[x265] [PATCH] get same output between single and multi threading

[Min Chen]

# HG changeset patch
# User Min Chen <chenm003 at 163.com>
# Date 1380521904 -28800
# Node ID cd4d431e292d0935c154daf0917dbb6816cdbca8
# Parent  55edc34e253c14d3eccb83a7d1db43774349ff9a
get same output between single and multi threading

Output mistake reasons:

1. CABAC Table Initialize
   The HM have decide best table for CABAC, but in frame parallelism,
   we can't get this information before thread start, we have to
   disable it now.
Side effect: maybe lost some compress performance

2. SAO Global Disable
   The HM decide SAO global disable flag depend previous same type of
   slice. but here, we can get right statistics informat before start,
   so we have to disable it.
Side effect: more computer cost since we always try SAO on every LCU

3. CABAC status m_frac not reset.
    This HM bug still alive, we found more here.

diff -r 55edc34e253c -r cd4d431e292d source/Lib/TLibEncoder/TEncTop.cpp
--- a/source/Lib/TLibEncoder/TEncTop.cpp	Sat Sep 28 22:54:44 2013 -0500
+++ b/source/Lib/TLibEncoder/TEncTop.cpp	Mon Sep 30 14:18:24 2013 +0800
@@ -823,7 +823,7 @@
     pps->setDeblockingFilterOverrideEnabledFlag(!m_loopFilterOffsetInPPS);
     pps->setPicDisableDeblockingFilterFlag(!param.bEnableLoopFilter);
     pps->setLog2ParallelMergeLevelMinus2(m_log2ParallelMergeLevelMinus2);
-    pps->setCabacInitPresentFlag(CABAC_INIT_PRESENT_FLAG);
+    pps->setCabacInitPresentFlag(param.frameNumThreads > 1 ? 0 : CABAC_INIT_PRESENT_FLAG);
 
     pps->setNumRefIdxL0DefaultActive(1);
     pps->setNumRefIdxL1DefaultActive(1);
diff -r 55edc34e253c -r cd4d431e292d source/encoder/cturow.h
--- a/source/encoder/cturow.h	Sat Sep 28 22:54:44 2013 -0500
+++ b/source/encoder/cturow.h	Mon Sep 30 14:18:24 2013 +0800
@@ -63,9 +63,22 @@
 
     void destroy();
 
-    void init()
+    void init(TComSlice *slice)
     {
         m_active = 0;
+
+        // Note: Reset status to avoid frame parallelism output mistake on different thread number
+        for (UInt depth = 0; depth < g_maxCUDepth + 1; depth++)
+        {
+            for (int ciIdx = 0; ciIdx < CI_NUM; ciIdx++)
+            {
+                m_rdSbacCoders[depth][ciIdx]->setSlice(slice);
+                m_rdSbacCoders[depth][ciIdx]->resetEntropy();
+                m_binCodersCABAC[depth][ciIdx]->m_fracBits = 0;
+            }
+        }
+        m_rdGoOnSbacCoder.setSlice(slice);
+        m_rdGoOnSbacCoder.resetEntropy();
     }
 
     void processCU(TComDataCU *cu, TComSlice *slice, TEncSbac *bufferSBac, bool bSaveCabac);
diff -r 55edc34e253c -r cd4d431e292d source/encoder/frameencoder.cpp
--- a/source/encoder/frameencoder.cpp	Sat Sep 28 22:54:44 2013 -0500
+++ b/source/encoder/frameencoder.cpp	Mon Sep 30 14:18:24 2013 +0800
@@ -877,12 +877,14 @@
 void FrameEncoder::compressCTURows()
 {
     PPAScopeEvent(FrameEncoder_compressRows);
+    TComSlice* slice = m_pic->getSlice();
+
     // reset entropy coders
     m_sbacCoder.init(&m_binCoderCABAC);
     for (int i = 0; i < this->m_numRows; i++)
     {
-        m_rows[i].init();
-        m_rows[i].m_entropyCoder.setEntropyCoder(&m_sbacCoder, m_pic->getSlice());
+        m_rows[i].init(slice);
+        m_rows[i].m_entropyCoder.setEntropyCoder(&m_sbacCoder, slice);
         m_rows[i].m_entropyCoder.resetEntropy();
 
         m_rows[i].m_rdSbacCoders[0][CI_CURR_BEST]->load(&m_sbacCoder);
@@ -891,7 +893,6 @@
     }
 
     UInt refLagRows = ((m_cfg->param.searchRange + NTAPS_LUMA/2 + g_maxCUHeight - 1) / g_maxCUHeight) + 1;
-    TComSlice* slice = m_pic->getSlice();
     int numPredDir = slice->isInterP() ? 1 : slice->isInterB() ? 2 : 0;
 
     m_frameFilter.start(m_pic);
diff -r 55edc34e253c -r cd4d431e292d source/encoder/framefilter.cpp
--- a/source/encoder/framefilter.cpp	Sat Sep 28 22:54:44 2013 -0500
+++ b/source/encoder/framefilter.cpp	Mon Sep 30 14:18:24 2013 +0800
@@ -98,6 +98,13 @@
         SAOParam* saoParam = pic->getPicSym()->getSaoParam();
         m_sao.resetSAOParam(saoParam);
         m_sao.rdoSaoUnitRowInit(saoParam);
+
+        // NOTE: Disable SAO automatic turn off when Frame Parallelism for output exact
+        if (m_cfg->param.frameNumThreads)
+        {
+            saoParam->bSaoFlag[0] = true;
+            saoParam->bSaoFlag[1] = true;
+        }
     }
 }