00001 /* ffmpeg/libavcodec/ppc/fdct_altivec.c, this file is part of the 00002 * AltiVec optimized library for the FFMPEG Multimedia System 00003 * Copyright (C) 2003 James Klicman <james@klicman.org> 00004 * 00005 * This file is part of FFmpeg. 00006 * 00007 * FFmpeg is free software; you can redistribute it and/or 00008 * modify it under the terms of the GNU Lesser General Public 00009 * License as published by the Free Software Foundation; either 00010 * version 2.1 of the License, or (at your option) any later version. 00011 * 00012 * FFmpeg is distributed in the hope that it will be useful, 00013 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00014 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00015 * Lesser General Public License for more details. 00016 * 00017 * You should have received a copy of the GNU Lesser General Public 00018 * License along with FFmpeg; if not, write to the Free Software 00019 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00020 */ 00021 00022 00023 #include "libavutil/common.h" 00024 #include "libavcodec/dsputil.h" 00025 #include "dsputil_ppc.h" 00026 #include "gcc_fixes.h" 00027 00028 00029 #define vs16(v) ((vector signed short)(v)) 00030 #define vs32(v) ((vector signed int)(v)) 00031 #define vu8(v) ((vector unsigned char)(v)) 00032 #define vu16(v) ((vector unsigned short)(v)) 00033 #define vu32(v) ((vector unsigned int)(v)) 00034 00035 00036 #define C1 0.98078525066375732421875000 /* cos(1*PI/16) */ 00037 #define C2 0.92387950420379638671875000 /* cos(2*PI/16) */ 00038 #define C3 0.83146959543228149414062500 /* cos(3*PI/16) */ 00039 #define C4 0.70710676908493041992187500 /* cos(4*PI/16) */ 00040 #define C5 0.55557024478912353515625000 /* cos(5*PI/16) */ 00041 #define C6 0.38268342614173889160156250 /* cos(6*PI/16) */ 00042 #define C7 0.19509032368659973144531250 /* cos(7*PI/16) */ 00043 #define SQRT_2 1.41421353816986083984375000 /* sqrt(2) */ 00044 00045 00046 #define W0 -(2 * C2) 00047 #define W1 (2 * C6) 00048 #define W2 (SQRT_2 * C6) 00049 #define W3 (SQRT_2 * C3) 00050 #define W4 (SQRT_2 * (-C1 + C3 + C5 - C7)) 00051 #define W5 (SQRT_2 * ( C1 + C3 - C5 + C7)) 00052 #define W6 (SQRT_2 * ( C1 + C3 + C5 - C7)) 00053 #define W7 (SQRT_2 * ( C1 + C3 - C5 - C7)) 00054 #define W8 (SQRT_2 * ( C7 - C3)) 00055 #define W9 (SQRT_2 * (-C1 - C3)) 00056 #define WA (SQRT_2 * (-C3 - C5)) 00057 #define WB (SQRT_2 * ( C5 - C3)) 00058 00059 00060 static vector float fdctconsts[3] = { 00061 { W0, W1, W2, W3 }, 00062 { W4, W5, W6, W7 }, 00063 { W8, W9, WA, WB } 00064 }; 00065 00066 #define LD_W0 vec_splat(cnsts0, 0) 00067 #define LD_W1 vec_splat(cnsts0, 1) 00068 #define LD_W2 vec_splat(cnsts0, 2) 00069 #define LD_W3 vec_splat(cnsts0, 3) 00070 #define LD_W4 vec_splat(cnsts1, 0) 00071 #define LD_W5 vec_splat(cnsts1, 1) 00072 #define LD_W6 vec_splat(cnsts1, 2) 00073 #define LD_W7 vec_splat(cnsts1, 3) 00074 #define LD_W8 vec_splat(cnsts2, 0) 00075 #define LD_W9 vec_splat(cnsts2, 1) 00076 #define LD_WA vec_splat(cnsts2, 2) 00077 #define LD_WB vec_splat(cnsts2, 3) 00078 00079 00080 #define FDCTROW(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */ \ 00081 x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \ 00082 x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \ 00083 x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \ 00084 x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \ 00085 x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \ 00086 x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \ 00087 x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \ 00088 x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \ 00089 \ 00090 b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \ 00091 b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \ 00092 b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \ 00093 b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \ 00094 \ 00095 b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \ 00096 b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \ 00097 b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \ 00098 cnst = LD_W2; \ 00099 b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \ 00100 cnst = LD_W1; \ 00101 b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \ 00102 cnst = LD_W0; \ 00103 b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \ 00104 \ 00105 x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \ 00106 x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \ 00107 x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \ 00108 x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \ 00109 x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \ 00110 cnst = LD_W3; \ 00111 x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \ 00112 \ 00113 cnst = LD_W8; \ 00114 x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \ 00115 cnst = LD_W9; \ 00116 x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \ 00117 cnst = LD_WA; \ 00118 x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \ 00119 cnst = LD_WB; \ 00120 x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \ 00121 \ 00122 cnst = LD_W4; \ 00123 b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \ 00124 cnst = LD_W5; \ 00125 b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \ 00126 cnst = LD_W6; \ 00127 b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \ 00128 cnst = LD_W7; \ 00129 b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \ 00130 \ 00131 b7 = vec_add(b7, x2); /* b7 = b7 + x2; */ \ 00132 b5 = vec_add(b5, x3); /* b5 = b5 + x3; */ \ 00133 b3 = vec_add(b3, x2); /* b3 = b3 + x2; */ \ 00134 b1 = vec_add(b1, x3); /* b1 = b1 + x3; */ \ 00135 /* }}} */ 00136 00137 #define FDCTCOL(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */ \ 00138 x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \ 00139 x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \ 00140 x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \ 00141 x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \ 00142 x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \ 00143 x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \ 00144 x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \ 00145 x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \ 00146 \ 00147 b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \ 00148 b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \ 00149 b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \ 00150 b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \ 00151 \ 00152 b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \ 00153 b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \ 00154 b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \ 00155 cnst = LD_W2; \ 00156 b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \ 00157 cnst = LD_W1; \ 00158 b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \ 00159 cnst = LD_W0; \ 00160 b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \ 00161 \ 00162 x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \ 00163 x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \ 00164 x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \ 00165 x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \ 00166 x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \ 00167 cnst = LD_W3; \ 00168 x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \ 00169 \ 00170 cnst = LD_W8; \ 00171 x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \ 00172 cnst = LD_W9; \ 00173 x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \ 00174 cnst = LD_WA; \ 00175 x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \ 00176 cnst = LD_WB; \ 00177 x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \ 00178 \ 00179 cnst = LD_W4; \ 00180 b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \ 00181 cnst = LD_W5; \ 00182 b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \ 00183 cnst = LD_W6; \ 00184 b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \ 00185 cnst = LD_W7; \ 00186 b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \ 00187 \ 00188 b7 = vec_add(b7, x2); /* b7 += x2; */ \ 00189 b5 = vec_add(b5, x3); /* b5 += x3; */ \ 00190 b3 = vec_add(b3, x2); /* b3 += x2; */ \ 00191 b1 = vec_add(b1, x3); /* b1 += x3; */ \ 00192 /* }}} */ 00193 00194 00195 00196 /* two dimensional discrete cosine transform */ 00197 00198 void fdct_altivec(int16_t *block) 00199 { 00200 POWERPC_PERF_DECLARE(altivec_fdct, 1); 00201 vector signed short *bp; 00202 vector float *cp; 00203 vector float b00, b10, b20, b30, b40, b50, b60, b70; 00204 vector float b01, b11, b21, b31, b41, b51, b61, b71; 00205 vector float mzero, cnst, cnsts0, cnsts1, cnsts2; 00206 vector float x0, x1, x2, x3, x4, x5, x6, x7, x8; 00207 00208 POWERPC_PERF_START_COUNT(altivec_fdct, 1); 00209 00210 00211 /* setup constants {{{ */ 00212 /* mzero = -0.0 */ 00213 mzero = ((vector float)vec_splat_u32(-1)); 00214 mzero = ((vector float)vec_sl(vu32(mzero), vu32(mzero))); 00215 cp = fdctconsts; 00216 cnsts0 = vec_ld(0, cp); cp++; 00217 cnsts1 = vec_ld(0, cp); cp++; 00218 cnsts2 = vec_ld(0, cp); 00219 /* }}} */ 00220 00221 00222 /* 8x8 matrix transpose (vector short[8]) {{{ */ 00223 #define MERGE_S16(hl,a,b) vec_merge##hl(vs16(a), vs16(b)) 00224 00225 bp = (vector signed short*)block; 00226 b00 = ((vector float)vec_ld(0, bp)); 00227 b40 = ((vector float)vec_ld(16*4, bp)); 00228 b01 = ((vector float)MERGE_S16(h, b00, b40)); 00229 b11 = ((vector float)MERGE_S16(l, b00, b40)); 00230 bp++; 00231 b10 = ((vector float)vec_ld(0, bp)); 00232 b50 = ((vector float)vec_ld(16*4, bp)); 00233 b21 = ((vector float)MERGE_S16(h, b10, b50)); 00234 b31 = ((vector float)MERGE_S16(l, b10, b50)); 00235 bp++; 00236 b20 = ((vector float)vec_ld(0, bp)); 00237 b60 = ((vector float)vec_ld(16*4, bp)); 00238 b41 = ((vector float)MERGE_S16(h, b20, b60)); 00239 b51 = ((vector float)MERGE_S16(l, b20, b60)); 00240 bp++; 00241 b30 = ((vector float)vec_ld(0, bp)); 00242 b70 = ((vector float)vec_ld(16*4, bp)); 00243 b61 = ((vector float)MERGE_S16(h, b30, b70)); 00244 b71 = ((vector float)MERGE_S16(l, b30, b70)); 00245 00246 x0 = ((vector float)MERGE_S16(h, b01, b41)); 00247 x1 = ((vector float)MERGE_S16(l, b01, b41)); 00248 x2 = ((vector float)MERGE_S16(h, b11, b51)); 00249 x3 = ((vector float)MERGE_S16(l, b11, b51)); 00250 x4 = ((vector float)MERGE_S16(h, b21, b61)); 00251 x5 = ((vector float)MERGE_S16(l, b21, b61)); 00252 x6 = ((vector float)MERGE_S16(h, b31, b71)); 00253 x7 = ((vector float)MERGE_S16(l, b31, b71)); 00254 00255 b00 = ((vector float)MERGE_S16(h, x0, x4)); 00256 b10 = ((vector float)MERGE_S16(l, x0, x4)); 00257 b20 = ((vector float)MERGE_S16(h, x1, x5)); 00258 b30 = ((vector float)MERGE_S16(l, x1, x5)); 00259 b40 = ((vector float)MERGE_S16(h, x2, x6)); 00260 b50 = ((vector float)MERGE_S16(l, x2, x6)); 00261 b60 = ((vector float)MERGE_S16(h, x3, x7)); 00262 b70 = ((vector float)MERGE_S16(l, x3, x7)); 00263 00264 #undef MERGE_S16 00265 /* }}} */ 00266 00267 00268 /* Some of the initial calculations can be done as vector short before 00269 * conversion to vector float. The following code section takes advantage 00270 * of this. 00271 */ 00272 #if 1 00273 /* fdct rows {{{ */ 00274 x0 = ((vector float)vec_add(vs16(b00), vs16(b70))); 00275 x7 = ((vector float)vec_sub(vs16(b00), vs16(b70))); 00276 x1 = ((vector float)vec_add(vs16(b10), vs16(b60))); 00277 x6 = ((vector float)vec_sub(vs16(b10), vs16(b60))); 00278 x2 = ((vector float)vec_add(vs16(b20), vs16(b50))); 00279 x5 = ((vector float)vec_sub(vs16(b20), vs16(b50))); 00280 x3 = ((vector float)vec_add(vs16(b30), vs16(b40))); 00281 x4 = ((vector float)vec_sub(vs16(b30), vs16(b40))); 00282 00283 b70 = ((vector float)vec_add(vs16(x0), vs16(x3))); 00284 b10 = ((vector float)vec_add(vs16(x1), vs16(x2))); 00285 00286 b00 = ((vector float)vec_add(vs16(b70), vs16(b10))); 00287 b40 = ((vector float)vec_sub(vs16(b70), vs16(b10))); 00288 00289 #define CTF0(n) \ 00290 b##n##1 = ((vector float)vec_unpackl(vs16(b##n##0))); \ 00291 b##n##0 = ((vector float)vec_unpackh(vs16(b##n##0))); \ 00292 b##n##1 = vec_ctf(vs32(b##n##1), 0); \ 00293 b##n##0 = vec_ctf(vs32(b##n##0), 0); 00294 00295 CTF0(0); 00296 CTF0(4); 00297 00298 b20 = ((vector float)vec_sub(vs16(x0), vs16(x3))); 00299 b60 = ((vector float)vec_sub(vs16(x1), vs16(x2))); 00300 00301 CTF0(2); 00302 CTF0(6); 00303 00304 #undef CTF0 00305 00306 x0 = vec_add(b60, b20); 00307 x1 = vec_add(b61, b21); 00308 00309 cnst = LD_W2; 00310 x0 = vec_madd(cnst, x0, mzero); 00311 x1 = vec_madd(cnst, x1, mzero); 00312 cnst = LD_W1; 00313 b20 = vec_madd(cnst, b20, x0); 00314 b21 = vec_madd(cnst, b21, x1); 00315 cnst = LD_W0; 00316 b60 = vec_madd(cnst, b60, x0); 00317 b61 = vec_madd(cnst, b61, x1); 00318 00319 #define CTFX(x,b) \ 00320 b##0 = ((vector float)vec_unpackh(vs16(x))); \ 00321 b##1 = ((vector float)vec_unpackl(vs16(x))); \ 00322 b##0 = vec_ctf(vs32(b##0), 0); \ 00323 b##1 = vec_ctf(vs32(b##1), 0); \ 00324 00325 CTFX(x4, b7); 00326 CTFX(x5, b5); 00327 CTFX(x6, b3); 00328 CTFX(x7, b1); 00329 00330 #undef CTFX 00331 00332 00333 x0 = vec_add(b70, b10); 00334 x1 = vec_add(b50, b30); 00335 x2 = vec_add(b70, b30); 00336 x3 = vec_add(b50, b10); 00337 x8 = vec_add(x2, x3); 00338 cnst = LD_W3; 00339 x8 = vec_madd(cnst, x8, mzero); 00340 00341 cnst = LD_W8; 00342 x0 = vec_madd(cnst, x0, mzero); 00343 cnst = LD_W9; 00344 x1 = vec_madd(cnst, x1, mzero); 00345 cnst = LD_WA; 00346 x2 = vec_madd(cnst, x2, x8); 00347 cnst = LD_WB; 00348 x3 = vec_madd(cnst, x3, x8); 00349 00350 cnst = LD_W4; 00351 b70 = vec_madd(cnst, b70, x0); 00352 cnst = LD_W5; 00353 b50 = vec_madd(cnst, b50, x1); 00354 cnst = LD_W6; 00355 b30 = vec_madd(cnst, b30, x1); 00356 cnst = LD_W7; 00357 b10 = vec_madd(cnst, b10, x0); 00358 00359 b70 = vec_add(b70, x2); 00360 b50 = vec_add(b50, x3); 00361 b30 = vec_add(b30, x2); 00362 b10 = vec_add(b10, x3); 00363 00364 00365 x0 = vec_add(b71, b11); 00366 x1 = vec_add(b51, b31); 00367 x2 = vec_add(b71, b31); 00368 x3 = vec_add(b51, b11); 00369 x8 = vec_add(x2, x3); 00370 cnst = LD_W3; 00371 x8 = vec_madd(cnst, x8, mzero); 00372 00373 cnst = LD_W8; 00374 x0 = vec_madd(cnst, x0, mzero); 00375 cnst = LD_W9; 00376 x1 = vec_madd(cnst, x1, mzero); 00377 cnst = LD_WA; 00378 x2 = vec_madd(cnst, x2, x8); 00379 cnst = LD_WB; 00380 x3 = vec_madd(cnst, x3, x8); 00381 00382 cnst = LD_W4; 00383 b71 = vec_madd(cnst, b71, x0); 00384 cnst = LD_W5; 00385 b51 = vec_madd(cnst, b51, x1); 00386 cnst = LD_W6; 00387 b31 = vec_madd(cnst, b31, x1); 00388 cnst = LD_W7; 00389 b11 = vec_madd(cnst, b11, x0); 00390 00391 b71 = vec_add(b71, x2); 00392 b51 = vec_add(b51, x3); 00393 b31 = vec_add(b31, x2); 00394 b11 = vec_add(b11, x3); 00395 /* }}} */ 00396 #else 00397 /* convert to float {{{ */ 00398 #define CTF(n) \ 00399 vs32(b##n##1) = vec_unpackl(vs16(b##n##0)); \ 00400 vs32(b##n##0) = vec_unpackh(vs16(b##n##0)); \ 00401 b##n##1 = vec_ctf(vs32(b##n##1), 0); \ 00402 b##n##0 = vec_ctf(vs32(b##n##0), 0); \ 00403 00404 CTF(0); 00405 CTF(1); 00406 CTF(2); 00407 CTF(3); 00408 CTF(4); 00409 CTF(5); 00410 CTF(6); 00411 CTF(7); 00412 00413 #undef CTF 00414 /* }}} */ 00415 00416 FDCTROW(b00, b10, b20, b30, b40, b50, b60, b70); 00417 FDCTROW(b01, b11, b21, b31, b41, b51, b61, b71); 00418 #endif 00419 00420 00421 /* 8x8 matrix transpose (vector float[8][2]) {{{ */ 00422 x0 = vec_mergel(b00, b20); 00423 x1 = vec_mergeh(b00, b20); 00424 x2 = vec_mergel(b10, b30); 00425 x3 = vec_mergeh(b10, b30); 00426 00427 b00 = vec_mergeh(x1, x3); 00428 b10 = vec_mergel(x1, x3); 00429 b20 = vec_mergeh(x0, x2); 00430 b30 = vec_mergel(x0, x2); 00431 00432 x4 = vec_mergel(b41, b61); 00433 x5 = vec_mergeh(b41, b61); 00434 x6 = vec_mergel(b51, b71); 00435 x7 = vec_mergeh(b51, b71); 00436 00437 b41 = vec_mergeh(x5, x7); 00438 b51 = vec_mergel(x5, x7); 00439 b61 = vec_mergeh(x4, x6); 00440 b71 = vec_mergel(x4, x6); 00441 00442 x0 = vec_mergel(b01, b21); 00443 x1 = vec_mergeh(b01, b21); 00444 x2 = vec_mergel(b11, b31); 00445 x3 = vec_mergeh(b11, b31); 00446 00447 x4 = vec_mergel(b40, b60); 00448 x5 = vec_mergeh(b40, b60); 00449 x6 = vec_mergel(b50, b70); 00450 x7 = vec_mergeh(b50, b70); 00451 00452 b40 = vec_mergeh(x1, x3); 00453 b50 = vec_mergel(x1, x3); 00454 b60 = vec_mergeh(x0, x2); 00455 b70 = vec_mergel(x0, x2); 00456 00457 b01 = vec_mergeh(x5, x7); 00458 b11 = vec_mergel(x5, x7); 00459 b21 = vec_mergeh(x4, x6); 00460 b31 = vec_mergel(x4, x6); 00461 /* }}} */ 00462 00463 00464 FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70); 00465 FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71); 00466 00467 00468 /* round, convert back to short {{{ */ 00469 #define CTS(n) \ 00470 b##n##0 = vec_round(b##n##0); \ 00471 b##n##1 = vec_round(b##n##1); \ 00472 b##n##0 = ((vector float)vec_cts(b##n##0, 0)); \ 00473 b##n##1 = ((vector float)vec_cts(b##n##1, 0)); \ 00474 b##n##0 = ((vector float)vec_pack(vs32(b##n##0), vs32(b##n##1))); \ 00475 vec_st(vs16(b##n##0), 0, bp); 00476 00477 bp = (vector signed short*)block; 00478 CTS(0); bp++; 00479 CTS(1); bp++; 00480 CTS(2); bp++; 00481 CTS(3); bp++; 00482 CTS(4); bp++; 00483 CTS(5); bp++; 00484 CTS(6); bp++; 00485 CTS(7); 00486 00487 #undef CTS 00488 /* }}} */ 00489 00490 POWERPC_PERF_STOP_COUNT(altivec_fdct, 1); 00491 } 00492 00493 /* vim:set foldmethod=marker foldlevel=0: */