| The Android Open Source Project | dd7bc33 | 2009-03-03 19:32:55 -0800 | [diff] [blame] | 1 | /* libs/pixelflinger/fixed.cpp | 
|  | 2 | ** | 
|  | 3 | ** Copyright 2006, The Android Open Source Project | 
|  | 4 | ** | 
|  | 5 | ** Licensed under the Apache License, Version 2.0 (the "License"); | 
|  | 6 | ** you may not use this file except in compliance with the License. | 
|  | 7 | ** You may obtain a copy of the License at | 
|  | 8 | ** | 
|  | 9 | **     http://www.apache.org/licenses/LICENSE-2.0 | 
|  | 10 | ** | 
|  | 11 | ** Unless required by applicable law or agreed to in writing, software | 
|  | 12 | ** distributed under the License is distributed on an "AS IS" BASIS, | 
|  | 13 | ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | 
|  | 14 | ** See the License for the specific language governing permissions and | 
|  | 15 | ** limitations under the License. | 
|  | 16 | */ | 
|  | 17 |  | 
|  | 18 | #include <stdio.h> | 
|  | 19 |  | 
|  | 20 | #include <private/pixelflinger/ggl_context.h> | 
|  | 21 | #include <private/pixelflinger/ggl_fixed.h> | 
|  | 22 |  | 
|  | 23 |  | 
|  | 24 | // ------------------------------------------------------------------------ | 
|  | 25 |  | 
|  | 26 | int32_t gglRecipQNormalized(int32_t x, int* exponent) | 
|  | 27 | { | 
|  | 28 | const int32_t s = x>>31; | 
|  | 29 | uint32_t a = s ? -x : x; | 
|  | 30 |  | 
|  | 31 | // the result will overflow, so just set it to the biggest/inf value | 
|  | 32 | if (ggl_unlikely(a <= 2LU)) { | 
|  | 33 | *exponent = 0; | 
|  | 34 | return s ? FIXED_MIN : FIXED_MAX; | 
|  | 35 | } | 
|  | 36 |  | 
|  | 37 | // Newton-Raphson iteration: | 
|  | 38 | // x = r*(2 - a*r) | 
|  | 39 |  | 
|  | 40 | const int32_t lz = gglClz(a); | 
|  | 41 | a <<= lz;  // 0.32 | 
|  | 42 | uint32_t r = a; | 
|  | 43 | // note: if a == 0x80000000, this means x was a power-of-2, in this | 
|  | 44 | // case we don't need to compute anything. We get the reciprocal for | 
|  | 45 | // (almost) free. | 
|  | 46 | if (a != 0x80000000) { | 
|  | 47 | r = (0x2E800 << (30-16)) - (r>>(2-1)); // 2.30, r = 2.90625 - 2*a | 
|  | 48 | // 0.32 + 2.30 = 2.62 -> 2.30 | 
|  | 49 | // 2.30 + 2.30 = 4.60 -> 2.30 | 
|  | 50 | r = (((2LU<<30) - uint32_t((uint64_t(a)*r) >> 32)) * uint64_t(r)) >> 30; | 
|  | 51 | r = (((2LU<<30) - uint32_t((uint64_t(a)*r) >> 32)) * uint64_t(r)) >> 30; | 
|  | 52 | } | 
|  | 53 |  | 
|  | 54 | // shift right 1-bit to make room for the sign bit | 
|  | 55 | *exponent = 30-lz-1; | 
|  | 56 | r >>= 1; | 
|  | 57 | return s ? -r : r; | 
|  | 58 | } | 
|  | 59 |  | 
|  | 60 | int32_t gglRecipQ(GGLfixed x, int q) | 
|  | 61 | { | 
|  | 62 | int shift; | 
|  | 63 | x = gglRecipQNormalized(x, &shift); | 
|  | 64 | shift += 16-q; | 
| Bhanu Chetlapalli | 65026f9 | 2012-01-25 14:45:30 -0800 | [diff] [blame] | 65 | if (shift > 0) | 
|  | 66 | x += 1L << (shift-1);   // rounding | 
| The Android Open Source Project | dd7bc33 | 2009-03-03 19:32:55 -0800 | [diff] [blame] | 67 | x >>= shift; | 
|  | 68 | return x; | 
|  | 69 | } | 
|  | 70 |  | 
|  | 71 | // ------------------------------------------------------------------------ | 
|  | 72 |  | 
| The Android Open Source Project | dd7bc33 | 2009-03-03 19:32:55 -0800 | [diff] [blame] | 73 | static const GGLfixed ggl_sqrt_reciproc_approx_tab[8] = { | 
|  | 74 | // 1/sqrt(x) with x = 1-N/16, N=[8...1] | 
|  | 75 | 0x16A09, 0x15555, 0x143D1, 0x134BF, 0x1279A, 0x11C01, 0x111AC, 0x10865 | 
|  | 76 | }; | 
|  | 77 |  | 
|  | 78 | GGLfixed gglSqrtRecipx(GGLfixed x) | 
|  | 79 | { | 
|  | 80 | if (x == 0)         return FIXED_MAX; | 
|  | 81 | if (x == FIXED_ONE) return x; | 
|  | 82 | const GGLfixed a = x; | 
|  | 83 | const int32_t lz = gglClz(x); | 
|  | 84 | x = ggl_sqrt_reciproc_approx_tab[(a>>(28-lz))&0x7]; | 
|  | 85 | const int32_t exp = lz - 16; | 
|  | 86 | if (exp <= 0)   x >>= -exp>>1; | 
|  | 87 | else            x <<= (exp>>1) + (exp & 1); | 
|  | 88 | if (exp & 1) { | 
|  | 89 | x = gglMulx(x, ggl_sqrt_reciproc_approx_tab[0])>>1; | 
|  | 90 | } | 
|  | 91 | // 2 Newton-Raphson iterations: x = x/2*(3-(a*x)*x) | 
|  | 92 | x = gglMulx((x>>1),(0x30000 - gglMulx(gglMulx(a,x),x))); | 
|  | 93 | x = gglMulx((x>>1),(0x30000 - gglMulx(gglMulx(a,x),x))); | 
|  | 94 | return x; | 
|  | 95 | } | 
|  | 96 |  | 
|  | 97 | GGLfixed gglSqrtx(GGLfixed a) | 
|  | 98 | { | 
|  | 99 | // Compute a full precision square-root (24 bits accuracy) | 
|  | 100 | GGLfixed r = 0; | 
|  | 101 | GGLfixed bit = 0x800000; | 
|  | 102 | int32_t bshift = 15; | 
|  | 103 | do { | 
|  | 104 | GGLfixed temp = bit + (r<<1); | 
|  | 105 | if (bshift >= 8)    temp <<= (bshift-8); | 
|  | 106 | else                temp >>= (8-bshift); | 
|  | 107 | if (a >= temp) { | 
|  | 108 | r += bit; | 
|  | 109 | a -= temp; | 
|  | 110 | } | 
|  | 111 | bshift--; | 
|  | 112 | } while (bit>>=1); | 
|  | 113 | return r; | 
|  | 114 | } | 
|  | 115 |  | 
|  | 116 | // ------------------------------------------------------------------------ | 
|  | 117 |  | 
|  | 118 | static const GGLfixed ggl_log_approx_tab[] = { | 
|  | 119 | // -ln(x)/ln(2) with x = N/16, N=[8...16] | 
|  | 120 | 0xFFFF, 0xd47f, 0xad96, 0x8a62, 0x6a3f, 0x4caf, 0x3151, 0x17d6, 0x0000 | 
|  | 121 | }; | 
|  | 122 |  | 
|  | 123 | static const GGLfixed ggl_alog_approx_tab[] = { // domain [0 - 1.0] | 
|  | 124 | 0xffff, 0xeac0, 0xd744, 0xc567, 0xb504, 0xa5fe, 0x9837, 0x8b95, 0x8000 | 
|  | 125 | }; | 
|  | 126 |  | 
|  | 127 | GGLfixed gglPowx(GGLfixed x, GGLfixed y) | 
|  | 128 | { | 
|  | 129 | // prerequisite: 0 <= x <= 1, and y >=0 | 
|  | 130 |  | 
|  | 131 | // pow(x,y) = 2^(y*log2(x)) | 
|  | 132 | // =  2^(y*log2(x*(2^exp)*(2^-exp)))) | 
|  | 133 | // =  2^(y*(log2(X)-exp)) | 
|  | 134 | // =  2^(log2(X)*y - y*exp) | 
|  | 135 | // =  2^( - (-log2(X)*y + y*exp) ) | 
|  | 136 |  | 
|  | 137 | int32_t exp = gglClz(x) - 16; | 
|  | 138 | GGLfixed f = x << exp; | 
|  | 139 | x = (f & 0x0FFF)<<4; | 
|  | 140 | f = (f >> 12) & 0x7; | 
|  | 141 | GGLfixed p = gglMulAddx( | 
|  | 142 | ggl_log_approx_tab[f+1] - ggl_log_approx_tab[f], x, | 
|  | 143 | ggl_log_approx_tab[f]); | 
|  | 144 | p = gglMulAddx(p, y, y*exp); | 
|  | 145 | exp = gglFixedToIntFloor(p); | 
|  | 146 | if (exp < 31) { | 
|  | 147 | p = gglFracx(p); | 
|  | 148 | x = (p & 0x1FFF)<<3; | 
|  | 149 | p >>= 13; | 
|  | 150 | p = gglMulAddx( | 
|  | 151 | ggl_alog_approx_tab[p+1] - ggl_alog_approx_tab[p], x, | 
|  | 152 | ggl_alog_approx_tab[p]); | 
|  | 153 | p >>= exp; | 
|  | 154 | } else { | 
|  | 155 | p = 0; | 
|  | 156 | } | 
|  | 157 | return p; | 
|  | 158 | // ( powf((a*65536.0f), (b*65536.0f)) ) * 65536.0f; | 
|  | 159 | } | 
|  | 160 |  | 
|  | 161 | // ------------------------------------------------------------------------ | 
|  | 162 |  | 
|  | 163 | int32_t gglDivQ(GGLfixed n, GGLfixed d, int32_t i) | 
|  | 164 | { | 
|  | 165 | //int32_t r =int32_t((int64_t(n)<<i)/d); | 
|  | 166 | const int32_t ds = n^d; | 
|  | 167 | if (n<0) n = -n; | 
|  | 168 | if (d<0) d = -d; | 
|  | 169 | int nd = gglClz(d) - gglClz(n); | 
|  | 170 | i += nd + 1; | 
|  | 171 | if (nd > 0) d <<= nd; | 
|  | 172 | else        n <<= -nd; | 
|  | 173 | uint32_t q = 0; | 
|  | 174 |  | 
|  | 175 | int j = i & 7; | 
|  | 176 | i >>= 3; | 
|  | 177 |  | 
|  | 178 | // gcc deals with the code below pretty well. | 
|  | 179 | // we get 3.75 cycles per bit in the main loop | 
|  | 180 | // and 8 cycles per bit in the termination loop | 
|  | 181 | if (ggl_likely(i)) { | 
|  | 182 | n -= d; | 
|  | 183 | do { | 
|  | 184 | q <<= 8; | 
|  | 185 | if (n>=0)   q |= 128; | 
|  | 186 | else        n += d; | 
|  | 187 | n = n*2 - d; | 
|  | 188 | if (n>=0)   q |= 64; | 
|  | 189 | else        n += d; | 
|  | 190 | n = n*2 - d; | 
|  | 191 | if (n>=0)   q |= 32; | 
|  | 192 | else        n += d; | 
|  | 193 | n = n*2 - d; | 
|  | 194 | if (n>=0)   q |= 16; | 
|  | 195 | else        n += d; | 
|  | 196 | n = n*2 - d; | 
|  | 197 | if (n>=0)   q |= 8; | 
|  | 198 | else        n += d; | 
|  | 199 | n = n*2 - d; | 
|  | 200 | if (n>=0)   q |= 4; | 
|  | 201 | else        n += d; | 
|  | 202 | n = n*2 - d; | 
|  | 203 | if (n>=0)   q |= 2; | 
|  | 204 | else        n += d; | 
|  | 205 | n = n*2 - d; | 
|  | 206 | if (n>=0)   q |= 1; | 
|  | 207 | else        n += d; | 
|  | 208 |  | 
|  | 209 | if (--i == 0) | 
|  | 210 | goto finish; | 
|  | 211 |  | 
|  | 212 | n = n*2 - d; | 
|  | 213 | } while(true); | 
|  | 214 | do { | 
|  | 215 | q <<= 1; | 
|  | 216 | n = n*2 - d; | 
|  | 217 | if (n>=0)   q |= 1; | 
|  | 218 | else        n += d; | 
|  | 219 | finish: ; | 
|  | 220 | } while (j--); | 
|  | 221 | return (ds<0) ? -q : q; | 
|  | 222 | } | 
|  | 223 |  | 
|  | 224 | n -= d; | 
|  | 225 | if (n>=0)   q |= 1; | 
|  | 226 | else        n += d; | 
|  | 227 | j--; | 
|  | 228 | goto finish; | 
|  | 229 | } | 
|  | 230 |  | 
|  | 231 | // ------------------------------------------------------------------------ | 
|  | 232 |  | 
|  | 233 | // assumes that the int32_t values of a, b, and c are all positive | 
|  | 234 | // use when both a and b are larger than c | 
|  | 235 |  | 
|  | 236 | template <typename T> | 
|  | 237 | static inline void swap(T& a, T& b) { | 
|  | 238 | T t(a); | 
|  | 239 | a = b; | 
|  | 240 | b = t; | 
|  | 241 | } | 
|  | 242 |  | 
|  | 243 | static __attribute__((noinline)) | 
|  | 244 | int32_t slow_muldiv(uint32_t a, uint32_t b, uint32_t c) | 
|  | 245 | { | 
|  | 246 | // first we compute a*b as a 64-bit integer | 
|  | 247 | // (GCC generates umull with the code below) | 
|  | 248 | uint64_t ab = uint64_t(a)*b; | 
|  | 249 | uint32_t hi = ab>>32; | 
|  | 250 | uint32_t lo = ab; | 
|  | 251 | uint32_t result; | 
|  | 252 |  | 
|  | 253 | // now perform the division | 
|  | 254 | if (hi >= c) { | 
|  | 255 | overflow: | 
|  | 256 | result = 0x7fffffff;  // basic overflow | 
|  | 257 | } else if (hi == 0) { | 
|  | 258 | result = lo/c;  // note: c can't be 0 | 
|  | 259 | if ((result >> 31) != 0)  // result must fit in 31 bits | 
|  | 260 | goto overflow; | 
|  | 261 | } else { | 
|  | 262 | uint32_t r = hi; | 
|  | 263 | int bits = 31; | 
|  | 264 | result = 0; | 
|  | 265 | do { | 
|  | 266 | r = (r << 1) | (lo >> 31); | 
|  | 267 | lo <<= 1; | 
|  | 268 | result <<= 1; | 
|  | 269 | if (r >= c) { | 
|  | 270 | r -= c; | 
|  | 271 | result |= 1; | 
|  | 272 | } | 
|  | 273 | } while (bits--); | 
|  | 274 | } | 
|  | 275 | return int32_t(result); | 
|  | 276 | } | 
|  | 277 |  | 
|  | 278 | // assumes a >= 0 and c >= b >= 0 | 
|  | 279 | static inline | 
|  | 280 | int32_t quick_muldiv(int32_t a, int32_t b, int32_t c) | 
|  | 281 | { | 
|  | 282 | int32_t r = 0, q = 0, i; | 
|  | 283 | int leading = gglClz(a); | 
|  | 284 | i = 32 - leading; | 
|  | 285 | a <<= leading; | 
|  | 286 | do { | 
|  | 287 | r <<= 1; | 
|  | 288 | if (a < 0) | 
|  | 289 | r += b; | 
|  | 290 | a <<= 1; | 
|  | 291 | q <<= 1; | 
|  | 292 | if (r >= c) { | 
|  | 293 | r -= c; | 
|  | 294 | q++; | 
|  | 295 | } | 
|  | 296 | asm(""::); // gcc generates better code this way | 
|  | 297 | if (r >= c) { | 
|  | 298 | r -= c; | 
|  | 299 | q++; | 
|  | 300 | } | 
|  | 301 | } | 
|  | 302 | while (--i); | 
|  | 303 | return q; | 
|  | 304 | } | 
|  | 305 |  | 
|  | 306 | // this function computes a*b/c with 64-bit intermediate accuracy | 
|  | 307 | // overflows (e.g. division by 0) are handled and return INT_MAX | 
|  | 308 |  | 
|  | 309 | int32_t gglMulDivi(int32_t a, int32_t b, int32_t c) | 
|  | 310 | { | 
|  | 311 | int32_t result; | 
|  | 312 | int32_t sign = a^b^c; | 
|  | 313 |  | 
|  | 314 | if (a < 0) a = -a; | 
|  | 315 | if (b < 0) b = -b; | 
|  | 316 | if (c < 0) c = -c; | 
|  | 317 |  | 
|  | 318 | if (a < b) { | 
|  | 319 | swap(a, b); | 
|  | 320 | } | 
|  | 321 |  | 
|  | 322 | if (b <= c) result = quick_muldiv(a, b, c); | 
|  | 323 | else        result = slow_muldiv((uint32_t)a, (uint32_t)b, (uint32_t)c); | 
|  | 324 |  | 
|  | 325 | if (sign < 0) | 
|  | 326 | result = -result; | 
|  | 327 |  | 
|  | 328 | return result; | 
|  | 329 | } |