FFmpeg: libavcodec/ppc/fdct_altivec.c Source File

00001 /*
00002  * Copyright (C) 2003  James Klicman <james@klicman.org>
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * FFmpeg is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #include "config.h"
00022 #if HAVE_ALTIVEC_H
00023 #include <altivec.h>
00024 #endif
00025 #include "libavutil/common.h"
00026 #include "libavcodec/dsputil.h"
00027 #include "dsputil_ppc.h"
00028 #include "dsputil_altivec.h"
00029 
00030 #define vs16(v) ((vector signed short)(v))
00031 #define vs32(v) ((vector signed int)(v))
00032 #define vu8(v)  ((vector unsigned char)(v))
00033 #define vu16(v) ((vector unsigned short)(v))
00034 #define vu32(v) ((vector unsigned int)(v))
00035 
00036 
00037 #define C1     0.98078525066375732421875000 /* cos(1*PI/16) */
00038 #define C2     0.92387950420379638671875000 /* cos(2*PI/16) */
00039 #define C3     0.83146959543228149414062500 /* cos(3*PI/16) */
00040 #define C4     0.70710676908493041992187500 /* cos(4*PI/16) */
00041 #define C5     0.55557024478912353515625000 /* cos(5*PI/16) */
00042 #define C6     0.38268342614173889160156250 /* cos(6*PI/16) */
00043 #define C7     0.19509032368659973144531250 /* cos(7*PI/16) */
00044 #define SQRT_2 1.41421353816986083984375000 /* sqrt(2)      */
00045 
00046 
00047 #define W0 -(2 * C2)
00048 #define W1 (2 * C6)
00049 #define W2 (SQRT_2 * C6)
00050 #define W3 (SQRT_2 * C3)
00051 #define W4 (SQRT_2 * (-C1 + C3 + C5 - C7))
00052 #define W5 (SQRT_2 * ( C1 + C3 - C5 + C7))
00053 #define W6 (SQRT_2 * ( C1 + C3 + C5 - C7))
00054 #define W7 (SQRT_2 * ( C1 + C3 - C5 - C7))
00055 #define W8 (SQRT_2 * ( C7 - C3))
00056 #define W9 (SQRT_2 * (-C1 - C3))
00057 #define WA (SQRT_2 * (-C3 - C5))
00058 #define WB (SQRT_2 * ( C5 - C3))
00059 
00060 
00061 static vector float fdctconsts[3] = {
00062     { W0, W1, W2, W3 },
00063     { W4, W5, W6, W7 },
00064     { W8, W9, WA, WB }
00065 };
00066 
00067 #define LD_W0 vec_splat(cnsts0, 0)
00068 #define LD_W1 vec_splat(cnsts0, 1)
00069 #define LD_W2 vec_splat(cnsts0, 2)
00070 #define LD_W3 vec_splat(cnsts0, 3)
00071 #define LD_W4 vec_splat(cnsts1, 0)
00072 #define LD_W5 vec_splat(cnsts1, 1)
00073 #define LD_W6 vec_splat(cnsts1, 2)
00074 #define LD_W7 vec_splat(cnsts1, 3)
00075 #define LD_W8 vec_splat(cnsts2, 0)
00076 #define LD_W9 vec_splat(cnsts2, 1)
00077 #define LD_WA vec_splat(cnsts2, 2)
00078 #define LD_WB vec_splat(cnsts2, 3)
00079 
00080 
00081 #define FDCTROW(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */                  \
00082     x0 = vec_add(b0, b7);               /* x0 = b0 + b7; */         \
00083     x7 = vec_sub(b0, b7);               /* x7 = b0 - b7; */         \
00084     x1 = vec_add(b1, b6);               /* x1 = b1 + b6; */         \
00085     x6 = vec_sub(b1, b6);               /* x6 = b1 - b6; */         \
00086     x2 = vec_add(b2, b5);               /* x2 = b2 + b5; */         \
00087     x5 = vec_sub(b2, b5);               /* x5 = b2 - b5; */         \
00088     x3 = vec_add(b3, b4);               /* x3 = b3 + b4; */         \
00089     x4 = vec_sub(b3, b4);               /* x4 = b3 - b4; */         \
00090                                                                     \
00091     b7 = vec_add(x0, x3);               /* b7 = x0 + x3; */         \
00092     b1 = vec_add(x1, x2);               /* b1 = x1 + x2; */         \
00093     b0 = vec_add(b7, b1);               /* b0 = b7 + b1; */         \
00094     b4 = vec_sub(b7, b1);               /* b4 = b7 - b1; */         \
00095                                                                     \
00096     b2 = vec_sub(x0, x3);               /* b2 = x0 - x3; */         \
00097     b6 = vec_sub(x1, x2);               /* b6 = x1 - x2; */         \
00098     b5 = vec_add(b6, b2);               /* b5 = b6 + b2; */         \
00099     cnst = LD_W2;                                                   \
00100     b5 = vec_madd(cnst, b5, mzero);     /* b5 = b5 * W2; */         \
00101     cnst = LD_W1;                                                   \
00102     b2 = vec_madd(cnst, b2, b5);        /* b2 = b5 + b2 * W1; */    \
00103     cnst = LD_W0;                                                   \
00104     b6 = vec_madd(cnst, b6, b5);        /* b6 = b5 + b6 * W0; */    \
00105                                                                     \
00106     x0 = vec_add(x4, x7);               /* x0 = x4 + x7; */         \
00107     x1 = vec_add(x5, x6);               /* x1 = x5 + x6; */         \
00108     x2 = vec_add(x4, x6);               /* x2 = x4 + x6; */         \
00109     x3 = vec_add(x5, x7);               /* x3 = x5 + x7; */         \
00110     x8 = vec_add(x2, x3);               /* x8 = x2 + x3; */         \
00111     cnst = LD_W3;                                                   \
00112     x8 = vec_madd(cnst, x8, mzero);     /* x8 = x8 * W3; */         \
00113                                                                     \
00114     cnst = LD_W8;                                                   \
00115     x0 = vec_madd(cnst, x0, mzero);     /* x0 *= W8; */             \
00116     cnst = LD_W9;                                                   \
00117     x1 = vec_madd(cnst, x1, mzero);     /* x1 *= W9; */             \
00118     cnst = LD_WA;                                                   \
00119     x2 = vec_madd(cnst, x2, x8);        /* x2 = x2 * WA + x8; */    \
00120     cnst = LD_WB;                                                   \
00121     x3 = vec_madd(cnst, x3, x8);        /* x3 = x3 * WB + x8; */    \
00122                                                                     \
00123     cnst = LD_W4;                                                   \
00124     b7 = vec_madd(cnst, x4, x0);        /* b7 = x4 * W4 + x0; */    \
00125     cnst = LD_W5;                                                   \
00126     b5 = vec_madd(cnst, x5, x1);        /* b5 = x5 * W5 + x1; */    \
00127     cnst = LD_W6;                                                   \
00128     b3 = vec_madd(cnst, x6, x1);        /* b3 = x6 * W6 + x1; */    \
00129     cnst = LD_W7;                                                   \
00130     b1 = vec_madd(cnst, x7, x0);        /* b1 = x7 * W7 + x0; */    \
00131                                                                     \
00132     b7 = vec_add(b7, x2);               /* b7 = b7 + x2; */         \
00133     b5 = vec_add(b5, x3);               /* b5 = b5 + x3; */         \
00134     b3 = vec_add(b3, x2);               /* b3 = b3 + x2; */         \
00135     b1 = vec_add(b1, x3);               /* b1 = b1 + x3; */         \
00136     /* }}} */
00137 
00138 #define FDCTCOL(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */                  \
00139     x0 = vec_add(b0, b7);               /* x0 = b0 + b7; */         \
00140     x7 = vec_sub(b0, b7);               /* x7 = b0 - b7; */         \
00141     x1 = vec_add(b1, b6);               /* x1 = b1 + b6; */         \
00142     x6 = vec_sub(b1, b6);               /* x6 = b1 - b6; */         \
00143     x2 = vec_add(b2, b5);               /* x2 = b2 + b5; */         \
00144     x5 = vec_sub(b2, b5);               /* x5 = b2 - b5; */         \
00145     x3 = vec_add(b3, b4);               /* x3 = b3 + b4; */         \
00146     x4 = vec_sub(b3, b4);               /* x4 = b3 - b4; */         \
00147                                                                     \
00148     b7 = vec_add(x0, x3);               /* b7 = x0 + x3; */         \
00149     b1 = vec_add(x1, x2);               /* b1 = x1 + x2; */         \
00150     b0 = vec_add(b7, b1);               /* b0 = b7 + b1; */         \
00151     b4 = vec_sub(b7, b1);               /* b4 = b7 - b1; */         \
00152                                                                     \
00153     b2 = vec_sub(x0, x3);               /* b2 = x0 - x3; */         \
00154     b6 = vec_sub(x1, x2);               /* b6 = x1 - x2; */         \
00155     b5 = vec_add(b6, b2);               /* b5 = b6 + b2; */         \
00156     cnst = LD_W2;                                                   \
00157     b5 = vec_madd(cnst, b5, mzero);     /* b5 = b5 * W2; */         \
00158     cnst = LD_W1;                                                   \
00159     b2 = vec_madd(cnst, b2, b5);        /* b2 = b5 + b2 * W1; */    \
00160     cnst = LD_W0;                                                   \
00161     b6 = vec_madd(cnst, b6, b5);        /* b6 = b5 + b6 * W0; */    \
00162                                                                     \
00163     x0 = vec_add(x4, x7);               /* x0 = x4 + x7; */         \
00164     x1 = vec_add(x5, x6);               /* x1 = x5 + x6; */         \
00165     x2 = vec_add(x4, x6);               /* x2 = x4 + x6; */         \
00166     x3 = vec_add(x5, x7);               /* x3 = x5 + x7; */         \
00167     x8 = vec_add(x2, x3);               /* x8 = x2 + x3; */         \
00168     cnst = LD_W3;                                                   \
00169     x8 = vec_madd(cnst, x8, mzero);     /* x8 = x8 * W3; */         \
00170                                                                     \
00171     cnst = LD_W8;                                                   \
00172     x0 = vec_madd(cnst, x0, mzero);     /* x0 *= W8; */             \
00173     cnst = LD_W9;                                                   \
00174     x1 = vec_madd(cnst, x1, mzero);     /* x1 *= W9; */             \
00175     cnst = LD_WA;                                                   \
00176     x2 = vec_madd(cnst, x2, x8);        /* x2 = x2 * WA + x8; */    \
00177     cnst = LD_WB;                                                   \
00178     x3 = vec_madd(cnst, x3, x8);        /* x3 = x3 * WB + x8; */    \
00179                                                                     \
00180     cnst = LD_W4;                                                   \
00181     b7 = vec_madd(cnst, x4, x0);        /* b7 = x4 * W4 + x0; */    \
00182     cnst = LD_W5;                                                   \
00183     b5 = vec_madd(cnst, x5, x1);        /* b5 = x5 * W5 + x1; */    \
00184     cnst = LD_W6;                                                   \
00185     b3 = vec_madd(cnst, x6, x1);        /* b3 = x6 * W6 + x1; */    \
00186     cnst = LD_W7;                                                   \
00187     b1 = vec_madd(cnst, x7, x0);        /* b1 = x7 * W7 + x0; */    \
00188                                                                     \
00189     b7 = vec_add(b7, x2);               /* b7 += x2; */             \
00190     b5 = vec_add(b5, x3);               /* b5 += x3; */             \
00191     b3 = vec_add(b3, x2);               /* b3 += x2; */             \
00192     b1 = vec_add(b1, x3);               /* b1 += x3; */             \
00193     /* }}} */
00194 
00195 
00196 
00197 /* two dimensional discrete cosine transform */
00198 
00199 void fdct_altivec(int16_t *block)
00200 {
00201 POWERPC_PERF_DECLARE(altivec_fdct, 1);
00202     vector signed short *bp;
00203     vector float *cp;
00204     vector float b00, b10, b20, b30, b40, b50, b60, b70;
00205     vector float b01, b11, b21, b31, b41, b51, b61, b71;
00206     vector float mzero, cnst, cnsts0, cnsts1, cnsts2;
00207     vector float x0, x1, x2, x3, x4, x5, x6, x7, x8;
00208 
00209     POWERPC_PERF_START_COUNT(altivec_fdct, 1);
00210 
00211 
00212     /* setup constants {{{ */
00213     /* mzero = -0.0 */
00214     mzero = ((vector float)vec_splat_u32(-1));
00215     mzero = ((vector float)vec_sl(vu32(mzero), vu32(mzero)));
00216     cp = fdctconsts;
00217     cnsts0 = vec_ld(0, cp); cp++;
00218     cnsts1 = vec_ld(0, cp); cp++;
00219     cnsts2 = vec_ld(0, cp);
00220     /* }}} */
00221 
00222 
00223     /* 8x8 matrix transpose (vector short[8]) {{{ */
00224 #define MERGE_S16(hl,a,b) vec_merge##hl(vs16(a), vs16(b))
00225 
00226     bp = (vector signed short*)block;
00227     b00 = ((vector float)vec_ld(0,    bp));
00228     b40 = ((vector float)vec_ld(16*4, bp));
00229     b01 = ((vector float)MERGE_S16(h, b00, b40));
00230     b11 = ((vector float)MERGE_S16(l, b00, b40));
00231     bp++;
00232     b10 = ((vector float)vec_ld(0,    bp));
00233     b50 = ((vector float)vec_ld(16*4, bp));
00234     b21 = ((vector float)MERGE_S16(h, b10, b50));
00235     b31 = ((vector float)MERGE_S16(l, b10, b50));
00236     bp++;
00237     b20 = ((vector float)vec_ld(0,    bp));
00238     b60 = ((vector float)vec_ld(16*4, bp));
00239     b41 = ((vector float)MERGE_S16(h, b20, b60));
00240     b51 = ((vector float)MERGE_S16(l, b20, b60));
00241     bp++;
00242     b30 = ((vector float)vec_ld(0,    bp));
00243     b70 = ((vector float)vec_ld(16*4, bp));
00244     b61 = ((vector float)MERGE_S16(h, b30, b70));
00245     b71 = ((vector float)MERGE_S16(l, b30, b70));
00246 
00247     x0 = ((vector float)MERGE_S16(h, b01, b41));
00248     x1 = ((vector float)MERGE_S16(l, b01, b41));
00249     x2 = ((vector float)MERGE_S16(h, b11, b51));
00250     x3 = ((vector float)MERGE_S16(l, b11, b51));
00251     x4 = ((vector float)MERGE_S16(h, b21, b61));
00252     x5 = ((vector float)MERGE_S16(l, b21, b61));
00253     x6 = ((vector float)MERGE_S16(h, b31, b71));
00254     x7 = ((vector float)MERGE_S16(l, b31, b71));
00255 
00256     b00 = ((vector float)MERGE_S16(h, x0, x4));
00257     b10 = ((vector float)MERGE_S16(l, x0, x4));
00258     b20 = ((vector float)MERGE_S16(h, x1, x5));
00259     b30 = ((vector float)MERGE_S16(l, x1, x5));
00260     b40 = ((vector float)MERGE_S16(h, x2, x6));
00261     b50 = ((vector float)MERGE_S16(l, x2, x6));
00262     b60 = ((vector float)MERGE_S16(h, x3, x7));
00263     b70 = ((vector float)MERGE_S16(l, x3, x7));
00264 
00265 #undef MERGE_S16
00266     /* }}} */
00267 
00268 
00269 /* Some of the initial calculations can be done as vector short before
00270  * conversion to vector float.  The following code section takes advantage
00271  * of this.
00272  */
00273 #if 1
00274     /* fdct rows {{{ */
00275     x0 = ((vector float)vec_add(vs16(b00), vs16(b70)));
00276     x7 = ((vector float)vec_sub(vs16(b00), vs16(b70)));
00277     x1 = ((vector float)vec_add(vs16(b10), vs16(b60)));
00278     x6 = ((vector float)vec_sub(vs16(b10), vs16(b60)));
00279     x2 = ((vector float)vec_add(vs16(b20), vs16(b50)));
00280     x5 = ((vector float)vec_sub(vs16(b20), vs16(b50)));
00281     x3 = ((vector float)vec_add(vs16(b30), vs16(b40)));
00282     x4 = ((vector float)vec_sub(vs16(b30), vs16(b40)));
00283 
00284     b70 = ((vector float)vec_add(vs16(x0), vs16(x3)));
00285     b10 = ((vector float)vec_add(vs16(x1), vs16(x2)));
00286 
00287     b00 = ((vector float)vec_add(vs16(b70), vs16(b10)));
00288     b40 = ((vector float)vec_sub(vs16(b70), vs16(b10)));
00289 
00290 #define CTF0(n) \
00291     b##n##1 = ((vector float)vec_unpackl(vs16(b##n##0))); \
00292     b##n##0 = ((vector float)vec_unpackh(vs16(b##n##0))); \
00293     b##n##1 = vec_ctf(vs32(b##n##1), 0); \
00294     b##n##0 = vec_ctf(vs32(b##n##0), 0);
00295 
00296     CTF0(0);
00297     CTF0(4);
00298 
00299     b20 = ((vector float)vec_sub(vs16(x0), vs16(x3)));
00300     b60 = ((vector float)vec_sub(vs16(x1), vs16(x2)));
00301 
00302     CTF0(2);
00303     CTF0(6);
00304 
00305 #undef CTF0
00306 
00307     x0 = vec_add(b60, b20);
00308     x1 = vec_add(b61, b21);
00309 
00310     cnst = LD_W2;
00311     x0 = vec_madd(cnst, x0, mzero);
00312     x1 = vec_madd(cnst, x1, mzero);
00313     cnst = LD_W1;
00314     b20 = vec_madd(cnst, b20, x0);
00315     b21 = vec_madd(cnst, b21, x1);
00316     cnst = LD_W0;
00317     b60 = vec_madd(cnst, b60, x0);
00318     b61 = vec_madd(cnst, b61, x1);
00319 
00320 #define CTFX(x,b) \
00321     b##0 = ((vector float)vec_unpackh(vs16(x))); \
00322     b##1 = ((vector float)vec_unpackl(vs16(x))); \
00323     b##0 = vec_ctf(vs32(b##0), 0); \
00324     b##1 = vec_ctf(vs32(b##1), 0); \
00325 
00326     CTFX(x4, b7);
00327     CTFX(x5, b5);
00328     CTFX(x6, b3);
00329     CTFX(x7, b1);
00330 
00331 #undef CTFX
00332 
00333 
00334     x0 = vec_add(b70, b10);
00335     x1 = vec_add(b50, b30);
00336     x2 = vec_add(b70, b30);
00337     x3 = vec_add(b50, b10);
00338     x8 = vec_add(x2, x3);
00339     cnst = LD_W3;
00340     x8 = vec_madd(cnst, x8, mzero);
00341 
00342     cnst = LD_W8;
00343     x0 = vec_madd(cnst, x0, mzero);
00344     cnst = LD_W9;
00345     x1 = vec_madd(cnst, x1, mzero);
00346     cnst = LD_WA;
00347     x2 = vec_madd(cnst, x2, x8);
00348     cnst = LD_WB;
00349     x3 = vec_madd(cnst, x3, x8);
00350 
00351     cnst = LD_W4;
00352     b70 = vec_madd(cnst, b70, x0);
00353     cnst = LD_W5;
00354     b50 = vec_madd(cnst, b50, x1);
00355     cnst = LD_W6;
00356     b30 = vec_madd(cnst, b30, x1);
00357     cnst = LD_W7;
00358     b10 = vec_madd(cnst, b10, x0);
00359 
00360     b70 = vec_add(b70, x2);
00361     b50 = vec_add(b50, x3);
00362     b30 = vec_add(b30, x2);
00363     b10 = vec_add(b10, x3);
00364 
00365 
00366     x0 = vec_add(b71, b11);
00367     x1 = vec_add(b51, b31);
00368     x2 = vec_add(b71, b31);
00369     x3 = vec_add(b51, b11);
00370     x8 = vec_add(x2, x3);
00371     cnst = LD_W3;
00372     x8 = vec_madd(cnst, x8, mzero);
00373 
00374     cnst = LD_W8;
00375     x0 = vec_madd(cnst, x0, mzero);
00376     cnst = LD_W9;
00377     x1 = vec_madd(cnst, x1, mzero);
00378     cnst = LD_WA;
00379     x2 = vec_madd(cnst, x2, x8);
00380     cnst = LD_WB;
00381     x3 = vec_madd(cnst, x3, x8);
00382 
00383     cnst = LD_W4;
00384     b71 = vec_madd(cnst, b71, x0);
00385     cnst = LD_W5;
00386     b51 = vec_madd(cnst, b51, x1);
00387     cnst = LD_W6;
00388     b31 = vec_madd(cnst, b31, x1);
00389     cnst = LD_W7;
00390     b11 = vec_madd(cnst, b11, x0);
00391 
00392     b71 = vec_add(b71, x2);
00393     b51 = vec_add(b51, x3);
00394     b31 = vec_add(b31, x2);
00395     b11 = vec_add(b11, x3);
00396     /* }}} */
00397 #else
00398     /* convert to float {{{ */
00399 #define CTF(n) \
00400     vs32(b##n##1) = vec_unpackl(vs16(b##n##0)); \
00401     vs32(b##n##0) = vec_unpackh(vs16(b##n##0)); \
00402     b##n##1 = vec_ctf(vs32(b##n##1), 0); \
00403     b##n##0 = vec_ctf(vs32(b##n##0), 0); \
00404 
00405     CTF(0);
00406     CTF(1);
00407     CTF(2);
00408     CTF(3);
00409     CTF(4);
00410     CTF(5);
00411     CTF(6);
00412     CTF(7);
00413 
00414 #undef CTF
00415     /* }}} */
00416 
00417     FDCTROW(b00, b10, b20, b30, b40, b50, b60, b70);
00418     FDCTROW(b01, b11, b21, b31, b41, b51, b61, b71);
00419 #endif
00420 
00421 
00422     /* 8x8 matrix transpose (vector float[8][2]) {{{ */
00423     x0 = vec_mergel(b00, b20);
00424     x1 = vec_mergeh(b00, b20);
00425     x2 = vec_mergel(b10, b30);
00426     x3 = vec_mergeh(b10, b30);
00427 
00428     b00 = vec_mergeh(x1, x3);
00429     b10 = vec_mergel(x1, x3);
00430     b20 = vec_mergeh(x0, x2);
00431     b30 = vec_mergel(x0, x2);
00432 
00433     x4 = vec_mergel(b41, b61);
00434     x5 = vec_mergeh(b41, b61);
00435     x6 = vec_mergel(b51, b71);
00436     x7 = vec_mergeh(b51, b71);
00437 
00438     b41 = vec_mergeh(x5, x7);
00439     b51 = vec_mergel(x5, x7);
00440     b61 = vec_mergeh(x4, x6);
00441     b71 = vec_mergel(x4, x6);
00442 
00443     x0 = vec_mergel(b01, b21);
00444     x1 = vec_mergeh(b01, b21);
00445     x2 = vec_mergel(b11, b31);
00446     x3 = vec_mergeh(b11, b31);
00447 
00448     x4 = vec_mergel(b40, b60);
00449     x5 = vec_mergeh(b40, b60);
00450     x6 = vec_mergel(b50, b70);
00451     x7 = vec_mergeh(b50, b70);
00452 
00453     b40 = vec_mergeh(x1, x3);
00454     b50 = vec_mergel(x1, x3);
00455     b60 = vec_mergeh(x0, x2);
00456     b70 = vec_mergel(x0, x2);
00457 
00458     b01 = vec_mergeh(x5, x7);
00459     b11 = vec_mergel(x5, x7);
00460     b21 = vec_mergeh(x4, x6);
00461     b31 = vec_mergel(x4, x6);
00462     /* }}} */
00463 
00464 
00465     FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70);
00466     FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71);
00467 
00468 
00469     /* round, convert back to short {{{ */
00470 #define CTS(n) \
00471     b##n##0 = vec_round(b##n##0); \
00472     b##n##1 = vec_round(b##n##1); \
00473     b##n##0 = ((vector float)vec_cts(b##n##0, 0)); \
00474     b##n##1 = ((vector float)vec_cts(b##n##1, 0)); \
00475     b##n##0 = ((vector float)vec_pack(vs32(b##n##0), vs32(b##n##1))); \
00476     vec_st(vs16(b##n##0), 0, bp);
00477 
00478     bp = (vector signed short*)block;
00479     CTS(0); bp++;
00480     CTS(1); bp++;
00481     CTS(2); bp++;
00482     CTS(3); bp++;
00483     CTS(4); bp++;
00484     CTS(5); bp++;
00485     CTS(6); bp++;
00486     CTS(7);
00487 
00488 #undef CTS
00489     /* }}} */
00490 
00491 POWERPC_PERF_STOP_COUNT(altivec_fdct, 1);
00492 }
00493 
00494 /* vim:set foldmethod=marker foldlevel=0: */