diff options
author | Fernando Sahmkow <[email protected]> | 2023-06-28 19:32:50 +0200 |
---|---|---|
committer | Fernando Sahmkow <[email protected]> | 2023-06-28 21:32:45 +0200 |
commit | da440da9f54cc860f3c69da685a415d5ec9d7b64 (patch) | |
tree | 5a7a4a56462244970e1356a723e6a8a77477f820 /src/core/memory.cpp | |
parent | 47d0d292d5cc5f0404e126023279db7decd532ac (diff) | |
download | yuzu-mainline-da440da9f54cc860f3c69da685a415d5ec9d7b64.tar.gz yuzu-mainline-da440da9f54cc860f3c69da685a415d5ec9d7b64.zip |
Memory Tracking: Optimize tracking to only use atomic writes when contested with the host GPU
Diffstat (limited to 'src/core/memory.cpp')
-rw-r--r-- | src/core/memory.cpp | 39 |
1 files changed, 33 insertions, 6 deletions
diff --git a/src/core/memory.cpp b/src/core/memory.cpp index 60b246bdd..257406f09 100644 --- a/src/core/memory.cpp +++ b/src/core/memory.cpp @@ -3,6 +3,7 @@ #include <algorithm> #include <cstring> +#include <span> #include "common/assert.h" #include "common/atomic_ops.h" @@ -679,7 +680,7 @@ struct Memory::Impl { LOG_ERROR(HW_Memory, "Unmapped Write{} @ 0x{:016X} = 0x{:016X}", sizeof(T) * 8, GetInteger(vaddr), static_cast<u64>(data)); }, - [&]() { system.CurrentGPUDirtyMemoryManager().Collect(GetInteger(vaddr), sizeof(T)); }); + [&]() { HandleRasterizerWrite(GetInteger(vaddr), sizeof(T)); }); if (ptr) { std::memcpy(ptr, &data, sizeof(T)); } @@ -693,7 +694,7 @@ struct Memory::Impl { LOG_ERROR(HW_Memory, "Unmapped WriteExclusive{} @ 0x{:016X} = 0x{:016X}", sizeof(T) * 8, GetInteger(vaddr), static_cast<u64>(data)); }, - [&]() { system.CurrentGPUDirtyMemoryManager().Collect(GetInteger(vaddr), sizeof(T)); }); + [&]() { HandleRasterizerWrite(GetInteger(vaddr), sizeof(T)); }); if (ptr) { const auto volatile_pointer = reinterpret_cast<volatile T*>(ptr); return Common::AtomicCompareAndSwap(volatile_pointer, data, expected); @@ -708,7 +709,7 @@ struct Memory::Impl { LOG_ERROR(HW_Memory, "Unmapped WriteExclusive128 @ 0x{:016X} = 0x{:016X}{:016X}", GetInteger(vaddr), static_cast<u64>(data[1]), static_cast<u64>(data[0])); }, - [&]() { system.CurrentGPUDirtyMemoryManager().Collect(GetInteger(vaddr), sizeof(u128)); }); + [&]() { HandleRasterizerWrite(GetInteger(vaddr), sizeof(u128)); }); if (ptr) { const auto volatile_pointer = reinterpret_cast<volatile u64*>(ptr); return Common::AtomicCompareAndSwap(volatile_pointer, data, expected); @@ -718,7 +719,7 @@ struct Memory::Impl { void HandleRasterizerDownload(VAddr address, size_t size) { const size_t core = system.GetCurrentHostThreadID(); - auto& current_area = rasterizer_areas[core]; + auto& current_area = rasterizer_read_areas[core]; const VAddr end_address = address + size; if (current_area.start_address <= address && end_address <= current_area.end_address) [[likely]] { @@ -727,9 +728,31 @@ struct Memory::Impl { current_area = system.GPU().OnCPURead(address, size); } - Common::PageTable* current_page_table = nullptr; - std::array<VideoCore::RasterizerDownloadArea, Core::Hardware::NUM_CPU_CORES> rasterizer_areas{}; + void HandleRasterizerWrite(VAddr address, size_t size) { + const size_t core = system.GetCurrentHostThreadID(); + auto& current_area = rasterizer_write_areas[core]; + VAddr subaddress = address >> YUZU_PAGEBITS; + bool do_collection = current_area.last_address == subaddress; + if (!do_collection) [[unlikely]] { + do_collection = system.GPU().OnCPUWrite(address, size); + if (!do_collection) { + return; + } + current_area.last_address = subaddress; + } + gpu_dirty_managers[core].Collect(address, size); + } + + struct GPUDirtyState { + VAddr last_address; + }; + Core::System& system; + Common::PageTable* current_page_table = nullptr; + std::array<VideoCore::RasterizerDownloadArea, Core::Hardware::NUM_CPU_CORES> + rasterizer_read_areas{}; + std::array<GPUDirtyState, Core::Hardware::NUM_CPU_CORES> rasterizer_write_areas{}; + std::span<Core::GPUDirtyMemoryManager> gpu_dirty_managers; }; Memory::Memory(Core::System& system_) : system{system_} { @@ -877,6 +900,10 @@ void Memory::ZeroBlock(Common::ProcessAddress dest_addr, const std::size_t size) impl->ZeroBlock(*system.ApplicationProcess(), dest_addr, size); } +void Memory::SetGPUDirtyManagers(std::span<Core::GPUDirtyMemoryManager> managers) { + impl->gpu_dirty_managers = managers; +} + Result Memory::InvalidateDataCache(Common::ProcessAddress dest_addr, const std::size_t size) { return impl->InvalidateDataCache(*system.ApplicationProcess(), dest_addr, size); } |