Changes between Initial Version and Version 1 of Ticket #14, comment 2


Ignore:
Timestamp:
20 Feb 2016, 04:14:31 (8 years ago)
Author:
ksuehring
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #14, comment 2

    initial v1  
    22
    33Basically, what I reported seem to be two separate problems.
    4 (1)     The attached valgrind memory check report is generated from unmodified JEM1.0 by using command
     41. The attached valgrind memory check report is generated from unmodified JEM1.0 by using command
    55
     6{{{
    67valgrind --leak-check=yes  ../bin/TAppEncoderStaticd S15BQSquare.cfg -c encoder_intra_jvet10_LCU16.cfg -q 33
     8}}}
    79
     102. As to the problem of windows and linux mismatch, it happened to the code I modified which I forced intra coding not split. For JEM1.0, I don’t have an example to repeat the mismatch just using JEM1.0.   I think the bug can be seen by examining the code.
    811
    9 (2)     As to the problem of windows and linux mismatch, it happened to the code I modified which I forced intra coding not split. For JEM1.0, I don’t have an example to repeat the mismatch just using JEM1.0.   I think the bug can be seen by examining the code.
     12 Basically TEncSearch.cpp Line #1850 CABAC state is only stored if bCheckSplit is true.
    1013
    11         Basically TEncSearch.cpp Line #1850 CABAC state is only stored if bCheckSplit is true.
    12 
     14{{{
    1315  if( bCheckFull )
    1416  {     
     
    2022          ….
    2123  }
     24}}}
    2225
    23 But it is loaded several times for bfilter loop and emtloop. So if bCheckSplit is false, the CABAC state loaded will be uninitialized.
     26 But it is loaded several times for bfilter loop and emtloop. So if bCheckSplit is false, the CABAC state loaded will be uninitialized.
    2427
    25 If you use attached config file which try to limit LCU to 16x16 and with minimum split, step into it using visual studio, the first time the code stop at line # 1850, you will see the CABAC state loaded from CI_QT_TRAFO_ROOT have some weird uninitialized numbers.
    26 I think this bug can be fixed by eliminating the if(bCheckSplit) before store(m_pppcRDSbacCoder[ uiFullDepth ][ CI_QT_TRAFO_ROOT ]
     28 If you use attached config file which try to limit LCU to 16x16 and with minimum split, step into it using visual studio, the first time the code stop at line # 1850, you will see the CABAC state loaded from CI_QT_TRAFO_ROOT have some weird uninitialized numbers.
     29 I think this bug can be fixed by eliminating the if(bCheckSplit) before store(m_pppcRDSbacCoder[ uiFullDepth ][ CI_QT_TRAFO_ROOT ]
    2730
    2831Best.