1 /*****************************************************************************
2 * idctmmxext.c : MMX EXT IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999-2001 VideoLAN
5 * $Id: idctmmxext.c,v 1.24 2002/06/01 12:31:59 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 /*****************************************************************************
30 *****************************************************************************/
39 #include "block_mmx.h"
41 /*****************************************************************************
43 *****************************************************************************/
44 static void idct_getfunctions( function_list_t * p_function_list );
46 /*****************************************************************************
47 * Build configuration tree.
48 *****************************************************************************/
53 SET_DESCRIPTION( _("MMX EXT IDCT module") )
54 ADD_CAPABILITY( IDCT, 200 )
55 ADD_REQUIREMENT( MMXEXT )
56 ADD_SHORTCUT( "mmxext" )
60 idct_getfunctions( &p_module->p_functions->idct );
63 MODULE_DEACTIVATE_START
64 MODULE_DEACTIVATE_STOP
66 /* Following functions are local */
68 /*****************************************************************************
69 * NormScan : This IDCT uses reordered coeffs, so we patch the scan table
70 *****************************************************************************/
71 static void NormScan( u8 ppi_scan[2][64] )
75 for( i = 0; i < 64; i++ )
78 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
81 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
85 /*****************************************************************************
87 *****************************************************************************/
91 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
92 #define rounder(bias) {round (bias), round (bias)}
94 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, -c4, -c2, \
103 static inline void RowHead( dctelem_t * row, int offset, dctelem_t * table )
105 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
107 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
108 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
110 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
111 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
113 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
114 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
116 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
119 static inline void Row( dctelem_t * table, s32 * rounder )
121 movq_m2r (*(table+8), mm1); // mm1 = -C5 -C1 C3 C1
122 pmaddwd_r2r (mm2, mm4); // mm4 = C4*x0+C6*x2 C4*x4+C6*x6
124 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x4-C6*x6 C4*x0-C6*x2
125 pshufw_r2r (mm6, mm6, 0x4e); // mm6 = x3 x1 x7 x5
127 movq_m2r (*(table+12), mm7); // mm7 = -C7 C3 C7 C5
128 pmaddwd_r2r (mm5, mm1); // mm1 = -C1*x5-C5*x7 C1*x1+C3*x3
130 paddd_m2r (*rounder, mm3); // mm3 += rounder
131 pmaddwd_r2r (mm6, mm7); // mm7 = C3*x1-C7*x3 C5*x5+C7*x7
133 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x0-C2*x2 -C4*x4+C2*x6
134 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
136 pmaddwd_m2r (*(table+24), mm5); // mm5 = C3*x5-C1*x7 C5*x1-C1*x3
137 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
139 pmaddwd_m2r (*(table+28), mm6); // mm6 = C7*x1-C5*x3 C7*x5+C3*x7
140 paddd_r2r (mm7, mm1); // mm1 = b1 b0
142 paddd_m2r (*rounder, mm0); // mm0 += rounder
143 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
145 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
146 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
148 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
149 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
151 paddd_r2r (mm6, mm5); // mm5 = b3 b2
152 movq_r2r (mm0, mm4); // mm4 = a3 a2 + rounder
154 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
155 psubd_r2r (mm5, mm4); // mm4 = a3-b3 a2-b2 + rounder
158 static inline void RowTail( dctelem_t * row, int store )
160 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
162 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
164 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
166 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
168 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
169 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
172 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
175 static inline void RowMid( dctelem_t * row, int store,
176 int offset, dctelem_t * table)
178 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
179 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
181 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
182 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
184 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
185 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
187 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
188 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
190 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
191 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
193 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
194 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
196 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
198 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
199 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
202 static inline void Col( dctelem_t * col, int offset )
209 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
210 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
211 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
212 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
213 static mmx_t scratch0, scratch1;
215 /* column code adapted from peter gubanov */
216 /* http://www.elecard.com/peter/idct.shtml */
218 movq_m2r (*_T1, mm0); // mm0 = T1
220 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
221 movq_r2r (mm0, mm2); // mm2 = T1
223 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
224 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
226 movq_m2r (*_T3, mm5); // mm5 = T3
227 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
229 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
230 movq_r2r (mm5, mm7); // mm7 = T3-1
232 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
233 psubsw_r2r (mm4, mm0); // mm0 = v17
235 movq_m2r (*_T2, mm4); // mm4 = T2
236 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
238 paddsw_r2r (mm2, mm1); // mm1 = u17
239 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
243 movq_r2r (mm4, mm2); // mm2 = T2
244 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
246 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
247 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
249 psubsw_r2r (mm6, mm5); // mm5 = v35
250 paddsw_r2r (mm3, mm7); // mm7 = u35
252 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
253 movq_r2r (mm0, mm6); // mm6 = v17
255 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
256 psubsw_r2r (mm5, mm0); // mm0 = b3
258 psubsw_r2r (mm3, mm4); // mm4 = v26
259 paddsw_r2r (mm6, mm5); // mm5 = v12
261 movq_r2m (mm0, scratch0); // save b3
262 movq_r2r (mm1, mm6); // mm6 = u17
264 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
265 paddsw_r2r (mm7, mm6); // mm6 = b0
267 psubsw_r2r (mm7, mm1); // mm1 = u12
268 movq_r2r (mm1, mm7); // mm7 = u12
270 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
271 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
273 movq_m2r (*_C4, mm0); // mm0 = C4/2
274 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
276 movq_r2m (mm6, scratch1); // save b0
277 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
279 movq_r2r (mm4, mm6); // mm6 = v26
280 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
282 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
283 movq_r2r (mm3, mm0); // mm0 = x0
285 psubsw_r2r (mm5, mm3); // mm3 = v04
286 paddsw_r2r (mm5, mm0); // mm0 = u04
288 paddsw_r2r (mm3, mm4); // mm4 = a1
289 movq_r2r (mm0, mm5); // mm5 = u04
291 psubsw_r2r (mm6, mm3); // mm3 = a2
292 paddsw_r2r (mm2, mm5); // mm5 = a0
294 paddsw_r2r (mm1, mm1); // mm1 = b1
295 psubsw_r2r (mm2, mm0); // mm0 = a3
297 paddsw_r2r (mm7, mm7); // mm7 = b2
298 movq_r2r (mm3, mm2); // mm2 = a2
300 movq_r2r (mm4, mm6); // mm6 = a1
301 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
303 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
304 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
306 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
307 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
309 movq_m2r (scratch1, mm1); // mm1 = b0
310 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
312 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
313 movq_r2r (mm5, mm7); // mm7 = a0
315 movq_r2m (mm4, *(col+offset+1*8)); // save y1
316 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
318 movq_r2m (mm3, *(col+offset+2*8)); // save y2
319 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
321 movq_m2r (scratch0, mm4); // mm4 = b3
322 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
324 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
325 movq_r2r (mm0, mm3); // mm3 = a3
327 movq_r2m (mm2, *(col+offset+5*8)); // save y5
328 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
330 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
331 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
333 movq_r2m (mm5, *(col+offset+0*8)); // save y0
334 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
336 movq_r2m (mm6, *(col+offset+6*8)); // save y6
337 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
339 movq_r2m (mm7, *(col+offset+7*8)); // save y7
341 movq_r2m (mm3, *(col+offset+4*8)); // save y4
343 movq_r2m (mm4, *(col+offset+3*8)); // save y3
347 static s32 rounder0[] ATTR_ALIGN(8) =
348 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
349 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
350 static s32 rounder1[] ATTR_ALIGN(8) =
351 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
352 static s32 rounder7[] ATTR_ALIGN(8) =
353 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
354 static s32 rounder2[] ATTR_ALIGN(8) =
355 rounder (0.60355339059); // C2 * (C6+C2)/2
356 static s32 rounder6[] ATTR_ALIGN(8) =
357 rounder (-0.25); // C2 * (C6-C2)/2
358 static s32 rounder3[] ATTR_ALIGN(8) =
359 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
360 static s32 rounder5[] ATTR_ALIGN(8) =
361 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
363 static void IDCT( dctelem_t * p_block )
365 static dctelem_t table04[] ATTR_ALIGN(16) =
366 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
367 static dctelem_t table17[] ATTR_ALIGN(16) =
368 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
369 static dctelem_t table26[] ATTR_ALIGN(16) =
370 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
371 static dctelem_t table35[] ATTR_ALIGN(16) =
372 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
374 RowHead( p_block, 0*8, table04 );
375 Row( table04, rounder0 );
376 RowMid( p_block, 0*8, 4*8, table04 );
377 Row( table04, rounder4 );
378 RowMid( p_block, 4*8, 1*8, table17 );
379 Row( table17, rounder1 );
380 RowMid( p_block, 1*8, 7*8, table17 );
381 Row( table17, rounder7 );
382 RowMid( p_block, 7*8, 2*8, table26 );
383 Row( table26, rounder2 );
384 RowMid( p_block, 2*8, 6*8, table26 );
385 Row( table26, rounder6 );
386 RowMid( p_block, 6*8, 3*8, table35 );
387 Row( table35, rounder3 );
388 RowMid( p_block, 3*8, 5*8, table35 );
389 Row( table35, rounder5 );
390 RowTail( p_block, 5*8);
396 static inline void RestoreCPUState( )
398 /* reenables the FPU */
399 __asm__ __volatile__ ("emms");
402 #include "idct_sparse.h"
403 #include "idct_decl.h"