[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 1.57, Sun Dec 5 13:56:13 2004 UTC revision 1.72, Sun Nov 12 01:40:36 2006 UTC
# Line 40  Line 40 
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 51  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 76  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 89  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 126  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 183  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 210  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 221  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;          interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_c;
232          interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_c;          interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_c;
233          interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_c;          interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_c;
# Line 250  Line 257 
257          yv12_to_yv12    = yv12_to_yv12_c;          yv12_to_yv12    = yv12_to_yv12_c;
258          rgb555_to_yv12  = rgb555_to_yv12_c;          rgb555_to_yv12  = rgb555_to_yv12_c;
259          rgb565_to_yv12  = rgb565_to_yv12_c;          rgb565_to_yv12  = rgb565_to_yv12_c;
260            rgb_to_yv12     = rgb_to_yv12_c;
261          bgr_to_yv12     = bgr_to_yv12_c;          bgr_to_yv12     = bgr_to_yv12_c;
262          bgra_to_yv12    = bgra_to_yv12_c;          bgra_to_yv12    = bgra_to_yv12_c;
263          abgr_to_yv12    = abgr_to_yv12_c;          abgr_to_yv12    = abgr_to_yv12_c;
# Line 271  Line 279 
279          /* All colorspace transformation functions YV12->User format */          /* All colorspace transformation functions YV12->User format */
280          yv12_to_rgb555  = yv12_to_rgb555_c;          yv12_to_rgb555  = yv12_to_rgb555_c;
281          yv12_to_rgb565  = yv12_to_rgb565_c;          yv12_to_rgb565  = yv12_to_rgb565_c;
282            yv12_to_rgb     = yv12_to_rgb_c;
283          yv12_to_bgr     = yv12_to_bgr_c;          yv12_to_bgr     = yv12_to_bgr_c;
284          yv12_to_bgra    = yv12_to_bgra_c;          yv12_to_bgra    = yv12_to_bgra_c;
285          yv12_to_abgr    = yv12_to_abgr_c;          yv12_to_abgr    = yv12_to_abgr_c;
# Line 300  Line 309 
309          sse8_16bit = sse8_16bit_c;          sse8_16bit = sse8_16bit_c;
310          sse8_8bit  = sse8_8bit_c;          sse8_8bit  = sse8_8bit_c;
311    
312            init_GMC(cpu_flags);
313    
314  #if defined(ARCH_IS_IA32)  #if defined(ARCH_IS_IA32)
315    
316          if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) ||          if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) ||
# Line 339  Line 350 
350                  transfer_8to16sub2 = transfer_8to16sub2_mmx;                  transfer_8to16sub2 = transfer_8to16sub2_mmx;
351                  transfer_16to8add  = transfer_16to8add_mmx;                  transfer_16to8add  = transfer_16to8add_mmx;
352                  transfer8x8_copy   = transfer8x8_copy_mmx;                  transfer8x8_copy   = transfer8x8_copy_mmx;
353                    transfer8x4_copy   = transfer8x4_copy_mmx;
354    
355                  /* Interlacing Functions */                  /* Interlacing Functions */
356                  MBFieldTest = MBFieldTest_mmx;                  MBFieldTest = MBFieldTest_mmx;
# Line 348  Line 360 
360                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_mmx;                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_mmx;
361                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx;
362    
363                    interpolate8x4_halfpel_h  = interpolate8x4_halfpel_h_mmx;
364                    interpolate8x4_halfpel_v  = interpolate8x4_halfpel_v_mmx;
365                    interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_mmx;
366    
367                  interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_mmx;                  interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_mmx;
368                  interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_mmx;                  interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_mmx;
369                  interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_mmx;                  interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_mmx;
# Line 365  Line 381 
381                  /* image input xxx_to_yv12 related functions */                  /* image input xxx_to_yv12 related functions */
382                  yv12_to_yv12  = yv12_to_yv12_mmx;                  yv12_to_yv12  = yv12_to_yv12_mmx;
383                  bgr_to_yv12   = bgr_to_yv12_mmx;                  bgr_to_yv12   = bgr_to_yv12_mmx;
384                    rgb_to_yv12   = rgb_to_yv12_mmx;
385                  bgra_to_yv12  = bgra_to_yv12_mmx;                  bgra_to_yv12  = bgra_to_yv12_mmx;
386                    rgba_to_yv12  = rgba_to_yv12_mmx;
387                  yuyv_to_yv12  = yuyv_to_yv12_mmx;                  yuyv_to_yv12  = yuyv_to_yv12_mmx;
388                  uyvy_to_yv12  = uyvy_to_yv12_mmx;                  uyvy_to_yv12  = uyvy_to_yv12_mmx;
389    
# Line 415  Line 433 
433                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_xmm;                  interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_xmm;
434                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm;
435    
436                    interpolate8x4_halfpel_h  = interpolate8x4_halfpel_h_xmm;
437                    interpolate8x4_halfpel_v  = interpolate8x4_halfpel_v_xmm;
438                    interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_xmm;
439    
440                  interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_xmm;                  interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_xmm;
441                  interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_xmm;                  interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_xmm;
442                  interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_xmm;                  interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_xmm;
443                  interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_xmm;                  interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_xmm;
444    
445                  /* Quantization */                  /* Quantization */
                 quant_mpeg_intra = quant_mpeg_intra_xmm;  
446                  quant_mpeg_inter = quant_mpeg_inter_xmm;                  quant_mpeg_inter = quant_mpeg_inter_xmm;
447    
448                  dequant_h263_intra = dequant_h263_intra_xmm;                  dequant_h263_intra = dequant_h263_intra_xmm;
# Line 429  Line 450 
450    
451                  /* Buffer transfer */                  /* Buffer transfer */
452                  transfer_8to16sub2 = transfer_8to16sub2_xmm;                  transfer_8to16sub2 = transfer_8to16sub2_xmm;
453                    transfer_8to16sub2ro = transfer_8to16sub2ro_xmm;
454    
455                  /* Colorspace transformation */                  /* Colorspace transformation */
456                  yv12_to_yv12  = yv12_to_yv12_xmm;                  yv12_to_yv12  = yv12_to_yv12_xmm;
# Line 450  Line 472 
472                  interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn;                  interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn;
473                  interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dn;                  interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dn;
474                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn;
475    
476                    interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_3dn;
477                    interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_3dn;
478                    interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_3dn;
479          }          }
480    
481          if ((cpu_flags & XVID_CPU_3DNOWEXT)) {          if ((cpu_flags & XVID_CPU_3DNOWEXT)) {
# Line 461  Line 487 
487                  transfer_8to16subro =  transfer_8to16subro_3dne;                  transfer_8to16subro =  transfer_8to16subro_3dne;
488                  transfer_16to8add = transfer_16to8add_3dne;                  transfer_16to8add = transfer_16to8add_3dne;
489                  transfer8x8_copy = transfer8x8_copy_3dne;                  transfer8x8_copy = transfer8x8_copy_3dne;
490                    transfer8x4_copy = transfer8x4_copy_3dne;
491    
492                  if ((cpu_flags & XVID_CPU_MMXEXT)) {                  if ((cpu_flags & XVID_CPU_MMXEXT)) {
493                          /* Inverse DCT */                          /* Inverse DCT */
# Line 474  Line 501 
501                          interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne;                          interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne;
502                          interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;                          interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;
503    
504                            interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_3dne;
505                            interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_3dne;
506                            interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_3dne;
507    
508                          /* Quantization */                          /* Quantization */
509                          quant_h263_intra = quant_h263_intra_3dne;               /* cmov only */                          quant_h263_intra = quant_h263_intra_3dne;               /* cmov only */
510                          quant_h263_inter = quant_h263_inter_3dne;                          quant_h263_inter = quant_h263_inter_3dne;
# Line 507  Line 538 
538                  sad16    = sad16_sse2;                  sad16    = sad16_sse2;
539                  dev16    = dev16_sse2;                  dev16    = dev16_sse2;
540    
541                  /* DCT operators                  /* DCT operators */
                  * no iDCT because it's not "Walken matching" */  
542                  fdct = fdct_sse2_skal;                  fdct = fdct_sse2_skal;
543        /* idct = idct_sse2_skal; */   /* Is now IEEE1180 and Walken compliant. Disabled until fully tested. */
544    
545                  /* postprocessing */                  /* postprocessing */
546                  image_brightness = image_brightness_sse2;                  image_brightness = image_brightness_sse2;
# Line 595  Line 626 
626            dequant_h263_intra = dequant_h263_intra_altivec_c;            dequant_h263_intra = dequant_h263_intra_altivec_c;
627            dequant_h263_inter = dequant_h263_inter_altivec_c;            dequant_h263_inter = dequant_h263_inter_altivec_c;
628    
629                      dequant_mpeg_intra = dequant_mpeg_intra_altivec_c;
630                      dequant_mpeg_inter = dequant_mpeg_inter_altivec_c;
631    
632                    /* Qpel stuff */                    /* Qpel stuff */
633                    xvid_QP_Funcs = &xvid_QP_Funcs_Altivec_C;                    xvid_QP_Funcs = &xvid_QP_Funcs_Altivec_C;
634                    xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_Altivec_C;                    xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_Altivec_C;
635          }          }
636  #endif  #endif
637    
638    #if defined(ARCH_IS_X86_64)
639            /* For now, only XVID_CPU_ASM is looked for, so user can still
640             * disable asm usage the usual way. When Intel EMT64 cpus will
641             * be out, maybe we'll have to check more precisely what cpu
642             * features there really are. */
643            if (cpu_flags & XVID_CPU_ASM) {
644                    /* SIMD state flusher */
645                    emms = emms_mmx;
646    
647                    /* DCT operators */
648                    fdct = fdct_skal_x86_64;
649                    idct = idct_x86_64;
650    
651                    /* SAD operators */
652                    sad16      = sad16_x86_64;
653                    sad8       = sad8_x86_64;
654                    sad16bi    = sad16bi_x86_64;
655                    sad8bi     = sad8bi_x86_64;
656                    dev16      = dev16_x86_64;
657                    sad16v     = sad16v_x86_64;
658                    sse8_16bit = sse8_16bit_x86_64;
659                    sse8_8bit  = sse8_8bit_x86_64;
660    
661                    /* Interpolation operators */
662                    interpolate8x8_halfpel_h  = interpolate8x8_halfpel_h_x86_64;
663                    interpolate8x8_halfpel_v  = interpolate8x8_halfpel_v_x86_64;
664                    interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_x86_64;
665    
666                    interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_x86_64;
667                    interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_x86_64;
668                    interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_x86_64;
669                    interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_x86_64;
670    
671                    interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_x86_64;
672                    interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_x86_64;
673    
674                    interpolate8x8_avg2 = interpolate8x8_avg2_x86_64;
675                    interpolate8x8_avg4 = interpolate8x8_avg4_x86_64;
676    
677                    /* Quantization related functions */
678                    quant_h263_intra   = quant_h263_intra_x86_64;
679                    quant_h263_inter   = quant_h263_inter_x86_64;
680                    dequant_h263_intra = dequant_h263_intra_x86_64;
681                    dequant_h263_inter = dequant_h263_inter_x86_64;
682                    /*quant_mpeg_intra   = quant_mpeg_intra_x86_64; fix me! */
683                    quant_mpeg_inter   = quant_mpeg_inter_x86_64;
684                    dequant_mpeg_intra   = dequant_mpeg_intra_x86_64;
685                    dequant_mpeg_inter   = dequant_mpeg_inter_x86_64;
686    
687                    /* Block related functions */
688                    transfer_8to16copy  = transfer_8to16copy_x86_64;
689                    transfer_16to8copy  = transfer_16to8copy_x86_64;
690                    transfer_8to16sub   = transfer_8to16sub_x86_64;
691                    transfer_8to16subro = transfer_8to16subro_x86_64;
692                    transfer_8to16sub2  = transfer_8to16sub2_x86_64;
693                    transfer_8to16sub2ro= transfer_8to16sub2ro_x86_64;
694                    transfer_16to8add   = transfer_16to8add_x86_64;
695                    transfer8x8_copy    = transfer8x8_copy_x86_64;
696    
697                    /* Qpel stuff */
698                    xvid_QP_Funcs = &xvid_QP_Funcs_x86_64;
699                    xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_x86_64;
700    
701                    /* Interlacing Functions */
702                    MBFieldTest = MBFieldTest_x86_64;
703            }
704    #endif
705    
706  #if defined(_DEBUG)  #if defined(_DEBUG)
707      xvid_debug = init->debug;      xvid_debug = init->debug;
708  #endif  #endif
# Line 616  Line 718 
718                  return XVID_ERR_VERSION;                  return XVID_ERR_VERSION;
719    
720          info->actual_version = XVID_VERSION;          info->actual_version = XVID_VERSION;
721          info->build = "xvid-1.1-cvshead";          info->build = "xvid-1.2.0-dev";
722          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  
723          info->num_threads = 0;          info->num_threads = 0;
724    
725    #if defined(WIN32)
726      {
727        DWORD dwProcessAffinityMask, dwSystemAffinityMask;
728        if (GetProcessAffinityMask(GetCurrentProcess(), &dwProcessAffinityMask, &dwSystemAffinityMask)) {
729          int i;
730          for(i=0; i<32; i++) {
731            if ((dwProcessAffinityMask & (1<<i)))
732              info->num_threads++;
733          }
734        }
735      }
736  #endif  #endif
737    
738          return 0;          return 0;

Legend:
Removed from v.1.57  
changed lines
  Added in v.1.72

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