1 /*****************************************************************************
2 * idctmmxext.c : MMX EXT IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmxext.c,v 1.10 2001/06/03 12:47:21 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
30 /*****************************************************************************
32 *****************************************************************************/
41 #include "tests.h" /* TestCPU() */
44 #include "video_output.h"
46 #include "video_decoder.h"
49 #include "modules_inner.h"
51 #include "vdec_block.h"
52 #include "vdec_idct.h"
54 #include "attributes.h"
57 #include "modules_export.h"
59 /*****************************************************************************
61 *****************************************************************************/
62 static void idct_getfunctions( function_list_t * p_function_list );
63 static int idct_Probe ( probedata_t *p_data );
64 static void vdec_NormScan ( u8 ppi_scan[2][64] );
66 /*****************************************************************************
67 * Build configuration tree.
68 *****************************************************************************/
70 ADD_WINDOW( "Configuration for MMX EXT IDCT module" )
71 ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
75 p_module->i_capabilities = MODULE_CAPABILITY_NULL
76 | MODULE_CAPABILITY_IDCT;
77 p_module->psz_longname = "MMX EXT IDCT module";
81 idct_getfunctions( &p_module->p_functions->idct );
84 MODULE_DEACTIVATE_START
85 MODULE_DEACTIVATE_STOP
87 /* Following functions are local */
89 /*****************************************************************************
90 * Functions exported as capabilities.
91 *****************************************************************************/
92 static void idct_getfunctions( function_list_t * p_function_list )
94 p_function_list->pf_probe = idct_Probe;
95 #define F p_function_list->functions.idct
96 F.pf_idct_init = _M( vdec_InitIDCT );
97 F.pf_sparse_idct = _M( vdec_SparseIDCT );
98 F.pf_idct = _M( vdec_IDCT );
99 F.pf_norm_scan = vdec_NormScan;
100 F.pf_decode_init = _M( vdec_InitDecode );
101 F.pf_decode_mb_c = _M( vdec_DecodeMacroblockC );
102 F.pf_decode_mb_bw = _M( vdec_DecodeMacroblockBW );
106 /*****************************************************************************
107 * idct_Probe: return a preference score
108 *****************************************************************************/
109 static int idct_Probe( probedata_t *p_data )
111 if( !TestCPU( CPU_CAPABILITY_MMXEXT ) )
116 if( TestMethod( IDCT_METHOD_VAR, "idctmmxext" ) )
125 /*****************************************************************************
126 * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table
127 *****************************************************************************/
128 static void vdec_NormScan( u8 ppi_scan[2][64] )
132 for( i = 0; i < 64; i++ )
135 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
138 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
142 /*****************************************************************************
144 *****************************************************************************/
148 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
149 #define rounder(bias) {round (bias), round (bias)}
151 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, -c4, -c2, \
160 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
162 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
164 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
165 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
167 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
168 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
170 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
171 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
173 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
176 static __inline__ void Row( dctelem_t * table, s32 * rounder )
178 movq_m2r (*(table+8), mm1); // mm1 = -C5 -C1 C3 C1
179 pmaddwd_r2r (mm2, mm4); // mm4 = C4*x0+C6*x2 C4*x4+C6*x6
181 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x4-C6*x6 C4*x0-C6*x2
182 pshufw_r2r (mm6, mm6, 0x4e); // mm6 = x3 x1 x7 x5
184 movq_m2r (*(table+12), mm7); // mm7 = -C7 C3 C7 C5
185 pmaddwd_r2r (mm5, mm1); // mm1 = -C1*x5-C5*x7 C1*x1+C3*x3
187 paddd_m2r (*rounder, mm3); // mm3 += rounder
188 pmaddwd_r2r (mm6, mm7); // mm7 = C3*x1-C7*x3 C5*x5+C7*x7
190 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x0-C2*x2 -C4*x4+C2*x6
191 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
193 pmaddwd_m2r (*(table+24), mm5); // mm5 = C3*x5-C1*x7 C5*x1-C1*x3
194 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
196 pmaddwd_m2r (*(table+28), mm6); // mm6 = C7*x1-C5*x3 C7*x5+C3*x7
197 paddd_r2r (mm7, mm1); // mm1 = b1 b0
199 paddd_m2r (*rounder, mm0); // mm0 += rounder
200 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
202 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
203 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
205 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
206 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
208 paddd_r2r (mm6, mm5); // mm5 = b3 b2
209 movq_r2r (mm0, mm4); // mm4 = a3 a2 + rounder
211 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
212 psubd_r2r (mm5, mm4); // mm4 = a3-b3 a2-b2 + rounder
215 static __inline__ void RowTail( dctelem_t * row, int store )
217 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
219 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
221 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
223 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
225 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
226 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
229 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
232 static __inline__ void RowMid( dctelem_t * row, int store,
233 int offset, dctelem_t * table)
235 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
236 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
238 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
239 psrad_i2r (ROW_SHIFT, mm4); // mm4 = y4 y5
241 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
242 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
244 packssdw_r2r (mm3, mm4); // mm4 = y6 y7 y4 y5
245 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
247 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
248 pshufw_r2r (mm4, mm4, 0xb1); // mm4 = y7 y6 y5 y4
250 movq_m2r (*table, mm3); // mm3 = -C2 -C4 C2 C4
251 movq_r2m (mm4, *(row+store+4)); // save y7 y6 y5 y4
253 pmaddwd_r2r (mm0, mm3); // mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
255 movq_m2r (*(table+4), mm4); // mm4 = C6 C4 C6 C4
256 pshufw_r2r (mm2, mm2, 0x4e); // mm2 = x2 x0 x6 x4
259 static __inline__ void Col( dctelem_t * col, int offset )
266 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
267 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
268 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
269 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
270 static mmx_t scratch0, scratch1;
272 /* column code adapted from peter gubanov */
273 /* http://www.elecard.com/peter/idct.shtml */
275 movq_m2r (*_T1, mm0); // mm0 = T1
277 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
278 movq_r2r (mm0, mm2); // mm2 = T1
280 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
281 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
283 movq_m2r (*_T3, mm5); // mm5 = T3
284 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
286 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
287 movq_r2r (mm5, mm7); // mm7 = T3-1
289 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
290 psubsw_r2r (mm4, mm0); // mm0 = v17
292 movq_m2r (*_T2, mm4); // mm4 = T2
293 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
295 paddsw_r2r (mm2, mm1); // mm1 = u17
296 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
300 movq_r2r (mm4, mm2); // mm2 = T2
301 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
303 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
304 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
306 psubsw_r2r (mm6, mm5); // mm5 = v35
307 paddsw_r2r (mm3, mm7); // mm7 = u35
309 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
310 movq_r2r (mm0, mm6); // mm6 = v17
312 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
313 psubsw_r2r (mm5, mm0); // mm0 = b3
315 psubsw_r2r (mm3, mm4); // mm4 = v26
316 paddsw_r2r (mm6, mm5); // mm5 = v12
318 movq_r2m (mm0, scratch0); // save b3
319 movq_r2r (mm1, mm6); // mm6 = u17
321 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
322 paddsw_r2r (mm7, mm6); // mm6 = b0
324 psubsw_r2r (mm7, mm1); // mm1 = u12
325 movq_r2r (mm1, mm7); // mm7 = u12
327 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
328 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
330 movq_m2r (*_C4, mm0); // mm0 = C4/2
331 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
333 movq_r2m (mm6, scratch1); // save b0
334 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
336 movq_r2r (mm4, mm6); // mm6 = v26
337 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
339 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
340 movq_r2r (mm3, mm0); // mm0 = x0
342 psubsw_r2r (mm5, mm3); // mm3 = v04
343 paddsw_r2r (mm5, mm0); // mm0 = u04
345 paddsw_r2r (mm3, mm4); // mm4 = a1
346 movq_r2r (mm0, mm5); // mm5 = u04
348 psubsw_r2r (mm6, mm3); // mm3 = a2
349 paddsw_r2r (mm2, mm5); // mm5 = a0
351 paddsw_r2r (mm1, mm1); // mm1 = b1
352 psubsw_r2r (mm2, mm0); // mm0 = a3
354 paddsw_r2r (mm7, mm7); // mm7 = b2
355 movq_r2r (mm3, mm2); // mm2 = a2
357 movq_r2r (mm4, mm6); // mm6 = a1
358 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
360 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
361 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
363 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
364 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
366 movq_m2r (scratch1, mm1); // mm1 = b0
367 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
369 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
370 movq_r2r (mm5, mm7); // mm7 = a0
372 movq_r2m (mm4, *(col+offset+1*8)); // save y1
373 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
375 movq_r2m (mm3, *(col+offset+2*8)); // save y2
376 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
378 movq_m2r (scratch0, mm4); // mm4 = b3
379 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
381 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
382 movq_r2r (mm0, mm3); // mm3 = a3
384 movq_r2m (mm2, *(col+offset+5*8)); // save y5
385 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
387 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
388 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
390 movq_r2m (mm5, *(col+offset+0*8)); // save y0
391 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
393 movq_r2m (mm6, *(col+offset+6*8)); // save y6
394 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
396 movq_r2m (mm7, *(col+offset+7*8)); // save y7
398 movq_r2m (mm3, *(col+offset+4*8)); // save y4
400 movq_r2m (mm4, *(col+offset+3*8)); // save y3
404 static s32 rounder0[] ATTR_ALIGN(8) =
405 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
406 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
407 static s32 rounder1[] ATTR_ALIGN(8) =
408 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
409 static s32 rounder7[] ATTR_ALIGN(8) =
410 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
411 static s32 rounder2[] ATTR_ALIGN(8) =
412 rounder (0.60355339059); // C2 * (C6+C2)/2
413 static s32 rounder6[] ATTR_ALIGN(8) =
414 rounder (-0.25); // C2 * (C6-C2)/2
415 static s32 rounder3[] ATTR_ALIGN(8) =
416 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
417 static s32 rounder5[] ATTR_ALIGN(8) =
418 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
420 void _M( vdec_IDCT )( vdec_thread_t * p_vdec, dctelem_t * p_block,
423 static dctelem_t table04[] ATTR_ALIGN(16) =
424 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
425 static dctelem_t table17[] ATTR_ALIGN(16) =
426 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
427 static dctelem_t table26[] ATTR_ALIGN(16) =
428 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
429 static dctelem_t table35[] ATTR_ALIGN(16) =
430 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
432 RowHead( p_block, 0*8, table04 );
433 Row( table04, rounder0 );
434 RowMid( p_block, 0*8, 4*8, table04 );
435 Row( table04, rounder4 );
436 RowMid( p_block, 4*8, 1*8, table17 );
437 Row( table17, rounder1 );
438 RowMid( p_block, 1*8, 7*8, table17 );
439 Row( table17, rounder7 );
440 RowMid( p_block, 7*8, 2*8, table26 );
441 Row( table26, rounder2 );
442 RowMid( p_block, 2*8, 6*8, table26 );
443 Row( table26, rounder6 );
444 RowMid( p_block, 6*8, 3*8, table35 );
445 Row( table35, rounder3 );
446 RowMid( p_block, 3*8, 5*8, table35 );
447 Row( table35, rounder5 );
448 RowTail( p_block, 5*8);