1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmx.c,v 1.16 2001/08/22 17:21:45 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 idctmmx
29 #include "modules_inner.h"
31 /*****************************************************************************
33 *****************************************************************************/
42 #include "tests.h" /* TestCPU() */
44 #include "vdec_idct.h"
49 #include "modules_export.h"
51 /*****************************************************************************
53 *****************************************************************************/
54 static void idct_getfunctions( function_list_t * p_function_list );
55 static int idct_Probe ( probedata_t *p_data );
56 static void vdec_NormScan ( u8 ppi_scan[2][64] );
58 /*****************************************************************************
59 * Build configuration tree.
60 *****************************************************************************/
62 ADD_WINDOW( "Configuration for MMX IDCT module" )
63 ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
67 p_module->i_capabilities = MODULE_CAPABILITY_NULL
68 | MODULE_CAPABILITY_IDCT;
69 p_module->psz_longname = "MMX IDCT module";
73 idct_getfunctions( &p_module->p_functions->idct );
76 MODULE_DEACTIVATE_START
77 MODULE_DEACTIVATE_STOP
79 /* Following functions are local */
81 /*****************************************************************************
82 * Functions exported as capabilities.
83 *****************************************************************************/
84 static void idct_getfunctions( function_list_t * p_function_list )
86 p_function_list->pf_probe = idct_Probe;
87 #define F p_function_list->functions.idct
88 F.pf_idct_init = _M( vdec_InitIDCT );
89 F.pf_sparse_idct = _M( vdec_SparseIDCT );
90 F.pf_idct = _M( vdec_IDCT );
91 F.pf_norm_scan = vdec_NormScan;
92 F.pf_decode_init = _M( vdec_InitDecode );
93 F.pf_addblock = _M( vdec_AddBlock );
94 F.pf_copyblock = _M( vdec_CopyBlock );
98 /*****************************************************************************
99 * idct_Probe: return a preference score
100 *****************************************************************************/
101 static int idct_Probe( probedata_t *p_data )
103 if( !TestCPU( CPU_CAPABILITY_MMX ) )
108 if( TestMethod( IDCT_METHOD_VAR, "idctmmx" )
109 || TestMethod( IDCT_METHOD_VAR, "mmx" ) )
117 /*****************************************************************************
118 * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table
119 *****************************************************************************/
120 static void vdec_NormScan( u8 ppi_scan[2][64] )
124 for( i = 0; i < 64; i++ )
127 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
130 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
134 /*****************************************************************************
136 *****************************************************************************/
140 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
141 #define rounder(bias) {round (bias), round (bias)}
143 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \
152 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
154 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
156 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
157 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
159 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
160 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
162 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
164 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
165 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
167 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
168 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
171 static __inline__ void Row( dctelem_t * table, s32 * rounder )
173 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
174 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
176 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
177 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
179 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
180 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
182 paddd_m2r (*rounder, mm3); // mm3 += rounder
183 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
185 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
186 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
188 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
189 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
191 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
192 paddd_r2r (mm7, mm1); // mm1 = b1 b0
194 paddd_m2r (*rounder, mm0); // mm0 += rounder
195 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
197 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
198 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
200 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
201 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
203 paddd_r2r (mm6, mm5); // mm5 = b3 b2
204 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
206 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
207 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
210 static __inline__ void RowTail( dctelem_t * row, int store )
212 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
214 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
216 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
218 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
220 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
221 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
223 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
225 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
227 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
231 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
234 static __inline__ void RowMid( dctelem_t * row, int store,
235 int offset, dctelem_t * table )
237 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
238 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
240 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
241 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
243 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
244 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
246 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
247 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
249 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
250 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
252 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
253 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
255 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
256 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
258 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
259 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
261 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
262 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
264 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
265 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
268 static __inline__ void Col( dctelem_t * col, int offset )
275 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
276 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
277 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
278 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
279 static mmx_t scratch0, scratch1;
281 /* column code adapted from peter gubanov */
282 /* http://www.elecard.com/peter/idct.shtml */
284 movq_m2r (*_T1, mm0); // mm0 = T1
286 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
287 movq_r2r (mm0, mm2); // mm2 = T1
289 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
290 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
292 movq_m2r (*_T3, mm5); // mm5 = T3
293 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
295 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
296 movq_r2r (mm5, mm7); // mm7 = T3-1
298 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
299 psubsw_r2r (mm4, mm0); // mm0 = v17
301 movq_m2r (*_T2, mm4); // mm4 = T2
302 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
304 paddsw_r2r (mm2, mm1); // mm1 = u17
305 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
309 movq_r2r (mm4, mm2); // mm2 = T2
310 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
312 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
313 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
315 psubsw_r2r (mm6, mm5); // mm5 = v35
316 paddsw_r2r (mm3, mm7); // mm7 = u35
318 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
319 movq_r2r (mm0, mm6); // mm6 = v17
321 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
322 psubsw_r2r (mm5, mm0); // mm0 = b3
324 psubsw_r2r (mm3, mm4); // mm4 = v26
325 paddsw_r2r (mm6, mm5); // mm5 = v12
327 movq_r2m (mm0, scratch0); // save b3
328 movq_r2r (mm1, mm6); // mm6 = u17
330 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
331 paddsw_r2r (mm7, mm6); // mm6 = b0
333 psubsw_r2r (mm7, mm1); // mm1 = u12
334 movq_r2r (mm1, mm7); // mm7 = u12
336 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
337 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
339 movq_m2r (*_C4, mm0); // mm0 = C4/2
340 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
342 movq_r2m (mm6, scratch1); // save b0
343 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
345 movq_r2r (mm4, mm6); // mm6 = v26
346 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
348 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
349 movq_r2r (mm3, mm0); // mm0 = x0
351 psubsw_r2r (mm5, mm3); // mm3 = v04
352 paddsw_r2r (mm5, mm0); // mm0 = u04
354 paddsw_r2r (mm3, mm4); // mm4 = a1
355 movq_r2r (mm0, mm5); // mm5 = u04
357 psubsw_r2r (mm6, mm3); // mm3 = a2
358 paddsw_r2r (mm2, mm5); // mm5 = a0
360 paddsw_r2r (mm1, mm1); // mm1 = b1
361 psubsw_r2r (mm2, mm0); // mm0 = a3
363 paddsw_r2r (mm7, mm7); // mm7 = b2
364 movq_r2r (mm3, mm2); // mm2 = a2
366 movq_r2r (mm4, mm6); // mm6 = a1
367 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
369 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
370 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
372 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
373 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
375 movq_m2r (scratch1, mm1); // mm1 = b0
376 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
378 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
379 movq_r2r (mm5, mm7); // mm7 = a0
381 movq_r2m (mm4, *(col+offset+1*8)); // save y1
382 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
384 movq_r2m (mm3, *(col+offset+2*8)); // save y2
385 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
387 movq_m2r (scratch0, mm4); // mm4 = b3
388 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
390 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
391 movq_r2r (mm0, mm3); // mm3 = a3
393 movq_r2m (mm2, *(col+offset+5*8)); // save y5
394 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
396 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
397 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
399 movq_r2m (mm5, *(col+offset+0*8)); // save y0
400 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
402 movq_r2m (mm6, *(col+offset+6*8)); // save y6
403 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
405 movq_r2m (mm7, *(col+offset+7*8)); // save y7
407 movq_r2m (mm3, *(col+offset+4*8)); // save y4
409 movq_r2m (mm4, *(col+offset+3*8)); // save y3
413 static s32 rounder0[] ATTR_ALIGN(8) =
414 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
415 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
416 static s32 rounder1[] ATTR_ALIGN(8) =
417 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
418 static s32 rounder7[] ATTR_ALIGN(8) =
419 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
420 static s32 rounder2[] ATTR_ALIGN(8) =
421 rounder (0.60355339059); // C2 * (C6+C2)/2
422 static s32 rounder6[] ATTR_ALIGN(8) =
423 rounder (-0.25); // C2 * (C6-C2)/2
424 static s32 rounder3[] ATTR_ALIGN(8) =
425 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
426 static s32 rounder5[] ATTR_ALIGN(8) =
427 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
429 void _M( vdec_IDCT )( void * p_unused_data, dctelem_t * p_block,
432 static dctelem_t table04[] ATTR_ALIGN(16) =
433 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
434 static dctelem_t table17[] ATTR_ALIGN(16) =
435 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
436 static dctelem_t table26[] ATTR_ALIGN(16) =
437 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
438 static dctelem_t table35[] ATTR_ALIGN(16) =
439 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
441 RowHead( p_block, 0*8, table04 );
442 Row( table04, rounder0 );
443 RowMid( p_block, 0*8, 4*8, table04 );
444 Row( table04, rounder4 );
445 RowMid( p_block, 4*8, 1*8, table17 );
446 Row( table17, rounder1 );
447 RowMid( p_block, 1*8, 7*8, table17 );
448 Row( table17, rounder7 );
449 RowMid( p_block, 7*8, 2*8, table26 );
450 Row( table26, rounder2 );
451 RowMid( p_block, 2*8, 6*8, table26 );
452 Row( table26, rounder6 );
453 RowMid( p_block, 6*8, 3*8, table35 );
454 Row( table35, rounder3 );
455 RowMid( p_block, 3*8, 5*8, table35 );
456 Row( table35, rounder5 );
457 RowTail( p_block, 5*8);