1 /*****************************************************************************
2 * idctmmxext.c : MMX EXT IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999-2001 VideoLAN
5 * $Id: idctmmxext.c,v 1.19 2001/12/30 07:09:55 sam Exp $
7 * Authors: Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
8 * Michel Lespinasse <walken@zoy.org>
9 * Peter Gubanov <peter@elecard.net.ru>
10 * (from the LiViD project)
11 * Christophe Massiot <massiot@via.ecp.fr>
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
23 * You should have received a copy of the GNU General Public License
24 * along with this program; if not, write to the Free Software
25 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
26 *****************************************************************************/
28 /*****************************************************************************
30 *****************************************************************************/
34 #include <videolan/vlc.h>
39 #include "block_mmx.h"
41 /*****************************************************************************
43 *****************************************************************************/
44 static void idct_getfunctions( function_list_t * p_function_list );
46 /*****************************************************************************
47 * Build configuration tree.
48 *****************************************************************************/
53 SET_DESCRIPTION( "MMX EXT IDCT module" )
54 ADD_CAPABILITY( IDCT, 200 )
55 ADD_REQUIREMENT( MMXEXT )
56 ADD_SHORTCUT( "mmxext" )
57 ADD_SHORTCUT( "idctmmxext" )
61 idct_getfunctions( &p_module->p_functions->idct );
64 MODULE_DEACTIVATE_START
65 MODULE_DEACTIVATE_STOP
67 /* Following functions are local */
69 /*****************************************************************************
70 * idct_Probe: return a preference score
71 *****************************************************************************/
72 static int idct_Probe( probedata_t *p_data )
77 /*****************************************************************************
78 * NormScan : This IDCT uses reordered coeffs, so we patch the scan table
79 *****************************************************************************/
80 static void NormScan( u8 ppi_scan[2][64] )
84 for( i = 0; i < 64; i++ )
87 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
90 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
94 /*****************************************************************************
96 *****************************************************************************/
100 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
101 #define rounder(bias) {round (bias), round (bias)}
103 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, -c4, -c2, \
112 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
114 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
116 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
117 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
119 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
120 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
122 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
123 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
125 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
128 static __inline__ void Row( dctelem_t * table, s32 * rounder )
130 movq_m2r (*(table+8), mm1); // mm1 = -C5 -C1 C3 C1
131 pmaddwd_r2r (mm2, mm4); // mm4 = C4*x0+C6*x2 C4*x4+C6*x6
133 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x4-C6*x6 C4*x0-C6*x2
134 pshufw_r2r (mm6, mm6, 0x4e); // mm6 = x3 x1 x7 x5
136 movq_m2r (*(table+12), mm7); // mm7 = -C7 C3 C7 C5
137 pmaddwd_r2r (mm5, mm1); // mm1 = -C1*x5-C5*x7 C1*x1+C3*x3
139 paddd_m2r (*rounder, mm3); // mm3 += rounder
140 pmaddwd_r2r (mm6, mm7); // mm7 = C3*x1-C7*x3 C5*x5+C7*x7
142 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x0-C2*x2 -C4*x4+C2*x6
143 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
145 pmaddwd_m2r (*(table+24), mm5); // mm5 = C3*x5-C1*x7 C5*x1-C1*x3
146 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
148 pmaddwd_m2r (*(table+28), mm6); // mm6 = C7*x1-C5*x3 C7*x5+C3*x7
149 paddd_r2r (mm7, mm1); // mm1 = b1 b0
151 paddd_m2r (*rounder, mm0); // mm0 += rounder
152 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
154 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
155 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
157 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
158 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
160 paddd_r2r (mm6, mm5); // mm5 = b3 b2
161 movq_r2r (mm0, mm4); // mm4 = a3 a2 + rounder
163 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
164 psubd_r2r (mm5, mm4); // mm4 = a3-b3 a2-b2 + rounder
167 static __inline__ void RowTail( dctelem_t * row, int store )
169 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
171 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
173 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
175 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
177 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
178 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
181 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
184 static __inline__ void RowMid( dctelem_t * row, int store,
185 int offset, dctelem_t * table)
187 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
188 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
190 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
191 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
193 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
194 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
196 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
197 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
199 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
200 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
202 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
203 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
205 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
207 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
208 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
211 static __inline__ void Col( dctelem_t * col, int offset )
218 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
219 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
220 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
221 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
222 static mmx_t scratch0, scratch1;
224 /* column code adapted from peter gubanov */
225 /* http://www.elecard.com/peter/idct.shtml */
227 movq_m2r (*_T1, mm0); // mm0 = T1
229 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
230 movq_r2r (mm0, mm2); // mm2 = T1
232 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
233 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
235 movq_m2r (*_T3, mm5); // mm5 = T3
236 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
238 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
239 movq_r2r (mm5, mm7); // mm7 = T3-1
241 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
242 psubsw_r2r (mm4, mm0); // mm0 = v17
244 movq_m2r (*_T2, mm4); // mm4 = T2
245 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
247 paddsw_r2r (mm2, mm1); // mm1 = u17
248 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
252 movq_r2r (mm4, mm2); // mm2 = T2
253 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
255 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
256 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
258 psubsw_r2r (mm6, mm5); // mm5 = v35
259 paddsw_r2r (mm3, mm7); // mm7 = u35
261 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
262 movq_r2r (mm0, mm6); // mm6 = v17
264 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
265 psubsw_r2r (mm5, mm0); // mm0 = b3
267 psubsw_r2r (mm3, mm4); // mm4 = v26
268 paddsw_r2r (mm6, mm5); // mm5 = v12
270 movq_r2m (mm0, scratch0); // save b3
271 movq_r2r (mm1, mm6); // mm6 = u17
273 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
274 paddsw_r2r (mm7, mm6); // mm6 = b0
276 psubsw_r2r (mm7, mm1); // mm1 = u12
277 movq_r2r (mm1, mm7); // mm7 = u12
279 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
280 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
282 movq_m2r (*_C4, mm0); // mm0 = C4/2
283 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
285 movq_r2m (mm6, scratch1); // save b0
286 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
288 movq_r2r (mm4, mm6); // mm6 = v26
289 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
291 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
292 movq_r2r (mm3, mm0); // mm0 = x0
294 psubsw_r2r (mm5, mm3); // mm3 = v04
295 paddsw_r2r (mm5, mm0); // mm0 = u04
297 paddsw_r2r (mm3, mm4); // mm4 = a1
298 movq_r2r (mm0, mm5); // mm5 = u04
300 psubsw_r2r (mm6, mm3); // mm3 = a2
301 paddsw_r2r (mm2, mm5); // mm5 = a0
303 paddsw_r2r (mm1, mm1); // mm1 = b1
304 psubsw_r2r (mm2, mm0); // mm0 = a3
306 paddsw_r2r (mm7, mm7); // mm7 = b2
307 movq_r2r (mm3, mm2); // mm2 = a2
309 movq_r2r (mm4, mm6); // mm6 = a1
310 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
312 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
313 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
315 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
316 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
318 movq_m2r (scratch1, mm1); // mm1 = b0
319 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
321 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
322 movq_r2r (mm5, mm7); // mm7 = a0
324 movq_r2m (mm4, *(col+offset+1*8)); // save y1
325 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
327 movq_r2m (mm3, *(col+offset+2*8)); // save y2
328 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
330 movq_m2r (scratch0, mm4); // mm4 = b3
331 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
333 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
334 movq_r2r (mm0, mm3); // mm3 = a3
336 movq_r2m (mm2, *(col+offset+5*8)); // save y5
337 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
339 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
340 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
342 movq_r2m (mm5, *(col+offset+0*8)); // save y0
343 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
345 movq_r2m (mm6, *(col+offset+6*8)); // save y6
346 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
348 movq_r2m (mm7, *(col+offset+7*8)); // save y7
350 movq_r2m (mm3, *(col+offset+4*8)); // save y4
352 movq_r2m (mm4, *(col+offset+3*8)); // save y3
356 static s32 rounder0[] ATTR_ALIGN(8) =
357 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
358 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
359 static s32 rounder1[] ATTR_ALIGN(8) =
360 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
361 static s32 rounder7[] ATTR_ALIGN(8) =
362 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
363 static s32 rounder2[] ATTR_ALIGN(8) =
364 rounder (0.60355339059); // C2 * (C6+C2)/2
365 static s32 rounder6[] ATTR_ALIGN(8) =
366 rounder (-0.25); // C2 * (C6-C2)/2
367 static s32 rounder3[] ATTR_ALIGN(8) =
368 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
369 static s32 rounder5[] ATTR_ALIGN(8) =
370 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
372 static void IDCT( dctelem_t * p_block )
374 static dctelem_t table04[] ATTR_ALIGN(16) =
375 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
376 static dctelem_t table17[] ATTR_ALIGN(16) =
377 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
378 static dctelem_t table26[] ATTR_ALIGN(16) =
379 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
380 static dctelem_t table35[] ATTR_ALIGN(16) =
381 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
383 RowHead( p_block, 0*8, table04 );
384 Row( table04, rounder0 );
385 RowMid( p_block, 0*8, 4*8, table04 );
386 Row( table04, rounder4 );
387 RowMid( p_block, 4*8, 1*8, table17 );
388 Row( table17, rounder1 );
389 RowMid( p_block, 1*8, 7*8, table17 );
390 Row( table17, rounder7 );
391 RowMid( p_block, 7*8, 2*8, table26 );
392 Row( table26, rounder2 );
393 RowMid( p_block, 2*8, 6*8, table26 );
394 Row( table26, rounder6 );
395 RowMid( p_block, 6*8, 3*8, table35 );
396 Row( table35, rounder3 );
397 RowMid( p_block, 3*8, 5*8, table35 );
398 Row( table35, rounder5 );
399 RowTail( p_block, 5*8);
405 #include "idct_sparse.h"
406 #include "idct_decl.h"