2 * Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #ifndef AVUTIL_SOFTFLOAT_H
22 #define AVUTIL_SOFTFLOAT_H
28 #include "softfloat_tables.h"
34 typedef struct SoftFloat{
39 static av_const SoftFloat av_normalize_sf(SoftFloat a){
42 while((a.mant + 0x20000000U)<0x40000000U){
47 int s=ONE_BITS + 1 - av_log2(a.mant ^ (a.mant<<1));
61 static inline av_const SoftFloat av_normalize1_sf(SoftFloat a){
63 if((int32_t)(a.mant + 0x40000000U) < 0){
67 av_assert2(a.mant < 0x40000000 && a.mant > -0x40000000);
70 int t= a.mant + 0x40000000 < 0;
71 return (SoftFloat){ a.mant>>t, a.exp+t};
73 int t= (a.mant + 0x40000000U)>>31;
74 return (SoftFloat){a.mant>>t, a.exp+t};
79 * @return Will not be more denormalized than a+b. So if either input is
80 * normalized, then the output will not be worse then the other input.
81 * If both are normalized, then the output will be normalized.
83 static inline av_const SoftFloat av_mul_sf(SoftFloat a, SoftFloat b){
85 av_assert2((int32_t)((a.mant * (int64_t)b.mant) >> ONE_BITS) == (a.mant * (int64_t)b.mant) >> ONE_BITS);
86 a.mant = (a.mant * (int64_t)b.mant) >> ONE_BITS;
87 return av_normalize1_sf((SoftFloat){a.mant, a.exp - 1});
91 * b has to be normalized and not zero.
92 * @return Will not be more denormalized than a.
94 static av_const SoftFloat av_div_sf(SoftFloat a, SoftFloat b){
96 a.mant = ((int64_t)a.mant<<(ONE_BITS+1)) / b.mant;
97 return av_normalize1_sf(a);
100 static inline av_const int av_cmp_sf(SoftFloat a, SoftFloat b){
101 int t= a.exp - b.exp;
102 if(t<0) return (a.mant >> (-t)) - b.mant ;
103 else return a.mant - (b.mant >> t);
106 static inline av_const int av_gt_sf(SoftFloat a, SoftFloat b)
108 int t= a.exp - b.exp;
109 if(t<0) return (a.mant >> (-t)) > b.mant ;
110 else return a.mant > (b.mant >> t);
113 static inline av_const SoftFloat av_add_sf(SoftFloat a, SoftFloat b){
114 int t= a.exp - b.exp;
115 if (t <-31) return b;
116 else if (t < 0) return av_normalize_sf(av_normalize1_sf((SoftFloat){ b.mant + (a.mant >> (-t)), b.exp}));
117 else if (t < 32) return av_normalize_sf(av_normalize1_sf((SoftFloat){ a.mant + (b.mant >> t ), a.exp}));
121 static inline av_const SoftFloat av_sub_sf(SoftFloat a, SoftFloat b){
122 return av_add_sf(a, (SoftFloat){ -b.mant, b.exp});
125 //FIXME log, exp, pow
128 * Converts a mantisse and exponent to a SoftFloat
129 * @returns a SoftFloat with value v * 2^frac_bits
131 static inline av_const SoftFloat av_int2sf(int v, int frac_bits){
132 return av_normalize_sf((SoftFloat){v, ONE_BITS + 1 - frac_bits});
136 * Rounding is to -inf.
138 static inline av_const int av_sf2int(SoftFloat v, int frac_bits){
139 v.exp += frac_bits - (ONE_BITS + 1);
140 if(v.exp >= 0) return v.mant << v.exp ;
141 else return v.mant >>(-v.exp);
145 * Rounding-to-nearest used.
147 static av_always_inline SoftFloat av_sqrt_sf(SoftFloat val)
155 tabIndex = (val.mant - 0x20000000) >> 20;
157 rem = val.mant & 0xFFFFF;
158 val.mant = (int)(((int64_t)av_sqrttbl_sf[tabIndex] * (0x100000 - rem) +
159 (int64_t)av_sqrttbl_sf[tabIndex + 1] * rem +
161 val.mant = (int)(((int64_t)av_sqr_exp_multbl_sf[val.exp & 1] * val.mant +
164 if (val.mant < 0x40000000)
169 val.exp = (val.exp >> 1) + 1;
176 * Rounding-to-nearest used.
178 static av_always_inline void av_sincos_sf(int a, int *s, int *c)
185 sign = (idx << 27) >> 31;
186 cv = av_costbl_1_sf[idx & 0xf];
187 cv = (cv ^ sign) - sign;
190 sign = (idx << 27) >> 31;
191 sv = av_costbl_1_sf[idx & 0xf];
192 sv = (sv ^ sign) - sign;
195 ct = av_costbl_2_sf[idx & 0x1f];
196 st = av_sintbl_2_sf[idx & 0x1f];
198 idx = (int)(((int64_t)cv * ct - (int64_t)sv * st + 0x20000000) >> 30);
200 sv = (int)(((int64_t)cv * st + (int64_t)sv * ct + 0x20000000) >> 30);
205 ct = av_costbl_3_sf[idx & 0x1f];
206 st = av_sintbl_3_sf[idx & 0x1f];
208 idx = (int)(((int64_t)cv * ct - (int64_t)sv * st + 0x20000000) >> 30);
210 sv = (int)(((int64_t)cv * st + (int64_t)sv * ct + 0x20000000) >> 30);
215 ct = (int)(((int64_t)av_costbl_4_sf[idx & 0x1f] * (0x800 - (a & 0x7ff)) +
216 (int64_t)av_costbl_4_sf[(idx & 0x1f)+1]*(a & 0x7ff) +
218 st = (int)(((int64_t)av_sintbl_4_sf[idx & 0x1f] * (0x800 - (a & 0x7ff)) +
219 (int64_t)av_sintbl_4_sf[(idx & 0x1f) + 1] * (a & 0x7ff) +
222 *c = (int)(((int64_t)cv * ct + (int64_t)sv * st + 0x20000000) >> 30);
224 *s = (int)(((int64_t)cv * st + (int64_t)sv * ct + 0x20000000) >> 30);
227 #endif /* AVUTIL_SOFTFLOAT_H */