[host] added multithreaded memcopy implementation

host/Capture/NvFBC.cpp

@@ -123,12 +123,21 @@ namespace Capture
     m_grabFrameParams.eGMode = NVFBC_TOSYS_SOURCEMODE_SCALE;
     m_grabFrameParams.pNvFBCFrameGrabInfo = &m_grabInfo;
 
+    if (!m_memcpy.Initialize())
+    {
+      DEBUG_ERROR("Failed to initialize MTMemcpy");
+      DeInitialize();
+      return false;
+    }
+
     m_initialized = true;
     return true;
   }
 
   void NvFBC::DeInitialize()
   {
+    m_memcpy.DeInitialize();
+
     m_frameBuffer = NULL;
 
     if (m_nvFBC)
@@ -197,7 +206,11 @@ namespace Capture
         frame.height  = m_grabInfo.dwHeight;
         frame.stride  = m_grabInfo.dwBufferWidth;
         frame.outSize = m_grabInfo.dwBufferWidth * m_grabInfo.dwHeight * 3;
-        memcpy(frame.buffer, m_frameBuffer, frame.outSize);
+        if (!m_memcpy.Copy(frame.buffer, m_frameBuffer, frame.outSize))
+        {
+          DEBUG_ERROR("Memory copy failed");
+          return false;
+        }
         return true;
       }
 

host/Capture/NvFBC.h

@@ -1,5 +1,6 @@
 #pragma once
 #include "ICapture.h"
+#include "MTMemcpy.h"
 
 #define W32_LEAN_AND_MEAN
 #include <Windows.h>
@@ -24,7 +25,8 @@ namespace Capture
 
   private:
     bool m_initialized;
-    HMODULE m_hDLL;
+    HMODULE  m_hDLL;
+    MTMemcpy m_memcpy;
 
     NvFBC_CreateFunctionExType    m_fnCreateEx;
     NvFBC_SetGlobalFlagsType      m_fnSetGlobalFlags;

host/MTMemcpy.cpp

@@ -0,0 +1,79 @@
+#include "MTMemcpy.h"
+
+MTMemcpy::MTMemcpy() :
+  m_initialized(false)
+{
+
+}
+
+MTMemcpy::~MTMemcpy()
+{
+  DeInitialize();
+}
+
+DWORD WINAPI MTMemcpy::thread_copy_proc(LPVOID param)
+{
+  mt_cpy_t * p = (mt_cpy_t *)param;
+
+  while (1)
+  {
+    WaitForSingleObject(p->s->hCopyStartSemaphores[p->ct], INFINITE);
+    memcpy(p->dest, p->src, p->size);
+    ReleaseSemaphore(p->s->hCopyStopSemaphores[p->ct], 1, NULL);
+  }
+
+  return 0;
+}
+
+bool MTMemcpy::Initialize()
+{
+  if (m_initialized)
+    DeInitialize();
+
+  for (int ctr = 0; ctr < NUM_CPY_THREADS; ctr++)
+  {
+    hCopyStartSemaphores[ctr] = CreateSemaphore(NULL, 0, 1, NULL);
+    hCopyStopSemaphores[ctr] = CreateSemaphore(NULL, 0, 1, NULL);
+    mtParamters[ctr].s  = this;
+    mtParamters[ctr].ct = ctr;
+    hCopyThreads[ctr] = CreateThread(0, 0, thread_copy_proc, &mtParamters[ctr], 0, NULL);
+  }
+
+  m_initialized = true;
+  return true;
+}
+
+bool MTMemcpy::Copy(void * dest, void * src, size_t bytes)
+{
+  if (!m_initialized)
+    return false;
+
+  //set up parameters
+  for (int ctr = 0; ctr < NUM_CPY_THREADS; ctr++)
+  {
+    mtParamters[ctr].dest = (char *)dest + ctr * bytes / NUM_CPY_THREADS;
+    mtParamters[ctr].src = (char *)src + ctr * bytes / NUM_CPY_THREADS;
+    mtParamters[ctr].size = (ctr + 1) * bytes / NUM_CPY_THREADS - ctr * bytes / NUM_CPY_THREADS;
+    ReleaseSemaphore(hCopyStartSemaphores[ctr], 1, NULL);
+  }
+
+  //wait for all threads to finish
+  WaitForMultipleObjects(NUM_CPY_THREADS, hCopyStopSemaphores, TRUE, INFINITE);
+
+  return true;
+}
+
+void MTMemcpy::DeInitialize()
+{
+  if (!m_initialized)
+    return;
+
+  for (int ctr = 0; ctr < NUM_CPY_THREADS; ctr++)
+  {
+    TerminateThread(hCopyThreads[ctr], 0);
+    CloseHandle(hCopyStartSemaphores[ctr]);
+    CloseHandle(hCopyStopSemaphores[ctr]);
+  }
+
+  m_initialized = false;
+}

host/MTMemcpy.h

@@ -0,0 +1,37 @@
+#pragma once
+
+#define W32_LEAN_AND_MEAN
+#include <Windows.h>
+
+#define NUM_CPY_THREADS 4
+
+class MTMemcpy
+{
+public:
+  bool MTMemcpy::Initialize();
+  void MTMemcpy::DeInitialize();
+  bool MTMemcpy::Copy(void * dest, void * src, size_t bytes);
+
+  MTMemcpy();
+  ~MTMemcpy();
+
+private:
+  bool m_initialized;
+  static DWORD WINAPI MTMemcpy::thread_copy_proc(LPVOID param);
+
+  typedef struct
+  {
+    MTMemcpy * s;
+    int        ct;
+    void     * src;
+    void     * dest;
+    size_t     size;
+  }
+  mt_cpy_t;
+
+  HANDLE hCopyThreads[NUM_CPY_THREADS] = { 0 };
+  HANDLE hCopyStartSemaphores[NUM_CPY_THREADS] = { 0 };
+  HANDLE hCopyStopSemaphores[NUM_CPY_THREADS] = { 0 };
+
+  mt_cpy_t mtParamters[NUM_CPY_THREADS] = { 0 };
+};

host/kvm-ivshmem-host.vcxproj

@@ -163,6 +163,7 @@
     <ClCompile Include="Capture\NvFBC.cpp" />
     <ClCompile Include="ivshmem.cpp" />
     <ClCompile Include="main.cpp" />
+    <ClCompile Include="MTMemcpy.cpp" />
     <ClCompile Include="Service.cpp" />
   </ItemGroup>
   <ItemGroup>
@@ -170,6 +171,7 @@
     <ClInclude Include="Capture\NvFBC.h" />
     <ClInclude Include="ICapture.h" />
     <ClInclude Include="ivshmem.h" />
+    <ClInclude Include="MTMemcpy.h" />
     <ClInclude Include="Service.h" />
     <ClInclude Include="Util.h" />
   </ItemGroup>

host/kvm-ivshmem-host.vcxproj.filters

@@ -33,6 +33,9 @@
     <ClCompile Include="Capture\NvFBC.cpp">
       <Filter>Source Files\Capture</Filter>
     </ClCompile>
+    <ClCompile Include="MTMemcpy.cpp">
+      <Filter>Source Files</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="ivshmem.h">
@@ -53,5 +56,8 @@
     <ClInclude Include="Util.h">
       <Filter>Header Files</Filter>
     </ClInclude>
+    <ClInclude Include="MTMemcpy.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
   </ItemGroup>
 </Project>