Cinelerra: hvirtual/quicktime/ffmpeg/libavcodec/ppc/fdct

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 library is free software; you can redistribute it and/or
00006  * modify it under the terms of the GNU Lesser General Public
00007  * License as published by the Free Software Foundation; either
00008  * version 2.1 of the License, or (at your option) any later version.
00009  *
00010  * This library is distributed in the hope that it will be useful,
00011  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00012  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00013  * Lesser General Public License for more details.
00014  *
00015  * You should have received a copy of the GNU Lesser General Public
00016  * License along with this library; if not, write to the Free Software
00017  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
00018  */
00019 
00020 
00021 #include "../common.h"
00022 #include "../dsputil.h"
00023 #include "dsputil_altivec.h"
00024 #include "gcc_fixes.h"
00025 
00026 
00027 #define vs16(v) ((vector signed short)(v))
00028 #define vs32(v) ((vector signed int)(v))
00029 #define vu8(v)  ((vector unsigned char)(v))
00030 #define vu16(v) ((vector unsigned short)(v))
00031 #define vu32(v) ((vector unsigned int)(v))
00032 
00033 
00034 #define C1     0.98078525066375732421875000 /* cos(1*PI/16) */
00035 #define C2     0.92387950420379638671875000 /* cos(2*PI/16) */
00036 #define C3     0.83146959543228149414062500 /* cos(3*PI/16) */
00037 #define C4     0.70710676908493041992187500 /* cos(4*PI/16) */
00038 #define C5     0.55557024478912353515625000 /* cos(5*PI/16) */
00039 #define C6     0.38268342614173889160156250 /* cos(6*PI/16) */
00040 #define C7     0.19509032368659973144531250 /* cos(7*PI/16) */
00041 #define SQRT_2 1.41421353816986083984375000 /* sqrt(2)      */
00042 
00043 
00044 #define W0 -(2 * C2)
00045 #define W1 (2 * C6)
00046 #define W2 (SQRT_2 * C6)
00047 #define W3 (SQRT_2 * C3)
00048 #define W4 (SQRT_2 * (-C1 + C3 + C5 - C7))
00049 #define W5 (SQRT_2 * ( C1 + C3 - C5 + C7))
00050 #define W6 (SQRT_2 * ( C1 + C3 + C5 - C7))
00051 #define W7 (SQRT_2 * ( C1 + C3 - C5 - C7))
00052 #define W8 (SQRT_2 * ( C7 - C3))
00053 #define W9 (SQRT_2 * (-C1 - C3))
00054 #define WA (SQRT_2 * (-C3 - C5))
00055 #define WB (SQRT_2 * ( C5 - C3))
00056 
00057 
00058 static vector float fdctconsts[3] = {
00059     (vector float)AVV( W0, W1, W2, W3 ),
00060     (vector float)AVV( W4, W5, W6, W7 ),
00061     (vector float)AVV( W8, W9, WA, WB )
00062 };
00063 
00064 #define LD_W0 vec_splat(cnsts0, 0)
00065 #define LD_W1 vec_splat(cnsts0, 1)
00066 #define LD_W2 vec_splat(cnsts0, 2)
00067 #define LD_W3 vec_splat(cnsts0, 3)
00068 #define LD_W4 vec_splat(cnsts1, 0)
00069 #define LD_W5 vec_splat(cnsts1, 1)
00070 #define LD_W6 vec_splat(cnsts1, 2)
00071 #define LD_W7 vec_splat(cnsts1, 3)
00072 #define LD_W8 vec_splat(cnsts2, 0)
00073 #define LD_W9 vec_splat(cnsts2, 1)
00074 #define LD_WA vec_splat(cnsts2, 2)
00075 #define LD_WB vec_splat(cnsts2, 3)
00076 
00077 
00078 #define FDCTROW(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */                  \
00079     x0 = vec_add(b0, b7);               /* x0 = b0 + b7; */         \
00080     x7 = vec_sub(b0, b7);               /* x7 = b0 - b7; */         \
00081     x1 = vec_add(b1, b6);               /* x1 = b1 + b6; */         \
00082     x6 = vec_sub(b1, b6);               /* x6 = b1 - b6; */         \
00083     x2 = vec_add(b2, b5);               /* x2 = b2 + b5; */         \
00084     x5 = vec_sub(b2, b5);               /* x5 = b2 - b5; */         \
00085     x3 = vec_add(b3, b4);               /* x3 = b3 + b4; */         \
00086     x4 = vec_sub(b3, b4);               /* x4 = b3 - b4; */         \
00087                                                                     \
00088     b7 = vec_add(x0, x3);               /* b7 = x0 + x3; */         \
00089     b1 = vec_add(x1, x2);               /* b1 = x1 + x2; */         \
00090     b0 = vec_add(b7, b1);               /* b0 = b7 + b1; */         \
00091     b4 = vec_sub(b7, b1);               /* b4 = b7 - b1; */         \
00092                                                                     \
00093     b2 = vec_sub(x0, x3);               /* b2 = x0 - x3; */         \
00094     b6 = vec_sub(x1, x2);               /* b6 = x1 - x2; */         \
00095     b5 = vec_add(b6, b2);               /* b5 = b6 + b2; */         \
00096     cnst = LD_W2;                                                   \
00097     b5 = vec_madd(cnst, b5, mzero);     /* b5 = b5 * W2; */         \
00098     cnst = LD_W1;                                                   \
00099     b2 = vec_madd(cnst, b2, b5);        /* b2 = b5 + b2 * W1; */    \
00100     cnst = LD_W0;                                                   \
00101     b6 = vec_madd(cnst, b6, b5);        /* b6 = b5 + b6 * W0; */    \
00102                                                                     \
00103     x0 = vec_add(x4, x7);               /* x0 = x4 + x7; */         \
00104     x1 = vec_add(x5, x6);               /* x1 = x5 + x6; */         \
00105     x2 = vec_add(x4, x6);               /* x2 = x4 + x6; */         \
00106     x3 = vec_add(x5, x7);               /* x3 = x5 + x7; */         \
00107     x8 = vec_add(x2, x3);               /* x8 = x2 + x3; */         \
00108     cnst = LD_W3;                                                   \
00109     x8 = vec_madd(cnst, x8, mzero);     /* x8 = x8 * W3; */         \
00110                                                                     \
00111     cnst = LD_W8;                                                   \
00112     x0 = vec_madd(cnst, x0, mzero);     /* x0 *= W8; */             \
00113     cnst = LD_W9;                                                   \
00114     x1 = vec_madd(cnst, x1, mzero);     /* x1 *= W9; */             \
00115     cnst = LD_WA;                                                   \
00116     x2 = vec_madd(cnst, x2, x8);        /* x2 = x2 * WA + x8; */    \
00117     cnst = LD_WB;                                                   \
00118     x3 = vec_madd(cnst, x3, x8);        /* x3 = x3 * WB + x8; */    \
00119                                                                     \
00120     cnst = LD_W4;                                                   \
00121     b7 = vec_madd(cnst, x4, x0);        /* b7 = x4 * W4 + x0; */    \
00122     cnst = LD_W5;                                                   \
00123     b5 = vec_madd(cnst, x5, x1);        /* b5 = x5 * W5 + x1; */    \
00124     cnst = LD_W6;                                                   \
00125     b3 = vec_madd(cnst, x6, x1);        /* b3 = x6 * W6 + x1; */    \
00126     cnst = LD_W7;                                                   \
00127     b1 = vec_madd(cnst, x7, x0);        /* b1 = x7 * W7 + x0; */    \
00128                                                                     \
00129     b7 = vec_add(b7, x2);               /* b7 = b7 + x2; */         \
00130     b5 = vec_add(b5, x3);               /* b5 = b5 + x3; */         \
00131     b3 = vec_add(b3, x2);               /* b3 = b3 + x2; */         \
00132     b1 = vec_add(b1, x3);               /* b1 = b1 + x3; */         \
00133     /* }}} */
00134 
00135 #define FDCTCOL(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */                  \
00136     x0 = vec_add(b0, b7);               /* x0 = b0 + b7; */         \
00137     x7 = vec_sub(b0, b7);               /* x7 = b0 - b7; */         \
00138     x1 = vec_add(b1, b6);               /* x1 = b1 + b6; */         \
00139     x6 = vec_sub(b1, b6);               /* x6 = b1 - b6; */         \
00140     x2 = vec_add(b2, b5);               /* x2 = b2 + b5; */         \
00141     x5 = vec_sub(b2, b5);               /* x5 = b2 - b5; */         \
00142     x3 = vec_add(b3, b4);               /* x3 = b3 + b4; */         \
00143     x4 = vec_sub(b3, b4);               /* x4 = b3 - b4; */         \
00144                                                                     \
00145     b7 = vec_add(x0, x3);               /* b7 = x0 + x3; */         \
00146     b1 = vec_add(x1, x2);               /* b1 = x1 + x2; */         \
00147     b0 = vec_add(b7, b1);               /* b0 = b7 + b1; */         \
00148     b4 = vec_sub(b7, b1);               /* b4 = b7 - b1; */         \
00149                                                                     \
00150     b2 = vec_sub(x0, x3);               /* b2 = x0 - x3; */         \
00151     b6 = vec_sub(x1, x2);               /* b6 = x1 - x2; */         \
00152     b5 = vec_add(b6, b2);               /* b5 = b6 + b2; */         \
00153     cnst = LD_W2;                                                   \
00154     b5 = vec_madd(cnst, b5, mzero);     /* b5 = b5 * W2; */         \
00155     cnst = LD_W1;                                                   \
00156     b2 = vec_madd(cnst, b2, b5);        /* b2 = b5 + b2 * W1; */    \
00157     cnst = LD_W0;                                                   \
00158     b6 = vec_madd(cnst, b6, b5);        /* b6 = b5 + b6 * W0; */    \
00159                                                                     \
00160     x0 = vec_add(x4, x7);               /* x0 = x4 + x7; */         \
00161     x1 = vec_add(x5, x6);               /* x1 = x5 + x6; */         \
00162     x2 = vec_add(x4, x6);               /* x2 = x4 + x6; */         \
00163     x3 = vec_add(x5, x7);               /* x3 = x5 + x7; */         \
00164     x8 = vec_add(x2, x3);               /* x8 = x2 + x3; */         \
00165     cnst = LD_W3;                                                   \
00166     x8 = vec_madd(cnst, x8, mzero);     /* x8 = x8 * W3; */         \
00167                                                                     \
00168     cnst = LD_W8;                                                   \
00169     x0 = vec_madd(cnst, x0, mzero);     /* x0 *= W8; */             \
00170     cnst = LD_W9;                                                   \
00171     x1 = vec_madd(cnst, x1, mzero);     /* x1 *= W9; */             \
00172     cnst = LD_WA;                                                   \
00173     x2 = vec_madd(cnst, x2, x8);        /* x2 = x2 * WA + x8; */    \
00174     cnst = LD_WB;                                                   \
00175     x3 = vec_madd(cnst, x3, x8);        /* x3 = x3 * WB + x8; */    \
00176                                                                     \
00177     cnst = LD_W4;                                                   \
00178     b7 = vec_madd(cnst, x4, x0);        /* b7 = x4 * W4 + x0; */    \
00179     cnst = LD_W5;                                                   \
00180     b5 = vec_madd(cnst, x5, x1);        /* b5 = x5 * W5 + x1; */    \
00181     cnst = LD_W6;                                                   \
00182     b3 = vec_madd(cnst, x6, x1);        /* b3 = x6 * W6 + x1; */    \
00183     cnst = LD_W7;                                                   \
00184     b1 = vec_madd(cnst, x7, x0);        /* b1 = x7 * W7 + x0; */    \
00185                                                                     \
00186     b7 = vec_add(b7, x2);               /* b7 += x2; */             \
00187     b5 = vec_add(b5, x3);               /* b5 += x3; */             \
00188     b3 = vec_add(b3, x2);               /* b3 += x2; */             \
00189     b1 = vec_add(b1, x3);               /* b1 += x3; */             \
00190     /* }}} */
00191 
00192 
00193 
00194 /* two dimensional discrete cosine transform */
00195 
00196 void fdct_altivec(int16_t *block)
00197 {
00198 POWERPC_PERF_DECLARE(altivec_fdct, 1);
00199 #ifdef ALTIVEC_USE_REFERENCE_C_CODE
00200 POWERPC_PERF_START_COUNT(altivec_fdct, 1);
00201     void ff_jpeg_fdct_islow(int16_t *block);
00202     ff_jpeg_fdct_islow(block);
00203 POWERPC_PERF_STOP_COUNT(altivec_fdct, 1);
00204 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
00205     vector signed short *bp;
00206     vector float *cp;
00207     vector float b00, b10, b20, b30, b40, b50, b60, b70;
00208     vector float b01, b11, b21, b31, b41, b51, b61, b71;
00209     vector float mzero, cnst, cnsts0, cnsts1, cnsts2;
00210     vector float x0, x1, x2, x3, x4, x5, x6, x7, x8;
00211 
00212     POWERPC_PERF_START_COUNT(altivec_fdct, 1);
00213 
00214 
00215     /* setup constants {{{ */
00216     /* mzero = -0.0 */
00217     mzero = ((vector float)vec_splat_u32(-1));
00218     mzero = ((vector float)vec_sl(vu32(mzero), vu32(mzero)));
00219     cp = fdctconsts;
00220     cnsts0 = vec_ld(0, cp); cp++;
00221     cnsts1 = vec_ld(0, cp); cp++;
00222     cnsts2 = vec_ld(0, cp);
00223     /* }}} */
00224 
00225 
00226     /* 8x8 matrix transpose (vector short[8]) {{{ */
00227 #define MERGE_S16(hl,a,b) vec_merge##hl(vs16(a), vs16(b))
00228 
00229     bp = (vector signed short*)block;
00230     b00 = ((vector float)vec_ld(0,    bp));
00231     b40 = ((vector float)vec_ld(16*4, bp));
00232     b01 = ((vector float)MERGE_S16(h, b00, b40));
00233     b11 = ((vector float)MERGE_S16(l, b00, b40));
00234     bp++;
00235     b10 = ((vector float)vec_ld(0,    bp));
00236     b50 = ((vector float)vec_ld(16*4, bp));
00237     b21 = ((vector float)MERGE_S16(h, b10, b50));
00238     b31 = ((vector float)MERGE_S16(l, b10, b50));
00239     bp++;
00240     b20 = ((vector float)vec_ld(0,    bp));
00241     b60 = ((vector float)vec_ld(16*4, bp));
00242     b41 = ((vector float)MERGE_S16(h, b20, b60));
00243     b51 = ((vector float)MERGE_S16(l, b20, b60));
00244     bp++;
00245     b30 = ((vector float)vec_ld(0,    bp));
00246     b70 = ((vector float)vec_ld(16*4, bp));
00247     b61 = ((vector float)MERGE_S16(h, b30, b70));
00248     b71 = ((vector float)MERGE_S16(l, b30, b70));
00249 
00250     x0 = ((vector float)MERGE_S16(h, b01, b41));
00251     x1 = ((vector float)MERGE_S16(l, b01, b41));
00252     x2 = ((vector float)MERGE_S16(h, b11, b51));
00253     x3 = ((vector float)MERGE_S16(l, b11, b51));
00254     x4 = ((vector float)MERGE_S16(h, b21, b61));
00255     x5 = ((vector float)MERGE_S16(l, b21, b61));
00256     x6 = ((vector float)MERGE_S16(h, b31, b71));
00257     x7 = ((vector float)MERGE_S16(l, b31, b71));
00258 
00259     b00 = ((vector float)MERGE_S16(h, x0, x4));
00260     b10 = ((vector float)MERGE_S16(l, x0, x4));
00261     b20 = ((vector float)MERGE_S16(h, x1, x5));
00262     b30 = ((vector float)MERGE_S16(l, x1, x5));
00263     b40 = ((vector float)MERGE_S16(h, x2, x6));
00264     b50 = ((vector float)MERGE_S16(l, x2, x6));
00265     b60 = ((vector float)MERGE_S16(h, x3, x7));
00266     b70 = ((vector float)MERGE_S16(l, x3, x7));
00267 
00268 #undef MERGE_S16
00269     /* }}} */
00270 
00271 
00272 /* Some of the initial calculations can be done as vector short before
00273  * conversion to vector float.  The following code section takes advantage
00274  * of this.
00275  */
00276 #if 1
00277     /* fdct rows {{{ */
00278     x0 = ((vector float)vec_add(vs16(b00), vs16(b70)));
00279     x7 = ((vector float)vec_sub(vs16(b00), vs16(b70)));
00280     x1 = ((vector float)vec_add(vs16(b10), vs16(b60)));
00281     x6 = ((vector float)vec_sub(vs16(b10), vs16(b60)));
00282     x2 = ((vector float)vec_add(vs16(b20), vs16(b50)));
00283     x5 = ((vector float)vec_sub(vs16(b20), vs16(b50)));
00284     x3 = ((vector float)vec_add(vs16(b30), vs16(b40)));
00285     x4 = ((vector float)vec_sub(vs16(b30), vs16(b40)));
00286 
00287     b70 = ((vector float)vec_add(vs16(x0), vs16(x3)));
00288     b10 = ((vector float)vec_add(vs16(x1), vs16(x2)));
00289 
00290     b00 = ((vector float)vec_add(vs16(b70), vs16(b10)));
00291     b40 = ((vector float)vec_sub(vs16(b70), vs16(b10)));
00292 
00293 #define CTF0(n) \
00294     b##n##1 = ((vector float)vec_unpackl(vs16(b##n##0))); \
00295     b##n##0 = ((vector float)vec_unpackh(vs16(b##n##0))); \
00296     b##n##1 = vec_ctf(vs32(b##n##1), 0); \
00297     b##n##0 = vec_ctf(vs32(b##n##0), 0);
00298 
00299     CTF0(0);
00300     CTF0(4);
00301 
00302     b20 = ((vector float)vec_sub(vs16(x0), vs16(x3)));
00303     b60 = ((vector float)vec_sub(vs16(x1), vs16(x2)));
00304 
00305     CTF0(2);
00306     CTF0(6);
00307 
00308 #undef CTF0
00309 
00310     x0 = vec_add(b60, b20);
00311     x1 = vec_add(b61, b21);
00312 
00313     cnst = LD_W2;
00314     x0 = vec_madd(cnst, x0, mzero);
00315     x1 = vec_madd(cnst, x1, mzero);
00316     cnst = LD_W1;
00317     b20 = vec_madd(cnst, b20, x0);
00318     b21 = vec_madd(cnst, b21, x1);
00319     cnst = LD_W0;
00320     b60 = vec_madd(cnst, b60, x0);
00321     b61 = vec_madd(cnst, b61, x1);
00322 
00323 #define CTFX(x,b) \
00324     b##0 = ((vector float)vec_unpackh(vs16(x))); \
00325     b##1 = ((vector float)vec_unpackl(vs16(x))); \
00326     b##0 = vec_ctf(vs32(b##0), 0); \
00327     b##1 = vec_ctf(vs32(b##1), 0); \
00328 
00329     CTFX(x4, b7);
00330     CTFX(x5, b5);
00331     CTFX(x6, b3);
00332     CTFX(x7, b1);
00333 
00334 #undef CTFX
00335 
00336 
00337     x0 = vec_add(b70, b10);
00338     x1 = vec_add(b50, b30);
00339     x2 = vec_add(b70, b30);
00340     x3 = vec_add(b50, b10);
00341     x8 = vec_add(x2, x3);
00342     cnst = LD_W3;
00343     x8 = vec_madd(cnst, x8, mzero);
00344 
00345     cnst = LD_W8;
00346     x0 = vec_madd(cnst, x0, mzero);
00347     cnst = LD_W9;
00348     x1 = vec_madd(cnst, x1, mzero);
00349     cnst = LD_WA;
00350     x2 = vec_madd(cnst, x2, x8);
00351     cnst = LD_WB;
00352     x3 = vec_madd(cnst, x3, x8);
00353 
00354     cnst = LD_W4;
00355     b70 = vec_madd(cnst, b70, x0);
00356     cnst = LD_W5;
00357     b50 = vec_madd(cnst, b50, x1);
00358     cnst = LD_W6;
00359     b30 = vec_madd(cnst, b30, x1);
00360     cnst = LD_W7;
00361     b10 = vec_madd(cnst, b10, x0);
00362 
00363     b70 = vec_add(b70, x2);
00364     b50 = vec_add(b50, x3);
00365     b30 = vec_add(b30, x2);
00366     b10 = vec_add(b10, x3);
00367 
00368 
00369     x0 = vec_add(b71, b11);
00370     x1 = vec_add(b51, b31);
00371     x2 = vec_add(b71, b31);
00372     x3 = vec_add(b51, b11);
00373     x8 = vec_add(x2, x3);
00374     cnst = LD_W3;
00375     x8 = vec_madd(cnst, x8, mzero);
00376 
00377     cnst = LD_W8;
00378     x0 = vec_madd(cnst, x0, mzero);
00379     cnst = LD_W9;
00380     x1 = vec_madd(cnst, x1, mzero);
00381     cnst = LD_WA;
00382     x2 = vec_madd(cnst, x2, x8);
00383     cnst = LD_WB;
00384     x3 = vec_madd(cnst, x3, x8);
00385 
00386     cnst = LD_W4;
00387     b71 = vec_madd(cnst, b71, x0);
00388     cnst = LD_W5;
00389     b51 = vec_madd(cnst, b51, x1);
00390     cnst = LD_W6;
00391     b31 = vec_madd(cnst, b31, x1);
00392     cnst = LD_W7;
00393     b11 = vec_madd(cnst, b11, x0);
00394 
00395     b71 = vec_add(b71, x2);
00396     b51 = vec_add(b51, x3);
00397     b31 = vec_add(b31, x2);
00398     b11 = vec_add(b11, x3);
00399     /* }}} */
00400 #else
00401     /* convert to float {{{ */
00402 #define CTF(n) \
00403     vs32(b##n##1) = vec_unpackl(vs16(b##n##0)); \
00404     vs32(b##n##0) = vec_unpackh(vs16(b##n##0)); \
00405     b##n##1 = vec_ctf(vs32(b##n##1), 0); \
00406     b##n##0 = vec_ctf(vs32(b##n##0), 0); \
00407 
00408     CTF(0);
00409     CTF(1);
00410     CTF(2);
00411     CTF(3);
00412     CTF(4);
00413     CTF(5);
00414     CTF(6);
00415     CTF(7);
00416 
00417 #undef CTF
00418     /* }}} */
00419 
00420     FDCTROW(b00, b10, b20, b30, b40, b50, b60, b70);
00421     FDCTROW(b01, b11, b21, b31, b41, b51, b61, b71);
00422 #endif
00423 
00424 
00425     /* 8x8 matrix transpose (vector float[8][2]) {{{ */
00426     x0 = vec_mergel(b00, b20);
00427     x1 = vec_mergeh(b00, b20);
00428     x2 = vec_mergel(b10, b30);
00429     x3 = vec_mergeh(b10, b30);
00430 
00431     b00 = vec_mergeh(x1, x3);
00432     b10 = vec_mergel(x1, x3);
00433     b20 = vec_mergeh(x0, x2);
00434     b30 = vec_mergel(x0, x2);
00435 
00436     x4 = vec_mergel(b41, b61);
00437     x5 = vec_mergeh(b41, b61);
00438     x6 = vec_mergel(b51, b71);
00439     x7 = vec_mergeh(b51, b71);
00440 
00441     b41 = vec_mergeh(x5, x7);
00442     b51 = vec_mergel(x5, x7);
00443     b61 = vec_mergeh(x4, x6);
00444     b71 = vec_mergel(x4, x6);
00445 
00446     x0 = vec_mergel(b01, b21);
00447     x1 = vec_mergeh(b01, b21);
00448     x2 = vec_mergel(b11, b31);
00449     x3 = vec_mergeh(b11, b31);
00450 
00451     x4 = vec_mergel(b40, b60);
00452     x5 = vec_mergeh(b40, b60);
00453     x6 = vec_mergel(b50, b70);
00454     x7 = vec_mergeh(b50, b70);
00455 
00456     b40 = vec_mergeh(x1, x3);
00457     b50 = vec_mergel(x1, x3);
00458     b60 = vec_mergeh(x0, x2);
00459     b70 = vec_mergel(x0, x2);
00460 
00461     b01 = vec_mergeh(x5, x7);
00462     b11 = vec_mergel(x5, x7);
00463     b21 = vec_mergeh(x4, x6);
00464     b31 = vec_mergel(x4, x6);
00465     /* }}} */
00466 
00467 
00468     FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70);
00469     FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71);
00470 
00471 
00472     /* round, convert back to short {{{ */
00473 #define CTS(n) \
00474     b##n##0 = vec_round(b##n##0); \
00475     b##n##1 = vec_round(b##n##1); \
00476     b##n##0 = ((vector float)vec_cts(b##n##0, 0)); \
00477     b##n##1 = ((vector float)vec_cts(b##n##1, 0)); \
00478     b##n##0 = ((vector float)vec_pack(vs32(b##n##0), vs32(b##n##1))); \
00479     vec_st(vs16(b##n##0), 0, bp);
00480 
00481     bp = (vector signed short*)block;
00482     CTS(0); bp++;
00483     CTS(1); bp++;
00484     CTS(2); bp++;
00485     CTS(3); bp++;
00486     CTS(4); bp++;
00487     CTS(5); bp++;
00488     CTS(6); bp++;
00489     CTS(7);
00490 
00491 #undef CTS
00492     /* }}} */
00493 
00494 POWERPC_PERF_STOP_COUNT(altivec_fdct, 1);
00495 #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
00496 }
00497 
00498 /* vim:set foldmethod=marker foldlevel=0: */
hvirtual/quicktime/ffmpeg/libavcodec/ppc/fdct_altivec.c
