1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999-2001 VideoLAN
5 * $Id: idctmmx.c,v 1.22 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 IDCT module" )
54 ADD_CAPABILITY( IDCT, 150 )
55 ADD_REQUIREMENT( MMX )
57 ADD_SHORTCUT( "imdctmmx" )
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, c6, \
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 = C6 C4 C2 C4
120 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
122 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
124 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
125 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
127 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
128 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
131 static __inline__ void Row( dctelem_t * table, s32 * rounder )
133 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
134 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
136 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
137 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
139 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
140 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
142 paddd_m2r (*rounder, mm3); // mm3 += rounder
143 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
145 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
146 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
148 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
149 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
151 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
152 paddd_r2r (mm7, mm1); // mm1 = b1 b0
154 paddd_m2r (*rounder, mm0); // mm0 += rounder
155 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
157 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
158 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
160 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
161 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
163 paddd_r2r (mm6, mm5); // mm5 = b3 b2
164 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
166 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
167 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
170 static __inline__ void RowTail( dctelem_t * row, int store )
172 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
174 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
176 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
178 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
180 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
181 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
183 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
185 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
187 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
191 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
194 static __inline__ void RowMid( dctelem_t * row, int store,
195 int offset, dctelem_t * table )
197 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
198 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
200 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
201 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
203 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
204 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
206 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
207 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
209 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
210 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
212 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
213 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
215 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
216 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
218 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
219 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
221 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
222 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
224 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
225 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
228 static __inline__ void Col( dctelem_t * col, int offset )
235 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
236 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
237 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
238 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
239 static mmx_t scratch0, scratch1;
241 /* column code adapted from peter gubanov */
242 /* http://www.elecard.com/peter/idct.shtml */
244 movq_m2r (*_T1, mm0); // mm0 = T1
246 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
247 movq_r2r (mm0, mm2); // mm2 = T1
249 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
250 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
252 movq_m2r (*_T3, mm5); // mm5 = T3
253 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
255 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
256 movq_r2r (mm5, mm7); // mm7 = T3-1
258 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
259 psubsw_r2r (mm4, mm0); // mm0 = v17
261 movq_m2r (*_T2, mm4); // mm4 = T2
262 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
264 paddsw_r2r (mm2, mm1); // mm1 = u17
265 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
269 movq_r2r (mm4, mm2); // mm2 = T2
270 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
272 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
273 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
275 psubsw_r2r (mm6, mm5); // mm5 = v35
276 paddsw_r2r (mm3, mm7); // mm7 = u35
278 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
279 movq_r2r (mm0, mm6); // mm6 = v17
281 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
282 psubsw_r2r (mm5, mm0); // mm0 = b3
284 psubsw_r2r (mm3, mm4); // mm4 = v26
285 paddsw_r2r (mm6, mm5); // mm5 = v12
287 movq_r2m (mm0, scratch0); // save b3
288 movq_r2r (mm1, mm6); // mm6 = u17
290 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
291 paddsw_r2r (mm7, mm6); // mm6 = b0
293 psubsw_r2r (mm7, mm1); // mm1 = u12
294 movq_r2r (mm1, mm7); // mm7 = u12
296 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
297 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
299 movq_m2r (*_C4, mm0); // mm0 = C4/2
300 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
302 movq_r2m (mm6, scratch1); // save b0
303 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
305 movq_r2r (mm4, mm6); // mm6 = v26
306 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
308 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
309 movq_r2r (mm3, mm0); // mm0 = x0
311 psubsw_r2r (mm5, mm3); // mm3 = v04
312 paddsw_r2r (mm5, mm0); // mm0 = u04
314 paddsw_r2r (mm3, mm4); // mm4 = a1
315 movq_r2r (mm0, mm5); // mm5 = u04
317 psubsw_r2r (mm6, mm3); // mm3 = a2
318 paddsw_r2r (mm2, mm5); // mm5 = a0
320 paddsw_r2r (mm1, mm1); // mm1 = b1
321 psubsw_r2r (mm2, mm0); // mm0 = a3
323 paddsw_r2r (mm7, mm7); // mm7 = b2
324 movq_r2r (mm3, mm2); // mm2 = a2
326 movq_r2r (mm4, mm6); // mm6 = a1
327 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
329 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
330 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
332 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
333 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
335 movq_m2r (scratch1, mm1); // mm1 = b0
336 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
338 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
339 movq_r2r (mm5, mm7); // mm7 = a0
341 movq_r2m (mm4, *(col+offset+1*8)); // save y1
342 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
344 movq_r2m (mm3, *(col+offset+2*8)); // save y2
345 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
347 movq_m2r (scratch0, mm4); // mm4 = b3
348 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
350 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
351 movq_r2r (mm0, mm3); // mm3 = a3
353 movq_r2m (mm2, *(col+offset+5*8)); // save y5
354 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
356 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
357 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
359 movq_r2m (mm5, *(col+offset+0*8)); // save y0
360 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
362 movq_r2m (mm6, *(col+offset+6*8)); // save y6
363 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
365 movq_r2m (mm7, *(col+offset+7*8)); // save y7
367 movq_r2m (mm3, *(col+offset+4*8)); // save y4
369 movq_r2m (mm4, *(col+offset+3*8)); // save y3
373 static s32 rounder0[] ATTR_ALIGN(8) =
374 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
375 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
376 static s32 rounder1[] ATTR_ALIGN(8) =
377 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
378 static s32 rounder7[] ATTR_ALIGN(8) =
379 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
380 static s32 rounder2[] ATTR_ALIGN(8) =
381 rounder (0.60355339059); // C2 * (C6+C2)/2
382 static s32 rounder6[] ATTR_ALIGN(8) =
383 rounder (-0.25); // C2 * (C6-C2)/2
384 static s32 rounder3[] ATTR_ALIGN(8) =
385 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
386 static s32 rounder5[] ATTR_ALIGN(8) =
387 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
389 static void IDCT( dctelem_t * p_block )
391 static dctelem_t table04[] ATTR_ALIGN(16) =
392 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
393 static dctelem_t table17[] ATTR_ALIGN(16) =
394 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
395 static dctelem_t table26[] ATTR_ALIGN(16) =
396 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
397 static dctelem_t table35[] ATTR_ALIGN(16) =
398 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
400 RowHead( p_block, 0*8, table04 );
401 Row( table04, rounder0 );
402 RowMid( p_block, 0*8, 4*8, table04 );
403 Row( table04, rounder4 );
404 RowMid( p_block, 4*8, 1*8, table17 );
405 Row( table17, rounder1 );
406 RowMid( p_block, 1*8, 7*8, table17 );
407 Row( table17, rounder7 );
408 RowMid( p_block, 7*8, 2*8, table26 );
409 Row( table26, rounder2 );
410 RowMid( p_block, 2*8, 6*8, table26 );
411 Row( table26, rounder6 );
412 RowMid( p_block, 6*8, 3*8, table35 );
413 Row( table35, rounder3 );
414 RowMid( p_block, 3*8, 5*8, table35 );
415 Row( table35, rounder5 );
416 RowTail( p_block, 5*8);
422 #include "idct_sparse.h"
423 #include "idct_decl.h"