Skip to content

W
wine-winehq
Commit bfc3a2bd authored by Alexandre Julliard's avatar Alexandre Julliard
Browse files
Options

msvcrt: Use the erf()/erfc()/erff()/erfcf() implementation from the bundled musl library.

parent 3b1e2a7b
Hide whitespace changes
Inline Side-by-side
Showing with 8 additions and 371 deletions
dlls/msvcrt/math.c
View file @ bfc3a2bd
......@@ -77,15 +77,6 @@ void msvcrt_init_math( void *module )
#endif
}
/* Copied from musl: src/internal/libm.h */
#if _MSVCR_VER >= 120
static inline double fp_barrier(double x)
{
volatile double y = x;
return y;
}
#endif
static inline double ret_nan( BOOL update_sw )
{
double x = 1.0;
......@@ -2874,365 +2865,6 @@ short CDECL _fdtest(float *x)
return _fdclass(*x);
}
static double erfc1(double x)
{
static const double erx = 8.45062911510467529297e-01,
pa0 = -2.36211856075265944077e-03,
pa1 = 4.14856118683748331666e-01,
pa2 = -3.72207876035701323847e-01,
pa3 = 3.18346619901161753674e-01,
pa4 = -1.10894694282396677476e-01,
pa5 = 3.54783043256182359371e-02,
pa6 = -2.16637559486879084300e-03,
qa1 = 1.06420880400844228286e-01,
qa2 = 5.40397917702171048937e-01,
qa3 = 7.18286544141962662868e-02,
qa4 = 1.26171219808761642112e-01,
qa5 = 1.36370839120290507362e-02,
qa6 = 1.19844998467991074170e-02;
double s, P, Q;
s = fabs(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 double erfc2(UINT32 ix, double x)
{
static const double ra0 = -9.86494403484714822705e-03,
ra1 = -6.93858572707181764372e-01,
ra2 = -1.05586262253232909814e+01,
ra3 = -6.23753324503260060396e+01,
ra4 = -1.62396669462573470355e+02,
ra5 = -1.84605092906711035994e+02,
ra6 = -8.12874355063065934246e+01,
ra7 = -9.81432934416914548592e+00,
sa1 = 1.96512716674392571292e+01,
sa2 = 1.37657754143519042600e+02,
sa3 = 4.34565877475229228821e+02,
sa4 = 6.45387271733267880336e+02,
sa5 = 4.29008140027567833386e+02,
sa6 = 1.08635005541779435134e+02,
sa7 = 6.57024977031928170135e+00,
sa8 = -6.04244152148580987438e-02,
rb0 = -9.86494292470009928597e-03,
rb1 = -7.99283237680523006574e-01,
rb2 = -1.77579549177547519889e+01,
rb3 = -1.60636384855821916062e+02,
rb4 = -6.37566443368389627722e+02,
rb5 = -1.02509513161107724954e+03,
rb6 = -4.83519191608651397019e+02,
sb1 = 3.03380607434824582924e+01,
sb2 = 3.25792512996573918826e+02,
sb3 = 1.53672958608443695994e+03,
sb4 = 3.19985821950859553908e+03,
sb5 = 2.55305040643316442583e+03,
sb6 = 4.74528541206955367215e+02,
sb7 = -2.24409524465858183362e+01;
double s, R, S, z;
UINT64 iz;
if (ix < 0x3ff40000) /* |x| < 1.25 */
return erfc1(x);
x = fabs(x);
s = 1 / (x * x);
if (ix < 0x4006db6d) { /* |x| < 1/.35 ~ 2.85714 */
R = ra0 + s * (ra1 + s * (ra2 + s * (ra3 + s * (ra4 + s *
(ra5 + s * (ra6 + s * ra7))))));
S = 1.0 + s * (sa1 + s * (sa2 + s * (sa3 + s * (sa4 + s *
(sa5 + s * (sa6 + s * (sa7 + s * sa8)))))));
} else { /* |x| > 1/.35 */
R = rb0 + s * (rb1 + s * (rb2 + s * (rb3 + s * (rb4 + s *
(rb5 + s * rb6)))));
S = 1.0 + s * (sb1 + s * (sb2 + s * (sb3 + s * (sb4 + s *
(sb5 + s * (sb6 + s * sb7))))));
}
z = x;
iz = *(ULONGLONG*)&z;
iz &= 0xffffffff00000000ULL;
z = *(double*)&iz;
return exp(-z * z - 0.5625) * exp((z - x) * (z + x) + R / S) / x;
}
/*********************************************************************
* erf (MSVCR120.@)
*/
double CDECL erf(double x)
{
static const double efx8 = 1.02703333676410069053e+00,
pp0 = 1.28379167095512558561e-01,
pp1 = -3.25042107247001499370e-01,
pp2 = -2.84817495755985104766e-02,
pp3 = -5.77027029648944159157e-03,
pp4 = -2.37630166566501626084e-05,
qq1 = 3.97917223959155352819e-01,
qq2 = 6.50222499887672944485e-02,
qq3 = 5.08130628187576562776e-03,
qq4 = 1.32494738004321644526e-04,
qq5 = -3.96022827877536812320e-06;
double r, s, z, y;
UINT32 ix;
int sign;
ix = *(UINT64*)&x >> 32;
sign = ix >> 31;
ix &= 0x7fffffff;
if (ix >= 0x7ff00000) {
/* erf(nan)=nan, erf(+-inf)=+-1 */
return 1 - 2 * sign + 1 / x;
}
if (ix < 0x3feb0000) { /* |x| < 0.84375 */
if (ix < 0x3e300000) { /* |x| < 2**-28 */
/* avoid underflow */
return 0.125 * (8 * x + efx8 * x);
}
z = x * x;
r = pp0 + z * (pp1 + z * (pp2 + z * (pp3 + z * pp4)));
s = 1.0 + z * (qq1 + z * (qq2 + z * (qq3 + z * (qq4 + z * qq5))));
y = r / s;
return x + x * y;
}
if (ix < 0x40180000) /* 0.84375 <= |x| < 6 */
y = 1 - erfc2(ix, x);
else
y = 1 - DBL_MIN;
return sign ? -y : y;
}
static float erfc1f(float x)
{
static const float erx = 8.4506291151e-01,
pa0 = -2.3621185683e-03,
pa1 = 4.1485610604e-01,
pa2 = -3.7220788002e-01,
pa3 = 3.1834661961e-01,
pa4 = -1.1089469492e-01,
pa5 = 3.5478305072e-02,
pa6 = -2.1663755178e-03,
qa1 = 1.0642088205e-01,
qa2 = 5.4039794207e-01,
qa3 = 7.1828655899e-02,
qa4 = 1.2617121637e-01,
qa5 = 1.3637083583e-02,
qa6 = 1.1984500103e-02;
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 erfc2f(UINT32 ix, float x)
{
static const float ra0 = -9.8649440333e-03,
ra1 = -6.9385856390e-01,
ra2 = -1.0558626175e+01,
ra3 = -6.2375331879e+01,
ra4 = -1.6239666748e+02,
ra5 = -1.8460508728e+02,
ra6 = -8.1287437439e+01,
ra7 = -9.8143291473e+00,
sa1 = 1.9651271820e+01,
sa2 = 1.3765776062e+02,
sa3 = 4.3456588745e+02,
sa4 = 6.4538726807e+02,
sa5 = 4.2900814819e+02,
sa6 = 1.0863500214e+02,
sa7 = 6.5702495575e+00,
sa8 = -6.0424413532e-02,
rb0 = -9.8649431020e-03,
rb1 = -7.9928326607e-01,
rb2 = -1.7757955551e+01,
rb3 = -1.6063638306e+02,
rb4 = -6.3756646729e+02,
rb5 = -1.0250950928e+03,
rb6 = -4.8351919556e+02,
sb1 = 3.0338060379e+01,
sb2 = 3.2579251099e+02,
sb3 = 1.5367296143e+03,
sb4 = 3.1998581543e+03,
sb5 = 2.5530502930e+03,
sb6 = 4.7452853394e+02,
sb7 = -2.2440952301e+01;
float s, R, S, z;
if (ix < 0x3fa00000) /* |x| < 1.25 */
return erfc1f(x);
x = fabsf(x);
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 *
(sa5 + s * (sa6 + s * (sa7 + s * sa8)))))));
} else { /* |x| >= 1/0.35 */
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))))));
}
ix = *(UINT32*)&x & 0xffffe000;
z = *(float*)&ix;
return expf(-z * z - 0.5625f) * expf((z - x) * (z + x) + R / S) / x;
}
/*********************************************************************
* erff (MSVCR120.@)
*
* Copied from musl: src/math/erff.c
*/
float CDECL erff(float x)
{
static const float efx8 = 1.0270333290e+00,
pp0 = 1.2837916613e-01,
pp1 = -3.2504209876e-01,
pp2 = -2.8481749818e-02,
pp3 = -5.7702702470e-03,
pp4 = -2.3763017452e-05,
qq1 = 3.9791721106e-01,
qq2 = 6.5022252500e-02,
qq3 = 5.0813062117e-03,
qq4 = 1.3249473704e-04,
qq5 = -3.9602282413e-06;
float r, s, z, y;
UINT32 ix;
int sign;
ix = *(UINT32*)&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 - erfc2f(ix, x);
else
y = 1 - FLT_MIN;
return sign ? -y : y;
}
/*********************************************************************
* erfc (MSVCR120.@)
*
* Copied from musl: src/math/erf.c
*/
double CDECL erfc(double x)
{
static const double pp0 = 1.28379167095512558561e-01,
pp1 = -3.25042107247001499370e-01,
pp2 = -2.84817495755985104766e-02,
pp3 = -5.77027029648944159157e-03,
pp4 = -2.37630166566501626084e-05,
qq1 = 3.97917223959155352819e-01,
qq2 = 6.50222499887672944485e-02,
qq3 = 5.08130628187576562776e-03,
qq4 = 1.32494738004321644526e-04,
qq5 = -3.96022827877536812320e-06;
double r, s, z, y;
UINT32 ix;
int sign;
ix = *(ULONGLONG*)&x >> 32;
sign = ix >> 31;
ix &= 0x7fffffff;
if (ix >= 0x7ff00000) {
/* erfc(nan)=nan, erfc(+-inf)=0,2 */
return 2 * sign + 1 / x;
}
if (ix < 0x3feb0000) { /* |x| < 0.84375 */
if (ix < 0x3c700000) /* |x| < 2**-56 */
return 1.0 - x;
z = x * x;
r = pp0 + z * (pp1 + z * (pp2 + z * (pp3 + z * pp4)));
s = 1.0 + z * (qq1 + z * (qq2 + z * (qq3 + z * (qq4 + z * qq5))));
y = r / s;
if (sign || ix < 0x3fd00000) { /* x < 1/4 */
return 1.0 - (x + x * y);
}
return 0.5 - (x - 0.5 + x * y);
}
if (ix < 0x403c0000) { /* 0.84375 <= |x| < 28 */
return sign ? 2 - erfc2(ix, x) : erfc2(ix, x);
}
if (sign)
return 2 - DBL_MIN;
*_errno() = ERANGE;
return fp_barrier(DBL_MIN) * DBL_MIN;
}
/*********************************************************************
* erfcf (MSVCR120.@)
*
* Copied from musl: src/math/erff.c
*/
float CDECL erfcf(float x)
{
static const float pp0 = 1.2837916613e-01,
pp1 = -3.2504209876e-01,
pp2 = -2.8481749818e-02,
pp3 = -5.7702702470e-03,
pp4 = -2.3763017452e-05,
qq1 = 3.9791721106e-01,
qq2 = 6.5022252500e-02,
qq3 = 5.0813062117e-03,
qq4 = 1.3249473704e-04,
qq5 = -3.9602282413e-06;
float r, s, z, y;
UINT32 ix;
int sign;
ix = *(UINT32*)&x;
sign = ix >> 31;
ix &= 0x7fffffff;
if (ix >= 0x7f800000) {
/* erfc(nan)=nan, erfc(+-inf)=0,2 */
return 2 * sign + 1 / x;
}
if (ix < 0x3f580000) { /* |x| < 0.84375 */
if (ix < 0x23800000) /* |x| < 2**-56 */
return 1.0f - 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;
if (sign || ix < 0x3e800000) /* x < 1/4 */
return 1.0f - (x + x * y);
return 0.5f - (x - 0.5f + x * y);
}
if (ix < 0x41e00000) { /* |x| < 28 */
return sign ? 2 - erfc2f(ix, x) : erfc2f(ix, x);
}
if (sign)
return 2 - FLT_MIN;
*_errno() = ERANGE;
return FLT_MIN * FLT_MIN;
}
/*********************************************************************
* _fdsign (MSVCR120.@)
*/
......@@ -3251,7 +2883,6 @@ int CDECL _dsign(double x)
return (u.i >> 48) & 0x8000;
}
/*********************************************************************
* _dpcomp (MSVCR120.@)
*/
......
libs/musl/src/math/erf.c
View file @ bfc3a2bd
......@@ -269,5 +269,8 @@ double __cdecl erfc(double x)
if (ix < 0x403c0000) { /* 0.84375 <= |x| < 28 */
return sign ? 2 - erfc2(ix,x) : erfc2(ix,x);
}
return sign ? 2 - 0x1p-1022 : 0x1p-1022*0x1p-1022;
if (sign)
return 2 - DBL_MIN;
errno = ERANGE;
return fp_barrier(DBL_MIN) * DBL_MIN;
}
libs/musl/src/math/erff.c
View file @ bfc3a2bd
......@@ -179,5 +179,8 @@ float __cdecl erfcf(float x)
if (ix < 0x41e00000) { /* |x| < 28 */
return sign ? 2 - erfc2(ix,x) : erfc2(ix,x);
}
return sign ? 2 - 0x1p-120f : 0x1p-120f*0x1p-120f;
if (sign)
return 2 - FLT_MIN;
errno = ERANGE;
return FLT_MIN * FLT_MIN;
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment