[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 1.69, Wed Jun 14 21:44:07 2006 UTC revision 1.83, Tue Mar 9 10:00:14 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 54  Line 75 
75    
76  #if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && 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_X86_64) || 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 90  Line 110 
110          }          }
111          return(0);          return(0);
112  }  }
113  #elif defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) || 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 137  Line 157 
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 257  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 278  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 309  Line 341 
341    
342          init_GMC(cpu_flags);          init_GMC(cpu_flags);
343    
344  #if defined(ARCH_IS_IA32)  #if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
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 334  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 377  Line 410 
410                  image_brightness = image_brightness_mmx;                  image_brightness = image_brightness_mmx;
411    
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 415  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 439  Line 477 
477                  interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_xmm;                  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 450  Line 487 
487                  transfer_8to16sub2ro = transfer_8to16sub2ro_xmm;                  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 511  Line 548 
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;
# Line 537  Line 572 
572    
573                  /* DCT operators */                  /* DCT operators */
574                  fdct = fdct_sse2_skal;                  fdct = fdct_sse2_skal;
575      /* idct = idct_sse2_skal; */   /* Is now IEEE1180 and Walken compliant. Disabled until fully tested. */                  idct = idct_sse2_skal;   /* Is now IEEE1180 and Walken compliant. */
576    
577                  /* postprocessing */                  /* postprocessing */
578                  image_brightness = image_brightness_sse2;                  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 632  Line 676 
676          }          }
677  #endif  #endif
678    
 #if defined(ARCH_IS_X86_64)  
         /* For now, only XVID_CPU_ASM is looked for, so user can still  
          * disable asm usage the usual way. When Intel EMT64 cpus will  
          * be out, maybe we'll have to check more precisely what cpu  
          * features there really are. */  
         if (cpu_flags & XVID_CPU_ASM) {  
                 /* SIMD state flusher */  
                 emms = emms_mmx;  
   
                 /* DCT operators */  
                 fdct = fdct_skal_x86_64;  
                 idct = idct_x86_64;  
   
                 /* SAD operators */  
                 sad16      = sad16_x86_64;  
                 sad8       = sad8_x86_64;  
                 sad16bi    = sad16bi_x86_64;  
                 sad8bi     = sad8bi_x86_64;  
                 dev16      = dev16_x86_64;  
                 sad16v     = sad16v_x86_64;  
                 sse8_16bit = sse8_16bit_x86_64;  
                 sse8_8bit  = sse8_8bit_x86_64;  
   
                 /* Interpolation operators */  
                 interpolate8x8_halfpel_h  = interpolate8x8_halfpel_h_x86_64;  
                 interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_x86_64;  
                 interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_x86_64;  
   
                 interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_x86_64;  
                 interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_x86_64;  
                 interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_x86_64;  
                 interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_x86_64;  
   
                 interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_x86_64;  
                 interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_x86_64;  
   
                 interpolate8x8_avg2 = interpolate8x8_avg2_x86_64;  
                 interpolate8x8_avg4 = interpolate8x8_avg4_x86_64;  
   
                 /* Quantization related functions */  
                 quant_h263_intra   = quant_h263_intra_x86_64;  
                 quant_h263_inter   = quant_h263_inter_x86_64;  
                 dequant_h263_intra = dequant_h263_intra_x86_64;  
                 dequant_h263_inter = dequant_h263_inter_x86_64;  
                 quant_mpeg_intra   = quant_mpeg_intra_x86_64;  
                 quant_mpeg_inter   = quant_mpeg_inter_x86_64;  
                 dequant_mpeg_intra   = dequant_mpeg_intra_x86_64;  
                 dequant_mpeg_inter   = dequant_mpeg_inter_x86_64;  
   
                 /* Block related functions */  
                 transfer_8to16copy  = transfer_8to16copy_x86_64;  
                 transfer_16to8copy  = transfer_16to8copy_x86_64;  
                 transfer_8to16sub   = transfer_8to16sub_x86_64;  
                 transfer_8to16subro = transfer_8to16subro_x86_64;  
                 transfer_8to16sub2  = transfer_8to16sub2_x86_64;  
                 transfer_8to16sub2ro= transfer_8to16sub2ro_x86_64;  
                 transfer_16to8add   = transfer_16to8add_x86_64;  
                 transfer8x8_copy    = transfer8x8_copy_x86_64;  
   
                 /* Qpel stuff */  
                 xvid_QP_Funcs = &xvid_QP_Funcs_x86_64;  
                 xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_x86_64;  
   
                 /* Interlacing Functions */  
                 MBFieldTest = MBFieldTest_x86_64;  
         }  
 #endif  
   
679  #if defined(_DEBUG)  #if defined(_DEBUG)
680      xvid_debug = init->debug;      xvid_debug = init->debug;
681  #endif  #endif
# Line 715  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-1.2.0-dev";          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;          info->num_threads = 0; /* single-thread */
697    
698    #if defined(_WIN32)
699    
 #if defined(WIN32)  
700    {    {
701      DWORD dwProcessAffinityMask, dwSystemAffinityMask;          SYSTEM_INFO siSysInfo;
702      if (GetProcessAffinityMask(GetCurrentProcess(), &dwProcessAffinityMask, &dwSystemAffinityMask)) {          GetSystemInfo(&siSysInfo);
703        int i;          info->num_threads = siSysInfo.dwNumberOfProcessors; /* number of _logical_ cores */
       for(i=0; i<32; i++) {  
         if ((dwProcessAffinityMask & (1<<i)))  
           info->num_threads++;  
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    
733  #endif  #endif
734    
735          return 0;          return 0;
# Line 777  Line 777 
777  }  }
778    
779  /*****************************************************************************  /*****************************************************************************
780   * XviD Global Entry point   * Xvid Global Entry point
781   *   *
782   * Well this function initialize all internal function pointers according   * Well this function initialize all internal function pointers according
783   * 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 810  Line 810 
810  }  }
811    
812  /*****************************************************************************  /*****************************************************************************
813   * XviD Native decoder entry point   * Xvid Native decoder entry point
814   *   *
815   * This function is just a wrapper to all the option cases.   * This function is just a wrapper to all the option cases.
816   *   *
# Line 842  Line 842 
842    
843    
844  /*****************************************************************************  /*****************************************************************************
845   * XviD Native encoder entry point   * Xvid Native encoder entry point
846   *   *
847   * This function is just a wrapper to all the option cases.   * This function is just a wrapper to all the option cases.
848   *   *

Legend:
Removed from v.1.69  
changed lines
  Added in v.1.83

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