• Main Page
  • Related Pages
  • Modules
  • Data Structures
  • Files
  • Examples
  • File List
  • Globals

libavcodec/sh4/idct_sh4.c

Go to the documentation of this file.
00001 /*
00002  * idct for sh4
00003  *
00004  * Copyright (c) 2001-2003 BERO <bero@geocities.co.jp>
00005  *
00006  * This file is part of Libav.
00007  *
00008  * Libav is free software; you can redistribute it and/or
00009  * modify it under the terms of the GNU Lesser General Public
00010  * License as published by the Free Software Foundation; either
00011  * version 2.1 of the License, or (at your option) any later version.
00012  *
00013  * Libav is distributed in the hope that it will be useful,
00014  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00015  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00016  * Lesser General Public License for more details.
00017  *
00018  * You should have received a copy of the GNU Lesser General Public
00019  * License along with Libav; if not, write to the Free Software
00020  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00021  */
00022 
00023 #include "libavcodec/dsputil.h"
00024 #include "dsputil_sh4.h"
00025 #include "sh4.h"
00026 
00027 #define c1      1.38703984532214752434  /* sqrt(2)*cos(1*pi/16) */
00028 #define c2      1.30656296487637657577  /* sqrt(2)*cos(2*pi/16) */
00029 #define c3      1.17587560241935884520  /* sqrt(2)*cos(3*pi/16) */
00030 #define c4      1.00000000000000000000  /* sqrt(2)*cos(4*pi/16) */
00031 #define c5      0.78569495838710234903  /* sqrt(2)*cos(5*pi/16) */
00032 #define c6      0.54119610014619712324  /* sqrt(2)*cos(6*pi/16) */
00033 #define c7      0.27589937928294311353  /* sqrt(2)*cos(7*pi/16) */
00034 
00035 static const float even_table[] __attribute__ ((aligned(8))) = {
00036         c4, c4, c4, c4,
00037         c2, c6,-c6,-c2,
00038         c4,-c4,-c4, c4,
00039         c6,-c2, c2,-c6
00040 };
00041 
00042 static const float odd_table[] __attribute__ ((aligned(8))) = {
00043         c1, c3, c5, c7,
00044         c3,-c7,-c1,-c5,
00045         c5,-c1, c7, c3,
00046         c7,-c5, c3,-c1
00047 };
00048 
00049 #undef  c1
00050 #undef  c2
00051 #undef  c3
00052 #undef  c4
00053 #undef  c5
00054 #undef  c6
00055 #undef  c7
00056 
00057 #define         load_matrix(table) \
00058     do { \
00059         const float *t = table; \
00060         __asm__ volatile( \
00061         "       fschg\n" \
00062         "       fmov   @%0+,xd0\n" \
00063         "       fmov   @%0+,xd2\n" \
00064         "       fmov   @%0+,xd4\n" \
00065         "       fmov   @%0+,xd6\n" \
00066         "       fmov   @%0+,xd8\n" \
00067         "       fmov   @%0+,xd10\n" \
00068         "       fmov   @%0+,xd12\n" \
00069         "       fmov   @%0+,xd14\n" \
00070         "       fschg\n" \
00071         : "+r"(t) \
00072         ); \
00073     } while (0)
00074 
00075 #define         ftrv() \
00076                 __asm__ volatile("ftrv xmtrx,fv0" \
00077                 : "+f"(fr0),"+f"(fr1),"+f"(fr2),"+f"(fr3));
00078 
00079 #define         DEFREG        \
00080         register float fr0 __asm__("fr0"); \
00081         register float fr1 __asm__("fr1"); \
00082         register float fr2 __asm__("fr2"); \
00083         register float fr3 __asm__("fr3")
00084 
00085 #define         DESCALE(x,n)    (x)*(1.0f/(1<<(n)))
00086 
00087 /* this code work worse on gcc cvs. 3.2.3 work fine */
00088 
00089 
00090 //optimized
00091 
00092 void idct_sh4(DCTELEM *block)
00093 {
00094         DEFREG;
00095 
00096         int i;
00097         float        tblock[8*8],*fblock;
00098         int ofs1,ofs2,ofs3;
00099         int fpscr;
00100 
00101         fp_single_enter(fpscr);
00102 
00103         /* row */
00104 
00105         /* even part */
00106         load_matrix(even_table);
00107 
00108         fblock = tblock+4;
00109         i = 8;
00110         do {
00111                 fr0 = block[0];
00112                 fr1 = block[2];
00113                 fr2 = block[4];
00114                 fr3 = block[6];
00115                 block+=8;
00116                 ftrv();
00117                 *--fblock = fr3;
00118                 *--fblock = fr2;
00119                 *--fblock = fr1;
00120                 *--fblock = fr0;
00121                 fblock+=8+4;
00122         } while(--i);
00123         block-=8*8;
00124         fblock-=8*8+4;
00125 
00126         load_matrix(odd_table);
00127 
00128         i = 8;
00129 
00130         do {
00131                 float t0,t1,t2,t3;
00132                 fr0 = block[1];
00133                 fr1 = block[3];
00134                 fr2 = block[5];
00135                 fr3 = block[7];
00136                 block+=8;
00137                 ftrv();
00138                 t0 = *fblock++;
00139                 t1 = *fblock++;
00140                 t2 = *fblock++;
00141                 t3 = *fblock++;
00142                 fblock+=4;
00143                 *--fblock = t0 - fr0;
00144                 *--fblock = t1 - fr1;
00145                 *--fblock = t2 - fr2;
00146                 *--fblock = t3 - fr3;
00147                 *--fblock = t3 + fr3;
00148                 *--fblock = t2 + fr2;
00149                 *--fblock = t1 + fr1;
00150                 *--fblock = t0 + fr0;
00151                 fblock+=8;
00152         } while(--i);
00153         block-=8*8;
00154         fblock-=8*8;
00155 
00156         /* col */
00157 
00158         /* even part */
00159         load_matrix(even_table);
00160 
00161         ofs1 = sizeof(float)*2*8;
00162         ofs2 = sizeof(float)*4*8;
00163         ofs3 = sizeof(float)*6*8;
00164 
00165         i = 8;
00166 
00167 #define        OA(fblock,ofs)   *(float*)((char*)fblock + ofs)
00168 
00169         do {
00170                 fr0 = OA(fblock,   0);
00171                 fr1 = OA(fblock,ofs1);
00172                 fr2 = OA(fblock,ofs2);
00173                 fr3 = OA(fblock,ofs3);
00174                 ftrv();
00175                 OA(fblock,0   ) = fr0;
00176                 OA(fblock,ofs1) = fr1;
00177                 OA(fblock,ofs2) = fr2;
00178                 OA(fblock,ofs3) = fr3;
00179                 fblock++;
00180         } while(--i);
00181         fblock-=8;
00182 
00183         load_matrix(odd_table);
00184 
00185         i=8;
00186         do {
00187                 float t0,t1,t2,t3;
00188                 t0 = OA(fblock,   0); /* [8*0] */
00189                 t1 = OA(fblock,ofs1); /* [8*2] */
00190                 t2 = OA(fblock,ofs2); /* [8*4] */
00191                 t3 = OA(fblock,ofs3); /* [8*6] */
00192                 fblock+=8;
00193                 fr0 = OA(fblock,   0); /* [8*1] */
00194                 fr1 = OA(fblock,ofs1); /* [8*3] */
00195                 fr2 = OA(fblock,ofs2); /* [8*5] */
00196                 fr3 = OA(fblock,ofs3); /* [8*7] */
00197                 fblock+=-8+1;
00198                 ftrv();
00199                 block[8*0] = DESCALE(t0 + fr0,3);
00200                 block[8*7] = DESCALE(t0 - fr0,3);
00201                 block[8*1] = DESCALE(t1 + fr1,3);
00202                 block[8*6] = DESCALE(t1 - fr1,3);
00203                 block[8*2] = DESCALE(t2 + fr2,3);
00204                 block[8*5] = DESCALE(t2 - fr2,3);
00205                 block[8*3] = DESCALE(t3 + fr3,3);
00206                 block[8*4] = DESCALE(t3 - fr3,3);
00207                 block++;
00208         } while(--i);
00209 
00210         fp_single_leave(fpscr);
00211 }
Generated on Thu Jan 24 2013 17:08:53 for Libav by doxygen 1.7.1