1 /*****************************************************************************
2 * idctmmxext.c : MMX EXT IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmxext.c,v 1.14 2001/09/05 16:07:49 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 idctmmxext
29 #include "modules_inner.h"
31 /*****************************************************************************
33 *****************************************************************************/
42 #include "tests.h" /* TestCPU() */
47 #include "block_mmx.h"
50 #include "modules_export.h"
52 /*****************************************************************************
54 *****************************************************************************/
55 static void idct_getfunctions( function_list_t * p_function_list );
57 /*****************************************************************************
58 * Build configuration tree.
59 *****************************************************************************/
61 ADD_WINDOW( "Configuration for MMX EXT IDCT module" )
62 ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
66 p_module->i_capabilities = MODULE_CAPABILITY_NULL
67 | MODULE_CAPABILITY_IDCT;
68 p_module->psz_longname = "MMX EXT IDCT module";
72 idct_getfunctions( &p_module->p_functions->idct );
75 MODULE_DEACTIVATE_START
76 MODULE_DEACTIVATE_STOP
78 /* Following functions are local */
80 /*****************************************************************************
81 * idct_Probe: return a preference score
82 *****************************************************************************/
83 static int idct_Probe( probedata_t *p_data )
85 if( !TestCPU( CPU_CAPABILITY_MMXEXT ) )
90 if( TestMethod( IDCT_METHOD_VAR, "idctmmxext" )
91 || TestMethod( IDCT_METHOD_VAR, "mmxext" ) )
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, -c2, \
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 = -C2 -C4 C2 C4
143 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
145 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
146 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
148 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
151 static __inline__ void Row( dctelem_t * table, s32 * rounder )
153 movq_m2r (*(table+8), mm1); // mm1 = -C5 -C1 C3 C1
154 pmaddwd_r2r (mm2, mm4); // mm4 = C4*x0+C6*x2 C4*x4+C6*x6
156 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x4-C6*x6 C4*x0-C6*x2
157 pshufw_r2r (mm6, mm6, 0x4e); // mm6 = x3 x1 x7 x5
159 movq_m2r (*(table+12), mm7); // mm7 = -C7 C3 C7 C5
160 pmaddwd_r2r (mm5, mm1); // mm1 = -C1*x5-C5*x7 C1*x1+C3*x3
162 paddd_m2r (*rounder, mm3); // mm3 += rounder
163 pmaddwd_r2r (mm6, mm7); // mm7 = C3*x1-C7*x3 C5*x5+C7*x7
165 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x0-C2*x2 -C4*x4+C2*x6
166 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
168 pmaddwd_m2r (*(table+24), mm5); // mm5 = C3*x5-C1*x7 C5*x1-C1*x3
169 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
171 pmaddwd_m2r (*(table+28), mm6); // mm6 = C7*x1-C5*x3 C7*x5+C3*x7
172 paddd_r2r (mm7, mm1); // mm1 = b1 b0
174 paddd_m2r (*rounder, mm0); // mm0 += rounder
175 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
177 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
178 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
180 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
181 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
183 paddd_r2r (mm6, mm5); // mm5 = b3 b2
184 movq_r2r (mm0, mm4); // mm4 = a3 a2 + rounder
186 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
187 psubd_r2r (mm5, mm4); // mm4 = a3-b3 a2-b2 + rounder
190 static __inline__ void RowTail( dctelem_t * row, int store )
192 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
194 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
196 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
198 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
200 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
201 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
204 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
207 static __inline__ void RowMid( dctelem_t * row, int store,
208 int offset, dctelem_t * table)
210 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
211 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
213 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
214 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
216 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
217 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
219 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
220 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
222 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
223 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
225 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
226 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
228 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
230 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
231 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
234 static __inline__ void Col( dctelem_t * col, int offset )
241 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
242 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
243 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
244 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
245 static mmx_t scratch0, scratch1;
247 /* column code adapted from peter gubanov */
248 /* http://www.elecard.com/peter/idct.shtml */
250 movq_m2r (*_T1, mm0); // mm0 = T1
252 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
253 movq_r2r (mm0, mm2); // mm2 = T1
255 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
256 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
258 movq_m2r (*_T3, mm5); // mm5 = T3
259 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
261 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
262 movq_r2r (mm5, mm7); // mm7 = T3-1
264 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
265 psubsw_r2r (mm4, mm0); // mm0 = v17
267 movq_m2r (*_T2, mm4); // mm4 = T2
268 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
270 paddsw_r2r (mm2, mm1); // mm1 = u17
271 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
275 movq_r2r (mm4, mm2); // mm2 = T2
276 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
278 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
279 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
281 psubsw_r2r (mm6, mm5); // mm5 = v35
282 paddsw_r2r (mm3, mm7); // mm7 = u35
284 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
285 movq_r2r (mm0, mm6); // mm6 = v17
287 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
288 psubsw_r2r (mm5, mm0); // mm0 = b3
290 psubsw_r2r (mm3, mm4); // mm4 = v26
291 paddsw_r2r (mm6, mm5); // mm5 = v12
293 movq_r2m (mm0, scratch0); // save b3
294 movq_r2r (mm1, mm6); // mm6 = u17
296 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
297 paddsw_r2r (mm7, mm6); // mm6 = b0
299 psubsw_r2r (mm7, mm1); // mm1 = u12
300 movq_r2r (mm1, mm7); // mm7 = u12
302 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
303 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
305 movq_m2r (*_C4, mm0); // mm0 = C4/2
306 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
308 movq_r2m (mm6, scratch1); // save b0
309 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
311 movq_r2r (mm4, mm6); // mm6 = v26
312 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
314 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
315 movq_r2r (mm3, mm0); // mm0 = x0
317 psubsw_r2r (mm5, mm3); // mm3 = v04
318 paddsw_r2r (mm5, mm0); // mm0 = u04
320 paddsw_r2r (mm3, mm4); // mm4 = a1
321 movq_r2r (mm0, mm5); // mm5 = u04
323 psubsw_r2r (mm6, mm3); // mm3 = a2
324 paddsw_r2r (mm2, mm5); // mm5 = a0
326 paddsw_r2r (mm1, mm1); // mm1 = b1
327 psubsw_r2r (mm2, mm0); // mm0 = a3
329 paddsw_r2r (mm7, mm7); // mm7 = b2
330 movq_r2r (mm3, mm2); // mm2 = a2
332 movq_r2r (mm4, mm6); // mm6 = a1
333 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
335 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
336 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
338 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
339 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
341 movq_m2r (scratch1, mm1); // mm1 = b0
342 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
344 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
345 movq_r2r (mm5, mm7); // mm7 = a0
347 movq_r2m (mm4, *(col+offset+1*8)); // save y1
348 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
350 movq_r2m (mm3, *(col+offset+2*8)); // save y2
351 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
353 movq_m2r (scratch0, mm4); // mm4 = b3
354 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
356 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
357 movq_r2r (mm0, mm3); // mm3 = a3
359 movq_r2m (mm2, *(col+offset+5*8)); // save y5
360 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
362 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
363 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
365 movq_r2m (mm5, *(col+offset+0*8)); // save y0
366 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
368 movq_r2m (mm6, *(col+offset+6*8)); // save y6
369 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
371 movq_r2m (mm7, *(col+offset+7*8)); // save y7
373 movq_r2m (mm3, *(col+offset+4*8)); // save y4
375 movq_r2m (mm4, *(col+offset+3*8)); // save y3
379 static s32 rounder0[] ATTR_ALIGN(8) =
380 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
381 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
382 static s32 rounder1[] ATTR_ALIGN(8) =
383 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
384 static s32 rounder7[] ATTR_ALIGN(8) =
385 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
386 static s32 rounder2[] ATTR_ALIGN(8) =
387 rounder (0.60355339059); // C2 * (C6+C2)/2
388 static s32 rounder6[] ATTR_ALIGN(8) =
389 rounder (-0.25); // C2 * (C6-C2)/2
390 static s32 rounder3[] ATTR_ALIGN(8) =
391 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
392 static s32 rounder5[] ATTR_ALIGN(8) =
393 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
395 static __inline__ void IDCT( dctelem_t * p_block )
397 static dctelem_t table04[] ATTR_ALIGN(16) =
398 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
399 static dctelem_t table17[] ATTR_ALIGN(16) =
400 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
401 static dctelem_t table26[] ATTR_ALIGN(16) =
402 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
403 static dctelem_t table35[] ATTR_ALIGN(16) =
404 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
406 RowHead( p_block, 0*8, table04 );
407 Row( table04, rounder0 );
408 RowMid( p_block, 0*8, 4*8, table04 );
409 Row( table04, rounder4 );
410 RowMid( p_block, 4*8, 1*8, table17 );
411 Row( table17, rounder1 );
412 RowMid( p_block, 1*8, 7*8, table17 );
413 Row( table17, rounder7 );
414 RowMid( p_block, 7*8, 2*8, table26 );
415 Row( table26, rounder2 );
416 RowMid( p_block, 2*8, 6*8, table26 );
417 Row( table26, rounder6 );
418 RowMid( p_block, 6*8, 3*8, table35 );
419 Row( table35, rounder3 );
420 RowMid( p_block, 3*8, 5*8, table35 );
421 Row( table35, rounder5 );
422 RowTail( p_block, 5*8);
428 #include "idct_sparse.h"
429 #include "idct_decl.h"