1 /*****************************************************************************
2 * idctmmx.c : MMX IDCT module
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idctmmx.c,v 1.17 2001/09/05 16:07:49 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 idctmmx
29 #include "modules_inner.h"
31 /*****************************************************************************
33 *****************************************************************************/
42 #include "tests.h" /* TestCPU() */
47 #include "block_mmx.h"
50 #include "modules_export.h"
52 /*****************************************************************************
54 *****************************************************************************/
55 static void idct_getfunctions( function_list_t * p_function_list );
57 /*****************************************************************************
58 * Build configuration tree.
59 *****************************************************************************/
61 ADD_WINDOW( "Configuration for MMX IDCT module" )
62 ADD_COMMENT( "Ha, ha -- nothing to configure yet" )
66 p_module->i_capabilities = MODULE_CAPABILITY_NULL
67 | MODULE_CAPABILITY_IDCT;
68 p_module->psz_longname = "MMX IDCT module";
72 idct_getfunctions( &p_module->p_functions->idct );
75 MODULE_DEACTIVATE_START
76 MODULE_DEACTIVATE_STOP
78 /* Following functions are local */
80 /*****************************************************************************
81 * idct_Probe: return a preference score
82 *****************************************************************************/
83 static int idct_Probe( probedata_t *p_data )
85 if( !TestCPU( CPU_CAPABILITY_MMX ) )
90 if( TestMethod( IDCT_METHOD_VAR, "idctmmx" )
91 || TestMethod( IDCT_METHOD_VAR, "mmx" ) )
99 /*****************************************************************************
100 * NormScan : This IDCT uses reordered coeffs, so we patch the scan table
101 *****************************************************************************/
102 static void NormScan( u8 ppi_scan[2][64] )
106 for( i = 0; i < 64; i++ )
109 ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
112 ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);
116 /*****************************************************************************
118 *****************************************************************************/
122 #define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
123 #define rounder(bias) {round (bias), round (bias)}
125 #define table(c1,c2,c3,c4,c5,c6,c7) { c4, c2, c4, c6, \
134 static __inline__ void RowHead( dctelem_t * row, int offset, dctelem_t * table )
136 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
138 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
139 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
141 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
142 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
144 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
146 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
147 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
149 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
150 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
153 static __inline__ void Row( dctelem_t * table, s32 * rounder )
155 pmaddwd_r2r (mm2, mm4); // mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
156 punpckldq_r2r (mm5, mm5); // mm5 = x3 x1 x3 x1
158 pmaddwd_m2r (*(table+16), mm0); // mm0 = C4*x0-C2*x2 C4*x0-C6*x2
159 punpckhdq_r2r (mm6, mm6); // mm6 = x7 x5 x7 x5
161 movq_m2r (*(table+12), mm7); // mm7 = -C5 -C1 C7 C5
162 pmaddwd_r2r (mm5, mm1); // mm1 = C3*x1-C7*x3 C1*x1+C3*x3
164 paddd_m2r (*rounder, mm3); // mm3 += rounder
165 pmaddwd_r2r (mm6, mm7); // mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
167 pmaddwd_m2r (*(table+20), mm2); // mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
168 paddd_r2r (mm4, mm3); // mm3 = a1 a0 + rounder
170 pmaddwd_m2r (*(table+24), mm5); // mm5 = C7*x1-C5*x3 C5*x1-C1*x3
171 movq_r2r (mm3, mm4); // mm4 = a1 a0 + rounder
173 pmaddwd_m2r (*(table+28), mm6); // mm6 = C3*x5-C1*x7 C7*x5+C3*x7
174 paddd_r2r (mm7, mm1); // mm1 = b1 b0
176 paddd_m2r (*rounder, mm0); // mm0 += rounder
177 psubd_r2r (mm1, mm3); // mm3 = a1-b1 a0-b0 + rounder
179 psrad_i2r (ROW_SHIFT, mm3); // mm3 = y6 y7
180 paddd_r2r (mm4, mm1); // mm1 = a1+b1 a0+b0 + rounder
182 paddd_r2r (mm2, mm0); // mm0 = a3 a2 + rounder
183 psrad_i2r (ROW_SHIFT, mm1); // mm1 = y1 y0
185 paddd_r2r (mm6, mm5); // mm5 = b3 b2
186 movq_r2r (mm0, mm7); // mm7 = a3 a2 + rounder
188 paddd_r2r (mm5, mm0); // mm0 = a3+b3 a2+b2 + rounder
189 psubd_r2r (mm5, mm7); // mm7 = a3-b3 a2-b2 + rounder
192 static __inline__ void RowTail( dctelem_t * row, int store )
194 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
196 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
198 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
200 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
202 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
203 movq_r2r (mm7, mm4); // mm4 = y6 y7 y4 y5
205 pslld_i2r (16, mm7); // mm7 = y7 0 y5 0
207 psrld_i2r (16, mm4); // mm4 = 0 y6 0 y4
209 por_r2r (mm4, mm7); // mm7 = y7 y6 y5 y4
213 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
216 static __inline__ void RowMid( dctelem_t * row, int store,
217 int offset, dctelem_t * table )
219 movq_m2r (*(row+offset), mm2); // mm2 = x6 x4 x2 x0
220 psrad_i2r (ROW_SHIFT, mm0); // mm0 = y3 y2
222 movq_m2r (*(row+offset+4), mm5); // mm5 = x7 x5 x3 x1
223 psrad_i2r (ROW_SHIFT, mm7); // mm7 = y4 y5
225 packssdw_r2r (mm0, mm1); // mm1 = y3 y2 y1 y0
226 movq_r2r (mm5, mm6); // mm6 = x7 x5 x3 x1
228 packssdw_r2r (mm3, mm7); // mm7 = y6 y7 y4 y5
229 movq_r2r (mm2, mm0); // mm0 = x6 x4 x2 x0
231 movq_r2m (mm1, *(row+store)); // save y3 y2 y1 y0
232 movq_r2r (mm7, mm1); // mm1 = y6 y7 y4 y5
234 punpckldq_r2r (mm0, mm0); // mm0 = x2 x0 x2 x0
235 psrld_i2r (16, mm7); // mm7 = 0 y6 0 y4
237 movq_m2r (*table, mm3); // mm3 = C6 C4 C2 C4
238 pslld_i2r (16, mm1); // mm1 = y7 0 y5 0
240 movq_m2r (*(table+4), mm4); // mm4 = -C2 -C4 C6 C4
241 por_r2r (mm1, mm7); // mm7 = y7 y6 y5 y4
243 movq_m2r (*(table+8), mm1); // mm1 = -C7 C3 C3 C1
244 punpckhdq_r2r (mm2, mm2); // mm2 = x6 x4 x6 x4
246 movq_r2m (mm7, *(row+store+4)); // save y7 y6 y5 y4
247 pmaddwd_r2r (mm0, mm3); // mm3 = C4*x0+C6*x2 C4*x0+C2*x2
250 static __inline__ void Col( dctelem_t * col, int offset )
257 static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
258 static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
259 static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
260 static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
261 static mmx_t scratch0, scratch1;
263 /* column code adapted from peter gubanov */
264 /* http://www.elecard.com/peter/idct.shtml */
266 movq_m2r (*_T1, mm0); // mm0 = T1
268 movq_m2r (*(col+offset+1*8), mm1); // mm1 = x1
269 movq_r2r (mm0, mm2); // mm2 = T1
271 movq_m2r (*(col+offset+7*8), mm4); // mm4 = x7
272 pmulhw_r2r (mm1, mm0); // mm0 = T1*x1
274 movq_m2r (*_T3, mm5); // mm5 = T3
275 pmulhw_r2r (mm4, mm2); // mm2 = T1*x7
277 movq_m2r (*(col+offset+5*8), mm6); // mm6 = x5
278 movq_r2r (mm5, mm7); // mm7 = T3-1
280 movq_m2r (*(col+offset+3*8), mm3); // mm3 = x3
281 psubsw_r2r (mm4, mm0); // mm0 = v17
283 movq_m2r (*_T2, mm4); // mm4 = T2
284 pmulhw_r2r (mm3, mm5); // mm5 = (T3-1)*x3
286 paddsw_r2r (mm2, mm1); // mm1 = u17
287 pmulhw_r2r (mm6, mm7); // mm7 = (T3-1)*x5
291 movq_r2r (mm4, mm2); // mm2 = T2
292 paddsw_r2r (mm3, mm5); // mm5 = T3*x3
294 pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
295 paddsw_r2r (mm6, mm7); // mm7 = T3*x5
297 psubsw_r2r (mm6, mm5); // mm5 = v35
298 paddsw_r2r (mm3, mm7); // mm7 = u35
300 movq_m2r (*(col+offset+6*8), mm3); // mm3 = x6
301 movq_r2r (mm0, mm6); // mm6 = v17
303 pmulhw_r2r (mm3, mm2); // mm2 = T2*x6
304 psubsw_r2r (mm5, mm0); // mm0 = b3
306 psubsw_r2r (mm3, mm4); // mm4 = v26
307 paddsw_r2r (mm6, mm5); // mm5 = v12
309 movq_r2m (mm0, scratch0); // save b3
310 movq_r2r (mm1, mm6); // mm6 = u17
312 paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
313 paddsw_r2r (mm7, mm6); // mm6 = b0
315 psubsw_r2r (mm7, mm1); // mm1 = u12
316 movq_r2r (mm1, mm7); // mm7 = u12
318 movq_m2r (*(col+offset+0*8), mm3); // mm3 = x0
319 paddsw_r2r (mm5, mm1); // mm1 = u12+v12
321 movq_m2r (*_C4, mm0); // mm0 = C4/2
322 psubsw_r2r (mm5, mm7); // mm7 = u12-v12
324 movq_r2m (mm6, scratch1); // save b0
325 pmulhw_r2r (mm0, mm1); // mm1 = b1/2
327 movq_r2r (mm4, mm6); // mm6 = v26
328 pmulhw_r2r (mm0, mm7); // mm7 = b2/2
330 movq_m2r (*(col+offset+4*8), mm5); // mm5 = x4
331 movq_r2r (mm3, mm0); // mm0 = x0
333 psubsw_r2r (mm5, mm3); // mm3 = v04
334 paddsw_r2r (mm5, mm0); // mm0 = u04
336 paddsw_r2r (mm3, mm4); // mm4 = a1
337 movq_r2r (mm0, mm5); // mm5 = u04
339 psubsw_r2r (mm6, mm3); // mm3 = a2
340 paddsw_r2r (mm2, mm5); // mm5 = a0
342 paddsw_r2r (mm1, mm1); // mm1 = b1
343 psubsw_r2r (mm2, mm0); // mm0 = a3
345 paddsw_r2r (mm7, mm7); // mm7 = b2
346 movq_r2r (mm3, mm2); // mm2 = a2
348 movq_r2r (mm4, mm6); // mm6 = a1
349 paddsw_r2r (mm7, mm3); // mm3 = a2+b2
351 psraw_i2r (COL_SHIFT, mm3); // mm3 = y2
352 paddsw_r2r (mm1, mm4); // mm4 = a1+b1
354 psraw_i2r (COL_SHIFT, mm4); // mm4 = y1
355 psubsw_r2r (mm1, mm6); // mm6 = a1-b1
357 movq_m2r (scratch1, mm1); // mm1 = b0
358 psubsw_r2r (mm7, mm2); // mm2 = a2-b2
360 psraw_i2r (COL_SHIFT, mm6); // mm6 = y6
361 movq_r2r (mm5, mm7); // mm7 = a0
363 movq_r2m (mm4, *(col+offset+1*8)); // save y1
364 psraw_i2r (COL_SHIFT, mm2); // mm2 = y5
366 movq_r2m (mm3, *(col+offset+2*8)); // save y2
367 paddsw_r2r (mm1, mm5); // mm5 = a0+b0
369 movq_m2r (scratch0, mm4); // mm4 = b3
370 psubsw_r2r (mm1, mm7); // mm7 = a0-b0
372 psraw_i2r (COL_SHIFT, mm5); // mm5 = y0
373 movq_r2r (mm0, mm3); // mm3 = a3
375 movq_r2m (mm2, *(col+offset+5*8)); // save y5
376 psubsw_r2r (mm4, mm3); // mm3 = a3-b3
378 psraw_i2r (COL_SHIFT, mm7); // mm7 = y7
379 paddsw_r2r (mm0, mm4); // mm4 = a3+b3
381 movq_r2m (mm5, *(col+offset+0*8)); // save y0
382 psraw_i2r (COL_SHIFT, mm3); // mm3 = y4
384 movq_r2m (mm6, *(col+offset+6*8)); // save y6
385 psraw_i2r (COL_SHIFT, mm4); // mm4 = y3
387 movq_r2m (mm7, *(col+offset+7*8)); // save y7
389 movq_r2m (mm3, *(col+offset+4*8)); // save y4
391 movq_r2m (mm4, *(col+offset+3*8)); // save y3
395 static s32 rounder0[] ATTR_ALIGN(8) =
396 rounder ((1 << (COL_SHIFT - 1)) - 0.5);
397 static s32 rounder4[] ATTR_ALIGN(8) = rounder (0);
398 static s32 rounder1[] ATTR_ALIGN(8) =
399 rounder (1.25683487303); // C1*(C1/C4+C1+C7)/2
400 static s32 rounder7[] ATTR_ALIGN(8) =
401 rounder (-0.25); // C1*(C7/C4+C7-C1)/2
402 static s32 rounder2[] ATTR_ALIGN(8) =
403 rounder (0.60355339059); // C2 * (C6+C2)/2
404 static s32 rounder6[] ATTR_ALIGN(8) =
405 rounder (-0.25); // C2 * (C6-C2)/2
406 static s32 rounder3[] ATTR_ALIGN(8) =
407 rounder (0.087788325588); // C3*(-C3/C4+C3+C5)/2
408 static s32 rounder5[] ATTR_ALIGN(8) =
409 rounder (-0.441341716183); // C3*(-C5/C4+C5-C3)/2
411 static __inline__ void IDCT( dctelem_t * p_block )
413 static dctelem_t table04[] ATTR_ALIGN(16) =
414 table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
415 static dctelem_t table17[] ATTR_ALIGN(16) =
416 table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
417 static dctelem_t table26[] ATTR_ALIGN(16) =
418 table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
419 static dctelem_t table35[] ATTR_ALIGN(16) =
420 table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
422 RowHead( p_block, 0*8, table04 );
423 Row( table04, rounder0 );
424 RowMid( p_block, 0*8, 4*8, table04 );
425 Row( table04, rounder4 );
426 RowMid( p_block, 4*8, 1*8, table17 );
427 Row( table17, rounder1 );
428 RowMid( p_block, 1*8, 7*8, table17 );
429 Row( table17, rounder7 );
430 RowMid( p_block, 7*8, 2*8, table26 );
431 Row( table26, rounder2 );
432 RowMid( p_block, 2*8, 6*8, table26 );
433 Row( table26, rounder6 );
434 RowMid( p_block, 6*8, 3*8, table35 );
435 Row( table35, rounder3 );
436 RowMid( p_block, 3*8, 5*8, table35 );
437 Row( table35, rounder5 );
438 RowTail( p_block, 5*8);
444 #include "idct_sparse.h"
445 #include "idct_decl.h"