1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmx.c,v 1.19 2001/09/06 14:02:56 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 *****************************************************************************/
43 #include "tests.h" /* TestCPU() */
48 #include "block_mmx.h"
51 #include "modules_export.h"
53 /*****************************************************************************
55 *****************************************************************************/
56 static void idct_getfunctions( function_list_t * p_function_list );
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 * idct_Probe: return a preference score
83 *****************************************************************************/
84 static int idct_Probe( probedata_t *p_data )
86 if( !TestCPU( CPU_CAPABILITY_MMX ) )
91 if( TestMethod( IDCT_METHOD_VAR, "idctmmx" )
92 || TestMethod( IDCT_METHOD_VAR, "mmx" ) )
100 /*****************************************************************************
101 * NormScan : This IDCT uses reordered coeffs, so we patch the scan table
102 *****************************************************************************/
103 static void NormScan( u8 ppi_scan[2][64] )
107 for( i = 0; i < 64; i++ )
110 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
113 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
117 /*****************************************************************************
119 *****************************************************************************/
123 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
124 #define rounder(bias) {round (bias), round (bias)}
126 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \
135 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
137 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
139 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
140 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
142 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
143 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
145 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
147 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
148 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
150 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
151 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
154 static __inline__ void Row( dctelem_t * table, s32 * rounder )
156 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
157 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
159 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
160 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
162 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
163 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
165 paddd_m2r (*rounder, mm3); // mm3 += rounder
166 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
168 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
169 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
171 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
172 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
174 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
175 paddd_r2r (mm7, mm1); // mm1 = b1 b0
177 paddd_m2r (*rounder, mm0); // mm0 += rounder
178 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
180 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
181 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
183 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
184 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
186 paddd_r2r (mm6, mm5); // mm5 = b3 b2
187 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
189 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
190 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
193 static __inline__ void RowTail( dctelem_t * row, int store )
195 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
197 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
199 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
201 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
203 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
204 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
206 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
208 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
210 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
214 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
217 static __inline__ void RowMid( dctelem_t * row, int store,
218 int offset, dctelem_t * table )
220 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
221 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
223 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
224 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
226 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
227 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
229 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
230 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
232 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
233 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
235 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
236 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
238 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
239 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
241 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
242 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
244 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
245 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
247 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
248 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
251 static __inline__ void Col( dctelem_t * col, int offset )
258 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
259 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
260 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
261 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
262 static mmx_t scratch0, scratch1;
264 /* column code adapted from peter gubanov */
265 /* http://www.elecard.com/peter/idct.shtml */
267 movq_m2r (*_T1, mm0); // mm0 = T1
269 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
270 movq_r2r (mm0, mm2); // mm2 = T1
272 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
273 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
275 movq_m2r (*_T3, mm5); // mm5 = T3
276 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
278 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
279 movq_r2r (mm5, mm7); // mm7 = T3-1
281 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
282 psubsw_r2r (mm4, mm0); // mm0 = v17
284 movq_m2r (*_T2, mm4); // mm4 = T2
285 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
287 paddsw_r2r (mm2, mm1); // mm1 = u17
288 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
292 movq_r2r (mm4, mm2); // mm2 = T2
293 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
295 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
296 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
298 psubsw_r2r (mm6, mm5); // mm5 = v35
299 paddsw_r2r (mm3, mm7); // mm7 = u35
301 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
302 movq_r2r (mm0, mm6); // mm6 = v17
304 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
305 psubsw_r2r (mm5, mm0); // mm0 = b3
307 psubsw_r2r (mm3, mm4); // mm4 = v26
308 paddsw_r2r (mm6, mm5); // mm5 = v12
310 movq_r2m (mm0, scratch0); // save b3
311 movq_r2r (mm1, mm6); // mm6 = u17
313 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
314 paddsw_r2r (mm7, mm6); // mm6 = b0
316 psubsw_r2r (mm7, mm1); // mm1 = u12
317 movq_r2r (mm1, mm7); // mm7 = u12
319 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
320 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
322 movq_m2r (*_C4, mm0); // mm0 = C4/2
323 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
325 movq_r2m (mm6, scratch1); // save b0
326 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
328 movq_r2r (mm4, mm6); // mm6 = v26
329 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
331 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
332 movq_r2r (mm3, mm0); // mm0 = x0
334 psubsw_r2r (mm5, mm3); // mm3 = v04
335 paddsw_r2r (mm5, mm0); // mm0 = u04
337 paddsw_r2r (mm3, mm4); // mm4 = a1
338 movq_r2r (mm0, mm5); // mm5 = u04
340 psubsw_r2r (mm6, mm3); // mm3 = a2
341 paddsw_r2r (mm2, mm5); // mm5 = a0
343 paddsw_r2r (mm1, mm1); // mm1 = b1
344 psubsw_r2r (mm2, mm0); // mm0 = a3
346 paddsw_r2r (mm7, mm7); // mm7 = b2
347 movq_r2r (mm3, mm2); // mm2 = a2
349 movq_r2r (mm4, mm6); // mm6 = a1
350 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
352 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
353 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
355 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
356 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
358 movq_m2r (scratch1, mm1); // mm1 = b0
359 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
361 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
362 movq_r2r (mm5, mm7); // mm7 = a0
364 movq_r2m (mm4, *(col+offset+1*8)); // save y1
365 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
367 movq_r2m (mm3, *(col+offset+2*8)); // save y2
368 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
370 movq_m2r (scratch0, mm4); // mm4 = b3
371 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
373 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
374 movq_r2r (mm0, mm3); // mm3 = a3
376 movq_r2m (mm2, *(col+offset+5*8)); // save y5
377 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
379 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
380 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
382 movq_r2m (mm5, *(col+offset+0*8)); // save y0
383 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
385 movq_r2m (mm6, *(col+offset+6*8)); // save y6
386 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
388 movq_r2m (mm7, *(col+offset+7*8)); // save y7
390 movq_r2m (mm3, *(col+offset+4*8)); // save y4
392 movq_r2m (mm4, *(col+offset+3*8)); // save y3
396 static s32 rounder0[] ATTR_ALIGN(8) =
397 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
398 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
399 static s32 rounder1[] ATTR_ALIGN(8) =
400 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
401 static s32 rounder7[] ATTR_ALIGN(8) =
402 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
403 static s32 rounder2[] ATTR_ALIGN(8) =
404 rounder (0.60355339059); // C2 * (C6+C2)/2
405 static s32 rounder6[] ATTR_ALIGN(8) =
406 rounder (-0.25); // C2 * (C6-C2)/2
407 static s32 rounder3[] ATTR_ALIGN(8) =
408 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
409 static s32 rounder5[] ATTR_ALIGN(8) =
410 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
412 static void IDCT( dctelem_t * p_block )
414 static dctelem_t table04[] ATTR_ALIGN(16) =
415 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
416 static dctelem_t table17[] ATTR_ALIGN(16) =
417 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
418 static dctelem_t table26[] ATTR_ALIGN(16) =
419 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
420 static dctelem_t table35[] ATTR_ALIGN(16) =
421 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
423 RowHead( p_block, 0*8, table04 );
424 Row( table04, rounder0 );
425 RowMid( p_block, 0*8, 4*8, table04 );
426 Row( table04, rounder4 );
427 RowMid( p_block, 4*8, 1*8, table17 );
428 Row( table17, rounder1 );
429 RowMid( p_block, 1*8, 7*8, table17 );
430 Row( table17, rounder7 );
431 RowMid( p_block, 7*8, 2*8, table26 );
432 Row( table26, rounder2 );
433 RowMid( p_block, 2*8, 6*8, table26 );
434 Row( table26, rounder6 );
435 RowMid( p_block, 6*8, 3*8, table35 );
436 Row( table35, rounder3 );
437 RowMid( p_block, 3*8, 5*8, table35 );
438 Row( table35, rounder5 );
439 RowTail( p_block, 5*8);
445 #include "idct_sparse.h"
446 #include "idct_decl.h"