1 /*****************************************************************************
2 * idctmmxext.c : MMX EXT IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999-2001 VideoLAN
5 * $Id: idctmmxext.c,v 1.18 2001/12/09 17:01:36 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 idctmmxext
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 EXT 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 EXT 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_MMXEXT ) )
91 if( TestMethod( IDCT_METHOD_VAR, "idctmmxext" )
92 || TestMethod( IDCT_METHOD_VAR, "mmxext" ) )
101 /*****************************************************************************
102 * NormScan : This IDCT uses reordered coeffs, so we patch the scan table
103 *****************************************************************************/
104 static void NormScan( u8 ppi_scan[2][64] )
108 for( i = 0; i < 64; i++ )
111 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
114 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
118 /*****************************************************************************
120 *****************************************************************************/
124 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
125 #define rounder(bias) {round (bias), round (bias)}
127 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, -c4, -c2, \
136 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
138 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
140 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
141 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
143 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
144 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
146 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
147 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
149 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
152 static __inline__ void Row( dctelem_t * table, s32 * rounder )
154 movq_m2r (*(table+8), mm1); // mm1 = -C5 -C1 C3 C1
155 pmaddwd_r2r (mm2, mm4); // mm4 = C4*x0+C6*x2 C4*x4+C6*x6
157 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x4-C6*x6 C4*x0-C6*x2
158 pshufw_r2r (mm6, mm6, 0x4e); // mm6 = x3 x1 x7 x5
160 movq_m2r (*(table+12), mm7); // mm7 = -C7 C3 C7 C5
161 pmaddwd_r2r (mm5, mm1); // mm1 = -C1*x5-C5*x7 C1*x1+C3*x3
163 paddd_m2r (*rounder, mm3); // mm3 += rounder
164 pmaddwd_r2r (mm6, mm7); // mm7 = C3*x1-C7*x3 C5*x5+C7*x7
166 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x0-C2*x2 -C4*x4+C2*x6
167 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
169 pmaddwd_m2r (*(table+24), mm5); // mm5 = C3*x5-C1*x7 C5*x1-C1*x3
170 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
172 pmaddwd_m2r (*(table+28), mm6); // mm6 = C7*x1-C5*x3 C7*x5+C3*x7
173 paddd_r2r (mm7, mm1); // mm1 = b1 b0
175 paddd_m2r (*rounder, mm0); // mm0 += rounder
176 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
178 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
179 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
181 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
182 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
184 paddd_r2r (mm6, mm5); // mm5 = b3 b2
185 movq_r2r (mm0, mm4); // mm4 = a3 a2 + rounder
187 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
188 psubd_r2r (mm5, mm4); // mm4 = a3-b3 a2-b2 + rounder
191 static __inline__ void RowTail( dctelem_t * row, int store )
193 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
195 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
197 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
199 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
201 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
202 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
205 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
208 static __inline__ void RowMid( dctelem_t * row, int store,
209 int offset, dctelem_t * table)
211 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
212 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
214 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
215 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
217 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
218 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
220 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
221 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
223 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
224 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
226 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
227 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
229 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
231 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
232 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
235 static __inline__ void Col( dctelem_t * col, int offset )
242 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
243 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
244 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
245 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
246 static mmx_t scratch0, scratch1;
248 /* column code adapted from peter gubanov */
249 /* http://www.elecard.com/peter/idct.shtml */
251 movq_m2r (*_T1, mm0); // mm0 = T1
253 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
254 movq_r2r (mm0, mm2); // mm2 = T1
256 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
257 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
259 movq_m2r (*_T3, mm5); // mm5 = T3
260 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
262 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
263 movq_r2r (mm5, mm7); // mm7 = T3-1
265 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
266 psubsw_r2r (mm4, mm0); // mm0 = v17
268 movq_m2r (*_T2, mm4); // mm4 = T2
269 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
271 paddsw_r2r (mm2, mm1); // mm1 = u17
272 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
276 movq_r2r (mm4, mm2); // mm2 = T2
277 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
279 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
280 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
282 psubsw_r2r (mm6, mm5); // mm5 = v35
283 paddsw_r2r (mm3, mm7); // mm7 = u35
285 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
286 movq_r2r (mm0, mm6); // mm6 = v17
288 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
289 psubsw_r2r (mm5, mm0); // mm0 = b3
291 psubsw_r2r (mm3, mm4); // mm4 = v26
292 paddsw_r2r (mm6, mm5); // mm5 = v12
294 movq_r2m (mm0, scratch0); // save b3
295 movq_r2r (mm1, mm6); // mm6 = u17
297 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
298 paddsw_r2r (mm7, mm6); // mm6 = b0
300 psubsw_r2r (mm7, mm1); // mm1 = u12
301 movq_r2r (mm1, mm7); // mm7 = u12
303 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
304 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
306 movq_m2r (*_C4, mm0); // mm0 = C4/2
307 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
309 movq_r2m (mm6, scratch1); // save b0
310 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
312 movq_r2r (mm4, mm6); // mm6 = v26
313 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
315 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
316 movq_r2r (mm3, mm0); // mm0 = x0
318 psubsw_r2r (mm5, mm3); // mm3 = v04
319 paddsw_r2r (mm5, mm0); // mm0 = u04
321 paddsw_r2r (mm3, mm4); // mm4 = a1
322 movq_r2r (mm0, mm5); // mm5 = u04
324 psubsw_r2r (mm6, mm3); // mm3 = a2
325 paddsw_r2r (mm2, mm5); // mm5 = a0
327 paddsw_r2r (mm1, mm1); // mm1 = b1
328 psubsw_r2r (mm2, mm0); // mm0 = a3
330 paddsw_r2r (mm7, mm7); // mm7 = b2
331 movq_r2r (mm3, mm2); // mm2 = a2
333 movq_r2r (mm4, mm6); // mm6 = a1
334 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
336 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
337 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
339 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
340 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
342 movq_m2r (scratch1, mm1); // mm1 = b0
343 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
345 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
346 movq_r2r (mm5, mm7); // mm7 = a0
348 movq_r2m (mm4, *(col+offset+1*8)); // save y1
349 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
351 movq_r2m (mm3, *(col+offset+2*8)); // save y2
352 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
354 movq_m2r (scratch0, mm4); // mm4 = b3
355 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
357 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
358 movq_r2r (mm0, mm3); // mm3 = a3
360 movq_r2m (mm2, *(col+offset+5*8)); // save y5
361 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
363 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
364 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
366 movq_r2m (mm5, *(col+offset+0*8)); // save y0
367 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
369 movq_r2m (mm6, *(col+offset+6*8)); // save y6
370 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
372 movq_r2m (mm7, *(col+offset+7*8)); // save y7
374 movq_r2m (mm3, *(col+offset+4*8)); // save y4
376 movq_r2m (mm4, *(col+offset+3*8)); // save y3
380 static s32 rounder0[] ATTR_ALIGN(8) =
381 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
382 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
383 static s32 rounder1[] ATTR_ALIGN(8) =
384 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
385 static s32 rounder7[] ATTR_ALIGN(8) =
386 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
387 static s32 rounder2[] ATTR_ALIGN(8) =
388 rounder (0.60355339059); // C2 * (C6+C2)/2
389 static s32 rounder6[] ATTR_ALIGN(8) =
390 rounder (-0.25); // C2 * (C6-C2)/2
391 static s32 rounder3[] ATTR_ALIGN(8) =
392 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
393 static s32 rounder5[] ATTR_ALIGN(8) =
394 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
396 static void IDCT( dctelem_t * p_block )
398 static dctelem_t table04[] ATTR_ALIGN(16) =
399 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
400 static dctelem_t table17[] ATTR_ALIGN(16) =
401 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
402 static dctelem_t table26[] ATTR_ALIGN(16) =
403 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
404 static dctelem_t table35[] ATTR_ALIGN(16) =
405 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
407 RowHead( p_block, 0*8, table04 );
408 Row( table04, rounder0 );
409 RowMid( p_block, 0*8, 4*8, table04 );
410 Row( table04, rounder4 );
411 RowMid( p_block, 4*8, 1*8, table17 );
412 Row( table17, rounder1 );
413 RowMid( p_block, 1*8, 7*8, table17 );
414 Row( table17, rounder7 );
415 RowMid( p_block, 7*8, 2*8, table26 );
416 Row( table26, rounder2 );
417 RowMid( p_block, 2*8, 6*8, table26 );
418 Row( table26, rounder6 );
419 RowMid( p_block, 6*8, 3*8, table35 );
420 Row( table35, rounder3 );
421 RowMid( p_block, 3*8, 5*8, table35 );
422 Row( table35, rounder5 );
423 RowTail( p_block, 5*8);
429 #include "idct_sparse.h"
430 #include "idct_decl.h"