[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 1.54, Thu Jul 15 10:08:21 2004 UTC revision 1.70, Mon Jul 10 08:09:59 2006 UTC
# Line 36  Line 36 
36  #include "dct/fdct.h"  #include "dct/fdct.h"
37  #include "image/colorspace.h"  #include "image/colorspace.h"
38  #include "image/interpolate8x8.h"  #include "image/interpolate8x8.h"
 #include "image/reduced.h"  
39  #include "utils/mem_transfer.h"  #include "utils/mem_transfer.h"
40  #include "utils/mbfunctions.h"  #include "utils/mbfunctions.h"
41  #include "quant/quant.h"  #include "quant/quant.h"
42  #include "motion/motion.h"  #include "motion/motion.h"
43    #include "motion/gmc.h"
44  #include "motion/sad.h"  #include "motion/sad.h"
45  #include "utils/emms.h"  #include "utils/emms.h"
46  #include "utils/timer.h"  #include "utils/timer.h"
# Line 52  Line 52 
52  unsigned int xvid_debug = 0; /* xvid debug mask */  unsigned int xvid_debug = 0; /* xvid debug mask */
53  #endif  #endif
54    
55  #if defined(ARCH_IS_IA32) && defined(_MSC_VER)  #if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && defined(_MSC_VER)
56  #       include <windows.h>  #       include <windows.h>
57  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_PPC)  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) || defined(ARCH_IS_PPC)
58  #       include <signal.h>  #       include <signal.h>
59  #       include <setjmp.h>  #       include <setjmp.h>
60    
# Line 77  Line 77 
77   *   0 : SIGILL was *not* signalled   *   0 : SIGILL was *not* signalled
78   *   1 : SIGILL was signalled   *   1 : SIGILL was signalled
79   */   */
80  #if defined(ARCH_IS_IA32) && defined(_MSC_VER)  #if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && defined(_MSC_VER)
81  static int  static int
82  sigill_check(void (*func)())  sigill_check(void (*func)())
83  {  {
# Line 90  Line 90 
90          }          }
91          return(0);          return(0);
92  }  }
93  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_PPC)  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) || defined(ARCH_IS_PPC)
94  static int  static int
95  sigill_check(void (*func)())  sigill_check(void (*func)())
96  {  {
# Line 127  Line 127 
127    
128  /* detect cpu flags  */  /* detect cpu flags  */
129  static unsigned int  static unsigned int
130  detect_cpu_flags()  detect_cpu_flags(void)
131  {  {
132          /* enable native assembly optimizations by default */          /* enable native assembly optimizations by default */
133          unsigned int cpu_flags = XVID_CPU_ASM;          unsigned int cpu_flags = XVID_CPU_ASM;
134    
135  #if defined(ARCH_IS_IA32)  #if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
136          cpu_flags |= check_cpu_features();          cpu_flags |= check_cpu_features();
137          if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger))          if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger))
138                  cpu_flags &= ~XVID_CPU_SSE;                  cpu_flags &= ~XVID_CPU_SSE;
# Line 184  Line 184 
184          idct = idct_int32;          idct = idct_int32;
185    
186          /* Only needed on PPC Altivec archs */          /* Only needed on PPC Altivec archs */
187          sadInit = 0;          sadInit = NULL;
188    
189          /* Restore FPU context : emms_c is a nop functions */          /* Restore FPU context : emms_c is a nop functions */
190          emms = emms_c;          emms = emms_c;
# Line 211  Line 211 
211          transfer_8to16sub  = transfer_8to16sub_c;          transfer_8to16sub  = transfer_8to16sub_c;
212          transfer_8to16subro  = transfer_8to16subro_c;          transfer_8to16subro  = transfer_8to16subro_c;
213          transfer_8to16sub2 = transfer_8to16sub2_c;          transfer_8to16sub2 = transfer_8to16sub2_c;
214            transfer_8to16sub2ro = transfer_8to16sub2ro_c;
215          transfer_16to8add  = transfer_16to8add_c;          transfer_16to8add  = transfer_16to8add_c;
216          transfer8x8_copy   = transfer8x8_copy_c;          transfer8x8_copy   = transfer8x8_copy_c;
217            transfer8x4_copy   = transfer8x4_copy_c;
218    
219          /* Interlacing functions */          /* Interlacing functions */
220          MBFieldTest = MBFieldTest_c;          MBFieldTest = MBFieldTest_c;
# Line 222  Line 224 
224          interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_c;          interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_c;
225          interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_c;          interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_c;
226    
227            interpolate8x4_halfpel_h  = interpolate8x4_halfpel_h_c;
228            interpolate8x4_halfpel_v  = interpolate8x4_halfpel_v_c;
229            interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_c;
230    
231            interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_c;
232            interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_c;
233            interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_c;
234            interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_c;
235    
236          interpolate16x16_lowpass_h = interpolate16x16_lowpass_h_c;          interpolate16x16_lowpass_h = interpolate16x16_lowpass_h_c;
237          interpolate16x16_lowpass_v = interpolate16x16_lowpass_v_c;          interpolate16x16_lowpass_v = interpolate16x16_lowpass_v_c;
238          interpolate16x16_lowpass_hv = interpolate16x16_lowpass_hv_c;          interpolate16x16_lowpass_hv = interpolate16x16_lowpass_hv_c;
# Line 239  Line 250 
250          /* postprocessing */          /* postprocessing */
251          image_brightness = image_brightness_c;          image_brightness = image_brightness_c;
252    
         /* 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;  
   
253          /* Initialize internal colorspace transformation tables */          /* Initialize internal colorspace transformation tables */
254          colorspace_init();          colorspace_init();
255    
# Line 304  Line 307 
307          sse8_16bit = sse8_16bit_c;          sse8_16bit = sse8_16bit_c;
308          sse8_8bit  = sse8_8bit_c;          sse8_8bit  = sse8_8bit_c;
309    
310  #if defined(ARCH_IS_IA32)          init_GMC(cpu_flags);
311    
312          if ((cpu_flags & XVID_CPU_ASM)) {  #if defined(ARCH_IS_IA32)
                 vfilter_31 = xvid_VFilter_31_x86;  
                 hfilter_31 = xvid_HFilter_31_x86;  
         }  
313    
314          if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) ||          if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) ||
315                  (cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) ||                  (cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) ||
# Line 348  Line 348 
348                  transfer_8to16sub2 = transfer_8to16sub2_mmx;                  transfer_8to16sub2 = transfer_8to16sub2_mmx;
349                  transfer_16to8add  = transfer_16to8add_mmx;                  transfer_16to8add  = transfer_16to8add_mmx;
350                  transfer8x8_copy   = transfer8x8_copy_mmx;                  transfer8x8_copy   = transfer8x8_copy_mmx;
351                    transfer8x4_copy   = transfer8x4_copy_mmx;
352    
353                  /* Interlacing Functions */                  /* Interlacing Functions */
354                  MBFieldTest = MBFieldTest_mmx;                  MBFieldTest = MBFieldTest_mmx;
# Line 357  Line 358 
358                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_mmx;                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_mmx;
359                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx;
360    
361                    interpolate8x4_halfpel_h  = interpolate8x4_halfpel_h_mmx;
362                    interpolate8x4_halfpel_v  = interpolate8x4_halfpel_v_mmx;
363                    interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_mmx;
364    
365                    interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_mmx;
366                    interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_mmx;
367                    interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_mmx;
368                    interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_mmx;
369    
370                  interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_mmx;                  interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_mmx;
371                  interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_mmx;                  interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_mmx;
372    
# Line 366  Line 376 
376                  /* postprocessing */                  /* postprocessing */
377                  image_brightness = image_brightness_mmx;                  image_brightness = image_brightness_mmx;
378    
                 /* 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;  
   
379                  /* image input xxx_to_yv12 related functions */                  /* image input xxx_to_yv12 related functions */
380                  yv12_to_yv12  = yv12_to_yv12_mmx;                  yv12_to_yv12  = yv12_to_yv12_mmx;
381                  bgr_to_yv12   = bgr_to_yv12_mmx;                  bgr_to_yv12   = bgr_to_yv12_mmx;
# Line 426  Line 429 
429                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_xmm;                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_xmm;
430                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm;
431    
432                  /* reduced resolution */                  interpolate8x4_halfpel_h  = interpolate8x4_halfpel_h_xmm;
433                  copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_xmm;                  interpolate8x4_halfpel_v  = interpolate8x4_halfpel_v_xmm;
434                  add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm;                  interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_xmm;
435    
436                    interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_xmm;
437                    interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_xmm;
438                    interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_xmm;
439                    interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_xmm;
440    
441                  /* Quantization */                  /* Quantization */
                 quant_mpeg_intra = quant_mpeg_intra_xmm;  
442                  quant_mpeg_inter = quant_mpeg_inter_xmm;                  quant_mpeg_inter = quant_mpeg_inter_xmm;
443    
444                  dequant_h263_intra = dequant_h263_intra_xmm;                  dequant_h263_intra = dequant_h263_intra_xmm;
# Line 439  Line 446 
446    
447                  /* Buffer transfer */                  /* Buffer transfer */
448                  transfer_8to16sub2 = transfer_8to16sub2_xmm;                  transfer_8to16sub2 = transfer_8to16sub2_xmm;
449                    transfer_8to16sub2ro = transfer_8to16sub2ro_xmm;
450    
451                  /* Colorspace transformation */                  /* Colorspace transformation */
452                  yv12_to_yv12  = yv12_to_yv12_xmm;                  yv12_to_yv12  = yv12_to_yv12_xmm;
# Line 460  Line 468 
468                  interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn;                  interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn;
469                  interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dn;                  interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dn;
470                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn;
471    
472                    interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_3dn;
473                    interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_3dn;
474                    interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_3dn;
475          }          }
476    
477          if ((cpu_flags & XVID_CPU_3DNOWEXT)) {          if ((cpu_flags & XVID_CPU_3DNOWEXT)) {
# Line 471  Line 483 
483                  transfer_8to16subro =  transfer_8to16subro_3dne;                  transfer_8to16subro =  transfer_8to16subro_3dne;
484                  transfer_16to8add = transfer_16to8add_3dne;                  transfer_16to8add = transfer_16to8add_3dne;
485                  transfer8x8_copy = transfer8x8_copy_3dne;                  transfer8x8_copy = transfer8x8_copy_3dne;
486                    transfer8x4_copy = transfer8x4_copy_3dne;
487    
488                  if ((cpu_flags & XVID_CPU_MMXEXT)) {                  if ((cpu_flags & XVID_CPU_MMXEXT)) {
489                          /* Inverse DCT */                          /* Inverse DCT */
# Line 484  Line 497 
497                          interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne;                          interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne;
498                          interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;                          interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;
499    
500                            interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_3dne;
501                            interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_3dne;
502                            interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_3dne;
503    
504                          /* Quantization */                          /* Quantization */
505                          quant_h263_intra = quant_h263_intra_3dne;               /* cmov only */                          quant_h263_intra = quant_h263_intra_3dne;               /* cmov only */
506                          quant_h263_inter = quant_h263_inter_3dne;                          quant_h263_inter = quant_h263_inter_3dne;
# Line 517  Line 534 
534                  sad16    = sad16_sse2;                  sad16    = sad16_sse2;
535                  dev16    = dev16_sse2;                  dev16    = dev16_sse2;
536    
537                  /* DCT operators                  /* DCT operators */
                  * no iDCT because it's not "Walken matching" */  
538                  fdct = fdct_sse2_skal;                  fdct = fdct_sse2_skal;
539        /* idct = idct_sse2_skal; */   /* Is now IEEE1180 and Walken compliant. Disabled until fully tested. */
540    
541                  /* postprocessing */                  /* postprocessing */
542                  image_brightness = image_brightness_sse2;                  image_brightness = image_brightness_sse2;
# Line 582  Line 599 
599            interpolate8x8_avg2 = interpolate8x8_avg2_altivec_c;            interpolate8x8_avg2 = interpolate8x8_avg2_altivec_c;
600            interpolate8x8_avg4 = interpolate8x8_avg4_altivec_c;            interpolate8x8_avg4 = interpolate8x8_avg4_altivec_c;
601    
602            interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_altivec_c;                    interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_altivec_c;
603                      interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_altivec_c;
604                      interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_altivec_c;
605                      interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_altivec_c;
606    
607            /* Colorspace conversion */            /* Colorspace conversion */
608            bgra_to_yv12 = bgra_to_yv12_altivec_c;            bgra_to_yv12 = bgra_to_yv12_altivec_c;
# Line 601  Line 621 
621            quant_h263_inter = quant_h263_inter_altivec_c;            quant_h263_inter = quant_h263_inter_altivec_c;
622            dequant_h263_intra = dequant_h263_intra_altivec_c;            dequant_h263_intra = dequant_h263_intra_altivec_c;
623            dequant_h263_inter = dequant_h263_inter_altivec_c;            dequant_h263_inter = dequant_h263_inter_altivec_c;
624    
625                      dequant_mpeg_intra = dequant_mpeg_intra_altivec_c;
626                      dequant_mpeg_inter = dequant_mpeg_inter_altivec_c;
627    
628                      /* Qpel stuff */
629                      xvid_QP_Funcs = &xvid_QP_Funcs_Altivec_C;
630                      xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_Altivec_C;
631            }
632    #endif
633    
634    #if defined(ARCH_IS_X86_64)
635            /* For now, only XVID_CPU_ASM is looked for, so user can still
636             * disable asm usage the usual way. When Intel EMT64 cpus will
637             * be out, maybe we'll have to check more precisely what cpu
638             * features there really are. */
639            if (cpu_flags & XVID_CPU_ASM) {
640                    /* SIMD state flusher */
641                    emms = emms_mmx;
642    
643                    /* DCT operators */
644                    fdct = fdct_skal_x86_64;
645                    idct = idct_x86_64;
646    
647                    /* SAD operators */
648                    sad16      = sad16_x86_64;
649                    sad8       = sad8_x86_64;
650                    sad16bi    = sad16bi_x86_64;
651                    sad8bi     = sad8bi_x86_64;
652                    dev16      = dev16_x86_64;
653                    sad16v     = sad16v_x86_64;
654                    sse8_16bit = sse8_16bit_x86_64;
655                    sse8_8bit  = sse8_8bit_x86_64;
656    
657                    /* Interpolation operators */
658                    interpolate8x8_halfpel_h  = interpolate8x8_halfpel_h_x86_64;
659                    interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_x86_64;
660                    interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_x86_64;
661    
662                    interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_x86_64;
663                    interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_x86_64;
664                    interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_x86_64;
665                    interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_x86_64;
666    
667                    interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_x86_64;
668                    interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_x86_64;
669    
670                    interpolate8x8_avg2 = interpolate8x8_avg2_x86_64;
671                    interpolate8x8_avg4 = interpolate8x8_avg4_x86_64;
672    
673                    /* Quantization related functions */
674                    quant_h263_intra   = quant_h263_intra_x86_64;
675                    quant_h263_inter   = quant_h263_inter_x86_64;
676                    dequant_h263_intra = dequant_h263_intra_x86_64;
677                    dequant_h263_inter = dequant_h263_inter_x86_64;
678                    /*quant_mpeg_intra   = quant_mpeg_intra_x86_64; fix me! */
679                    quant_mpeg_inter   = quant_mpeg_inter_x86_64;
680                    dequant_mpeg_intra   = dequant_mpeg_intra_x86_64;
681                    dequant_mpeg_inter   = dequant_mpeg_inter_x86_64;
682    
683                    /* Block related functions */
684                    transfer_8to16copy  = transfer_8to16copy_x86_64;
685                    transfer_16to8copy  = transfer_16to8copy_x86_64;
686                    transfer_8to16sub   = transfer_8to16sub_x86_64;
687                    transfer_8to16subro = transfer_8to16subro_x86_64;
688                    transfer_8to16sub2  = transfer_8to16sub2_x86_64;
689                    transfer_8to16sub2ro= transfer_8to16sub2ro_x86_64;
690                    transfer_16to8add   = transfer_16to8add_x86_64;
691                    transfer8x8_copy    = transfer8x8_copy_x86_64;
692    
693                    /* Qpel stuff */
694                    xvid_QP_Funcs = &xvid_QP_Funcs_x86_64;
695                    xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_x86_64;
696    
697                    /* Interlacing Functions */
698                    MBFieldTest = MBFieldTest_x86_64;
699          }          }
700  #endif  #endif
701    
# Line 619  Line 714 
714                  return XVID_ERR_VERSION;                  return XVID_ERR_VERSION;
715    
716          info->actual_version = XVID_VERSION;          info->actual_version = XVID_VERSION;
717          info->build = "xvid-1.1-cvshead";          info->build = "xvid-1.2.0-dev";
718          info->cpu_flags = detect_cpu_flags();          info->cpu_flags = detect_cpu_flags();
   
 #if defined(_SMP) && defined(WIN32)  
         info->num_threads = pthread_num_processors_np();;  
 #else  
719          info->num_threads = 0;          info->num_threads = 0;
720    
721    #if defined(WIN32)
722      {
723        DWORD dwProcessAffinityMask, dwSystemAffinityMask;
724        if (GetProcessAffinityMask(GetCurrentProcess(), &dwProcessAffinityMask, &dwSystemAffinityMask)) {
725          int i;
726          for(i=0; i<32; i++) {
727            if ((dwProcessAffinityMask & (1<<i)))
728              info->num_threads++;
729          }
730        }
731      }
732  #endif  #endif
733    
734          return 0;          return 0;

Legend:
Removed from v.1.54  
changed lines
  Added in v.1.70

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