2 * Copyright 2012-15 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
26 #include "dm_services.h"
27 #include "include/fixed31_32.h"
29 static const struct fixed31_32 dc_fixpt_two_pi = { 26986075409LL };
30 static const struct fixed31_32 dc_fixpt_ln2 = { 2977044471LL };
31 static const struct fixed31_32 dc_fixpt_ln2_div_2 = { 1488522236LL };
33 static inline unsigned long long abs_i64(
37 return (unsigned long long)arg;
39 return (unsigned long long)(-arg);
44 * result = dividend / divisor
45 * *remainder = dividend % divisor
47 static inline unsigned long long complete_integer_division_u64(
48 unsigned long long dividend,
49 unsigned long long divisor,
50 unsigned long long *remainder)
52 unsigned long long result;
56 result = div64_u64_rem(dividend, divisor, remainder);
62 #define FRACTIONAL_PART_MASK \
63 ((1ULL << FIXED31_32_BITS_PER_FRACTIONAL_PART) - 1)
65 #define GET_INTEGER_PART(x) \
66 ((x) >> FIXED31_32_BITS_PER_FRACTIONAL_PART)
68 #define GET_FRACTIONAL_PART(x) \
69 (FRACTIONAL_PART_MASK & (x))
71 struct fixed31_32 dc_fixpt_from_fraction(long long numerator, long long denominator)
73 struct fixed31_32 res;
75 bool arg1_negative = numerator < 0;
76 bool arg2_negative = denominator < 0;
78 unsigned long long arg1_value = arg1_negative ? -numerator : numerator;
79 unsigned long long arg2_value = arg2_negative ? -denominator : denominator;
81 unsigned long long remainder;
83 /* determine integer part */
85 unsigned long long res_value = complete_integer_division_u64(
86 arg1_value, arg2_value, &remainder);
88 ASSERT(res_value <= LONG_MAX);
90 /* determine fractional part */
92 unsigned int i = FIXED31_32_BITS_PER_FRACTIONAL_PART;
99 if (remainder >= arg2_value) {
101 remainder -= arg2_value;
108 unsigned long long summand = (remainder << 1) >= arg2_value;
110 ASSERT(res_value <= LLONG_MAX - summand);
112 res_value += summand;
115 res.value = (long long)res_value;
117 if (arg1_negative ^ arg2_negative)
118 res.value = -res.value;
123 struct fixed31_32 dc_fixpt_mul(struct fixed31_32 arg1, struct fixed31_32 arg2)
125 struct fixed31_32 res;
127 bool arg1_negative = arg1.value < 0;
128 bool arg2_negative = arg2.value < 0;
130 unsigned long long arg1_value = arg1_negative ? -arg1.value : arg1.value;
131 unsigned long long arg2_value = arg2_negative ? -arg2.value : arg2.value;
133 unsigned long long arg1_int = GET_INTEGER_PART(arg1_value);
134 unsigned long long arg2_int = GET_INTEGER_PART(arg2_value);
136 unsigned long long arg1_fra = GET_FRACTIONAL_PART(arg1_value);
137 unsigned long long arg2_fra = GET_FRACTIONAL_PART(arg2_value);
139 unsigned long long tmp;
141 res.value = arg1_int * arg2_int;
143 ASSERT(res.value <= LONG_MAX);
145 res.value <<= FIXED31_32_BITS_PER_FRACTIONAL_PART;
147 tmp = arg1_int * arg2_fra;
149 ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
153 tmp = arg2_int * arg1_fra;
155 ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
159 tmp = arg1_fra * arg2_fra;
161 tmp = (tmp >> FIXED31_32_BITS_PER_FRACTIONAL_PART) +
162 (tmp >= (unsigned long long)dc_fixpt_half.value);
164 ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
168 if (arg1_negative ^ arg2_negative)
169 res.value = -res.value;
174 struct fixed31_32 dc_fixpt_sqr(struct fixed31_32 arg)
176 struct fixed31_32 res;
178 unsigned long long arg_value = abs_i64(arg.value);
180 unsigned long long arg_int = GET_INTEGER_PART(arg_value);
182 unsigned long long arg_fra = GET_FRACTIONAL_PART(arg_value);
184 unsigned long long tmp;
186 res.value = arg_int * arg_int;
188 ASSERT(res.value <= LONG_MAX);
190 res.value <<= FIXED31_32_BITS_PER_FRACTIONAL_PART;
192 tmp = arg_int * arg_fra;
194 ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
198 ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
202 tmp = arg_fra * arg_fra;
204 tmp = (tmp >> FIXED31_32_BITS_PER_FRACTIONAL_PART) +
205 (tmp >= (unsigned long long)dc_fixpt_half.value);
207 ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
214 struct fixed31_32 dc_fixpt_recip(struct fixed31_32 arg)
218 * Good idea to use Newton's method
223 return dc_fixpt_from_fraction(
228 struct fixed31_32 dc_fixpt_sinc(struct fixed31_32 arg)
230 struct fixed31_32 square;
232 struct fixed31_32 res = dc_fixpt_one;
236 struct fixed31_32 arg_norm = arg;
240 dc_fixpt_abs(arg))) {
241 arg_norm = dc_fixpt_sub(
247 dc_fixpt_two_pi.value)));
250 square = dc_fixpt_sqr(arg_norm);
264 if (arg.value != arg_norm.value)
266 dc_fixpt_mul(res, arg_norm),
272 struct fixed31_32 dc_fixpt_sin(struct fixed31_32 arg)
279 struct fixed31_32 dc_fixpt_cos(struct fixed31_32 arg)
281 /* TODO implement argument normalization */
283 const struct fixed31_32 square = dc_fixpt_sqr(arg);
285 struct fixed31_32 res = dc_fixpt_one;
309 * Calculated as Taylor series.
311 static struct fixed31_32 fixed31_32_exp_from_taylor_series(struct fixed31_32 arg)
315 struct fixed31_32 res = dc_fixpt_from_fraction(
318 /* TODO find correct res */
320 ASSERT(dc_fixpt_lt(arg, dc_fixpt_one));
339 struct fixed31_32 dc_fixpt_exp(struct fixed31_32 arg)
344 * exp(x) = exp(r + m * ln(2)) = (1 << m) * exp(r),
345 * where m = round(x / ln(2)), r = x - m * ln(2)
350 dc_fixpt_abs(arg))) {
351 int m = dc_fixpt_round(
356 struct fixed31_32 r = dc_fixpt_sub(
370 fixed31_32_exp_from_taylor_series(r),
373 return dc_fixpt_div_int(
374 fixed31_32_exp_from_taylor_series(r),
376 } else if (arg.value != 0)
377 return fixed31_32_exp_from_taylor_series(arg);
382 struct fixed31_32 dc_fixpt_log(struct fixed31_32 arg)
384 struct fixed31_32 res = dc_fixpt_neg(dc_fixpt_one);
385 /* TODO improve 1st estimation */
387 struct fixed31_32 error;
389 ASSERT(arg.value > 0);
390 /* TODO if arg is negative, return NaN */
391 /* TODO if arg is zero, return -INF */
394 struct fixed31_32 res1 = dc_fixpt_add(
402 error = dc_fixpt_sub(
407 /* TODO determine max_allowed_error based on quality of exp() */
408 } while (abs_i64(error.value) > 100ULL);
414 /* this function is a generic helper to translate fixed point value to
415 * specified integer format that will consist of integer_bits integer part and
416 * fractional_bits fractional part. For example it is used in
417 * dc_fixpt_u2d19 to receive 2 bits integer part and 19 bits fractional
418 * part in 32 bits. It is used in hw programming (scaler)
421 static inline unsigned int ux_dy(
423 unsigned int integer_bits,
424 unsigned int fractional_bits)
426 /* 1. create mask of integer part */
427 unsigned int result = (1 << integer_bits) - 1;
428 /* 2. mask out fractional part */
429 unsigned int fractional_part = FRACTIONAL_PART_MASK & value;
430 /* 3. shrink fixed point integer part to be of integer_bits width*/
431 result &= GET_INTEGER_PART(value);
432 /* 4. make space for fractional part to be filled in after integer */
433 result <<= fractional_bits;
434 /* 5. shrink fixed point fractional part to of fractional_bits width*/
435 fractional_part >>= FIXED31_32_BITS_PER_FRACTIONAL_PART - fractional_bits;
436 /* 6. merge the result */
437 return result | fractional_part;
440 static inline unsigned int clamp_ux_dy(
442 unsigned int integer_bits,
443 unsigned int fractional_bits,
444 unsigned int min_clamp)
446 unsigned int truncated_val = ux_dy(value, integer_bits, fractional_bits);
448 if (value >= (1LL << (integer_bits + FIXED31_32_BITS_PER_FRACTIONAL_PART)))
449 return (1 << (integer_bits + fractional_bits)) - 1;
450 else if (truncated_val > min_clamp)
451 return truncated_val;
456 unsigned int dc_fixpt_u4d19(struct fixed31_32 arg)
458 return ux_dy(arg.value, 4, 19);
461 unsigned int dc_fixpt_u3d19(struct fixed31_32 arg)
463 return ux_dy(arg.value, 3, 19);
466 unsigned int dc_fixpt_u2d19(struct fixed31_32 arg)
468 return ux_dy(arg.value, 2, 19);
471 unsigned int dc_fixpt_u0d19(struct fixed31_32 arg)
473 return ux_dy(arg.value, 0, 19);
476 unsigned int dc_fixpt_clamp_u0d14(struct fixed31_32 arg)
478 return clamp_ux_dy(arg.value, 0, 14, 1);
481 unsigned int dc_fixpt_clamp_u0d10(struct fixed31_32 arg)
483 return clamp_ux_dy(arg.value, 0, 10, 1);
486 int dc_fixpt_s4d19(struct fixed31_32 arg)
489 return -(int)ux_dy(dc_fixpt_abs(arg).value, 4, 19);
491 return ux_dy(arg.value, 4, 19);