summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorandreykproductengine <andreykproductengine@lindenlab.com>2019-10-16 22:36:10 +0300
committerandreykproductengine <andreykproductengine@lindenlab.com>2019-10-16 22:36:10 +0300
commit94d4364084329cc6d16af7a126148117ad13555a (patch)
tree61c7d216b6c9e22e3f149d357ff45042145f7a21
parent63a3d9b4d99d244481dbd33c184d3ff8039ca7ad (diff)
SL-12103 More reliable memory detection
-rw-r--r--indra/llrender/llgl.cpp83
-rw-r--r--indra/llrender/llgl.h1
-rw-r--r--indra/llrender/llglheaders.h2
-rw-r--r--indra/llwindow/lldxhardware.cpp70
-rw-r--r--indra/llwindow/lldxhardware.h4
-rw-r--r--indra/newview/llviewertexturelist.cpp7
6 files changed, 144 insertions, 23 deletions
diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index c0f0cec80b..bcc702d31f 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -49,6 +49,10 @@
#include "llglheaders.h"
#include "llglslshader.h"
+#if LL_WINDOWS
+#include "lldxhardware.h"
+#endif
+
#ifdef _DEBUG
//#define GL_STATE_VERIFY
#endif
@@ -394,6 +398,8 @@ PFNGLGETACTIVEATTRIBARBPROC glGetActiveAttribARB = NULL;
PFNGLGETATTRIBLOCATIONARBPROC glGetAttribLocationARB = NULL;
#if LL_WINDOWS
+PFNWGLGETGPUIDSAMDPROC wglGetGPUIDsAMD = NULL;
+PFNWGLGETGPUINFOAMDPROC wglGetGPUInfoAMD = NULL;
PFNWGLSWAPINTERVALEXTPROC wglSwapIntervalEXT = NULL;
#endif
@@ -413,6 +419,7 @@ LLGLManager::LLGLManager() :
mHasMultitexture(FALSE),
mHasATIMemInfo(FALSE),
+ mHasAMDAssociations(FALSE),
mHasNVXMemInfo(FALSE),
mNumTextureUnits(1),
mHasMipMapGeneration(FALSE),
@@ -497,7 +504,16 @@ void LLGLManager::initWGL()
{
LL_WARNS("RenderInit") << "No ARB create context extensions" << LL_ENDL;
}
-
+
+ // For retreiving information per AMD adapter,
+ // because we can't trust curently selected/default one when there are multiple
+ mHasAMDAssociations = ExtensionExists("WGL_AMD_gpu_association", gGLHExts.mSysExts);
+ if (mHasAMDAssociations)
+ {
+ GLH_EXT_NAME(wglGetGPUIDsAMD) = (PFNWGLGETGPUIDSAMDPROC)GLH_EXT_GET_PROC_ADDRESS("wglGetGPUIDsAMD");
+ GLH_EXT_NAME(wglGetGPUInfoAMD) = (PFNWGLGETGPUINFOAMDPROC)GLH_EXT_GET_PROC_ADDRESS("wglGetGPUInfoAMD");
+ }
+
if (ExtensionExists("WGL_EXT_swap_control", gGLHExts.mSysExts))
{
GLH_EXT_NAME(wglSwapIntervalEXT) = (PFNWGLSWAPINTERVALEXTPROC)GLH_EXT_GET_PROC_ADDRESS("wglSwapIntervalEXT");
@@ -683,23 +699,78 @@ bool LLGLManager::initGL()
stop_glerror();
S32 old_vram = mVRAM;
+ mVRAM = 0;
- if (mHasATIMemInfo)
+#if LL_WINDOWS
+ if (mHasAMDAssociations)
+ {
+ GLuint gl_gpus_count = wglGetGPUIDsAMD(0, 0);
+ if (gl_gpus_count > 0)
+ {
+ GLuint* ids = new GLuint[gl_gpus_count];
+ wglGetGPUIDsAMD(gl_gpus_count, ids);
+
+ GLuint mem_mb = 0;
+ for (U32 i = 0; i < gl_gpus_count; i++)
+ {
+ wglGetGPUInfoAMD(ids[i],
+ WGL_GPU_RAM_AMD,
+ GL_UNSIGNED_INT,
+ sizeof(GLuint),
+ &mem_mb);
+ if (mVRAM < mem_mb)
+ {
+ // basically pick the best AMD and trust driver/OS to know to switch
+ mVRAM = mem_mb;
+ }
+ }
+ }
+ if (mVRAM != 0)
+ {
+ LL_WARNS("RenderInit") << "VRAM Detected (AMDAssociations):" << mVRAM << LL_ENDL;
+ }
+ }
+#endif
+
+ if (mHasATIMemInfo && mVRAM == 0)
{ //ask the gl how much vram is free at startup and attempt to use no more than half of that
S32 meminfo[4];
glGetIntegerv(GL_TEXTURE_FREE_MEMORY_ATI, meminfo);
- mVRAM = meminfo[0]/1024;
+ mVRAM = meminfo[0] / 1024;
+ LL_WARNS("RenderInit") << "VRAM Detected (ATIMemInfo):" << mVRAM << LL_ENDL;
}
- else if (mHasNVXMemInfo)
+
+ if (mHasNVXMemInfo && mVRAM == 0)
{
S32 dedicated_memory;
glGetIntegerv(GL_GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX, &dedicated_memory);
mVRAM = dedicated_memory/1024;
+ LL_WARNS("RenderInit") << "VRAM Detected (NVXMemInfo):" << mVRAM << LL_ENDL;
}
+#if LL_WINDOWS
if (mVRAM < 256)
- { //something likely went wrong using the above extensions, fall back to old method
+ {
+ // Something likely went wrong using the above extensions
+ // try WMI first and fall back to old method (from dxdiag) if all else fails
+ // Function will check all GPUs WMI knows of and will pick up the one with most
+ // memory. We need to check all GPUs because system can switch active GPU to
+ // weaker one, to preserve power when not under load.
+ S32 mem = LLDXHardware::getMBVideoMemoryViaWMI();
+ if (mem != 0)
+ {
+ mVRAM = mem;
+ LL_WARNS("RenderInit") << "VRAM Detected (WMI):" << mVRAM<< LL_ENDL;
+ }
+ }
+#endif
+
+ if (mVRAM < 256 && old_vram > 0)
+ {
+ // fall back to old method
+ // Note: on Windows value will be from LLDXHardware.
+ // Either received via dxdiag or via WMI by id from dxdiag.
mVRAM = old_vram;
}
@@ -961,7 +1032,7 @@ void LLGLManager::initExtensions()
mHasTextureRectangle = FALSE;
#else // LL_MESA_HEADLESS //important, gGLHExts.mSysExts is uninitialized until after glh_init_extensions is called
mHasMultitexture = glh_init_extensions("GL_ARB_multitexture");
- mHasATIMemInfo = ExtensionExists("GL_ATI_meminfo", gGLHExts.mSysExts);
+ mHasATIMemInfo = ExtensionExists("GL_ATI_meminfo", gGLHExts.mSysExts); //Basic AMD method, also see mHasAMDAssociations
mHasNVXMemInfo = ExtensionExists("GL_NVX_gpu_memory_info", gGLHExts.mSysExts);
mHasSeparateSpecularColor = glh_init_extensions("GL_EXT_separate_specular_color");
mHasAnisotropic = glh_init_extensions("GL_EXT_texture_filter_anisotropic");
diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h
index 4c4302d05b..6db1fe9c90 100644
--- a/indra/llrender/llgl.h
+++ b/indra/llrender/llgl.h
@@ -78,6 +78,7 @@ public:
// Extensions used by everyone
BOOL mHasMultitexture;
BOOL mHasATIMemInfo;
+ BOOL mHasAMDAssociations;
BOOL mHasNVXMemInfo;
S32 mNumTextureUnits;
BOOL mHasMipMapGeneration;
diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h
index 722dd9050b..36fbb381bb 100644
--- a/indra/llrender/llglheaders.h
+++ b/indra/llrender/llglheaders.h
@@ -618,6 +618,8 @@ extern PFNGLVARIANTARRAYOBJECTATIPROC glVariantObjectArrayATI;
extern PFNGLGETVARIANTARRAYOBJECTFVATIPROC glGetVariantArrayObjectfvATI;
extern PFNGLGETVARIANTARRAYOBJECTIVATIPROC glGetVariantArrayObjectivATI;
+extern PFNWGLGETGPUIDSAMDPROC wglGetGPUIDsAMD;
+extern PFNWGLGETGPUINFOAMDPROC wglGetGPUInfoAMD;
extern PFNWGLSWAPINTERVALEXTPROC wglSwapIntervalEXT;
// GL_ARB_occlusion_query
diff --git a/indra/llwindow/lldxhardware.cpp b/indra/llwindow/lldxhardware.cpp
index 9397b831f7..12a6baa3e6 100644
--- a/indra/llwindow/lldxhardware.cpp
+++ b/indra/llwindow/lldxhardware.cpp
@@ -61,7 +61,7 @@ typedef BOOL ( WINAPI* PfnCoSetProxyBlanket )( IUnknown* pProxy, DWORD dwAuthnSv
OLECHAR* pServerPrincName, DWORD dwAuthnLevel, DWORD dwImpLevel,
RPC_AUTH_IDENTITY_HANDLE pAuthInfo, DWORD dwCapabilities );
-HRESULT GetVideoMemoryViaWMI( WCHAR* strInputDeviceID, DWORD* pdwAdapterRam )
+HRESULT GetVideoMemoryViaWMI(WCHAR* strInputDeviceID, DWORD* pdwAdapterRam)
{
HRESULT hr;
bool bGotMemory = false;
@@ -149,21 +149,26 @@ HRESULT GetVideoMemoryViaWMI( WCHAR* strInputDeviceID, DWORD* pdwAdapterRam )
if ( !pVideoControllers[iController] )
continue;
- pPropName = SysAllocString( L"PNPDeviceID" );
- hr = pVideoControllers[iController]->Get( pPropName, 0L, &var, nullptr, nullptr );
+ // if strInputDeviceID is set find this specific device and return memory or specific device
+ // if strInputDeviceID is not set return the best device
+ if (strInputDeviceID)
+ {
+ pPropName = SysAllocString( L"PNPDeviceID" );
+ hr = pVideoControllers[iController]->Get( pPropName, 0L, &var, nullptr, nullptr );
#ifdef PRINTF_DEBUGGING
- if( FAILED( hr ) )
- wprintf( L"WMI: pVideoControllers[iController]->Get PNPDeviceID failed: 0x%0.8x\n", hr );
+ if( FAILED( hr ) )
+ wprintf( L"WMI: pVideoControllers[iController]->Get PNPDeviceID failed: 0x%0.8x\n", hr );
#endif
- if( SUCCEEDED( hr ) )
- {
- if( wcsstr( var.bstrVal, strInputDeviceID ) != 0 )
- bFound = true;
+ if( SUCCEEDED( hr ) && strInputDeviceID)
+ {
+ if( wcsstr( var.bstrVal, strInputDeviceID ) != 0 )
+ bFound = true;
+ }
+ VariantClear( &var );
+ if( pPropName ) SysFreeString( pPropName );
}
- VariantClear( &var );
- if( pPropName ) SysFreeString( pPropName );
- if( bFound )
+ if( bFound || !strInputDeviceID )
{
pPropName = SysAllocString( L"AdapterRAM" );
hr = pVideoControllers[iController]->Get( pPropName, 0L, &var, nullptr, nullptr );
@@ -175,13 +180,18 @@ HRESULT GetVideoMemoryViaWMI( WCHAR* strInputDeviceID, DWORD* pdwAdapterRam )
if( SUCCEEDED( hr ) )
{
bGotMemory = true;
- *pdwAdapterRam = var.ulVal;
+ *pdwAdapterRam = llmax(var.ulVal, *pdwAdapterRam);
}
VariantClear( &var );
if( pPropName ) SysFreeString( pPropName );
- break;
}
+
SAFE_RELEASE( pVideoControllers[iController] );
+
+ if (bFound)
+ {
+ break;
+ }
}
}
}
@@ -207,6 +217,17 @@ HRESULT GetVideoMemoryViaWMI( WCHAR* strInputDeviceID, DWORD* pdwAdapterRam )
return E_FAIL;
}
+//static
+S32 LLDXHardware::getMBVideoMemoryViaWMI()
+{
+ DWORD vram = 0;
+ if (SUCCEEDED(GetVideoMemoryViaWMI(NULL, &vram)))
+ {
+ return vram / (1024 * 1024);;
+ }
+ return 0;
+}
+
//Getting the version of graphics controller driver via WMI
std::string LLDXHardware::getDriverVersionWMI()
{
@@ -615,6 +636,8 @@ BOOL LLDXHardware::getInfo(BOOL vram_only)
IDxDiagContainer *driver_containerp = NULL;
DWORD dw_device_count;
+ mVRAM = 0;
+
// CoCreate a IDxDiagProvider*
LL_DEBUGS("AppInit") << "CoCreateInstance IID_IDxDiagProvider" << LL_ENDL;
hr = CoCreateInstance(CLSID_DxDiagProvider,
@@ -677,6 +700,8 @@ BOOL LLDXHardware::getInfo(BOOL vram_only)
}
// Get device 0
+ // By default 0 device is the primary one, howhever in case of various hybrid graphics
+ // like itegrated AMD and PCI AMD GPUs system might switch.
LL_DEBUGS("AppInit") << "devices_containerp->GetChildContainer" << LL_ENDL;
hr = devices_containerp->GetChildContainer(L"0", &device_containerp);
if(FAILED(hr) || !device_containerp)
@@ -689,12 +714,27 @@ BOOL LLDXHardware::getInfo(BOOL vram_only)
WCHAR deviceID[512];
get_wstring(device_containerp, L"szDeviceID", deviceID, 512);
-
+ // Example: searches id like 1F06 in pnp string (aka VEN_10DE&DEV_1F06)
+ // doesn't seem to work on some systems since format is unrecognizable
+ // but in such case keyDeviceID works
if (SUCCEEDED(GetVideoMemoryViaWMI(deviceID, &vram)))
{
mVRAM = vram/(1024*1024);
}
else
+ {
+ get_wstring(device_containerp, L"szKeyDeviceID", deviceID, 512);
+ LL_WARNS() << "szDeviceID" << deviceID << LL_ENDL;
+ // '+9' to avoid ENUM\\PCI\\ prefix
+ // Returns string like Enum\\PCI\\VEN_10DE&DEV_1F06&SUBSYS...
+ // and since GetVideoMemoryViaWMI searches by PNPDeviceID it is sufficient
+ if (SUCCEEDED(GetVideoMemoryViaWMI(deviceID + 9, &vram)))
+ {
+ mVRAM = vram / (1024 * 1024);
+ }
+ }
+
+ if (mVRAM == 0)
{ // Get the English VRAM string
std::string ram_str = get_string(device_containerp, L"szDisplayMemoryEnglish");
diff --git a/indra/llwindow/lldxhardware.h b/indra/llwindow/lldxhardware.h
index cf33db8b37..1cb687e3b6 100644
--- a/indra/llwindow/lldxhardware.h
+++ b/indra/llwindow/lldxhardware.h
@@ -94,6 +94,10 @@ public:
LLSD getDisplayInfo();
+ // Will get memory of best GPU in MB, return memory on sucsess, 0 on failure
+ // Note: WMI is not accurate in some cases
+ static S32 getMBVideoMemoryViaWMI();
+
// Find a particular device that matches the following specs.
// Empty strings indicate that you don't care.
// You can separate multiple devices with '|' chars to indicate you want
diff --git a/indra/newview/llviewertexturelist.cpp b/indra/newview/llviewertexturelist.cpp
index 06524847d1..561319ca5d 100644
--- a/indra/newview/llviewertexturelist.cpp
+++ b/indra/newview/llviewertexturelist.cpp
@@ -1385,8 +1385,6 @@ S32Megabytes LLViewerTextureList::getMaxVideoRamSetting(bool get_recommended, fl
{
max_texmem = (S32Megabytes)128;
}
-
- LL_WARNS() << "VRAM amount not detected, defaulting to " << max_texmem << " MB" << LL_ENDL;
}
S32Megabytes system_ram = gSysMemory.getPhysicalMemoryKB(); // In MB
@@ -1428,6 +1426,11 @@ void LLViewerTextureList::updateMaxResidentTexMem(S32Megabytes mem)
return; //listener will re-enter this function
}
+ if (gGLManager.mVRAM == 0)
+ {
+ LL_WARNS() << "VRAM amount not detected, defaulting to " << mem << " MB" << LL_ENDL;
+ }
+
// TODO: set available resident texture mem based on use by other subsystems
// currently max(12MB, VRAM/4) assumed...