1 /*****************************************************************************
2 * idctmmxext.c : MMX EXT IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmxext.c,v 1.13 2001/08/22 17:21:45 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() */
44 #include "vdec_idct.h"
49 #include "modules_export.h"
51 /*****************************************************************************
53 *****************************************************************************/
54 static void idct_getfunctions( function_list_t * p_function_list );
55 static int idct_Probe ( probedata_t *p_data );
56 static void vdec_NormScan ( u8 ppi_scan[2][64] );
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 * Functions exported as capabilities.
83 *****************************************************************************/
84 static void idct_getfunctions( function_list_t * p_function_list )
86 p_function_list->pf_probe = idct_Probe;
87 #define F p_function_list->functions.idct
88 F.pf_idct_init = _M( vdec_InitIDCT );
89 F.pf_sparse_idct = _M( vdec_SparseIDCT );
90 F.pf_idct = _M( vdec_IDCT );
91 F.pf_norm_scan = vdec_NormScan;
92 F.pf_decode_init = _M( vdec_InitDecode );
93 F.pf_addblock = _M( vdec_AddBlock );
94 F.pf_copyblock = _M( vdec_CopyBlock );
98 /*****************************************************************************
99 * idct_Probe: return a preference score
100 *****************************************************************************/
101 static int idct_Probe( probedata_t *p_data )
103 if( !TestCPU( CPU_CAPABILITY_MMXEXT ) )
108 if( TestMethod( IDCT_METHOD_VAR, "idctmmxext" )
109 || TestMethod( IDCT_METHOD_VAR, "mmxext" ) )
118 /*****************************************************************************
119 * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table
120 *****************************************************************************/
121 static void vdec_NormScan( u8 ppi_scan[2][64] )
125 for( i = 0; i < 64; i++ )
128 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
131 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
135 /*****************************************************************************
137 *****************************************************************************/
141 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
142 #define rounder(bias) {round (bias), round (bias)}
144 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, -c4, -c2, \
153 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
155 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
157 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
158 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
160 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
161 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
163 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
164 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
166 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
169 static __inline__ void Row( dctelem_t * table, s32 * rounder )
171 movq_m2r (*(table+8), mm1); // mm1 = -C5 -C1 C3 C1
172 pmaddwd_r2r (mm2, mm4); // mm4 = C4*x0+C6*x2 C4*x4+C6*x6
174 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x4-C6*x6 C4*x0-C6*x2
175 pshufw_r2r (mm6, mm6, 0x4e); // mm6 = x3 x1 x7 x5
177 movq_m2r (*(table+12), mm7); // mm7 = -C7 C3 C7 C5
178 pmaddwd_r2r (mm5, mm1); // mm1 = -C1*x5-C5*x7 C1*x1+C3*x3
180 paddd_m2r (*rounder, mm3); // mm3 += rounder
181 pmaddwd_r2r (mm6, mm7); // mm7 = C3*x1-C7*x3 C5*x5+C7*x7
183 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x0-C2*x2 -C4*x4+C2*x6
184 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
186 pmaddwd_m2r (*(table+24), mm5); // mm5 = C3*x5-C1*x7 C5*x1-C1*x3
187 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
189 pmaddwd_m2r (*(table+28), mm6); // mm6 = C7*x1-C5*x3 C7*x5+C3*x7
190 paddd_r2r (mm7, mm1); // mm1 = b1 b0
192 paddd_m2r (*rounder, mm0); // mm0 += rounder
193 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
195 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
196 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
198 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
199 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
201 paddd_r2r (mm6, mm5); // mm5 = b3 b2
202 movq_r2r (mm0, mm4); // mm4 = a3 a2 + rounder
204 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
205 psubd_r2r (mm5, mm4); // mm4 = a3-b3 a2-b2 + rounder
208 static __inline__ void RowTail( dctelem_t * row, int store )
210 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
212 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
214 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
216 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
218 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
219 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
222 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
225 static __inline__ void RowMid( dctelem_t * row, int store,
226 int offset, dctelem_t * table)
228 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
229 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
231 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
232 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
234 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
235 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
237 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
238 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
240 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
241 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
243 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
244 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
246 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
248 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
249 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
252 static __inline__ void Col( dctelem_t * col, int offset )
259 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
260 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
261 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
262 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
263 static mmx_t scratch0, scratch1;
265 /* column code adapted from peter gubanov */
266 /* http://www.elecard.com/peter/idct.shtml */
268 movq_m2r (*_T1, mm0); // mm0 = T1
270 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
271 movq_r2r (mm0, mm2); // mm2 = T1
273 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
274 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
276 movq_m2r (*_T3, mm5); // mm5 = T3
277 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
279 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
280 movq_r2r (mm5, mm7); // mm7 = T3-1
282 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
283 psubsw_r2r (mm4, mm0); // mm0 = v17
285 movq_m2r (*_T2, mm4); // mm4 = T2
286 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
288 paddsw_r2r (mm2, mm1); // mm1 = u17
289 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
293 movq_r2r (mm4, mm2); // mm2 = T2
294 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
296 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
297 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
299 psubsw_r2r (mm6, mm5); // mm5 = v35
300 paddsw_r2r (mm3, mm7); // mm7 = u35
302 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
303 movq_r2r (mm0, mm6); // mm6 = v17
305 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
306 psubsw_r2r (mm5, mm0); // mm0 = b3
308 psubsw_r2r (mm3, mm4); // mm4 = v26
309 paddsw_r2r (mm6, mm5); // mm5 = v12
311 movq_r2m (mm0, scratch0); // save b3
312 movq_r2r (mm1, mm6); // mm6 = u17
314 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
315 paddsw_r2r (mm7, mm6); // mm6 = b0
317 psubsw_r2r (mm7, mm1); // mm1 = u12
318 movq_r2r (mm1, mm7); // mm7 = u12
320 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
321 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
323 movq_m2r (*_C4, mm0); // mm0 = C4/2
324 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
326 movq_r2m (mm6, scratch1); // save b0
327 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
329 movq_r2r (mm4, mm6); // mm6 = v26
330 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
332 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
333 movq_r2r (mm3, mm0); // mm0 = x0
335 psubsw_r2r (mm5, mm3); // mm3 = v04
336 paddsw_r2r (mm5, mm0); // mm0 = u04
338 paddsw_r2r (mm3, mm4); // mm4 = a1
339 movq_r2r (mm0, mm5); // mm5 = u04
341 psubsw_r2r (mm6, mm3); // mm3 = a2
342 paddsw_r2r (mm2, mm5); // mm5 = a0
344 paddsw_r2r (mm1, mm1); // mm1 = b1
345 psubsw_r2r (mm2, mm0); // mm0 = a3
347 paddsw_r2r (mm7, mm7); // mm7 = b2
348 movq_r2r (mm3, mm2); // mm2 = a2
350 movq_r2r (mm4, mm6); // mm6 = a1
351 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
353 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
354 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
356 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
357 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
359 movq_m2r (scratch1, mm1); // mm1 = b0
360 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
362 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
363 movq_r2r (mm5, mm7); // mm7 = a0
365 movq_r2m (mm4, *(col+offset+1*8)); // save y1
366 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
368 movq_r2m (mm3, *(col+offset+2*8)); // save y2
369 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
371 movq_m2r (scratch0, mm4); // mm4 = b3
372 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
374 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
375 movq_r2r (mm0, mm3); // mm3 = a3
377 movq_r2m (mm2, *(col+offset+5*8)); // save y5
378 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
380 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
381 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
383 movq_r2m (mm5, *(col+offset+0*8)); // save y0
384 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
386 movq_r2m (mm6, *(col+offset+6*8)); // save y6
387 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
389 movq_r2m (mm7, *(col+offset+7*8)); // save y7
391 movq_r2m (mm3, *(col+offset+4*8)); // save y4
393 movq_r2m (mm4, *(col+offset+3*8)); // save y3
397 static s32 rounder0[] ATTR_ALIGN(8) =
398 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
399 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
400 static s32 rounder1[] ATTR_ALIGN(8) =
401 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
402 static s32 rounder7[] ATTR_ALIGN(8) =
403 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
404 static s32 rounder2[] ATTR_ALIGN(8) =
405 rounder (0.60355339059); // C2 * (C6+C2)/2
406 static s32 rounder6[] ATTR_ALIGN(8) =
407 rounder (-0.25); // C2 * (C6-C2)/2
408 static s32 rounder3[] ATTR_ALIGN(8) =
409 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
410 static s32 rounder5[] ATTR_ALIGN(8) =
411 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
413 void _M( vdec_IDCT )( void * p_unused_data, dctelem_t * p_block,
416 static dctelem_t table04[] ATTR_ALIGN(16) =
417 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
418 static dctelem_t table17[] ATTR_ALIGN(16) =
419 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
420 static dctelem_t table26[] ATTR_ALIGN(16) =
421 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
422 static dctelem_t table35[] ATTR_ALIGN(16) =
423 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
425 RowHead( p_block, 0*8, table04 );
426 Row( table04, rounder0 );
427 RowMid( p_block, 0*8, 4*8, table04 );
428 Row( table04, rounder4 );
429 RowMid( p_block, 4*8, 1*8, table17 );
430 Row( table17, rounder1 );
431 RowMid( p_block, 1*8, 7*8, table17 );
432 Row( table17, rounder7 );
433 RowMid( p_block, 7*8, 2*8, table26 );
434 Row( table26, rounder2 );
435 RowMid( p_block, 2*8, 6*8, table26 );
436 Row( table26, rounder6 );
437 RowMid( p_block, 6*8, 3*8, table35 );
438 Row( table35, rounder3 );
439 RowMid( p_block, 3*8, 5*8, table35 );
440 Row( table35, rounder5 );
441 RowTail( p_block, 5*8);