1 | /* |
2 | * Copyright 2012-15 Advanced Micro Devices, Inc. |
3 | * |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
5 | * copy of this software and associated documentation files (the "Software"), |
6 | * to deal in the Software without restriction, including without limitation |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
8 | * and/or sell copies of the Software, and to permit persons to whom the |
9 | * Software is furnished to do so, subject to the following conditions: |
10 | * |
11 | * The above copyright notice and this permission notice shall be included in |
12 | * all copies or substantial portions of the Software. |
13 | * |
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
17 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
18 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
19 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
20 | * OTHER DEALINGS IN THE SOFTWARE. |
21 | * |
22 | * Authors: AMD |
23 | * |
24 | */ |
25 | |
26 | #ifndef __DAL_FIXED31_32_H__ |
27 | #define __DAL_FIXED31_32_H__ |
28 | |
29 | #ifndef LLONG_MAX |
30 | #define LLONG_MAX 9223372036854775807ll |
31 | #endif |
32 | #ifndef LLONG_MIN |
33 | #define LLONG_MIN (-LLONG_MAX - 1ll) |
34 | #endif |
35 | |
36 | #define FIXED31_32_BITS_PER_FRACTIONAL_PART 32 |
37 | #ifndef LLONG_MIN |
38 | #define LLONG_MIN (1LL<<63) |
39 | #endif |
40 | #ifndef LLONG_MAX |
41 | #define LLONG_MAX (-1LL>>1) |
42 | #endif |
43 | |
44 | /* |
45 | * @brief |
46 | * Arithmetic operations on real numbers |
47 | * represented as fixed-point numbers. |
48 | * There are: 1 bit for sign, |
49 | * 31 bit for integer part, |
50 | * 32 bits for fractional part. |
51 | * |
52 | * @note |
53 | * Currently, overflows and underflows are asserted; |
54 | * no special result returned. |
55 | */ |
56 | |
57 | struct fixed31_32 { |
58 | long long value; |
59 | }; |
60 | |
61 | |
62 | /* |
63 | * @brief |
64 | * Useful constants |
65 | */ |
66 | |
67 | static const struct fixed31_32 dc_fixpt_zero = { 0 }; |
68 | static const struct fixed31_32 dc_fixpt_epsilon = { 1LL }; |
69 | static const struct fixed31_32 dc_fixpt_half = { 0x80000000LL }; |
70 | static const struct fixed31_32 dc_fixpt_one = { 0x100000000LL }; |
71 | |
72 | static const struct fixed31_32 dc_fixpt_pi = { 13493037705LL }; |
73 | static const struct fixed31_32 dc_fixpt_two_pi = { 26986075409LL }; |
74 | static const struct fixed31_32 dc_fixpt_e = { 11674931555LL }; |
75 | static const struct fixed31_32 dc_fixpt_ln2 = { 2977044471LL }; |
76 | static const struct fixed31_32 dc_fixpt_ln2_div_2 = { 1488522236LL }; |
77 | |
78 | /* |
79 | * @brief |
80 | * Initialization routines |
81 | */ |
82 | |
83 | /* |
84 | * @brief |
85 | * result = numerator / denominator |
86 | */ |
87 | struct fixed31_32 dc_fixpt_from_fraction(long long numerator, long long denominator); |
88 | |
89 | /* |
90 | * @brief |
91 | * result = arg |
92 | */ |
93 | static inline struct fixed31_32 dc_fixpt_from_int(int arg) |
94 | { |
95 | struct fixed31_32 res; |
96 | |
97 | res.value = (long long) arg << FIXED31_32_BITS_PER_FRACTIONAL_PART; |
98 | |
99 | return res; |
100 | } |
101 | |
102 | /* |
103 | * @brief |
104 | * Unary operators |
105 | */ |
106 | |
107 | /* |
108 | * @brief |
109 | * result = -arg |
110 | */ |
111 | static inline struct fixed31_32 dc_fixpt_neg(struct fixed31_32 arg) |
112 | { |
113 | struct fixed31_32 res; |
114 | |
115 | res.value = -arg.value; |
116 | |
117 | return res; |
118 | } |
119 | |
120 | /* |
121 | * @brief |
122 | * result = abs(arg) := (arg >= 0) ? arg : -arg |
123 | */ |
124 | static inline struct fixed31_32 dc_fixpt_abs(struct fixed31_32 arg) |
125 | { |
126 | if (arg.value < 0) |
127 | return dc_fixpt_neg(arg); |
128 | else |
129 | return arg; |
130 | } |
131 | |
132 | /* |
133 | * @brief |
134 | * Binary relational operators |
135 | */ |
136 | |
137 | /* |
138 | * @brief |
139 | * result = arg1 < arg2 |
140 | */ |
141 | static inline bool dc_fixpt_lt(struct fixed31_32 arg1, struct fixed31_32 arg2) |
142 | { |
143 | return arg1.value < arg2.value; |
144 | } |
145 | |
146 | /* |
147 | * @brief |
148 | * result = arg1 <= arg2 |
149 | */ |
150 | static inline bool dc_fixpt_le(struct fixed31_32 arg1, struct fixed31_32 arg2) |
151 | { |
152 | return arg1.value <= arg2.value; |
153 | } |
154 | |
155 | /* |
156 | * @brief |
157 | * result = arg1 == arg2 |
158 | */ |
159 | static inline bool dc_fixpt_eq(struct fixed31_32 arg1, struct fixed31_32 arg2) |
160 | { |
161 | return arg1.value == arg2.value; |
162 | } |
163 | |
164 | /* |
165 | * @brief |
166 | * result = min(arg1, arg2) := (arg1 <= arg2) ? arg1 : arg2 |
167 | */ |
168 | static inline struct fixed31_32 dc_fixpt_min(struct fixed31_32 arg1, struct fixed31_32 arg2) |
169 | { |
170 | if (arg1.value <= arg2.value) |
171 | return arg1; |
172 | else |
173 | return arg2; |
174 | } |
175 | |
176 | /* |
177 | * @brief |
178 | * result = max(arg1, arg2) := (arg1 <= arg2) ? arg2 : arg1 |
179 | */ |
180 | static inline struct fixed31_32 dc_fixpt_max(struct fixed31_32 arg1, struct fixed31_32 arg2) |
181 | { |
182 | if (arg1.value <= arg2.value) |
183 | return arg2; |
184 | else |
185 | return arg1; |
186 | } |
187 | |
188 | /* |
189 | * @brief |
190 | * | min_value, when arg <= min_value |
191 | * result = | arg, when min_value < arg < max_value |
192 | * | max_value, when arg >= max_value |
193 | */ |
194 | static inline struct fixed31_32 dc_fixpt_clamp( |
195 | struct fixed31_32 arg, |
196 | struct fixed31_32 min_value, |
197 | struct fixed31_32 max_value) |
198 | { |
199 | if (dc_fixpt_le(arg, min_value)) |
200 | return min_value; |
201 | else if (dc_fixpt_le(max_value, arg)) |
202 | return max_value; |
203 | else |
204 | return arg; |
205 | } |
206 | |
207 | /* |
208 | * @brief |
209 | * Binary shift operators |
210 | */ |
211 | |
212 | /* |
213 | * @brief |
214 | * result = arg << shift |
215 | */ |
216 | static inline struct fixed31_32 dc_fixpt_shl(struct fixed31_32 arg, unsigned char shift) |
217 | { |
218 | ASSERT(((arg.value >= 0) && (arg.value <= LLONG_MAX >> shift)) || |
219 | ((arg.value < 0) && (arg.value >= ~(LLONG_MAX >> shift)))); |
220 | |
221 | arg.value = arg.value << shift; |
222 | |
223 | return arg; |
224 | } |
225 | |
226 | /* |
227 | * @brief |
228 | * result = arg >> shift |
229 | */ |
230 | static inline struct fixed31_32 dc_fixpt_shr(struct fixed31_32 arg, unsigned char shift) |
231 | { |
232 | bool negative = arg.value < 0; |
233 | |
234 | if (negative) |
235 | arg.value = -arg.value; |
236 | arg.value = arg.value >> shift; |
237 | if (negative) |
238 | arg.value = -arg.value; |
239 | return arg; |
240 | } |
241 | |
242 | /* |
243 | * @brief |
244 | * Binary additive operators |
245 | */ |
246 | |
247 | /* |
248 | * @brief |
249 | * result = arg1 + arg2 |
250 | */ |
251 | static inline struct fixed31_32 dc_fixpt_add(struct fixed31_32 arg1, struct fixed31_32 arg2) |
252 | { |
253 | struct fixed31_32 res; |
254 | |
255 | ASSERT(((arg1.value >= 0) && (LLONG_MAX - arg1.value >= arg2.value)) || |
256 | ((arg1.value < 0) && (LLONG_MIN - arg1.value <= arg2.value))); |
257 | |
258 | res.value = arg1.value + arg2.value; |
259 | |
260 | return res; |
261 | } |
262 | |
263 | /* |
264 | * @brief |
265 | * result = arg1 + arg2 |
266 | */ |
267 | static inline struct fixed31_32 dc_fixpt_add_int(struct fixed31_32 arg1, int arg2) |
268 | { |
269 | return dc_fixpt_add(arg1, dc_fixpt_from_int(arg2)); |
270 | } |
271 | |
272 | /* |
273 | * @brief |
274 | * result = arg1 - arg2 |
275 | */ |
276 | static inline struct fixed31_32 dc_fixpt_sub(struct fixed31_32 arg1, struct fixed31_32 arg2) |
277 | { |
278 | struct fixed31_32 res; |
279 | |
280 | ASSERT(((arg2.value >= 0) && (LLONG_MIN + arg2.value <= arg1.value)) || |
281 | ((arg2.value < 0) && (LLONG_MAX + arg2.value >= arg1.value))); |
282 | |
283 | res.value = arg1.value - arg2.value; |
284 | |
285 | return res; |
286 | } |
287 | |
288 | /* |
289 | * @brief |
290 | * result = arg1 - arg2 |
291 | */ |
292 | static inline struct fixed31_32 dc_fixpt_sub_int(struct fixed31_32 arg1, int arg2) |
293 | { |
294 | return dc_fixpt_sub(arg1, dc_fixpt_from_int(arg2)); |
295 | } |
296 | |
297 | |
298 | /* |
299 | * @brief |
300 | * Binary multiplicative operators |
301 | */ |
302 | |
303 | /* |
304 | * @brief |
305 | * result = arg1 * arg2 |
306 | */ |
307 | struct fixed31_32 dc_fixpt_mul(struct fixed31_32 arg1, struct fixed31_32 arg2); |
308 | |
309 | |
310 | /* |
311 | * @brief |
312 | * result = arg1 * arg2 |
313 | */ |
314 | static inline struct fixed31_32 dc_fixpt_mul_int(struct fixed31_32 arg1, int arg2) |
315 | { |
316 | return dc_fixpt_mul(arg1, dc_fixpt_from_int(arg2)); |
317 | } |
318 | |
319 | /* |
320 | * @brief |
321 | * result = square(arg) := arg * arg |
322 | */ |
323 | struct fixed31_32 dc_fixpt_sqr(struct fixed31_32 arg); |
324 | |
325 | /* |
326 | * @brief |
327 | * result = arg1 / arg2 |
328 | */ |
329 | static inline struct fixed31_32 dc_fixpt_div_int(struct fixed31_32 arg1, long long arg2) |
330 | { |
331 | return dc_fixpt_from_fraction(arg1.value, dc_fixpt_from_int(arg2).value); |
332 | } |
333 | |
334 | /* |
335 | * @brief |
336 | * result = arg1 / arg2 |
337 | */ |
338 | static inline struct fixed31_32 dc_fixpt_div(struct fixed31_32 arg1, struct fixed31_32 arg2) |
339 | { |
340 | return dc_fixpt_from_fraction(arg1.value, arg2.value); |
341 | } |
342 | |
343 | /* |
344 | * @brief |
345 | * Reciprocal function |
346 | */ |
347 | |
348 | /* |
349 | * @brief |
350 | * result = reciprocal(arg) := 1 / arg |
351 | * |
352 | * @note |
353 | * No special actions taken in case argument is zero. |
354 | */ |
355 | struct fixed31_32 dc_fixpt_recip(struct fixed31_32 arg); |
356 | |
357 | /* |
358 | * @brief |
359 | * Trigonometric functions |
360 | */ |
361 | |
362 | /* |
363 | * @brief |
364 | * result = sinc(arg) := sin(arg) / arg |
365 | * |
366 | * @note |
367 | * Argument specified in radians, |
368 | * internally it's normalized to [-2pi...2pi] range. |
369 | */ |
370 | struct fixed31_32 dc_fixpt_sinc(struct fixed31_32 arg); |
371 | |
372 | /* |
373 | * @brief |
374 | * result = sin(arg) |
375 | * |
376 | * @note |
377 | * Argument specified in radians, |
378 | * internally it's normalized to [-2pi...2pi] range. |
379 | */ |
380 | struct fixed31_32 dc_fixpt_sin(struct fixed31_32 arg); |
381 | |
382 | /* |
383 | * @brief |
384 | * result = cos(arg) |
385 | * |
386 | * @note |
387 | * Argument specified in radians |
388 | * and should be in [-2pi...2pi] range - |
389 | * passing arguments outside that range |
390 | * will cause incorrect result! |
391 | */ |
392 | struct fixed31_32 dc_fixpt_cos(struct fixed31_32 arg); |
393 | |
394 | /* |
395 | * @brief |
396 | * Transcendent functions |
397 | */ |
398 | |
399 | /* |
400 | * @brief |
401 | * result = exp(arg) |
402 | * |
403 | * @note |
404 | * Currently, function is verified for abs(arg) <= 1. |
405 | */ |
406 | struct fixed31_32 dc_fixpt_exp(struct fixed31_32 arg); |
407 | |
408 | /* |
409 | * @brief |
410 | * result = log(arg) |
411 | * |
412 | * @note |
413 | * Currently, abs(arg) should be less than 1. |
414 | * No normalization is done. |
415 | * Currently, no special actions taken |
416 | * in case of invalid argument(s). Take care! |
417 | */ |
418 | struct fixed31_32 dc_fixpt_log(struct fixed31_32 arg); |
419 | |
420 | /* |
421 | * @brief |
422 | * Power function |
423 | */ |
424 | |
425 | /* |
426 | * @brief |
427 | * result = pow(arg1, arg2) |
428 | * |
429 | * @note |
430 | * Currently, abs(arg1) should be less than 1. Take care! |
431 | */ |
432 | static inline struct fixed31_32 dc_fixpt_pow(struct fixed31_32 arg1, struct fixed31_32 arg2) |
433 | { |
434 | return dc_fixpt_exp( |
435 | dc_fixpt_mul( |
436 | dc_fixpt_log(arg1), |
437 | arg2)); |
438 | } |
439 | |
440 | /* |
441 | * @brief |
442 | * Rounding functions |
443 | */ |
444 | |
445 | /* |
446 | * @brief |
447 | * result = floor(arg) := greatest integer lower than or equal to arg |
448 | */ |
449 | static inline int dc_fixpt_floor(struct fixed31_32 arg) |
450 | { |
451 | unsigned long long arg_value = arg.value > 0 ? arg.value : -arg.value; |
452 | |
453 | if (arg.value >= 0) |
454 | return (int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART); |
455 | else |
456 | return -(int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART); |
457 | } |
458 | |
459 | /* |
460 | * @brief |
461 | * result = round(arg) := integer nearest to arg |
462 | */ |
463 | static inline int dc_fixpt_round(struct fixed31_32 arg) |
464 | { |
465 | unsigned long long arg_value = arg.value > 0 ? arg.value : -arg.value; |
466 | |
467 | const long long summand = dc_fixpt_half.value; |
468 | |
469 | ASSERT(LLONG_MAX - (long long)arg_value >= summand); |
470 | |
471 | arg_value += summand; |
472 | |
473 | if (arg.value >= 0) |
474 | return (int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART); |
475 | else |
476 | return -(int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART); |
477 | } |
478 | |
479 | /* |
480 | * @brief |
481 | * result = ceil(arg) := lowest integer greater than or equal to arg |
482 | */ |
483 | static inline int dc_fixpt_ceil(struct fixed31_32 arg) |
484 | { |
485 | unsigned long long arg_value = arg.value > 0 ? arg.value : -arg.value; |
486 | |
487 | const long long summand = dc_fixpt_one.value - |
488 | dc_fixpt_epsilon.value; |
489 | |
490 | ASSERT(LLONG_MAX - (long long)arg_value >= summand); |
491 | |
492 | arg_value += summand; |
493 | |
494 | if (arg.value >= 0) |
495 | return (int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART); |
496 | else |
497 | return -(int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART); |
498 | } |
499 | |
500 | /* the following two function are used in scaler hw programming to convert fixed |
501 | * point value to format 2 bits from integer part and 19 bits from fractional |
502 | * part. The same applies for u0d19, 0 bits from integer part and 19 bits from |
503 | * fractional |
504 | */ |
505 | |
506 | unsigned int dc_fixpt_u3d19(struct fixed31_32 arg); |
507 | |
508 | unsigned int dc_fixpt_u2d19(struct fixed31_32 arg); |
509 | |
510 | unsigned int dc_fixpt_u0d19(struct fixed31_32 arg); |
511 | |
512 | unsigned int dc_fixpt_clamp_u0d14(struct fixed31_32 arg); |
513 | |
514 | unsigned int dc_fixpt_clamp_u0d10(struct fixed31_32 arg); |
515 | |
516 | int dc_fixpt_s4d19(struct fixed31_32 arg); |
517 | |
518 | static inline struct fixed31_32 dc_fixpt_truncate(struct fixed31_32 arg, unsigned int frac_bits) |
519 | { |
520 | bool negative = arg.value < 0; |
521 | |
522 | if (frac_bits >= FIXED31_32_BITS_PER_FRACTIONAL_PART) { |
523 | ASSERT(frac_bits == FIXED31_32_BITS_PER_FRACTIONAL_PART); |
524 | return arg; |
525 | } |
526 | |
527 | if (negative) |
528 | arg.value = -arg.value; |
529 | arg.value &= (~0LL) << (FIXED31_32_BITS_PER_FRACTIONAL_PART - frac_bits); |
530 | if (negative) |
531 | arg.value = -arg.value; |
532 | return arg; |
533 | } |
534 | |
535 | #endif |
536 | |