1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmx.c,v 1.10 2001/05/06 04:32:02 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"
51 #include "vdec_block.h"
52 #include "vdec_idct.h"
54 #include "attributes.h"
57 /*****************************************************************************
59 *****************************************************************************/
60 static void idct_getfunctions( function_list_t * p_function_list );
61 static int idct_Probe ( probedata_t *p_data );
62 static void vdec_NormScan ( u8 ppi_scan[2][64] );
65 /*****************************************************************************
66 * Build configuration tree.
67 *****************************************************************************/
69 ADD_WINDOW( "Configuration for MMX IDCT module" )
70 ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
73 /*****************************************************************************
74 * InitModule: get the module structure and configuration.
75 *****************************************************************************
76 * We have to fill psz_name, psz_longname and psz_version. These variables
77 * will be strdup()ed later by the main application because the module can
78 * be unloaded later to save memory, and we want to be able to access this
79 * data even after the module has been unloaded.
80 *****************************************************************************/
83 p_module->psz_name = MODULE_STRING;
84 p_module->psz_longname = "MMX IDCT module";
85 p_module->psz_version = VERSION;
87 p_module->i_capabilities = MODULE_CAPABILITY_NULL
88 | MODULE_CAPABILITY_IDCT;
93 /*****************************************************************************
94 * ActivateModule: set the module to an usable state.
95 *****************************************************************************
96 * This function fills the capability functions and the configuration
97 * structure. Once ActivateModule() has been called, the i_usage can
98 * be set to 0 and calls to NeedModule() be made to increment it. To unload
99 * the module, one has to wait until i_usage == 0 and call DeactivateModule().
100 *****************************************************************************/
103 p_module->p_functions = malloc( sizeof( module_functions_t ) );
104 if( p_module->p_functions == NULL )
109 idct_getfunctions( &p_module->p_functions->idct );
111 p_module->p_config = p_config;
116 /*****************************************************************************
117 * DeactivateModule: make sure the module can be unloaded.
118 *****************************************************************************
119 * This function must only be called when i_usage == 0. If it successfully
120 * returns, i_usage can be set to -1 and the module unloaded. Be careful to
121 * lock usage_lock during the whole process.
122 *****************************************************************************/
125 free( p_module->p_functions );
130 /* Following functions are local */
132 /*****************************************************************************
133 * Functions exported as capabilities.
134 *****************************************************************************/
135 static void idct_getfunctions( function_list_t * p_function_list )
137 p_function_list->pf_probe = idct_Probe;
138 #define F p_function_list->functions.idct
139 F.pf_idct_init = _M( vdec_InitIDCT );
140 F.pf_sparse_idct = _M( vdec_SparseIDCT );
141 F.pf_idct = _M( vdec_IDCT );
142 F.pf_norm_scan = vdec_NormScan;
143 F.pf_vdec_init = _M( vdec_Init );
144 F.pf_decode_mb_c = _M( vdec_DecodeMacroblockC );
145 F.pf_decode_mb_bw = _M( vdec_DecodeMacroblockBW );
149 /*****************************************************************************
150 * idct_Probe: return a preference score
151 *****************************************************************************/
152 static int idct_Probe( probedata_t *p_data )
154 if( TestCPU( CPU_CAPABILITY_MMX ) )
156 if( TestMethod( IDCT_METHOD_VAR, "idctmmx" ) )
171 /*****************************************************************************
172 * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table
173 *****************************************************************************/
174 static void vdec_NormScan( u8 ppi_scan[2][64] )
178 for( i = 0; i < 64; i++ )
181 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
184 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
188 /*****************************************************************************
190 *****************************************************************************/
194 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
195 #define rounder(bias) {round (bias), round (bias)}
197 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \
206 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
208 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
210 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
211 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
213 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
214 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
216 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
218 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
219 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
221 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
222 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
225 static __inline__ void Row( dctelem_t * table, s32 * rounder )
227 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
228 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
230 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
231 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
233 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
234 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
236 paddd_m2r (*rounder, mm3); // mm3 += rounder
237 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
239 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
240 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
242 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
243 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
245 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
246 paddd_r2r (mm7, mm1); // mm1 = b1 b0
248 paddd_m2r (*rounder, mm0); // mm0 += rounder
249 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
251 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
252 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
254 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
255 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
257 paddd_r2r (mm6, mm5); // mm5 = b3 b2
258 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
260 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
261 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
264 static __inline__ void RowTail( dctelem_t * row, int store )
266 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
268 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
270 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
272 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
274 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
275 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
277 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
279 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
281 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
285 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
288 static __inline__ void RowMid( dctelem_t * row, int store,
289 int offset, dctelem_t * table )
291 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
292 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
294 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
295 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
297 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
298 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
300 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
301 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
303 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
304 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
306 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
307 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
309 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
310 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
312 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
313 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
315 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
316 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
318 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
319 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
322 static __inline__ void Col( dctelem_t * col, int offset )
329 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
330 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
331 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
332 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
333 static mmx_t scratch0, scratch1;
335 /* column code adapted from peter gubanov */
336 /* http://www.elecard.com/peter/idct.shtml */
338 movq_m2r (*_T1, mm0); // mm0 = T1
340 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
341 movq_r2r (mm0, mm2); // mm2 = T1
343 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
344 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
346 movq_m2r (*_T3, mm5); // mm5 = T3
347 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
349 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
350 movq_r2r (mm5, mm7); // mm7 = T3-1
352 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
353 psubsw_r2r (mm4, mm0); // mm0 = v17
355 movq_m2r (*_T2, mm4); // mm4 = T2
356 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
358 paddsw_r2r (mm2, mm1); // mm1 = u17
359 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
363 movq_r2r (mm4, mm2); // mm2 = T2
364 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
366 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
367 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
369 psubsw_r2r (mm6, mm5); // mm5 = v35
370 paddsw_r2r (mm3, mm7); // mm7 = u35
372 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
373 movq_r2r (mm0, mm6); // mm6 = v17
375 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
376 psubsw_r2r (mm5, mm0); // mm0 = b3
378 psubsw_r2r (mm3, mm4); // mm4 = v26
379 paddsw_r2r (mm6, mm5); // mm5 = v12
381 movq_r2m (mm0, scratch0); // save b3
382 movq_r2r (mm1, mm6); // mm6 = u17
384 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
385 paddsw_r2r (mm7, mm6); // mm6 = b0
387 psubsw_r2r (mm7, mm1); // mm1 = u12
388 movq_r2r (mm1, mm7); // mm7 = u12
390 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
391 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
393 movq_m2r (*_C4, mm0); // mm0 = C4/2
394 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
396 movq_r2m (mm6, scratch1); // save b0
397 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
399 movq_r2r (mm4, mm6); // mm6 = v26
400 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
402 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
403 movq_r2r (mm3, mm0); // mm0 = x0
405 psubsw_r2r (mm5, mm3); // mm3 = v04
406 paddsw_r2r (mm5, mm0); // mm0 = u04
408 paddsw_r2r (mm3, mm4); // mm4 = a1
409 movq_r2r (mm0, mm5); // mm5 = u04
411 psubsw_r2r (mm6, mm3); // mm3 = a2
412 paddsw_r2r (mm2, mm5); // mm5 = a0
414 paddsw_r2r (mm1, mm1); // mm1 = b1
415 psubsw_r2r (mm2, mm0); // mm0 = a3
417 paddsw_r2r (mm7, mm7); // mm7 = b2
418 movq_r2r (mm3, mm2); // mm2 = a2
420 movq_r2r (mm4, mm6); // mm6 = a1
421 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
423 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
424 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
426 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
427 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
429 movq_m2r (scratch1, mm1); // mm1 = b0
430 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
432 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
433 movq_r2r (mm5, mm7); // mm7 = a0
435 movq_r2m (mm4, *(col+offset+1*8)); // save y1
436 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
438 movq_r2m (mm3, *(col+offset+2*8)); // save y2
439 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
441 movq_m2r (scratch0, mm4); // mm4 = b3
442 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
444 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
445 movq_r2r (mm0, mm3); // mm3 = a3
447 movq_r2m (mm2, *(col+offset+5*8)); // save y5
448 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
450 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
451 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
453 movq_r2m (mm5, *(col+offset+0*8)); // save y0
454 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
456 movq_r2m (mm6, *(col+offset+6*8)); // save y6
457 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
459 movq_r2m (mm7, *(col+offset+7*8)); // save y7
461 movq_r2m (mm3, *(col+offset+4*8)); // save y4
463 movq_r2m (mm4, *(col+offset+3*8)); // save y3
467 static s32 rounder0[] ATTR_ALIGN(8) =
468 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
469 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
470 static s32 rounder1[] ATTR_ALIGN(8) =
471 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
472 static s32 rounder7[] ATTR_ALIGN(8) =
473 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
474 static s32 rounder2[] ATTR_ALIGN(8) =
475 rounder (0.60355339059); // C2 * (C6+C2)/2
476 static s32 rounder6[] ATTR_ALIGN(8) =
477 rounder (-0.25); // C2 * (C6-C2)/2
478 static s32 rounder3[] ATTR_ALIGN(8) =
479 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
480 static s32 rounder5[] ATTR_ALIGN(8) =
481 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
483 void _M( vdec_IDCT )( vdec_thread_t * p_vdec, dctelem_t * p_block,
486 static dctelem_t table04[] ATTR_ALIGN(16) =
487 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
488 static dctelem_t table17[] ATTR_ALIGN(16) =
489 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
490 static dctelem_t table26[] ATTR_ALIGN(16) =
491 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
492 static dctelem_t table35[] ATTR_ALIGN(16) =
493 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
495 RowHead( p_block, 0*8, table04 );
496 Row( table04, rounder0 );
497 RowMid( p_block, 0*8, 4*8, table04 );
498 Row( table04, rounder4 );
499 RowMid( p_block, 4*8, 1*8, table17 );
500 Row( table17, rounder1 );
501 RowMid( p_block, 1*8, 7*8, table17 );
502 Row( table17, rounder7 );
503 RowMid( p_block, 7*8, 2*8, table26 );
504 Row( table26, rounder2 );
505 RowMid( p_block, 2*8, 6*8, table26 );
506 Row( table26, rounder6 );
507 RowMid( p_block, 6*8, 3*8, table35 );
508 Row( table35, rounder3 );
509 RowMid( p_block, 3*8, 5*8, table35 );
510 Row( table35, rounder5 );
511 RowTail( p_block, 5*8);