1 /* 128-bit long double support routines for Darwin.
2    Copyright (C) 1993-2014 Free Software Foundation, Inc.
3 
4 This file is part of GCC.
5 
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
10 
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
14 for more details.
15 
16 Under Section 7 of GPL version 3, you are granted additional
17 permissions described in the GCC Runtime Library Exception, version
18 3.1, as published by the Free Software Foundation.
19 
20 You should have received a copy of the GNU General Public License and
21 a copy of the GCC Runtime Library Exception along with this program;
22 see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
23 <http://www.gnu.org/licenses/>.  */
24 
25 
26 /* Implementations of floating-point long double basic arithmetic
27    functions called by the IBM C compiler when generating code for
28    PowerPC platforms.  In particular, the following functions are
29    implemented: __gcc_qadd, __gcc_qsub, __gcc_qmul, and __gcc_qdiv.
30    Double-double algorithms are based on the paper "Doubled-Precision
31    IEEE Standard 754 Floating-Point Arithmetic" by W. Kahan, February 26,
32    1987.  An alternative published reference is "Software for
33    Doubled-Precision Floating-Point Computations", by Seppo Linnainmaa,
34    ACM TOMS vol 7 no 3, September 1981, pages 272-283.  */
35 
36 /* Each long double is made up of two IEEE doubles.  The value of the
37    long double is the sum of the values of the two parts.  The most
38    significant part is required to be the value of the long double
39    rounded to the nearest double, as specified by IEEE.  For Inf
40    values, the least significant part is required to be one of +0.0 or
41    -0.0.  No other requirements are made; so, for example, 1.0 may be
42    represented as (1.0, +0.0) or (1.0, -0.0), and the low part of a
43    NaN is don't-care.
44 
45    This code currently assumes the most significant double is in
46    the lower numbered register or lower addressed memory.  */
47 
48 #if defined (__MACH__) || defined (__powerpc__) || defined (_AIX)
49 
50 #define fabs(x) __builtin_fabs(x)
51 #define isless(x, y) __builtin_isless (x, y)
52 #define inf() __builtin_inf()
53 
54 #define unlikely(x) __builtin_expect ((x), 0)
55 
56 #define nonfinite(a) unlikely (! isless (fabs (a), inf ()))
57 
58 /* Define ALIASNAME as a strong alias for NAME.  */
59 # define strong_alias(name, aliasname) _strong_alias(name, aliasname)
60 # define _strong_alias(name, aliasname) \
61   extern __typeof (name) aliasname __attribute__ ((alias (#name)));
62 
63 /* All these routines actually take two long doubles as parameters,
64    but GCC currently generates poor code when a union is used to turn
65    a long double into a pair of doubles.  */
66 
67 long double __gcc_qadd (double, double, double, double);
68 long double __gcc_qsub (double, double, double, double);
69 long double __gcc_qmul (double, double, double, double);
70 long double __gcc_qdiv (double, double, double, double);
71 
72 #if defined __ELF__ && defined SHARED \
73     && (defined __powerpc64__ || !(defined __linux__ || defined __gnu_hurd__))
74 /* Provide definitions of the old symbol names to satisfy apps and
75    shared libs built against an older libgcc.  To access the _xlq
76    symbols an explicit version reference is needed, so these won't
77    satisfy an unadorned reference like _xlqadd.  If dot symbols are
78    not needed, the assembler will remove the aliases from the symbol
79    table.  */
80 __asm__ (".symver __gcc_qadd,_xlqadd@GCC_3.4\n\t"
81 	 ".symver __gcc_qsub,_xlqsub@GCC_3.4\n\t"
82 	 ".symver __gcc_qmul,_xlqmul@GCC_3.4\n\t"
83 	 ".symver __gcc_qdiv,_xlqdiv@GCC_3.4\n\t"
84 	 ".symver .__gcc_qadd,._xlqadd@GCC_3.4\n\t"
85 	 ".symver .__gcc_qsub,._xlqsub@GCC_3.4\n\t"
86 	 ".symver .__gcc_qmul,._xlqmul@GCC_3.4\n\t"
87 	 ".symver .__gcc_qdiv,._xlqdiv@GCC_3.4");
88 #endif
89 
90 typedef union
91 {
92   long double ldval;
93   double dval[2];
94 } longDblUnion;
95 
96 /* Add two 'long double' values and return the result.	*/
97 long double
__gcc_qadd(double a,double aa,double c,double cc)98 __gcc_qadd (double a, double aa, double c, double cc)
99 {
100   longDblUnion x;
101   double z, q, zz, xh;
102 
103   z = a + c;
104 
105   if (nonfinite (z))
106     {
107       if (fabs (z) != inf())
108 	return z;
109       z = cc + aa + c + a;
110       if (nonfinite (z))
111 	return z;
112       x.dval[0] = z;  /* Will always be DBL_MAX.  */
113       zz = aa + cc;
114       if (fabs(a) > fabs(c))
115 	x.dval[1] = a - z + c + zz;
116       else
117 	x.dval[1] = c - z + a + zz;
118     }
119   else
120     {
121       q = a - z;
122       zz = q + c + (a - (q + z)) + aa + cc;
123 
124       /* Keep -0 result.  */
125       if (zz == 0.0)
126 	return z;
127 
128       xh = z + zz;
129       if (nonfinite (xh))
130 	return xh;
131 
132       x.dval[0] = xh;
133       x.dval[1] = z - xh + zz;
134     }
135   return x.ldval;
136 }
137 
138 long double
__gcc_qsub(double a,double b,double c,double d)139 __gcc_qsub (double a, double b, double c, double d)
140 {
141   return __gcc_qadd (a, b, -c, -d);
142 }
143 
144 #ifdef __NO_FPRS__
145 static double fmsub (double, double, double);
146 #endif
147 
148 long double
__gcc_qmul(double a,double b,double c,double d)149 __gcc_qmul (double a, double b, double c, double d)
150 {
151   longDblUnion z;
152   double t, tau, u, v, w;
153 
154   t = a * c;			/* Highest order double term.  */
155 
156   if (unlikely (t == 0)		/* Preserve -0.  */
157       || nonfinite (t))
158     return t;
159 
160   /* Sum terms of two highest orders. */
161 
162   /* Use fused multiply-add to get low part of a * c.  */
163 #ifndef __NO_FPRS__
164   asm ("fmsub %0,%1,%2,%3" : "=f"(tau) : "f"(a), "f"(c), "f"(t));
165 #else
166   tau = fmsub (a, c, t);
167 #endif
168   v = a*d;
169   w = b*c;
170   tau += v + w;	    /* Add in other second-order terms.	 */
171   u = t + tau;
172 
173   /* Construct long double result.  */
174   if (nonfinite (u))
175     return u;
176   z.dval[0] = u;
177   z.dval[1] = (t - u) + tau;
178   return z.ldval;
179 }
180 
181 long double
__gcc_qdiv(double a,double b,double c,double d)182 __gcc_qdiv (double a, double b, double c, double d)
183 {
184   longDblUnion z;
185   double s, sigma, t, tau, u, v, w;
186 
187   t = a / c;                    /* highest order double term */
188 
189   if (unlikely (t == 0)		/* Preserve -0.  */
190       || nonfinite (t))
191     return t;
192 
193   /* Finite nonzero result requires corrections to the highest order
194      term.  These corrections require the low part of c * t to be
195      exactly represented in double.  */
196   if (fabs (a) <= 0x1p-969)
197     {
198       a *= 0x1p106;
199       b *= 0x1p106;
200       c *= 0x1p106;
201       d *= 0x1p106;
202     }
203 
204   s = c * t;                    /* (s,sigma) = c*t exactly.  */
205   w = -(-b + d * t);	/* Written to get fnmsub for speed, but not
206 			   numerically necessary.  */
207 
208   /* Use fused multiply-add to get low part of c * t.	 */
209 #ifndef __NO_FPRS__
210   asm ("fmsub %0,%1,%2,%3" : "=f"(sigma) : "f"(c), "f"(t), "f"(s));
211 #else
212   sigma = fmsub (c, t, s);
213 #endif
214   v = a - s;
215 
216   tau = ((v-sigma)+w)/c;   /* Correction to t.  */
217   u = t + tau;
218 
219   /* Construct long double result.  */
220   if (nonfinite (u))
221     return u;
222   z.dval[0] = u;
223   z.dval[1] = (t - u) + tau;
224   return z.ldval;
225 }
226 
227 #if defined (_SOFT_DOUBLE) && defined (__LONG_DOUBLE_128__)
228 
229 long double __gcc_qneg (double, double);
230 int __gcc_qeq (double, double, double, double);
231 int __gcc_qne (double, double, double, double);
232 int __gcc_qge (double, double, double, double);
233 int __gcc_qle (double, double, double, double);
234 long double __gcc_stoq (float);
235 long double __gcc_dtoq (double);
236 float __gcc_qtos (double, double);
237 double __gcc_qtod (double, double);
238 int __gcc_qtoi (double, double);
239 unsigned int __gcc_qtou (double, double);
240 long double __gcc_itoq (int);
241 long double __gcc_utoq (unsigned int);
242 
243 extern int __eqdf2 (double, double);
244 extern int __ledf2 (double, double);
245 extern int __gedf2 (double, double);
246 
247 /* Negate 'long double' value and return the result.	*/
248 long double
__gcc_qneg(double a,double aa)249 __gcc_qneg (double a, double aa)
250 {
251   longDblUnion x;
252 
253   x.dval[0] = -a;
254   x.dval[1] = -aa;
255   return x.ldval;
256 }
257 
258 /* Compare two 'long double' values for equality.  */
259 int
__gcc_qeq(double a,double aa,double c,double cc)260 __gcc_qeq (double a, double aa, double c, double cc)
261 {
262   if (__eqdf2 (a, c) == 0)
263     return __eqdf2 (aa, cc);
264   return 1;
265 }
266 
267 strong_alias (__gcc_qeq, __gcc_qne);
268 
269 /* Compare two 'long double' values for less than or equal.  */
270 int
__gcc_qle(double a,double aa,double c,double cc)271 __gcc_qle (double a, double aa, double c, double cc)
272 {
273   if (__eqdf2 (a, c) == 0)
274     return __ledf2 (aa, cc);
275   return __ledf2 (a, c);
276 }
277 
278 strong_alias (__gcc_qle, __gcc_qlt);
279 
280 /* Compare two 'long double' values for greater than or equal.  */
281 int
__gcc_qge(double a,double aa,double c,double cc)282 __gcc_qge (double a, double aa, double c, double cc)
283 {
284   if (__eqdf2 (a, c) == 0)
285     return __gedf2 (aa, cc);
286   return __gedf2 (a, c);
287 }
288 
289 strong_alias (__gcc_qge, __gcc_qgt);
290 
291 /* Convert single to long double.  */
292 long double
__gcc_stoq(float a)293 __gcc_stoq (float a)
294 {
295   longDblUnion x;
296 
297   x.dval[0] = (double) a;
298   x.dval[1] = 0.0;
299 
300   return x.ldval;
301 }
302 
303 /* Convert double to long double.  */
304 long double
__gcc_dtoq(double a)305 __gcc_dtoq (double a)
306 {
307   longDblUnion x;
308 
309   x.dval[0] = a;
310   x.dval[1] = 0.0;
311 
312   return x.ldval;
313 }
314 
315 /* Convert long double to single.  */
316 float
__gcc_qtos(double a,double aa)317 __gcc_qtos (double a, double aa __attribute__ ((__unused__)))
318 {
319   return (float) a;
320 }
321 
322 /* Convert long double to double.  */
323 double
__gcc_qtod(double a,double aa)324 __gcc_qtod (double a, double aa __attribute__ ((__unused__)))
325 {
326   return a;
327 }
328 
329 /* Convert long double to int.  */
330 int
__gcc_qtoi(double a,double aa)331 __gcc_qtoi (double a, double aa)
332 {
333   double z = a + aa;
334   return (int) z;
335 }
336 
337 /* Convert long double to unsigned int.  */
338 unsigned int
__gcc_qtou(double a,double aa)339 __gcc_qtou (double a, double aa)
340 {
341   double z = a + aa;
342   return (unsigned int) z;
343 }
344 
345 /* Convert int to long double.  */
346 long double
__gcc_itoq(int a)347 __gcc_itoq (int a)
348 {
349   return __gcc_dtoq ((double) a);
350 }
351 
352 /* Convert unsigned int to long double.  */
353 long double
__gcc_utoq(unsigned int a)354 __gcc_utoq (unsigned int a)
355 {
356   return __gcc_dtoq ((double) a);
357 }
358 
359 #endif
360 
361 #ifdef __NO_FPRS__
362 
363 int __gcc_qunord (double, double, double, double);
364 
365 extern int __eqdf2 (double, double);
366 extern int __unorddf2 (double, double);
367 
368 /* Compare two 'long double' values for unordered.  */
369 int
__gcc_qunord(double a,double aa,double c,double cc)370 __gcc_qunord (double a, double aa, double c, double cc)
371 {
372   if (__eqdf2 (a, c) == 0)
373     return __unorddf2 (aa, cc);
374   return __unorddf2 (a, c);
375 }
376 
377 #include "soft-fp/soft-fp.h"
378 #include "soft-fp/double.h"
379 #include "soft-fp/quad.h"
380 
381 /* Compute floating point multiply-subtract with higher (quad) precision.  */
382 static double
fmsub(double a,double b,double c)383 fmsub (double a, double b, double c)
384 {
385     FP_DECL_EX;
386     FP_DECL_D(A);
387     FP_DECL_D(B);
388     FP_DECL_D(C);
389     FP_DECL_Q(X);
390     FP_DECL_Q(Y);
391     FP_DECL_Q(Z);
392     FP_DECL_Q(U);
393     FP_DECL_Q(V);
394     FP_DECL_D(R);
395     double r;
396     long double u, x, y, z;
397 
398     FP_INIT_ROUNDMODE;
399     FP_UNPACK_RAW_D (A, a);
400     FP_UNPACK_RAW_D (B, b);
401     FP_UNPACK_RAW_D (C, c);
402 
403     /* Extend double to quad.  */
404 #if (2 * _FP_W_TYPE_SIZE) < _FP_FRACBITS_Q
405     FP_EXTEND(Q,D,4,2,X,A);
406     FP_EXTEND(Q,D,4,2,Y,B);
407     FP_EXTEND(Q,D,4,2,Z,C);
408 #else
409     FP_EXTEND(Q,D,2,1,X,A);
410     FP_EXTEND(Q,D,2,1,Y,B);
411     FP_EXTEND(Q,D,2,1,Z,C);
412 #endif
413     FP_PACK_RAW_Q(x,X);
414     FP_PACK_RAW_Q(y,Y);
415     FP_PACK_RAW_Q(z,Z);
416     FP_HANDLE_EXCEPTIONS;
417 
418     /* Multiply.  */
419     FP_INIT_ROUNDMODE;
420     FP_UNPACK_Q(X,x);
421     FP_UNPACK_Q(Y,y);
422     FP_MUL_Q(U,X,Y);
423     FP_PACK_Q(u,U);
424     FP_HANDLE_EXCEPTIONS;
425 
426     /* Subtract.  */
427     FP_INIT_ROUNDMODE;
428     FP_UNPACK_SEMIRAW_Q(U,u);
429     FP_UNPACK_SEMIRAW_Q(Z,z);
430     FP_SUB_Q(V,U,Z);
431 
432     /* Truncate quad to double.  */
433 #if (2 * _FP_W_TYPE_SIZE) < _FP_FRACBITS_Q
434     V_f[3] &= 0x0007ffff;
435     FP_TRUNC(D,Q,2,4,R,V);
436 #else
437     V_f1 &= 0x0007ffffffffffffL;
438     FP_TRUNC(D,Q,1,2,R,V);
439 #endif
440     FP_PACK_SEMIRAW_D(r,R);
441     FP_HANDLE_EXCEPTIONS;
442 
443     return r;
444 }
445 
446 #endif
447 
448 #endif
449