[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 1.45.2.8, Wed Jun 11 14:10:40 2003 UTC revision 1.49, Thu Apr 1 11:11:28 2004 UTC
# Line 3  Line 3 
3   *  XVID MPEG-4 VIDEO CODEC   *  XVID MPEG-4 VIDEO CODEC
4   *  - Native API implementation  -   *  - Native API implementation  -
5   *   *
6   *  Copyright(C) 2001-2003 Peter Ross <pross@xvid.org>   *  Copyright(C) 2001-2004 Peter Ross <pross@xvid.org>
7   *   *
8   *  This program is free software ; you can redistribute it and/or modify   *  This program is free software ; you can redistribute it and/or modify
9   *  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 39  Line 39 
39  #include "image/reduced.h"  #include "image/reduced.h"
40  #include "utils/mem_transfer.h"  #include "utils/mem_transfer.h"
41  #include "utils/mbfunctions.h"  #include "utils/mbfunctions.h"
42  #include "quant/quant_h263.h"  #include "quant/quant.h"
 #include "quant/quant_mpeg4.h"  
43  #include "motion/motion.h"  #include "motion/motion.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"
47  #include "bitstream/mbcoding.h"  #include "bitstream/mbcoding.h"
48    #include "image/qpel.h"
49    #include "image/postprocessing.h"
50    
51  #if defined(_DEBUG)  #if defined(_DEBUG)
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)  #if defined(ARCH_IS_IA32)
   
56  #if defined(_MSC_VER)  #if defined(_MSC_VER)
57  #       include <windows.h>  #       include <windows.h>
58  #else  #else
# Line 163  Line 163 
163  {  {
164          unsigned int cpu_flags;          unsigned int cpu_flags;
165    
166          if (XVID_MAJOR(init->version) != 1) /* v1.x.x */          if (XVID_VERSION_MAJOR(init->version) != 1) /* v1.x.x */
167                  return XVID_ERR_VERSION;                  return XVID_ERR_VERSION;
168    
169          cpu_flags = (init->cpu_flags & XVID_CPU_FORCE) ? init->cpu_flags : detect_cpu_flags();          cpu_flags = (init->cpu_flags & XVID_CPU_FORCE) ? init->cpu_flags : detect_cpu_flags();
# Line 182  Line 182 
182          /* Restore FPU context : emms_c is a nop functions */          /* Restore FPU context : emms_c is a nop functions */
183          emms = emms_c;          emms = emms_c;
184    
185            /* Qpel stuff */
186            xvid_QP_Funcs = &xvid_QP_Funcs_C;
187            xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_C;
188            xvid_Init_QP();
189    
190          /* Quantization functions */          /* Quantization functions */
191          quant_intra   = quant_intra_c;          quant_h263_intra   = quant_h263_intra_c;
192          dequant_intra = dequant_intra_c;          quant_h263_inter   = quant_h263_inter_c;
193          quant_inter   = quant_inter_c;          dequant_h263_intra = dequant_h263_intra_c;
194          dequant_inter = dequant_inter_c;          dequant_h263_inter = dequant_h263_inter_c;
195    
196          quant4_intra   = quant4_intra_c;          quant_mpeg_intra   = quant_mpeg_intra_c;
197          dequant4_intra = dequant4_intra_c;          quant_mpeg_inter   = quant_mpeg_inter_c;
198          quant4_inter   = quant4_inter_c;          dequant_mpeg_intra = dequant_mpeg_intra_c;
199          dequant4_inter = dequant4_inter_c;          dequant_mpeg_inter = dequant_mpeg_inter_c;
200    
201          /* Block transfer related functions */          /* Block transfer related functions */
202          transfer_8to16copy = transfer_8to16copy_c;          transfer_8to16copy = transfer_8to16copy_c;
# Line 224  Line 229 
229          interpolate8x8_avg2 = interpolate8x8_avg2_c;          interpolate8x8_avg2 = interpolate8x8_avg2_c;
230          interpolate8x8_avg4 = interpolate8x8_avg4_c;          interpolate8x8_avg4 = interpolate8x8_avg4_c;
231    
232          /* reduced resoltuion */          /* postprocessing */
233            image_brightness = image_brightness_c;
234    
235            /* reduced resolution */
236          copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_C;          copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_C;
237          add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_C;          add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_C;
238          vfilter_31 = xvid_VFilter_31_C;          vfilter_31 = xvid_VFilter_31_C;
# Line 243  Line 251 
251          bgra_to_yv12    = bgra_to_yv12_c;          bgra_to_yv12    = bgra_to_yv12_c;
252          abgr_to_yv12    = abgr_to_yv12_c;          abgr_to_yv12    = abgr_to_yv12_c;
253          rgba_to_yv12    = rgba_to_yv12_c;          rgba_to_yv12    = rgba_to_yv12_c;
254            argb_to_yv12    = argb_to_yv12_c;
255          yuyv_to_yv12    = yuyv_to_yv12_c;          yuyv_to_yv12    = yuyv_to_yv12_c;
256          uyvy_to_yv12    = uyvy_to_yv12_c;          uyvy_to_yv12    = uyvy_to_yv12_c;
257    
# Line 252  Line 261 
261          bgrai_to_yv12   = bgrai_to_yv12_c;          bgrai_to_yv12   = bgrai_to_yv12_c;
262          abgri_to_yv12   = abgri_to_yv12_c;          abgri_to_yv12   = abgri_to_yv12_c;
263          rgbai_to_yv12   = rgbai_to_yv12_c;          rgbai_to_yv12   = rgbai_to_yv12_c;
264            argbi_to_yv12   = argbi_to_yv12_c;
265          yuyvi_to_yv12   = yuyvi_to_yv12_c;          yuyvi_to_yv12   = yuyvi_to_yv12_c;
266          uyvyi_to_yv12   = uyvyi_to_yv12_c;          uyvyi_to_yv12   = uyvyi_to_yv12_c;
267    
   
268          /* All colorspace transformation functions YV12->User format */          /* All colorspace transformation functions YV12->User format */
269          yv12_to_rgb555  = yv12_to_rgb555_c;          yv12_to_rgb555  = yv12_to_rgb555_c;
270          yv12_to_rgb565  = yv12_to_rgb565_c;          yv12_to_rgb565  = yv12_to_rgb565_c;
# Line 263  Line 272 
272          yv12_to_bgra    = yv12_to_bgra_c;          yv12_to_bgra    = yv12_to_bgra_c;
273          yv12_to_abgr    = yv12_to_abgr_c;          yv12_to_abgr    = yv12_to_abgr_c;
274          yv12_to_rgba    = yv12_to_rgba_c;          yv12_to_rgba    = yv12_to_rgba_c;
275            yv12_to_argb    = yv12_to_argb_c;
276          yv12_to_yuyv    = yv12_to_yuyv_c;          yv12_to_yuyv    = yv12_to_yuyv_c;
277          yv12_to_uyvy    = yv12_to_uyvy_c;          yv12_to_uyvy    = yv12_to_uyvy_c;
278    
# Line 272  Line 282 
282          yv12_to_bgrai   = yv12_to_bgrai_c;          yv12_to_bgrai   = yv12_to_bgrai_c;
283          yv12_to_abgri   = yv12_to_abgri_c;          yv12_to_abgri   = yv12_to_abgri_c;
284          yv12_to_rgbai   = yv12_to_rgbai_c;          yv12_to_rgbai   = yv12_to_rgbai_c;
285            yv12_to_argbi   = yv12_to_argbi_c;
286          yv12_to_yuyvi   = yv12_to_yuyvi_c;          yv12_to_yuyvi   = yv12_to_yuyvi_c;
287          yv12_to_uyvyi   = yv12_to_uyvyi_c;          yv12_to_uyvyi   = yv12_to_uyvyi_c;
288    
# Line 283  Line 294 
294          sad8bi   = sad8bi_c;          sad8bi   = sad8bi_c;
295          dev16    = dev16_c;          dev16    = dev16_c;
296          sad16v   = sad16v_c;          sad16v   = sad16v_c;
297            sse8_16bit = sse8_16bit_c;
 /*      Halfpel8_Refine = Halfpel8_Refine_c; */  
298    
299  #if defined(ARCH_IS_IA32)  #if defined(ARCH_IS_IA32)
300    
# Line 304  Line 314 
314          if ((cpu_flags & XVID_CPU_MMX)) {          if ((cpu_flags & XVID_CPU_MMX)) {
315    
316                  /* Forward and Inverse Discrete Cosine Transformation functions */                  /* Forward and Inverse Discrete Cosine Transformation functions */
317                  fdct = fdct_mmx;                  fdct = fdct_mmx_skal;
318                  idct = idct_mmx;                  idct = idct_mmx;
319    
320                    /* Qpel stuff */
321                    xvid_QP_Funcs = &xvid_QP_Funcs_mmx;
322                    xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_mmx;
323    
324                  /* Quantization related functions */                  /* Quantization related functions */
325                  quant_intra   = quant_intra_mmx;                  quant_h263_intra   = quant_h263_intra_mmx;
326                  dequant_intra = dequant_intra_mmx;                  quant_h263_inter   = quant_h263_inter_mmx;
327                  quant_inter   = quant_inter_mmx;                  dequant_h263_intra = dequant_h263_intra_mmx;
328                  dequant_inter = dequant_inter_mmx;                  dequant_h263_inter = dequant_h263_inter_mmx;
329    
330                  quant4_intra   = quant4_intra_mmx;                  quant_mpeg_intra   = quant_mpeg_intra_mmx;
331                  dequant4_intra = dequant4_intra_mmx;                  quant_mpeg_inter   = quant_mpeg_inter_mmx;
332                  quant4_inter   = quant4_inter_mmx;                  dequant_mpeg_intra = dequant_mpeg_intra_mmx;
333                  dequant4_inter = dequant4_inter_mmx;                  dequant_mpeg_inter = dequant_mpeg_inter_mmx;
334    
335                  /* Block related functions */                  /* Block related functions */
336                  transfer_8to16copy = transfer_8to16copy_mmx;                  transfer_8to16copy = transfer_8to16copy_mmx;
# Line 341  Line 355 
355                  interpolate8x8_avg2 = interpolate8x8_avg2_mmx;                  interpolate8x8_avg2 = interpolate8x8_avg2_mmx;
356                  interpolate8x8_avg4 = interpolate8x8_avg4_mmx;                  interpolate8x8_avg4 = interpolate8x8_avg4_mmx;
357    
358                    /* postprocessing */
359                    image_brightness = image_brightness_mmx;
360    
361                  /* reduced resolution */                  /* reduced resolution */
362                  copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_mmx;                  copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_mmx;
363                  add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_mmx;                  add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_mmx;
# Line 372  Line 389 
389                  sad8bi  = sad8bi_mmx;                  sad8bi  = sad8bi_mmx;
390                  dev16    = dev16_mmx;                  dev16    = dev16_mmx;
391                  sad16v   = sad16v_mmx;                  sad16v   = sad16v_mmx;
392                    sse8_16bit = sse8_16bit_mmx;
393          }          }
394    
395          /* these 3dnow functions are faster than mmx, but slower than xmm. */          /* these 3dnow functions are faster than mmx, but slower than xmm. */
# Line 390  Line 408 
408    
409          if ((cpu_flags & XVID_CPU_MMXEXT)) {          if ((cpu_flags & XVID_CPU_MMXEXT)) {
410    
411                  /* Inverse DCT */                  /* DCT */
412                    fdct = fdct_xmm_skal;
413                  idct = idct_xmm;                  idct = idct_xmm;
414    
415                  /* Interpolation */                  /* Interpolation */
# Line 403  Line 422 
422                  add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm;                  add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm;
423    
424                  /* Quantization */                  /* Quantization */
425                  quant4_intra = quant4_intra_xmm;                  quant_mpeg_intra = quant_mpeg_intra_xmm;
426                  quant4_inter = quant4_inter_xmm;                  quant_mpeg_inter = quant_mpeg_inter_xmm;
427    
428                  dequant_intra = dequant_intra_xmm;                  dequant_h263_intra = dequant_h263_intra_xmm;
429                  dequant_inter = dequant_inter_xmm;                  dequant_h263_inter = dequant_h263_inter_xmm;
430    
431                  /* Buffer transfer */                  /* Buffer transfer */
432                  transfer_8to16sub2 = transfer_8to16sub2_xmm;                  transfer_8to16sub2 = transfer_8to16sub2_xmm;
# Line 449  Line 468 
468                  transfer8x8_copy = transfer8x8_copy_3dne;                  transfer8x8_copy = transfer8x8_copy_3dne;
469    
470                  /* Quantization */                  /* Quantization */
471                  dequant4_intra = dequant4_intra_3dne;                  quant_h263_intra = quant_h263_intra_3dne;
472                  dequant4_inter = dequant4_inter_3dne;                  quant_h263_inter = quant_h263_inter_3dne;
473                  quant_intra = quant_intra_3dne;                  dequant_mpeg_intra = dequant_mpeg_intra_3dne;
474                  quant_inter = quant_inter_3dne;                  dequant_mpeg_inter = dequant_mpeg_inter_3dne;
475                  dequant_intra = dequant_intra_3dne;                  dequant_h263_intra = dequant_h263_intra_3dne;
476                  dequant_inter = dequant_inter_3dne;                  dequant_h263_inter = dequant_h263_inter_3dne;
477    
478                  /* ME functions */                  /* ME functions */
479                  calc_cbp = calc_cbp_3dne;                  calc_cbp = calc_cbp_3dne;
# Line 470  Line 489 
489                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;                  interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;
490          }          }
491    
   
492          if ((cpu_flags & XVID_CPU_SSE2)) {          if ((cpu_flags & XVID_CPU_SSE2)) {
493    
494                  calc_cbp = calc_cbp_sse2;                  calc_cbp = calc_cbp_sse2;
495    
496                  /* Quantization */                  /* Quantization */
497                  quant_intra   = quant_intra_sse2;                  quant_h263_intra   = quant_h263_intra_sse2;
498                  dequant_intra = dequant_intra_sse2;                  quant_h263_inter   = quant_h263_inter_sse2;
499                  quant_inter   = quant_inter_sse2;                  dequant_h263_intra = dequant_h263_intra_sse2;
500                  dequant_inter = dequant_inter_sse2;                  dequant_h263_inter = dequant_h263_inter_sse2;
501    
502  #if defined(EXPERIMENTAL_SSE2_CODE)                  /* SAD operators */
                 /* ME; slower than xmm */  
503                  sad16    = sad16_sse2;                  sad16    = sad16_sse2;
504                  dev16    = dev16_sse2;                  dev16    = dev16_sse2;
505  #endif  
506                  /* Forward and Inverse DCT */                  /* DCT operators
507  #if 0 /* Both function are known to be unprecise, better keep them deactivated */                   * no iDCT because it's not "Walken matching" */
508                  idct  = idct_sse2;                  fdct = fdct_sse2_skal;
                 fdct = fdct_sse2;  
 #endif  
509          }          }
510  #endif  #endif /* ARCH_IS_IA32 */
511    
512  #if defined(ARCH_IS_IA64)  #if defined(ARCH_IS_IA64)
513          if ((cpu_flags & XVID_CPU_ASM)) { /* use assembler routines? */          if ((cpu_flags & XVID_CPU_ASM)) { /* use assembler routines? */
# Line 507  Line 522 
522            sad8 = sad8_ia64;            sad8 = sad8_ia64;
523            dev16 = dev16_ia64;            dev16 = dev16_ia64;
524  /*        Halfpel8_Refine = Halfpel8_Refine_ia64; */  /*        Halfpel8_Refine = Halfpel8_Refine_ia64; */
525            quant_intra = quant_intra_ia64;            quant_h263_intra = quant_h263_intra_ia64;
526            dequant_intra = dequant_intra_ia64;            quant_h263_inter = quant_h263_inter_ia64;
527            quant_inter = quant_inter_ia64;            dequant_h263_intra = dequant_h263_intra_ia64;
528            dequant_inter = dequant_inter_ia64;            dequant_h263_inter = dequant_h263_inter_ia64;
529            transfer_8to16copy = transfer_8to16copy_ia64;            transfer_8to16copy = transfer_8to16copy_ia64;
530            transfer_16to8copy = transfer_16to8copy_ia64;            transfer_16to8copy = transfer_16to8copy_ia64;
531            transfer_8to16sub = transfer_8to16sub_ia64;            transfer_8to16sub = transfer_8to16sub_ia64;
532            transfer_8to16sub2 = transfer_8to16sub2_ia64;            transfer_8to16sub2 = transfer_8to16sub2_ia64;
533            transfer_16to8add = transfer_16to8add_ia64;            transfer_16to8add = transfer_16to8add_ia64;
534            transfer8x8_copy = transfer8x8_copy_ia64;            transfer8x8_copy = transfer8x8_copy_ia64;
           DPRINTF(DPRINTF_DEBUG, "Using IA-64 assembler routines.");  
535          }          }
536  #endif  #endif
537    
# Line 550  Line 564 
564  static int  static int
565  xvid_gbl_info(xvid_gbl_info_t * info)  xvid_gbl_info(xvid_gbl_info_t * info)
566  {  {
567          if (XVID_MAJOR(info->version) != 1) /* v1.x.x */          if (XVID_VERSION_MAJOR(info->version) != 1) /* v1.x.x */
568                  return XVID_ERR_VERSION;                  return XVID_ERR_VERSION;
569    
570          info->actual_version = XVID_VERSION;          info->actual_version = XVID_VERSION;
571          info->build = "dev-api-4";          info->build = "xvid-1.0.0";
572          info->cpu_flags = detect_cpu_flags();          info->cpu_flags = detect_cpu_flags();
573    
574  #if defined(_SMP) && defined(WIN32)  #if defined(_SMP) && defined(WIN32)
# Line 576  Line 590 
590          int height2;          int height2;
591          IMAGE img;          IMAGE img;
592    
593          if (XVID_MAJOR(convert->version) != 1)   /* v1.x.x */          if (XVID_VERSION_MAJOR(convert->version) != 1)   /* v1.x.x */
594                return XVID_ERR_VERSION;                return XVID_ERR_VERSION;
595    
596  #if 0  #if 0
# Line 607  Line 621 
621          return 0;          return 0;
622  }  }
623    
   
   
 void fill8(uint8_t * block, int size, int value)  
 {  
         int i;  
         for (i = 0; i < size; i++)  
                 block[i] = value;  
 }  
   
 void fill16(int16_t * block, int size, int value)  
 {  
         int i;  
         for (i = 0; i < size; i++)  
                 block[i] = value;  
 }  
   
 #define RANDOM(min,max) min + (rand() % (max-min))  
   
 void random8(uint8_t * block, int size, int min, int max)  
 {  
         int i;  
         for (i = 0; i < size; i++)  
                 block[i] = RANDOM(min,max);  
 }  
   
 void random16(int16_t * block, int size, int min, int max)  
 {  
         int i;  
         for (i = 0; i < size; i++)  
                 block[i] = RANDOM(min,max);  
 }  
   
 int compare16(const int16_t * blockA, const int16_t * blockB, int size)  
 {  
         int i;  
         for (i = 0; i < size; i++)  
                 if (blockA[i] != blockB[i])  
                         return 1;  
   
         return 0;  
 }  
   
 int diff16(const int16_t * blockA, const int16_t * blockB, int size)  
 {  
         int i, diff = 0;  
         for (i = 0; i < size; i++)  
                 diff += abs(blockA[i]-blockB[i]);  
         return diff;  
 }  
   
   
 #define XVID_TEST_RANDOM        0x00000001      /* random input data */  
 #define XVID_TEST_VERBOSE       0x00000002      /* verbose error output */  
   
   
 #define TEST_FORWARD    0x00000001      /* intra */  
 #define TEST_FDCT  (TEST_FORWARD)  
 #define TEST_IDCT  (0)  
   
 static int test_transform(void * funcA, void * funcB, const char * nameB,  
                                    int test, int flags)  
 {  
         int i;  
         int64_t timeSTART;  
         int64_t timeA = 0;  
         int64_t timeB = 0;  
         DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE);  
         DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE);  
         int min, max;  
         int count = 0;  
   
         int tmp;  
         int min_error = 0x10000*64;  
         int max_error = 0;  
   
   
         if ((test & TEST_FORWARD))      /* forward */  
         {  
                 min = -256;  
                 max = 255;  
         }else{          /* inverse */  
                 min = -2048;  
                 max = 2047;  
         }  
   
         for (i = 0; i < 64*64; i++)  
         {  
                 if ((flags & XVID_TEST_RANDOM))  
                 {  
                         random16(arrayA, 64, min, max);  
                 }else{  
                         fill16(arrayA, 64, i);  
                 }  
                 memcpy(arrayB, arrayA, 64*sizeof(int16_t));  
   
                 if ((test & TEST_FORWARD))  
                 {  
                         timeSTART = read_counter();  
                         ((fdctFunc*)funcA)(arrayA);  
                         timeA += read_counter() - timeSTART;  
   
                         timeSTART = read_counter();  
                         ((fdctFunc*)funcB)(arrayB);  
                         timeB += read_counter() - timeSTART;  
                 }  
                 else  
                 {  
                         timeSTART = read_counter();  
                         ((idctFunc*)funcA)(arrayA);  
                         timeA += read_counter() - timeSTART;  
   
                         timeSTART = read_counter();  
                         ((idctFunc*)funcB)(arrayB);  
                         timeB += read_counter() - timeSTART;  
                 }  
   
                 tmp = diff16(arrayA, arrayB, 64) / 64;  
                 if (tmp > max_error)  
                         max_error = tmp;  
                 if (tmp < min_error)  
                         min_error = tmp;  
   
                 count++;  
         }  
   
         /* print the "average difference" of best/worst transforms */  
         printf("%s:\t%i\t(min_error:%i, max_error:%i)\n", nameB, (int)(timeB / count), min_error, max_error);  
   
         return 0;  
 }  
   
   
 #define TEST_QUANT      0x00000001      /* forward quantization */  
 #define TEST_INTRA      0x00000002      /* intra */  
 #define TEST_QUANT_INTRA        (TEST_QUANT|TEST_INTRA)  
 #define TEST_QUANT_INTER        (TEST_QUANT)  
 #define TEST_DEQUANT_INTRA      (TEST_INTRA)  
 #define TEST_DEQUANT_INTER      (0)  
   
 static int test_quant(void * funcA, void * funcB, const char * nameB,  
                            int test, int flags)  
 {  
         int q,i;  
         int64_t timeSTART;  
         int64_t timeA = 0;  
         int64_t timeB = 0;  
         int retA = 0, retB = 0;  
         DECLARE_ALIGNED_MATRIX(arrayX, 1, 64, int16_t, CACHE_LINE);  
         DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE);  
         DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE);  
         int min, max;  
         int count = 0;  
         int errors = 0;  
   
         if ((test & TEST_QUANT))        /* quant */  
         {  
                 min = -2048;  
                 max = 2047;  
         }else{          /* dequant */  
                 min = -256;  
                 max = 255;  
         }  
   
         for (q = 1; q <= 31; q++)       /* quantizer */  
         {  
                 for (i = min; i < max; i++)     /* input coeff */  
                 {  
                         if ((flags & XVID_TEST_RANDOM))  
                         {  
                                 random16(arrayX, 64, min, max);  
                         }else{  
                                 fill16(arrayX, 64, i);  
                         }  
   
                         if ((test & TEST_INTRA))        /* intra */  
                         {  
                                 timeSTART = read_counter();  
                                 ((quanth263_intraFunc*)funcA)(arrayA, arrayX, q, q);  
                                 timeA += read_counter() - timeSTART;  
   
                                 timeSTART = read_counter();  
                                 ((quanth263_intraFunc*)funcB)(arrayB, arrayX, q, q);  
                                 timeB += read_counter() - timeSTART;  
                         }  
                         else    /* inter */  
                         {  
                                 timeSTART = read_counter();  
                                 retA = ((quanth263_interFunc*)funcA)(arrayA, arrayX, q);  
                                 timeA += read_counter() - timeSTART;  
   
                                 timeSTART = read_counter();  
                                 retB = ((quanth263_interFunc*)funcB)(arrayB, arrayX, q);  
                                 timeB += read_counter() - timeSTART;  
                         }  
   
                         /* compare return value from quant_inter, and compare (de)quantiz'd arrays */  
                         if ( ((test&TEST_QUANT) && !(test&TEST_INTRA) && retA != retB ) ||  
                                 compare16(arrayA, arrayB, 64))  
                         {  
                                 errors++;  
                                 if ((flags & XVID_TEST_VERBOSE))  
                                         printf("%s error: q=%i, i=%i\n", nameB, q, i);  
                         }  
   
                         count++;  
                 }  
         }  
   
         printf("%s:\t%i", nameB, (int)(timeB / count));  
         if (errors>0)  
                 printf("\t(%i errors out of %i)", errors, count);  
         printf("\n");  
   
         return 0;  
 }  
   
   
   
 int xvid_init_test(int flags)  
 {  
 #if defined(ARCH_IS_IA32)  
         int cpu_flags;  
 #endif  
   
         printf("XviD tests\n\n");  
   
 #if defined(ARCH_IS_IA32)  
         cpu_flags = detect_cpu_flags();  
 #endif  
   
         idct_int32_init();  
         emms();  
   
         srand(time(0));  
   
         /* fDCT test */  
         printf("--- fdct ---\n");  
                 test_transform(fdct_int32, fdct_int32, "c", TEST_FDCT, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_transform(fdct_int32, fdct_mmx, "mmx", TEST_FDCT, flags);  
         if (cpu_flags & XVID_CPU_SSE2)  
                 test_transform(fdct_int32, fdct_sse2, "sse2", TEST_FDCT, flags);  
 #endif  
   
         /* iDCT test */  
         printf("\n--- idct ---\n");  
                 test_transform(idct_int32, idct_int32, "c", TEST_IDCT, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_transform(idct_int32, idct_mmx, "mmx", TEST_IDCT, flags);  
         if (cpu_flags & XVID_CPU_MMXEXT)  
                 test_transform(idct_int32, idct_xmm, "xmm", TEST_IDCT, flags);  
         if (cpu_flags & XVID_CPU_3DNOWEXT)  
                 test_transform(idct_int32, idct_3dne, "3dne", TEST_IDCT, flags);  
         if (cpu_flags & XVID_CPU_SSE2)  
                 test_transform(idct_int32, idct_sse2, "sse2", TEST_IDCT, flags);  
 #endif  
   
         /* Intra quantization test */  
         printf("\n--- quant intra ---\n");  
                 test_quant(quant_intra_c, quant_intra_c, "c", TEST_QUANT_INTRA, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_quant(quant_intra_c, quant_intra_mmx, "mmx", TEST_QUANT_INTRA, flags);  
         if (cpu_flags & XVID_CPU_3DNOWEXT)  
                 test_quant(quant_intra_c, quant_intra_3dne, "3dne", TEST_QUANT_INTRA, flags);  
         if (cpu_flags & XVID_CPU_SSE2)  
                 test_quant(quant_intra_c, quant_intra_sse2, "sse2", TEST_QUANT_INTRA, flags);  
 #endif  
   
         /* Inter quantization test */  
         printf("\n--- quant inter ---\n");  
                 test_quant(quant_inter_c, quant_inter_c, "c", TEST_QUANT_INTER, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_quant(quant_inter_c, quant_inter_mmx, "mmx", TEST_QUANT_INTER, flags);  
         if (cpu_flags & XVID_CPU_3DNOWEXT)  
                 test_quant(quant_inter_c, quant_inter_3dne, "3dne", TEST_QUANT_INTER, flags);  
         if (cpu_flags & XVID_CPU_SSE2)  
                 test_quant(quant_inter_c, quant_inter_sse2, "sse2", TEST_QUANT_INTER, flags);  
 #endif  
   
         /* Intra dequantization test */  
         printf("\n--- dequant intra ---\n");  
                 test_quant(dequant_intra_c, dequant_intra_c, "c", TEST_DEQUANT_INTRA, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_quant(dequant_intra_c, dequant_intra_mmx, "mmx", TEST_DEQUANT_INTRA, flags);  
         if (cpu_flags & XVID_CPU_MMXEXT)  
                 test_quant(dequant_intra_c, dequant_intra_xmm, "xmm", TEST_DEQUANT_INTRA, flags);  
         if (cpu_flags & XVID_CPU_3DNOWEXT)  
                 test_quant(dequant_intra_c, dequant_intra_3dne, "3dne", TEST_DEQUANT_INTRA, flags);  
         if (cpu_flags & XVID_CPU_SSE2)  
                 test_quant(dequant_intra_c, dequant_intra_sse2, "sse2", TEST_DEQUANT_INTRA, flags);  
 #endif  
   
         /* Inter dequantization test */  
         printf("\n--- dequant inter ---\n");  
                 test_quant(dequant_inter_c, dequant_inter_c, "c", TEST_DEQUANT_INTER, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_quant(dequant_inter_c, dequant_inter_mmx, "mmx", TEST_DEQUANT_INTER, flags);  
         if (cpu_flags & XVID_CPU_MMXEXT)  
                 test_quant(dequant_inter_c, dequant_inter_xmm, "xmm", TEST_DEQUANT_INTER, flags);  
         if (cpu_flags & XVID_CPU_3DNOWEXT)  
                 test_quant(dequant_inter_c, dequant_inter_3dne, "3dne", TEST_DEQUANT_INTER, flags);  
         if (cpu_flags & XVID_CPU_SSE2)  
                 test_quant(dequant_inter_c, dequant_inter_sse2, "sse2", TEST_DEQUANT_INTER, flags);  
 #endif  
   
         /* Intra quantization test */  
         printf("\n--- quant4 intra ---\n");  
                 test_quant(quant4_intra_c, quant4_intra_c, "c", TEST_QUANT_INTRA, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_quant(quant4_intra_c, quant4_intra_mmx, "mmx", TEST_QUANT_INTRA, flags);  
         if (cpu_flags & XVID_CPU_MMXEXT)  
                 test_quant(quant4_intra_c, quant4_intra_xmm, "xmm", TEST_QUANT_INTRA, flags);  
 #endif  
   
         /* Inter quantization test */  
         printf("\n--- quant4 inter ---\n");  
                 test_quant(quant4_inter_c, quant4_inter_c, "c", TEST_QUANT_INTER, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_quant(quant4_inter_c, quant4_inter_mmx, "mmx", TEST_QUANT_INTER, flags);  
         if (cpu_flags & XVID_CPU_MMXEXT)  
                 test_quant(quant4_inter_c, quant4_inter_xmm, "xmm", TEST_QUANT_INTER, flags);  
 #endif  
   
         /* Intra dequantization test */  
         printf("\n--- dequant4 intra ---\n");  
                 test_quant(dequant4_intra_c, dequant4_intra_c, "c", TEST_DEQUANT_INTRA, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_quant(dequant4_intra_c, dequant4_intra_mmx, "mmx", TEST_DEQUANT_INTRA, flags);  
         if (cpu_flags & XVID_CPU_3DNOWEXT)  
                 test_quant(dequant4_intra_c, dequant4_intra_3dne, "3dne", TEST_DEQUANT_INTRA, flags);  
 #endif  
   
         /* Inter dequantization test */  
         printf("\n--- dequant4 inter ---\n");  
                 test_quant(dequant4_inter_c, dequant4_inter_c, "c", TEST_DEQUANT_INTER, flags);  
   
 #if defined(ARCH_IS_IA32)  
         if (cpu_flags & XVID_CPU_MMX)  
                 test_quant(dequant4_inter_c, dequant4_inter_mmx, "mmx", TEST_DEQUANT_INTER, flags);  
         if (cpu_flags & XVID_CPU_3DNOWEXT)  
                 test_quant(dequant4_inter_c, dequant4_inter_3dne, "3dne", TEST_DEQUANT_INTER, flags);  
 #endif  
   
         emms();  
   
         return 0;  
 }  
   
   
624  /*****************************************************************************  /*****************************************************************************
625   * XviD Global Entry point   * XviD Global Entry point
626   *   *
# Line 1002  Line 649 
649                  case XVID_GBL_CONVERT :                  case XVID_GBL_CONVERT :
650                          return xvid_gbl_convert((xvid_gbl_convert_t*)param1);                          return xvid_gbl_convert((xvid_gbl_convert_t*)param1);
651    
                 case XVID_GBL_TEST :  
                 {  
                         ptr_t flags = (ptr_t)param1;  
                         return xvid_init_test((int)flags);  
                 }  
652                  default :                  default :
653                          return XVID_ERR_FAIL;                          return XVID_ERR_FAIL;
654          }          }
Legend:
Removed from v.1.45.2.8  
changed lines
  Added in v.1.49
No admin address has been configured
 ViewVC Help
Powered by ViewVC 1.0.4