fpmath.h
1 /* 2 * 3 * Copyright (C) 2020 Google, Inc. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. The name of the author may not be used to endorse or promote products 14 * derived from this software without specific prior written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <stdint.h> 30 31 /* 32 * This file implements operations for a simple 32.32 fixed-point math type. 33 * This is intended for speed-critical stuff (e.g. graphics) so there are 34 * intentionally no overflow checks or assertions, and operations are written 35 * to prefer speed over precision (e.g. multiplying by 1 may lose precision). 36 * For best results, only use for applications where 16.16 would fit. 37 */ 38 39 typedef struct { /* wrap in struct to prevent direct access */ 40 int64_t v; 41 } fpmath_t; 42 43 #define FPMATH_SHIFT 32 /* define where to place the decimal point */ 44 45 /* Turn an integer into an fpmath_t. */ 46 static inline fpmath_t fp(int32_t a) 47 { 48 return (fpmath_t){ .v = (int64_t)a << FPMATH_SHIFT }; 49 } 50 51 /* Create an fpmath_t from a fraction. (numerator / denominator) */ 52 static inline fpmath_t fpfrac(int32_t numerator, int32_t denominator) 53 { 54 return (fpmath_t){ .v = ((int64_t)numerator << FPMATH_SHIFT) / denominator }; 55 } 56 57 /* Turn an fpmath_t back into an integer, rounding towards -INF. */ 58 static inline int32_t fpfloor(fpmath_t a) 59 { 60 return a.v >> FPMATH_SHIFT; 61 } 62 63 /* Turn an fpmath_t back into an integer, rounding towards nearest. */ 64 static inline int32_t fpround(fpmath_t a) 65 { 66 return (a.v + ((int64_t)1 << (FPMATH_SHIFT - 1))) >> FPMATH_SHIFT; 67 } 68 69 /* Turn an fpmath_t back into an integer, rounding towards +INF. */ 70 static inline int32_t fpceil(fpmath_t a) 71 { 72 return (a.v + ((int64_t)1 << FPMATH_SHIFT) - 1) >> FPMATH_SHIFT; 73 } 74 75 /* Add two fpmath_t. (a + b) */ 76 static inline fpmath_t fpadd(fpmath_t a, fpmath_t b) 77 { 78 return (fpmath_t){ .v = a.v + b.v }; 79 } 80 81 /* Add an fpmath_t and an integer. (a + b) */ 82 static inline fpmath_t fpaddi(fpmath_t a, int32_t b) 83 { 84 return (fpmath_t){ .v = a.v + ((int64_t)b << FPMATH_SHIFT) }; 85 } 86 87 /* Subtract one fpmath_t from another. (a + b) */ 88 static inline fpmath_t fpsub(fpmath_t a, fpmath_t b) 89 { 90 return (fpmath_t){ .v = a.v - b.v }; 91 } 92 93 /* Subtract an integer from an fpmath_t. (a - b) */ 94 static inline fpmath_t fpsubi(fpmath_t a, int32_t b) 95 { 96 return (fpmath_t){ .v = a.v - ((int64_t)b << FPMATH_SHIFT) }; 97 } 98 99 /* Subtract an fpmath_t from an integer. (a - b) */ 100 static inline fpmath_t fpisub(int32_t a, fpmath_t b) 101 { 102 return (fpmath_t){ .v = ((int64_t)a << FPMATH_SHIFT) - b.v }; 103 } 104 105 /* Multiply two fpmath_t. (a * b) 106 Looses 16 bits fractional precision on each. */ 107 static inline fpmath_t fpmul(fpmath_t a, fpmath_t b) 108 { 109 return (fpmath_t){ .v = (a.v >> (FPMATH_SHIFT/2)) * (b.v >> (FPMATH_SHIFT/2)) }; 110 } 111 112 /* Multiply an fpmath_t and an integer. (a * b) */ 113 static inline fpmath_t fpmuli(fpmath_t a, int32_t b) 114 { 115 return (fpmath_t){ .v = a.v * b }; 116 } 117 118 /* Divide an fpmath_t by another. (a / b) 119 Truncates integral part of a to 16 bits! Careful with this one! */ 120 static inline fpmath_t fpdiv(fpmath_t a, fpmath_t b) 121 { 122 return (fpmath_t){ .v = (a.v << (FPMATH_SHIFT/2)) / (b.v >> (FPMATH_SHIFT/2)) }; 123 } 124 125 /* Divide an fpmath_t by an integer. (a / b) */ 126 static inline fpmath_t fpdivi(fpmath_t a, int32_t b) 127 { 128 return (fpmath_t){ .v = a.v / b }; 129 } 130 131 /* Calculate absolute value of an fpmath_t. (ABS(a)) */ 132 static inline fpmath_t fpabs(fpmath_t a) 133 { 134 return (fpmath_t){ .v = (a.v < 0 ? -a.v : a.v) }; 135 } 136 137 /* Return true iff two fpmath_t are exactly equal. (a == b) 138 Like with floats, you probably don't want to use this most of the time. */ 139 static inline int fpequals(fpmath_t a, fpmath_t b) 140 { 141 return a.v == b.v; 142 } 143 144 /* Return true iff one fpmath_t is less than another. (a < b) */ 145 static inline int fpless(fpmath_t a, fpmath_t b) 146 { 147 return a.v < b.v; 148 } 149 150 /* Return true iff one fpmath_t is more than another. (a > b) */ 151 static inline int fpmore(fpmath_t a, fpmath_t b) 152 { 153 return a.v > b.v; 154 } 155 156 /* Return the smaller of two fpmath_t. (MIN(a, b)) */ 157 static inline fpmath_t fpmin(fpmath_t a, fpmath_t b) 158 { 159 if (a.v < b.v) 160 return a; 161 else 162 return b; 163 } 164 165 /* Return the larger of two fpmath_t. (MAX(a, b)) */ 166 static inline fpmath_t fpmax(fpmath_t a, fpmath_t b) 167 { 168 if (a.v > b.v) 169 return a; 170 else 171 return b; 172 } 173 174 /* Return the constant PI as an fpmath_t. */ 175 static inline fpmath_t fppi(void) 176 { 177 /* Rounded (uint64_t)(M_PI * (1UL << 60)) to nine hex digits. */ 178 return (fpmath_t){ .v = 0x3243f6a89 }; 179 } 180 181 /* 182 * Returns the "one-based" sine of an fpmath_t, meaning the input is interpreted as if the range 183 * 0.0-1.0 corresponded to 0.0-PI/2 for radians. This is mostly here as the base primitives for 184 * the other trig stuff, but it may be useful to use directly if your input value already needs 185 * to be multiplied by some factor of PI and you want to save the instructions (and precision) 186 * for multiplying it in just so that the trig functions can divide it right out again. 187 */ 188 fpmath_t fpsin1(fpmath_t x); 189 190 /* Returns the "one-based" cosine of an fpmath_t (analogous definition to fpsin1()). */ 191 static inline fpmath_t fpcos1(fpmath_t x) 192 { 193 return fpsin1(fpaddi(x, 1)); 194 } 195 196 /* Returns the sine of an fpmath_t interpreted as radians. */ 197 static inline fpmath_t fpsinr(fpmath_t radians) 198 { 199 return fpsin1(fpdiv(radians, fpdivi(fppi(), 2))); 200 } 201 202 /* Returns the sine of an fpmath_t interpreted as degrees. */ 203 static inline fpmath_t fpsind(fpmath_t degrees) 204 { 205 return fpsin1(fpdivi(degrees, 90)); 206 } 207 208 /* Returns the cosine of an fpmath_t interpreted as radians. */ 209 static inline fpmath_t fpcosr(fpmath_t radians) 210 { 211 return fpcos1(fpdiv(radians, fpdivi(fppi(), 2))); 212 } 213 214 /* Returns the cosine of an fpmath_t interpreted as degrees. */ 215 static inline fpmath_t fpcosd(fpmath_t degrees) 216 { 217 return fpcos1(fpdivi(degrees, 90)); 218 } 219 220 /* Returns the tangent of an fpmath_t interpreted as radians. 221 No guard rails, don't call this at the poles or you'll divide by 0! */ 222 static inline fpmath_t fptanr(fpmath_t radians) 223 { 224 fpmath_t one_based = fpdiv(radians, fpdivi(fppi(), 2)); 225 return fpdiv(fpsin1(one_based), fpcos1(one_based)); 226 } 227 228 /* Returns the tangent of an fpmath_t interpreted as degrees. 229 No guard rails, don't call this at the poles or you'll divide by 0! */ 230 static inline fpmath_t fptand(fpmath_t degrees) 231 { 232 fpmath_t one_based = fpdivi(degrees, 90); 233 return fpdiv(fpsin1(one_based), fpcos1(one_based)); 234 }