1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmx.c,v 1.9 2001/04/15 04:19:57 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 #define MODULE_NAME idctmmx
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 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 *****************************************************************************/
82 p_module->psz_name = MODULE_STRING;
83 p_module->psz_longname = "MMX 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 *****************************************************************************/
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 *****************************************************************************/
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 = _M( vdec_InitIDCT );
138 p_function_list->functions.idct.pf_sparse_idct = _M( vdec_SparseIDCT );
139 p_function_list->functions.idct.pf_idct = _M( 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_MMX ) )
150 if( TestMethod( IDCT_METHOD_VAR, "idctmmx" ) )
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, c6, \
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 = C6 C4 C2 C4
208 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
210 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
212 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
213 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
215 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
216 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
219 static __inline__ void Row( dctelem_t * table, s32 * rounder )
221 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
222 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
224 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
225 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
227 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
228 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
230 paddd_m2r (*rounder, mm3); // mm3 += rounder
231 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
233 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
234 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
236 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
237 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
239 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
240 paddd_r2r (mm7, mm1); // mm1 = b1 b0
242 paddd_m2r (*rounder, mm0); // mm0 += rounder
243 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
245 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
246 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
248 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
249 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
251 paddd_r2r (mm6, mm5); // mm5 = b3 b2
252 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
254 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
255 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
258 static __inline__ void RowTail( dctelem_t * row, int store )
260 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
262 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
264 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
266 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
268 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
269 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
271 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
273 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
275 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
279 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
282 static __inline__ void RowMid( dctelem_t * row, int store,
283 int offset, dctelem_t * table )
285 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
286 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
288 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
289 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
291 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
292 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
294 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
295 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
297 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
298 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
300 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
301 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
303 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
304 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
306 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
307 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
309 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
310 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
312 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
313 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
316 static __inline__ void Col( dctelem_t * col, int offset )
323 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
324 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
325 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
326 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
327 static mmx_t scratch0, scratch1;
329 /* column code adapted from peter gubanov */
330 /* http://www.elecard.com/peter/idct.shtml */
332 movq_m2r (*_T1, mm0); // mm0 = T1
334 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
335 movq_r2r (mm0, mm2); // mm2 = T1
337 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
338 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
340 movq_m2r (*_T3, mm5); // mm5 = T3
341 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
343 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
344 movq_r2r (mm5, mm7); // mm7 = T3-1
346 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
347 psubsw_r2r (mm4, mm0); // mm0 = v17
349 movq_m2r (*_T2, mm4); // mm4 = T2
350 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
352 paddsw_r2r (mm2, mm1); // mm1 = u17
353 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
357 movq_r2r (mm4, mm2); // mm2 = T2
358 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
360 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
361 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
363 psubsw_r2r (mm6, mm5); // mm5 = v35
364 paddsw_r2r (mm3, mm7); // mm7 = u35
366 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
367 movq_r2r (mm0, mm6); // mm6 = v17
369 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
370 psubsw_r2r (mm5, mm0); // mm0 = b3
372 psubsw_r2r (mm3, mm4); // mm4 = v26
373 paddsw_r2r (mm6, mm5); // mm5 = v12
375 movq_r2m (mm0, scratch0); // save b3
376 movq_r2r (mm1, mm6); // mm6 = u17
378 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
379 paddsw_r2r (mm7, mm6); // mm6 = b0
381 psubsw_r2r (mm7, mm1); // mm1 = u12
382 movq_r2r (mm1, mm7); // mm7 = u12
384 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
385 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
387 movq_m2r (*_C4, mm0); // mm0 = C4/2
388 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
390 movq_r2m (mm6, scratch1); // save b0
391 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
393 movq_r2r (mm4, mm6); // mm6 = v26
394 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
396 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
397 movq_r2r (mm3, mm0); // mm0 = x0
399 psubsw_r2r (mm5, mm3); // mm3 = v04
400 paddsw_r2r (mm5, mm0); // mm0 = u04
402 paddsw_r2r (mm3, mm4); // mm4 = a1
403 movq_r2r (mm0, mm5); // mm5 = u04
405 psubsw_r2r (mm6, mm3); // mm3 = a2
406 paddsw_r2r (mm2, mm5); // mm5 = a0
408 paddsw_r2r (mm1, mm1); // mm1 = b1
409 psubsw_r2r (mm2, mm0); // mm0 = a3
411 paddsw_r2r (mm7, mm7); // mm7 = b2
412 movq_r2r (mm3, mm2); // mm2 = a2
414 movq_r2r (mm4, mm6); // mm6 = a1
415 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
417 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
418 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
420 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
421 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
423 movq_m2r (scratch1, mm1); // mm1 = b0
424 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
426 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
427 movq_r2r (mm5, mm7); // mm7 = a0
429 movq_r2m (mm4, *(col+offset+1*8)); // save y1
430 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
432 movq_r2m (mm3, *(col+offset+2*8)); // save y2
433 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
435 movq_m2r (scratch0, mm4); // mm4 = b3
436 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
438 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
439 movq_r2r (mm0, mm3); // mm3 = a3
441 movq_r2m (mm2, *(col+offset+5*8)); // save y5
442 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
444 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
445 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
447 movq_r2m (mm5, *(col+offset+0*8)); // save y0
448 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
450 movq_r2m (mm6, *(col+offset+6*8)); // save y6
451 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
453 movq_r2m (mm7, *(col+offset+7*8)); // save y7
455 movq_r2m (mm3, *(col+offset+4*8)); // save y4
457 movq_r2m (mm4, *(col+offset+3*8)); // save y3
461 static s32 rounder0[] ATTR_ALIGN(8) =
462 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
463 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
464 static s32 rounder1[] ATTR_ALIGN(8) =
465 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
466 static s32 rounder7[] ATTR_ALIGN(8) =
467 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
468 static s32 rounder2[] ATTR_ALIGN(8) =
469 rounder (0.60355339059); // C2 * (C6+C2)/2
470 static s32 rounder6[] ATTR_ALIGN(8) =
471 rounder (-0.25); // C2 * (C6-C2)/2
472 static s32 rounder3[] ATTR_ALIGN(8) =
473 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
474 static s32 rounder5[] ATTR_ALIGN(8) =
475 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
477 void _M( vdec_IDCT )( vdec_thread_t * p_vdec, dctelem_t * p_block,
480 static dctelem_t table04[] ATTR_ALIGN(16) =
481 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
482 static dctelem_t table17[] ATTR_ALIGN(16) =
483 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
484 static dctelem_t table26[] ATTR_ALIGN(16) =
485 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
486 static dctelem_t table35[] ATTR_ALIGN(16) =
487 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
489 RowHead( p_block, 0*8, table04 );
490 Row( table04, rounder0 );
491 RowMid( p_block, 0*8, 4*8, table04 );
492 Row( table04, rounder4 );
493 RowMid( p_block, 4*8, 1*8, table17 );
494 Row( table17, rounder1 );
495 RowMid( p_block, 1*8, 7*8, table17 );
496 Row( table17, rounder7 );
497 RowMid( p_block, 7*8, 2*8, table26 );
498 Row( table26, rounder2 );
499 RowMid( p_block, 2*8, 6*8, table26 );
500 Row( table26, rounder6 );
501 RowMid( p_block, 6*8, 3*8, table35 );
502 Row( table35, rounder3 );
503 RowMid( p_block, 3*8, 5*8, table35 );
504 Row( table35, rounder5 );
505 RowTail( p_block, 5*8);