diff options
author | Szabolcs Nagy <nsz@port70.net> | 2013-01-07 23:54:13 +0100 |
---|---|---|
committer | Szabolcs Nagy <nsz@port70.net> | 2013-01-07 23:54:13 +0100 |
commit | 121e3a38a1d0b324696fc1671029e4b815ea7fa4 (patch) | |
tree | 608923bcb097799f051cb509af138d8a49239cc5 /src/math/erff.c | |
parent | d84923d89e4fe46877334cbaa5e4549f2c4cb8a6 (diff) | |
download | musl-121e3a38a1d0b324696fc1671029e4b815ea7fa4.tar.gz |
math: erf and erfc cleanup
common part of erf and erfc was put in a separate function which
saved some space and the new code is using unsigned arithmetics
erfcf had a bug: for some inputs in [7.95,8] the result had
more than 60ulp error: in expf(-z*z - 0.5625f) the argument
must be exact but not enough lowbits of z were zeroed,
-SET_FLOAT_WORD(z, ix&0xfffff000);
+SET_FLOAT_WORD(z, ix&0xffffe000);
fixed the issue
Diffstat (limited to 'src/math/erff.c')
-rw-r--r-- | src/math/erff.c | 154 |
1 files changed, 61 insertions, 93 deletions
diff --git a/src/math/erff.c b/src/math/erff.c index a0a304e5..e2cfc984 100644 --- a/src/math/erff.c +++ b/src/math/erff.c @@ -16,13 +16,10 @@ #include "libm.h" static const float -tiny = 1e-30, -/* c = (subfloat)0.84506291151 */ erx = 8.4506291151e-01, /* 0x3f58560b */ /* * Coefficients for approximation to erf on [0,0.84375] */ -efx = 1.2837916613e-01, /* 0x3e0375d4 */ efx8 = 1.0270333290e+00, /* 0x3f8375d4 */ pp0 = 1.2837916613e-01, /* 0x3e0375d4 */ pp1 = -3.2504209876e-01, /* 0xbea66beb */ @@ -87,47 +84,26 @@ sb5 = 2.5530502930e+03, /* 0x451f90ce */ sb6 = 4.7452853394e+02, /* 0x43ed43a7 */ sb7 = -2.2440952301e+01; /* 0xc1b38712 */ -float erff(float x) +static float erfc1(float x) { - int32_t hx,ix,i; - float R,S,P,Q,s,y,z,r; + float s,P,Q; + + s = fabsf(x) - 1; + P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); + Q = 1+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); + return 1 - erx - P/Q; +} + +static float erfc2(uint32_t ix, float x) +{ + float s,z,R,S; + + if (ix < 0x3fa00000) /* |x| < 1.25 */ + return erfc1(x); - GET_FLOAT_WORD(hx, x); - ix = hx & 0x7fffffff; - if (ix >= 0x7f800000) { - /* erf(nan)=nan, erf(+-inf)=+-1 */ - i = ((uint32_t)hx>>31)<<1; - return (float)(1-i)+1.0f/x; - } - if (ix < 0x3f580000) { /* |x| < 0.84375 */ - if (ix < 0x31800000) { /* |x| < 2**-28 */ - if (ix < 0x04000000) - /*avoid underflow */ - return 0.125f*(8.0f*x + efx8*x); - return x + efx*x; - } - z = x*x; - r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); - s = 1.0f+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); - y = r/s; - return x + x*y; - } - if (ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */ - s = fabsf(x)-1.0f; - P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); - Q = 1.0f+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); - if (hx >= 0) - return erx + P/Q; - return -erx - P/Q; - } - if (ix >= 0x40c00000) { /* inf > |x| >= 6 */ - if (hx >= 0) - return 1.0f - tiny; - return tiny - 1.0f; - } x = fabsf(x); - s = 1.0f/(x*x); - if (ix < 0x4036DB6E) { /* |x| < 1/0.35 */ + s = 1/(x*x); + if (ix < 0x4036db6d) { /* |x| < 1/0.35 */ R = ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*( ra5+s*(ra6+s*ra7)))))); S = 1.0f+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*( @@ -139,23 +115,53 @@ float erff(float x) sb5+s*(sb6+s*sb7)))))); } GET_FLOAT_WORD(ix, x); - SET_FLOAT_WORD(z, ix&0xfffff000); - r = expf(-z*z - 0.5625f) * expf((z-x)*(z+x) + R/S); - if (hx >= 0) - return 1.0f - r/x; - return r/x - 1.0f; + SET_FLOAT_WORD(z, ix&0xffffe000); + return expf(-z*z - 0.5625f) * expf((z-x)*(z+x) + R/S)/x; +} + +float erff(float x) +{ + float r,s,z,y; + uint32_t ix; + int sign; + + GET_FLOAT_WORD(ix, x); + sign = ix>>31; + ix &= 0x7fffffff; + if (ix >= 0x7f800000) { + /* erf(nan)=nan, erf(+-inf)=+-1 */ + return 1-2*sign + 1/x; + } + if (ix < 0x3f580000) { /* |x| < 0.84375 */ + if (ix < 0x31800000) { /* |x| < 2**-28 */ + /*avoid underflow */ + return 0.125f*(8*x + efx8*x); + } + z = x*x; + r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); + s = 1+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); + y = r/s; + return x + x*y; + } + if (ix < 0x40c00000) /* |x| < 6 */ + y = 1 - erfc2(ix,x); + else + y = 1 - 0x1p-120f; + return sign ? -y : y; } float erfcf(float x) { - int32_t hx,ix; - float R,S,P,Q,s,y,z,r; + float r,s,z,y; + uint32_t ix; + int sign; - GET_FLOAT_WORD(hx, x); - ix = hx & 0x7fffffff; + GET_FLOAT_WORD(ix, x); + sign = ix>>31; + ix &= 0x7fffffff; if (ix >= 0x7f800000) { /* erfc(nan)=nan, erfc(+-inf)=0,2 */ - return (float)(((uint32_t)hx>>31)<<1) + 1.0f/x; + return 2*sign + 1/x; } if (ix < 0x3f580000) { /* |x| < 0.84375 */ @@ -165,50 +171,12 @@ float erfcf(float x) r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); s = 1.0f+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); y = r/s; - if (hx < 0x3e800000) { /* x<1/4 */ + if (sign || ix < 0x3e800000) /* x < 1/4 */ return 1.0f - (x+x*y); - } else { - r = x*y; - r += (x-0.5f); - return 0.5f - r ; - } - } - if (ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */ - s = fabsf(x)-1.0f; - P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); - Q = 1.0f+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); - if(hx >= 0) { - z = 1.0f - erx; - return z - P/Q; - } else { - z = erx + P/Q; - return 1.0f + z; - } + return 0.5f - (x - 0.5f + x*y); } if (ix < 0x41e00000) { /* |x| < 28 */ - x = fabsf(x); - s = 1.0f/(x*x); - if (ix < 0x4036DB6D) { /* |x| < 1/.35 ~ 2.857143*/ - R = ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*( - ra5+s*(ra6+s*ra7)))))); - S = 1.0f+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*( - sa5+s*(sa6+s*(sa7+s*sa8))))))); - } else { /* |x| >= 1/.35 ~ 2.857143 */ - if (hx < 0 && ix >= 0x40c00000) /* x < -6 */ - return 2.0f-tiny; - R = rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*( - rb5+s*rb6))))); - S = 1.0f+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*( - sb5+s*(sb6+s*sb7)))))); - } - GET_FLOAT_WORD(ix, x); - SET_FLOAT_WORD(z, ix&0xfffff000); - r = expf(-z*z - 0.5625f) * expf((z-x)*(z+x) + R/S); - if (hx > 0) - return r/x; - return 2.0f - r/x; + return sign ? 2 - erfc2(ix,x) : erfc2(ix,x); } - if (hx > 0) - return tiny*tiny; - return 2.0f - tiny; + return sign ? 2 - 0x1p-120f : 0x1p-120f*0x1p-120f; } |