| 1 | /* |
| 2 | * Simple IDCT (Alpha optimized) |
| 3 | * |
| 4 | * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at> |
| 5 | * |
| 6 | * based upon some outcommented C code from mpeg2dec (idct_mmx.c |
| 7 | * written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>) |
| 8 | * |
| 9 | * Alpha optimizations by Måns Rullgård <mans@mansr.com> |
| 10 | * and Falk Hueffner <falk@debian.org> |
| 11 | * |
| 12 | * This file is part of FFmpeg. |
| 13 | * |
| 14 | * FFmpeg is free software; you can redistribute it and/or |
| 15 | * modify it under the terms of the GNU Lesser General Public |
| 16 | * License as published by the Free Software Foundation; either |
| 17 | * version 2.1 of the License, or (at your option) any later version. |
| 18 | * |
| 19 | * FFmpeg is distributed in the hope that it will be useful, |
| 20 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 21 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 22 | * Lesser General Public License for more details. |
| 23 | * |
| 24 | * You should have received a copy of the GNU Lesser General Public |
| 25 | * License along with FFmpeg; if not, write to the Free Software |
| 26 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 27 | */ |
| 28 | |
| 29 | #include "idctdsp_alpha.h" |
| 30 | #include "asm.h" |
| 31 | |
| 32 | // cos(i * M_PI / 16) * sqrt(2) * (1 << 14) |
| 33 | // W4 is actually exactly 16384, but using 16383 works around |
| 34 | // accumulating rounding errors for some encoders |
| 35 | #define W1 22725 |
| 36 | #define W2 21407 |
| 37 | #define W3 19266 |
| 38 | #define W4 16383 |
| 39 | #define W5 12873 |
| 40 | #define W6 8867 |
| 41 | #define W7 4520 |
| 42 | #define ROW_SHIFT 11 |
| 43 | #define COL_SHIFT 20 |
| 44 | |
| 45 | /* 0: all entries 0, 1: only first entry nonzero, 2: otherwise */ |
| 46 | static inline int idct_row(int16_t *row) |
| 47 | { |
| 48 | int a0, a1, a2, a3, b0, b1, b2, b3, t; |
| 49 | uint64_t l, r, t2; |
| 50 | l = ldq(row); |
| 51 | r = ldq(row + 4); |
| 52 | |
| 53 | if (l == 0 && r == 0) |
| 54 | return 0; |
| 55 | |
| 56 | a0 = W4 * sextw(l) + (1 << (ROW_SHIFT - 1)); |
| 57 | |
| 58 | if (((l & ~0xffffUL) | r) == 0) { |
| 59 | a0 >>= ROW_SHIFT; |
| 60 | t2 = (uint16_t) a0; |
| 61 | t2 |= t2 << 16; |
| 62 | t2 |= t2 << 32; |
| 63 | |
| 64 | stq(t2, row); |
| 65 | stq(t2, row + 4); |
| 66 | return 1; |
| 67 | } |
| 68 | |
| 69 | a1 = a0; |
| 70 | a2 = a0; |
| 71 | a3 = a0; |
| 72 | |
| 73 | t = extwl(l, 4); /* row[2] */ |
| 74 | if (t != 0) { |
| 75 | t = sextw(t); |
| 76 | a0 += W2 * t; |
| 77 | a1 += W6 * t; |
| 78 | a2 -= W6 * t; |
| 79 | a3 -= W2 * t; |
| 80 | } |
| 81 | |
| 82 | t = extwl(r, 0); /* row[4] */ |
| 83 | if (t != 0) { |
| 84 | t = sextw(t); |
| 85 | a0 += W4 * t; |
| 86 | a1 -= W4 * t; |
| 87 | a2 -= W4 * t; |
| 88 | a3 += W4 * t; |
| 89 | } |
| 90 | |
| 91 | t = extwl(r, 4); /* row[6] */ |
| 92 | if (t != 0) { |
| 93 | t = sextw(t); |
| 94 | a0 += W6 * t; |
| 95 | a1 -= W2 * t; |
| 96 | a2 += W2 * t; |
| 97 | a3 -= W6 * t; |
| 98 | } |
| 99 | |
| 100 | t = extwl(l, 2); /* row[1] */ |
| 101 | if (t != 0) { |
| 102 | t = sextw(t); |
| 103 | b0 = W1 * t; |
| 104 | b1 = W3 * t; |
| 105 | b2 = W5 * t; |
| 106 | b3 = W7 * t; |
| 107 | } else { |
| 108 | b0 = 0; |
| 109 | b1 = 0; |
| 110 | b2 = 0; |
| 111 | b3 = 0; |
| 112 | } |
| 113 | |
| 114 | t = extwl(l, 6); /* row[3] */ |
| 115 | if (t) { |
| 116 | t = sextw(t); |
| 117 | b0 += W3 * t; |
| 118 | b1 -= W7 * t; |
| 119 | b2 -= W1 * t; |
| 120 | b3 -= W5 * t; |
| 121 | } |
| 122 | |
| 123 | |
| 124 | t = extwl(r, 2); /* row[5] */ |
| 125 | if (t) { |
| 126 | t = sextw(t); |
| 127 | b0 += W5 * t; |
| 128 | b1 -= W1 * t; |
| 129 | b2 += W7 * t; |
| 130 | b3 += W3 * t; |
| 131 | } |
| 132 | |
| 133 | t = extwl(r, 6); /* row[7] */ |
| 134 | if (t) { |
| 135 | t = sextw(t); |
| 136 | b0 += W7 * t; |
| 137 | b1 -= W5 * t; |
| 138 | b2 += W3 * t; |
| 139 | b3 -= W1 * t; |
| 140 | } |
| 141 | |
| 142 | row[0] = (a0 + b0) >> ROW_SHIFT; |
| 143 | row[1] = (a1 + b1) >> ROW_SHIFT; |
| 144 | row[2] = (a2 + b2) >> ROW_SHIFT; |
| 145 | row[3] = (a3 + b3) >> ROW_SHIFT; |
| 146 | row[4] = (a3 - b3) >> ROW_SHIFT; |
| 147 | row[5] = (a2 - b2) >> ROW_SHIFT; |
| 148 | row[6] = (a1 - b1) >> ROW_SHIFT; |
| 149 | row[7] = (a0 - b0) >> ROW_SHIFT; |
| 150 | |
| 151 | return 2; |
| 152 | } |
| 153 | |
| 154 | static inline void idct_col(int16_t *col) |
| 155 | { |
| 156 | int a0, a1, a2, a3, b0, b1, b2, b3; |
| 157 | |
| 158 | col[0] += (1 << (COL_SHIFT - 1)) / W4; |
| 159 | |
| 160 | a0 = W4 * col[8 * 0]; |
| 161 | a1 = W4 * col[8 * 0]; |
| 162 | a2 = W4 * col[8 * 0]; |
| 163 | a3 = W4 * col[8 * 0]; |
| 164 | |
| 165 | if (col[8 * 2]) { |
| 166 | a0 += W2 * col[8 * 2]; |
| 167 | a1 += W6 * col[8 * 2]; |
| 168 | a2 -= W6 * col[8 * 2]; |
| 169 | a3 -= W2 * col[8 * 2]; |
| 170 | } |
| 171 | |
| 172 | if (col[8 * 4]) { |
| 173 | a0 += W4 * col[8 * 4]; |
| 174 | a1 -= W4 * col[8 * 4]; |
| 175 | a2 -= W4 * col[8 * 4]; |
| 176 | a3 += W4 * col[8 * 4]; |
| 177 | } |
| 178 | |
| 179 | if (col[8 * 6]) { |
| 180 | a0 += W6 * col[8 * 6]; |
| 181 | a1 -= W2 * col[8 * 6]; |
| 182 | a2 += W2 * col[8 * 6]; |
| 183 | a3 -= W6 * col[8 * 6]; |
| 184 | } |
| 185 | |
| 186 | if (col[8 * 1]) { |
| 187 | b0 = W1 * col[8 * 1]; |
| 188 | b1 = W3 * col[8 * 1]; |
| 189 | b2 = W5 * col[8 * 1]; |
| 190 | b3 = W7 * col[8 * 1]; |
| 191 | } else { |
| 192 | b0 = 0; |
| 193 | b1 = 0; |
| 194 | b2 = 0; |
| 195 | b3 = 0; |
| 196 | } |
| 197 | |
| 198 | if (col[8 * 3]) { |
| 199 | b0 += W3 * col[8 * 3]; |
| 200 | b1 -= W7 * col[8 * 3]; |
| 201 | b2 -= W1 * col[8 * 3]; |
| 202 | b3 -= W5 * col[8 * 3]; |
| 203 | } |
| 204 | |
| 205 | if (col[8 * 5]) { |
| 206 | b0 += W5 * col[8 * 5]; |
| 207 | b1 -= W1 * col[8 * 5]; |
| 208 | b2 += W7 * col[8 * 5]; |
| 209 | b3 += W3 * col[8 * 5]; |
| 210 | } |
| 211 | |
| 212 | if (col[8 * 7]) { |
| 213 | b0 += W7 * col[8 * 7]; |
| 214 | b1 -= W5 * col[8 * 7]; |
| 215 | b2 += W3 * col[8 * 7]; |
| 216 | b3 -= W1 * col[8 * 7]; |
| 217 | } |
| 218 | |
| 219 | col[8 * 0] = (a0 + b0) >> COL_SHIFT; |
| 220 | col[8 * 7] = (a0 - b0) >> COL_SHIFT; |
| 221 | col[8 * 1] = (a1 + b1) >> COL_SHIFT; |
| 222 | col[8 * 6] = (a1 - b1) >> COL_SHIFT; |
| 223 | col[8 * 2] = (a2 + b2) >> COL_SHIFT; |
| 224 | col[8 * 5] = (a2 - b2) >> COL_SHIFT; |
| 225 | col[8 * 3] = (a3 + b3) >> COL_SHIFT; |
| 226 | col[8 * 4] = (a3 - b3) >> COL_SHIFT; |
| 227 | } |
| 228 | |
| 229 | /* If all rows but the first one are zero after row transformation, |
| 230 | all rows will be identical after column transformation. */ |
| 231 | static inline void idct_col2(int16_t *col) |
| 232 | { |
| 233 | int i; |
| 234 | uint64_t l, r; |
| 235 | |
| 236 | for (i = 0; i < 8; ++i) { |
| 237 | int a0 = col[i] + (1 << (COL_SHIFT - 1)) / W4; |
| 238 | |
| 239 | a0 *= W4; |
| 240 | col[i] = a0 >> COL_SHIFT; |
| 241 | } |
| 242 | |
| 243 | l = ldq(col + 0 * 4); r = ldq(col + 1 * 4); |
| 244 | stq(l, col + 2 * 4); stq(r, col + 3 * 4); |
| 245 | stq(l, col + 4 * 4); stq(r, col + 5 * 4); |
| 246 | stq(l, col + 6 * 4); stq(r, col + 7 * 4); |
| 247 | stq(l, col + 8 * 4); stq(r, col + 9 * 4); |
| 248 | stq(l, col + 10 * 4); stq(r, col + 11 * 4); |
| 249 | stq(l, col + 12 * 4); stq(r, col + 13 * 4); |
| 250 | stq(l, col + 14 * 4); stq(r, col + 15 * 4); |
| 251 | } |
| 252 | |
| 253 | void ff_simple_idct_axp(int16_t *block) |
| 254 | { |
| 255 | |
| 256 | int i; |
| 257 | int rowsZero = 1; /* all rows except row 0 zero */ |
| 258 | int rowsConstant = 1; /* all rows consist of a constant value */ |
| 259 | |
| 260 | for (i = 0; i < 8; i++) { |
| 261 | int sparseness = idct_row(block + 8 * i); |
| 262 | |
| 263 | if (i > 0 && sparseness > 0) |
| 264 | rowsZero = 0; |
| 265 | if (sparseness == 2) |
| 266 | rowsConstant = 0; |
| 267 | } |
| 268 | |
| 269 | if (rowsZero) { |
| 270 | idct_col2(block); |
| 271 | } else if (rowsConstant) { |
| 272 | idct_col(block); |
| 273 | for (i = 0; i < 8; i += 2) { |
| 274 | uint64_t v = (uint16_t) block[0]; |
| 275 | uint64_t w = (uint16_t) block[8]; |
| 276 | |
| 277 | v |= v << 16; |
| 278 | w |= w << 16; |
| 279 | v |= v << 32; |
| 280 | w |= w << 32; |
| 281 | stq(v, block + 0 * 4); |
| 282 | stq(v, block + 1 * 4); |
| 283 | stq(w, block + 2 * 4); |
| 284 | stq(w, block + 3 * 4); |
| 285 | block += 4 * 4; |
| 286 | } |
| 287 | } else { |
| 288 | for (i = 0; i < 8; i++) |
| 289 | idct_col(block + i); |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | void ff_simple_idct_put_axp(uint8_t *dest, int line_size, int16_t *block) |
| 294 | { |
| 295 | ff_simple_idct_axp(block); |
| 296 | put_pixels_clamped_axp_p(block, dest, line_size); |
| 297 | } |
| 298 | |
| 299 | void ff_simple_idct_add_axp(uint8_t *dest, int line_size, int16_t *block) |
| 300 | { |
| 301 | ff_simple_idct_axp(block); |
| 302 | add_pixels_clamped_axp_p(block, dest, line_size); |
| 303 | } |