[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 1.61, Sun Jan 16 10:34:52 2005 UTC revision 1.73, Wed Dec 6 19:55:07 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 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;
# Line 136  Line 137 
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;
139    
140          if ((cpu_flags & XVID_CPU_SSE2) && sigill_check(sse2_os_trigger))          if ((cpu_flags & (XVID_CPU_SSE2|XVID_CPU_SSE3)) && sigill_check(sse2_os_trigger))
141                  cpu_flags &= ~XVID_CPU_SSE2;                  cpu_flags &= ~(XVID_CPU_SSE2|XVID_CPU_SSE3);
142  #endif  #endif
143    
144  #if defined(ARCH_IS_PPC)  #if defined(ARCH_IS_PPC)
# 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 213  Line 214 
214          transfer_8to16sub2ro = transfer_8to16sub2ro_c;          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;          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 251  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 272  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 301  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 340  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 349  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 366  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 416  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 452  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 463  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 476  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 509  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;
547          }          }
548    
549    #if 0 // TODO: test...
550            if ((cpu_flags & XVID_CPU_SSE3)) {
551    
552                    /* SAD operators */
553                    sad16    = sad16_sse3;
554                    dev16    = dev16_sse3;
555            }
556    #endif
557  #endif /* ARCH_IS_IA32 */  #endif /* ARCH_IS_IA32 */
558    
559  #if defined(ARCH_IS_IA64)  #if defined(ARCH_IS_IA64)
# Line 597  Line 635 
635            dequant_h263_intra = dequant_h263_intra_altivec_c;            dequant_h263_intra = dequant_h263_intra_altivec_c;
636            dequant_h263_inter = dequant_h263_inter_altivec_c;            dequant_h263_inter = dequant_h263_inter_altivec_c;
637    
638                      dequant_mpeg_intra = dequant_mpeg_intra_altivec_c;
639                      dequant_mpeg_inter = dequant_mpeg_inter_altivec_c;
640    
641                    /* Qpel stuff */                    /* Qpel stuff */
642                    xvid_QP_Funcs = &xvid_QP_Funcs_Altivec_C;                    xvid_QP_Funcs = &xvid_QP_Funcs_Altivec_C;
643                    xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_Altivec_C;                    xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_Altivec_C;
# Line 610  Line 651 
651           * features there really are. */           * features there really are. */
652          if (cpu_flags & XVID_CPU_ASM) {          if (cpu_flags & XVID_CPU_ASM) {
653                  /* SIMD state flusher */                  /* SIMD state flusher */
654                  emms = emms_3dn;                  emms = emms_mmx;
655    
656                  /* DCT operators */                  /* DCT operators */
657                  fdct = fdct_skal_x86_64;                  fdct = fdct_skal_x86_64;
# Line 647  Line 688 
688                  quant_h263_inter   = quant_h263_inter_x86_64;                  quant_h263_inter   = quant_h263_inter_x86_64;
689                  dequant_h263_intra = dequant_h263_intra_x86_64;                  dequant_h263_intra = dequant_h263_intra_x86_64;
690                  dequant_h263_inter = dequant_h263_inter_x86_64;                  dequant_h263_inter = dequant_h263_inter_x86_64;
691                  quant_mpeg_intra   = quant_mpeg_intra_x86_64;                  /*quant_mpeg_intra   = quant_mpeg_intra_x86_64; fix me! */
692                  quant_mpeg_inter   = quant_mpeg_inter_x86_64;                  quant_mpeg_inter   = quant_mpeg_inter_x86_64;
693                  dequant_mpeg_intra   = dequant_mpeg_intra_x86_64;                  dequant_mpeg_intra   = dequant_mpeg_intra_x86_64;
694                  dequant_mpeg_inter   = dequant_mpeg_inter_x86_64;                  dequant_mpeg_inter   = dequant_mpeg_inter_x86_64;
# Line 686  Line 727 
727                  return XVID_ERR_VERSION;                  return XVID_ERR_VERSION;
728    
729          info->actual_version = XVID_VERSION;          info->actual_version = XVID_VERSION;
730          info->build = "xvid-1.1.0-beta1";          info->build = "xvid-1.2.0-dev";
731          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  
732          info->num_threads = 0;          info->num_threads = 0;
733    
734    #if defined(WIN32)
735      {
736        DWORD dwProcessAffinityMask, dwSystemAffinityMask;
737        if (GetProcessAffinityMask(GetCurrentProcess(), &dwProcessAffinityMask, &dwSystemAffinityMask)) {
738          int i;
739          for(i=0; i<32; i++) {
740            if ((dwProcessAffinityMask & (1<<i)))
741              info->num_threads++;
742          }
743        }
744      }
745  #endif  #endif
746    
747          return 0;          return 0;

Legend:
Removed from v.1.61  
changed lines
  Added in v.1.73

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