construct_long_double): Fix test for overflow of high mantissa and correct normalisation. Avoid undefined shift.
2013-08-17 08:49:44 +00:00
double two52 = 0x1p52;
[BZ #2423] 2006-03-07 Jakub Jelinek <jakub@redhat.com> [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Only run some of the new tests if LDBL_MANT_DIG > 100. 2006-03-03 Steven Munroe <sjmunroe@us.ibm.com> Alan Modra <amodra@bigpond.net.au> * sysdeps/powerpc/fpu/fenv_libc.h (__fegetround, __fesetround): Define inline implementations. * sysdeps/powerpc/fpu/fegetround.c: Use __fegetround. * sysdeps/powerpc/fpu/fesetround.c: Use __fesetround. * sysdeps/powerpc/fpu/math_ldbl.h: New file. [BZ #2423] * math/libm-test.inc [TEST_LDOUBLE] (ceil_test, floor_test, rint_test, round_test, trunc_test): Add new tests. * sysdeps/ieee754/ldbl-128ibm/math_ldbl.h (EXTRACT_IBM_EXTENDED_MANTISSA, INSERT_IBM_EXTENDED_MANTISSA): Removed, replaced with ... (ldbl_extract_mantissa, ldbl_insert_mantissa, ldbl_pack, ldbl_unpack, ldbl_canonicalise, ldbl_nearbyint): New functions. * sysdeps/ieee754/ldbl-128ibm/e_fmodl.c (__ieee754_fmodl): Replace EXTRACT_IBM_EXTENDED_MANTISSA and INSERT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa and ldbl_insert_mantissa. * sysdeps/ieee754/ldbl-128ibm/e_rem_pio2l.c (__ieee754_rem_pio2l): Replace EXTRACT_IBM_EXTENDED_MANTISSA with ldbl_extract_mantissa. (ldbl_extract_mantissa, ldbl_insert_mantissa): New inline functions. * sysdeps/ieee754/ldbl-128ibm/s_ceill.c (__ceill): Handle rounding that spans doubles in IBM long double format. * sysdeps/ieee754/ldbl-128ibm/s_floorl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_rintl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_roundl.c: Likewise. * sysdeps/ieee754/ldbl-128ibm/s_truncl.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_rintl.S: File removed.
2006-03-16 11:47:24 +00:00
if (__builtin_expect ((__builtin_fabs (a) < two52), 1))
{
if (__builtin_expect ((a > 0.0), 1))
{
a += two52;
a -= two52;
}
else if (__builtin_expect ((a < 0.0), 1))
{
a = two52 - a;
a = -(a - two52);
}
}
return a;
}