diff options
author | zimmerma <zimmerma@280ebfd0-de03-0410-8827-d642c229c3f4> | 2007-05-29 12:24:20 +0000 |
---|---|---|
committer | zimmerma <zimmerma@280ebfd0-de03-0410-8827-d642c229c3f4> | 2007-05-29 12:24:20 +0000 |
commit | 10fa8f5c223a066a1bbe4c290777aa782384f836 (patch) | |
tree | 6ef4019ac93c72f1c2e43479855047402f54fc70 /csch.c | |
parent | 7b8b57034998cd6e2fdd9ce3012d8fd923b3e342 (diff) | |
download | mpfr-10fa8f5c223a066a1bbe4c290777aa782384f836.tar.gz |
fixed tiny input problem with csch, j0, j1
git-svn-id: svn://scm.gforge.inria.fr/svn/mpfr/trunk@4502 280ebfd0-de03-0410-8827-d642c229c3f4
Diffstat (limited to 'csch.c')
-rw-r--r-- | csch.c | 37 |
1 files changed, 37 insertions, 0 deletions
@@ -36,4 +36,41 @@ MA 02110-1301, USA. */ #define ACTION_ZERO(y,x) do { MPFR_SET_SAME_SIGN(y,x); MPFR_SET_INF(y); \ MPFR_RET(0); } while (1) +/* (This analysis is adapted from that for mpfr_csc.) + Near x=0, we have csch(x) = 1/x - x/6 + ..., more precisely we have + |csch(x) - 1/x| <= 0.2 for |x| <= 1. The error term has the opposite + sign as 1/x, thus |csch(x)| <= |1/x|. Then: + (i) either x is a power of two, then 1/x is exactly representable, and + as long as 1/2*ulp(1/x) > 0.2, we can conclude; + (ii) otherwise assume x has <= n bits, and y has <= n+1 bits, then + |y - 1/x| >= 2^(-2n) ufp(y), where ufp means unit in first place. + Since |csch(x) - 1/x| <= 0.2, if 2^(-2n) ufp(y) >= 0.4, then + |y - csch(x)| >= 2^(-2n-1) ufp(y), and rounding 1/x gives the correct + result. If x < 2^E, then y > 2^(-E), thus ufp(y) > 2^(-E-1). + A sufficient condition is thus EXP(x) <= -2 MAX(PREC(x),PREC(Y)). */ +#define ACTION_TINY(y,x,r) \ + if (MPFR_EXP(x) <= -2 * (mp_exp_t) MAX(MPFR_PREC(x), MPFR_PREC(y))) \ + { \ + int signx = MPFR_SIGN(x); \ + inexact = mpfr_ui_div (y, 1, x, r); \ + if (inexact == 0) /* x is a power of two */ \ + { /* result always 1/x, except when rounding to zero */ \ + if (rnd_mode == GMP_RNDU || (rnd_mode == GMP_RNDZ && signx < 0)) \ + { \ + if (signx < 0) \ + mpfr_nextabove (y); /* -2^k + epsilon */ \ + inexact = 1; \ + } \ + else if (rnd_mode == GMP_RNDD || rnd_mode == GMP_RNDZ) \ + { \ + if (signx > 0) \ + mpfr_nextbelow (y); /* 2^k - epsilon */ \ + inexact = -1; \ + } \ + else /* round to nearest */ \ + inexact = signx; \ + } \ + goto end; \ + } + #include "gen_inverse.h" |