May 2021 - Wine-commits - list.winehq.org

Piotr Caban : msvcrt: Import cosh implementation from musl.
by Alexandre Julliard 31 May '21

31 May '21

Module: wine Branch: master Commit: b431d7515da309ff85e3677fe98fce2218c890ab URL: https://source.winehq.org/git/wine.git/?a=commit;h=b431d7515da309ff85e3677f… Author: Piotr Caban <piotr(a)codeweavers.com> Date: Sat May 29 19:13:39 2021 +0200 msvcrt: Import cosh implementation from musl. Signed-off-by: Piotr Caban <piotr(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/msvcrt/math.c | 231 ++++++++++++++++++++++++++++++-------------------- dlls/msvcrt/unixlib.c | 9 -- dlls/msvcrt/unixlib.h | 1 - 3 files changed, 137 insertions(+), 104 deletions(-) diff --git a/dlls/msvcrt/math.c b/dlls/msvcrt/math.c index 06f1dc26838..760f3834761 100644 --- a/dlls/msvcrt/math.c +++ b/dlls/msvcrt/math.c @@ -2095,14 +2095,147 @@ double CDECL cos( double x ) } } +/* Copied from musl: src/math/expm1.c */ +static double CDECL __expm1(double x) +{ + static const double o_threshold = 7.09782712893383973096e+02, + ln2_hi = 6.93147180369123816490e-01, + ln2_lo = 1.90821492927058770002e-10, + invln2 = 1.44269504088896338700e+00, + Q1 = -3.33333333333331316428e-02, + Q2 = 1.58730158725481460165e-03, + Q3 = -7.93650757867487942473e-05, + Q4 = 4.00821782732936239552e-06, + Q5 = -2.01099218183624371326e-07; + + double y, hi, lo, c, t, e, hxs, hfx, r1, twopk; + union {double f; UINT64 i;} u = {x}; + UINT32 hx = u.i >> 32 & 0x7fffffff; + int k, sign = u.i >> 63; + + /* filter out huge and non-finite argument */ + if (hx >= 0x4043687A) { /* if |x|>=56*ln2 */ + if (isnan(x)) + return x; + if (isinf(x)) + return sign ? -1 : x; + if (sign) + return math_error(_UNDERFLOW, "exp", x, 0, -1); + if (x > o_threshold) + return math_error(_OVERFLOW, "exp", x, 0, x * 0x1p1023); + } + + /* argument reduction */ + if (hx > 0x3fd62e42) { /* if |x| > 0.5 ln2 */ + if (hx < 0x3FF0A2B2) { /* and |x| < 1.5 ln2 */ + if (!sign) { + hi = x - ln2_hi; + lo = ln2_lo; + k = 1; + } else { + hi = x + ln2_hi; + lo = -ln2_lo; + k = -1; + } + } else { + k = invln2 * x + (sign ? -0.5 : 0.5); + t = k; + hi = x - t * ln2_hi; /* t*ln2_hi is exact here */ + lo = t * ln2_lo; + } + x = hi - lo; + c = (hi - x) - lo; + } else if (hx < 0x3c900000) { /* |x| < 2**-54, return x */ + fp_barrier(x + 0x1p120f); + if (hx < 0x00100000) + fp_barrier((float)x); + return x; + } else + k = 0; + + /* x is now in primary range */ + hfx = 0.5 * x; + hxs = x * hfx; + r1 = 1.0 + hxs * (Q1 + hxs * (Q2 + hxs * (Q3 + hxs * (Q4 + hxs * Q5)))); + t = 3.0 - r1 * hfx; + e = hxs * ((r1 - t) / (6.0 - x * t)); + if (k == 0) /* c is 0 */ + return x - (x * e - hxs); + e = x * (e - c) - c; + e -= hxs; + /* exp(x) ~ 2^k (x_reduced - e + 1) */ + if (k == -1) + return 0.5 * (x - e) - 0.5; + if (k == 1) { + if (x < -0.25) + return -2.0 * (e - (x + 0.5)); + return 1.0 + 2.0 * (x - e); + } + u.i = (UINT64)(0x3ff + k) << 52; /* 2^k */ + twopk = u.f; + if (k < 0 || k > 56) { /* suffice to return exp(x)-1 */ + y = x - e + 1.0; + if (k == 1024) + y = y * 2.0 * 0x1p1023; + else + y = y * twopk; + return y - 1.0; + } + u.i = (UINT64)(0x3ff - k) << 52; /* 2^-k */ + if (k < 20) + y = (x - e + (1 - u.f)) * twopk; + else + y = (x - (e + u.f) + 1) * twopk; + return y; +} + +static double __expo2(double x, double sign) +{ + static const int k = 2043; + static const double kln2 = 0x1.62066151add8bp+10; + double scale; + + *(UINT64*)&scale = (UINT64)(0x3ff + k / 2) << 52; + return exp(x - kln2) * (sign * scale) * scale; +} + /********************************************************************* * cosh (MSVCRT.@) + * + * Copied from musl: src/math/cosh.c */ double CDECL cosh( double x ) { - double ret = unix_funcs->cosh( x ); - if (isnan(x)) return math_error(_DOMAIN, "cosh", x, 0, ret); - return ret; + UINT64 ux = *(UINT64*)&x; + UINT32 w; + double t; + + /* |x| */ + ux &= (uint64_t)-1 / 2; + x = *(double*)&ux; + w = ux >> 32; + + /* |x| < log(2) */ + if (w < 0x3fe62e42) { + if (w < 0x3ff00000 - (26 << 20)) { + fp_barrier(x + 0x1p120f); + return 1; + } + t = __expm1(x); + return 1 + t * t / (2 * (1 + t)); + } + + /* |x| < log(DBL_MAX) */ + if (w < 0x40862e42) { + t = exp(x); + /* note: if x>log(0x1p26) then the 1/t is not needed */ + return 0.5 * (t + 1 / t); + } + + /* |x| > log(DBL_MAX) or nan */ + /* note: the result is stored to handle overflow */ + t = __expo2(x, 1.0); + return t; } /********************************************************************* @@ -5758,100 +5891,10 @@ float CDECL exp2f(float x) /********************************************************************* * expm1 (MSVCR120.@) - * - * Copied from musl: src/math/expm1.c */ double CDECL expm1(double x) { - static const double o_threshold = 7.09782712893383973096e+02, - ln2_hi = 6.93147180369123816490e-01, - ln2_lo = 1.90821492927058770002e-10, - invln2 = 1.44269504088896338700e+00, - Q1 = -3.33333333333331316428e-02, - Q2 = 1.58730158725481460165e-03, - Q3 = -7.93650757867487942473e-05, - Q4 = 4.00821782732936239552e-06, - Q5 = -2.01099218183624371326e-07; - - double y, hi, lo, c, t, e, hxs, hfx, r1, twopk; - union {double f; UINT64 i;} u = {x}; - UINT32 hx = u.i >> 32 & 0x7fffffff; - int k, sign = u.i >> 63; - - /* filter out huge and non-finite argument */ - if (hx >= 0x4043687A) { /* if |x|>=56*ln2 */ - if (isnan(x)) - return x; - if (isinf(x)) - return sign ? -1 : x; - if (sign) - return math_error(_UNDERFLOW, "exp", x, 0, -1); - if (x > o_threshold) - return math_error(_OVERFLOW, "exp", x, 0, x * 0x1p1023); - } - - /* argument reduction */ - if (hx > 0x3fd62e42) { /* if |x| > 0.5 ln2 */ - if (hx < 0x3FF0A2B2) { /* and |x| < 1.5 ln2 */ - if (!sign) { - hi = x - ln2_hi; - lo = ln2_lo; - k = 1; - } else { - hi = x + ln2_hi; - lo = -ln2_lo; - k = -1; - } - } else { - k = invln2 * x + (sign ? -0.5 : 0.5); - t = k; - hi = x - t * ln2_hi; /* t*ln2_hi is exact here */ - lo = t * ln2_lo; - } - x = hi - lo; - c = (hi - x) - lo; - } else if (hx < 0x3c900000) { /* |x| < 2**-54, return x */ - fp_barrier(x + 0x1p120f); - if (hx < 0x00100000) - fp_barrier((float)x); - return x; - } else - k = 0; - - /* x is now in primary range */ - hfx = 0.5 * x; - hxs = x * hfx; - r1 = 1.0 + hxs * (Q1 + hxs * (Q2 + hxs * (Q3 + hxs * (Q4 + hxs * Q5)))); - t = 3.0 - r1 * hfx; - e = hxs * ((r1 - t) / (6.0 - x * t)); - if (k == 0) /* c is 0 */ - return x - (x * e - hxs); - e = x * (e - c) - c; - e -= hxs; - /* exp(x) ~ 2^k (x_reduced - e + 1) */ - if (k == -1) - return 0.5 * (x - e) - 0.5; - if (k == 1) { - if (x < -0.25) - return -2.0 * (e - (x + 0.5)); - return 1.0 + 2.0 * (x - e); - } - u.i = (UINT64)(0x3ff + k) << 52; /* 2^k */ - twopk = u.f; - if (k < 0 || k > 56) { /* suffice to return exp(x)-1 */ - y = x - e + 1.0; - if (k == 1024) - y = y * 2.0 * 0x1p1023; - else - y = y * twopk; - return y - 1.0; - } - u.i = (UINT64)(0x3ff - k) << 52; /* 2^-k */ - if (k < 20) - y = (x - e + (1 - u.f)) * twopk; - else - y = (x - (e + u.f) + 1) * twopk; - return y; + return __expm1(x); } /********************************************************************* diff --git a/dlls/msvcrt/unixlib.c b/dlls/msvcrt/unixlib.c index c0c3a2b1636..6af252e5e9b 100644 --- a/dlls/msvcrt/unixlib.c +++ b/dlls/msvcrt/unixlib.c @@ -42,14 +42,6 @@ WINE_DEFAULT_DEBUG_CHANNEL(msvcrt); -/********************************************************************* - * cosh - */ -static double CDECL unix_cosh( double x ) -{ - return cosh( x ); -} - /********************************************************************* * coshf */ @@ -344,7 +336,6 @@ static float CDECL unix_tgammaf(float x) static const struct unix_funcs funcs = { - unix_cosh, unix_coshf, unix_exp, unix_expf, diff --git a/dlls/msvcrt/unixlib.h b/dlls/msvcrt/unixlib.h index 6f8347b066a..67dd1f98b84 100644 --- a/dlls/msvcrt/unixlib.h +++ b/dlls/msvcrt/unixlib.h @@ -23,7 +23,6 @@ struct unix_funcs { - double (CDECL *cosh)(double x); float (CDECL *coshf)(float x); double (CDECL *exp)(double x); float (CDECL *expf)(float x);

1 0

Piotr Caban : msvcrt: Import coshf implementation from musl.
by Alexandre Julliard 31 May '21

31 May '21

Module: wine Branch: master Commit: c395ec77d7b1b601def6324778535fea0fabd525 URL: https://source.winehq.org/git/wine.git/?a=commit;h=c395ec77d7b1b601def63247… Author: Piotr Caban <piotr(a)codeweavers.com> Date: Sat May 29 19:13:43 2021 +0200 msvcrt: Import coshf implementation from musl. Signed-off-by: Piotr Caban <piotr(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/msvcrt/math.c | 216 ++++++++++++++++++++++++++++++-------------------- dlls/msvcrt/unixlib.c | 9 --- dlls/msvcrt/unixlib.h | 1 - 3 files changed, 129 insertions(+), 97 deletions(-) diff --git a/dlls/msvcrt/math.c b/dlls/msvcrt/math.c index 760f3834761..9aadc6dea1e 100644 --- a/dlls/msvcrt/math.c +++ b/dlls/msvcrt/math.c @@ -549,6 +549,95 @@ recompute: return n & 7; } +#if !defined(__i386__) || _MSVCR_VER >= 120 +/* Copied from musl: src/math/expm1f.c */ +static float __expm1f(float x) +{ + static const float ln2_hi = 6.9313812256e-01, + ln2_lo = 9.0580006145e-06, + invln2 = 1.4426950216e+00, + Q1 = -3.3333212137e-2, + Q2 = 1.5807170421e-3; + + float y, hi, lo, c, t, e, hxs, hfx, r1, twopk; + union {float f; UINT32 i;} u = {x}; + UINT32 hx = u.i & 0x7fffffff; + int k, sign = u.i >> 31; + + /* filter out huge and non-finite argument */ + if (hx >= 0x4195b844) { /* if |x|>=27*ln2 */ + if (hx >= 0x7f800000) /* NaN */ + return u.i == 0xff800000 ? -1 : x; + if (sign) + return math_error(_UNDERFLOW, "exp", x, 0, -1); + if (hx > 0x42b17217) /* x > log(FLT_MAX) */ + return math_error(_OVERFLOW, "exp", x, 0, fp_barrierf(x * FLT_MAX)); + } + + /* argument reduction */ + if (hx > 0x3eb17218) { /* if |x| > 0.5 ln2 */ + if (hx < 0x3F851592) { /* and |x| < 1.5 ln2 */ + if (!sign) { + hi = x - ln2_hi; + lo = ln2_lo; + k = 1; + } else { + hi = x + ln2_hi; + lo = -ln2_lo; + k = -1; + } + } else { + k = invln2 * x + (sign ? -0.5f : 0.5f); + t = k; + hi = x - t * ln2_hi; /* t*ln2_hi is exact here */ + lo = t * ln2_lo; + } + x = hi - lo; + c = (hi - x) - lo; + } else if (hx < 0x33000000) { /* when |x|<2**-25, return x */ + if (hx < 0x00800000) + fp_barrierf(x * x); + return x; + } else + k = 0; + + /* x is now in primary range */ + hfx = 0.5f * x; + hxs = x * hfx; + r1 = 1.0f + hxs * (Q1 + hxs * Q2); + t = 3.0f - r1 * hfx; + e = hxs * ((r1 - t) / (6.0f - x * t)); + if (k == 0) /* c is 0 */ + return x - (x * e - hxs); + e = x * (e - c) - c; + e -= hxs; + /* exp(x) ~ 2^k (x_reduced - e + 1) */ + if (k == -1) + return 0.5f * (x - e) - 0.5f; + if (k == 1) { + if (x < -0.25f) + return -2.0f * (e - (x + 0.5f)); + return 1.0f + 2.0f * (x - e); + } + u.i = (0x7f + k) << 23; /* 2^k */ + twopk = u.f; + if (k < 0 || k > 56) { /* suffice to return exp(x)-1 */ + y = x - e + 1.0f; + if (k == 128) + y = y * 2.0f * 0x1p127f; + else + y = y * twopk; + return y - 1.0f; + } + u.i = (0x7f-k) << 23; /* 2^-k */ + if (k < 23) + y = (x - e + (1 - u.f)) * twopk; + else + y = (x - (e + u.f) + 1) * twopk; + return y; +} +#endif + #ifndef __i386__ /********************************************************************* @@ -994,14 +1083,50 @@ float CDECL cosf( float x ) } } +/* Copied from musl: src/math/__expo2f.c */ +static float __expo2f(float x, float sign) +{ + static const int k = 235; + static const float kln2 = 0x1.45c778p+7f; + float scale; + + *(UINT32*)&scale = (UINT32)(0x7f + k/2) << 23; + return expf(x - kln2) * (sign * scale) * scale; +} + /********************************************************************* * coshf (MSVCRT.@) + * + * Copied from musl: src/math/coshf.c */ float CDECL coshf( float x ) { - float ret = unix_funcs->coshf( x ); - if (isnan(x)) return math_error(_DOMAIN, "coshf", x, 0, ret); - return ret; + UINT32 ui = *(UINT32*)&x; + float t; + + /* |x| */ + ui &= 0x7fffffff; + x = *(float*)&ui; + + /* |x| < log(2) */ + if (ui < 0x3f317217) { + if (ui < 0x3f800000 - (12 << 23)) { + fp_barrierf(x + 0x1p120f); + return 1; + } + t = __expm1f(x); + return 1 + t * t / (2 * (1 + t)); + } + + /* |x| < log(FLT_MAX) */ + if (ui < 0x42b17217) { + t = expf(x); + return 0.5f * (t + 1 / t); + } + + /* |x| > log(FLT_MAX) or nan */ + t = __expo2f(x, 1.0f); + return t; } /********************************************************************* @@ -5899,93 +6024,10 @@ double CDECL expm1(double x) /********************************************************************* * expm1f (MSVCR120.@) - * - * Copied from musl: src/math/expm1f.c */ float CDECL expm1f(float x) { - static const float ln2_hi = 6.9313812256e-01, - ln2_lo = 9.0580006145e-06, - invln2 = 1.4426950216e+00, - Q1 = -3.3333212137e-2, - Q2 = 1.5807170421e-3; - - float y, hi, lo, c, t, e, hxs, hfx, r1, twopk; - union {float f; UINT32 i;} u = {x}; - UINT32 hx = u.i & 0x7fffffff; - int k, sign = u.i >> 31; - - /* filter out huge and non-finite argument */ - if (hx >= 0x4195b844) { /* if |x|>=27*ln2 */ - if (hx >= 0x7f800000) /* NaN */ - return u.i == 0xff800000 ? -1 : x; - if (sign) - return math_error(_UNDERFLOW, "exp", x, 0, -1); - if (hx > 0x42b17217) /* x > log(FLT_MAX) */ - return math_error(_OVERFLOW, "exp", x, 0, fp_barrierf(x * FLT_MAX)); - } - - /* argument reduction */ - if (hx > 0x3eb17218) { /* if |x| > 0.5 ln2 */ - if (hx < 0x3F851592) { /* and |x| < 1.5 ln2 */ - if (!sign) { - hi = x - ln2_hi; - lo = ln2_lo; - k = 1; - } else { - hi = x + ln2_hi; - lo = -ln2_lo; - k = -1; - } - } else { - k = invln2 * x + (sign ? -0.5f : 0.5f); - t = k; - hi = x - t * ln2_hi; /* t*ln2_hi is exact here */ - lo = t * ln2_lo; - } - x = hi - lo; - c = (hi - x) - lo; - } else if (hx < 0x33000000) { /* when |x|<2**-25, return x */ - if (hx < 0x00800000) - fp_barrierf(x * x); - return x; - } else - k = 0; - - /* x is now in primary range */ - hfx = 0.5f * x; - hxs = x * hfx; - r1 = 1.0f + hxs * (Q1 + hxs * Q2); - t = 3.0f - r1 * hfx; - e = hxs * ((r1 - t) / (6.0f - x * t)); - if (k == 0) /* c is 0 */ - return x - (x * e - hxs); - e = x * (e - c) - c; - e -= hxs; - /* exp(x) ~ 2^k (x_reduced - e + 1) */ - if (k == -1) - return 0.5f * (x - e) - 0.5f; - if (k == 1) { - if (x < -0.25f) - return -2.0f * (e - (x + 0.5f)); - return 1.0f + 2.0f * (x - e); - } - u.i = (0x7f + k) << 23; /* 2^k */ - twopk = u.f; - if (k < 0 || k > 56) { /* suffice to return exp(x)-1 */ - y = x - e + 1.0f; - if (k == 128) - y = y * 2.0f * 0x1p127f; - else - y = y * twopk; - return y - 1.0f; - } - u.i = (0x7f-k) << 23; /* 2^-k */ - if (k < 23) - y = (x - e + (1 - u.f)) * twopk; - else - y = (x - (e + u.f) + 1) * twopk; - return y; + return __expm1f(x); } /********************************************************************* diff --git a/dlls/msvcrt/unixlib.c b/dlls/msvcrt/unixlib.c index 6af252e5e9b..e5e4233d12f 100644 --- a/dlls/msvcrt/unixlib.c +++ b/dlls/msvcrt/unixlib.c @@ -42,14 +42,6 @@ WINE_DEFAULT_DEBUG_CHANNEL(msvcrt); -/********************************************************************* - * coshf - */ -static float CDECL unix_coshf( float x ) -{ - return coshf( x ); -} - /********************************************************************* * exp */ @@ -336,7 +328,6 @@ static float CDECL unix_tgammaf(float x) static const struct unix_funcs funcs = { - unix_coshf, unix_exp, unix_expf, unix_exp2, diff --git a/dlls/msvcrt/unixlib.h b/dlls/msvcrt/unixlib.h index 67dd1f98b84..4c30fc220fe 100644 --- a/dlls/msvcrt/unixlib.h +++ b/dlls/msvcrt/unixlib.h @@ -23,7 +23,6 @@ struct unix_funcs { - float (CDECL *coshf)(float x); double (CDECL *exp)(double x); float (CDECL *expf)(float x); double (CDECL *exp2)(double x);

1 0

Piotr Caban : msvcrt: Import sinh implementation from musl.
by Alexandre Julliard 31 May '21

31 May '21

Module: wine Branch: master Commit: 739d4f0c943732b88ee135884a51ccd272c46047 URL: https://source.winehq.org/git/wine.git/?a=commit;h=739d4f0c943732b88ee13588… Author: Piotr Caban <piotr(a)codeweavers.com> Date: Sat May 29 19:13:46 2021 +0200 msvcrt: Import sinh implementation from musl. Signed-off-by: Piotr Caban <piotr(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/msvcrt/math.c | 30 +++++++++++++++++++++++++++--- dlls/msvcrt/unixlib.c | 9 --------- dlls/msvcrt/unixlib.h | 1 - 3 files changed, 27 insertions(+), 13 deletions(-) diff --git a/dlls/msvcrt/math.c b/dlls/msvcrt/math.c index 9aadc6dea1e..5ecbb8bf037 100644 --- a/dlls/msvcrt/math.c +++ b/dlls/msvcrt/math.c @@ -2527,9 +2527,33 @@ double CDECL sin( double x ) */ double CDECL sinh( double x ) { - double ret = unix_funcs->sinh( x ); - if (isnan(x)) return math_error(_DOMAIN, "sinh", x, 0, ret); - return ret; + UINT64 ux = *(UINT64*)&x; + UINT32 w; + double t, h, absx; + + h = 0.5; + if (ux >> 63) + h = -h; + /* |x| */ + ux &= (UINT64)-1 / 2; + absx = *(double*)&ux; + w = ux >> 32; + + /* |x| < log(DBL_MAX) */ + if (w < 0x40862e42) { + t = __expm1(absx); + if (w < 0x3ff00000) { + if (w < 0x3ff00000 - (26 << 20)) + return x; + return h * (2 * t - t * t / (t + 1)); + } + return h * (t + t / (t + 1)); + } + + /* |x| > log(DBL_MAX) or nan */ + /* note: the result is stored to handle overflow */ + t = __expo2(absx, 2 * h); + return t; } static BOOL sqrt_validate( double *x, BOOL update_sw ) diff --git a/dlls/msvcrt/unixlib.c b/dlls/msvcrt/unixlib.c index e5e4233d12f..2f447533fb6 100644 --- a/dlls/msvcrt/unixlib.c +++ b/dlls/msvcrt/unixlib.c @@ -268,14 +268,6 @@ static float CDECL unix_powf( float x, float y ) return powf( x, y ); } -/********************************************************************* - * sinh - */ -static double CDECL unix_sinh( double x ) -{ - return sinh( x ); -} - /********************************************************************* * sinhf */ @@ -351,7 +343,6 @@ static const struct unix_funcs funcs = unix_log2f, unix_pow, unix_powf, - unix_sinh, unix_sinhf, unix_tanh, unix_tanhf, diff --git a/dlls/msvcrt/unixlib.h b/dlls/msvcrt/unixlib.h index 4c30fc220fe..80a150b9031 100644 --- a/dlls/msvcrt/unixlib.h +++ b/dlls/msvcrt/unixlib.h @@ -46,7 +46,6 @@ struct unix_funcs float (CDECL *log2f)(float x); double (CDECL *pow)(double x, double y); float (CDECL *powf)(float x, float y); - double (CDECL *sinh)(double x); float (CDECL *sinhf)(float x); double (CDECL *tanh)(double x); float (CDECL *tanhf)(float x);

1 0

Piotr Caban : msvcrt: Import sinhf implementation from musl.
by Alexandre Julliard 31 May '21

31 May '21

Module: wine Branch: master Commit: ae48d58362d92b378cb47e07b46ff16cb156fd68 URL: https://source.winehq.org/git/wine.git/?a=commit;h=ae48d58362d92b378cb47e07… Author: Piotr Caban <piotr(a)codeweavers.com> Date: Sat May 29 19:13:49 2021 +0200 msvcrt: Import sinhf implementation from musl. Signed-off-by: Piotr Caban <piotr(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/msvcrt/math.c | 27 ++++++++++++++++++++++++--- dlls/msvcrt/unixlib.c | 9 --------- dlls/msvcrt/unixlib.h | 1 - 3 files changed, 24 insertions(+), 13 deletions(-) diff --git a/dlls/msvcrt/math.c b/dlls/msvcrt/math.c index 5ecbb8bf037..05571a7fe20 100644 --- a/dlls/msvcrt/math.c +++ b/dlls/msvcrt/math.c @@ -1312,9 +1312,30 @@ float CDECL sinf( float x ) */ float CDECL sinhf( float x ) { - float ret = unix_funcs->sinhf( x ); - if (isnan(x)) return math_error(_DOMAIN, "sinhf", x, 0, ret); - return ret; + UINT32 ui = *(UINT32*)&x; + float t, h, absx; + + h = 0.5; + if (ui >> 31) + h = -h; + /* |x| */ + ui &= 0x7fffffff; + absx = *(float*)&ui; + + /* |x| < log(FLT_MAX) */ + if (ui < 0x42b17217) { + t = __expm1f(absx); + if (ui < 0x3f800000) { + if (ui < 0x3f800000 - (12 << 23)) + return x; + return h * (2 * t - t * t / (t + 1)); + } + return h * (t + t / (t + 1)); + } + + /* |x| > logf(FLT_MAX) or nan */ + t = __expo2f(absx, 2 * h); + return t; } static BOOL sqrtf_validate( float *x ) diff --git a/dlls/msvcrt/unixlib.c b/dlls/msvcrt/unixlib.c index 2f447533fb6..918c7e6ed71 100644 --- a/dlls/msvcrt/unixlib.c +++ b/dlls/msvcrt/unixlib.c @@ -268,14 +268,6 @@ static float CDECL unix_powf( float x, float y ) return powf( x, y ); } -/********************************************************************* - * sinhf - */ -static float CDECL unix_sinhf( float x ) -{ - return sinhf( x ); -} - /********************************************************************* * tanh */ @@ -343,7 +335,6 @@ static const struct unix_funcs funcs = unix_log2f, unix_pow, unix_powf, - unix_sinhf, unix_tanh, unix_tanhf, unix_tgamma, diff --git a/dlls/msvcrt/unixlib.h b/dlls/msvcrt/unixlib.h index 80a150b9031..0e3074be0c5 100644 --- a/dlls/msvcrt/unixlib.h +++ b/dlls/msvcrt/unixlib.h @@ -46,7 +46,6 @@ struct unix_funcs float (CDECL *log2f)(float x); double (CDECL *pow)(double x, double y); float (CDECL *powf)(float x, float y); - float (CDECL *sinhf)(float x); double (CDECL *tanh)(double x); float (CDECL *tanhf)(float x); double (CDECL *tgamma)(double x);

1 0

Zebediah Figura : vkd3d-shader: Lower DIV to RCP + MUL for SM1.
by Alexandre Julliard 31 May '21

31 May '21

Module: vkd3d Branch: master Commit: 100c65c967704411b751fee17e9c0b165be9c6d7 URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=100c65c967704411b751fee… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri May 21 00:32:24 2021 -0500 vkd3d-shader: Lower DIV to RCP + MUL for SM1. Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Matteo Bruni <mbruni(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d-shader/hlsl_codegen.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index b4ea839..afa096b 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -388,6 +388,27 @@ static bool fold_constants(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, voi return true; } +/* Lower DIV to RCP + MUL. */ +static bool lower_division(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) +{ + struct hlsl_ir_expr *expr; + struct hlsl_ir_node *rcp; + + if (instr->type != HLSL_IR_EXPR) + return false; + expr = hlsl_ir_expr(instr); + if (expr->op != HLSL_IR_BINOP_DIV) + return false; + + if (!(rcp = hlsl_new_unary_expr(ctx, HLSL_IR_UNOP_RCP, expr->operands[1].node, instr->loc))) + return false; + list_add_before(&expr->node.entry, &rcp->entry); + expr->op = HLSL_IR_BINOP_MUL; + hlsl_src_remove(&expr->operands[1]); + hlsl_src_from_node(&expr->operands[1], rcp); + return true; +} + static bool dce(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) { switch (instr->type) @@ -1954,6 +1975,9 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun while (transform_ir(ctx, split_struct_copies, entry_func->body, NULL)); while (transform_ir(ctx, fold_constants, entry_func->body, NULL)); + if (ctx->profile->major_version < 4) + transform_ir(ctx, lower_division, entry_func->body, NULL); + do compute_liveness(ctx, entry_func); while (transform_ir(ctx, dce, entry_func->body, NULL));

1 0

Zebediah Figura : vkd3d-shader: Introduce HLSL allocation helpers.
by Alexandre Julliard 31 May '21

31 May '21

Module: vkd3d Branch: master Commit: 49726b0e649f196841abf8ab1b3604ac7c65ab1c URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=49726b0e649f196841abf8a… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri May 21 00:32:20 2021 -0500 vkd3d-shader: Introduce HLSL allocation helpers. To automatically put the compilation context in a failed state. Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Matteo Bruni <mbruni(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d-shader/hlsl.c | 102 ++++++++++++++------------- libs/vkd3d-shader/hlsl.h | 57 +++++++++++---- libs/vkd3d-shader/hlsl.l | 5 +- libs/vkd3d-shader/hlsl.y | 137 ++++++++++++++++++------------------ libs/vkd3d-shader/hlsl_codegen.c | 147 +++++++++++++-------------------------- 5 files changed, 216 insertions(+), 232 deletions(-) Diff: https://source.winehq.org/git/vkd3d.git/?a=commitdiff;h=49726b0e649f196841a…

1 0

Zebediah Figura : vkd3d-shader: Introduce HLSL string buffer helpers.
by Alexandre Julliard 31 May '21

31 May '21

Module: vkd3d Branch: master Commit: 4fe1e86310eeabd2ed49f25bb4f076a509cc881e URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=4fe1e86310eeabd2ed49f25… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri May 21 00:32:21 2021 -0500 vkd3d-shader: Introduce HLSL string buffer helpers. Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Matteo Bruni <mbruni(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d-shader/hlsl.h | 14 ++++++++++++++ libs/vkd3d-shader/hlsl_codegen.c | 33 ++++++++++++--------------------- 2 files changed, 26 insertions(+), 21 deletions(-) diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 3e7d244..3e6d1c3 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -567,6 +567,20 @@ static inline bool hlsl_array_reserve(struct hlsl_ctx *ctx, void **elements, return ret; } +static inline struct vkd3d_string_buffer *hlsl_get_string_buffer(struct hlsl_ctx *ctx) +{ + struct vkd3d_string_buffer *ret = vkd3d_string_buffer_get(&ctx->string_buffers); + + if (!ret) + ctx->failed = true; + return ret; +} + +static inline void hlsl_release_string_buffer(struct hlsl_ctx *ctx, struct vkd3d_string_buffer *buffer) +{ + vkd3d_string_buffer_release(&ctx->string_buffers, buffer); +} + const char *debug_hlsl_type(const struct hlsl_type *type) DECLSPEC_HIDDEN; const char *debug_hlsl_writemask(unsigned int writemask) DECLSPEC_HIDDEN; diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 1705529..84a47ed 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -42,14 +42,11 @@ static void prepend_uniform_copy(struct hlsl_ctx *ctx, struct list *instrs, stru uniform->is_uniform = 1; uniform->is_param = temp->is_param; - if (!(name = vkd3d_string_buffer_get(&ctx->string_buffers))) - { - ctx->failed = true; + if (!(name = hlsl_get_string_buffer(ctx))) return; - } vkd3d_string_buffer_printf(name, "<temp-%s>", temp->name); temp->name = hlsl_strdup(ctx, name->buffer); - vkd3d_string_buffer_release(&ctx->string_buffers, name); + hlsl_release_string_buffer(ctx, name); if (!(load = hlsl_new_var_load(ctx, uniform, temp->loc))) return; @@ -70,25 +67,22 @@ static void prepend_input_copy(struct hlsl_ctx *ctx, struct list *instrs, struct struct hlsl_ir_load *load; struct hlsl_ir_var *input; - if (!(name = vkd3d_string_buffer_get(&ctx->string_buffers))) - { - ctx->failed = true; + if (!(name = hlsl_get_string_buffer(ctx))) return; - } vkd3d_string_buffer_printf(name, "<input-%s%u>", semantic->name, semantic->index); if (!(new_semantic.name = hlsl_strdup(ctx, semantic->name))) { - vkd3d_string_buffer_release(&ctx->string_buffers, name); + hlsl_release_string_buffer(ctx, name); return; } new_semantic.index = semantic->index; if (!(input = hlsl_new_var(ctx, hlsl_strdup(ctx, name->buffer), type, var->loc, &new_semantic, 0, NULL))) { - vkd3d_string_buffer_release(&ctx->string_buffers, name); + hlsl_release_string_buffer(ctx, name); vkd3d_free((void *)new_semantic.name); return; } - vkd3d_string_buffer_release(&ctx->string_buffers, name); + hlsl_release_string_buffer(ctx, name); input->is_input_semantic = 1; input->is_param = var->is_param; list_add_before(&var->scope_entry, &input->scope_entry); @@ -144,25 +138,22 @@ static void append_output_copy(struct hlsl_ctx *ctx, struct list *instrs, struct struct hlsl_ir_var *output; struct hlsl_ir_load *load; - if (!(name = vkd3d_string_buffer_get(&ctx->string_buffers))) - { - ctx->failed = true; + if (!(name = hlsl_get_string_buffer(ctx))) return; - } vkd3d_string_buffer_printf(name, "<output-%s%u>", semantic->name, semantic->index); if (!(new_semantic.name = hlsl_strdup(ctx, semantic->name))) { - vkd3d_string_buffer_release(&ctx->string_buffers, name); + hlsl_release_string_buffer(ctx, name); return; } new_semantic.index = semantic->index; if (!(output = hlsl_new_var(ctx, hlsl_strdup(ctx, name->buffer), type, var->loc, &new_semantic, 0, NULL))) { vkd3d_free((void *)new_semantic.name); - vkd3d_string_buffer_release(&ctx->string_buffers, name); + hlsl_release_string_buffer(ctx, name); return; } - vkd3d_string_buffer_release(&ctx->string_buffers, name); + hlsl_release_string_buffer(ctx, name); output->is_output_semantic = 1; output->is_param = var->is_param; list_add_before(&var->scope_entry, &output->scope_entry); @@ -1422,7 +1413,7 @@ static void write_sm1_uniforms(struct hlsl_ctx *ctx, struct bytecode_buffer *buf { struct vkd3d_string_buffer *name; - if (!(name = vkd3d_string_buffer_get(&ctx->string_buffers))) + if (!(name = hlsl_get_string_buffer(ctx))) { buffer->status = VKD3D_ERROR_OUT_OF_MEMORY; return; @@ -1430,7 +1421,7 @@ static void write_sm1_uniforms(struct hlsl_ctx *ctx, struct bytecode_buffer *buf vkd3d_string_buffer_printf(name, "$%s", var->name); vkd3d_free((char *)var->name); var->name = hlsl_strdup(ctx, name->buffer); - vkd3d_string_buffer_release(&ctx->string_buffers, name); + hlsl_release_string_buffer(ctx, name); } } }

1 0

Zebediah Figura : vkd3d-shader: Use the HLSL string buffer cache in hlsl_type_to_string() and hlsl_modifiers_to_string().
by Alexandre Julliard 31 May '21

31 May '21

Module: vkd3d Branch: master Commit: ba1cc670e68175e5506dfcc41ea0ad6df8d81eab URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=ba1cc670e68175e5506dfcc… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri May 21 00:32:22 2021 -0500 vkd3d-shader: Use the HLSL string buffer cache in hlsl_type_to_string() and hlsl_modifiers_to_string(). Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Matteo Bruni <mbruni(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d-shader/hlsl.c | 66 +++++++++++++++++--------------------- libs/vkd3d-shader/hlsl.h | 10 +++--- libs/vkd3d-shader/hlsl.y | 68 ++++++++++++++++++++-------------------- libs/vkd3d-shader/hlsl_codegen.c | 9 ++++-- 4 files changed, 73 insertions(+), 80 deletions(-) Diff: https://source.winehq.org/git/vkd3d.git/?a=commitdiff;h=ba1cc670e68175e5506…

1 0

Zebediah Figura : vkd3d-shader: Distinguish between resource exhaustion and invalid shaders when returning failure.
by Alexandre Julliard 31 May '21

31 May '21

Module: vkd3d Branch: master Commit: 6493953516f9c798f221da0f0cf0a58d97034798 URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=6493953516f9c798f221da0… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri May 21 00:32:23 2021 -0500 vkd3d-shader: Distinguish between resource exhaustion and invalid shaders when returning failure. Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Matteo Bruni <mbruni(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d-shader/hlsl.c | 7 ++++--- libs/vkd3d-shader/hlsl.h | 10 +++++----- libs/vkd3d-shader/hlsl_codegen.c | 4 ++-- 3 files changed, 11 insertions(+), 10 deletions(-) diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index f7e2ad5..441fd6c 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -39,7 +39,8 @@ void hlsl_error(struct hlsl_ctx *ctx, const struct vkd3d_shader_location loc, vkd3d_shader_verror(ctx->message_context, &loc, error, fmt, args); va_end(args); - ctx->failed = true; + if (!ctx->result) + ctx->result = VKD3D_ERROR_INVALID_SHADER; } void hlsl_warning(struct hlsl_ctx *ctx, const struct vkd3d_shader_location loc, @@ -1648,10 +1649,10 @@ int hlsl_compile_shader(const struct vkd3d_shader_code *hlsl, const struct vkd3d return VKD3D_ERROR_OUT_OF_MEMORY; } - if (ctx.failed) + if (ctx.result) { hlsl_ctx_cleanup(&ctx); - return VKD3D_ERROR_INVALID_SHADER; + return ctx.result; } if (!(entry_func = hlsl_get_func_decl(&ctx, entry_point))) diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 55665c1..4fe90a9 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -428,7 +428,7 @@ struct hlsl_ctx struct vkd3d_shader_location location; struct vkd3d_shader_message_context *message_context; struct vkd3d_string_buffer_cache string_buffers; - bool failed; + int result; void *scanner; @@ -544,7 +544,7 @@ static inline void *hlsl_alloc(struct hlsl_ctx *ctx, size_t size) void *ptr = vkd3d_calloc(1, size); if (!ptr) - ctx->failed = true; + ctx->result = VKD3D_ERROR_OUT_OF_MEMORY; return ptr; } @@ -553,7 +553,7 @@ static inline char *hlsl_strdup(struct hlsl_ctx *ctx, const char *string) char *ptr = vkd3d_strdup(string); if (!ptr) - ctx->failed = true; + ctx->result = VKD3D_ERROR_OUT_OF_MEMORY; return ptr; } @@ -563,7 +563,7 @@ static inline bool hlsl_array_reserve(struct hlsl_ctx *ctx, void **elements, bool ret = vkd3d_array_reserve(elements, capacity, element_count, element_size); if (!ret) - ctx->failed = true; + ctx->result = VKD3D_ERROR_OUT_OF_MEMORY; return ret; } @@ -572,7 +572,7 @@ static inline struct vkd3d_string_buffer *hlsl_get_string_buffer(struct hlsl_ctx struct vkd3d_string_buffer *ret = vkd3d_string_buffer_get(&ctx->string_buffers); if (!ret) - ctx->failed = true; + ctx->result = VKD3D_ERROR_OUT_OF_MEMORY; return ret; } diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 243deec..b4ea839 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -1968,8 +1968,8 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun allocate_const_registers(ctx, entry_func); allocate_semantic_registers(ctx); - if (ctx->failed) - return VKD3D_ERROR_INVALID_SHADER; + if (ctx->result) + return ctx->result; if (ctx->profile->major_version < 4) return write_sm1_shader(ctx, entry_func, out);

1 0

Nikolay Sivov : mf/session: Do not drop pending commands when clearing current presentation.
by Alexandre Julliard 28 May '21

28 May '21

Module: wine Branch: master Commit: 20a1eb3b252b404bd2ed9a11dc03312ea11fff0d URL: https://source.winehq.org/git/wine.git/?a=commit;h=20a1eb3b252b404bd2ed9a11… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Fri May 28 15:49:02 2021 +0300 mf/session: Do not drop pending commands when clearing current presentation. Calling Start() short after SetTopology() has a good chance of getting lost, because clearing logic will drop all pending commands. This is easy to reproduce with IMFPMediaPlayer, which works reliably on Windows. Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/mf/session.c | 21 +++++++++++++-------- 1 file changed, 13 insertions(+), 8 deletions(-) diff --git a/dlls/mf/session.c b/dlls/mf/session.c index 925a8c93d20..41caf02d059 100644 --- a/dlls/mf/session.c +++ b/dlls/mf/session.c @@ -855,12 +855,22 @@ static void session_shutdown_current_topology(struct media_session *session) } } +static void session_clear_command_list(struct media_session *session) +{ + struct session_op *op, *op2; + + LIST_FOR_EACH_ENTRY_SAFE(op, op2, &session->commands, struct session_op, entry) + { + list_remove(&op->entry); + IUnknown_Release(&op->IUnknown_iface); + } +} + static void session_clear_presentation(struct media_session *session) { struct media_source *source, *source2; struct media_sink *sink, *sink2; struct topo_node *node, *node2; - struct session_op *op, *op2; session_shutdown_current_topology(session); @@ -895,12 +905,6 @@ static void session_clear_presentation(struct media_session *session) IMFMediaEventGenerator_Release(sink->event_generator); heap_free(sink); } - - LIST_FOR_EACH_ENTRY_SAFE(op, op2, &session->commands, struct session_op, entry) - { - list_remove(&op->entry); - IUnknown_Release(&op->IUnknown_iface); - } } static struct topo_node *session_get_node_by_id(const struct media_session *session, TOPOID id) @@ -1730,6 +1734,7 @@ static ULONG WINAPI mfsession_Release(IMFMediaSession *iface) { session_clear_topologies(session); session_clear_presentation(session); + session_clear_command_list(session); if (session->presentation.current_topology) IMFTopology_Release(session->presentation.current_topology); if (session->event_queue) @@ -1847,7 +1852,6 @@ static HRESULT WINAPI mfsession_Start(IMFMediaSession *iface, const GUID *format hr = session_submit_command(session, op); IUnknown_Release(&op->IUnknown_iface); - return hr; } @@ -1896,6 +1900,7 @@ static HRESULT WINAPI mfsession_Shutdown(IMFMediaSession *iface) IMFPresentationClock_Release(session->clock); session->clock = NULL; session_clear_presentation(session); + session_clear_command_list(session); } LeaveCriticalSection(&session->cs);

1 0