1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999-2001 VideoLAN
5 * $Id: idctmmx.c,v 1.28 2002/07/31 20:56:51 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 static int Open( vlc_object_t *p_this );
43 /*****************************************************************************
45 *****************************************************************************/
47 set_description( _("MMX IDCT module") );
48 set_capability( "idct", 150 );
49 add_shortcut( "mmx" );
50 add_requirement( MMX );
51 set_callbacks( Open, NULL );
54 /*****************************************************************************
55 * NormScan : This IDCT uses reordered coeffs, so we patch the scan table
56 *****************************************************************************/
57 static void NormScan( u8 ppi_scan[2][64] )
61 for( i = 0; i < 64; i++ )
64 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
67 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
71 /*****************************************************************************
73 *****************************************************************************/
77 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
78 #define rounder(bias) {round (bias), round (bias)}
80 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \
89 static inline void RowHead( dctelem_t * row, int offset, dctelem_t * table )
91 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
93 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
94 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
96 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
97 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
99 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
101 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
102 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
104 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
105 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
108 static inline void Row( dctelem_t * table, s32 * rounder )
110 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
111 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
113 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
114 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
116 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
117 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
119 paddd_m2r (*rounder, mm3); // mm3 += rounder
120 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
122 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
123 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
125 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
126 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
128 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
129 paddd_r2r (mm7, mm1); // mm1 = b1 b0
131 paddd_m2r (*rounder, mm0); // mm0 += rounder
132 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
134 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
135 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
137 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
138 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
140 paddd_r2r (mm6, mm5); // mm5 = b3 b2
141 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
143 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
144 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
147 static inline void RowTail( dctelem_t * row, int store )
149 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
151 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
153 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
155 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
157 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
158 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
160 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
162 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
164 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
168 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
171 static inline void RowMid( dctelem_t * row, int store,
172 int offset, dctelem_t * table )
174 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
175 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
177 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
178 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
180 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
181 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
183 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
184 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
186 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
187 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
189 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
190 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
192 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
193 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
195 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
196 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
198 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
199 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
201 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
202 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
205 static inline void Col( dctelem_t * col, int offset )
212 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
213 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
214 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
215 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
216 static mmx_t scratch0, scratch1;
218 /* column code adapted from peter gubanov */
219 /* http://www.elecard.com/peter/idct.shtml */
221 movq_m2r (*_T1, mm0); // mm0 = T1
223 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
224 movq_r2r (mm0, mm2); // mm2 = T1
226 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
227 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
229 movq_m2r (*_T3, mm5); // mm5 = T3
230 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
232 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
233 movq_r2r (mm5, mm7); // mm7 = T3-1
235 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
236 psubsw_r2r (mm4, mm0); // mm0 = v17
238 movq_m2r (*_T2, mm4); // mm4 = T2
239 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
241 paddsw_r2r (mm2, mm1); // mm1 = u17
242 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
246 movq_r2r (mm4, mm2); // mm2 = T2
247 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
249 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
250 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
252 psubsw_r2r (mm6, mm5); // mm5 = v35
253 paddsw_r2r (mm3, mm7); // mm7 = u35
255 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
256 movq_r2r (mm0, mm6); // mm6 = v17
258 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
259 psubsw_r2r (mm5, mm0); // mm0 = b3
261 psubsw_r2r (mm3, mm4); // mm4 = v26
262 paddsw_r2r (mm6, mm5); // mm5 = v12
264 movq_r2m (mm0, scratch0); // save b3
265 movq_r2r (mm1, mm6); // mm6 = u17
267 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
268 paddsw_r2r (mm7, mm6); // mm6 = b0
270 psubsw_r2r (mm7, mm1); // mm1 = u12
271 movq_r2r (mm1, mm7); // mm7 = u12
273 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
274 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
276 movq_m2r (*_C4, mm0); // mm0 = C4/2
277 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
279 movq_r2m (mm6, scratch1); // save b0
280 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
282 movq_r2r (mm4, mm6); // mm6 = v26
283 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
285 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
286 movq_r2r (mm3, mm0); // mm0 = x0
288 psubsw_r2r (mm5, mm3); // mm3 = v04
289 paddsw_r2r (mm5, mm0); // mm0 = u04
291 paddsw_r2r (mm3, mm4); // mm4 = a1
292 movq_r2r (mm0, mm5); // mm5 = u04
294 psubsw_r2r (mm6, mm3); // mm3 = a2
295 paddsw_r2r (mm2, mm5); // mm5 = a0
297 paddsw_r2r (mm1, mm1); // mm1 = b1
298 psubsw_r2r (mm2, mm0); // mm0 = a3
300 paddsw_r2r (mm7, mm7); // mm7 = b2
301 movq_r2r (mm3, mm2); // mm2 = a2
303 movq_r2r (mm4, mm6); // mm6 = a1
304 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
306 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
307 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
309 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
310 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
312 movq_m2r (scratch1, mm1); // mm1 = b0
313 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
315 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
316 movq_r2r (mm5, mm7); // mm7 = a0
318 movq_r2m (mm4, *(col+offset+1*8)); // save y1
319 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
321 movq_r2m (mm3, *(col+offset+2*8)); // save y2
322 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
324 movq_m2r (scratch0, mm4); // mm4 = b3
325 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
327 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
328 movq_r2r (mm0, mm3); // mm3 = a3
330 movq_r2m (mm2, *(col+offset+5*8)); // save y5
331 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
333 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
334 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
336 movq_r2m (mm5, *(col+offset+0*8)); // save y0
337 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
339 movq_r2m (mm6, *(col+offset+6*8)); // save y6
340 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
342 movq_r2m (mm7, *(col+offset+7*8)); // save y7
344 movq_r2m (mm3, *(col+offset+4*8)); // save y4
346 movq_r2m (mm4, *(col+offset+3*8)); // save y3
350 static s32 rounder0[] ATTR_ALIGN(8) =
351 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
352 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
353 static s32 rounder1[] ATTR_ALIGN(8) =
354 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
355 static s32 rounder7[] ATTR_ALIGN(8) =
356 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
357 static s32 rounder2[] ATTR_ALIGN(8) =
358 rounder (0.60355339059); // C2 * (C6+C2)/2
359 static s32 rounder6[] ATTR_ALIGN(8) =
360 rounder (-0.25); // C2 * (C6-C2)/2
361 static s32 rounder3[] ATTR_ALIGN(8) =
362 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
363 static s32 rounder5[] ATTR_ALIGN(8) =
364 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
366 static void IDCT( dctelem_t * p_block )
368 static dctelem_t table04[] ATTR_ALIGN(16) =
369 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
370 static dctelem_t table17[] ATTR_ALIGN(16) =
371 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
372 static dctelem_t table26[] ATTR_ALIGN(16) =
373 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
374 static dctelem_t table35[] ATTR_ALIGN(16) =
375 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
377 RowHead( p_block, 0*8, table04 );
378 Row( table04, rounder0 );
379 RowMid( p_block, 0*8, 4*8, table04 );
380 Row( table04, rounder4 );
381 RowMid( p_block, 4*8, 1*8, table17 );
382 Row( table17, rounder1 );
383 RowMid( p_block, 1*8, 7*8, table17 );
384 Row( table17, rounder7 );
385 RowMid( p_block, 7*8, 2*8, table26 );
386 Row( table26, rounder2 );
387 RowMid( p_block, 2*8, 6*8, table26 );
388 Row( table26, rounder6 );
389 RowMid( p_block, 6*8, 3*8, table35 );
390 Row( table35, rounder3 );
391 RowMid( p_block, 3*8, 5*8, table35 );
392 Row( table35, rounder5 );
393 RowTail( p_block, 5*8);
399 static inline void RestoreCPUState( )
401 /* reenables the FPU */
402 __asm__ __volatile__ ("emms");
405 #include "idct_sparse.h"
406 #include "idct_decl.h"