xref: /freebsd/contrib/arm-optimized-routines/math/math_config.h (revision 18054d0220cfc8df9c9568c437bd6fbb59d53c3c)
1 /*
2  * Configuration for math routines.
3  *
4  * Copyright (c) 2017-2020, Arm Limited.
5  * SPDX-License-Identifier: MIT
6  */
7 
8 #ifndef _MATH_CONFIG_H
9 #define _MATH_CONFIG_H
10 
11 #include <math.h>
12 #include <stdint.h>
13 
14 #ifndef WANT_ROUNDING
15 /* If defined to 1, return correct results for special cases in non-nearest
16    rounding modes (logf (1.0f) returns 0.0f with FE_DOWNWARD rather than -0.0f).
17    This may be set to 0 if there is no fenv support or if math functions only
18    get called in round to nearest mode.  */
19 # define WANT_ROUNDING 1
20 #endif
21 #ifndef WANT_ERRNO
22 /* If defined to 1, set errno in math functions according to ISO C.  Many math
23    libraries do not set errno, so this is 0 by default.  It may need to be
24    set to 1 if math.h has (math_errhandling & MATH_ERRNO) != 0.  */
25 # define WANT_ERRNO 0
26 #endif
27 #ifndef WANT_ERRNO_UFLOW
28 /* Set errno to ERANGE if result underflows to 0 (in all rounding modes).  */
29 # define WANT_ERRNO_UFLOW (WANT_ROUNDING && WANT_ERRNO)
30 #endif
31 
32 /* Compiler can inline round as a single instruction.  */
33 #ifndef HAVE_FAST_ROUND
34 # if __aarch64__
35 #   define HAVE_FAST_ROUND 1
36 # else
37 #   define HAVE_FAST_ROUND 0
38 # endif
39 #endif
40 
41 /* Compiler can inline lround, but not (long)round(x).  */
42 #ifndef HAVE_FAST_LROUND
43 # if __aarch64__ && (100*__GNUC__ + __GNUC_MINOR__) >= 408 && __NO_MATH_ERRNO__
44 #   define HAVE_FAST_LROUND 1
45 # else
46 #   define HAVE_FAST_LROUND 0
47 # endif
48 #endif
49 
50 /* Compiler can inline fma as a single instruction.  */
51 #ifndef HAVE_FAST_FMA
52 # if defined FP_FAST_FMA || __aarch64__
53 #   define HAVE_FAST_FMA 1
54 # else
55 #   define HAVE_FAST_FMA 0
56 # endif
57 #endif
58 
59 /* Provide *_finite symbols and some of the glibc hidden symbols
60    so libmathlib can be used with binaries compiled against glibc
61    to interpose math functions with both static and dynamic linking.  */
62 #ifndef USE_GLIBC_ABI
63 # if __GNUC__
64 #   define USE_GLIBC_ABI 1
65 # else
66 #   define USE_GLIBC_ABI 0
67 # endif
68 #endif
69 
70 /* Optionally used extensions.  */
71 #ifdef __GNUC__
72 # define HIDDEN __attribute__ ((__visibility__ ("hidden")))
73 # define NOINLINE __attribute__ ((noinline))
74 # define UNUSED __attribute__ ((unused))
75 # define likely(x) __builtin_expect (!!(x), 1)
76 # define unlikely(x) __builtin_expect (x, 0)
77 # if __GNUC__ >= 9
78 #   define attribute_copy(f) __attribute__ ((copy (f)))
79 # else
80 #   define attribute_copy(f)
81 # endif
82 # define strong_alias(f, a) \
83   extern __typeof (f) a __attribute__ ((alias (#f))) attribute_copy (f);
84 # define hidden_alias(f, a) \
85   extern __typeof (f) a __attribute__ ((alias (#f), visibility ("hidden"))) \
86   attribute_copy (f);
87 #else
88 # define HIDDEN
89 # define NOINLINE
90 # define UNUSED
91 # define likely(x) (x)
92 # define unlikely(x) (x)
93 #endif
94 
95 #if HAVE_FAST_ROUND
96 /* When set, the roundtoint and converttoint functions are provided with
97    the semantics documented below.  */
98 # define TOINT_INTRINSICS 1
99 
100 /* Round x to nearest int in all rounding modes, ties have to be rounded
101    consistently with converttoint so the results match.  If the result
102    would be outside of [-2^31, 2^31-1] then the semantics is unspecified.  */
103 static inline double_t
104 roundtoint (double_t x)
105 {
106   return round (x);
107 }
108 
109 /* Convert x to nearest int in all rounding modes, ties have to be rounded
110    consistently with roundtoint.  If the result is not representible in an
111    int32_t then the semantics is unspecified.  */
112 static inline int32_t
113 converttoint (double_t x)
114 {
115 # if HAVE_FAST_LROUND
116   return lround (x);
117 # else
118   return (long) round (x);
119 # endif
120 }
121 #endif
122 
123 static inline uint32_t
124 asuint (float f)
125 {
126   union
127   {
128     float f;
129     uint32_t i;
130   } u = {f};
131   return u.i;
132 }
133 
134 static inline float
135 asfloat (uint32_t i)
136 {
137   union
138   {
139     uint32_t i;
140     float f;
141   } u = {i};
142   return u.f;
143 }
144 
145 static inline uint64_t
146 asuint64 (double f)
147 {
148   union
149   {
150     double f;
151     uint64_t i;
152   } u = {f};
153   return u.i;
154 }
155 
156 static inline double
157 asdouble (uint64_t i)
158 {
159   union
160   {
161     uint64_t i;
162     double f;
163   } u = {i};
164   return u.f;
165 }
166 
167 #ifndef IEEE_754_2008_SNAN
168 # define IEEE_754_2008_SNAN 1
169 #endif
170 static inline int
171 issignalingf_inline (float x)
172 {
173   uint32_t ix = asuint (x);
174   if (!IEEE_754_2008_SNAN)
175     return (ix & 0x7fc00000) == 0x7fc00000;
176   return 2 * (ix ^ 0x00400000) > 2u * 0x7fc00000;
177 }
178 
179 static inline int
180 issignaling_inline (double x)
181 {
182   uint64_t ix = asuint64 (x);
183   if (!IEEE_754_2008_SNAN)
184     return (ix & 0x7ff8000000000000) == 0x7ff8000000000000;
185   return 2 * (ix ^ 0x0008000000000000) > 2 * 0x7ff8000000000000ULL;
186 }
187 
188 #if __aarch64__ && __GNUC__
189 /* Prevent the optimization of a floating-point expression.  */
190 static inline float
191 opt_barrier_float (float x)
192 {
193   __asm__ __volatile__ ("" : "+w" (x));
194   return x;
195 }
196 static inline double
197 opt_barrier_double (double x)
198 {
199   __asm__ __volatile__ ("" : "+w" (x));
200   return x;
201 }
202 /* Force the evaluation of a floating-point expression for its side-effect.  */
203 static inline void
204 force_eval_float (float x)
205 {
206   __asm__ __volatile__ ("" : "+w" (x));
207 }
208 static inline void
209 force_eval_double (double x)
210 {
211   __asm__ __volatile__ ("" : "+w" (x));
212 }
213 #else
214 static inline float
215 opt_barrier_float (float x)
216 {
217   volatile float y = x;
218   return y;
219 }
220 static inline double
221 opt_barrier_double (double x)
222 {
223   volatile double y = x;
224   return y;
225 }
226 static inline void
227 force_eval_float (float x)
228 {
229   volatile float y UNUSED = x;
230 }
231 static inline void
232 force_eval_double (double x)
233 {
234   volatile double y UNUSED = x;
235 }
236 #endif
237 
238 /* Evaluate an expression as the specified type, normally a type
239    cast should be enough, but compilers implement non-standard
240    excess-precision handling, so when FLT_EVAL_METHOD != 0 then
241    these functions may need to be customized.  */
242 static inline float
243 eval_as_float (float x)
244 {
245   return x;
246 }
247 static inline double
248 eval_as_double (double x)
249 {
250   return x;
251 }
252 
253 /* Error handling tail calls for special cases, with a sign argument.
254    The sign of the return value is set if the argument is non-zero.  */
255 
256 /* The result overflows.  */
257 HIDDEN float __math_oflowf (uint32_t);
258 /* The result underflows to 0 in nearest rounding mode.  */
259 HIDDEN float __math_uflowf (uint32_t);
260 /* The result underflows to 0 in some directed rounding mode only.  */
261 HIDDEN float __math_may_uflowf (uint32_t);
262 /* Division by zero.  */
263 HIDDEN float __math_divzerof (uint32_t);
264 /* The result overflows.  */
265 HIDDEN double __math_oflow (uint32_t);
266 /* The result underflows to 0 in nearest rounding mode.  */
267 HIDDEN double __math_uflow (uint32_t);
268 /* The result underflows to 0 in some directed rounding mode only.  */
269 HIDDEN double __math_may_uflow (uint32_t);
270 /* Division by zero.  */
271 HIDDEN double __math_divzero (uint32_t);
272 
273 /* Error handling using input checking.  */
274 
275 /* Invalid input unless it is a quiet NaN.  */
276 HIDDEN float __math_invalidf (float);
277 /* Invalid input unless it is a quiet NaN.  */
278 HIDDEN double __math_invalid (double);
279 
280 /* Error handling using output checking, only for errno setting.  */
281 
282 /* Check if the result overflowed to infinity.  */
283 HIDDEN double __math_check_oflow (double);
284 /* Check if the result underflowed to 0.  */
285 HIDDEN double __math_check_uflow (double);
286 
287 /* Check if the result overflowed to infinity.  */
288 static inline double
289 check_oflow (double x)
290 {
291   return WANT_ERRNO ? __math_check_oflow (x) : x;
292 }
293 
294 /* Check if the result underflowed to 0.  */
295 static inline double
296 check_uflow (double x)
297 {
298   return WANT_ERRNO ? __math_check_uflow (x) : x;
299 }
300 
301 /* Check if the result overflowed to infinity.  */
302 HIDDEN float __math_check_oflowf (float);
303 /* Check if the result underflowed to 0.  */
304 HIDDEN float __math_check_uflowf (float);
305 
306 /* Check if the result overflowed to infinity.  */
307 static inline float
308 check_oflowf (float x)
309 {
310   return WANT_ERRNO ? __math_check_oflowf (x) : x;
311 }
312 
313 /* Check if the result underflowed to 0.  */
314 static inline float
315 check_uflowf (float x)
316 {
317   return WANT_ERRNO ? __math_check_uflowf (x) : x;
318 }
319 
320 /* Shared between expf, exp2f and powf.  */
321 #define EXP2F_TABLE_BITS 5
322 #define EXP2F_POLY_ORDER 3
323 extern const struct exp2f_data
324 {
325   uint64_t tab[1 << EXP2F_TABLE_BITS];
326   double shift_scaled;
327   double poly[EXP2F_POLY_ORDER];
328   double shift;
329   double invln2_scaled;
330   double poly_scaled[EXP2F_POLY_ORDER];
331 } __exp2f_data HIDDEN;
332 
333 #define LOGF_TABLE_BITS 4
334 #define LOGF_POLY_ORDER 4
335 extern const struct logf_data
336 {
337   struct
338   {
339     double invc, logc;
340   } tab[1 << LOGF_TABLE_BITS];
341   double ln2;
342   double poly[LOGF_POLY_ORDER - 1]; /* First order coefficient is 1.  */
343 } __logf_data HIDDEN;
344 
345 #define LOG2F_TABLE_BITS 4
346 #define LOG2F_POLY_ORDER 4
347 extern const struct log2f_data
348 {
349   struct
350   {
351     double invc, logc;
352   } tab[1 << LOG2F_TABLE_BITS];
353   double poly[LOG2F_POLY_ORDER];
354 } __log2f_data HIDDEN;
355 
356 #define POWF_LOG2_TABLE_BITS 4
357 #define POWF_LOG2_POLY_ORDER 5
358 #if TOINT_INTRINSICS
359 # define POWF_SCALE_BITS EXP2F_TABLE_BITS
360 #else
361 # define POWF_SCALE_BITS 0
362 #endif
363 #define POWF_SCALE ((double) (1 << POWF_SCALE_BITS))
364 extern const struct powf_log2_data
365 {
366   struct
367   {
368     double invc, logc;
369   } tab[1 << POWF_LOG2_TABLE_BITS];
370   double poly[POWF_LOG2_POLY_ORDER];
371 } __powf_log2_data HIDDEN;
372 
373 
374 #define EXP_TABLE_BITS 7
375 #define EXP_POLY_ORDER 5
376 /* Use polynomial that is optimized for a wider input range.  This may be
377    needed for good precision in non-nearest rounding and !TOINT_INTRINSICS.  */
378 #define EXP_POLY_WIDE 0
379 /* Use close to nearest rounding toint when !TOINT_INTRINSICS.  This may be
380    needed for good precision in non-nearest rouning and !EXP_POLY_WIDE.  */
381 #define EXP_USE_TOINT_NARROW 0
382 #define EXP2_POLY_ORDER 5
383 #define EXP2_POLY_WIDE 0
384 extern const struct exp_data
385 {
386   double invln2N;
387   double shift;
388   double negln2hiN;
389   double negln2loN;
390   double poly[4]; /* Last four coefficients.  */
391   double exp2_shift;
392   double exp2_poly[EXP2_POLY_ORDER];
393   uint64_t tab[2*(1 << EXP_TABLE_BITS)];
394 } __exp_data HIDDEN;
395 
396 #define LOG_TABLE_BITS 7
397 #define LOG_POLY_ORDER 6
398 #define LOG_POLY1_ORDER 12
399 extern const struct log_data
400 {
401   double ln2hi;
402   double ln2lo;
403   double poly[LOG_POLY_ORDER - 1]; /* First coefficient is 1.  */
404   double poly1[LOG_POLY1_ORDER - 1];
405   struct {double invc, logc;} tab[1 << LOG_TABLE_BITS];
406 #if !HAVE_FAST_FMA
407   struct {double chi, clo;} tab2[1 << LOG_TABLE_BITS];
408 #endif
409 } __log_data HIDDEN;
410 
411 #define LOG2_TABLE_BITS 6
412 #define LOG2_POLY_ORDER 7
413 #define LOG2_POLY1_ORDER 11
414 extern const struct log2_data
415 {
416   double invln2hi;
417   double invln2lo;
418   double poly[LOG2_POLY_ORDER - 1];
419   double poly1[LOG2_POLY1_ORDER - 1];
420   struct {double invc, logc;} tab[1 << LOG2_TABLE_BITS];
421 #if !HAVE_FAST_FMA
422   struct {double chi, clo;} tab2[1 << LOG2_TABLE_BITS];
423 #endif
424 } __log2_data HIDDEN;
425 
426 #define POW_LOG_TABLE_BITS 7
427 #define POW_LOG_POLY_ORDER 8
428 extern const struct pow_log_data
429 {
430   double ln2hi;
431   double ln2lo;
432   double poly[POW_LOG_POLY_ORDER - 1]; /* First coefficient is 1.  */
433   /* Note: the pad field is unused, but allows slightly faster indexing.  */
434   struct {double invc, pad, logc, logctail;} tab[1 << POW_LOG_TABLE_BITS];
435 } __pow_log_data HIDDEN;
436 
437 extern const struct erff_data
438 {
439   float erff_poly_A[6];
440   float erff_poly_B[7];
441 } __erff_data HIDDEN;
442 
443 #define ERF_POLY_A_ORDER 19
444 #define ERF_POLY_A_NCOEFFS 10
445 #define ERFC_POLY_C_NCOEFFS 16
446 #define ERFC_POLY_D_NCOEFFS 18
447 #define ERFC_POLY_E_NCOEFFS 14
448 #define ERFC_POLY_F_NCOEFFS 17
449 extern const struct erf_data
450 {
451   double erf_poly_A[ERF_POLY_A_NCOEFFS];
452   double erf_ratio_N_A[5];
453   double erf_ratio_D_A[5];
454   double erf_ratio_N_B[7];
455   double erf_ratio_D_B[6];
456   double erfc_poly_C[ERFC_POLY_C_NCOEFFS];
457   double erfc_poly_D[ERFC_POLY_D_NCOEFFS];
458   double erfc_poly_E[ERFC_POLY_E_NCOEFFS];
459   double erfc_poly_F[ERFC_POLY_F_NCOEFFS];
460 } __erf_data HIDDEN;
461 
462 #endif
463