mirror of
git://sourceware.org/git/newlib-cygwin.git
synced 2025-01-23 23:47:22 +08:00
e18743072b
This implements a set of vectorized math routines to be used by the compiler auto-vectorizer. Versions for vectors with 2 lanes up to 64 lanes (in powers of 2) are provided. These routines are based on the scalar versions of the math routines in libm/common, libm/math and libm/mathfp. They make extensive use of the GCC C vector extensions and GCN-specific builtins in GCC.
307 lines
11 KiB
C
307 lines
11 KiB
C
/*
|
|
* Copyright 2023 Siemens
|
|
*
|
|
* The authors hereby grant permission to use, copy, modify, distribute,
|
|
* and license this software and its documentation for any purpose, provided
|
|
* that existing copyright notices are retained in all copies and that this
|
|
* notice is included verbatim in any distributions. No written agreement,
|
|
* license, or royalty fee is required for any of the authorized uses.
|
|
* Modifications to this software may be copyrighted by their authors
|
|
* and need not follow the licensing terms described here, provided that
|
|
* the new terms are clearly indicated on the first page of each file where
|
|
* they apply.
|
|
*/
|
|
|
|
/*
|
|
* ====================================================
|
|
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
|
|
*
|
|
* Developed at SunPro, a Sun Microsystems, Inc. business.
|
|
* Permission to use, copy, modify, and distribute this
|
|
* software is freely granted, provided that this notice
|
|
* is preserved.
|
|
* ====================================================
|
|
*/
|
|
|
|
/* Based on newlib/libm/math/ef_pow.c in Newlib. */
|
|
|
|
#include "amdgcnmach.h"
|
|
|
|
static const float
|
|
bp[] = {1.0, 1.5,},
|
|
dp_h[] = { 0.0, 5.84960938e-01,}, /* 0x3f15c000 */
|
|
dp_l[] = { 0.0, 1.56322085e-06,}, /* 0x35d1cfdc */
|
|
zero = 0.0,
|
|
one = 1.0,
|
|
two = 2.0,
|
|
two24 = 16777216.0, /* 0x4b800000 */
|
|
/* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
|
|
L1 = 6.0000002384e-01, /* 0x3f19999a */
|
|
L2 = 4.2857143283e-01, /* 0x3edb6db7 */
|
|
L3 = 3.3333334327e-01, /* 0x3eaaaaab */
|
|
L4 = 2.7272811532e-01, /* 0x3e8ba305 */
|
|
L5 = 2.3066075146e-01, /* 0x3e6c3255 */
|
|
L6 = 2.0697501302e-01, /* 0x3e53f142 */
|
|
P1 = 1.6666667163e-01, /* 0x3e2aaaab */
|
|
P2 = -2.7777778450e-03, /* 0xbb360b61 */
|
|
P3 = 6.6137559770e-05, /* 0x388ab355 */
|
|
P4 = -1.6533901999e-06, /* 0xb5ddea0e */
|
|
P5 = 4.1381369442e-08, /* 0x3331bb4c */
|
|
lg2 = 6.9314718246e-01, /* 0x3f317218 */
|
|
lg2_h = 6.93145752e-01, /* 0x3f317200 */
|
|
lg2_l = 1.42860654e-06, /* 0x35bfbe8c */
|
|
ovt = 4.2995665694e-08, /* -(128-log2(ovfl+.5ulp)) */
|
|
cp = 9.6179670095e-01, /* 0x3f76384f =2/(3ln2) */
|
|
cp_h = 9.6179199219e-01, /* 0x3f763800 =head of cp */
|
|
cp_l = 4.7017383622e-06, /* 0x369dc3a0 =tail of cp_h */
|
|
ivln2 = 1.4426950216e+00, /* 0x3fb8aa3b =1/ln2 */
|
|
ivln2_h = 1.4426879883e+00, /* 0x3fb8aa00 =16b 1/ln2*/
|
|
ivln2_l = 7.0526075433e-06; /* 0x36eca570 =1/ln2 tail*/
|
|
|
|
v64sf v64sf_sqrtf_aux (v64sf, v64si);
|
|
v64sf v64sf_scalbnf_aux (v64sf, v64si, v64si);
|
|
|
|
#if defined (__has_builtin) && __has_builtin (__builtin_gcn_fabsvf)
|
|
|
|
DEF_VS_MATH_FUNC (v64sf, powf, v64sf x, v64sf y)
|
|
{
|
|
FUNCTION_INIT (v64sf);
|
|
|
|
v64si hx, hy;
|
|
GET_FLOAT_WORD (hx,x, NO_COND);
|
|
GET_FLOAT_WORD (hy,y, NO_COND);
|
|
v64si ix = hx&0x7fffffff;
|
|
v64si iy = hy&0x7fffffff;
|
|
|
|
/* y==zero: x**0 = 1 */
|
|
VECTOR_IF (FLT_UWORD_IS_ZERO(iy), cond)
|
|
VECTOR_RETURN (x + y, cond & v64sf_issignalingf_inline(x));
|
|
VECTOR_RETURN (VECTOR_INIT (1.0f), cond);
|
|
VECTOR_ENDIF
|
|
|
|
/* x|y==NaN return NaN unless x==1 then return 1 */
|
|
VECTOR_IF (FLT_UWORD_IS_NAN(ix) | FLT_UWORD_IS_NAN(iy), cond)
|
|
VECTOR_IF2 (hx==0x3f800000 & ~v64sf_issignalingf_inline(y), cond2, cond)
|
|
VECTOR_RETURN (VECTOR_INIT (1.0f), cond2);
|
|
VECTOR_ELSE2 (cond2, cond)
|
|
VECTOR_RETURN (x + y, cond2);
|
|
VECTOR_ENDIF
|
|
VECTOR_ENDIF
|
|
|
|
/* determine if y is an odd int when x < 0
|
|
* yisint = 0 ... y is not an integer
|
|
* yisint = 1 ... y is an odd int
|
|
* yisint = 2 ... y is an even int
|
|
*/
|
|
v64si yisint = VECTOR_INIT (0);
|
|
|
|
VECTOR_IF (hx < 0, cond)
|
|
VECTOR_IF2 (iy >= 0x4b800000, cond2, cond)
|
|
VECTOR_COND_MOVE (yisint, VECTOR_INIT (2), cond2); /* even integer y */
|
|
VECTOR_ELSEIF2 (iy >= 0x3f800000, cond2, cond)
|
|
v64si k = (iy>>23)-0x7f; /* exponent */
|
|
v64si j = iy>>(23-k);
|
|
VECTOR_COND_MOVE (yisint, 2-(j&1), cond2 & (j<<(23-k))==iy);
|
|
VECTOR_ENDIF
|
|
VECTOR_ENDIF
|
|
|
|
/* special value of y */
|
|
VECTOR_IF (FLT_UWORD_IS_INFINITE(iy), cond) /* y is +-inf */
|
|
VECTOR_IF2 (ix==0x3f800000, cond2, cond)
|
|
VECTOR_RETURN (VECTOR_INIT (1.0f), cond2); /* +-1**+-inf = 1 */
|
|
VECTOR_ELSEIF2 (ix > 0x3f800000, cond2, cond) /* (|x|>1)**+-inf = inf,0 */
|
|
VECTOR_RETURN (y, cond2 & (hy >= 0));
|
|
VECTOR_RETURN (VECTOR_INIT (0.0f), cond2);
|
|
VECTOR_ELSE2 (cond2, cond) /* (|x|<1)**-,+inf = inf,0 */
|
|
VECTOR_RETURN (-y, cond2 & (hy<0));
|
|
VECTOR_RETURN (VECTOR_INIT (0.0f), cond2);
|
|
VECTOR_ENDIF
|
|
VECTOR_ENDIF
|
|
VECTOR_IF (iy==0x3f800000, cond) /* y is +-1 */
|
|
VECTOR_RETURN (VECTOR_INIT (1.0f) / x, cond & (hy<0));
|
|
VECTOR_RETURN (x, cond);
|
|
VECTOR_ENDIF
|
|
|
|
VECTOR_RETURN (x*x, hy==0x40000000); /* y is 2 */
|
|
/* y is 0.5 */
|
|
/* x >= +0 */
|
|
VECTOR_RETURN (v64sf_sqrtf_aux (x, __mask), (hy==0x3f000000) & (hx >= 0));
|
|
|
|
v64sf ax = __builtin_gcn_fabsvf(x);
|
|
/* special value of x */
|
|
VECTOR_IF (FLT_UWORD_IS_INFINITE(ix)|FLT_UWORD_IS_ZERO(ix)|ix==0x3f800000, cond)
|
|
v64sf z = ax; /*x is +-0,+-inf,+-1*/
|
|
VECTOR_COND_MOVE (z, VECTOR_INIT (1.0f) / z, cond & (hy < 0)); /* z = (1/|x|) */
|
|
VECTOR_IF2 (hx<0, cond2, cond)
|
|
VECTOR_IF2 (((ix-0x3f800000)|yisint)==0, cond3, cond2)
|
|
/* (-1)**non-int is NaN */
|
|
VECTOR_COND_MOVE (z, (z-z)/(z-z), cond3);
|
|
VECTOR_ELSEIF2 (yisint==1, cond3, cond2)
|
|
/* (x<0)**odd = -(|x|**odd) */
|
|
VECTOR_COND_MOVE (z, -z, cond3);
|
|
VECTOR_ENDIF
|
|
VECTOR_ENDIF
|
|
VECTOR_RETURN (z, cond);
|
|
VECTOR_ENDIF
|
|
|
|
/* (x<0)**(non-int) is NaN */
|
|
VECTOR_RETURN ((x-x)/(x-x), ((((hx >> 31) & 1) - 1)|yisint)==0);
|
|
|
|
v64sf t1, t2;
|
|
|
|
/* |y| is huge */
|
|
VECTOR_IF (iy>0x4d000000, cond) /* if |y| > 2**27 */
|
|
/* over/underflow if x is not close to one */
|
|
VECTOR_IF2(ix<0x3f7ffff4, cond2, cond)
|
|
VECTOR_RETURN (v64sf_math_oflowf(VECTOR_INIT (0)), cond2 & (hy < 0));
|
|
VECTOR_RETURN (v64sf_math_uflowf(VECTOR_INIT (0)), cond2);
|
|
VECTOR_ENDIF
|
|
VECTOR_IF2(ix>0x3f800007, cond2, cond)
|
|
VECTOR_RETURN (v64sf_math_oflowf(VECTOR_INIT (0)), cond2 & (hy > 0));
|
|
VECTOR_RETURN (v64sf_math_uflowf(VECTOR_INIT (0)), cond2);
|
|
VECTOR_ENDIF
|
|
/* now |1-x| is tiny <= 2**-20, suffice to compute
|
|
log(x) by x-x^2/2+x^3/3-x^4/4 */
|
|
v64sf t = ax-1; /* t has 20 trailing zeros */
|
|
v64sf w = (t*t)*(0.5f-t*(0.333333333333f-t*0.25f));
|
|
v64sf u = ivln2_h*t; /* ivln2_h has 16 sig. bits */
|
|
v64sf v = t*ivln2_l-w*ivln2;
|
|
VECTOR_COND_MOVE (t1, u+v, cond);
|
|
v64si is;
|
|
GET_FLOAT_WORD(is,t1, cond);
|
|
SET_FLOAT_WORD(t1,is&0xfffff000, cond);
|
|
VECTOR_COND_MOVE (t2, v-(t1-u), cond);
|
|
VECTOR_RETURN (VECTOR_INIT (0.123456f), cond);
|
|
VECTOR_ELSE (cond)
|
|
v64si n = VECTOR_INIT (0);
|
|
/* take care subnormal number */
|
|
VECTOR_IF2 (FLT_UWORD_IS_SUBNORMAL(ix), cond2, cond)
|
|
VECTOR_COND_MOVE (ax, ax * two24, cond);
|
|
VECTOR_COND_MOVE (n, n - 24, cond2);
|
|
GET_FLOAT_WORD(ix,ax, cond2);
|
|
VECTOR_ENDIF
|
|
n += (ix>>23)-0x7f;
|
|
v64si j = ix&0x007fffff;
|
|
/* determine interval */
|
|
v64si ix = j|0x3f800000; /* normalize ix */
|
|
v64si k;
|
|
VECTOR_IF2 (j<=0x1cc471, cond2, cond)
|
|
VECTOR_COND_MOVE (k, VECTOR_INIT (0), cond2); /* |x|<sqrt(3/2) */
|
|
VECTOR_ELSEIF2 (j<0x5db3d7, cond2, cond)
|
|
VECTOR_COND_MOVE (k, VECTOR_INIT (1), cond2); /* |x|<sqrt(3) */
|
|
VECTOR_ELSE2 (cond2, cond)
|
|
VECTOR_COND_MOVE (k, VECTOR_INIT (0), cond2);
|
|
VECTOR_COND_MOVE (n, n + 1, cond2);
|
|
VECTOR_COND_MOVE (ix, ix - 0x00800000, cond2);
|
|
VECTOR_ENDIF
|
|
SET_FLOAT_WORD(ax,ix, cond);
|
|
|
|
/* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
|
|
/* bp[0]=1.0, bp[1]=1.5 */
|
|
v64sf bp_k = VECTOR_MERGE (VECTOR_INIT (bp[1]), VECTOR_INIT (bp[0]), k == 1);
|
|
v64sf u = ax-bp_k;
|
|
v64sf v = 1.0f/(ax+bp_k);
|
|
v64sf s = u*v;
|
|
v64sf s_h = s;
|
|
v64si is;
|
|
GET_FLOAT_WORD(is,s_h, cond);
|
|
SET_FLOAT_WORD(s_h,is&0xfffff000, cond);
|
|
/* t_h=ax+bp[k] High */
|
|
v64sf t_h;
|
|
SET_FLOAT_WORD(t_h,((ix>>1)|0x20000000)+0x0040000+(k<<21), cond);
|
|
v64sf t_l = ax - (t_h-bp_k);
|
|
v64sf s_l = v*((u-s_h*t_h)-s_h*t_l);
|
|
/* compute log(ax) */
|
|
v64sf s2 = s*s;
|
|
v64sf r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
|
|
r += s_l*(s_h+s);
|
|
s2 = s_h*s_h;
|
|
t_h = __builtin_convertvector(3.0f+s2+r, v64sf);
|
|
GET_FLOAT_WORD(is,t_h, cond);
|
|
SET_FLOAT_WORD(t_h,is&0xfffff000, cond);
|
|
t_l = r-((t_h-3.0f)-s2);
|
|
/* u+v = s*(1+...) */
|
|
u = s_h*t_h;
|
|
v = s_l*t_h+t_l*s;
|
|
/* 2/(3log2)*(s+...) */
|
|
v64sf p_h = u+v;
|
|
GET_FLOAT_WORD(is,p_h, cond);
|
|
SET_FLOAT_WORD(p_h,is&0xfffff000, cond);
|
|
v64sf p_l = v-(p_h-u);
|
|
v64sf z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */
|
|
v64sf dp_l_k = VECTOR_MERGE (VECTOR_INIT (dp_l[1]), VECTOR_INIT (dp_l[0]), k == 1);
|
|
v64sf z_l = cp_l*p_h+p_l*cp+dp_l_k;
|
|
/* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
|
|
v64sf t = __builtin_convertvector (n, v64sf);
|
|
v64sf dp_h_k = VECTOR_MERGE (VECTOR_INIT (dp_h[1]), VECTOR_INIT (dp_h[0]), k == 1);
|
|
VECTOR_COND_MOVE (t1, (((z_h+z_l)+dp_h_k)+t), cond);
|
|
GET_FLOAT_WORD(is,t1, cond);
|
|
SET_FLOAT_WORD(t1,is&0xfffff000, cond);
|
|
VECTOR_COND_MOVE (t2, z_l-(((t1-t)-dp_h_k)-z_h), cond);
|
|
VECTOR_ENDIF
|
|
|
|
v64sf s = VECTOR_INIT (1.0f); /* s (sign of result -ve**odd) = -1 else = 1 */
|
|
VECTOR_COND_MOVE (s, VECTOR_INIT (-1.0f), /* (-ve)**(odd int) */
|
|
((hx>>31) != 0)&(yisint == 1));
|
|
|
|
/* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
|
|
v64si is;
|
|
GET_FLOAT_WORD(is,y, NO_COND);
|
|
v64sf y1;
|
|
SET_FLOAT_WORD(y1,is&0xfffff000, NO_COND);
|
|
v64sf p_l = (y-y1)*t1+y*t2;
|
|
v64sf p_h = y1*t1;
|
|
v64sf z = p_l+p_h;
|
|
v64si j;
|
|
GET_FLOAT_WORD(j,z, NO_COND);
|
|
v64si i = j&0x7fffffff;
|
|
VECTOR_IF (j>0, cond)
|
|
VECTOR_RETURN (v64sf_math_oflowf(s<0), cond & i>FLT_UWORD_EXP_MAX); /* overflow */
|
|
VECTOR_RETURN (v64sf_math_oflowf(s<0), cond & (i==FLT_UWORD_EXP_MAX)
|
|
& (p_l+ovt>z-p_h)); /* overflow */
|
|
VECTOR_ELSE (cond)
|
|
VECTOR_RETURN (v64sf_math_uflowf(s<0), cond & (i>FLT_UWORD_EXP_MIN)); /* underflow */
|
|
VECTOR_RETURN (v64sf_math_uflowf(s<0), cond & (i==FLT_UWORD_EXP_MIN)
|
|
& (p_l<=z-p_h)); /* underflow */
|
|
VECTOR_ENDIF
|
|
/*
|
|
* compute 2**(p_h+p_l)
|
|
*/
|
|
v64si k = (i>>23)-0x7f;
|
|
v64si n = VECTOR_INIT (0);
|
|
VECTOR_IF (i>0x3f000000, cond) /* if |z| > 0.5, set n = [z+0.5] */
|
|
VECTOR_COND_MOVE (n, j+(0x00800000>>(k+1)), cond);
|
|
k = ((n&0x7fffffff)>>23)-0x7f; /* new k for n */
|
|
v64sf t;
|
|
SET_FLOAT_WORD(t,n&~(0x007fffff>>k), cond);
|
|
VECTOR_COND_MOVE (n, ((n&0x007fffff)|0x00800000)>>(23-k), cond);
|
|
VECTOR_COND_MOVE (n, -n, cond & (j<0));
|
|
VECTOR_COND_MOVE (p_h, p_h - t, cond);
|
|
VECTOR_ENDIF
|
|
v64sf t = p_l+p_h;
|
|
GET_FLOAT_WORD(is,t, NO_COND);
|
|
SET_FLOAT_WORD(t,is&0xfffff000, NO_COND);
|
|
v64sf u = t*lg2_h;
|
|
v64sf v = (p_l-(t-p_h))*lg2+t*lg2_l;
|
|
z = u+v;
|
|
v64sf w = v-(z-u);
|
|
t = z*z;
|
|
t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
|
|
v64sf r = (z*t1)/(t1-2.0f)-(w+z*w);
|
|
z = VECTOR_INIT (1.0f)-(r-z);
|
|
GET_FLOAT_WORD(j,z, NO_COND);
|
|
j += (n<<23);
|
|
VECTOR_IF ((j>>23)<=0, cond)
|
|
VECTOR_COND_MOVE (z, v64sf_scalbnf_aux(z, n, __mask), cond); /* subnormal output */
|
|
VECTOR_ELSE (cond)
|
|
SET_FLOAT_WORD(z, j, cond);
|
|
VECTOR_ENDIF
|
|
VECTOR_RETURN (s*z, NO_COND);
|
|
|
|
FUNCTION_RETURN;
|
|
}
|
|
|
|
DEF_VARIANTS2 (powf, sf, sf)
|
|
|
|
#endif
|