1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmx.c,v 1.12 2001/05/30 17:03:12 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] );
64 /*****************************************************************************
65 * Build configuration tree.
66 *****************************************************************************/
68 ADD_WINDOW( "Configuration for MMX IDCT module" )
69 ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
73 p_module->i_capabilities = MODULE_CAPABILITY_NULL
74 | MODULE_CAPABILITY_IDCT;
75 p_module->psz_longname = "MMX IDCT module";
79 idct_getfunctions( &p_module->p_functions->idct );
82 MODULE_DEACTIVATE_START
83 MODULE_DEACTIVATE_STOP
85 /* Following functions are local */
87 /*****************************************************************************
88 * Functions exported as capabilities.
89 *****************************************************************************/
90 static void idct_getfunctions( function_list_t * p_function_list )
92 p_function_list->pf_probe = idct_Probe;
93 #define F p_function_list->functions.idct
94 F.pf_idct_init = _M( vdec_InitIDCT );
95 F.pf_sparse_idct = _M( vdec_SparseIDCT );
96 F.pf_idct = _M( vdec_IDCT );
97 F.pf_norm_scan = vdec_NormScan;
98 F.pf_decode_init = _M( vdec_InitDecode );
99 F.pf_decode_mb_c = _M( vdec_DecodeMacroblockC );
100 F.pf_decode_mb_bw = _M( vdec_DecodeMacroblockBW );
104 /*****************************************************************************
105 * idct_Probe: return a preference score
106 *****************************************************************************/
107 static int idct_Probe( probedata_t *p_data )
109 if( !TestCPU( CPU_CAPABILITY_MMX ) )
114 if( TestMethod( IDCT_METHOD_VAR, "idctmmx" ) )
122 /*****************************************************************************
123 * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table
124 *****************************************************************************/
125 static void vdec_NormScan( u8 ppi_scan[2][64] )
129 for( i = 0; i < 64; i++ )
132 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
135 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
139 /*****************************************************************************
141 *****************************************************************************/
145 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
146 #define rounder(bias) {round (bias), round (bias)}
148 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \
157 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
159 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
161 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
162 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
164 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
165 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
167 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
169 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
170 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
172 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
173 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
176 static __inline__ void Row( dctelem_t * table, s32 * rounder )
178 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
179 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
181 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
182 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
184 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
185 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
187 paddd_m2r (*rounder, mm3); // mm3 += rounder
188 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
190 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
191 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
193 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
194 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
196 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
197 paddd_r2r (mm7, mm1); // mm1 = b1 b0
199 paddd_m2r (*rounder, mm0); // mm0 += rounder
200 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
202 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
203 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
205 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
206 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
208 paddd_r2r (mm6, mm5); // mm5 = b3 b2
209 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
211 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
212 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
215 static __inline__ void RowTail( dctelem_t * row, int store )
217 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
219 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
221 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
223 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
225 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
226 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
228 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
230 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
232 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
236 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
239 static __inline__ void RowMid( dctelem_t * row, int store,
240 int offset, dctelem_t * table )
242 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
243 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
245 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
246 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
248 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
249 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
251 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
252 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
254 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
255 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
257 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
258 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
260 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
261 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
263 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
264 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
266 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
267 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
269 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
270 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
273 static __inline__ void Col( dctelem_t * col, int offset )
280 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
281 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
282 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
283 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
284 static mmx_t scratch0, scratch1;
286 /* column code adapted from peter gubanov */
287 /* http://www.elecard.com/peter/idct.shtml */
289 movq_m2r (*_T1, mm0); // mm0 = T1
291 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
292 movq_r2r (mm0, mm2); // mm2 = T1
294 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
295 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
297 movq_m2r (*_T3, mm5); // mm5 = T3
298 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
300 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
301 movq_r2r (mm5, mm7); // mm7 = T3-1
303 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
304 psubsw_r2r (mm4, mm0); // mm0 = v17
306 movq_m2r (*_T2, mm4); // mm4 = T2
307 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
309 paddsw_r2r (mm2, mm1); // mm1 = u17
310 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
314 movq_r2r (mm4, mm2); // mm2 = T2
315 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
317 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
318 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
320 psubsw_r2r (mm6, mm5); // mm5 = v35
321 paddsw_r2r (mm3, mm7); // mm7 = u35
323 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
324 movq_r2r (mm0, mm6); // mm6 = v17
326 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
327 psubsw_r2r (mm5, mm0); // mm0 = b3
329 psubsw_r2r (mm3, mm4); // mm4 = v26
330 paddsw_r2r (mm6, mm5); // mm5 = v12
332 movq_r2m (mm0, scratch0); // save b3
333 movq_r2r (mm1, mm6); // mm6 = u17
335 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
336 paddsw_r2r (mm7, mm6); // mm6 = b0
338 psubsw_r2r (mm7, mm1); // mm1 = u12
339 movq_r2r (mm1, mm7); // mm7 = u12
341 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
342 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
344 movq_m2r (*_C4, mm0); // mm0 = C4/2
345 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
347 movq_r2m (mm6, scratch1); // save b0
348 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
350 movq_r2r (mm4, mm6); // mm6 = v26
351 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
353 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
354 movq_r2r (mm3, mm0); // mm0 = x0
356 psubsw_r2r (mm5, mm3); // mm3 = v04
357 paddsw_r2r (mm5, mm0); // mm0 = u04
359 paddsw_r2r (mm3, mm4); // mm4 = a1
360 movq_r2r (mm0, mm5); // mm5 = u04
362 psubsw_r2r (mm6, mm3); // mm3 = a2
363 paddsw_r2r (mm2, mm5); // mm5 = a0
365 paddsw_r2r (mm1, mm1); // mm1 = b1
366 psubsw_r2r (mm2, mm0); // mm0 = a3
368 paddsw_r2r (mm7, mm7); // mm7 = b2
369 movq_r2r (mm3, mm2); // mm2 = a2
371 movq_r2r (mm4, mm6); // mm6 = a1
372 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
374 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
375 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
377 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
378 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
380 movq_m2r (scratch1, mm1); // mm1 = b0
381 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
383 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
384 movq_r2r (mm5, mm7); // mm7 = a0
386 movq_r2m (mm4, *(col+offset+1*8)); // save y1
387 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
389 movq_r2m (mm3, *(col+offset+2*8)); // save y2
390 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
392 movq_m2r (scratch0, mm4); // mm4 = b3
393 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
395 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
396 movq_r2r (mm0, mm3); // mm3 = a3
398 movq_r2m (mm2, *(col+offset+5*8)); // save y5
399 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
401 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
402 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
404 movq_r2m (mm5, *(col+offset+0*8)); // save y0
405 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
407 movq_r2m (mm6, *(col+offset+6*8)); // save y6
408 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
410 movq_r2m (mm7, *(col+offset+7*8)); // save y7
412 movq_r2m (mm3, *(col+offset+4*8)); // save y4
414 movq_r2m (mm4, *(col+offset+3*8)); // save y3
418 static s32 rounder0[] ATTR_ALIGN(8) =
419 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
420 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
421 static s32 rounder1[] ATTR_ALIGN(8) =
422 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
423 static s32 rounder7[] ATTR_ALIGN(8) =
424 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
425 static s32 rounder2[] ATTR_ALIGN(8) =
426 rounder (0.60355339059); // C2 * (C6+C2)/2
427 static s32 rounder6[] ATTR_ALIGN(8) =
428 rounder (-0.25); // C2 * (C6-C2)/2
429 static s32 rounder3[] ATTR_ALIGN(8) =
430 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
431 static s32 rounder5[] ATTR_ALIGN(8) =
432 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
434 void _M( vdec_IDCT )( vdec_thread_t * p_vdec, dctelem_t * p_block,
437 static dctelem_t table04[] ATTR_ALIGN(16) =
438 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
439 static dctelem_t table17[] ATTR_ALIGN(16) =
440 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
441 static dctelem_t table26[] ATTR_ALIGN(16) =
442 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
443 static dctelem_t table35[] ATTR_ALIGN(16) =
444 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
446 RowHead( p_block, 0*8, table04 );
447 Row( table04, rounder0 );
448 RowMid( p_block, 0*8, 4*8, table04 );
449 Row( table04, rounder4 );
450 RowMid( p_block, 4*8, 1*8, table17 );
451 Row( table17, rounder1 );
452 RowMid( p_block, 1*8, 7*8, table17 );
453 Row( table17, rounder7 );
454 RowMid( p_block, 7*8, 2*8, table26 );
455 Row( table26, rounder2 );
456 RowMid( p_block, 2*8, 6*8, table26 );
457 Row( table26, rounder6 );
458 RowMid( p_block, 6*8, 3*8, table35 );
459 Row( table35, rounder3 );
460 RowMid( p_block, 3*8, 5*8, table35 );
461 Row( table35, rounder5 );
462 RowTail( p_block, 5*8);