From 3d643fbc0bbb80f17ed14cfcdfdba5f6d4fd5e83 Mon Sep 17 00:00:00 2001 From: Vincent Lejeune Date: Sat, 19 Sep 2015 19:17:23 +0200 Subject: [PATCH] d3d12: Factorise texture cache management. --- rpcs3/Emu/RSX/D3D12/D3D12GSRender.cpp | 56 ++++----------- rpcs3/Emu/RSX/D3D12/D3D12GSRender.h | 99 +++++++++++++++++++++++---- rpcs3/Emu/RSX/D3D12/D3D12Texture.cpp | 27 +++----- 3 files changed, 110 insertions(+), 72 deletions(-) diff --git a/rpcs3/Emu/RSX/D3D12/D3D12GSRender.cpp b/rpcs3/Emu/RSX/D3D12/D3D12GSRender.cpp index f8b10e4971..12d382e556 100644 --- a/rpcs3/Emu/RSX/D3D12/D3D12GSRender.cpp +++ b/rpcs3/Emu/RSX/D3D12/D3D12GSRender.cpp @@ -93,8 +93,7 @@ void D3D12GSRender::ResourceStorage::WaitAndClean() Reset(); - for (auto tmp : m_dirtyTextures) - tmp->Release(); + m_dirtyTextures.clear(); m_RAMFramebuffer = nullptr; } @@ -119,30 +118,11 @@ void D3D12GSRender::Shader::Release() extern std::function gfxHandler; -bool D3D12GSRender::invalidateTexture(u32 addr) +bool D3D12GSRender::invalidateAddress(u32 addr) { - bool handled = false; - auto It = m_protectedTextures.begin(), E = m_protectedTextures.end(); - for (; It != E;) - { - auto currentIt = It; - ++It; - auto protectedTexture = *currentIt; - u32 protectedRangeStart = std::get<1>(protectedTexture), protectedRangeSize = std::get<2>(protectedTexture); - if (addr - protectedRangeStart < protectedRangeSize) - { - std::lock_guard lock(mut); - u32 texadrr = std::get<0>(protectedTexture); - ID3D12Resource *texToErase = m_texturesCache[texadrr]; - m_texturesCache.erase(texadrr); - m_texToClean.push_back(texToErase); - - vm::page_protect(protectedRangeStart, protectedRangeSize, 0, vm::page_writable, 0); - m_protectedTextures.erase(currentIt); - handled = true; - } - } - return handled; + bool result = false; + result |= m_textureCache.invalidateAddress(addr); + return result; } D3D12DLLManagement::D3D12DLLManagement() @@ -159,7 +139,7 @@ D3D12GSRender::D3D12GSRender() : GSRender(), m_D3D12Lib(), m_PSO(nullptr) { gfxHandler = [this](u32 addr) { - bool result = invalidateTexture(addr); + bool result = invalidateAddress(addr); if (result) LOG_WARNING(RSX, "Reporting Cell writing to %x", addr); return result; @@ -305,12 +285,7 @@ D3D12GSRender::~D3D12GSRender() CloseHandle(handle); { - std::lock_guard lock(mut); - for (auto &protectedTexture : m_protectedTextures) - { - u32 protectedRangeStart = std::get<1>(protectedTexture), protectedRangeSize = std::get<2>(protectedTexture); - vm::page_protect(protectedRangeStart, protectedRangeSize, 0, vm::page_writable, 0); - } + m_textureCache.unprotedAll(); } gfxHandler = [this](u32) { return false; }; @@ -326,10 +301,6 @@ D3D12GSRender::~D3D12GSRender() m_perFrameStorage[0].Release(); m_perFrameStorage[1].Release(); m_rtts.Release(); - for (auto &tmp : m_texToClean) - tmp->Release(); - for (auto &tmp : m_texturesCache) - tmp.second->Release(); m_outputScalingPass.Release(); ReleaseD2DStructures(); @@ -851,9 +822,8 @@ void D3D12GSRender::Flip() storage.m_frameFinishedFence->SetEventOnCompletion(storage.m_fenceValue, storage.m_frameFinishedHandle); storage.m_fenceValue++; - storage.m_dirtyTextures = m_texToClean; + storage.m_dirtyTextures = m_textureCache.getCurrentDisposableTexture(); storage.m_inUse = true; - m_texToClean.clear(); // Get the put pos - 1. This way after cleaning we can set the get ptr to // this value, allowing heap to proceed even if we cleant before allocating @@ -1072,7 +1042,7 @@ void D3D12GSRender::semaphorePGRAPHBackendRelease(u32 offset, u32 value) getCurrentResourceStorage().m_commandList->CopyTextureRegion(&CD3DX12_TEXTURE_COPY_LOCATION(writeDest.Get(), { 0, { DXGI_FORMAT_R8_UNORM, m_surface_clip_w, m_surface_clip_h, 1, (UINT)depthRowPitch } }), 0, 0, 0, &CD3DX12_TEXTURE_COPY_LOCATION(depthConverted.Get(), 0), nullptr); - invalidateTexture(GetAddress(m_surface_offset_z, m_context_dma_z - 0xfeed0000)); + invalidateAddress(GetAddress(m_surface_offset_z, m_context_dma_z - 0xfeed0000)); } ID3D12Resource *rtt0, *rtt1, *rtt2, *rtt3; @@ -1110,10 +1080,10 @@ void D3D12GSRender::semaphorePGRAPHBackendRelease(u32 offset, u32 value) break; } - if (m_context_dma_color_a) invalidateTexture(GetAddress(m_surface_offset_a, m_context_dma_color_a - 0xfeed0000)); - if (m_context_dma_color_b) invalidateTexture(GetAddress(m_surface_offset_b, m_context_dma_color_b - 0xfeed0000)); - if (m_context_dma_color_c) invalidateTexture(GetAddress(m_surface_offset_c, m_context_dma_color_c - 0xfeed0000)); - if (m_context_dma_color_d) invalidateTexture(GetAddress(m_surface_offset_d, m_context_dma_color_d - 0xfeed0000)); + if (m_context_dma_color_a) invalidateAddress(GetAddress(m_surface_offset_a, m_context_dma_color_a - 0xfeed0000)); + if (m_context_dma_color_b) invalidateAddress(GetAddress(m_surface_offset_b, m_context_dma_color_b - 0xfeed0000)); + if (m_context_dma_color_c) invalidateAddress(GetAddress(m_surface_offset_c, m_context_dma_color_c - 0xfeed0000)); + if (m_context_dma_color_d) invalidateAddress(GetAddress(m_surface_offset_d, m_context_dma_color_d - 0xfeed0000)); } if (needTransfer) { diff --git a/rpcs3/Emu/RSX/D3D12/D3D12GSRender.h b/rpcs3/Emu/RSX/D3D12/D3D12GSRender.h index f0a11c1d14..a515813b9b 100644 --- a/rpcs3/Emu/RSX/D3D12/D3D12GSRender.h +++ b/rpcs3/Emu/RSX/D3D12/D3D12GSRender.h @@ -190,6 +190,87 @@ struct DataHeap } }; +/** + * Manages cache of data (texture/vertex/index) + */ +struct DataCache +{ +private: + /** + * Mutex protecting m_dataCache access + * Memory protection fault catch can be generated by any thread and + * modifies it. + */ + std::mutex mut; + + std::unordered_map > m_dataCache; // Storage + std::list > m_protectedRange; // address, start of protected range, size of protected range + std::list > m_dataToDispose; +public: + void storeAndProtectData(u64 key, u32 start, size_t size, ComPtr data) + { + std::lock_guard lock(mut); + m_dataCache[key] = data; + /// align start to 4096 byte + u32 protected_range_start = align(start, 4096); + u32 protected_range_size = (u32)align(size, 4096); + m_protectedRange.push_back(std::make_tuple(key, protected_range_start, protected_range_size)); + vm::page_protect(protected_range_start, protected_range_size, 0, 0, vm::page_writable); + } + + /// remove all data containing addr from cache, unprotect them. Returns false if no data is modified. + bool invalidateAddress(u32 addr) + { + bool handled = false; + auto It = m_protectedRange.begin(), E = m_protectedRange.end(); + for (; It != E;) + { + auto currentIt = It; + ++It; + auto protectedTexture = *currentIt; + u32 protectedRangeStart = std::get<1>(protectedTexture), protectedRangeSize = std::get<2>(protectedTexture); + if (addr >= protectedRangeStart && addr <= protectedRangeSize + protectedRangeStart) + { + std::lock_guard lock(mut); + u64 texadrr = std::get<0>(protectedTexture); + m_dataToDispose.push_back(m_dataCache[texadrr]); + m_dataCache.erase(texadrr); + + vm::page_protect(protectedRangeStart, protectedRangeSize, 0, vm::page_writable, 0); + m_protectedRange.erase(currentIt); + handled = true; + } + } + return handled; + } + + ID3D12Resource *findDataIfAvailable(u64 key) + { + std::lock_guard lock(mut); + std::unordered_map >::const_iterator It = m_dataCache.find(key); + if (It == m_dataCache.end()) + return nullptr; + return It->second.Get(); + } + + void unprotedAll() + { + std::lock_guard lock(mut); + for (auto &protectedTexture : m_protectedRange) + { + u32 protectedRangeStart = std::get<1>(protectedTexture), protectedRangeSize = std::get<2>(protectedTexture); + vm::page_protect(protectedRangeStart, protectedRangeSize, 0, vm::page_writable, 0); + } + } + + std::list > getCurrentDisposableTexture() + { + std::list > result; + result.swap(m_dataToDispose); + return result; + } +}; + /** * Structure used to load/unload D3D12 lib. */ @@ -213,22 +294,14 @@ private: ComPtr m_backbufferAsRendertarget[2]; // m_rootSignatures[N] is RS with N texture/sample ComPtr m_rootSignatures[17]; - /** - * Mutex protecting m_texturesCache and m_Textoclean access - * Memory protection fault catch can be generated by any thread and - * modifies these two members. - */ - std::mutex mut; - std::list > m_protectedTextures; // Texaddress, start of protected range, size of protected range - std::vector m_texToClean; - bool invalidateTexture(u32 addr); + + // TODO: Use a tree structure to parse more efficiently + DataCache m_textureCache; + bool invalidateAddress(u32 addr); // Copy of RTT to be used as texture std::unordered_map m_texturesRTTs; - std::unordered_map m_texturesCache; - // std::vector m_post_draw_objs; - PipelineStateObjectCache m_cachePSO; std::pair *m_PSO; @@ -311,7 +384,7 @@ private: /// Texture that were invalidated - std::vector m_dirtyTextures; + std::list > m_dirtyTextures; size_t m_getPosConstantsHeap; size_t m_getPosVertexIndexHeap; diff --git a/rpcs3/Emu/RSX/D3D12/D3D12Texture.cpp b/rpcs3/Emu/RSX/D3D12/D3D12Texture.cpp index 5c4f6e14c6..fc37b25557 100644 --- a/rpcs3/Emu/RSX/D3D12/D3D12Texture.cpp +++ b/rpcs3/Emu/RSX/D3D12/D3D12Texture.cpp @@ -126,13 +126,13 @@ D3D12_SAMPLER_DESC getSamplerDesc(const RSXTexture &texture) * using a temporary texture buffer. */ static -ID3D12Resource *uploadSingleTexture( +ComPtr uploadSingleTexture( const RSXTexture &texture, ID3D12Device *device, ID3D12GraphicsCommandList *commandList, DataHeap &textureBuffersHeap) { - ID3D12Resource *vramTexture; + ComPtr vramTexture; size_t w = texture.GetWidth(), h = texture.GetHeight(); int format = texture.GetFormat() & ~(CELL_GCM_TEXTURE_LN | CELL_GCM_TEXTURE_UN); @@ -157,18 +157,18 @@ ID3D12Resource *uploadSingleTexture( &texturedesc, D3D12_RESOURCE_STATE_COPY_DEST, nullptr, - IID_PPV_ARGS(&vramTexture) + IID_PPV_ARGS(vramTexture.GetAddressOf()) )); size_t miplevel = 0; for (const MipmapLevelInfo mli : mipInfos) { - commandList->CopyTextureRegion(&CD3DX12_TEXTURE_COPY_LOCATION(vramTexture, (UINT)miplevel), 0, 0, 0, + commandList->CopyTextureRegion(&CD3DX12_TEXTURE_COPY_LOCATION(vramTexture.Get(), (UINT)miplevel), 0, 0, 0, &CD3DX12_TEXTURE_COPY_LOCATION(textureBuffersHeap.m_heap, { heapOffset + mli.offset, { dxgiFormat, (UINT)mli.width, (UINT)mli.height, 1, (UINT)mli.rowPitch } }), nullptr); miplevel++; } - commandList->ResourceBarrier(1, &CD3DX12_RESOURCE_BARRIER::Transition(vramTexture, D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_GENERIC_READ)); + commandList->ResourceBarrier(1, &CD3DX12_RESOURCE_BARRIER::Transition(vramTexture.Get(), D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_GENERIC_READ)); return vramTexture; } @@ -246,7 +246,6 @@ size_t getTextureSize(const RSXTexture &texture) size_t D3D12GSRender::UploadTextures(ID3D12GraphicsCommandList *cmdlist) { - std::lock_guard lock(mut); size_t usedTexture = 0; for (u32 i = 0; i < m_textures_count; ++i) @@ -263,26 +262,22 @@ size_t D3D12GSRender::UploadTextures(ID3D12GraphicsCommandList *cmdlist) ID3D12Resource *vramTexture; std::unordered_map::const_iterator ItRTT = m_rtts.m_renderTargets.find(texaddr); - std::unordered_map::const_iterator ItCache = m_texturesCache.find(texaddr); + ID3D12Resource *cachedTex = m_textureCache.findDataIfAvailable(texaddr); bool isRenderTarget = false; if (ItRTT != m_rtts.m_renderTargets.end()) { vramTexture = ItRTT->second; isRenderTarget = true; } - else if (ItCache != m_texturesCache.end()) + else if (cachedTex != nullptr) { - vramTexture = ItCache->second; + vramTexture = cachedTex; } else { - vramTexture = uploadSingleTexture(m_textures[i], m_device.Get(), cmdlist, m_textureUploadData); - m_texturesCache[texaddr] = vramTexture; - - u32 s = (u32)align(getTextureSize(m_textures[i]), 4096); - LOG_WARNING(RSX, "PROTECTING %x of size %d", align(texaddr, 4096), s); - m_protectedTextures.push_back(std::make_tuple(texaddr, align(texaddr, 4096), s)); - vm::page_protect(align(texaddr, 4096), s, 0, 0, vm::page_writable); + ComPtr tex = uploadSingleTexture(m_textures[i], m_device.Get(), cmdlist, m_textureUploadData); + vramTexture = tex.Get(); + m_textureCache.storeAndProtectData(texaddr, texaddr, getTextureSize(m_textures[i]), tex); } D3D12_SHADER_RESOURCE_VIEW_DESC srvDesc = {};