lee/OptiScaler
1
0
Fork 0
mirror of https://github.com/optiscaler/OptiScaler.git synced 2026-09-23 22:05:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

add mutex to preven race condutions according to https://gpuopen.com/manuals/fidelityfx_sdk/fidelityfx_sdk-page_techniques_super-resolution-interpolation/#id11

Browse Source
This commit is contained in:
cdozdil
2024-11-04 11:52:06 +03:00
parent 961f57be57
commit 17b3d0c0a4
3 changed files with 71 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
OptiScaler/NVNGX_DLSS_Dx12.cpp
+3 -1
View File
@@ -1412,7 +1412,7 @@ NVSDK_NGX_API NVSDK_NGX_Result NVSDK_NGX_D3D12_EvaluateFeature(ID3D12GraphicsCom
if (Config::Instance()->CurrentFeature != nullptr)
fgLastFGFrame = Config::Instance()->CurrentFeature->FrameCount();
auto dispatchResult = FfxApiProxy::D3D12_Dispatch()(reinterpret_cast<ffxContext*>(pUserCtx), &params->header);
LOG_DEBUG("(FG) D3D12_Dispatch result: {}", (UINT)dispatchResult);
@@ -1535,9 +1535,11 @@ NVSDK_NGX_API NVSDK_NGX_Result NVSDK_NGX_D3D12_EvaluateFeature(ID3D12GraphicsCom
dfgPrepare.viewSpaceToMetersFactor = 1.0;
dfgPrepare.frameTimeDelta = msDelta;
FrameGen_Dx12::ffxMutex.lock();
Config::Instance()->dxgiSkipSpoofing = true;
retCode = FfxApiProxy::D3D12_Dispatch()(&FrameGen_Dx12::fgContext, &dfgPrepare.header);
Config::Instance()->dxgiSkipSpoofing = false;
FrameGen_Dx12::ffxMutex.unlock();
if (retCode != FFX_API_RETURN_OK)
LOG_ERROR("(FG) D3D12_Dispatch result: {}({})", retCode, FfxApiProxy::ReturnCodeToString(retCode));
OptiScaler/hooks/HooksDx.cpp
+63 -13
View File
@@ -122,7 +122,8 @@ typedef void(*PFN_CreateUnorderedAccessView)(ID3D12Device* This, ID3D12Resource*
typedef HRESULT(*PFN_CreateDescriptorHeap)(ID3D12Device* This, D3D12_DESCRIPTOR_HEAP_DESC* pDescriptorHeapDesc, REFIID riid, void** ppvHeap);
typedef void(*PFN_CopyDescriptors)(ID3D12Device* This, UINT NumDestDescriptorRanges, D3D12_CPU_DESCRIPTOR_HANDLE* pDestDescriptorRangeStarts, UINT* pDestDescriptorRangeSizes, UINT NumSrcDescriptorRanges, D3D12_CPU_DESCRIPTOR_HANDLE* pSrcDescriptorRangeStarts, UINT* pSrcDescriptorRangeSizes, D3D12_DESCRIPTOR_HEAP_TYPE DescriptorHeapsType);
typedef void(*PFN_CopyDescriptorsSimple)(ID3D12Device* This, UINT NumDescriptors, D3D12_CPU_DESCRIPTOR_HANDLE DestDescriptorRangeStart, D3D12_CPU_DESCRIPTOR_HANDLE SrcDescriptorRangeStart, D3D12_DESCRIPTOR_HEAP_TYPE DescriptorHeapsType);
typedef UINT(*PFN_slInit)(const void* pref, uint64_t sdkVersion);
typedef HRESULT(*PFN_Present)(void* This, UINT SyncInterval, UINT Flags);
// Command list hooks for FG
typedef void(*PFN_OMSetRenderTargets)(ID3D12GraphicsCommandList* This, UINT NumRenderTargetDescriptors, D3D12_CPU_DESCRIPTOR_HANDLE* pRenderTargetDescriptors, BOOL RTsSingleHandleToDescriptorRange, D3D12_CPU_DESCRIPTOR_HANDLE* pDepthStencilDescriptor);
@@ -146,6 +147,7 @@ static PFN_CreateUnorderedAccessView o_CreateUnorderedAccessView = nullptr;
static PFN_CreateDescriptorHeap o_CreateDescriptorHeap = nullptr;
static PFN_CopyDescriptors o_CopyDescriptors = nullptr;
static PFN_CopyDescriptorsSimple o_CopyDescriptorsSimple = nullptr;
static PFN_Present o_FGSCPresent = nullptr;
// Original method calls for command list
static PFN_OMSetRenderTargets o_OMSetRenderTargets = nullptr;
@@ -182,7 +184,6 @@ static ShaderType fgSourceType = None;
// mutexes
static std::shared_mutex heapMutex;
static std::shared_mutex resourceMutex;
static std::shared_mutex ffxMutex;
static std::shared_mutex hudlessMutex[FrameGen_Dx12::FG_BUFFER_SIZE];
static std::shared_mutex counterMutex[FrameGen_Dx12::FG_BUFFER_SIZE];
@@ -233,7 +234,6 @@ static PFN_CreateDXGIFactory2 o_CreateDXGIFactory2 = nullptr;
static PFN_CreateDXGIFactory o_SL_CreateDXGIFactory = nullptr;
static PFN_CreateDXGIFactory1 o_SL_CreateDXGIFactory1 = nullptr;
static PFN_CreateDXGIFactory2 o_SL_CreateDXGIFactory2 = nullptr;
static PFN_slInit o_SL_slInit = nullptr;
inline static PFN_EnumAdapters2 ptrEnumAdapters = nullptr;
inline static PFN_EnumAdapters12 ptrEnumAdapters1 = nullptr;
@@ -519,8 +519,6 @@ static void GetHudless(ID3D12GraphicsCommandList* This)
if (Config::Instance()->CurrentFeature != nullptr)
fgLastFGFrame = Config::Instance()->CurrentFeature->FrameCount();
std::unique_lock<std::shared_mutex> lock(ffxMutex);
dispatchResult = FfxApiProxy::D3D12_Dispatch()(reinterpret_cast<ffxContext*>(pUserCtx), &params->header);
ID3D12CommandList* cl[1] = { nullptr };
result = FrameGen_Dx12::fgCopyCommandList->Close();
@@ -599,10 +597,14 @@ static void GetHudless(ID3D12GraphicsCommandList* This)
return;
}
FrameGen_Dx12::ffxMutex.lock();
Config::Instance()->dxgiSkipSpoofing = true;
retCode = FfxApiProxy::D3D12_Dispatch()(&FrameGen_Dx12::fgContext, &dfgPrepare.header);
fgDispatchCalled = true;
Config::Instance()->dxgiSkipSpoofing = false;
FrameGen_Dx12::ffxMutex.unlock();
fgDispatchCalled = true;
LOG_DEBUG("D3D12_Dispatch result: {0}, frame: {1}", retCode, frame);
}
}
@@ -1496,17 +1498,24 @@ static void hkDispatch(ID3D12GraphicsCommandList* This, UINT ThreadGroupCountX,
#pragma region Callbacks for wrapped swapchain
static HRESULT hkFGPresent(void* This, UINT SyncInterval, UINT Flags)
{
FrameGen_Dx12::ffxMutex.lock();
LOG_DEBUG("");
auto result = o_FGSCPresent(This, SyncInterval, Flags);
FrameGen_Dx12::ffxMutex.unlock();
return result;
}
static HRESULT Present(IDXGISwapChain* pSwapChain, UINT SyncInterval, UINT Flags, const DXGI_PRESENT_PARAMETERS* pPresentParameters, IUnknown* pDevice, HWND hWnd)
{
LOG_DEBUG("");
HRESULT presentResult;
auto fIndex = fgFrameIndex;
if (hWnd != Util::GetProcessWindow())
{
std::unique_lock<std::shared_mutex> lock(ffxMutex);
if (pPresentParameters == nullptr)
presentResult = pSwapChain->Present(SyncInterval, Flags);
else
@@ -1577,8 +1586,6 @@ static HRESULT Present(IDXGISwapChain* pSwapChain, UINT SyncInterval, UINT Flags
if (device12 != nullptr)
device12->Release();
std::unique_lock<std::shared_mutex> lock(ffxMutex);
if (pPresentParameters == nullptr)
presentResult = pSwapChain->Present(SyncInterval, Flags);
else
@@ -1678,8 +1685,6 @@ static HRESULT Present(IDXGISwapChain* pSwapChain, UINT SyncInterval, UINT Flags
frameCounter++;
std::unique_lock<std::shared_mutex> lock(ffxMutex);
// swapchain present
if (pPresentParameters == nullptr)
presentResult = pSwapChain->Present(SyncInterval, Flags);
@@ -1876,6 +1881,27 @@ static HRESULT hkCreateSwapChain(IDXGIFactory* pFactory, IUnknown* pDevice, DXGI
if (result == FFX_API_RETURN_OK)
{
// Hooking FG Swapchain present
// for using ffxMutex during calls
if (o_FGSCPresent == nullptr && *ppSwapChain != nullptr)
{
void** pFactoryVTable = *reinterpret_cast<void***>(*ppSwapChain);
o_FGSCPresent = (PFN_Present)pFactoryVTable[8];
if (o_FGSCPresent != nullptr)
{
LOG_INFO("Hooking native FG SwapChain present");
DetourTransactionBegin();
DetourUpdateThread(GetCurrentThread());
DetourAttach(&(PVOID&)o_FGSCPresent, hkFGPresent);
DetourTransactionCommit();
}
}
scInfo.swapChainFormat = pDesc->BufferDesc.Format;
scInfo.swapChainBufferCount = pDesc->BufferCount;
scInfo.swapChain = (IDXGISwapChain4*)*ppSwapChain;
@@ -2040,6 +2066,27 @@ static HRESULT hkCreateSwapChainForHwnd(IDXGIFactory* This, IUnknown* pDevice, H
if (result == FFX_API_RETURN_OK)
{
// Hooking FG Swapchain present
// for using ffxMutex during calls
if (o_FGSCPresent == nullptr && *ppSwapChain != nullptr)
{
void** pFactoryVTable = *reinterpret_cast<void***>(*ppSwapChain);
o_FGSCPresent = (PFN_Present)pFactoryVTable[8];
if (o_FGSCPresent != nullptr)
{
LOG_INFO("Hooking native FG SwapChain present");
DetourTransactionBegin();
DetourUpdateThread(GetCurrentThread());
DetourAttach(&(PVOID&)o_FGSCPresent, hkFGPresent);
DetourTransactionCommit();
}
}
scInfo.swapChainFormat = pDesc->Format;
scInfo.swapChainBufferCount = pDesc->BufferCount;
scInfo.swapChain = (IDXGISwapChain4*)*ppSwapChain;
@@ -2987,7 +3034,10 @@ void FrameGen_Dx12::ReleaseFGSwapchain(HWND hWnd)
if (FrameGen_Dx12::fgSwapChainContext != nullptr)
{
FrameGen_Dx12::ffxMutex.lock();
auto result = FfxApiProxy::D3D12_DestroyContext()(&FrameGen_Dx12::fgSwapChainContext, nullptr);
FrameGen_Dx12::ffxMutex.unlock();
LOG_INFO("Destroy Ffx Swapchain Result: {}({})", result, FfxApiProxy::ReturnCodeToString(result));
FrameGen_Dx12::fgSwapChainContext = nullptr;
fgSwapChains.erase(hWnd);
OptiScaler/hooks/HooksDx.h
+5 -2
View File
@@ -8,7 +8,7 @@
#include <d3d12.h>
#include <dxgi1_6.h>
//#define ENABLE_DEBUG_LAYER
#define ENABLE_DEBUG_LAYER
//#define ENABLE_GPU_VALIDATION
#ifdef ENABLE_DEBUG_LAYER
@@ -43,7 +43,6 @@ namespace HooksDx
void HookDx11();
void HookDx12();
void HookDxgi();
void HookSLDxgi();
}
namespace FrameGen_Dx12
@@ -71,6 +70,10 @@ namespace FrameGen_Dx12
inline FT_Dx12* fgFormatTransfer = nullptr;
inline bool fgIsActive = false;
// According to https://gpuopen.com/manuals/fidelityfx_sdk/fidelityfx_sdk-page_techniques_super-resolution-interpolation/#id11
// will use this mutex to prevent race condutions
inline std::mutex ffxMutex;
UINT ClearFrameResources();
UINT GetFrame();
void NewFrame();