[x265] [PATCH 1 of 7] threadpool.cpp: get correct CPU count for multisocket machines -> windows system fix

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari <praveen at multicorewareinc.com>
# Date 1463655478 -19800
#      Thu May 19 16:27:58 2016 +0530
# Node ID 9a6ab28b736e1167ac26977d7da8ab2d23cc296f
# Parent  aca781339b4c8dae94ff7da73f18cd4439757e87
threadpool.cpp: get correct CPU count for multisocket machines -> windows system fix

diff -r aca781339b4c -r 9a6ab28b736e source/common/threadpool.cpp
--- a/source/common/threadpool.cpp	Tue May 10 15:33:17 2016 +0530
+++ b/source/common/threadpool.cpp	Thu May 19 16:27:58 2016 +0530
@@ -64,6 +64,19 @@
 # define strcasecmp _stricmp
 #endif
 
+const uint64_t m1 = 0x5555555555555555; //binary: 0101...
+const uint64_t m2 = 0x3333333333333333; //binary: 00110011..
+const uint64_t m3 = 0x0f0f0f0f0f0f0f0f; //binary:  4 zeros,  4 ones ...
+const uint64_t h01 = 0x0101010101010101; //the sum of 256 to the power of 0,1,2,3...
+
+int popCount(uint64_t x)
+{
+    x -= (x >> 1) & m1;
+    x = (x & m2) + ((x >> 2) & m2);
+    x = (x + (x >> 4)) & m3;
+    return (x * h01) >> 56;
+}
+
 namespace X265_NS {
 // x265 private namespace
 
@@ -525,9 +538,17 @@
 int ThreadPool::getCpuCount()
 {
 #if _WIN32
-    SYSTEM_INFO sysinfo;
-    GetSystemInfo(&sysinfo);
-    return sysinfo.dwNumberOfProcessors;
+    enum { MAX_NODE_NUM = 127 };
+    int cpus = 0;
+    int numNumaNodes = X265_MIN(getNumaNodeCount(), MAX_NODE_NUM);
+    PGROUP_AFFINITY groupAffinityPointer = new GROUP_AFFINITY;
+    for (int i = 0; i < numNumaNodes; i++)
+    {
+        GetNumaNodeProcessorMaskEx((UCHAR)i, groupAffinityPointer);
+        cpus += popCount(groupAffinityPointer->Mask);
+    }
+    delete groupAffinityPointer;
+    return cpus;
 #elif __unix__ && X265_ARCH_ARM
     /* Return the number of processors configured by OS. Because, most embedded linux distributions
      * uses only one processor as the scheduler doesn't have enough work to utilize all processors */