1 /*****************************************************************************
2 * idctmmxext.c : MMX EXT IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999-2001 VideoLAN
5 * $Id: idctmmxext.c,v 1.2 2002/08/08 00:35:11 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 EXT IDCT module") );
48 set_capability( "idct", 200 );
49 add_shortcut( "mmxext" );
50 add_requirement( MMXEXT );
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, -c2, \
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 = -C2 -C4 C2 C4 */
97 movq_r2r (mm5, mm6); /* mm6 = x7 x5 x3 x1 */
99 movq_m2r (*(table+4), mm4); /* mm4 = C6 C4 C6 C4 */
100 pmaddwd_r2r (mm0, mm3); /* mm3 = -C4*x4-C2*x6 C4*x0+C2*x2 */
102 pshufw_r2r (mm2, mm2, 0x4e); /* mm2 = x2 x0 x6 x4 */
105 static inline void Row( dctelem_t * table, s32 * rounder )
107 movq_m2r (*(table+8), mm1); /* mm1 = -C5 -C1 C3 C1 */
108 pmaddwd_r2r (mm2, mm4); /* mm4 = C4*x0+C6*x2 C4*x4+C6*x6 */
110 pmaddwd_m2r (*(table+16), mm0); /* mm0 = C4*x4-C6*x6 C4*x0-C6*x2 */
111 pshufw_r2r (mm6, mm6, 0x4e); /* mm6 = x3 x1 x7 x5 */
113 movq_m2r (*(table+12), mm7); /* mm7 = -C7 C3 C7 C5 */
114 pmaddwd_r2r (mm5, mm1); /* mm1 = -C1*x5-C5*x7 C1*x1+C3*x3 */
116 paddd_m2r (*rounder, mm3); /* mm3 += rounder */
117 pmaddwd_r2r (mm6, mm7); /* mm7 = C3*x1-C7*x3 C5*x5+C7*x7 */
119 pmaddwd_m2r (*(table+20), mm2); /* mm2 = C4*x0-C2*x2 -C4*x4+C2*x6 */
120 paddd_r2r (mm4, mm3); /* mm3 = a1 a0 + rounder */
122 pmaddwd_m2r (*(table+24), mm5); /* mm5 = C3*x5-C1*x7 C5*x1-C1*x3 */
123 movq_r2r (mm3, mm4); /* mm4 = a1 a0 + rounder */
125 pmaddwd_m2r (*(table+28), mm6); /* mm6 = C7*x1-C5*x3 C7*x5+C3*x7 */
126 paddd_r2r (mm7, mm1); /* mm1 = b1 b0 */
128 paddd_m2r (*rounder, mm0); /* mm0 += rounder */
129 psubd_r2r (mm1, mm3); /* mm3 = a1-b1 a0-b0 + rounder */
131 psrad_i2r (ROW_SHIFT, mm3); /* mm3 = y6 y7 */
132 paddd_r2r (mm4, mm1); /* mm1 = a1+b1 a0+b0 + rounder */
134 paddd_r2r (mm2, mm0); /* mm0 = a3 a2 + rounder */
135 psrad_i2r (ROW_SHIFT, mm1); /* mm1 = y1 y0 */
137 paddd_r2r (mm6, mm5); /* mm5 = b3 b2 */
138 movq_r2r (mm0, mm4); /* mm4 = a3 a2 + rounder */
140 paddd_r2r (mm5, mm0); /* mm0 = a3+b3 a2+b2 + rounder */
141 psubd_r2r (mm5, mm4); /* mm4 = a3-b3 a2-b2 + rounder */
144 static inline void RowTail( dctelem_t * row, int store )
146 psrad_i2r (ROW_SHIFT, mm0); /* mm0 = y3 y2 */
148 psrad_i2r (ROW_SHIFT, mm4); /* mm4 = y4 y5 */
150 packssdw_r2r (mm0, mm1); /* mm1 = y3 y2 y1 y0 */
152 packssdw_r2r (mm3, mm4); /* mm4 = y6 y7 y4 y5 */
154 movq_r2m (mm1, *(row+store)); /* save y3 y2 y1 y0 */
155 pshufw_r2r (mm4, mm4, 0xb1); /* mm4 = y7 y6 y5 y4 */
158 movq_r2m (mm4, *(row+store+4)); /* save y7 y6 y5 y4 */
161 static inline void RowMid( dctelem_t * row, int store,
162 int offset, dctelem_t * table)
164 movq_m2r (*(row+offset), mm2); /* mm2 = x6 x4 x2 x0 */
165 psrad_i2r (ROW_SHIFT, mm0); /* mm0 = y3 y2 */
167 movq_m2r (*(row+offset+4), mm5); /* mm5 = x7 x5 x3 x1 */
168 psrad_i2r (ROW_SHIFT, mm4); /* mm4 = y4 y5 */
170 packssdw_r2r (mm0, mm1); /* mm1 = y3 y2 y1 y0 */
171 movq_r2r (mm5, mm6); /* mm6 = x7 x5 x3 x1 */
173 packssdw_r2r (mm3, mm4); /* mm4 = y6 y7 y4 y5 */
174 movq_r2r (mm2, mm0); /* mm0 = x6 x4 x2 x0 */
176 movq_r2m (mm1, *(row+store)); /* save y3 y2 y1 y0 */
177 pshufw_r2r (mm4, mm4, 0xb1); /* mm4 = y7 y6 y5 y4 */
179 movq_m2r (*table, mm3); /* mm3 = -C2 -C4 C2 C4 */
180 movq_r2m (mm4, *(row+store+4)); /* save y7 y6 y5 y4 */
182 pmaddwd_r2r (mm0, mm3); /* mm3 = -C4*x4-C2*x6 C4*x0+C2*x2 */
184 movq_m2r (*(table+4), mm4); /* mm4 = C6 C4 C6 C4 */
185 pshufw_r2r (mm2, mm2, 0x4e); /* mm2 = x2 x0 x6 x4 */
188 static inline void Col( dctelem_t * col, int offset )
195 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
196 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
197 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
198 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
199 static mmx_t scratch0, scratch1;
201 /* column code adapted from peter gubanov */
202 /* http://www.elecard.com/peter/idct.shtml */
204 movq_m2r (*_T1, mm0); /* mm0 = T1 */
206 movq_m2r (*(col+offset+1*8), mm1); /* mm1 = x1 */
207 movq_r2r (mm0, mm2); /* mm2 = T1 */
209 movq_m2r (*(col+offset+7*8), mm4); /* mm4 = x7 */
210 pmulhw_r2r (mm1, mm0); /* mm0 = T1*x1 */
212 movq_m2r (*_T3, mm5); /* mm5 = T3 */
213 pmulhw_r2r (mm4, mm2); /* mm2 = T1*x7 */
215 movq_m2r (*(col+offset+5*8), mm6); /* mm6 = x5 */
216 movq_r2r (mm5, mm7); /* mm7 = T3-1 */
218 movq_m2r (*(col+offset+3*8), mm3); /* mm3 = x3 */
219 psubsw_r2r (mm4, mm0); /* mm0 = v17 */
221 movq_m2r (*_T2, mm4); /* mm4 = T2 */
222 pmulhw_r2r (mm3, mm5); /* mm5 = (T3-1)*x3 */
224 paddsw_r2r (mm2, mm1); /* mm1 = u17 */
225 pmulhw_r2r (mm6, mm7); /* mm7 = (T3-1)*x5 */
229 movq_r2r (mm4, mm2); /* mm2 = T2 */
230 paddsw_r2r (mm3, mm5); /* mm5 = T3*x3 */
232 pmulhw_m2r (*(col+offset+2*8), mm4);/* mm4 = T2*x2 */
233 paddsw_r2r (mm6, mm7); /* mm7 = T3*x5 */
235 psubsw_r2r (mm6, mm5); /* mm5 = v35 */
236 paddsw_r2r (mm3, mm7); /* mm7 = u35 */
238 movq_m2r (*(col+offset+6*8), mm3); /* mm3 = x6 */
239 movq_r2r (mm0, mm6); /* mm6 = v17 */
241 pmulhw_r2r (mm3, mm2); /* mm2 = T2*x6 */
242 psubsw_r2r (mm5, mm0); /* mm0 = b3 */
244 psubsw_r2r (mm3, mm4); /* mm4 = v26 */
245 paddsw_r2r (mm6, mm5); /* mm5 = v12 */
247 movq_r2m (mm0, scratch0); /* save b3 */
248 movq_r2r (mm1, mm6); /* mm6 = u17 */
250 paddsw_m2r (*(col+offset+2*8), mm2);/* mm2 = u26 */
251 paddsw_r2r (mm7, mm6); /* mm6 = b0 */
253 psubsw_r2r (mm7, mm1); /* mm1 = u12 */
254 movq_r2r (mm1, mm7); /* mm7 = u12 */
256 movq_m2r (*(col+offset+0*8), mm3); /* mm3 = x0 */
257 paddsw_r2r (mm5, mm1); /* mm1 = u12+v12 */
259 movq_m2r (*_C4, mm0); /* mm0 = C4/2 */
260 psubsw_r2r (mm5, mm7); /* mm7 = u12-v12 */
262 movq_r2m (mm6, scratch1); /* save b0 */
263 pmulhw_r2r (mm0, mm1); /* mm1 = b1/2 */
265 movq_r2r (mm4, mm6); /* mm6 = v26 */
266 pmulhw_r2r (mm0, mm7); /* mm7 = b2/2 */
268 movq_m2r (*(col+offset+4*8), mm5); /* mm5 = x4 */
269 movq_r2r (mm3, mm0); /* mm0 = x0 */
271 psubsw_r2r (mm5, mm3); /* mm3 = v04 */
272 paddsw_r2r (mm5, mm0); /* mm0 = u04 */
274 paddsw_r2r (mm3, mm4); /* mm4 = a1 */
275 movq_r2r (mm0, mm5); /* mm5 = u04 */
277 psubsw_r2r (mm6, mm3); /* mm3 = a2 */
278 paddsw_r2r (mm2, mm5); /* mm5 = a0 */
280 paddsw_r2r (mm1, mm1); /* mm1 = b1 */
281 psubsw_r2r (mm2, mm0); /* mm0 = a3 */
283 paddsw_r2r (mm7, mm7); /* mm7 = b2 */
284 movq_r2r (mm3, mm2); /* mm2 = a2 */
286 movq_r2r (mm4, mm6); /* mm6 = a1 */
287 paddsw_r2r (mm7, mm3); /* mm3 = a2+b2 */
289 psraw_i2r (COL_SHIFT, mm3); /* mm3 = y2 */
290 paddsw_r2r (mm1, mm4); /* mm4 = a1+b1 */
292 psraw_i2r (COL_SHIFT, mm4); /* mm4 = y1 */
293 psubsw_r2r (mm1, mm6); /* mm6 = a1-b1 */
295 movq_m2r (scratch1, mm1); /* mm1 = b0 */
296 psubsw_r2r (mm7, mm2); /* mm2 = a2-b2 */
298 psraw_i2r (COL_SHIFT, mm6); /* mm6 = y6 */
299 movq_r2r (mm5, mm7); /* mm7 = a0 */
301 movq_r2m (mm4, *(col+offset+1*8)); /* save y1 */
302 psraw_i2r (COL_SHIFT, mm2); /* mm2 = y5 */
304 movq_r2m (mm3, *(col+offset+2*8)); /* save y2 */
305 paddsw_r2r (mm1, mm5); /* mm5 = a0+b0 */
307 movq_m2r (scratch0, mm4); /* mm4 = b3 */
308 psubsw_r2r (mm1, mm7); /* mm7 = a0-b0 */
310 psraw_i2r (COL_SHIFT, mm5); /* mm5 = y0 */
311 movq_r2r (mm0, mm3); /* mm3 = a3 */
313 movq_r2m (mm2, *(col+offset+5*8)); /* save y5 */
314 psubsw_r2r (mm4, mm3); /* mm3 = a3-b3 */
316 psraw_i2r (COL_SHIFT, mm7); /* mm7 = y7 */
317 paddsw_r2r (mm0, mm4); /* mm4 = a3+b3 */
319 movq_r2m (mm5, *(col+offset+0*8)); /* save y0 */
320 psraw_i2r (COL_SHIFT, mm3); /* mm3 = y4 */
322 movq_r2m (mm6, *(col+offset+6*8)); /* save y6 */
323 psraw_i2r (COL_SHIFT, mm4); /* mm4 = y3 */
325 movq_r2m (mm7, *(col+offset+7*8)); /* save y7 */
327 movq_r2m (mm3, *(col+offset+4*8)); /* save y4 */
329 movq_r2m (mm4, *(col+offset+3*8)); /* save y3 */
333 static s32 rounder0[] ATTR_ALIGN(8) =
334 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
335 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
336 static s32 rounder1[] ATTR_ALIGN(8) =
337 rounder (1.25683487303); /* C1*(C1/C4+C1+C7)/2 */
338 static s32 rounder7[] ATTR_ALIGN(8) =
339 rounder (-0.25); /* C1*(C7/C4+C7-C1)/2 */
340 static s32 rounder2[] ATTR_ALIGN(8) =
341 rounder (0.60355339059); /* C2 * (C6+C2)/2 */
342 static s32 rounder6[] ATTR_ALIGN(8) =
343 rounder (-0.25); /* C2 * (C6-C2)/2 */
344 static s32 rounder3[] ATTR_ALIGN(8) =
345 rounder (0.087788325588); /* C3*(-C3/C4+C3+C5)/2 */
346 static s32 rounder5[] ATTR_ALIGN(8) =
347 rounder (-0.441341716183); /* C3*(-C5/C4+C5-C3)/2 */
349 static void IDCT( dctelem_t * p_block )
351 static dctelem_t table04[] ATTR_ALIGN(16) =
352 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
353 static dctelem_t table17[] ATTR_ALIGN(16) =
354 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
355 static dctelem_t table26[] ATTR_ALIGN(16) =
356 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
357 static dctelem_t table35[] ATTR_ALIGN(16) =
358 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
360 RowHead( p_block, 0*8, table04 );
361 Row( table04, rounder0 );
362 RowMid( p_block, 0*8, 4*8, table04 );
363 Row( table04, rounder4 );
364 RowMid( p_block, 4*8, 1*8, table17 );
365 Row( table17, rounder1 );
366 RowMid( p_block, 1*8, 7*8, table17 );
367 Row( table17, rounder7 );
368 RowMid( p_block, 7*8, 2*8, table26 );
369 Row( table26, rounder2 );
370 RowMid( p_block, 2*8, 6*8, table26 );
371 Row( table26, rounder6 );
372 RowMid( p_block, 6*8, 3*8, table35 );
373 Row( table35, rounder3 );
374 RowMid( p_block, 3*8, 5*8, table35 );
375 Row( table35, rounder5 );
376 RowTail( p_block, 5*8);
382 static inline void RestoreCPUState( )
384 /* reenables the FPU */
385 __asm__ __volatile__ ("emms");
388 #include "idct_sparse.h"
389 #include "idct_decl.h"