plugins/idct/idctclassic.c

   1 /*****************************************************************************
   2  * idctclassic.c : Classic IDCT module
   3  *****************************************************************************
   4  * Copyright (C) 1999, 2000 VideoLAN
   5  * $Id: idctclassic.c,v 1.12 2001/07/11 02:01:04 sam Exp $
   6  *
   7  * Authors: Gaël Hendryckx <jimmy@via.ecp.fr>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
  22  *****************************************************************************/
  23
  24 #define MODULE_NAME idctclassic
  25
  26 /*****************************************************************************
  27  * Preamble
  28  *****************************************************************************/
  29 #include "defs.h"
  30
  31 #include <stdlib.h>
  32
  33 #include "config.h"
  34 #include "common.h"
  35 #include "threads.h"
  36 #include "mtime.h"
  37 #include "tests.h"
  38
  39 #include "video.h"
  40 #include "video_output.h"
  41
  42 #include "video_decoder.h"
  43
  44 #include "modules.h"
  45 #include "modules_inner.h"
  46
  47 #include "vdec_block.h"
  48 #include "vdec_idct.h"
  49
  50 #include "modules_export.h"
  51
  52 /*****************************************************************************
  53  * Local and extern prototypes.
  54  *****************************************************************************/
  55 static void idct_getfunctions( function_list_t * p_function_list );
  56 static int  idct_Probe      ( probedata_t *p_data );
  57 static void vdec_NormScan   ( u8 ppi_scan[2][64] );
  58
  59
  60 /*****************************************************************************
  61  * Build configuration tree.
  62  *****************************************************************************/
  63 MODULE_CONFIG_START
  64 ADD_WINDOW( "Configuration for classic IDCT module" )
  65     ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
  66 MODULE_CONFIG_STOP
  67
  68 MODULE_INIT_START
  69     p_module->i_capabilities = MODULE_CAPABILITY_NULL
  70                                 | MODULE_CAPABILITY_IDCT;
  71     p_module->psz_longname = "classic IDCT module";
  72 MODULE_INIT_STOP
  73
  74 MODULE_ACTIVATE_START
  75     idct_getfunctions( &p_module->p_functions->idct );
  76 MODULE_ACTIVATE_STOP
  77
  78 MODULE_DEACTIVATE_START
  79 MODULE_DEACTIVATE_STOP
  80
  81 /* Following functions are local */
  82
  83 /*****************************************************************************
  84  * Functions exported as capabilities. They are declared as static so that
  85  * we don't pollute the namespace too much.
  86  *****************************************************************************/
  87 static void idct_getfunctions( function_list_t * p_function_list )
  88 {
  89     p_function_list->pf_probe = idct_Probe;
  90 #define F p_function_list->functions.idct
  91     F.pf_idct_init = _M( vdec_InitIDCT );
  92     F.pf_sparse_idct = _M( vdec_SparseIDCT );
  93     F.pf_idct = _M( vdec_IDCT );
  94     F.pf_norm_scan = vdec_NormScan;
  95     F.pf_decode_init = _M( vdec_InitDecode );
  96     F.pf_decode_mb_c = _M( vdec_DecodeMacroblockC );
  97     F.pf_decode_mb_bw = _M( vdec_DecodeMacroblockBW );
  98 #undef F
  99 }
 100
 101 /*****************************************************************************
 102  * idct_Probe: returns a preference score
 103  *****************************************************************************/
 104 static int idct_Probe( probedata_t *p_data )
 105 {
 106     if( TestMethod( IDCT_METHOD_VAR, "idctclassic" )
 107          || TestMethod( IDCT_METHOD_VAR, "classic" ) )
 108     {
 109         return( 999 );
 110     }
 111
 112     /* This plugin always works */
 113     return( 100 );
 114 }
 115
 116 /*****************************************************************************
 117  * vdec_NormScan : Unused in this IDCT
 118  *****************************************************************************/
 119 static void vdec_NormScan( u8 ppi_scan[2][64] )
 120 {
 121 }
 122
 123 /*****************************************************************************
 124  * vdec_IDCT : IDCT function for normal matrices
 125  *****************************************************************************/
 126 void _M( vdec_IDCT )( vdec_thread_t * p_vdec, dctelem_t * p_block,
 127                 int i_idontcare )
 128 {
 129     /* dct classique: pour tester la meilleure entre la classique et la */
 130     /* no classique */
 131     s32 tmp0, tmp1, tmp2, tmp3;
 132     s32 tmp10, tmp11, tmp12, tmp13;
 133     s32 z1, z2, z3, z4, z5;
 134     dctelem_t * dataptr;
 135     int rowctr;
 136     SHIFT_TEMPS
 137
 138   /* Pass 1: process rows. */
 139   /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
 140   /* furthermore, we scale the results by 2**PASS1_BITS. */
 141
 142     dataptr = p_block;
 143     for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--)
 144     {
 145     /* Due to quantization, we will usually find that many of the input
 146      * coefficients are zero, especially the AC terms.  We can exploit this
 147      * by short-circuiting the IDCT calculation for any row in which all
 148      * the AC terms are zero.  In that case each output is equal to the
 149      * DC coefficient (with scale factor as needed).
 150      * With typical images and quantization tables, half or more of the
 151      * row DCT calculations can be simplified this way.
 152      */
 153
 154         if ((dataptr[1] | dataptr[2] | dataptr[3] | dataptr[4] |
 155                 dataptr[5] | dataptr[6] | dataptr[7]) == 0)
 156         {
 157       /* AC terms all zero */
 158             dctelem_t dcval = (dctelem_t) (dataptr[0] << PASS1_BITS);
 159
 160             dataptr[0] = dcval;
 161             dataptr[1] = dcval;
 162             dataptr[2] = dcval;
 163             dataptr[3] = dcval;
 164             dataptr[4] = dcval;
 165             dataptr[5] = dcval;
 166             dataptr[6] = dcval;
 167             dataptr[7] = dcval;
 168
 169             dataptr += DCTSIZE; /* advance pointer to next row */
 170             continue;
 171         }
 172
 173     /* Even part: reverse the even part of the forward DCT. */
 174     /* The rotator is sqrt(2)*c(-6). */
 175
 176         z2 = (s32) dataptr[2];
 177         z3 = (s32) dataptr[6];
 178
 179         z1 = MULTIPLY(z2 + z3, FIX(0.541196100));
 180         tmp2 = z1 + MULTIPLY(z3, - FIX(1.847759065));
 181         tmp3 = z1 + MULTIPLY(z2, FIX(0.765366865));
 182
 183         tmp0 = ((s32) dataptr[0] + (s32) dataptr[4]) << CONST_BITS;
 184         tmp1 = ((s32) dataptr[0] - (s32) dataptr[4]) << CONST_BITS;
 185
 186         tmp10 = tmp0 + tmp3;
 187         tmp13 = tmp0 - tmp3;
 188         tmp11 = tmp1 + tmp2;
 189         tmp12 = tmp1 - tmp2;
 190
 191     /* Odd part per figure 8; the matrix is unitary and hence its
 192      * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
 193      */
 194
 195         tmp0 = (s32) dataptr[7];
 196         tmp1 = (s32) dataptr[5];
 197         tmp2 = (s32) dataptr[3];
 198         tmp3 = (s32) dataptr[1];
 199
 200         z1 = tmp0 + tmp3;
 201         z2 = tmp1 + tmp2;
 202         z3 = tmp0 + tmp2;
 203         z4 = tmp1 + tmp3;
 204         z5 = MULTIPLY(z3 + z4, FIX(1.175875602)); /* sqrt(2) * c3 */
 205
 206         tmp0 = MULTIPLY(tmp0, FIX(0.298631336)); /* sqrt(2) * (-c1+c3+c5-c7) */
 207         tmp1 = MULTIPLY(tmp1, FIX(2.053119869)); /* sqrt(2) * ( c1+c3-c5+c7) */
 208         tmp2 = MULTIPLY(tmp2, FIX(3.072711026)); /* sqrt(2) * ( c1+c3+c5-c7) */
 209         tmp3 = MULTIPLY(tmp3, FIX(1.501321110)); /* sqrt(2) * ( c1+c3-c5-c7) */
 210         z1 = MULTIPLY(z1, - FIX(0.899976223)); /* sqrt(2) * (c7-c3) */
 211         z2 = MULTIPLY(z2, - FIX(2.562915447)); /* sqrt(2) * (-c1-c3) */
 212         z3 = MULTIPLY(z3, - FIX(1.961570560)); /* sqrt(2) * (-c3-c5) */
 213         z4 = MULTIPLY(z4, - FIX(0.390180644)); /* sqrt(2) * (c5-c3) */
 214
 215         z3 += z5;
 216         z4 += z5;
 217
 218         tmp0 += z1 + z3;
 219         tmp1 += z2 + z4;
 220         tmp2 += z2 + z3;
 221         tmp3 += z1 + z4;
 222
 223     /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
 224
 225         dataptr[0] = (dctelem_t) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
 226         dataptr[7] = (dctelem_t) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
 227         dataptr[1] = (dctelem_t) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
 228         dataptr[6] = (dctelem_t) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
 229         dataptr[2] = (dctelem_t) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
 230         dataptr[5] = (dctelem_t) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
 231         dataptr[3] = (dctelem_t) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
 232         dataptr[4] = (dctelem_t) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
 233
 234         dataptr += DCTSIZE;             /* advance pointer to next row */
 235     }
 236
 237   /* Pass 2: process columns. */
 238   /* Note that we must descale the results by a factor of 8 == 2**3, */
 239   /* and also undo the PASS1_BITS scaling. */
 240
 241     dataptr = p_block;
 242     for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--)
 243     {
 244     /* Columns of zeroes can be exploited in the same way as we did with rows.
 245      * However, the row calculation has created many nonzero AC terms, so the
 246      * simplification applies less often (typically 5% to 10% of the time).
 247      * On machines with very fast multiplication, it's possible that the
 248      * test takes more time than it's worth.  In that case this section
 249      * may be commented out.
 250      */
 251
 252 #ifndef NO_ZERO_COLUMN_TEST /*ajoute un test mais evite des calculs */
 253         if ((dataptr[DCTSIZE*1] | dataptr[DCTSIZE*2] | dataptr[DCTSIZE*3] |
 254             dataptr[DCTSIZE*4] | dataptr[DCTSIZE*5] | dataptr[DCTSIZE*6] |
 255             dataptr[DCTSIZE*7]) == 0)
 256         {
 257       /* AC terms all zero */
 258             dctelem_t dcval = (dctelem_t) DESCALE((s32) dataptr[0], PASS1_BITS+3);
 259
 260             dataptr[DCTSIZE*0] = dcval;
 261             dataptr[DCTSIZE*1] = dcval;
 262             dataptr[DCTSIZE*2] = dcval;
 263             dataptr[DCTSIZE*3] = dcval;
 264             dataptr[DCTSIZE*4] = dcval;
 265             dataptr[DCTSIZE*5] = dcval;
 266             dataptr[DCTSIZE*6] = dcval;
 267             dataptr[DCTSIZE*7] = dcval;
 268
 269             dataptr++;          /* advance pointer to next column */
 270             continue;
 271         }
 272 #endif
 273
 274     /* Even part: reverse the even part of the forward DCT. */
 275     /* The rotator is sqrt(2)*c(-6). */
 276
 277         z2 = (s32) dataptr[DCTSIZE*2];
 278         z3 = (s32) dataptr[DCTSIZE*6];
 279
 280         z1 = MULTIPLY(z2 + z3, FIX(0.541196100));
 281         tmp2 = z1 + MULTIPLY(z3, - FIX(1.847759065));
 282         tmp3 = z1 + MULTIPLY(z2, FIX(0.765366865));
 283
 284         tmp0 = ((s32) dataptr[DCTSIZE*0] + (s32) dataptr[DCTSIZE*4]) << CONST_BITS;
 285         tmp1 = ((s32) dataptr[DCTSIZE*0] - (s32) dataptr[DCTSIZE*4]) << CONST_BITS;
 286
 287         tmp10 = tmp0 + tmp3;
 288         tmp13 = tmp0 - tmp3;
 289         tmp11 = tmp1 + tmp2;
 290         tmp12 = tmp1 - tmp2;
 291
 292     /* Odd part per figure 8; the matrix is unitary and hence its
 293      * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
 294      */
 295
 296         tmp0 = (s32) dataptr[DCTSIZE*7];
 297         tmp1 = (s32) dataptr[DCTSIZE*5];
 298         tmp2 = (s32) dataptr[DCTSIZE*3];
 299         tmp3 = (s32) dataptr[DCTSIZE*1];
 300
 301         z1 = tmp0 + tmp3;
 302         z2 = tmp1 + tmp2;
 303         z3 = tmp0 + tmp2;
 304         z4 = tmp1 + tmp3;
 305         z5 = MULTIPLY(z3 + z4, FIX(1.175875602)); /* sqrt(2) * c3 */
 306
 307         tmp0 = MULTIPLY(tmp0, FIX(0.298631336)); /* sqrt(2) * (-c1+c3+c5-c7) */
 308         tmp1 = MULTIPLY(tmp1, FIX(2.053119869)); /* sqrt(2) * ( c1+c3-c5+c7) */
 309         tmp2 = MULTIPLY(tmp2, FIX(3.072711026)); /* sqrt(2) * ( c1+c3+c5-c7) */
 310         tmp3 = MULTIPLY(tmp3, FIX(1.501321110)); /* sqrt(2) * ( c1+c3-c5-c7) */
 311         z1 = MULTIPLY(z1, - FIX(0.899976223)); /* sqrt(2) * (c7-c3) */
 312         z2 = MULTIPLY(z2, - FIX(2.562915447)); /* sqrt(2) * (-c1-c3) */
 313         z3 = MULTIPLY(z3, - FIX(1.961570560)); /* sqrt(2) * (-c3-c5) */
 314         z4 = MULTIPLY(z4, - FIX(0.390180644)); /* sqrt(2) * (c5-c3) */
 315
 316         z3 += z5;
 317         z4 += z5;
 318
 319         tmp0 += z1 + z3;
 320         tmp1 += z2 + z4;
 321         tmp2 += z2 + z3;
 322         tmp3 += z1 + z4;
 323
 324     /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
 325
 326         dataptr[DCTSIZE*0] = (dctelem_t) DESCALE(tmp10 + tmp3,
 327                                            CONST_BITS+PASS1_BITS+3);
 328         dataptr[DCTSIZE*7] = (dctelem_t) DESCALE(tmp10 - tmp3,
 329                                            CONST_BITS+PASS1_BITS+3);
 330         dataptr[DCTSIZE*1] = (dctelem_t) DESCALE(tmp11 + tmp2,
 331                                            CONST_BITS+PASS1_BITS+3);
 332         dataptr[DCTSIZE*6] = (dctelem_t) DESCALE(tmp11 - tmp2,
 333                                            CONST_BITS+PASS1_BITS+3);
 334         dataptr[DCTSIZE*2] = (dctelem_t) DESCALE(tmp12 + tmp1,
 335                                            CONST_BITS+PASS1_BITS+3);
 336         dataptr[DCTSIZE*5] = (dctelem_t) DESCALE(tmp12 - tmp1,
 337                                            CONST_BITS+PASS1_BITS+3);
 338         dataptr[DCTSIZE*3] = (dctelem_t) DESCALE(tmp13 + tmp0,
 339                                            CONST_BITS+PASS1_BITS+3);
 340         dataptr[DCTSIZE*4] = (dctelem_t) DESCALE(tmp13 - tmp0,
 341                                            CONST_BITS+PASS1_BITS+3);
 342
 343         dataptr++;                      /* advance pointer to next column */
 344     }
 345 }
 346