[cvs] / xvidcore / src / xvid.c Repository:
ViewVC logotype

Diff of /xvidcore/src/xvid.c

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 1.50, Mon Apr 5 20:36:36 2004 UTC revision 1.84, Sun Oct 17 17:46:43 2010 UTC
# Line 4  Line 4 
4   *  - Native API implementation  -   *  - Native API implementation  -
5   *   *
6   *  Copyright(C) 2001-2004 Peter Ross <pross@xvid.org>   *  Copyright(C) 2001-2004 Peter Ross <pross@xvid.org>
7     *               2002-2010 Michael Militzer <isibaar@xvid.org>
8   *   *
9   *  This program is free software ; you can redistribute it and/or modify   *  This program is free software ; you can redistribute it and/or modify
10   *  it under the terms of the GNU General Public License as published by   *  it under the terms of the GNU General Public License as published by
# Line 28  Line 29 
29  #include <string.h>  #include <string.h>
30  #include <time.h>  #include <time.h>
31    
32    #if !defined(_WIN32)
33    #include <unistd.h>
34    #endif
35    
36    #if defined(__APPLE__) && defined(__MACH__) && !defined(_SC_NPROCESSORS_CONF)
37    #include <sys/types.h>
38    #include <sys/sysctl.h>
39    #ifdef MAX
40    #undef MAX
41    #endif
42    #ifdef MIN
43    #undef MIN
44    #endif
45    #endif
46    
47    #if defined(__amigaos4__)
48    #include <exec/exec.h>
49    #include <proto/exec.h>
50    #endif
51    
52  #include "xvid.h"  #include "xvid.h"
53  #include "decoder.h"  #include "decoder.h"
54  #include "encoder.h"  #include "encoder.h"
# Line 36  Line 57 
57  #include "dct/fdct.h"  #include "dct/fdct.h"
58  #include "image/colorspace.h"  #include "image/colorspace.h"
59  #include "image/interpolate8x8.h"  #include "image/interpolate8x8.h"
 #include "image/reduced.h"  
60  #include "utils/mem_transfer.h"  #include "utils/mem_transfer.h"
61  #include "utils/mbfunctions.h"  #include "utils/mbfunctions.h"
62  #include "quant/quant.h"  #include "quant/quant.h"
63  #include "motion/motion.h"  #include "motion/motion.h"
64    #include "motion/gmc.h"
65  #include "motion/sad.h"  #include "motion/sad.h"
66  #include "utils/emms.h"  #include "utils/emms.h"
67  #include "utils/timer.h"  #include "utils/timer.h"
# Line 52  Line 73 
73  unsigned int xvid_debug = 0; /* xvid debug mask */  unsigned int xvid_debug = 0; /* xvid debug mask */
74  #endif  #endif
75    
76  #if defined(ARCH_IS_IA32) && defined(_MSC_VER)  #if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && defined(_MSC_VER)
77  #       include <windows.h>  #       include <windows.h>
78  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_PPC)  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) || (defined(ARCH_IS_PPC) && !defined(__amigaos4__))
79  #       include <signal.h>  #       include <signal.h>
80  #       include <setjmp.h>  #       include <setjmp.h>
81    
# Line 67  Line 88 
88          }          }
89  #endif  #endif
90    
   
91  /*  /*
92   * Calls the funcptr, and returns whether SIGILL (illegal instruction) was   * Calls the funcptr, and returns whether SIGILL (illegal instruction) was
93   * signalled   * signalled
# Line 77  Line 97 
97   *   0 : SIGILL was *not* signalled   *   0 : SIGILL was *not* signalled
98   *   1 : SIGILL was signalled   *   1 : SIGILL was signalled
99   */   */
100  #if defined(ARCH_IS_IA32) && defined(_MSC_VER)  #if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && defined(_MSC_VER)
101  static int  static int
102  sigill_check(void (*func)())  sigill_check(void (*func)())
103  {  {
# Line 90  Line 110 
110          }          }
111          return(0);          return(0);
112  }  }
113  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_PPC)  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) || (defined(ARCH_IS_PPC) && !defined(__amigaos4__))
114  static int  static int
115  sigill_check(void (*func)())  sigill_check(void (*func)())
116  {  {
# Line 127  Line 147 
147    
148  /* detect cpu flags  */  /* detect cpu flags  */
149  static unsigned int  static unsigned int
150  detect_cpu_flags()  detect_cpu_flags(void)
151  {  {
152          /* enable native assembly optimizations by default */          /* enable native assembly optimizations by default */
153          unsigned int cpu_flags = XVID_CPU_ASM;          unsigned int cpu_flags = XVID_CPU_ASM;
154    
155  #if defined(ARCH_IS_IA32)  #if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
156          cpu_flags |= check_cpu_features();          cpu_flags |= check_cpu_features();
157          if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger))          if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger))
158                  cpu_flags &= ~XVID_CPU_SSE;                  cpu_flags &= ~XVID_CPU_SSE;
159    
160          if ((cpu_flags & XVID_CPU_SSE2) && sigill_check(sse2_os_trigger))          if ((cpu_flags & (XVID_CPU_SSE2|XVID_CPU_SSE3|XVID_CPU_SSE41)) && sigill_check(sse2_os_trigger))
161                  cpu_flags &= ~XVID_CPU_SSE2;                  cpu_flags &= ~(XVID_CPU_SSE2|XVID_CPU_SSE3|XVID_CPU_SSE41);
162  #endif  #endif
163    
164  #if defined(ARCH_IS_PPC)  #if defined(ARCH_IS_PPC)
165    #if defined(__amigaos4__)
166            {
167                    uint32_t vector_unit = VECTORTYPE_NONE;
168                    IExec->GetCPUInfoTags(GCIT_VectorUnit, &vector_unit, TAG_END);
169                    if (vector_unit == VECTORTYPE_ALTIVEC) {
170                            cpu_flags |= XVID_CPU_ALTIVEC;
171                    }
172            }
173    #else
174          if (!sigill_check(altivec_trigger))          if (!sigill_check(altivec_trigger))
175                  cpu_flags |= XVID_CPU_ALTIVEC;                  cpu_flags |= XVID_CPU_ALTIVEC;
176  #endif  #endif
177    #endif
178    
179          return cpu_flags;          return cpu_flags;
180  }  }
181    
182    
183  /*****************************************************************************  /*****************************************************************************
184   * XviD Init Entry point   * Xvid Init Entry point
185   *   *
186   * Well this function initialize all internal function pointers according   * Well this function initialize all internal function pointers according
187   * to the CPU features forced by the library client or autodetected (depending   * to the CPU features forced by the library client or autodetected (depending
# Line 184  Line 214 
214          idct = idct_int32;          idct = idct_int32;
215    
216          /* Only needed on PPC Altivec archs */          /* Only needed on PPC Altivec archs */
217          sadInit = 0;          sadInit = NULL;
218    
219          /* Restore FPU context : emms_c is a nop functions */          /* Restore FPU context : emms_c is a nop functions */
220          emms = emms_c;          emms = emms_c;
# Line 211  Line 241 
241          transfer_8to16sub  = transfer_8to16sub_c;          transfer_8to16sub  = transfer_8to16sub_c;
242          transfer_8to16subro  = transfer_8to16subro_c;          transfer_8to16subro  = transfer_8to16subro_c;
243          transfer_8to16sub2 = transfer_8to16sub2_c;          transfer_8to16sub2 = transfer_8to16sub2_c;
244            transfer_8to16sub2ro = transfer_8to16sub2ro_c;
245          transfer_16to8add  = transfer_16to8add_c;          transfer_16to8add  = transfer_16to8add_c;
246          transfer8x8_copy   = transfer8x8_copy_c;          transfer8x8_copy   = transfer8x8_copy_c;
247            transfer8x4_copy   = transfer8x4_copy_c;
248    
249          /* Interlacing functions */          /* Interlacing functions */
250          MBFieldTest = MBFieldTest_c;          MBFieldTest = MBFieldTest_c;
# Line 222  Line 254 
254          interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_c;          interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_c;
255          interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_c;          interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_c;
256    
257            interpolate8x4_halfpel_h  = interpolate8x4_halfpel_h_c;
258            interpolate8x4_halfpel_v  = interpolate8x4_halfpel_v_c;
259            interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_c;
260    
261            interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_c;
262            interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_c;
263            interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_c;
264            interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_c;
265    
266          interpolate16x16_lowpass_h = interpolate16x16_lowpass_h_c;          interpolate16x16_lowpass_h = interpolate16x16_lowpass_h_c;
267          interpolate16x16_lowpass_v = interpolate16x16_lowpass_v_c;          interpolate16x16_lowpass_v = interpolate16x16_lowpass_v_c;
268          interpolate16x16_lowpass_hv = interpolate16x16_lowpass_hv_c;          interpolate16x16_lowpass_hv = interpolate16x16_lowpass_hv_c;
# Line 239  Line 280 
280          /* postprocessing */          /* postprocessing */
281          image_brightness = image_brightness_c;          image_brightness = image_brightness_c;
282    
         /* reduced resolution */  
         copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_C;  
         add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_C;  
         vfilter_31 = xvid_VFilter_31_C;  
         hfilter_31 = xvid_HFilter_31_C;  
         filter_18x18_to_8x8 = xvid_Filter_18x18_To_8x8_C;  
         filter_diff_18x18_to_8x8 = xvid_Filter_Diff_18x18_To_8x8_C;  
   
283          /* Initialize internal colorspace transformation tables */          /* Initialize internal colorspace transformation tables */
284          colorspace_init();          colorspace_init();
285    
# Line 254  Line 287 
287          yv12_to_yv12    = yv12_to_yv12_c;          yv12_to_yv12    = yv12_to_yv12_c;
288          rgb555_to_yv12  = rgb555_to_yv12_c;          rgb555_to_yv12  = rgb555_to_yv12_c;
289          rgb565_to_yv12  = rgb565_to_yv12_c;          rgb565_to_yv12  = rgb565_to_yv12_c;
290            rgb_to_yv12     = rgb_to_yv12_c;
291          bgr_to_yv12     = bgr_to_yv12_c;          bgr_to_yv12     = bgr_to_yv12_c;
292          bgra_to_yv12    = bgra_to_yv12_c;          bgra_to_yv12    = bgra_to_yv12_c;
293          abgr_to_yv12    = abgr_to_yv12_c;          abgr_to_yv12    = abgr_to_yv12_c;
# Line 275  Line 309 
309          /* All colorspace transformation functions YV12->User format */          /* All colorspace transformation functions YV12->User format */
310          yv12_to_rgb555  = yv12_to_rgb555_c;          yv12_to_rgb555  = yv12_to_rgb555_c;
311          yv12_to_rgb565  = yv12_to_rgb565_c;          yv12_to_rgb565  = yv12_to_rgb565_c;
312            yv12_to_rgb     = yv12_to_rgb_c;
313          yv12_to_bgr     = yv12_to_bgr_c;          yv12_to_bgr     = yv12_to_bgr_c;
314          yv12_to_bgra    = yv12_to_bgra_c;          yv12_to_bgra    = yv12_to_bgra_c;
315          yv12_to_abgr    = yv12_to_abgr_c;          yv12_to_abgr    = yv12_to_abgr_c;
# Line 302  Line 337 
337          dev16      = dev16_c;          dev16      = dev16_c;
338          sad16v     = sad16v_c;          sad16v     = sad16v_c;
339          sse8_16bit = sse8_16bit_c;          sse8_16bit = sse8_16bit_c;
340            sse8_8bit  = sse8_8bit_c;
341    
342  #if defined(ARCH_IS_IA32)          init_GMC(cpu_flags);
343    
344          if ((cpu_flags & XVID_CPU_ASM)) {  #if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
                 vfilter_31 = xvid_VFilter_31_x86;  
                 hfilter_31 = xvid_HFilter_31_x86;  
         }  
345    
346          if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) ||          if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) ||
347                  (cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) ||                  (cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) ||
348                  (cpu_flags & XVID_CPU_SSE) || (cpu_flags & XVID_CPU_SSE2))                  (cpu_flags & XVID_CPU_SSE) || (cpu_flags & XVID_CPU_SSE2) ||
349            (cpu_flags & XVID_CPU_SSE3) || (cpu_flags & XVID_CPU_SSE41))
350          {          {
351                  /* Restore FPU context : emms_c is a nop functions */                  /* Restore FPU context : emms_c is a nop functions */
352                  emms = emms_mmx;                  emms = emms_mmx;
# Line 333  Line 367 
367                  quant_h263_inter   = quant_h263_inter_mmx;                  quant_h263_inter   = quant_h263_inter_mmx;
368                  dequant_h263_intra = dequant_h263_intra_mmx;                  dequant_h263_intra = dequant_h263_intra_mmx;
369                  dequant_h263_inter = dequant_h263_inter_mmx;                  dequant_h263_inter = dequant_h263_inter_mmx;
   
370                  quant_mpeg_intra   = quant_mpeg_intra_mmx;                  quant_mpeg_intra   = quant_mpeg_intra_mmx;
371                  quant_mpeg_inter   = quant_mpeg_inter_mmx;                  quant_mpeg_inter   = quant_mpeg_inter_mmx;
372                  dequant_mpeg_intra = dequant_mpeg_intra_mmx;                  dequant_mpeg_intra = dequant_mpeg_intra_mmx;
373                  dequant_mpeg_inter = dequant_mpeg_inter_mmx;                  dequant_mpeg_inter = dequant_mpeg_inter_mmx;
374    
375    
376                  /* Block related functions */                  /* Block related functions */
377                  transfer_8to16copy = transfer_8to16copy_mmx;                  transfer_8to16copy = transfer_8to16copy_mmx;
378                  transfer_16to8copy = transfer_16to8copy_mmx;                  transfer_16to8copy = transfer_16to8copy_mmx;
# Line 347  Line 381 
381                  transfer_8to16sub2 = transfer_8to16sub2_mmx;                  transfer_8to16sub2 = transfer_8to16sub2_mmx;
382                  transfer_16to8add  = transfer_16to8add_mmx;                  transfer_16to8add  = transfer_16to8add_mmx;
383                  transfer8x8_copy   = transfer8x8_copy_mmx;                  transfer8x8_copy   = transfer8x8_copy_mmx;
384                    transfer8x4_copy   = transfer8x4_copy_mmx;
385    
386                  /* Interlacing Functions */                  /* Interlacing Functions */
387                  MBFieldTest = MBFieldTest_mmx;                  MBFieldTest = MBFieldTest_mmx;
# Line 356  Line 391 
391                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_mmx;                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_mmx;
392                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx;
393    
394                    interpolate8x4_halfpel_h  = interpolate8x4_halfpel_h_mmx;
395                    interpolate8x4_halfpel_v  = interpolate8x4_halfpel_v_mmx;
396                    interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_mmx;
397    
398                    interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_mmx;
399                    interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_mmx;
400                    interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_mmx;
401                    interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_mmx;
402    
403                  interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_mmx;                  interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_mmx;
404                  interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_mmx;                  interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_mmx;
405    
# Line 365  Line 409 
409                  /* postprocessing */                  /* postprocessing */
410                  image_brightness = image_brightness_mmx;                  image_brightness = image_brightness_mmx;
411    
                 /* reduced resolution */  
                 copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_mmx;  
                 add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_mmx;  
                 hfilter_31 = xvid_HFilter_31_mmx;  
                 filter_18x18_to_8x8 = xvid_Filter_18x18_To_8x8_mmx;  
                 filter_diff_18x18_to_8x8 = xvid_Filter_Diff_18x18_To_8x8_mmx;  
   
412                  /* image input xxx_to_yv12 related functions */                  /* image input xxx_to_yv12 related functions */
413    
414                  yv12_to_yv12  = yv12_to_yv12_mmx;                  yv12_to_yv12  = yv12_to_yv12_mmx;
415    
416                  bgr_to_yv12   = bgr_to_yv12_mmx;                  bgr_to_yv12   = bgr_to_yv12_mmx;
417                    rgb_to_yv12   = rgb_to_yv12_mmx;
418                  bgra_to_yv12  = bgra_to_yv12_mmx;                  bgra_to_yv12  = bgra_to_yv12_mmx;
419                    rgba_to_yv12  = rgba_to_yv12_mmx;
420                  yuyv_to_yv12  = yuyv_to_yv12_mmx;                  yuyv_to_yv12  = yuyv_to_yv12_mmx;
421                  uyvy_to_yv12  = uyvy_to_yv12_mmx;                  uyvy_to_yv12  = uyvy_to_yv12_mmx;
422    
# Line 397  Line 438 
438                  dev16    = dev16_mmx;                  dev16    = dev16_mmx;
439                  sad16v   = sad16v_mmx;                  sad16v   = sad16v_mmx;
440                  sse8_16bit = sse8_16bit_mmx;                  sse8_16bit = sse8_16bit_mmx;
441                    sse8_8bit  = sse8_8bit_mmx;
442          }          }
443    
444          /* these 3dnow functions are faster than mmx, but slower than xmm. */          /* these 3dnow functions are faster than mmx, but slower than xmm. */
# Line 410  Line 452 
452    
453                  yuyv_to_yv12  = yuyv_to_yv12_3dn;                  yuyv_to_yv12  = yuyv_to_yv12_3dn;
454                  uyvy_to_yv12  = uyvy_to_yv12_3dn;                  uyvy_to_yv12  = uyvy_to_yv12_3dn;
455    
456          }          }
457    
458    
# Line 424  Line 467 
467                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_xmm;                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_xmm;
468                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm;
469    
470                  /* reduced resolution */                  interpolate8x4_halfpel_h  = interpolate8x4_halfpel_h_xmm;
471                  copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_xmm;                  interpolate8x4_halfpel_v  = interpolate8x4_halfpel_v_xmm;
472                  add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm;                  interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_xmm;
473    
474                    interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_xmm;
475                    interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_xmm;
476                    interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_xmm;
477                    interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_xmm;
478    
479                  /* Quantization */                  /* Quantization */
                 quant_mpeg_intra = quant_mpeg_intra_xmm;  
480                  quant_mpeg_inter = quant_mpeg_inter_xmm;                  quant_mpeg_inter = quant_mpeg_inter_xmm;
481    
482                  dequant_h263_intra = dequant_h263_intra_xmm;                  dequant_h263_intra = dequant_h263_intra_xmm;
# Line 437  Line 484 
484    
485                  /* Buffer transfer */                  /* Buffer transfer */
486                  transfer_8to16sub2 = transfer_8to16sub2_xmm;                  transfer_8to16sub2 = transfer_8to16sub2_xmm;
487                    transfer_8to16sub2ro = transfer_8to16sub2ro_xmm;
488    
489                  /* Colorspace transformation */                  /* Colorspace transformation */
490                  yv12_to_yv12  = yv12_to_yv12_xmm;                  /* yv12_to_yv12  = yv12_to_yv12_xmm; */ /* appears to be slow on many machines */
491                  yuyv_to_yv12  = yuyv_to_yv12_xmm;                  yuyv_to_yv12  = yuyv_to_yv12_xmm;
492                  uyvy_to_yv12  = uyvy_to_yv12_xmm;                  uyvy_to_yv12  = uyvy_to_yv12_xmm;
493    
# Line 458  Line 506 
506                  interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn;                  interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn;
507                  interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dn;                  interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dn;
508                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn;
509    
510                    interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_3dn;
511                    interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_3dn;
512                    interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_3dn;
513          }          }
514    
515          if ((cpu_flags & XVID_CPU_3DNOWEXT)) {          if ((cpu_flags & XVID_CPU_3DNOWEXT)) {
516    
                 /* Inverse DCT */  
                 idct =  idct_3dne;  
   
517                  /* Buffer transfer */                  /* Buffer transfer */
518                  transfer_8to16copy =  transfer_8to16copy_3dne;                  transfer_8to16copy =  transfer_8to16copy_3dne;
519                  transfer_16to8copy = transfer_16to8copy_3dne;                  transfer_16to8copy = transfer_16to8copy_3dne;
520                  transfer_8to16sub =  transfer_8to16sub_3dne;                  transfer_8to16sub =  transfer_8to16sub_3dne;
521                  transfer_8to16subro =  transfer_8to16subro_3dne;                  transfer_8to16subro =  transfer_8to16subro_3dne;
                 transfer_8to16sub2 =  transfer_8to16sub2_3dne;  
522                  transfer_16to8add = transfer_16to8add_3dne;                  transfer_16to8add = transfer_16to8add_3dne;
523                  transfer8x8_copy = transfer8x8_copy_3dne;                  transfer8x8_copy = transfer8x8_copy_3dne;
524                    transfer8x4_copy = transfer8x4_copy_3dne;
525    
526                    if ((cpu_flags & XVID_CPU_MMXEXT)) {
527                            /* Inverse DCT */
528                            idct =  idct_3dne;
529    
530                            /* Buffer transfer */
531                            transfer_8to16sub2 =  transfer_8to16sub2_3dne;
532    
533                            /* Interpolation */
534                            interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dne;
535                            interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne;
536                            interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;
537    
538                            interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_3dne;
539                            interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_3dne;
540                            interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_3dne;
541    
542                  /* Quantization */                  /* Quantization */
543                  quant_h263_intra = quant_h263_intra_3dne;                          quant_h263_intra = quant_h263_intra_3dne;               /* cmov only */
544                  quant_h263_inter = quant_h263_inter_3dne;                  quant_h263_inter = quant_h263_inter_3dne;
545                  dequant_mpeg_intra = dequant_mpeg_intra_3dne;                          dequant_mpeg_intra = dequant_mpeg_intra_3dne;   /* cmov only */
546                  dequant_mpeg_inter = dequant_mpeg_inter_3dne;                  dequant_mpeg_inter = dequant_mpeg_inter_3dne;
547                  dequant_h263_intra = dequant_h263_intra_3dne;                  dequant_h263_intra = dequant_h263_intra_3dne;
548                  dequant_h263_inter = dequant_h263_inter_3dne;                  dequant_h263_inter = dequant_h263_inter_3dne;
549    
550                  /* ME functions */                  /* ME functions */
                 calc_cbp = calc_cbp_3dne;  
551                  sad16 = sad16_3dne;                  sad16 = sad16_3dne;
552                  sad8 = sad8_3dne;                  sad8 = sad8_3dne;
553                  sad16bi = sad16bi_3dne;                  sad16bi = sad16bi_3dne;
554                  sad8bi = sad8bi_3dne;                  sad8bi = sad8bi_3dne;
555                  dev16 = dev16_3dne;                  dev16 = dev16_3dne;
556                    }
                 /* Interpolation */  
                 interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dne;  
                 interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne;  
                 interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;  
557          }          }
558    
559          if ((cpu_flags & XVID_CPU_SSE2)) {          if ((cpu_flags & XVID_CPU_SSE2)) {
# Line 510  Line 570 
570                  sad16    = sad16_sse2;                  sad16    = sad16_sse2;
571                  dev16    = dev16_sse2;                  dev16    = dev16_sse2;
572    
573                  /* DCT operators                  /* DCT operators */
                  * no iDCT because it's not "Walken matching" */  
574                  fdct = fdct_sse2_skal;                  fdct = fdct_sse2_skal;
575                    idct = idct_sse2_skal;   /* Is now IEEE1180 and Walken compliant. */
576    
577                    /* postprocessing */
578                    image_brightness = image_brightness_sse2;
579    
580            }
581    
582            if ((cpu_flags & XVID_CPU_SSE3)) {
583    
584                    /* SAD operators */
585                    sad16    = sad16_sse3;
586                    dev16    = dev16_sse3;
587          }          }
588    
589  #endif /* ARCH_IS_IA32 */  #endif /* ARCH_IS_IA32 */
590    
591  #if defined(ARCH_IS_IA64)  #if defined(ARCH_IS_IA64)
# Line 572  Line 644 
644            interpolate8x8_avg2 = interpolate8x8_avg2_altivec_c;            interpolate8x8_avg2 = interpolate8x8_avg2_altivec_c;
645            interpolate8x8_avg4 = interpolate8x8_avg4_altivec_c;            interpolate8x8_avg4 = interpolate8x8_avg4_altivec_c;
646    
647            interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_altivec_c;                    interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_altivec_c;
648                      interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_altivec_c;
649                      interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_altivec_c;
650                      interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_altivec_c;
651    
652            /* Colorspace conversion */            /* Colorspace conversion */
653            bgra_to_yv12 = bgra_to_yv12_altivec_c;            bgra_to_yv12 = bgra_to_yv12_altivec_c;
# Line 591  Line 666 
666            quant_h263_inter = quant_h263_inter_altivec_c;            quant_h263_inter = quant_h263_inter_altivec_c;
667            dequant_h263_intra = dequant_h263_intra_altivec_c;            dequant_h263_intra = dequant_h263_intra_altivec_c;
668            dequant_h263_inter = dequant_h263_inter_altivec_c;            dequant_h263_inter = dequant_h263_inter_altivec_c;
669    
670                      dequant_mpeg_intra = dequant_mpeg_intra_altivec_c;
671                      dequant_mpeg_inter = dequant_mpeg_inter_altivec_c;
672    
673                      /* Qpel stuff */
674                      xvid_QP_Funcs = &xvid_QP_Funcs_Altivec_C;
675                      xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_Altivec_C;
676          }          }
677  #endif  #endif
678    
# Line 609  Line 691 
691                  return XVID_ERR_VERSION;                  return XVID_ERR_VERSION;
692    
693          info->actual_version = XVID_VERSION;          info->actual_version = XVID_VERSION;
694          info->build = "xvid-1.0.0";          info->build = "xvid-1.3.0-dev";
695          info->cpu_flags = detect_cpu_flags();          info->cpu_flags = detect_cpu_flags();
696            info->num_threads = 0; /* single-thread */
697    
698    #if defined(_WIN32)
699    
700      {
701            SYSTEM_INFO siSysInfo;
702            GetSystemInfo(&siSysInfo);
703            info->num_threads = siSysInfo.dwNumberOfProcessors; /* number of _logical_ cores */
704      }
705    
706    #elif defined(_SC_NPROCESSORS_CONF) /* should be available on Apple too actually */
707    
708      info->num_threads = sysconf(_SC_NPROCESSORS_CONF);
709    
710    #elif defined(__APPLE__) && defined(__MACH__)
711    
712      {
713        size_t len;
714        int    mib[2], ncpu;
715    
716        mib[0] = CTL_HW;
717        mib[1] = HW_NCPU;
718        len    = sizeof(ncpu);
719        if (sysctl(mib, 2, &ncpu, &len, NULL, 0) == 0)
720          info -> num_threads = ncpu;
721        else
722          info -> num_threads = 1;
723      }
724    
725    #elif defined(__amigaos4__)
726    
727      {
728        uint32_t num_threads = 1;
729        IExec->GetCPUInfoTags(GCIT_NumberOfCPUs, &num_threads, TAG_END);
730        info->num_threads = num_threads;
731      }
732    
 #if defined(_SMP) && defined(WIN32)  
         info->num_threads = pthread_num_processors_np();;  
 #else  
         info->num_threads = 0;  
733  #endif  #endif
734    
735          return 0;          return 0;
# Line 653  Line 767 
767                                                  convert->output.csp, convert->interlacing);                                                  convert->output.csp, convert->interlacing);
768                          break;                          break;
769    
770                    case XVID_CSP_INTERNAL :
771                            img.y = (uint8_t*)convert->input.plane[0];
772                            img.u = (uint8_t*)convert->input.plane[1];
773                            img.v = (uint8_t*)convert->input.plane[2];
774                            image_output(&img, width, height, convert->input.stride[0],
775                                                    (uint8_t**)convert->output.plane, convert->output.stride,
776                                                    convert->output.csp, convert->interlacing);
777                            break;
778    
779                  default :                  default :
780                          return XVID_ERR_FORMAT;                          return XVID_ERR_FORMAT;
781          }          }
# Line 663  Line 786 
786  }  }
787    
788  /*****************************************************************************  /*****************************************************************************
789   * XviD Global Entry point   * Xvid Global Entry point
790   *   *
791   * Well this function initialize all internal function pointers according   * Well this function initialize all internal function pointers according
792   * to the CPU features forced by the library client or autodetected (depending   * to the CPU features forced by the library client or autodetected (depending
# Line 696  Line 819 
819  }  }
820    
821  /*****************************************************************************  /*****************************************************************************
822   * XviD Native decoder entry point   * Xvid Native decoder entry point
823   *   *
824   * This function is just a wrapper to all the option cases.   * This function is just a wrapper to all the option cases.
825   *   *
# Line 728  Line 851 
851    
852    
853  /*****************************************************************************  /*****************************************************************************
854   * XviD Native encoder entry point   * Xvid Native encoder entry point
855   *   *
856   * This function is just a wrapper to all the option cases.   * This function is just a wrapper to all the option cases.
857   *   *

Legend:
Removed from v.1.50  
changed lines
  Added in v.1.84

No admin address has been configured
ViewVC Help
Powered by ViewVC 1.0.4