1 /*****************************************************************************
2 * idctmmxext.c : MMX EXT IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmxext.c,v 1.4 2001/01/17 18:17:30 massiot 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 #define MODULE_NAME idctmmxext
30 /*****************************************************************************
32 *****************************************************************************/
41 #include "tests.h" /* TestCPU() */
44 #include "video_output.h"
46 #include "video_decoder.h"
49 #include "modules_inner.h"
53 #include "attributes.h"
56 /*****************************************************************************
58 *****************************************************************************/
59 static void idct_getfunctions( function_list_t * p_function_list );
60 static int idct_Probe ( probedata_t *p_data );
61 static void vdec_NormScan ( u8 ppi_scan[2][64] );
64 /*****************************************************************************
65 * Build configuration tree.
66 *****************************************************************************/
68 ADD_WINDOW( "Configuration for MMX EXT IDCT module" )
69 ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
72 /*****************************************************************************
73 * InitModule: get the module structure and configuration.
74 *****************************************************************************
75 * We have to fill psz_name, psz_longname and psz_version. These variables
76 * will be strdup()ed later by the main application because the module can
77 * be unloaded later to save memory, and we want to be able to access this
78 * data even after the module has been unloaded.
79 *****************************************************************************/
80 int InitModule( module_t * p_module )
82 p_module->psz_name = MODULE_STRING;
83 p_module->psz_longname = "MMX EXT IDCT module";
84 p_module->psz_version = VERSION;
86 p_module->i_capabilities = MODULE_CAPABILITY_NULL
87 | MODULE_CAPABILITY_IDCT;
92 /*****************************************************************************
93 * ActivateModule: set the module to an usable state.
94 *****************************************************************************
95 * This function fills the capability functions and the configuration
96 * structure. Once ActivateModule() has been called, the i_usage can
97 * be set to 0 and calls to NeedModule() be made to increment it. To unload
98 * the module, one has to wait until i_usage == 0 and call DeactivateModule().
99 *****************************************************************************/
100 int ActivateModule( module_t * p_module )
102 p_module->p_functions = malloc( sizeof( module_functions_t ) );
103 if( p_module->p_functions == NULL )
108 idct_getfunctions( &p_module->p_functions->idct );
110 p_module->p_config = p_config;
115 /*****************************************************************************
116 * DeactivateModule: make sure the module can be unloaded.
117 *****************************************************************************
118 * This function must only be called when i_usage == 0. If it successfully
119 * returns, i_usage can be set to -1 and the module unloaded. Be careful to
120 * lock usage_lock during the whole process.
121 *****************************************************************************/
122 int DeactivateModule( module_t * p_module )
124 free( p_module->p_functions );
129 /* Following functions are local */
131 /*****************************************************************************
132 * Functions exported as capabilities.
133 *****************************************************************************/
134 static void idct_getfunctions( function_list_t * p_function_list )
136 p_function_list->pf_probe = idct_Probe;
137 p_function_list->functions.idct.pf_init = vdec_InitIDCT;
138 p_function_list->functions.idct.pf_sparse_idct = vdec_SparseIDCT;
139 p_function_list->functions.idct.pf_idct = vdec_IDCT;
140 p_function_list->functions.idct.pf_norm_scan = vdec_NormScan;
143 /*****************************************************************************
144 * idct_Probe: return a preference score
145 *****************************************************************************/
146 static int idct_Probe( probedata_t *p_data )
148 if( TestCPU( CPU_CAPABILITY_MMXEXT ) )
150 if( TestMethod( IDCT_METHOD_VAR, "idctmmxext" ) )
165 /*****************************************************************************
166 * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table
167 *****************************************************************************/
168 static void vdec_NormScan( u8 ppi_scan[2][64] )
172 for( i = 0; i < 64; i++ )
175 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
178 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
182 /*****************************************************************************
184 *****************************************************************************/
188 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
189 #define rounder(bias) {round (bias), round (bias)}
191 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, -c4, -c2, \
200 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
202 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
204 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
205 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
207 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
208 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
210 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
211 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
213 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
216 static __inline__ void Row( dctelem_t * table, s32 * rounder )
218 movq_m2r (*(table+8), mm1); // mm1 = -C5 -C1 C3 C1
219 pmaddwd_r2r (mm2, mm4); // mm4 = C4*x0+C6*x2 C4*x4+C6*x6
221 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x4-C6*x6 C4*x0-C6*x2
222 pshufw_r2r (mm6, mm6, 0x4e); // mm6 = x3 x1 x7 x5
224 movq_m2r (*(table+12), mm7); // mm7 = -C7 C3 C7 C5
225 pmaddwd_r2r (mm5, mm1); // mm1 = -C1*x5-C5*x7 C1*x1+C3*x3
227 paddd_m2r (*rounder, mm3); // mm3 += rounder
228 pmaddwd_r2r (mm6, mm7); // mm7 = C3*x1-C7*x3 C5*x5+C7*x7
230 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x0-C2*x2 -C4*x4+C2*x6
231 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
233 pmaddwd_m2r (*(table+24), mm5); // mm5 = C3*x5-C1*x7 C5*x1-C1*x3
234 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
236 pmaddwd_m2r (*(table+28), mm6); // mm6 = C7*x1-C5*x3 C7*x5+C3*x7
237 paddd_r2r (mm7, mm1); // mm1 = b1 b0
239 paddd_m2r (*rounder, mm0); // mm0 += rounder
240 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
242 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
243 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
245 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
246 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
248 paddd_r2r (mm6, mm5); // mm5 = b3 b2
249 movq_r2r (mm0, mm4); // mm4 = a3 a2 + rounder
251 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
252 psubd_r2r (mm5, mm4); // mm4 = a3-b3 a2-b2 + rounder
255 static __inline__ void RowTail( dctelem_t * row, int store )
257 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
259 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
261 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
263 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
265 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
266 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
269 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
272 static __inline__ void RowMid( dctelem_t * row, int store,
273 int offset, dctelem_t * table)
275 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
276 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
278 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
279 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
281 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
282 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
284 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
285 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
287 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
288 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
290 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
291 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
293 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
295 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
296 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
299 static __inline__ void Col( dctelem_t * col, int offset )
306 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
307 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
308 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
309 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
310 static mmx_t scratch0, scratch1;
312 /* column code adapted from peter gubanov */
313 /* http://www.elecard.com/peter/idct.shtml */
315 movq_m2r (*_T1, mm0); // mm0 = T1
317 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
318 movq_r2r (mm0, mm2); // mm2 = T1
320 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
321 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
323 movq_m2r (*_T3, mm5); // mm5 = T3
324 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
326 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
327 movq_r2r (mm5, mm7); // mm7 = T3-1
329 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
330 psubsw_r2r (mm4, mm0); // mm0 = v17
332 movq_m2r (*_T2, mm4); // mm4 = T2
333 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
335 paddsw_r2r (mm2, mm1); // mm1 = u17
336 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
340 movq_r2r (mm4, mm2); // mm2 = T2
341 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
343 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
344 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
346 psubsw_r2r (mm6, mm5); // mm5 = v35
347 paddsw_r2r (mm3, mm7); // mm7 = u35
349 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
350 movq_r2r (mm0, mm6); // mm6 = v17
352 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
353 psubsw_r2r (mm5, mm0); // mm0 = b3
355 psubsw_r2r (mm3, mm4); // mm4 = v26
356 paddsw_r2r (mm6, mm5); // mm5 = v12
358 movq_r2m (mm0, scratch0); // save b3
359 movq_r2r (mm1, mm6); // mm6 = u17
361 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
362 paddsw_r2r (mm7, mm6); // mm6 = b0
364 psubsw_r2r (mm7, mm1); // mm1 = u12
365 movq_r2r (mm1, mm7); // mm7 = u12
367 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
368 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
370 movq_m2r (*_C4, mm0); // mm0 = C4/2
371 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
373 movq_r2m (mm6, scratch1); // save b0
374 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
376 movq_r2r (mm4, mm6); // mm6 = v26
377 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
379 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
380 movq_r2r (mm3, mm0); // mm0 = x0
382 psubsw_r2r (mm5, mm3); // mm3 = v04
383 paddsw_r2r (mm5, mm0); // mm0 = u04
385 paddsw_r2r (mm3, mm4); // mm4 = a1
386 movq_r2r (mm0, mm5); // mm5 = u04
388 psubsw_r2r (mm6, mm3); // mm3 = a2
389 paddsw_r2r (mm2, mm5); // mm5 = a0
391 paddsw_r2r (mm1, mm1); // mm1 = b1
392 psubsw_r2r (mm2, mm0); // mm0 = a3
394 paddsw_r2r (mm7, mm7); // mm7 = b2
395 movq_r2r (mm3, mm2); // mm2 = a2
397 movq_r2r (mm4, mm6); // mm6 = a1
398 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
400 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
401 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
403 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
404 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
406 movq_m2r (scratch1, mm1); // mm1 = b0
407 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
409 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
410 movq_r2r (mm5, mm7); // mm7 = a0
412 movq_r2m (mm4, *(col+offset+1*8)); // save y1
413 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
415 movq_r2m (mm3, *(col+offset+2*8)); // save y2
416 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
418 movq_m2r (scratch0, mm4); // mm4 = b3
419 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
421 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
422 movq_r2r (mm0, mm3); // mm3 = a3
424 movq_r2m (mm2, *(col+offset+5*8)); // save y5
425 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
427 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
428 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
430 movq_r2m (mm5, *(col+offset+0*8)); // save y0
431 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
433 movq_r2m (mm6, *(col+offset+6*8)); // save y6
434 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
436 movq_r2m (mm7, *(col+offset+7*8)); // save y7
438 movq_r2m (mm3, *(col+offset+4*8)); // save y4
440 movq_r2m (mm4, *(col+offset+3*8)); // save y3
444 static s32 rounder0[] ATTR_ALIGN(8) =
445 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
446 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
447 static s32 rounder1[] ATTR_ALIGN(8) =
448 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
449 static s32 rounder7[] ATTR_ALIGN(8) =
450 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
451 static s32 rounder2[] ATTR_ALIGN(8) =
452 rounder (0.60355339059); // C2 * (C6+C2)/2
453 static s32 rounder6[] ATTR_ALIGN(8) =
454 rounder (-0.25); // C2 * (C6-C2)/2
455 static s32 rounder3[] ATTR_ALIGN(8) =
456 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
457 static s32 rounder5[] ATTR_ALIGN(8) =
458 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
460 void vdec_IDCT( vdec_thread_t * p_vdec, dctelem_t * p_block,
463 static dctelem_t table04[] ATTR_ALIGN(16) =
464 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
465 static dctelem_t table17[] ATTR_ALIGN(16) =
466 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
467 static dctelem_t table26[] ATTR_ALIGN(16) =
468 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
469 static dctelem_t table35[] ATTR_ALIGN(16) =
470 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
472 RowHead( p_block, 0*8, table04 );
473 Row( table04, rounder0 );
474 RowMid( p_block, 0*8, 4*8, table04 );
475 Row( table04, rounder4 );
476 RowMid( p_block, 4*8, 1*8, table17 );
477 Row( table17, rounder1 );
478 RowMid( p_block, 1*8, 7*8, table17 );
479 Row( table17, rounder7 );
480 RowMid( p_block, 7*8, 2*8, table26 );
481 Row( table26, rounder2 );
482 RowMid( p_block, 2*8, 6*8, table26 );
483 Row( table26, rounder6 );
484 RowMid( p_block, 6*8, 3*8, table35 );
485 Row( table35, rounder3 );
486 RowMid( p_block, 3*8, 5*8, table35 );
487 Row( table35, rounder5 );
488 RowTail( p_block, 5*8);