SL-17005 WIP - Use D3D/DXGI to query for available VRAM on Windows

10 files changed, 153 insertions, 66 deletions

diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index c171f7051f..0a5d258b3a 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -1615,51 +1615,6 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_
     return TRUE;
 }
 
-void LLImageGLThread::updateClass()
-{
-    LL_PROFILE_ZONE_SCOPED;
-
-    // update available vram one per second
-    static LLFrameTimer sTimer;
-
-    if (sTimer.getElapsedSeconds() < 1.f)
-    {
-        return;
-    }
-    
-    sTimer.reset();
-
-    auto func = []()
-    {
-        if (gGLManager.mHasATIMemInfo)
-        {
-            S32 meminfo[4];
-            glGetIntegerv(GL_TEXTURE_FREE_MEMORY_ATI, meminfo);
-            LLImageGLThread::sFreeVRAMMegabytes = meminfo[0];
-
-        }
-        else if (gGLManager.mHasNVXMemInfo)
-        {
-            S32 free_memory;
-            glGetIntegerv(GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX, &free_memory);
-            LLImageGLThread::sFreeVRAMMegabytes = free_memory / 1024;
-        }
-    };
-
-    
-    // post update to background thread if available, otherwise execute immediately
-    auto queue = LL::WorkQueue::getInstance("LLImageGL");
-    if (sEnabled)
-    {
-        queue->post(func);
-    }
-    else
-    {
-        llassert(queue == nullptr);
-        func();
-    }
-}
-
 void LLImageGL::syncToMainThread(LLGLuint new_tex_name)
 {
     LL_PROFILE_ZONE_SCOPED;
@@ -2389,8 +2344,6 @@ void LLImageGL::checkActiveThread()
 		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL,  nummips);
 */  
 
-std::atomic<S32> LLImageGLThread::sFreeVRAMMegabytes(4096); //if free vram is unknown, default to 4GB
-
 LLImageGLThread::LLImageGLThread(LLWindow* window)
     // We want exactly one thread, but a very large capacity: we never want
     // anyone, especially inner-loop render code, to have to block on post()
@@ -2418,8 +2371,3 @@ void LLImageGLThread::run()
     mWindow->destroySharedContext(mContext);
 }
 
-S32 LLImageGLThread::getFreeVRAMMegabytes()
-{
-    return sFreeVRAMMegabytes;
-}
-
diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h
index ae4ab7c9c3..1716338c0b 100644
--- a/indra/llrender/llimagegl.h
+++ b/indra/llrender/llimagegl.h
@@ -321,12 +321,6 @@ public:
     // follows gSavedSettings "RenderGLMultiThreaded"
     static bool sEnabled;
     
-    // app should call this function periodically
-    static void updateClass();
-
-    // free video memory in megabytes
-    static std::atomic<S32> sFreeVRAMMegabytes;
-
     LLImageGLThread(LLWindow* window);
 
     // post a function to be executed on the LLImageGL background thread
@@ -338,8 +332,6 @@ public:
 
     void run() override;
 
-    static S32 getFreeVRAMMegabytes();
-
 private:
     LLWindow* mWindow;
     void* mContext = nullptr;
diff --git a/indra/llwindow/CMakeLists.txt b/indra/llwindow/CMakeLists.txt
index 70eb99c86c..32f0fa14c4 100644
--- a/indra/llwindow/CMakeLists.txt
+++ b/indra/llwindow/CMakeLists.txt
@@ -145,6 +145,8 @@ if (WINDOWS)
   list(APPEND llwindow_LINK_LIBRARIES
        comdlg32     # Common Dialogs for ChooseColor
        ole32
+       dxgi
+       d3d9
        )
 endif (WINDOWS)
 
diff --git a/indra/llwindow/llwindow.h b/indra/llwindow/llwindow.h
index 0edf39f6ef..507e6076d6 100644
--- a/indra/llwindow/llwindow.h
+++ b/indra/llwindow/llwindow.h
@@ -163,6 +163,9 @@ public:
 	virtual F32 getPixelAspectRatio() = 0;
 	virtual void setNativeAspectRatio(F32 aspect) = 0;
 	
+    // query VRAM usage
+    virtual U32 getAvailableVRAMMegabytes() = 0;
+
 	virtual void beforeDialog() {};	// prepare to put up an OS dialog (if special measures are required, such as in fullscreen mode)
 	virtual void afterDialog() {};	// undo whatever was done in beforeDialog()
 
diff --git a/indra/llwindow/llwindowheadless.h b/indra/llwindow/llwindowheadless.h
index 410da79623..0e1f603db7 100644
--- a/indra/llwindow/llwindowheadless.h
+++ b/indra/llwindow/llwindowheadless.h
@@ -96,6 +96,8 @@ public:
 	/*virtual*/ F32 getPixelAspectRatio() { return 1.0f; };
 	/*virtual*/ void setNativeAspectRatio(F32 ratio) {}
 
+    U32 getAvailableVRAMMegabytes() override { return 4096; }
+
 	/*virtual*/ void *getPlatformWindow() { return 0; };
 	/*virtual*/ void bringToFront() {};
 	
diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp
index 9feccd7874..3afb78d1d2 100644
--- a/indra/llwindow/llwindowwin32.cpp
+++ b/indra/llwindow/llwindowwin32.cpp
@@ -47,6 +47,7 @@
 #include "llglslshader.h"
 #include "llthreadsafequeue.h"
 #include "stringize.h"
+#include "llframetimer.h"
 
 // System includes
 #include <commdlg.h>
@@ -61,6 +62,10 @@
 #include <sstream>
 #include <utility>                  // std::pair
 
+#include <d3d9.h>
+#include <dxgi1_4.h>
+
+
 // Require DirectInput version 8
 #define DIRECTINPUT_VERSION 0x0800
 
@@ -347,6 +352,20 @@ struct LLWindowWin32::LLWindowWin32Thread : public LL::ThreadPool
 
     void run() override;
 
+    // initialzie DXGI adapter (for querying available VRAM)
+    void initDX();
+    
+    // initialize D3D (if DXGI cannot be used)
+    void initD3D();
+
+    // call periodically to update available VRAM
+    void updateVRAMUsage();
+
+    U32 getAvailableVRAMMegabytes()
+    {
+        return mAvailableVRAM;
+    }
+
     /// called by main thread to post work to this window thread
     template <typename CALLABLE>
     void post(CALLABLE&& func)
@@ -395,6 +414,12 @@ struct LLWindowWin32::LLWindowWin32Thread : public LL::ThreadPool
     void gatherInput();
     HWND mWindowHandle = NULL;
     HDC mhDC = 0;
+
+    std::atomic<U32> mAvailableVRAM;
+
+    IDXGIAdapter3* mDXGIAdapter = nullptr;
+    LPDIRECT3D9 mD3D = nullptr;
+    LPDIRECT3DDEVICE9 mD3DDevice = nullptr;
 };
 
 
@@ -4507,6 +4532,10 @@ std::vector<std::string> LLWindowWin32::getDynamicFallbackFontList()
 	return std::vector<std::string>();
 }
 
+U32 LLWindowWin32::getAvailableVRAMMegabytes()
+{
+    return mWindowThread ? mWindowThread->getAvailableVRAMMegabytes() : 0;
+}
 
 #endif // LL_WINDOWS
 
@@ -4562,17 +4591,97 @@ private:
     std::string mPrev;
 };
 
+void LLWindowWin32::LLWindowWin32Thread::initDX()
+{
+    if (mDXGIAdapter == NULL)
+    {
+        IDXGIFactory4* pFactory = nullptr;
+
+        HRESULT res = CreateDXGIFactory1(__uuidof(IDXGIFactory4), (void**)&pFactory);
+
+        if (FAILED(res))
+        {
+            LL_WARNS() << "CreateDXGIFactory1 failed: 0x" << std::hex << res << LL_ENDL;
+        }
+        else
+        {
+            res = pFactory->EnumAdapters(0, reinterpret_cast<IDXGIAdapter**>(&mDXGIAdapter));
+            if (FAILED(res))
+            {
+                LL_WARNS() << "EnumAdapters failed: 0x" << std::hex << res << LL_ENDL;
+            }
+        }
+
+        pFactory->Release();
+    }
+}
+
+void LLWindowWin32::LLWindowWin32Thread::initD3D()
+{
+    if (mDXGIAdapter == NULL && mD3DDevice == NULL && mWindowHandle != 0)
+    {
+        mD3D = Direct3DCreate9(D3D_SDK_VERSION);
+        
+        D3DPRESENT_PARAMETERS d3dpp;
+
+        ZeroMemory(&d3dpp, sizeof(d3dpp));
+        d3dpp.Windowed = TRUE;
+        d3dpp.SwapEffect = D3DSWAPEFFECT_DISCARD;
+
+        mD3D->CreateDevice(D3DADAPTER_DEFAULT, D3DDEVTYPE_HAL, mWindowHandle, D3DCREATE_SOFTWARE_VERTEXPROCESSING, &d3dpp, &mD3DDevice);
+    }
+}
+
+void LLWindowWin32::LLWindowWin32Thread::updateVRAMUsage()
+{
+    LL_PROFILE_ZONE_SCOPED;
+    if (mDXGIAdapter != nullptr)
+    {
+        DXGI_QUERY_VIDEO_MEMORY_INFO info;
+        mDXGIAdapter->QueryVideoMemoryInfo(0, DXGI_MEMORY_SEGMENT_GROUP_LOCAL, &info);
+
+        // try to use no more than the available reserve minus 10%
+        U32 target = info.AvailableForReservation / 1024 / 1024;
+        target -= target / 10;
+
+        U32 used_vram = info.CurrentUsage / 1024 / 1024;
+
+        mAvailableVRAM = used_vram < target ? target - used_vram : 0;
+
+        /*LL_INFOS() << "\nLocal\nAFR: " << info.AvailableForReservation / 1024 / 1024
+            << "\nBudget: " << info.Budget / 1024 / 1024
+            << "\nCR: " << info.CurrentReservation / 1024 / 1024
+            << "\nCU: " << info.CurrentUsage / 1024 / 1024 << LL_ENDL;
+
+        mDXGIAdapter->QueryVideoMemoryInfo(0, DXGI_MEMORY_SEGMENT_GROUP_NON_LOCAL, &info);
+        LL_INFOS() << "\nNon-Local\nAFR: " << info.AvailableForReservation / 1024 / 1024
+            << "\nBudget: " << info.Budget / 1024 / 1024
+            << "\nCR: " << info.CurrentReservation / 1024 / 1024
+            << "\nCU: " << info.CurrentUsage / 1024 / 1024 << LL_ENDL;*/
+    }
+    else if (mD3DDevice != NULL)
+    { // fallback to D3D9
+        mAvailableVRAM = mD3DDevice->GetAvailableTextureMem() / 1024 / 1024;
+    }
+
+}
+
 void LLWindowWin32::LLWindowWin32Thread::run()
 {
     sWindowThreadId = std::this_thread::get_id();
     LogChange logger("Window");
 
+    initDX();
+
     while (! getQueue().done())
     {
         LL_PROFILE_ZONE_SCOPED_CATEGORY_WIN32;
 
         if (mWindowHandle != 0)
         {
+            // lazily call initD3D inside this loop to catch when mWindowHandle has been set
+            initD3D();
+
             MSG msg;
             BOOL status;
             if (mhDC == 0)
@@ -4605,6 +4714,13 @@ void LLWindowWin32::LLWindowWin32Thread::run()
             getQueue().runPending();
         }
         
+        // update available vram once every 3 seconds
+        static LLFrameTimer vramTimer;
+        if (vramTimer.getElapsedTimeF32() > 3.f)
+        {
+            updateVRAMUsage();
+            vramTimer.reset();
+        }
 #if 0
         {
             LL_PROFILE_ZONE_NAMED_CATEGORY_WIN32("w32t - Sleep");
@@ -4613,6 +4729,26 @@ void LLWindowWin32::LLWindowWin32Thread::run()
         }
 #endif
     }
+
+    //clean up DXGI/D3D resources
+    if (mDXGIAdapter)
+    {
+        mDXGIAdapter->Release();
+        mDXGIAdapter = nullptr;
+    }
+
+    if (mD3DDevice)
+    {
+        mD3DDevice->Release();
+        mD3DDevice = nullptr;
+    }
+
+    if (mD3D)
+    {
+        mD3D->Release();
+        mD3D = nullptr;
+    }
+
 }
 
 void LLWindowWin32::post(const std::function<void()>& func)
diff --git a/indra/llwindow/llwindowwin32.h b/indra/llwindow/llwindowwin32.h
index b391acc12d..bd53b3e92a 100644
--- a/indra/llwindow/llwindowwin32.h
+++ b/indra/llwindow/llwindowwin32.h
@@ -108,7 +108,9 @@ public:
 	/*virtual*/ F32 getPixelAspectRatio();
 	/*virtual*/ void setNativeAspectRatio(F32 ratio) { mOverrideAspectRatio = ratio; }
 
-	/*virtual*/	BOOL dialogColorPicker(F32 *r, F32 *g, F32 *b );
+    U32 getAvailableVRAMMegabytes() override;
+	
+    /*virtual*/	BOOL dialogColorPicker(F32 *r, F32 *g, F32 *b );
 
 	/*virtual*/ void *getPlatformWindow();
 	/*virtual*/ void bringToFront();
diff --git a/indra/newview/lltextureview.cpp b/indra/newview/lltextureview.cpp
index b58070526c..99ab854293 100644
--- a/indra/newview/lltextureview.cpp
+++ b/indra/newview/lltextureview.cpp
@@ -49,6 +49,8 @@
 #include "llviewerobjectlist.h"
 #include "llviewertexture.h"
 #include "llviewertexturelist.h"
+#include "llviewerwindow.h"
+#include "llwindow.h"
 #include "llvovolume.h"
 #include "llviewerstats.h"
 #include "llworld.h"
@@ -545,8 +547,8 @@ void LLGLTexMemBar::draw()
     U32 texFetchLatMed = U32(recording.getMean(LLTextureFetch::sTexFetchLatency).value() * 1000.0f);
     U32 texFetchLatMax = U32(recording.getMax(LLTextureFetch::sTexFetchLatency).value() * 1000.0f);
 
-	text = llformat("GL Free: %d Sys Free: %d MB FBO: %d MB Bias: %.2f Cache: %.1f/%.1f MB",
-                    LLImageGLThread::getFreeVRAMMegabytes(),
+    text = llformat("GL Free: %d Sys Free: %d MB FBO: %d MB Bias: %.2f Cache: %.1f/%.1f MB",
+                    gViewerWindow->getWindow()->getAvailableVRAMMegabytes(),
                     LLMemory::getAvailableMemKB()/1024,
 					LLRenderTarget::sBytesAllocated/(1024*1024),
 					discard_bias,
diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp
index 6a2b06d9b5..4fc1bdbec3 100644
--- a/indra/newview/llviewerdisplay.cpp
+++ b/indra/newview/llviewerdisplay.cpp
@@ -788,8 +788,6 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot)
 				LLViewerTexture::updateClass();
 			}
 
-            LLImageGLThread::updateClass();
-
 			{
 				LL_RECORD_BLOCK_TIME(FTM_IMAGE_UPDATE_BUMP);
 				gBumpImageList.updateImages();  // must be called before gTextureList version so that it's textures are thrown out first.
diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp
index be7ee3660e..ac6c86d9c6 100644
--- a/indra/newview/llviewertexture.cpp
+++ b/indra/newview/llviewertexture.cpp
@@ -58,6 +58,8 @@
 #include "llvovolume.h"
 #include "llviewermedia.h"
 #include "lltexturecache.h"
+#include "llviewerwindow.h"
+#include "llwindow.h"
 ///////////////////////////////////////////////////////////////////////////////
 
 // extern
@@ -523,7 +525,7 @@ void LLViewerTexture::getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &p
     timer.reset();
 
     {
-        gpu_res = (S32Megabytes) LLImageGLThread::getFreeVRAMMegabytes();
+        gpu_res = (S32Megabytes)gViewerWindow->getWindow()->getAvailableVRAMMegabytes();
         
         //check main memory, only works for windows.
         LLMemory::updateMemoryInfo();