1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999-2001 VideoLAN
5 * $Id: idctmmx.c,v 1.26 2002/05/18 17:47:46 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 * NormScan : This IDCT uses reordered coeffs, so we patch the scan table
71 *****************************************************************************/
72 static void NormScan( u8 ppi_scan[2][64] )
76 for( i = 0; i < 64; i++ )
79 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
82 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
86 /*****************************************************************************
88 *****************************************************************************/
92 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
93 #define rounder(bias) {round (bias), round (bias)}
95 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \
104 static inline void RowHead( dctelem_t * row, int offset, dctelem_t * table )
106 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
108 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
109 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
111 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
112 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
114 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
116 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
117 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
119 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
120 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
123 static inline void Row( dctelem_t * table, s32 * rounder )
125 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
126 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
128 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
129 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
131 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
132 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
134 paddd_m2r (*rounder, mm3); // mm3 += rounder
135 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
137 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
138 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
140 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
141 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
143 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
144 paddd_r2r (mm7, mm1); // mm1 = b1 b0
146 paddd_m2r (*rounder, mm0); // mm0 += rounder
147 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
149 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
150 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
152 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
153 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
155 paddd_r2r (mm6, mm5); // mm5 = b3 b2
156 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
158 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
159 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
162 static inline void RowTail( dctelem_t * row, int store )
164 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
166 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
168 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
170 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
172 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
173 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
175 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
177 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
179 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
183 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
186 static inline void RowMid( dctelem_t * row, int store,
187 int offset, dctelem_t * table )
189 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
190 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
192 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
193 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
195 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
196 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
198 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
199 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
201 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
202 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
204 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
205 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
207 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
208 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
210 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
211 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
213 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
214 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
216 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
217 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
220 static inline void Col( dctelem_t * col, int offset )
227 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
228 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
229 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
230 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
231 static mmx_t scratch0, scratch1;
233 /* column code adapted from peter gubanov */
234 /* http://www.elecard.com/peter/idct.shtml */
236 movq_m2r (*_T1, mm0); // mm0 = T1
238 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
239 movq_r2r (mm0, mm2); // mm2 = T1
241 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
242 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
244 movq_m2r (*_T3, mm5); // mm5 = T3
245 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
247 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
248 movq_r2r (mm5, mm7); // mm7 = T3-1
250 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
251 psubsw_r2r (mm4, mm0); // mm0 = v17
253 movq_m2r (*_T2, mm4); // mm4 = T2
254 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
256 paddsw_r2r (mm2, mm1); // mm1 = u17
257 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
261 movq_r2r (mm4, mm2); // mm2 = T2
262 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
264 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
265 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
267 psubsw_r2r (mm6, mm5); // mm5 = v35
268 paddsw_r2r (mm3, mm7); // mm7 = u35
270 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
271 movq_r2r (mm0, mm6); // mm6 = v17
273 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
274 psubsw_r2r (mm5, mm0); // mm0 = b3
276 psubsw_r2r (mm3, mm4); // mm4 = v26
277 paddsw_r2r (mm6, mm5); // mm5 = v12
279 movq_r2m (mm0, scratch0); // save b3
280 movq_r2r (mm1, mm6); // mm6 = u17
282 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
283 paddsw_r2r (mm7, mm6); // mm6 = b0
285 psubsw_r2r (mm7, mm1); // mm1 = u12
286 movq_r2r (mm1, mm7); // mm7 = u12
288 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
289 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
291 movq_m2r (*_C4, mm0); // mm0 = C4/2
292 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
294 movq_r2m (mm6, scratch1); // save b0
295 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
297 movq_r2r (mm4, mm6); // mm6 = v26
298 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
300 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
301 movq_r2r (mm3, mm0); // mm0 = x0
303 psubsw_r2r (mm5, mm3); // mm3 = v04
304 paddsw_r2r (mm5, mm0); // mm0 = u04
306 paddsw_r2r (mm3, mm4); // mm4 = a1
307 movq_r2r (mm0, mm5); // mm5 = u04
309 psubsw_r2r (mm6, mm3); // mm3 = a2
310 paddsw_r2r (mm2, mm5); // mm5 = a0
312 paddsw_r2r (mm1, mm1); // mm1 = b1
313 psubsw_r2r (mm2, mm0); // mm0 = a3
315 paddsw_r2r (mm7, mm7); // mm7 = b2
316 movq_r2r (mm3, mm2); // mm2 = a2
318 movq_r2r (mm4, mm6); // mm6 = a1
319 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
321 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
322 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
324 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
325 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
327 movq_m2r (scratch1, mm1); // mm1 = b0
328 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
330 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
331 movq_r2r (mm5, mm7); // mm7 = a0
333 movq_r2m (mm4, *(col+offset+1*8)); // save y1
334 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
336 movq_r2m (mm3, *(col+offset+2*8)); // save y2
337 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
339 movq_m2r (scratch0, mm4); // mm4 = b3
340 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
342 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
343 movq_r2r (mm0, mm3); // mm3 = a3
345 movq_r2m (mm2, *(col+offset+5*8)); // save y5
346 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
348 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
349 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
351 movq_r2m (mm5, *(col+offset+0*8)); // save y0
352 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
354 movq_r2m (mm6, *(col+offset+6*8)); // save y6
355 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
357 movq_r2m (mm7, *(col+offset+7*8)); // save y7
359 movq_r2m (mm3, *(col+offset+4*8)); // save y4
361 movq_r2m (mm4, *(col+offset+3*8)); // save y3
365 static s32 rounder0[] ATTR_ALIGN(8) =
366 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
367 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
368 static s32 rounder1[] ATTR_ALIGN(8) =
369 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
370 static s32 rounder7[] ATTR_ALIGN(8) =
371 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
372 static s32 rounder2[] ATTR_ALIGN(8) =
373 rounder (0.60355339059); // C2 * (C6+C2)/2
374 static s32 rounder6[] ATTR_ALIGN(8) =
375 rounder (-0.25); // C2 * (C6-C2)/2
376 static s32 rounder3[] ATTR_ALIGN(8) =
377 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
378 static s32 rounder5[] ATTR_ALIGN(8) =
379 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
381 static void IDCT( dctelem_t * p_block )
383 static dctelem_t table04[] ATTR_ALIGN(16) =
384 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
385 static dctelem_t table17[] ATTR_ALIGN(16) =
386 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
387 static dctelem_t table26[] ATTR_ALIGN(16) =
388 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
389 static dctelem_t table35[] ATTR_ALIGN(16) =
390 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
392 RowHead( p_block, 0*8, table04 );
393 Row( table04, rounder0 );
394 RowMid( p_block, 0*8, 4*8, table04 );
395 Row( table04, rounder4 );
396 RowMid( p_block, 4*8, 1*8, table17 );
397 Row( table17, rounder1 );
398 RowMid( p_block, 1*8, 7*8, table17 );
399 Row( table17, rounder7 );
400 RowMid( p_block, 7*8, 2*8, table26 );
401 Row( table26, rounder2 );
402 RowMid( p_block, 2*8, 6*8, table26 );
403 Row( table26, rounder6 );
404 RowMid( p_block, 6*8, 3*8, table35 );
405 Row( table35, rounder3 );
406 RowMid( p_block, 3*8, 5*8, table35 );
407 Row( table35, rounder5 );
408 RowTail( p_block, 5*8);
414 static inline void RestoreCPUState( )
416 /* reenables the FPU */
417 __asm__ __volatile__ ("emms");
420 #include "idct_sparse.h"
421 #include "idct_decl.h"