1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmx.c,v 1.13 2001/06/03 12:47:21 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 #include "modules_export.h"
59 /*****************************************************************************
61 *****************************************************************************/
62 static void idct_getfunctions( function_list_t * p_function_list );
63 static int idct_Probe ( probedata_t *p_data );
64 static void vdec_NormScan ( u8 ppi_scan[2][64] );
66 /*****************************************************************************
67 * Build configuration tree.
68 *****************************************************************************/
70 ADD_WINDOW( "Configuration for MMX IDCT module" )
71 ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
75 p_module->i_capabilities = MODULE_CAPABILITY_NULL
76 | MODULE_CAPABILITY_IDCT;
77 p_module->psz_longname = "MMX IDCT module";
81 idct_getfunctions( &p_module->p_functions->idct );
84 MODULE_DEACTIVATE_START
85 MODULE_DEACTIVATE_STOP
87 /* Following functions are local */
89 /*****************************************************************************
90 * Functions exported as capabilities.
91 *****************************************************************************/
92 static void idct_getfunctions( function_list_t * p_function_list )
94 p_function_list->pf_probe = idct_Probe;
95 #define F p_function_list->functions.idct
96 F.pf_idct_init = _M( vdec_InitIDCT );
97 F.pf_sparse_idct = _M( vdec_SparseIDCT );
98 F.pf_idct = _M( vdec_IDCT );
99 F.pf_norm_scan = vdec_NormScan;
100 F.pf_decode_init = _M( vdec_InitDecode );
101 F.pf_decode_mb_c = _M( vdec_DecodeMacroblockC );
102 F.pf_decode_mb_bw = _M( vdec_DecodeMacroblockBW );
106 /*****************************************************************************
107 * idct_Probe: return a preference score
108 *****************************************************************************/
109 static int idct_Probe( probedata_t *p_data )
111 if( !TestCPU( CPU_CAPABILITY_MMX ) )
116 if( TestMethod( IDCT_METHOD_VAR, "idctmmx" ) )
124 /*****************************************************************************
125 * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table
126 *****************************************************************************/
127 static void vdec_NormScan( u8 ppi_scan[2][64] )
131 for( i = 0; i < 64; i++ )
134 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
137 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
141 /*****************************************************************************
143 *****************************************************************************/
147 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
148 #define rounder(bias) {round (bias), round (bias)}
150 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \
159 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
161 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
163 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
164 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
166 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
167 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
169 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
171 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
172 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
174 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
175 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
178 static __inline__ void Row( dctelem_t * table, s32 * rounder )
180 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
181 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
183 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
184 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
186 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
187 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
189 paddd_m2r (*rounder, mm3); // mm3 += rounder
190 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
192 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
193 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
195 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
196 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
198 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
199 paddd_r2r (mm7, mm1); // mm1 = b1 b0
201 paddd_m2r (*rounder, mm0); // mm0 += rounder
202 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
204 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
205 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
207 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
208 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
210 paddd_r2r (mm6, mm5); // mm5 = b3 b2
211 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
213 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
214 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
217 static __inline__ void RowTail( dctelem_t * row, int store )
219 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
221 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
223 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
225 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
227 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
228 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
230 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
232 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
234 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
238 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
241 static __inline__ void RowMid( dctelem_t * row, int store,
242 int offset, dctelem_t * table )
244 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
245 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
247 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
248 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
250 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
251 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
253 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
254 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
256 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
257 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
259 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
260 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
262 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
263 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
265 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
266 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
268 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
269 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
271 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
272 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
275 static __inline__ void Col( dctelem_t * col, int offset )
282 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
283 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
284 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
285 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
286 static mmx_t scratch0, scratch1;
288 /* column code adapted from peter gubanov */
289 /* http://www.elecard.com/peter/idct.shtml */
291 movq_m2r (*_T1, mm0); // mm0 = T1
293 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
294 movq_r2r (mm0, mm2); // mm2 = T1
296 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
297 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
299 movq_m2r (*_T3, mm5); // mm5 = T3
300 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
302 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
303 movq_r2r (mm5, mm7); // mm7 = T3-1
305 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
306 psubsw_r2r (mm4, mm0); // mm0 = v17
308 movq_m2r (*_T2, mm4); // mm4 = T2
309 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
311 paddsw_r2r (mm2, mm1); // mm1 = u17
312 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
316 movq_r2r (mm4, mm2); // mm2 = T2
317 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
319 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
320 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
322 psubsw_r2r (mm6, mm5); // mm5 = v35
323 paddsw_r2r (mm3, mm7); // mm7 = u35
325 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
326 movq_r2r (mm0, mm6); // mm6 = v17
328 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
329 psubsw_r2r (mm5, mm0); // mm0 = b3
331 psubsw_r2r (mm3, mm4); // mm4 = v26
332 paddsw_r2r (mm6, mm5); // mm5 = v12
334 movq_r2m (mm0, scratch0); // save b3
335 movq_r2r (mm1, mm6); // mm6 = u17
337 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
338 paddsw_r2r (mm7, mm6); // mm6 = b0
340 psubsw_r2r (mm7, mm1); // mm1 = u12
341 movq_r2r (mm1, mm7); // mm7 = u12
343 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
344 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
346 movq_m2r (*_C4, mm0); // mm0 = C4/2
347 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
349 movq_r2m (mm6, scratch1); // save b0
350 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
352 movq_r2r (mm4, mm6); // mm6 = v26
353 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
355 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
356 movq_r2r (mm3, mm0); // mm0 = x0
358 psubsw_r2r (mm5, mm3); // mm3 = v04
359 paddsw_r2r (mm5, mm0); // mm0 = u04
361 paddsw_r2r (mm3, mm4); // mm4 = a1
362 movq_r2r (mm0, mm5); // mm5 = u04
364 psubsw_r2r (mm6, mm3); // mm3 = a2
365 paddsw_r2r (mm2, mm5); // mm5 = a0
367 paddsw_r2r (mm1, mm1); // mm1 = b1
368 psubsw_r2r (mm2, mm0); // mm0 = a3
370 paddsw_r2r (mm7, mm7); // mm7 = b2
371 movq_r2r (mm3, mm2); // mm2 = a2
373 movq_r2r (mm4, mm6); // mm6 = a1
374 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
376 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
377 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
379 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
380 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
382 movq_m2r (scratch1, mm1); // mm1 = b0
383 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
385 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
386 movq_r2r (mm5, mm7); // mm7 = a0
388 movq_r2m (mm4, *(col+offset+1*8)); // save y1
389 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
391 movq_r2m (mm3, *(col+offset+2*8)); // save y2
392 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
394 movq_m2r (scratch0, mm4); // mm4 = b3
395 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
397 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
398 movq_r2r (mm0, mm3); // mm3 = a3
400 movq_r2m (mm2, *(col+offset+5*8)); // save y5
401 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
403 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
404 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
406 movq_r2m (mm5, *(col+offset+0*8)); // save y0
407 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
409 movq_r2m (mm6, *(col+offset+6*8)); // save y6
410 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
412 movq_r2m (mm7, *(col+offset+7*8)); // save y7
414 movq_r2m (mm3, *(col+offset+4*8)); // save y4
416 movq_r2m (mm4, *(col+offset+3*8)); // save y3
420 static s32 rounder0[] ATTR_ALIGN(8) =
421 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
422 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
423 static s32 rounder1[] ATTR_ALIGN(8) =
424 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
425 static s32 rounder7[] ATTR_ALIGN(8) =
426 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
427 static s32 rounder2[] ATTR_ALIGN(8) =
428 rounder (0.60355339059); // C2 * (C6+C2)/2
429 static s32 rounder6[] ATTR_ALIGN(8) =
430 rounder (-0.25); // C2 * (C6-C2)/2
431 static s32 rounder3[] ATTR_ALIGN(8) =
432 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
433 static s32 rounder5[] ATTR_ALIGN(8) =
434 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
436 void _M( vdec_IDCT )( vdec_thread_t * p_vdec, dctelem_t * p_block,
439 static dctelem_t table04[] ATTR_ALIGN(16) =
440 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
441 static dctelem_t table17[] ATTR_ALIGN(16) =
442 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
443 static dctelem_t table26[] ATTR_ALIGN(16) =
444 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
445 static dctelem_t table35[] ATTR_ALIGN(16) =
446 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
448 RowHead( p_block, 0*8, table04 );
449 Row( table04, rounder0 );
450 RowMid( p_block, 0*8, 4*8, table04 );
451 Row( table04, rounder4 );
452 RowMid( p_block, 4*8, 1*8, table17 );
453 Row( table17, rounder1 );
454 RowMid( p_block, 1*8, 7*8, table17 );
455 Row( table17, rounder7 );
456 RowMid( p_block, 7*8, 2*8, table26 );
457 Row( table26, rounder2 );
458 RowMid( p_block, 2*8, 6*8, table26 );
459 Row( table26, rounder6 );
460 RowMid( p_block, 6*8, 3*8, table35 );
461 Row( table35, rounder3 );
462 RowMid( p_block, 3*8, 5*8, table35 );
463 Row( table35, rounder5 );
464 RowTail( p_block, 5*8);