[x265] [PATCH 3 of 7] threadpool.cpp: utilize all available NUMA nodes for threadpool distribution -> windows system fix

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari <praveen at multicorewareinc.com>
# Date 1463738737 -19800
#      Fri May 20 15:35:37 2016 +0530
# Node ID ab1b27a1712d581c32b007f0f72e482f4a83905d
# Parent  e988eee2f0dc962b9b94c8cef6f739522bce9afb
threadpool.cpp: utilize all available NUMA nodes for threadpool distribution -> windows system fix

diff -r e988eee2f0dc -r ab1b27a1712d source/common/threadpool.cpp
--- a/source/common/threadpool.cpp	Fri May 20 15:35:37 2016 +0530
+++ b/source/common/threadpool.cpp	Fri May 20 15:35:37 2016 +0530
@@ -27,6 +27,7 @@
 #include "threading.h"
 
 #include <new>
+#include <winnt.h>
 
 #if X86_64
 
@@ -405,16 +406,15 @@
     X265_CHECK(numThreads <= MAX_POOL_THREADS, "a single thread pool cannot have more than MAX_POOL_THREADS threads\n");
 
 #if defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7 
-    m_winCpuMask = 0x0;
-    GROUP_AFFINITY groupAffinity;
+    memset(&m_groupAffinity, 0, sizeof(GROUP_AFFINITY));
     for (int i = 0; i < getNumaNodeCount(); i++)
     {
         int numaNode = ((nodeMask >> i) & 0x1U) ? i : -1;
         if (numaNode != -1)
-            if (GetNumaNodeProcessorMaskEx((USHORT)numaNode, &groupAffinity))
-                m_winCpuMask |= groupAffinity.Mask;
+        if (GetNumaNodeProcessorMaskEx((USHORT)numaNode, &m_groupAffinity))
+            break;
     }
-    m_numaMask = &m_winCpuMask;
+    m_numaMask = &m_groupAffinity.Mask;
 #elif HAVE_LIBNUMA
     if (numa_available() >= 0)
     {
@@ -496,11 +496,16 @@
     setThreadNodeAffinity(m_numaMask);
 }
 
-/* static */
 void ThreadPool::setThreadNodeAffinity(void *numaMask)
 {
 #if defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7 
-    if (SetThreadAffinityMask(GetCurrentThread(), *((DWORD_PTR*)numaMask)))
+    UNREFERENCED_PARAMETER(numaMask);
+    GROUP_AFFINITY groupAffinity;
+    memset(&groupAffinity, 0, sizeof(GROUP_AFFINITY));
+    groupAffinity.Group = m_groupAffinity.Group;
+    groupAffinity.Mask = m_groupAffinity.Mask;
+    const PGROUP_AFFINITY affinityPointer = &groupAffinity;
+    if (SetThreadGroupAffinity(GetCurrentThread(), affinityPointer, NULL))
         return;
     else
         x265_log(NULL, X265_LOG_ERROR, "unable to set thread affinity for NUMA node mask\n");
diff -r e988eee2f0dc -r ab1b27a1712d source/common/threadpool.h
--- a/source/common/threadpool.h	Fri May 20 15:35:37 2016 +0530
+++ b/source/common/threadpool.h	Fri May 20 15:35:37 2016 +0530
@@ -85,7 +85,7 @@
     int           m_numWorkers;
     void*         m_numaMask; // node mask in linux, cpu mask in windows
 #if defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7 
-    DWORD_PTR     m_winCpuMask;
+    GROUP_AFFINITY m_groupAffinity;
 #endif
     bool          m_isActive;
 
@@ -99,6 +99,7 @@
     bool start();
     void stopWorkers();
     void setCurrentThreadAffinity();
+    void setThreadNodeAffinity(void *numaMask);
     int  tryAcquireSleepingThread(sleepbitmap_t firstTryBitmap, sleepbitmap_t secondTryBitmap);
     int  tryBondPeers(int maxPeers, sleepbitmap_t peerBitmap, BondedTaskGroup& master);
 
@@ -106,7 +107,6 @@
 
     static int  getCpuCount();
     static int  getNumaNodeCount();
-    static void setThreadNodeAffinity(void *numaMask);
 };
 
 /* Any worker thread may enlist the help of idle worker threads from the same