[x265] [PATCH] Changed FrameEncoder::m_tld to a pointer and set it to one of Encoder's ThreadLocalData instances

[dtyx265 at gmail.com]

# HG changeset patch
# User David T Yuen <dtyx265 at gmail.com>
# Date 1411589843 25200
# Node ID fc82666c3f8fe258e99ffbb8398ae04fd90a4bee
# Parent  b2b7072ddbf73085d457bd6a71bca946e505dea8
Changed FrameEncoder::m_tld to a pointer and set it to one of Encoder's ThreadLocalData instances.

This uses less memory since m_tld isn't used in --wpp and Encoder's ThreadLocalData instances are not used in --no-wpp
Also there was a small performance increase on my system

diff -r b2b7072ddbf7 -r fc82666c3f8f source/encoder/encoder.cpp
--- a/source/encoder/encoder.cpp	Wed Sep 24 11:48:15 2014 +0530
+++ b/source/encoder/encoder.cpp	Wed Sep 24 13:17:23 2014 -0700
@@ -102,15 +102,20 @@
     m_scalingList.setupQuantMatrices();
 
     /* Allocate thread local data shared by all frame encoders */
-    ThreadPool *pool = ThreadPool::getThreadPool();
-    const int poolThreadCount = pool ? pool->getThreadCount() : 1;
-    m_threadLocalData = new ThreadLocalData[poolThreadCount];
-    for (int i = 0; i < poolThreadCount; i++)
+    const int poolThreadCount = ThreadPool::getThreadPool()->getThreadCount();
+    int numLocalData = m_param->frameNumThreads;
+    if (m_param->bEnableWavefront)
+        numLocalData = poolThreadCount;
+    m_threadLocalData = new ThreadLocalData[numLocalData];
+    for (int i = 0; i < numLocalData; i++)
     {
         m_threadLocalData[i].cuCoder.initSearch(m_param, m_scalingList);
         m_threadLocalData[i].cuCoder.create(g_maxCUDepth + 1, g_maxCUSize);
     }
 
+    for (int i = 0; i < m_param->frameNumThreads; i++)
+        m_frameEncoder[i].m_tld = &m_threadLocalData[i];
+
     m_lookahead = new Lookahead(m_param, m_threadPool, this);
     m_dpb = new DPB(m_param);
     m_rateControl = new RateControl(m_param);
diff -r b2b7072ddbf7 -r fc82666c3f8f source/encoder/frameencoder.cpp
--- a/source/encoder/frameencoder.cpp	Wed Sep 24 11:48:15 2014 +0530
+++ b/source/encoder/frameencoder.cpp	Wed Sep 24 13:17:23 2014 -0700
@@ -105,9 +105,6 @@
         m_pool = NULL;
     }
 
-    m_tld.cuCoder.initSearch(m_param, top->m_scalingList);
-    m_tld.cuCoder.create(g_maxCUDepth + 1, g_maxCUSize);
-
     m_frameFilter.init(top, this, numRows);
 
     // initialize HRD parameters of SPS
@@ -470,7 +467,7 @@
             }
         }
 
-        m_tld.cuCoder.loadCTUData(cu);
+        m_tld->cuCoder.loadCTUData(cu);
         // final coding (bitstream generation) for this CU
         m_entropyCoder.encodeCTU(cu);
 
@@ -581,8 +578,8 @@
     const int realRow = row >> 1;
     const int typeNum = row & 1;
 
-    ThreadLocalData& tld = threadId >= 0 ? m_top->m_threadLocalData[threadId] : m_tld;
-
+    ThreadLocalData& tld = threadId >= 0 ? m_top->m_threadLocalData[threadId] : *m_tld;
+    
     if (!typeNum)
         processRowEncoder(realRow, tld);
     else
diff -r b2b7072ddbf7 -r fc82666c3f8f source/encoder/frameencoder.h
--- a/source/encoder/frameencoder.h	Wed Sep 24 11:48:15 2014 +0530
+++ b/source/encoder/frameencoder.h	Wed Sep 24 13:17:23 2014 -0700
@@ -159,7 +159,7 @@
     uint32_t*                m_substreamSizes;
     NoiseReduction*          m_nr;
     NALList                  m_nalList;
-    ThreadLocalData          m_tld; /* for --no-wpp */
+    ThreadLocalData*         m_tld; /* for --no-wpp */
 
     int                      m_filterRowDelay;
     int                      m_filterRowDelayCus;