diff options
Diffstat (limited to 'src/video_core')
56 files changed, 3682 insertions, 803 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 44c761d3e..114bed20d 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -106,6 +106,8 @@ add_library(video_core STATIC textures/decoders.cpp textures/decoders.h textures/texture.h + texture_cache.cpp + texture_cache.h video_core.cpp video_core.h ) @@ -127,16 +129,21 @@ if (ENABLE_VULKAN) renderer_vulkan/vk_sampler_cache.h renderer_vulkan/vk_scheduler.cpp renderer_vulkan/vk_scheduler.h + renderer_vulkan/vk_shader_decompiler.cpp + renderer_vulkan/vk_shader_decompiler.h renderer_vulkan/vk_stream_buffer.cpp renderer_vulkan/vk_stream_buffer.h renderer_vulkan/vk_swapchain.cpp renderer_vulkan/vk_swapchain.h) - target_include_directories(video_core PRIVATE ../../externals/Vulkan-Headers/include) + target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include) target_compile_definitions(video_core PRIVATE HAS_VULKAN) endif() create_target_directory_groups(video_core) target_link_libraries(video_core PUBLIC common core) -target_link_libraries(video_core PRIVATE glad lz4_static) +target_link_libraries(video_core PRIVATE glad) +if (ENABLE_VULKAN) + target_link_libraries(video_core PRIVATE sirit) +endif() diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 8b1bea1ae..046d047cb 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -8,6 +8,7 @@ #include "video_core/dma_pusher.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/gpu.h" +#include "video_core/memory_manager.h" namespace Tegra { diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp index 03b7ee5d8..55966eef1 100644 --- a/src/video_core/engines/fermi_2d.cpp +++ b/src/video_core/engines/fermi_2d.cpp @@ -6,12 +6,13 @@ #include "common/logging/log.h" #include "common/math_util.h" #include "video_core/engines/fermi_2d.h" +#include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" namespace Tegra::Engines { Fermi2D::Fermi2D(VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager) - : memory_manager(memory_manager), rasterizer{rasterizer} {} + : rasterizer{rasterizer}, memory_manager{memory_manager} {} void Fermi2D::CallMethod(const GPU::MethodCall& method_call) { ASSERT_MSG(method_call.method < Regs::NUM_REGS, diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h index 80523e320..2e51b7f13 100644 --- a/src/video_core/engines/fermi_2d.h +++ b/src/video_core/engines/fermi_2d.h @@ -10,7 +10,10 @@ #include "common/common_funcs.h" #include "common/common_types.h" #include "video_core/gpu.h" -#include "video_core/memory_manager.h" + +namespace Tegra { +class MemoryManager; +} namespace VideoCore { class RasterizerInterface; @@ -115,10 +118,9 @@ public: }; } regs{}; - MemoryManager& memory_manager; - private: VideoCore::RasterizerInterface& rasterizer; + MemoryManager& memory_manager; /// Performs the copy from the source surface to the destination surface as configured in the /// registers. diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h index 6575afd0f..fb6cdf432 100644 --- a/src/video_core/engines/kepler_compute.h +++ b/src/video_core/engines/kepler_compute.h @@ -9,7 +9,10 @@ #include "common/common_funcs.h" #include "common/common_types.h" #include "video_core/gpu.h" -#include "video_core/memory_manager.h" + +namespace Tegra { +class MemoryManager; +} namespace Tegra::Engines { @@ -40,10 +43,11 @@ public: static_assert(sizeof(Regs) == Regs::NUM_REGS * sizeof(u32), "KeplerCompute Regs has wrong size"); - MemoryManager& memory_manager; - /// Write the value to the register identified by method. void CallMethod(const GPU::MethodCall& method_call); + +private: + MemoryManager& memory_manager; }; #define ASSERT_REG_POSITION(field_name, position) \ diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp index e259bf46b..cd51a31d7 100644 --- a/src/video_core/engines/kepler_memory.cpp +++ b/src/video_core/engines/kepler_memory.cpp @@ -5,9 +5,9 @@ #include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" -#include "core/memory.h" #include "video_core/engines/kepler_memory.h" #include "video_core/engines/maxwell_3d.h" +#include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_base.h" @@ -15,7 +15,7 @@ namespace Tegra::Engines { KeplerMemory::KeplerMemory(Core::System& system, VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager) - : system{system}, memory_manager(memory_manager), rasterizer{rasterizer} {} + : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager} {} KeplerMemory::~KeplerMemory() = default; diff --git a/src/video_core/engines/kepler_memory.h b/src/video_core/engines/kepler_memory.h index 9181e9d80..78b6c3e45 100644 --- a/src/video_core/engines/kepler_memory.h +++ b/src/video_core/engines/kepler_memory.h @@ -10,12 +10,15 @@ #include "common/common_funcs.h" #include "common/common_types.h" #include "video_core/gpu.h" -#include "video_core/memory_manager.h" namespace Core { class System; } +namespace Tegra { +class MemoryManager; +} + namespace VideoCore { class RasterizerInterface; } @@ -82,8 +85,8 @@ public: private: Core::System& system; - MemoryManager& memory_manager; VideoCore::RasterizerInterface& rasterizer; + MemoryManager& memory_manager; void ProcessData(u32 data); }; diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index defcfbd3f..74403eed4 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -7,11 +7,10 @@ #include "common/assert.h" #include "core/core.h" #include "core/core_timing.h" -#include "core/memory.h" #include "video_core/debug_utils/debug_utils.h" #include "video_core/engines/maxwell_3d.h" +#include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" -#include "video_core/renderer_base.h" #include "video_core/textures/texture.h" namespace Tegra::Engines { @@ -21,8 +20,8 @@ constexpr u32 MacroRegistersStart = 0xE00; Maxwell3D::Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager) - : memory_manager(memory_manager), system{system}, rasterizer{rasterizer}, - macro_interpreter(*this) { + : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager}, macro_interpreter{ + *this} { InitializeRegisterDefaults(); } @@ -250,6 +249,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) { ProcessQueryGet(); break; } + case MAXWELL3D_REG_INDEX(sync_info): { + ProcessSyncPoint(); + break; + } default: break; } @@ -327,6 +330,14 @@ void Maxwell3D::ProcessQueryGet() { } } +void Maxwell3D::ProcessSyncPoint() { + const u32 sync_point = regs.sync_info.sync_point.Value(); + const u32 increment = regs.sync_info.increment.Value(); + const u32 cache_flush = regs.sync_info.unknown.Value(); + LOG_DEBUG(HW_GPU, "Syncpoint set {}, increment: {}, unk: {}", sync_point, increment, + cache_flush); +} + void Maxwell3D::DrawArrays() { LOG_DEBUG(HW_GPU, "called, topology={}, count={}", static_cast<u32>(regs.draw.topology.Value()), regs.vertex_buffer.count); diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 7fbf1026e..321af3297 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -16,13 +16,16 @@ #include "common/math_util.h" #include "video_core/gpu.h" #include "video_core/macro_interpreter.h" -#include "video_core/memory_manager.h" #include "video_core/textures/texture.h" namespace Core { class System; } +namespace Tegra { +class MemoryManager; +} + namespace VideoCore { class RasterizerInterface; } @@ -576,7 +579,17 @@ public: u32 bind; } macros; - INSERT_PADDING_WORDS(0x188); + INSERT_PADDING_WORDS(0x69); + + struct { + union { + BitField<0, 16, u32> sync_point; + BitField<16, 1, u32> unknown; + BitField<20, 1, u32> increment; + }; + } sync_info; + + INSERT_PADDING_WORDS(0x11E); u32 tfb_enabled; @@ -1093,7 +1106,6 @@ public: }; State state{}; - MemoryManager& memory_manager; struct DirtyFlags { std::bitset<8> color_buffer{0xFF}; @@ -1141,6 +1153,8 @@ private: VideoCore::RasterizerInterface& rasterizer; + MemoryManager& memory_manager; + /// Start offsets of each macro in macro_memory std::unordered_map<u32, u32> macro_offsets; @@ -1180,6 +1194,9 @@ private: /// Handles a write to the QUERY_GET register. void ProcessQueryGet(); + /// Handles writes to syncing register. + void ProcessSyncPoint(); + /// Handles a write to the CB_DATA[i] register. void ProcessCBData(u32 value); @@ -1195,6 +1212,7 @@ private: "Field " #field_name " has invalid position") ASSERT_REG_POSITION(macros, 0x45); +ASSERT_REG_POSITION(sync_info, 0xB2); ASSERT_REG_POSITION(tfb_enabled, 0x1D1); ASSERT_REG_POSITION(rt, 0x200); ASSERT_REG_POSITION(viewport_transform, 0x280); diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index 5cca5c29a..2426d0067 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -5,9 +5,9 @@ #include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" -#include "core/memory.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/engines/maxwell_dma.h" +#include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_base.h" #include "video_core/textures/decoders.h" @@ -16,7 +16,7 @@ namespace Tegra::Engines { MaxwellDMA::MaxwellDMA(Core::System& system, VideoCore::RasterizerInterface& rasterizer, MemoryManager& memory_manager) - : memory_manager(memory_manager), system{system}, rasterizer{rasterizer} {} + : system{system}, rasterizer{rasterizer}, memory_manager{memory_manager} {} void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) { ASSERT_MSG(method_call.method < Regs::NUM_REGS, diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index 34c369320..c6b649842 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -10,12 +10,15 @@ #include "common/common_funcs.h" #include "common/common_types.h" #include "video_core/gpu.h" -#include "video_core/memory_manager.h" namespace Core { class System; } +namespace Tegra { +class MemoryManager; +} + namespace VideoCore { class RasterizerInterface; } @@ -139,13 +142,13 @@ public: }; } regs{}; - MemoryManager& memory_manager; - private: Core::System& system; VideoCore::RasterizerInterface& rasterizer; + MemoryManager& memory_manager; + /// Performs the copy from the source buffer to the destination buffer as configured in the /// registers. void HandleCopy(); diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index c41e3f4f0..38db4addd 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -806,6 +806,12 @@ union Instruction { } ldg; union { + BitField<48, 3, UniformType> type; + BitField<46, 2, u64> cache_mode; + BitField<20, 24, s64> immediate_offset; + } stg; + + union { BitField<0, 3, u64> pred0; BitField<3, 3, u64> pred3; BitField<7, 1, u64> abs_a; @@ -1252,13 +1258,16 @@ union Instruction { union { BitField<20, 16, u64> imm20_16; + BitField<35, 1, u64> high_b_rr; // used on RR BitField<36, 1, u64> product_shift_left; BitField<37, 1, u64> merge_37; BitField<48, 1, u64> sign_a; BitField<49, 1, u64> sign_b; + BitField<50, 2, XmadMode> mode_cbf; // used by CR, RC BitField<50, 3, XmadMode> mode; BitField<52, 1, u64> high_b; BitField<53, 1, u64> high_a; + BitField<55, 1, u64> product_shift_left_second; // used on CR BitField<56, 1, u64> merge_56; } xmad; @@ -1676,7 +1685,7 @@ private: INST("0011011-11110---", Id::BFI_IMM_R, Type::Bfi, "BFI_IMM_R"), INST("0100110001000---", Id::LOP_C, Type::ArithmeticInteger, "LOP_C"), INST("0101110001000---", Id::LOP_R, Type::ArithmeticInteger, "LOP_R"), - INST("0011100001000---", Id::LOP_IMM, Type::ArithmeticInteger, "LOP_IMM"), + INST("0011100-01000---", Id::LOP_IMM, Type::ArithmeticInteger, "LOP_IMM"), INST("000001----------", Id::LOP32I, Type::ArithmeticIntegerImmediate, "LOP32I"), INST("0000001---------", Id::LOP3_C, Type::ArithmeticInteger, "LOP3_C"), INST("0101101111100---", Id::LOP3_R, Type::ArithmeticInteger, "LOP3_R"), diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 30b29e14d..4461083ff 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -31,7 +31,7 @@ u32 FramebufferConfig::BytesPerPixel(PixelFormat format) { GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer) : renderer{renderer} { auto& rasterizer{renderer.Rasterizer()}; - memory_manager = std::make_unique<Tegra::MemoryManager>(); + memory_manager = std::make_unique<Tegra::MemoryManager>(rasterizer); dma_pusher = std::make_unique<Tegra::DmaPusher>(*this); maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager); fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer, *memory_manager); diff --git a/src/video_core/gpu_asynch.cpp b/src/video_core/gpu_asynch.cpp index 8b355cf7b..db507cf04 100644 --- a/src/video_core/gpu_asynch.cpp +++ b/src/video_core/gpu_asynch.cpp @@ -9,7 +9,7 @@ namespace VideoCommon { GPUAsynch::GPUAsynch(Core::System& system, VideoCore::RendererBase& renderer) - : Tegra::GPU(system, renderer), gpu_thread{renderer, *dma_pusher} {} + : Tegra::GPU(system, renderer), gpu_thread{system, renderer, *dma_pusher} {} GPUAsynch::~GPUAsynch() = default; diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp index c5dc199c5..cc56cf467 100644 --- a/src/video_core/gpu_thread.cpp +++ b/src/video_core/gpu_thread.cpp @@ -4,6 +4,9 @@ #include "common/assert.h" #include "common/microprofile.h" +#include "core/core.h" +#include "core/core_timing.h" +#include "core/core_timing_util.h" #include "core/frontend/scope_acquire_window_context.h" #include "video_core/dma_pusher.h" #include "video_core/gpu.h" @@ -36,7 +39,6 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p dma_pusher.Push(std::move(submit_list->entries)); dma_pusher.DispatchCalls(); } else if (const auto data = std::get_if<SwapBuffersCommand>(&next.data)) { - state.DecrementFramesCounter(); renderer.SwapBuffers(std::move(data->framebuffer)); } else if (const auto data = std::get_if<FlushRegionCommand>(&next.data)) { renderer.Rasterizer().FlushRegion(data->addr, data->size); @@ -47,13 +49,18 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p } else { UNREACHABLE(); } + state.signaled_fence = next.fence; + state.TrySynchronize(); } } } -ThreadManager::ThreadManager(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_pusher) - : renderer{renderer}, thread{RunThread, std::ref(renderer), std::ref(dma_pusher), - std::ref(state)} {} +ThreadManager::ThreadManager(Core::System& system, VideoCore::RendererBase& renderer, + Tegra::DmaPusher& dma_pusher) + : system{system}, thread{RunThread, std::ref(renderer), std::ref(dma_pusher), std::ref(state)} { + synchronization_event = system.CoreTiming().RegisterEvent( + "GPUThreadSynch", [this](u64 fence, s64) { state.WaitForSynchronization(fence); }); +} ThreadManager::~ThreadManager() { // Notify GPU thread that a shutdown is pending @@ -62,14 +69,14 @@ ThreadManager::~ThreadManager() { } void ThreadManager::SubmitList(Tegra::CommandList&& entries) { - PushCommand(SubmitListCommand(std::move(entries))); + const u64 fence{PushCommand(SubmitListCommand(std::move(entries)))}; + const s64 synchronization_ticks{Core::Timing::usToCycles(9000)}; + system.CoreTiming().ScheduleEvent(synchronization_ticks, synchronization_event, fence); } void ThreadManager::SwapBuffers( std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) { - state.IncrementFramesCounter(); PushCommand(SwapBuffersCommand(std::move(framebuffer))); - state.WaitForFrames(); } void ThreadManager::FlushRegion(CacheAddr addr, u64 size) { @@ -79,7 +86,7 @@ void ThreadManager::FlushRegion(CacheAddr addr, u64 size) { void ThreadManager::InvalidateRegion(CacheAddr addr, u64 size) { if (state.queue.Empty()) { // It's quicker to invalidate a single region on the CPU if the queue is already empty - renderer.Rasterizer().InvalidateRegion(addr, size); + system.Renderer().Rasterizer().InvalidateRegion(addr, size); } else { PushCommand(InvalidateRegionCommand(addr, size)); } @@ -90,9 +97,25 @@ void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) { InvalidateRegion(addr, size); } -void ThreadManager::PushCommand(CommandData&& command_data) { - state.queue.Push(CommandDataContainer(std::move(command_data))); +u64 ThreadManager::PushCommand(CommandData&& command_data) { + const u64 fence{++state.last_fence}; + state.queue.Push(CommandDataContainer(std::move(command_data), fence)); state.SignalCommands(); + return fence; +} + +MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192)); +void SynchState::WaitForSynchronization(u64 fence) { + if (signaled_fence >= fence) { + return; + } + + // Wait for the GPU to be idle (all commands to be executed) + { + MICROPROFILE_SCOPE(GPU_wait); + std::unique_lock<std::mutex> lock{synchronization_mutex}; + synchronization_condition.wait(lock, [this, fence] { return signaled_fence >= fence; }); + } } } // namespace VideoCommon::GPUThread diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h index 70acb2e79..62bcea5bb 100644 --- a/src/video_core/gpu_thread.h +++ b/src/video_core/gpu_thread.h @@ -19,9 +19,12 @@ struct FramebufferConfig; class DmaPusher; } // namespace Tegra -namespace VideoCore { -class RendererBase; -} // namespace VideoCore +namespace Core { +class System; +namespace Timing { +struct EventType; +} // namespace Timing +} // namespace Core namespace VideoCommon::GPUThread { @@ -75,63 +78,47 @@ using CommandData = struct CommandDataContainer { CommandDataContainer() = default; - CommandDataContainer(CommandData&& data) : data{std::move(data)} {} + CommandDataContainer(CommandData&& data, u64 next_fence) + : data{std::move(data)}, fence{next_fence} {} CommandDataContainer& operator=(const CommandDataContainer& t) { data = std::move(t.data); + fence = t.fence; return *this; } CommandData data; + u64 fence{}; }; /// Struct used to synchronize the GPU thread struct SynchState final { std::atomic_bool is_running{true}; std::atomic_int queued_frame_count{}; - std::mutex frames_mutex; + std::mutex synchronization_mutex; std::mutex commands_mutex; std::condition_variable commands_condition; - std::condition_variable frames_condition; + std::condition_variable synchronization_condition; - void IncrementFramesCounter() { - std::lock_guard lock{frames_mutex}; - ++queued_frame_count; + /// Returns true if the gap in GPU commands is small enough that we can consider the CPU and GPU + /// synchronized. This is entirely empirical. + bool IsSynchronized() const { + constexpr std::size_t max_queue_gap{5}; + return queue.Size() <= max_queue_gap; } - void DecrementFramesCounter() { - { - std::lock_guard lock{frames_mutex}; - --queued_frame_count; - - if (queued_frame_count) { - return; - } + void TrySynchronize() { + if (IsSynchronized()) { + std::lock_guard<std::mutex> lock{synchronization_mutex}; + synchronization_condition.notify_one(); } - frames_condition.notify_one(); } - void WaitForFrames() { - { - std::lock_guard lock{frames_mutex}; - if (!queued_frame_count) { - return; - } - } - - // Wait for the GPU to be idle (all commands to be executed) - { - std::unique_lock lock{frames_mutex}; - frames_condition.wait(lock, [this] { return !queued_frame_count; }); - } - } + void WaitForSynchronization(u64 fence); void SignalCommands() { - { - std::unique_lock lock{commands_mutex}; - if (queue.Empty()) { - return; - } + if (queue.Empty()) { + return; } commands_condition.notify_one(); @@ -144,12 +131,15 @@ struct SynchState final { using CommandQueue = Common::SPSCQueue<CommandDataContainer>; CommandQueue queue; + u64 last_fence{}; + std::atomic<u64> signaled_fence{}; }; /// Class used to manage the GPU thread class ThreadManager final { public: - explicit ThreadManager(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_pusher); + explicit ThreadManager(Core::System& system, VideoCore::RendererBase& renderer, + Tegra::DmaPusher& dma_pusher); ~ThreadManager(); /// Push GPU command entries to be processed @@ -170,11 +160,12 @@ public: private: /// Pushes a command to be executed by the GPU thread - void PushCommand(CommandData&& command_data); + u64 PushCommand(CommandData&& command_data); private: SynchState state; - VideoCore::RendererBase& renderer; + Core::System& system; + Core::Timing::EventType* synchronization_event{}; std::thread thread; std::thread::id thread_id; }; diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro_interpreter.cpp index 64f75db43..524d9ea5a 100644 --- a/src/video_core/macro_interpreter.cpp +++ b/src/video_core/macro_interpreter.cpp @@ -223,27 +223,21 @@ void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 res } u32 MacroInterpreter::FetchParameter() { - ASSERT(next_parameter_index < parameters.size()); - return parameters[next_parameter_index++]; + return parameters.at(next_parameter_index++); } u32 MacroInterpreter::GetRegister(u32 register_id) const { - // Register 0 is supposed to always return 0. - if (register_id == 0) - return 0; - - ASSERT(register_id < registers.size()); - return registers[register_id]; + return registers.at(register_id); } void MacroInterpreter::SetRegister(u32 register_id, u32 value) { - // Register 0 is supposed to always return 0. NOP is implemented as a store to the zero - // register. - if (register_id == 0) + // Register 0 is hardwired as the zero register. + // Ensure no writes to it actually occur. + if (register_id == 0) { return; + } - ASSERT(register_id < registers.size()); - registers[register_id] = value; + registers.at(register_id) = value; } void MacroInterpreter::SetMethodAddress(u32 address) { diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index e76b59842..0f4e820aa 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -5,16 +5,13 @@ #include "common/alignment.h" #include "common/assert.h" #include "common/logging/log.h" -#include "core/core.h" #include "core/memory.h" -#include "video_core/gpu.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" -#include "video_core/renderer_base.h" namespace Tegra { -MemoryManager::MemoryManager() { +MemoryManager::MemoryManager(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} { std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr); std::fill(page_table.attributes.begin(), page_table.attributes.end(), Common::PageType::Unmapped); @@ -70,23 +67,23 @@ GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) { const u64 aligned_size{Common::AlignUp(size, page_size)}; const CacheAddr cache_addr{ToCacheAddr(GetPointer(gpu_addr))}; - Core::System::GetInstance().Renderer().Rasterizer().FlushAndInvalidateRegion(cache_addr, - aligned_size); + rasterizer.FlushAndInvalidateRegion(cache_addr, aligned_size); UnmapRange(gpu_addr, aligned_size); return gpu_addr; } -GPUVAddr MemoryManager::FindFreeRegion(GPUVAddr region_start, u64 size) { +GPUVAddr MemoryManager::FindFreeRegion(GPUVAddr region_start, u64 size) const { // Find the first Free VMA. - const VMAHandle vma_handle{std::find_if(vma_map.begin(), vma_map.end(), [&](const auto& vma) { - if (vma.second.type != VirtualMemoryArea::Type::Unmapped) { - return false; - } + const VMAHandle vma_handle{ + std::find_if(vma_map.begin(), vma_map.end(), [region_start, size](const auto& vma) { + if (vma.second.type != VirtualMemoryArea::Type::Unmapped) { + return false; + } - const VAddr vma_end{vma.second.base + vma.second.size}; - return vma_end > region_start && vma_end >= region_start + size; - })}; + const VAddr vma_end{vma.second.base + vma.second.size}; + return vma_end > region_start && vma_end >= region_start + size; + })}; if (vma_handle == vma_map.end()) { return {}; @@ -99,12 +96,12 @@ bool MemoryManager::IsAddressValid(GPUVAddr addr) const { return (addr >> page_bits) < page_table.pointers.size(); } -std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr) { +std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr) const { if (!IsAddressValid(addr)) { return {}; } - VAddr cpu_addr{page_table.backing_addr[addr >> page_bits]}; + const VAddr cpu_addr{page_table.backing_addr[addr >> page_bits]}; if (cpu_addr) { return cpu_addr + (addr & page_mask); } @@ -113,7 +110,7 @@ std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr) { } template <typename T> -T MemoryManager::Read(GPUVAddr addr) { +T MemoryManager::Read(GPUVAddr addr) const { if (!IsAddressValid(addr)) { return {}; } @@ -165,10 +162,10 @@ void MemoryManager::Write(GPUVAddr addr, T data) { } } -template u8 MemoryManager::Read<u8>(GPUVAddr addr); -template u16 MemoryManager::Read<u16>(GPUVAddr addr); -template u32 MemoryManager::Read<u32>(GPUVAddr addr); -template u64 MemoryManager::Read<u64>(GPUVAddr addr); +template u8 MemoryManager::Read<u8>(GPUVAddr addr) const; +template u16 MemoryManager::Read<u16>(GPUVAddr addr) const; +template u32 MemoryManager::Read<u32>(GPUVAddr addr) const; +template u64 MemoryManager::Read<u64>(GPUVAddr addr) const; template void MemoryManager::Write<u8>(GPUVAddr addr, u8 data); template void MemoryManager::Write<u16>(GPUVAddr addr, u16 data); template void MemoryManager::Write<u32>(GPUVAddr addr, u32 data); @@ -179,8 +176,22 @@ u8* MemoryManager::GetPointer(GPUVAddr addr) { return {}; } - u8* page_pointer{page_table.pointers[addr >> page_bits]}; - if (page_pointer) { + u8* const page_pointer{page_table.pointers[addr >> page_bits]}; + if (page_pointer != nullptr) { + return page_pointer + (addr & page_mask); + } + + LOG_ERROR(HW_GPU, "Unknown GetPointer @ 0x{:016X}", addr); + return {}; +} + +const u8* MemoryManager::GetPointer(GPUVAddr addr) const { + if (!IsAddressValid(addr)) { + return {}; + } + + const u8* const page_pointer{page_table.pointers[addr >> page_bits]}; + if (page_pointer != nullptr) { return page_pointer + (addr & page_mask); } @@ -188,15 +199,86 @@ u8* MemoryManager::GetPointer(GPUVAddr addr) { return {}; } -void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size) { - std::memcpy(dest_buffer, GetPointer(src_addr), size); +void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size) const { + std::size_t remaining_size{size}; + std::size_t page_index{src_addr >> page_bits}; + std::size_t page_offset{src_addr & page_mask}; + + while (remaining_size > 0) { + const std::size_t copy_amount{ + std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; + + switch (page_table.attributes[page_index]) { + case Common::PageType::Memory: { + const u8* src_ptr{page_table.pointers[page_index] + page_offset}; + rasterizer.FlushRegion(ToCacheAddr(src_ptr), copy_amount); + std::memcpy(dest_buffer, src_ptr, copy_amount); + break; + } + default: + UNREACHABLE(); + } + + page_index++; + page_offset = 0; + dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount; + remaining_size -= copy_amount; + } } + void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size) { - std::memcpy(GetPointer(dest_addr), src_buffer, size); + std::size_t remaining_size{size}; + std::size_t page_index{dest_addr >> page_bits}; + std::size_t page_offset{dest_addr & page_mask}; + + while (remaining_size > 0) { + const std::size_t copy_amount{ + std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; + + switch (page_table.attributes[page_index]) { + case Common::PageType::Memory: { + u8* dest_ptr{page_table.pointers[page_index] + page_offset}; + rasterizer.InvalidateRegion(ToCacheAddr(dest_ptr), copy_amount); + std::memcpy(dest_ptr, src_buffer, copy_amount); + break; + } + default: + UNREACHABLE(); + } + + page_index++; + page_offset = 0; + src_buffer = static_cast<const u8*>(src_buffer) + copy_amount; + remaining_size -= copy_amount; + } } void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size) { - std::memcpy(GetPointer(dest_addr), GetPointer(src_addr), size); + std::size_t remaining_size{size}; + std::size_t page_index{src_addr >> page_bits}; + std::size_t page_offset{src_addr & page_mask}; + + while (remaining_size > 0) { + const std::size_t copy_amount{ + std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; + + switch (page_table.attributes[page_index]) { + case Common::PageType::Memory: { + const u8* src_ptr{page_table.pointers[page_index] + page_offset}; + rasterizer.FlushRegion(ToCacheAddr(src_ptr), copy_amount); + WriteBlock(dest_addr, src_ptr, copy_amount); + break; + } + default: + UNREACHABLE(); + } + + page_index++; + page_offset = 0; + dest_addr += static_cast<VAddr>(copy_amount); + src_addr += static_cast<VAddr>(copy_amount); + remaining_size -= copy_amount; + } } void MemoryManager::MapPages(GPUVAddr base, u64 size, u8* memory, Common::PageType type, @@ -336,7 +418,7 @@ MemoryManager::VMAIter MemoryManager::CarveVMA(GPUVAddr base, u64 size) { const VirtualMemoryArea& vma{vma_handle->second}; if (vma.type == VirtualMemoryArea::Type::Mapped) { // Region is already allocated - return {}; + return vma_handle; } const VAddr start_in_vma{base - vma.base}; diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index 34744bb27..647cbf93a 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -10,6 +10,10 @@ #include "common/common_types.h" #include "common/page_table.h" +namespace VideoCore { +class RasterizerInterface; +} + namespace Tegra { /** @@ -43,24 +47,25 @@ struct VirtualMemoryArea { class MemoryManager final { public: - MemoryManager(); + MemoryManager(VideoCore::RasterizerInterface& rasterizer); GPUVAddr AllocateSpace(u64 size, u64 align); GPUVAddr AllocateSpace(GPUVAddr addr, u64 size, u64 align); GPUVAddr MapBufferEx(VAddr cpu_addr, u64 size); GPUVAddr MapBufferEx(VAddr cpu_addr, GPUVAddr addr, u64 size); GPUVAddr UnmapBuffer(GPUVAddr addr, u64 size); - std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr); + std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr) const; template <typename T> - T Read(GPUVAddr addr); + T Read(GPUVAddr addr) const; template <typename T> void Write(GPUVAddr addr, T data); u8* GetPointer(GPUVAddr addr); + const u8* GetPointer(GPUVAddr addr) const; - void ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size); + void ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size) const; void WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size); void CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size); @@ -127,7 +132,7 @@ private: void UpdatePageTableForVMA(const VirtualMemoryArea& vma); /// Finds a free (unmapped region) of the specified size starting at the specified address. - GPUVAddr FindFreeRegion(GPUVAddr region_start, u64 size); + GPUVAddr FindFreeRegion(GPUVAddr region_start, u64 size) const; private: static constexpr u64 page_bits{16}; @@ -143,6 +148,7 @@ private: Common::PageTable page_table{page_bits}; VMAMap vma_map; + VideoCore::RasterizerInterface& rasterizer; }; } // namespace Tegra diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index fd091c84c..25652e794 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -7,7 +7,7 @@ #include "common/alignment.h" #include "core/core.h" -#include "core/memory.h" +#include "video_core/memory_manager.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_rasterizer.h" diff --git a/src/video_core/renderer_opengl/gl_global_cache.cpp b/src/video_core/renderer_opengl/gl_global_cache.cpp index da9326253..ea4a593af 100644 --- a/src/video_core/renderer_opengl/gl_global_cache.cpp +++ b/src/video_core/renderer_opengl/gl_global_cache.cpp @@ -4,9 +4,9 @@ #include <glad/glad.h> -#include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" +#include "video_core/memory_manager.h" #include "video_core/renderer_opengl/gl_global_cache.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" @@ -14,28 +14,28 @@ namespace OpenGL { -CachedGlobalRegion::CachedGlobalRegion(VAddr cpu_addr, u32 size, u8* host_ptr) - : RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, size{size} { +CachedGlobalRegion::CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size) + : RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, host_ptr{host_ptr}, size{size}, + max_size{max_size} { buffer.Create(); - // Bind and unbind the buffer so it gets allocated by the driver - glBindBuffer(GL_SHADER_STORAGE_BUFFER, buffer.handle); - glBindBuffer(GL_SHADER_STORAGE_BUFFER, 0); LabelGLObject(GL_BUFFER, buffer.handle, cpu_addr, "GlobalMemory"); } -void CachedGlobalRegion::Reload(u32 size_) { - constexpr auto max_size = static_cast<u32>(RasterizerOpenGL::MaxGlobalMemorySize); +CachedGlobalRegion::~CachedGlobalRegion() = default; +void CachedGlobalRegion::Reload(u32 size_) { size = size_; if (size > max_size) { size = max_size; - LOG_CRITICAL(HW_GPU, "Global region size {} exceeded the expected size {}!", size_, + LOG_CRITICAL(HW_GPU, "Global region size {} exceeded the supported size {}!", size_, max_size); } + glNamedBufferData(buffer.handle, size, host_ptr, GL_STREAM_DRAW); +} - // TODO(Rodrigo): Get rid of Memory::GetPointer with a staging buffer - glBindBuffer(GL_SHADER_STORAGE_BUFFER, buffer.handle); - glBufferData(GL_SHADER_STORAGE_BUFFER, size, GetHostPtr(), GL_DYNAMIC_DRAW); +void CachedGlobalRegion::Flush() { + LOG_DEBUG(Render_OpenGL, "Flushing {} bytes to CPU memory address 0x{:16}", size, cpu_addr); + glGetNamedBufferSubData(buffer.handle, 0, static_cast<GLsizeiptr>(size), host_ptr); } GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const { @@ -46,14 +46,16 @@ GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, return search->second; } -GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(GPUVAddr addr, u32 size, - u8* host_ptr) { +GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr, + u32 size) { GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)}; if (!region) { // No reserved surface available, create a new one and reserve it auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()}; - const auto cpu_addr = *memory_manager.GpuToCpuAddress(addr); - region = std::make_shared<CachedGlobalRegion>(cpu_addr, size, host_ptr); + const auto cpu_addr{memory_manager.GpuToCpuAddress(addr)}; + ASSERT(cpu_addr); + + region = std::make_shared<CachedGlobalRegion>(*cpu_addr, host_ptr, size, max_ssbo_size); ReserveGlobalRegion(region); } region->Reload(size); @@ -65,7 +67,11 @@ void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) { } GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer) - : RasterizerCache{rasterizer} {} + : RasterizerCache{rasterizer} { + GLint max_ssbo_size_; + glGetIntegerv(GL_MAX_SHADER_STORAGE_BLOCK_SIZE, &max_ssbo_size_); + max_ssbo_size = static_cast<u32>(max_ssbo_size_); +} GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion( const GLShader::GlobalMemoryEntry& global_region, @@ -73,7 +79,7 @@ GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion( auto& gpu{Core::System::GetInstance().GPU()}; auto& memory_manager{gpu.MemoryManager()}; - const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<u64>(stage)]}; + const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]}; const auto addr{cbufs.const_buffers[global_region.GetCbufIndex()].address + global_region.GetCbufOffset()}; const auto actual_addr{memory_manager.Read<u64>(addr)}; @@ -85,7 +91,7 @@ GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion( if (!region) { // No global region found - create a new one - region = GetUncachedGlobalRegion(actual_addr, size, host_ptr); + region = GetUncachedGlobalRegion(actual_addr, host_ptr, size); Register(region); } diff --git a/src/video_core/renderer_opengl/gl_global_cache.h b/src/video_core/renderer_opengl/gl_global_cache.h index 5a21ab66f..196e6e278 100644 --- a/src/video_core/renderer_opengl/gl_global_cache.h +++ b/src/video_core/renderer_opengl/gl_global_cache.h @@ -19,7 +19,7 @@ namespace OpenGL { namespace GLShader { class GlobalMemoryEntry; -} // namespace GLShader +} class RasterizerOpenGL; class CachedGlobalRegion; @@ -27,7 +27,8 @@ using GlobalRegion = std::shared_ptr<CachedGlobalRegion>; class CachedGlobalRegion final : public RasterizerCacheObject { public: - explicit CachedGlobalRegion(VAddr cpu_addr, u32 size, u8* host_ptr); + explicit CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size); + ~CachedGlobalRegion(); VAddr GetCpuAddr() const override { return cpu_addr; @@ -45,14 +46,14 @@ public: /// Reloads the global region from guest memory void Reload(u32 size_); - // TODO(Rodrigo): When global memory is written (STG), implement flushing - void Flush() override { - UNIMPLEMENTED(); - } + void Flush() override; private: VAddr cpu_addr{}; + u8* host_ptr{}; u32 size{}; + u32 max_size{}; + OGLBuffer buffer; }; @@ -66,10 +67,11 @@ public: private: GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const; - GlobalRegion GetUncachedGlobalRegion(GPUVAddr addr, u32 size, u8* host_ptr); + GlobalRegion GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr, u32 size); void ReserveGlobalRegion(GlobalRegion region); std::unordered_map<CacheAddr, GlobalRegion> reserve; + u32 max_ssbo_size{}; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_primitive_assembler.cpp b/src/video_core/renderer_opengl/gl_primitive_assembler.cpp index 2bcbd3da2..c3e94d917 100644 --- a/src/video_core/renderer_opengl/gl_primitive_assembler.cpp +++ b/src/video_core/renderer_opengl/gl_primitive_assembler.cpp @@ -7,7 +7,7 @@ #include "common/assert.h" #include "common/common_types.h" #include "core/core.h" -#include "core/memory.h" +#include "video_core/memory_manager.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_primitive_assembler.h" diff --git a/src/video_core/renderer_opengl/gl_primitive_assembler.h b/src/video_core/renderer_opengl/gl_primitive_assembler.h index 0e2e7dc36..4e87ce4d6 100644 --- a/src/video_core/renderer_opengl/gl_primitive_assembler.h +++ b/src/video_core/renderer_opengl/gl_primitive_assembler.h @@ -4,11 +4,9 @@ #pragma once -#include <vector> #include <glad/glad.h> #include "common/common_types.h" -#include "video_core/memory_manager.h" namespace OpenGL { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 046fc935b..0b1fe3494 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -17,7 +17,6 @@ #include "common/microprofile.h" #include "common/scope_exit.h" #include "core/core.h" -#include "core/frontend/emu_window.h" #include "core/hle/kernel/process.h" #include "core/settings.h" #include "video_core/engines/maxwell_3d.h" @@ -26,7 +25,6 @@ #include "video_core/renderer_opengl/gl_shader_gen.h" #include "video_core/renderer_opengl/maxwell_to_gl.h" #include "video_core/renderer_opengl/renderer_opengl.h" -#include "video_core/video_core.h" namespace OpenGL { @@ -301,6 +299,10 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { BaseBindings base_bindings; std::array<bool, Maxwell::NumClipDistances> clip_distances{}; + // Prepare packed bindings + bind_ubo_pushbuffer.Setup(base_bindings.cbuf); + bind_ssbo_pushbuffer.Setup(base_bindings.gmem); + for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { const auto& shader_config = gpu.regs.shader_config[index]; const Maxwell::ShaderProgram program{static_cast<Maxwell::ShaderProgram>(index)}; @@ -318,13 +320,13 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { const std::size_t stage{index == 0 ? 0 : index - 1}; // Stage indices are 0 - 5 GLShader::MaxwellUniformData ubo{}; - ubo.SetFromRegs(gpu.state.shader_stages[stage]); + ubo.SetFromRegs(gpu, stage); const GLintptr offset = buffer_cache.UploadHostMemory( &ubo, sizeof(ubo), static_cast<std::size_t>(uniform_buffer_alignment)); // Bind the emulation info buffer - glBindBufferRange(GL_UNIFORM_BUFFER, base_bindings.cbuf, buffer_cache.GetHandle(), offset, - static_cast<GLsizeiptr>(sizeof(ubo))); + bind_ubo_pushbuffer.Push(buffer_cache.GetHandle(), offset, + static_cast<GLsizeiptr>(sizeof(ubo))); Shader shader{shader_cache.GetStageProgram(program)}; const auto [program_handle, next_bindings] = @@ -368,6 +370,9 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { base_bindings = next_bindings; } + bind_ubo_pushbuffer.Bind(); + bind_ssbo_pushbuffer.Bind(); + SyncClipEnabled(clip_distances); gpu.dirty_flags.shaders = false; @@ -577,9 +582,6 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers( } void RasterizerOpenGL::Clear() { - const auto prev_state{state}; - SCOPE_EXIT({ prev_state.Apply(); }); - const auto& regs = system.GPU().Maxwell3D().regs; bool use_color{}; bool use_depth{}; @@ -651,7 +653,10 @@ void RasterizerOpenGL::Clear() { clear_state.EmulateViewportWithScissor(); } - clear_state.Apply(); + clear_state.ApplyColorMask(); + clear_state.ApplyDepth(); + clear_state.ApplyStencilTest(); + clear_state.ApplyViewport(); if (use_color) { glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color); @@ -751,6 +756,7 @@ void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) { return; } res_cache.FlushRegion(addr, size); + global_cache.FlushRegion(addr, size); } void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) { @@ -902,23 +908,14 @@ void RasterizerOpenGL::SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::Shader const auto& shader_stage = maxwell3d.state.shader_stages[static_cast<std::size_t>(stage)]; const auto& entries = shader->GetShaderEntries().const_buffers; - constexpr u64 max_binds = Tegra::Engines::Maxwell3D::Regs::MaxConstBuffers; - std::array<GLuint, max_binds> bind_buffers; - std::array<GLintptr, max_binds> bind_offsets; - std::array<GLsizeiptr, max_binds> bind_sizes; - - ASSERT_MSG(entries.size() <= max_binds, "Exceeded expected number of binding points."); - // Upload only the enabled buffers from the 16 constbuffers of each shader stage for (u32 bindpoint = 0; bindpoint < entries.size(); ++bindpoint) { const auto& used_buffer = entries[bindpoint]; const auto& buffer = shader_stage.const_buffers[used_buffer.GetIndex()]; if (!buffer.enabled) { - // With disabled buffers set values as zero to unbind them - bind_buffers[bindpoint] = 0; - bind_offsets[bindpoint] = 0; - bind_sizes[bindpoint] = 0; + // Set values to zero to unbind buffers + bind_ubo_pushbuffer.Push(0, 0, 0); continue; } @@ -946,30 +943,22 @@ void RasterizerOpenGL::SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::Shader const GLintptr const_buffer_offset = buffer_cache.UploadMemory( buffer.address, size, static_cast<std::size_t>(uniform_buffer_alignment)); - // Prepare values for multibind - bind_buffers[bindpoint] = buffer_cache.GetHandle(); - bind_offsets[bindpoint] = const_buffer_offset; - bind_sizes[bindpoint] = size; + bind_ubo_pushbuffer.Push(buffer_cache.GetHandle(), const_buffer_offset, size); } - - // The first binding is reserved for emulation values - const GLuint ubo_base_binding = base_bindings.cbuf + 1; - glBindBuffersRange(GL_UNIFORM_BUFFER, ubo_base_binding, static_cast<GLsizei>(entries.size()), - bind_buffers.data(), bind_offsets.data(), bind_sizes.data()); } void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, const Shader& shader, GLenum primitive_mode, BaseBindings base_bindings) { - // TODO(Rodrigo): Use ARB_multi_bind here const auto& entries = shader->GetShaderEntries().global_memory_entries; - - for (u32 bindpoint = 0; bindpoint < static_cast<u32>(entries.size()); ++bindpoint) { - const auto& entry = entries[bindpoint]; - const u32 current_bindpoint = base_bindings.gmem + bindpoint; - const auto& region = global_cache.GetGlobalRegion(entry, stage); - - glBindBufferBase(GL_SHADER_STORAGE_BUFFER, current_bindpoint, region->GetBufferHandle()); + for (std::size_t bindpoint = 0; bindpoint < entries.size(); ++bindpoint) { + const auto& entry{entries[bindpoint]}; + const auto& region{global_cache.GetGlobalRegion(entry, stage)}; + if (entry.IsWritten()) { + region->MarkAsModified(true, global_cache); + } + bind_ssbo_pushbuffer.Push(region->GetBufferHandle(), 0, + static_cast<GLsizeiptr>(region->GetSizeInBytes())); } } diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 4de565321..d4c2cf80e 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -12,15 +12,12 @@ #include <optional> #include <tuple> #include <utility> -#include <vector> #include <boost/icl/interval_map.hpp> -#include <boost/range/iterator_range.hpp> #include <glad/glad.h> #include "common/common_types.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/memory_manager.h" #include "video_core/rasterizer_cache.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" @@ -29,10 +26,9 @@ #include "video_core/renderer_opengl/gl_rasterizer_cache.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_cache.h" -#include "video_core/renderer_opengl/gl_shader_gen.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state.h" -#include "video_core/renderer_opengl/gl_stream_buffer.h" +#include "video_core/renderer_opengl/utils.h" namespace Core { class System; @@ -75,10 +71,6 @@ public: static_assert(MaxConstbufferSize % sizeof(GLvec4) == 0, "The maximum size of a constbuffer must be a multiple of the size of GLvec4"); - static constexpr std::size_t MaxGlobalMemorySize = 0x10000; - static_assert(MaxGlobalMemorySize % sizeof(float) == 0, - "The maximum size of a global memory must be a multiple of the size of float"); - private: class SamplerInfo { public: @@ -234,6 +226,9 @@ private: PrimitiveAssembler primitive_assembler{buffer_cache}; GLint uniform_buffer_alignment; + BindBuffersRangePushBuffer bind_ubo_pushbuffer{GL_UNIFORM_BUFFER}; + BindBuffersRangePushBuffer bind_ssbo_pushbuffer{GL_SHADER_STORAGE_BUFFER}; + std::size_t CalculateVertexArraysSize() const; std::size_t CalculateIndexBufferSize() const; diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index aba6ce731..7a68b8738 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -13,9 +13,9 @@ #include "common/scope_exit.h" #include "core/core.h" #include "core/hle/kernel/process.h" -#include "core/memory.h" #include "core/settings.h" #include "video_core/engines/maxwell_3d.h" +#include "video_core/memory_manager.h" #include "video_core/morton.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_rasterizer_cache.h" @@ -112,11 +112,26 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only, params.pixel_format = PixelFormatFromTextureFormat(config.tic.format, config.tic.r_type.Value(), params.srgb_conversion); - if (params.pixel_format == PixelFormat::R16U && config.tsc.depth_compare_enabled) { + if (config.tsc.depth_compare_enabled) { // Some titles create a 'R16U' (normalized 16-bit) texture with depth_compare enabled, // then attempt to sample from it via a shadow sampler. Convert format to Z16 (which also // causes GetFormatType to properly return 'Depth' below). - params.pixel_format = PixelFormat::Z16; + if (GetFormatType(params.pixel_format) == SurfaceType::ColorTexture) { + switch (params.pixel_format) { + case PixelFormat::R16S: + case PixelFormat::R16U: + case PixelFormat::R16F: + params.pixel_format = PixelFormat::Z16; + break; + case PixelFormat::R32F: + params.pixel_format = PixelFormat::Z32F; + break; + default: + LOG_WARNING(HW_GPU, "Color texture format being used with depth compare: {}", + static_cast<u32>(params.pixel_format)); + break; + } + } } params.component_type = ComponentTypeFromTexture(config.tic.r_type.Value()); @@ -266,6 +281,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only, params.component_type = ComponentTypeFromRenderTarget(config.format); params.type = GetFormatType(params.pixel_format); params.width = config.width; + params.pitch = config.pitch; params.height = config.height; params.unaligned_height = config.height; params.target = SurfaceTarget::Texture2D; @@ -662,8 +678,8 @@ void CachedSurface::FlushGLBuffer() { gl_buffer[0].resize(GetSizeInBytes()); const FormatTuple& tuple = GetFormatTuple(params.pixel_format, params.component_type); - // Ensure no bad interactions with GL_UNPACK_ALIGNMENT - ASSERT(params.width * GetBytesPerPixel(params.pixel_format) % 4 == 0); + const u32 align = std::clamp(params.RowAlign(0), 1U, 8U); + glPixelStorei(GL_PACK_ALIGNMENT, align); glPixelStorei(GL_PACK_ROW_LENGTH, static_cast<GLint>(params.width)); ASSERT(!tuple.compressed); glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); @@ -708,8 +724,8 @@ void CachedSurface::UploadGLMipmapTexture(u32 mip_map, GLuint read_fb_handle, const FormatTuple& tuple = GetFormatTuple(params.pixel_format, params.component_type); - // Ensure no bad interactions with GL_UNPACK_ALIGNMENT - ASSERT(params.MipWidth(mip_map) * GetBytesPerPixel(params.pixel_format) % 4 == 0); + const u32 align = std::clamp(params.RowAlign(mip_map), 1U, 8U); + glPixelStorei(GL_UNPACK_ALIGNMENT, align); glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(params.MipWidth(mip_map))); const auto image_size = static_cast<GLsizei>(params.GetMipmapSizeGL(mip_map, false)); @@ -1175,10 +1191,16 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface, return new_surface; } + const bool old_compressed = + GetFormatTuple(old_params.pixel_format, old_params.component_type).compressed; + const bool new_compressed = + GetFormatTuple(new_params.pixel_format, new_params.component_type).compressed; + const bool compatible_formats = + GetFormatBpp(old_params.pixel_format) == GetFormatBpp(new_params.pixel_format) && + !(old_compressed || new_compressed); // For compatible surfaces, we can just do fast glCopyImageSubData based copy - if (old_params.target == new_params.target && old_params.type == new_params.type && - old_params.depth == new_params.depth && old_params.depth == 1 && - GetFormatBpp(old_params.pixel_format) == GetFormatBpp(new_params.pixel_format)) { + if (old_params.target == new_params.target && old_params.depth == new_params.depth && + old_params.depth == 1 && compatible_formats) { FastCopySurface(old_surface, new_surface); return new_surface; } @@ -1193,7 +1215,7 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface, case SurfaceTarget::TextureCubemap: case SurfaceTarget::Texture2DArray: case SurfaceTarget::TextureCubeArray: - if (old_params.pixel_format == new_params.pixel_format) + if (compatible_formats) FastLayeredCopySurface(old_surface, new_surface); else { AccurateCopySurface(old_surface, new_surface); diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index e8073579f..db280dbb3 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -5,13 +5,13 @@ #pragma once #include <array> -#include <map> #include <memory> #include <string> -#include <unordered_set> +#include <tuple> #include <vector> #include "common/alignment.h" +#include "common/bit_util.h" #include "common/common_types.h" #include "common/hash.h" #include "common/math_util.h" @@ -206,6 +206,13 @@ struct SurfaceParams { return bd; } + u32 RowAlign(u32 mip_level) const { + const u32 m_width = MipWidth(mip_level); + const u32 bytes_per_pixel = GetBytesPerPixel(pixel_format); + const u32 l2 = Common::CountTrailingZeroes32(m_width * bytes_per_pixel); + return (1U << l2); + } + /// Creates SurfaceParams from a texture configuration static SurfaceParams CreateForTexture(const Tegra::Texture::FullTextureInfo& config, const GLShader::SamplerEntry& entry); diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 290e654bc..99f67494c 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -6,13 +6,12 @@ #include "common/assert.h" #include "common/hash.h" #include "core/core.h" -#include "core/memory.h" #include "video_core/engines/maxwell_3d.h" +#include "video_core/memory_manager.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_cache.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" #include "video_core/renderer_opengl/gl_shader_disk_cache.h" -#include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/utils.h" #include "video_core/shader/shader_ir.h" @@ -41,6 +40,10 @@ GPUVAddr GetShaderAddress(Maxwell::ShaderProgram program) { /// Gets the shader program code from memory for the specified address ProgramCode GetShaderCode(const u8* host_ptr) { ProgramCode program_code(VideoCommon::Shader::MAX_PROGRAM_LENGTH); + ASSERT_OR_EXECUTE(host_ptr != nullptr, { + std::fill(program_code.begin(), program_code.end(), 0); + return program_code; + }); std::memcpy(program_code.data(), host_ptr, program_code.size() * sizeof(u64)); return program_code; } diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index fd1c85115..0cf8e0b3d 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -5,21 +5,20 @@ #pragma once #include <array> +#include <atomic> #include <memory> #include <set> #include <tuple> #include <unordered_map> +#include <vector> #include <glad/glad.h> -#include "common/assert.h" #include "common/common_types.h" #include "video_core/rasterizer_cache.h" -#include "video_core/renderer_base.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" #include "video_core/renderer_opengl/gl_shader_disk_cache.h" -#include "video_core/renderer_opengl/gl_shader_gen.h" namespace Core { class System; diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 11d1169f0..445048daf 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -21,6 +21,8 @@ namespace OpenGL::GLShader { +namespace { + using Tegra::Shader::Attribute; using Tegra::Shader::AttributeUse; using Tegra::Shader::Header; @@ -34,13 +36,15 @@ using Maxwell = Tegra::Engines::Maxwell3D::Regs; using ShaderStage = Tegra::Engines::Maxwell3D::Regs::ShaderStage; using Operation = const OperationNode&; +enum class Type { Bool, Bool2, Float, Int, Uint, HalfFloat }; + +struct TextureAoffi {}; +using TextureArgument = std::pair<Type, Node>; +using TextureIR = std::variant<TextureAoffi, TextureArgument>; + enum : u32 { POSITION_VARYING_LOCATION = 0, GENERIC_VARYING_START_LOCATION = 1 }; constexpr u32 MAX_CONSTBUFFER_ELEMENTS = static_cast<u32>(RasterizerOpenGL::MaxConstbufferSize) / (4 * sizeof(float)); -constexpr u32 MAX_GLOBALMEMORY_ELEMENTS = - static_cast<u32>(RasterizerOpenGL::MaxGlobalMemorySize) / sizeof(float); - -enum class Type { Bool, Bool2, Float, Int, Uint, HalfFloat }; class ShaderWriter { public: @@ -69,10 +73,10 @@ public: shader_source += '\n'; } - std::string GenerateTemporal() { - std::string temporal = "tmp"; - temporal += std::to_string(temporal_index++); - return temporal; + std::string GenerateTemporary() { + std::string temporary = "tmp"; + temporary += std::to_string(temporary_index++); + return temporary; } std::string GetResult() { @@ -87,11 +91,11 @@ private: } std::string shader_source; - u32 temporal_index = 1; + u32 temporary_index = 1; }; /// Generates code to use for a swizzle operation. -static std::string GetSwizzle(u32 elem) { +std::string GetSwizzle(u32 elem) { ASSERT(elem <= 3); std::string swizzle = "."; swizzle += "xyzw"[elem]; @@ -99,7 +103,7 @@ static std::string GetSwizzle(u32 elem) { } /// Translate topology -static std::string GetTopologyName(Tegra::Shader::OutputTopology topology) { +std::string GetTopologyName(Tegra::Shader::OutputTopology topology) { switch (topology) { case Tegra::Shader::OutputTopology::PointList: return "points"; @@ -114,7 +118,7 @@ static std::string GetTopologyName(Tegra::Shader::OutputTopology topology) { } /// Returns true if an object has to be treated as precise -static bool IsPrecise(Operation operand) { +bool IsPrecise(Operation operand) { const auto& meta = operand.GetMeta(); if (const auto arithmetic = std::get_if<MetaArithmetic>(&meta)) { @@ -126,7 +130,7 @@ static bool IsPrecise(Operation operand) { return false; } -static bool IsPrecise(Node node) { +bool IsPrecise(Node node) { if (const auto operation = std::get_if<OperationNode>(node)) { return IsPrecise(*operation); } @@ -202,8 +206,10 @@ public: for (const auto& sampler : ir.GetSamplers()) { entries.samplers.emplace_back(sampler); } - for (const auto& gmem : ir.GetGlobalMemoryBases()) { - entries.global_memory_entries.emplace_back(gmem.cbuf_index, gmem.cbuf_offset); + for (const auto& gmem_pair : ir.GetGlobalMemory()) { + const auto& [base, usage] = gmem_pair; + entries.global_memory_entries.emplace_back(base.cbuf_index, base.cbuf_offset, + usage.is_read, usage.is_written); } entries.clip_distances = ir.GetClipDistances(); entries.shader_length = ir.GetLength(); @@ -374,12 +380,22 @@ private: } void DeclareGlobalMemory() { - for (const auto& entry : ir.GetGlobalMemoryBases()) { + for (const auto& gmem : ir.GetGlobalMemory()) { + const auto& [base, usage] = gmem; + + // Since we don't know how the shader will use the shader, hint the driver to disable as + // much optimizations as possible + std::string qualifier = "coherent volatile"; + if (usage.is_read && !usage.is_written) + qualifier += " readonly"; + else if (usage.is_written && !usage.is_read) + qualifier += " writeonly"; + const std::string binding = - fmt::format("GMEM_BINDING_{}_{}", entry.cbuf_index, entry.cbuf_offset); - code.AddLine("layout (std430, binding = " + binding + ") buffer " + - GetGlobalMemoryBlock(entry) + " {"); - code.AddLine(" float " + GetGlobalMemory(entry) + "[MAX_GLOBALMEMORY_ELEMENTS];"); + fmt::format("GMEM_BINDING_{}_{}", base.cbuf_index, base.cbuf_offset); + code.AddLine("layout (std430, binding = " + binding + ") " + qualifier + " buffer " + + GetGlobalMemoryBlock(base) + " {"); + code.AddLine(" float " + GetGlobalMemory(base) + "[];"); code.AddLine("};"); code.AddNewLine(); } @@ -426,9 +442,14 @@ private: std::string Visit(Node node) { if (const auto operation = std::get_if<OperationNode>(node)) { const auto operation_index = static_cast<std::size_t>(operation->GetCode()); + if (operation_index >= operation_decompilers.size()) { + UNREACHABLE_MSG("Out of bounds operation: {}", operation_index); + return {}; + } const auto decompiler = operation_decompilers[operation_index]; if (decompiler == nullptr) { - UNREACHABLE_MSG("Operation decompiler {} not defined", operation_index); + UNREACHABLE_MSG("Undefined operation: {}", operation_index); + return {}; } return (this->*decompiler)(*operation); @@ -540,9 +561,8 @@ private: } else if (std::holds_alternative<OperationNode>(*offset)) { // Indirect access - const std::string final_offset = code.GenerateTemporal(); - code.AddLine("uint " + final_offset + " = (ftou(" + Visit(offset) + ") / 4) & " + - std::to_string(MAX_CONSTBUFFER_ELEMENTS - 1) + ';'); + const std::string final_offset = code.GenerateTemporary(); + code.AddLine("uint " + final_offset + " = (ftou(" + Visit(offset) + ") / 4);"); return fmt::format("{}[{} / 4][{} % 4]", GetConstBuffer(cbuf->GetIndex()), final_offset, final_offset); @@ -587,9 +607,9 @@ private: // There's a bug in NVidia's proprietary drivers that makes precise fail on fragment shaders const std::string precise = stage != ShaderStage::Fragment ? "precise " : ""; - const std::string temporal = code.GenerateTemporal(); - code.AddLine(precise + "float " + temporal + " = " + value + ';'); - return temporal; + const std::string temporary = code.GenerateTemporary(); + code.AddLine(precise + "float " + temporary + " = " + value + ';'); + return temporary; } std::string VisitOperand(Operation operation, std::size_t operand_index) { @@ -601,9 +621,9 @@ private: return Visit(operand); } - const std::string temporal = code.GenerateTemporal(); - code.AddLine("float " + temporal + " = " + Visit(operand) + ';'); - return temporal; + const std::string temporary = code.GenerateTemporary(); + code.AddLine("float " + temporary + " = " + Visit(operand) + ';'); + return temporary; } std::string VisitOperand(Operation operation, std::size_t operand_index, Type type) { @@ -718,8 +738,8 @@ private: result_type)); } - std::string GenerateTexture(Operation operation, const std::string& func, - const std::vector<std::pair<Type, Node>>& extras) { + std::string GenerateTexture(Operation operation, const std::string& function_suffix, + const std::vector<TextureIR>& extras) { constexpr std::array<const char*, 4> coord_constructors = {"float", "vec2", "vec3", "vec4"}; const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); @@ -729,11 +749,11 @@ private: const bool has_array = meta->sampler.IsArray(); const bool has_shadow = meta->sampler.IsShadow(); - std::string expr = func; - expr += '('; - expr += GetSampler(meta->sampler); - expr += ", "; - + std::string expr = "texture" + function_suffix; + if (!meta->aoffi.empty()) { + expr += "Offset"; + } + expr += '(' + GetSampler(meta->sampler) + ", "; expr += coord_constructors.at(count + (has_array ? 1 : 0) + (has_shadow ? 1 : 0) - 1); expr += '('; for (std::size_t i = 0; i < count; ++i) { @@ -751,36 +771,74 @@ private: } expr += ')'; - for (const auto& extra_pair : extras) { - const auto [type, operand] = extra_pair; - if (operand == nullptr) { - continue; + for (const auto& variant : extras) { + if (const auto argument = std::get_if<TextureArgument>(&variant)) { + expr += GenerateTextureArgument(*argument); + } else if (std::get_if<TextureAoffi>(&variant)) { + expr += GenerateTextureAoffi(meta->aoffi); + } else { + UNREACHABLE(); } - expr += ", "; + } - switch (type) { - case Type::Int: - if (const auto immediate = std::get_if<ImmediateNode>(operand)) { - // Inline the string as an immediate integer in GLSL (some extra arguments are - // required to be constant) - expr += std::to_string(static_cast<s32>(immediate->GetValue())); - } else { - expr += "ftoi(" + Visit(operand) + ')'; - } - break; - case Type::Float: - expr += Visit(operand); - break; - default: { - const auto type_int = static_cast<u32>(type); - UNIMPLEMENTED_MSG("Unimplemented extra type={}", type_int); - expr += '0'; - break; + return expr + ')'; + } + + std::string GenerateTextureArgument(TextureArgument argument) { + const auto [type, operand] = argument; + if (operand == nullptr) { + return {}; + } + + std::string expr = ", "; + switch (type) { + case Type::Int: + if (const auto immediate = std::get_if<ImmediateNode>(operand)) { + // Inline the string as an immediate integer in GLSL (some extra arguments are + // required to be constant) + expr += std::to_string(static_cast<s32>(immediate->GetValue())); + } else { + expr += "ftoi(" + Visit(operand) + ')'; + } + break; + case Type::Float: + expr += Visit(operand); + break; + default: { + const auto type_int = static_cast<u32>(type); + UNIMPLEMENTED_MSG("Unimplemented extra type={}", type_int); + expr += '0'; + break; + } + } + return expr; + } + + std::string GenerateTextureAoffi(const std::vector<Node>& aoffi) { + if (aoffi.empty()) { + return {}; + } + constexpr std::array<const char*, 3> coord_constructors = {"int", "ivec2", "ivec3"}; + std::string expr = ", "; + expr += coord_constructors.at(aoffi.size() - 1); + expr += '('; + + for (std::size_t index = 0; index < aoffi.size(); ++index) { + const auto operand{aoffi.at(index)}; + if (const auto immediate = std::get_if<ImmediateNode>(operand)) { + // Inline the string as an immediate integer in GLSL (AOFFI arguments are required + // to be constant by the standard). + expr += std::to_string(static_cast<s32>(immediate->GetValue())); + } else { + expr += "ftoi(" + Visit(operand) + ')'; } + if (index + 1 < aoffi.size()) { + expr += ", "; } } + expr += ')'; - return expr + ')'; + return expr; } std::string Assign(Operation operation) { @@ -820,6 +878,12 @@ private: } else if (const auto lmem = std::get_if<LmemNode>(dest)) { target = GetLocalMemory() + "[ftou(" + Visit(lmem->GetAddress()) + ") / 4]"; + } else if (const auto gmem = std::get_if<GmemNode>(dest)) { + const std::string real = Visit(gmem->GetRealAddress()); + const std::string base = Visit(gmem->GetBaseAddress()); + const std::string final_offset = "(ftou(" + real + ") - ftou(" + base + ")) / 4"; + target = fmt::format("{}[{}]", GetGlobalMemory(gmem->GetDescriptor()), final_offset); + } else { UNREACHABLE_MSG("Assign called without a proper target"); } @@ -1159,7 +1223,8 @@ private: const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); ASSERT(meta); - std::string expr = GenerateTexture(operation, "texture", {{Type::Float, meta->bias}}); + std::string expr = GenerateTexture( + operation, "", {TextureAoffi{}, TextureArgument{Type::Float, meta->bias}}); if (meta->sampler.IsShadow()) { expr = "vec4(" + expr + ')'; } @@ -1170,7 +1235,8 @@ private: const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); ASSERT(meta); - std::string expr = GenerateTexture(operation, "textureLod", {{Type::Float, meta->lod}}); + std::string expr = GenerateTexture( + operation, "Lod", {TextureArgument{Type::Float, meta->lod}, TextureAoffi{}}); if (meta->sampler.IsShadow()) { expr = "vec4(" + expr + ')'; } @@ -1182,7 +1248,8 @@ private: ASSERT(meta); const auto type = meta->sampler.IsShadow() ? Type::Float : Type::Int; - return GenerateTexture(operation, "textureGather", {{type, meta->component}}) + + return GenerateTexture(operation, "Gather", + {TextureArgument{type, meta->component}, TextureAoffi{}}) + GetSwizzle(meta->element); } @@ -1196,11 +1263,12 @@ private: switch (meta->element) { case 0: case 1: - return "textureSize(" + sampler + ", " + lod + ')' + GetSwizzle(meta->element); + return "itof(int(textureSize(" + sampler + ", " + lod + ')' + + GetSwizzle(meta->element) + "))"; case 2: return "0"; case 3: - return "textureQueryLevels(" + sampler + ')'; + return "itof(textureQueryLevels(" + sampler + "))"; } UNREACHABLE(); return "0"; @@ -1211,8 +1279,8 @@ private: ASSERT(meta); if (meta->element < 2) { - return "itof(int((" + GenerateTexture(operation, "textureQueryLod", {}) + - " * vec2(256))" + GetSwizzle(meta->element) + "))"; + return "itof(int((" + GenerateTexture(operation, "QueryLod", {}) + " * vec2(256))" + + GetSwizzle(meta->element) + "))"; } return "0"; } @@ -1565,11 +1633,11 @@ private: ShaderWriter code; }; +} // Anonymous namespace + std::string GetCommonDeclarations() { const auto cbuf = std::to_string(MAX_CONSTBUFFER_ELEMENTS); - const auto gmem = std::to_string(MAX_GLOBALMEMORY_ELEMENTS); return "#define MAX_CONSTBUFFER_ELEMENTS " + cbuf + "\n" + - "#define MAX_GLOBALMEMORY_ELEMENTS " + gmem + "\n" + "#define ftoi floatBitsToInt\n" "#define ftou floatBitsToUint\n" "#define itof intBitsToFloat\n" diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h index 72aca4938..55b3d4d7b 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.h +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h @@ -5,7 +5,6 @@ #pragma once #include <array> -#include <set> #include <string> #include <utility> #include <vector> @@ -40,8 +39,9 @@ private: class GlobalMemoryEntry { public: - explicit GlobalMemoryEntry(u32 cbuf_index, u32 cbuf_offset) - : cbuf_index{cbuf_index}, cbuf_offset{cbuf_offset} {} + explicit GlobalMemoryEntry(u32 cbuf_index, u32 cbuf_offset, bool is_read, bool is_written) + : cbuf_index{cbuf_index}, cbuf_offset{cbuf_offset}, is_read{is_read}, is_written{ + is_written} {} u32 GetCbufIndex() const { return cbuf_index; @@ -51,9 +51,19 @@ public: return cbuf_offset; } + bool IsRead() const { + return is_read; + } + + bool IsWritten() const { + return is_written; + } + private: u32 cbuf_index{}; u32 cbuf_offset{}; + bool is_read{}; + bool is_written{}; }; struct ShaderEntries { diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp index 82fc4d44b..d5890a375 100644 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp @@ -4,7 +4,6 @@ #include <cstring> #include <fmt/format.h> -#include <lz4.h> #include "common/assert.h" #include "common/common_paths.h" @@ -12,6 +11,7 @@ #include "common/file_util.h" #include "common/logging/log.h" #include "common/scm_rev.h" +#include "common/zstd_compression.h" #include "core/core.h" #include "core/hle/kernel/process.h" @@ -49,39 +49,6 @@ ShaderCacheVersionHash GetShaderCacheVersionHash() { return hash; } -template <typename T> -std::vector<u8> CompressData(const T* source, std::size_t source_size) { - if (source_size > LZ4_MAX_INPUT_SIZE) { - // Source size exceeds LZ4 maximum input size - return {}; - } - const auto source_size_int = static_cast<int>(source_size); - const int max_compressed_size = LZ4_compressBound(source_size_int); - std::vector<u8> compressed(max_compressed_size); - const int compressed_size = LZ4_compress_default(reinterpret_cast<const char*>(source), - reinterpret_cast<char*>(compressed.data()), - source_size_int, max_compressed_size); - if (compressed_size <= 0) { - // Compression failed - return {}; - } - compressed.resize(compressed_size); - return compressed; -} - -std::vector<u8> DecompressData(const std::vector<u8>& compressed, std::size_t uncompressed_size) { - std::vector<u8> uncompressed(uncompressed_size); - const int size_check = LZ4_decompress_safe(reinterpret_cast<const char*>(compressed.data()), - reinterpret_cast<char*>(uncompressed.data()), - static_cast<int>(compressed.size()), - static_cast<int>(uncompressed.size())); - if (static_cast<int>(uncompressed_size) != size_check) { - // Decompression failed - return {}; - } - return uncompressed; -} - } // namespace ShaderDiskCacheRaw::ShaderDiskCacheRaw(u64 unique_identifier, Maxwell::ShaderProgram program_type, @@ -292,7 +259,7 @@ ShaderDiskCacheOpenGL::LoadPrecompiledFile(FileUtil::IOFile& file) { return {}; } - dump.binary = DecompressData(compressed_binary, binary_length); + dump.binary = Common::Compression::DecompressDataZSTD(compressed_binary); if (dump.binary.empty()) { return {}; } @@ -321,7 +288,7 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn return {}; } - const std::vector<u8> code = DecompressData(compressed_code, code_size); + const std::vector<u8> code = Common::Compression::DecompressDataZSTD(compressed_code); if (code.empty()) { return {}; } @@ -370,11 +337,16 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn for (u32 i = 0; i < global_memory_count; ++i) { u32 cbuf_index{}; u32 cbuf_offset{}; + u8 is_read{}; + u8 is_written{}; if (file.ReadBytes(&cbuf_index, sizeof(u32)) != sizeof(u32) || - file.ReadBytes(&cbuf_offset, sizeof(u32)) != sizeof(u32)) { + file.ReadBytes(&cbuf_offset, sizeof(u32)) != sizeof(u32) || + file.ReadBytes(&is_read, sizeof(u8)) != sizeof(u8) || + file.ReadBytes(&is_written, sizeof(u8)) != sizeof(u8)) { return {}; } - entry.entries.global_memory_entries.emplace_back(cbuf_index, cbuf_offset); + entry.entries.global_memory_entries.emplace_back(cbuf_index, cbuf_offset, is_read != 0, + is_written != 0); } for (auto& clip_distance : entry.entries.clip_distances) { @@ -430,7 +402,9 @@ bool ShaderDiskCacheOpenGL::SaveDecompiledFile(FileUtil::IOFile& file, u64 uniqu return false; for (const auto& gmem : entries.global_memory_entries) { if (file.WriteObject(static_cast<u32>(gmem.GetCbufIndex())) != 1 || - file.WriteObject(static_cast<u32>(gmem.GetCbufOffset())) != 1) { + file.WriteObject(static_cast<u32>(gmem.GetCbufOffset())) != 1 || + file.WriteObject(static_cast<u8>(gmem.IsRead() ? 1 : 0)) != 1 || + file.WriteObject(static_cast<u8>(gmem.IsWritten() ? 1 : 0)) != 1) { return false; } } @@ -507,7 +481,8 @@ void ShaderDiskCacheOpenGL::SaveDecompiled(u64 unique_identifier, const std::str if (!IsUsable()) return; - const std::vector<u8> compressed_code{CompressData(code.data(), code.size())}; + const std::vector<u8> compressed_code{Common::Compression::CompressDataZSTDDefault( + reinterpret_cast<const u8*>(code.data()), code.size())}; if (compressed_code.empty()) { LOG_ERROR(Render_OpenGL, "Failed to compress GLSL code - skipping shader {:016x}", unique_identifier); @@ -537,7 +512,9 @@ void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint p std::vector<u8> binary(binary_length); glGetProgramBinary(program, binary_length, nullptr, &binary_format, binary.data()); - const std::vector<u8> compressed_binary = CompressData(binary.data(), binary.size()); + const std::vector<u8> compressed_binary = + Common::Compression::CompressDataZSTDDefault(binary.data(), binary.size()); + if (compressed_binary.empty()) { LOG_ERROR(Render_OpenGL, "Failed to compress binary program in shader={:016x}", usage.unique_identifier); diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp index 7d96649af..8763d9c71 100644 --- a/src/video_core/renderer_opengl/gl_shader_gen.cpp +++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp @@ -3,7 +3,6 @@ // Refer to the license.txt file included. #include <fmt/format.h> -#include "common/assert.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" #include "video_core/renderer_opengl/gl_shader_gen.h" diff --git a/src/video_core/renderer_opengl/gl_shader_gen.h b/src/video_core/renderer_opengl/gl_shader_gen.h index fba8e681b..fad346b48 100644 --- a/src/video_core/renderer_opengl/gl_shader_gen.h +++ b/src/video_core/renderer_opengl/gl_shader_gen.h @@ -4,12 +4,9 @@ #pragma once -#include <array> -#include <string> #include <vector> #include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" #include "video_core/shader/shader_ir.h" diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp index 6a30c28d2..05ab01dcb 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.cpp +++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp @@ -2,23 +2,55 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include "core/core.h" +#include "common/common_types.h" +#include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_opengl/gl_shader_manager.h" namespace OpenGL::GLShader { -void MaxwellUniformData::SetFromRegs(const Maxwell3D::State::ShaderStageInfo& shader_stage) { - const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); - const auto& regs = gpu.regs; - const auto& state = gpu.state; +using Tegra::Engines::Maxwell3D; + +ProgramManager::ProgramManager() { + pipeline.Create(); +} + +ProgramManager::~ProgramManager() = default; + +void ProgramManager::ApplyTo(OpenGLState& state) { + UpdatePipeline(); + state.draw.shader_program = 0; + state.draw.program_pipeline = pipeline.handle; +} + +void ProgramManager::UpdatePipeline() { + // Avoid updating the pipeline when values have no changed + if (old_state == current_state) { + return; + } + + // Workaround for AMD bug + constexpr GLenum all_used_stages{GL_VERTEX_SHADER_BIT | GL_GEOMETRY_SHADER_BIT | + GL_FRAGMENT_SHADER_BIT}; + glUseProgramStages(pipeline.handle, all_used_stages, 0); + + glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, current_state.vertex_shader); + glUseProgramStages(pipeline.handle, GL_GEOMETRY_SHADER_BIT, current_state.geometry_shader); + glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, current_state.fragment_shader); + + old_state = current_state; +} + +void MaxwellUniformData::SetFromRegs(const Maxwell3D& maxwell, std::size_t shader_stage) { + const auto& regs = maxwell.regs; + const auto& state = maxwell.state; // TODO(bunnei): Support more than one viewport viewport_flip[0] = regs.viewport_transform[0].scale_x < 0.0 ? -1.0f : 1.0f; viewport_flip[1] = regs.viewport_transform[0].scale_y < 0.0 ? -1.0f : 1.0f; - u32 func = static_cast<u32>(regs.alpha_test_func); + auto func{static_cast<u32>(regs.alpha_test_func)}; // Normalize the gl variants of opCompare to be the same as the normal variants - u32 op_gl_variant_base = static_cast<u32>(Tegra::Engines::Maxwell3D::Regs::ComparisonOp::Never); + const u32 op_gl_variant_base = static_cast<u32>(Maxwell3D::Regs::ComparisonOp::Never); if (func >= op_gl_variant_base) { func = func - op_gl_variant_base + 1U; } @@ -31,8 +63,9 @@ void MaxwellUniformData::SetFromRegs(const Maxwell3D::State::ShaderStageInfo& sh // Assign in which stage the position has to be flipped // (the last stage before the fragment shader). - if (gpu.regs.shader_config[static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry)].enable) { - flip_stage = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry); + constexpr u32 geometry_index = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry); + if (maxwell.regs.shader_config[geometry_index].enable) { + flip_stage = geometry_index; } else { flip_stage = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::VertexB); } diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h index 4970aafed..cec18a832 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.h +++ b/src/video_core/renderer_opengl/gl_shader_manager.h @@ -4,6 +4,8 @@ #pragma once +#include <cstddef> + #include <glad/glad.h> #include "video_core/renderer_opengl/gl_resource_manager.h" @@ -12,14 +14,13 @@ namespace OpenGL::GLShader { -using Tegra::Engines::Maxwell3D; - /// Uniform structure for the Uniform Buffer Object, all vectors must be 16-byte aligned -// NOTE: Always keep a vec4 at the end. The GL spec is not clear whether the alignment at -// the end of a uniform block is included in UNIFORM_BLOCK_DATA_SIZE or not. -// Not following that rule will cause problems on some AMD drivers. +/// @note Always keep a vec4 at the end. The GL spec is not clear whether the alignment at +/// the end of a uniform block is included in UNIFORM_BLOCK_DATA_SIZE or not. +/// Not following that rule will cause problems on some AMD drivers. struct MaxwellUniformData { - void SetFromRegs(const Maxwell3D::State::ShaderStageInfo& shader_stage); + void SetFromRegs(const Tegra::Engines::Maxwell3D& maxwell, std::size_t shader_stage); + alignas(16) GLvec4 viewport_flip; struct alignas(16) { GLuint instance_id; @@ -39,56 +40,48 @@ static_assert(sizeof(MaxwellUniformData) < 16384, class ProgramManager { public: - ProgramManager() { - pipeline.Create(); - } + explicit ProgramManager(); + ~ProgramManager(); + + void ApplyTo(OpenGLState& state); void UseProgrammableVertexShader(GLuint program) { - vs = program; + current_state.vertex_shader = program; } void UseProgrammableGeometryShader(GLuint program) { - gs = program; + current_state.geometry_shader = program; } void UseProgrammableFragmentShader(GLuint program) { - fs = program; + current_state.fragment_shader = program; } void UseTrivialGeometryShader() { - gs = 0; - } - - void ApplyTo(OpenGLState& state) { - UpdatePipeline(); - state.draw.shader_program = 0; - state.draw.program_pipeline = pipeline.handle; - state.geometry_shaders.enabled = (gs != 0); + current_state.geometry_shader = 0; } private: - void UpdatePipeline() { - // Avoid updating the pipeline when values have no changed - if (old_vs == vs && old_fs == fs && old_gs == gs) - return; - // Workaround for AMD bug - glUseProgramStages(pipeline.handle, - GL_VERTEX_SHADER_BIT | GL_GEOMETRY_SHADER_BIT | GL_FRAGMENT_SHADER_BIT, - 0); - - glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vs); - glUseProgramStages(pipeline.handle, GL_GEOMETRY_SHADER_BIT, gs); - glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fs); - - // Update the old values - old_vs = vs; - old_fs = fs; - old_gs = gs; - } + struct PipelineState { + bool operator==(const PipelineState& rhs) const { + return vertex_shader == rhs.vertex_shader && fragment_shader == rhs.fragment_shader && + geometry_shader == rhs.geometry_shader; + } + + bool operator!=(const PipelineState& rhs) const { + return !operator==(rhs); + } + + GLuint vertex_shader{}; + GLuint fragment_shader{}; + GLuint geometry_shader{}; + }; + + void UpdatePipeline(); OGLPipeline pipeline; - GLuint vs{}, fs{}, gs{}; - GLuint old_vs{}, old_fs{}, old_gs{}; + PipelineState current_state; + PipelineState old_state; }; } // namespace OpenGL::GLShader diff --git a/src/video_core/renderer_opengl/gl_state.cpp b/src/video_core/renderer_opengl/gl_state.cpp index 9419326a3..52d569a1b 100644 --- a/src/video_core/renderer_opengl/gl_state.cpp +++ b/src/video_core/renderer_opengl/gl_state.cpp @@ -10,16 +10,62 @@ namespace OpenGL { -OpenGLState OpenGLState::cur_state; +using Maxwell = Tegra::Engines::Maxwell3D::Regs; +OpenGLState OpenGLState::cur_state; bool OpenGLState::s_rgb_used; +namespace { + +template <typename T> +bool UpdateValue(T& current_value, const T new_value) { + const bool changed = current_value != new_value; + current_value = new_value; + return changed; +} + +template <typename T1, typename T2> +bool UpdateTie(T1 current_value, const T2 new_value) { + const bool changed = current_value != new_value; + current_value = new_value; + return changed; +} + +void Enable(GLenum cap, bool enable) { + if (enable) { + glEnable(cap); + } else { + glDisable(cap); + } +} + +void Enable(GLenum cap, GLuint index, bool enable) { + if (enable) { + glEnablei(cap, index); + } else { + glDisablei(cap, index); + } +} + +void Enable(GLenum cap, bool& current_value, bool new_value) { + if (UpdateValue(current_value, new_value)) + Enable(cap, new_value); +} + +void Enable(GLenum cap, GLuint index, bool& current_value, bool new_value) { + if (UpdateValue(current_value, new_value)) + Enable(cap, index, new_value); +} + +} // namespace + OpenGLState::OpenGLState() { // These all match default OpenGL values - geometry_shaders.enabled = false; framebuffer_srgb.enabled = false; + multisample_control.alpha_to_coverage = false; multisample_control.alpha_to_one = false; + cull.enabled = false; cull.mode = GL_BACK; cull.front_face = GL_CCW; @@ -30,14 +76,15 @@ OpenGLState::OpenGLState() { primitive_restart.enabled = false; primitive_restart.index = 0; + for (auto& item : color_mask) { item.red_enabled = GL_TRUE; item.green_enabled = GL_TRUE; item.blue_enabled = GL_TRUE; item.alpha_enabled = GL_TRUE; } - stencil.test_enabled = false; - auto reset_stencil = [](auto& config) { + + const auto ResetStencil = [](auto& config) { config.test_func = GL_ALWAYS; config.test_ref = 0; config.test_mask = 0xFFFFFFFF; @@ -46,8 +93,10 @@ OpenGLState::OpenGLState() { config.action_depth_pass = GL_KEEP; config.action_stencil_fail = GL_KEEP; }; - reset_stencil(stencil.front); - reset_stencil(stencil.back); + stencil.test_enabled = false; + ResetStencil(stencil.front); + ResetStencil(stencil.back); + for (auto& item : viewports) { item.x = 0; item.y = 0; @@ -61,6 +110,7 @@ OpenGLState::OpenGLState() { item.scissor.width = 0; item.scissor.height = 0; } + for (auto& item : blend) { item.enabled = true; item.rgb_equation = GL_FUNC_ADD; @@ -70,11 +120,14 @@ OpenGLState::OpenGLState() { item.src_a_func = GL_ONE; item.dst_a_func = GL_ZERO; } + independant_blend.enabled = false; + blend_color.red = 0.0f; blend_color.green = 0.0f; blend_color.blue = 0.0f; blend_color.alpha = 0.0f; + logic_op.enabled = false; logic_op.operation = GL_COPY; @@ -91,9 +144,12 @@ OpenGLState::OpenGLState() { clip_distance = {}; point.size = 1; + fragment_color_clamp.enabled = false; + depth_clamp.far_plane = false; depth_clamp.near_plane = false; + polygon_offset.fill_enable = false; polygon_offset.line_enable = false; polygon_offset.point_enable = false; @@ -103,260 +159,255 @@ OpenGLState::OpenGLState() { } void OpenGLState::ApplyDefaultState() { + glEnable(GL_BLEND); glDisable(GL_FRAMEBUFFER_SRGB); glDisable(GL_CULL_FACE); glDisable(GL_DEPTH_TEST); glDisable(GL_PRIMITIVE_RESTART); glDisable(GL_STENCIL_TEST); - glEnable(GL_BLEND); glDisable(GL_COLOR_LOGIC_OP); glDisable(GL_SCISSOR_TEST); } +void OpenGLState::ApplyFramebufferState() const { + if (UpdateValue(cur_state.draw.read_framebuffer, draw.read_framebuffer)) { + glBindFramebuffer(GL_READ_FRAMEBUFFER, draw.read_framebuffer); + } + if (UpdateValue(cur_state.draw.draw_framebuffer, draw.draw_framebuffer)) { + glBindFramebuffer(GL_DRAW_FRAMEBUFFER, draw.draw_framebuffer); + } +} + +void OpenGLState::ApplyVertexArrayState() const { + if (UpdateValue(cur_state.draw.vertex_array, draw.vertex_array)) { + glBindVertexArray(draw.vertex_array); + } +} + +void OpenGLState::ApplyShaderProgram() const { + if (UpdateValue(cur_state.draw.shader_program, draw.shader_program)) { + glUseProgram(draw.shader_program); + } +} + +void OpenGLState::ApplyProgramPipeline() const { + if (UpdateValue(cur_state.draw.program_pipeline, draw.program_pipeline)) { + glBindProgramPipeline(draw.program_pipeline); + } +} + +void OpenGLState::ApplyClipDistances() const { + for (std::size_t i = 0; i < clip_distance.size(); ++i) { + Enable(GL_CLIP_DISTANCE0 + static_cast<GLenum>(i), cur_state.clip_distance[i], + clip_distance[i]); + } +} + +void OpenGLState::ApplyPointSize() const { + if (UpdateValue(cur_state.point.size, point.size)) { + glPointSize(point.size); + } +} + +void OpenGLState::ApplyFragmentColorClamp() const { + if (UpdateValue(cur_state.fragment_color_clamp.enabled, fragment_color_clamp.enabled)) { + glClampColor(GL_CLAMP_FRAGMENT_COLOR_ARB, + fragment_color_clamp.enabled ? GL_TRUE : GL_FALSE); + } +} + +void OpenGLState::ApplyMultisample() const { + Enable(GL_SAMPLE_ALPHA_TO_COVERAGE, cur_state.multisample_control.alpha_to_coverage, + multisample_control.alpha_to_coverage); + Enable(GL_SAMPLE_ALPHA_TO_ONE, cur_state.multisample_control.alpha_to_one, + multisample_control.alpha_to_one); +} + +void OpenGLState::ApplyDepthClamp() const { + if (depth_clamp.far_plane == cur_state.depth_clamp.far_plane && + depth_clamp.near_plane == cur_state.depth_clamp.near_plane) { + return; + } + cur_state.depth_clamp = depth_clamp; + + UNIMPLEMENTED_IF_MSG(depth_clamp.far_plane != depth_clamp.near_plane, + "Unimplemented Depth Clamp Separation!"); + + Enable(GL_DEPTH_CLAMP, depth_clamp.far_plane || depth_clamp.near_plane); +} + void OpenGLState::ApplySRgb() const { - if (framebuffer_srgb.enabled != cur_state.framebuffer_srgb.enabled) { - if (framebuffer_srgb.enabled) { - // Track if sRGB is used - s_rgb_used = true; - glEnable(GL_FRAMEBUFFER_SRGB); - } else { - glDisable(GL_FRAMEBUFFER_SRGB); - } + if (cur_state.framebuffer_srgb.enabled == framebuffer_srgb.enabled) + return; + cur_state.framebuffer_srgb.enabled = framebuffer_srgb.enabled; + if (framebuffer_srgb.enabled) { + // Track if sRGB is used + s_rgb_used = true; + glEnable(GL_FRAMEBUFFER_SRGB); + } else { + glDisable(GL_FRAMEBUFFER_SRGB); } } void OpenGLState::ApplyCulling() const { - if (cull.enabled != cur_state.cull.enabled) { - if (cull.enabled) { - glEnable(GL_CULL_FACE); - } else { - glDisable(GL_CULL_FACE); - } - } + Enable(GL_CULL_FACE, cur_state.cull.enabled, cull.enabled); - if (cull.mode != cur_state.cull.mode) { + if (UpdateValue(cur_state.cull.mode, cull.mode)) { glCullFace(cull.mode); } - if (cull.front_face != cur_state.cull.front_face) { + if (UpdateValue(cur_state.cull.front_face, cull.front_face)) { glFrontFace(cull.front_face); } } void OpenGLState::ApplyColorMask() const { - if (independant_blend.enabled) { - for (size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) { - const auto& updated = color_mask[i]; - const auto& current = cur_state.color_mask[i]; - if (updated.red_enabled != current.red_enabled || - updated.green_enabled != current.green_enabled || - updated.blue_enabled != current.blue_enabled || - updated.alpha_enabled != current.alpha_enabled) { - glColorMaski(static_cast<GLuint>(i), updated.red_enabled, updated.green_enabled, - updated.blue_enabled, updated.alpha_enabled); - } - } - } else { - const auto& updated = color_mask[0]; - const auto& current = cur_state.color_mask[0]; + for (std::size_t i = 0; i < Maxwell::NumRenderTargets; ++i) { + const auto& updated = color_mask[i]; + auto& current = cur_state.color_mask[i]; if (updated.red_enabled != current.red_enabled || updated.green_enabled != current.green_enabled || updated.blue_enabled != current.blue_enabled || updated.alpha_enabled != current.alpha_enabled) { - glColorMask(updated.red_enabled, updated.green_enabled, updated.blue_enabled, - updated.alpha_enabled); + current = updated; + glColorMaski(static_cast<GLuint>(i), updated.red_enabled, updated.green_enabled, + updated.blue_enabled, updated.alpha_enabled); } } } void OpenGLState::ApplyDepth() const { - if (depth.test_enabled != cur_state.depth.test_enabled) { - if (depth.test_enabled) { - glEnable(GL_DEPTH_TEST); - } else { - glDisable(GL_DEPTH_TEST); - } - } + Enable(GL_DEPTH_TEST, cur_state.depth.test_enabled, depth.test_enabled); - if (depth.test_func != cur_state.depth.test_func) { + if (cur_state.depth.test_func != depth.test_func) { + cur_state.depth.test_func = depth.test_func; glDepthFunc(depth.test_func); } - if (depth.write_mask != cur_state.depth.write_mask) { + if (cur_state.depth.write_mask != depth.write_mask) { + cur_state.depth.write_mask = depth.write_mask; glDepthMask(depth.write_mask); } } void OpenGLState::ApplyPrimitiveRestart() const { - if (primitive_restart.enabled != cur_state.primitive_restart.enabled) { - if (primitive_restart.enabled) { - glEnable(GL_PRIMITIVE_RESTART); - } else { - glDisable(GL_PRIMITIVE_RESTART); - } - } + Enable(GL_PRIMITIVE_RESTART, cur_state.primitive_restart.enabled, primitive_restart.enabled); - if (primitive_restart.index != cur_state.primitive_restart.index) { + if (cur_state.primitive_restart.index != primitive_restart.index) { + cur_state.primitive_restart.index = primitive_restart.index; glPrimitiveRestartIndex(primitive_restart.index); } } void OpenGLState::ApplyStencilTest() const { - if (stencil.test_enabled != cur_state.stencil.test_enabled) { - if (stencil.test_enabled) { - glEnable(GL_STENCIL_TEST); - } else { - glDisable(GL_STENCIL_TEST); - } - } - - const auto ConfigStencil = [](GLenum face, const auto& config, const auto& prev_config) { - if (config.test_func != prev_config.test_func || config.test_ref != prev_config.test_ref || - config.test_mask != prev_config.test_mask) { + Enable(GL_STENCIL_TEST, cur_state.stencil.test_enabled, stencil.test_enabled); + + const auto ConfigStencil = [](GLenum face, const auto& config, auto& current) { + if (current.test_func != config.test_func || current.test_ref != config.test_ref || + current.test_mask != config.test_mask) { + current.test_func = config.test_func; + current.test_ref = config.test_ref; + current.test_mask = config.test_mask; glStencilFuncSeparate(face, config.test_func, config.test_ref, config.test_mask); } - if (config.action_depth_fail != prev_config.action_depth_fail || - config.action_depth_pass != prev_config.action_depth_pass || - config.action_stencil_fail != prev_config.action_stencil_fail) { + if (current.action_depth_fail != config.action_depth_fail || + current.action_depth_pass != config.action_depth_pass || + current.action_stencil_fail != config.action_stencil_fail) { + current.action_depth_fail = config.action_depth_fail; + current.action_depth_pass = config.action_depth_pass; + current.action_stencil_fail = config.action_stencil_fail; glStencilOpSeparate(face, config.action_stencil_fail, config.action_depth_fail, config.action_depth_pass); } - if (config.write_mask != prev_config.write_mask) { + if (current.write_mask != config.write_mask) { + current.write_mask = config.write_mask; glStencilMaskSeparate(face, config.write_mask); } }; ConfigStencil(GL_FRONT, stencil.front, cur_state.stencil.front); ConfigStencil(GL_BACK, stencil.back, cur_state.stencil.back); } -// Viewport does not affects glClearBuffer so emulate viewport using scissor test -void OpenGLState::EmulateViewportWithScissor() { - auto& current = viewports[0]; - if (current.scissor.enabled) { - const GLint left = std::max(current.x, current.scissor.x); - const GLint right = - std::max(current.x + current.width, current.scissor.x + current.scissor.width); - const GLint bottom = std::max(current.y, current.scissor.y); - const GLint top = - std::max(current.y + current.height, current.scissor.y + current.scissor.height); - current.scissor.x = std::max(left, 0); - current.scissor.y = std::max(bottom, 0); - current.scissor.width = std::max(right - left, 0); - current.scissor.height = std::max(top - bottom, 0); - } else { - current.scissor.enabled = true; - current.scissor.x = current.x; - current.scissor.y = current.y; - current.scissor.width = current.width; - current.scissor.height = current.height; - } -} void OpenGLState::ApplyViewport() const { - if (geometry_shaders.enabled) { - for (GLuint i = 0; i < static_cast<GLuint>(Tegra::Engines::Maxwell3D::Regs::NumViewports); - i++) { - const auto& current = cur_state.viewports[i]; - const auto& updated = viewports[i]; - if (updated.x != current.x || updated.y != current.y || - updated.width != current.width || updated.height != current.height) { - glViewportIndexedf( - i, static_cast<GLfloat>(updated.x), static_cast<GLfloat>(updated.y), - static_cast<GLfloat>(updated.width), static_cast<GLfloat>(updated.height)); - } - if (updated.depth_range_near != current.depth_range_near || - updated.depth_range_far != current.depth_range_far) { - glDepthRangeIndexed(i, updated.depth_range_near, updated.depth_range_far); - } - - if (updated.scissor.enabled != current.scissor.enabled) { - if (updated.scissor.enabled) { - glEnablei(GL_SCISSOR_TEST, i); - } else { - glDisablei(GL_SCISSOR_TEST, i); - } - } - - if (updated.scissor.x != current.scissor.x || updated.scissor.y != current.scissor.y || - updated.scissor.width != current.scissor.width || - updated.scissor.height != current.scissor.height) { - glScissorIndexed(i, updated.scissor.x, updated.scissor.y, updated.scissor.width, - updated.scissor.height); - } - } - } else { - const auto& current = cur_state.viewports[0]; - const auto& updated = viewports[0]; - if (updated.x != current.x || updated.y != current.y || updated.width != current.width || - updated.height != current.height) { - glViewport(updated.x, updated.y, updated.width, updated.height); - } - - if (updated.depth_range_near != current.depth_range_near || - updated.depth_range_far != current.depth_range_far) { - glDepthRange(updated.depth_range_near, updated.depth_range_far); + for (GLuint i = 0; i < static_cast<GLuint>(Maxwell::NumViewports); ++i) { + const auto& updated = viewports[i]; + auto& current = cur_state.viewports[i]; + + if (current.x != updated.x || current.y != updated.y || current.width != updated.width || + current.height != updated.height) { + current.x = updated.x; + current.y = updated.y; + current.width = updated.width; + current.height = updated.height; + glViewportIndexedf(i, static_cast<GLfloat>(updated.x), static_cast<GLfloat>(updated.y), + static_cast<GLfloat>(updated.width), + static_cast<GLfloat>(updated.height)); } - - if (updated.scissor.enabled != current.scissor.enabled) { - if (updated.scissor.enabled) { - glEnable(GL_SCISSOR_TEST); - } else { - glDisable(GL_SCISSOR_TEST); - } + if (current.depth_range_near != updated.depth_range_near || + current.depth_range_far != updated.depth_range_far) { + current.depth_range_near = updated.depth_range_near; + current.depth_range_far = updated.depth_range_far; + glDepthRangeIndexed(i, updated.depth_range_near, updated.depth_range_far); } - if (updated.scissor.x != current.scissor.x || updated.scissor.y != current.scissor.y || - updated.scissor.width != current.scissor.width || - updated.scissor.height != current.scissor.height) { - glScissor(updated.scissor.x, updated.scissor.y, updated.scissor.width, - updated.scissor.height); + Enable(GL_SCISSOR_TEST, i, current.scissor.enabled, updated.scissor.enabled); + + if (current.scissor.x != updated.scissor.x || current.scissor.y != updated.scissor.y || + current.scissor.width != updated.scissor.width || + current.scissor.height != updated.scissor.height) { + current.scissor.x = updated.scissor.x; + current.scissor.y = updated.scissor.y; + current.scissor.width = updated.scissor.width; + current.scissor.height = updated.scissor.height; + glScissorIndexed(i, updated.scissor.x, updated.scissor.y, updated.scissor.width, + updated.scissor.height); } } } void OpenGLState::ApplyGlobalBlending() const { - const Blend& current = cur_state.blend[0]; const Blend& updated = blend[0]; - if (updated.enabled != current.enabled) { - if (updated.enabled) { - glEnable(GL_BLEND); - } else { - glDisable(GL_BLEND); - } - } - if (!updated.enabled) { - return; - } - if (updated.src_rgb_func != current.src_rgb_func || - updated.dst_rgb_func != current.dst_rgb_func || updated.src_a_func != current.src_a_func || - updated.dst_a_func != current.dst_a_func) { + Blend& current = cur_state.blend[0]; + + Enable(GL_BLEND, current.enabled, updated.enabled); + + if (current.src_rgb_func != updated.src_rgb_func || + current.dst_rgb_func != updated.dst_rgb_func || current.src_a_func != updated.src_a_func || + current.dst_a_func != updated.dst_a_func) { + current.src_rgb_func = updated.src_rgb_func; + current.dst_rgb_func = updated.dst_rgb_func; + current.src_a_func = updated.src_a_func; + current.dst_a_func = updated.dst_a_func; glBlendFuncSeparate(updated.src_rgb_func, updated.dst_rgb_func, updated.src_a_func, updated.dst_a_func); } - if (updated.rgb_equation != current.rgb_equation || updated.a_equation != current.a_equation) { + if (current.rgb_equation != updated.rgb_equation || current.a_equation != updated.a_equation) { + current.rgb_equation = updated.rgb_equation; + current.a_equation = updated.a_equation; glBlendEquationSeparate(updated.rgb_equation, updated.a_equation); } } void OpenGLState::ApplyTargetBlending(std::size_t target, bool force) const { const Blend& updated = blend[target]; - const Blend& current = cur_state.blend[target]; - if (updated.enabled != current.enabled || force) { - if (updated.enabled) { - glEnablei(GL_BLEND, static_cast<GLuint>(target)); - } else { - glDisablei(GL_BLEND, static_cast<GLuint>(target)); - } + Blend& current = cur_state.blend[target]; + + if (current.enabled != updated.enabled || force) { + current.enabled = updated.enabled; + Enable(GL_BLEND, static_cast<GLuint>(target), updated.enabled); } - if (updated.src_rgb_func != current.src_rgb_func || - updated.dst_rgb_func != current.dst_rgb_func || updated.src_a_func != current.src_a_func || - updated.dst_a_func != current.dst_a_func) { + if (UpdateTie(std::tie(current.src_rgb_func, current.dst_rgb_func, current.src_a_func, + current.dst_a_func), + std::tie(updated.src_rgb_func, updated.dst_rgb_func, updated.src_a_func, + updated.dst_a_func))) { glBlendFuncSeparatei(static_cast<GLuint>(target), updated.src_rgb_func, updated.dst_rgb_func, updated.src_a_func, updated.dst_a_func); } - if (updated.rgb_equation != current.rgb_equation || updated.a_equation != current.a_equation) { + if (UpdateTie(std::tie(current.rgb_equation, current.a_equation), + std::tie(updated.rgb_equation, updated.a_equation))) { glBlendEquationSeparatei(static_cast<GLuint>(target), updated.rgb_equation, updated.a_equation); } @@ -364,77 +415,48 @@ void OpenGLState::ApplyTargetBlending(std::size_t target, bool force) const { void OpenGLState::ApplyBlending() const { if (independant_blend.enabled) { - for (size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) { - ApplyTargetBlending(i, - independant_blend.enabled != cur_state.independant_blend.enabled); + const bool force = independant_blend.enabled != cur_state.independant_blend.enabled; + for (std::size_t target = 0; target < Maxwell::NumRenderTargets; ++target) { + ApplyTargetBlending(target, force); } } else { ApplyGlobalBlending(); } - if (blend_color.red != cur_state.blend_color.red || - blend_color.green != cur_state.blend_color.green || - blend_color.blue != cur_state.blend_color.blue || - blend_color.alpha != cur_state.blend_color.alpha) { + cur_state.independant_blend.enabled = independant_blend.enabled; + + if (UpdateTie( + std::tie(cur_state.blend_color.red, cur_state.blend_color.green, + cur_state.blend_color.blue, cur_state.blend_color.alpha), + std::tie(blend_color.red, blend_color.green, blend_color.blue, blend_color.alpha))) { glBlendColor(blend_color.red, blend_color.green, blend_color.blue, blend_color.alpha); } } void OpenGLState::ApplyLogicOp() const { - if (logic_op.enabled != cur_state.logic_op.enabled) { - if (logic_op.enabled) { - glEnable(GL_COLOR_LOGIC_OP); - } else { - glDisable(GL_COLOR_LOGIC_OP); - } - } + Enable(GL_COLOR_LOGIC_OP, cur_state.logic_op.enabled, logic_op.enabled); - if (logic_op.operation != cur_state.logic_op.operation) { + if (UpdateValue(cur_state.logic_op.operation, logic_op.operation)) { glLogicOp(logic_op.operation); } } void OpenGLState::ApplyPolygonOffset() const { - const bool fill_enable_changed = - polygon_offset.fill_enable != cur_state.polygon_offset.fill_enable; - const bool line_enable_changed = - polygon_offset.line_enable != cur_state.polygon_offset.line_enable; - const bool point_enable_changed = - polygon_offset.point_enable != cur_state.polygon_offset.point_enable; - const bool factor_changed = polygon_offset.factor != cur_state.polygon_offset.factor; - const bool units_changed = polygon_offset.units != cur_state.polygon_offset.units; - const bool clamp_changed = polygon_offset.clamp != cur_state.polygon_offset.clamp; - - if (fill_enable_changed) { - if (polygon_offset.fill_enable) { - glEnable(GL_POLYGON_OFFSET_FILL); - } else { - glDisable(GL_POLYGON_OFFSET_FILL); - } - } - - if (line_enable_changed) { - if (polygon_offset.line_enable) { - glEnable(GL_POLYGON_OFFSET_LINE); - } else { - glDisable(GL_POLYGON_OFFSET_LINE); - } - } - - if (point_enable_changed) { - if (polygon_offset.point_enable) { - glEnable(GL_POLYGON_OFFSET_POINT); - } else { - glDisable(GL_POLYGON_OFFSET_POINT); - } - } - - if (factor_changed || units_changed || clamp_changed) { + Enable(GL_POLYGON_OFFSET_FILL, cur_state.polygon_offset.fill_enable, + polygon_offset.fill_enable); + Enable(GL_POLYGON_OFFSET_LINE, cur_state.polygon_offset.line_enable, + polygon_offset.line_enable); + Enable(GL_POLYGON_OFFSET_POINT, cur_state.polygon_offset.point_enable, + polygon_offset.point_enable); + + if (UpdateTie(std::tie(cur_state.polygon_offset.factor, cur_state.polygon_offset.units, + cur_state.polygon_offset.clamp), + std::tie(polygon_offset.factor, polygon_offset.units, polygon_offset.clamp))) { if (GLAD_GL_EXT_polygon_offset_clamp && polygon_offset.clamp != 0) { glPolygonOffsetClamp(polygon_offset.factor, polygon_offset.units, polygon_offset.clamp); } else { - glPolygonOffset(polygon_offset.factor, polygon_offset.units); UNIMPLEMENTED_IF_MSG(polygon_offset.clamp != 0, "Unimplemented Depth polygon offset clamp."); + glPolygonOffset(polygon_offset.factor, polygon_offset.units); } } } @@ -443,22 +465,21 @@ void OpenGLState::ApplyTextures() const { bool has_delta{}; std::size_t first{}; std::size_t last{}; - std::array<GLuint, Tegra::Engines::Maxwell3D::Regs::NumTextureSamplers> textures; + std::array<GLuint, Maxwell::NumTextureSamplers> textures; for (std::size_t i = 0; i < std::size(texture_units); ++i) { const auto& texture_unit = texture_units[i]; - const auto& cur_state_texture_unit = cur_state.texture_units[i]; + auto& cur_state_texture_unit = cur_state.texture_units[i]; textures[i] = texture_unit.texture; - - if (textures[i] != cur_state_texture_unit.texture) { - if (!has_delta) { - first = i; - has_delta = true; - } - last = i; + if (cur_state_texture_unit.texture == textures[i]) + continue; + cur_state_texture_unit.texture = textures[i]; + if (!has_delta) { + first = i; + has_delta = true; } + last = i; } - if (has_delta) { glBindTextures(static_cast<GLuint>(first), static_cast<GLsizei>(last - first + 1), textures.data() + first); @@ -469,16 +490,18 @@ void OpenGLState::ApplySamplers() const { bool has_delta{}; std::size_t first{}; std::size_t last{}; - std::array<GLuint, Tegra::Engines::Maxwell3D::Regs::NumTextureSamplers> samplers; + std::array<GLuint, Maxwell::NumTextureSamplers> samplers; + for (std::size_t i = 0; i < std::size(samplers); ++i) { + if (cur_state.texture_units[i].sampler == texture_units[i].sampler) + continue; + cur_state.texture_units[i].sampler = texture_units[i].sampler; samplers[i] = texture_units[i].sampler; - if (samplers[i] != cur_state.texture_units[i].sampler) { - if (!has_delta) { - first = i; - has_delta = true; - } - last = i; + if (!has_delta) { + first = i; + has_delta = true; } + last = i; } if (has_delta) { glBindSamplers(static_cast<GLuint>(first), static_cast<GLsizei>(last - first + 1), @@ -486,81 +509,15 @@ void OpenGLState::ApplySamplers() const { } } -void OpenGLState::ApplyFramebufferState() const { - if (draw.read_framebuffer != cur_state.draw.read_framebuffer) { - glBindFramebuffer(GL_READ_FRAMEBUFFER, draw.read_framebuffer); - } - if (draw.draw_framebuffer != cur_state.draw.draw_framebuffer) { - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, draw.draw_framebuffer); - } -} - -void OpenGLState::ApplyVertexArrayState() const { - if (draw.vertex_array != cur_state.draw.vertex_array) { - glBindVertexArray(draw.vertex_array); - } -} - -void OpenGLState::ApplyDepthClamp() const { - if (depth_clamp.far_plane == cur_state.depth_clamp.far_plane && - depth_clamp.near_plane == cur_state.depth_clamp.near_plane) { - return; - } - UNIMPLEMENTED_IF_MSG(depth_clamp.far_plane != depth_clamp.near_plane, - "Unimplemented Depth Clamp Separation!"); - - if (depth_clamp.far_plane || depth_clamp.near_plane) { - glEnable(GL_DEPTH_CLAMP); - } else { - glDisable(GL_DEPTH_CLAMP); - } -} - void OpenGLState::Apply() const { ApplyFramebufferState(); ApplyVertexArrayState(); - - // Shader program - if (draw.shader_program != cur_state.draw.shader_program) { - glUseProgram(draw.shader_program); - } - - // Program pipeline - if (draw.program_pipeline != cur_state.draw.program_pipeline) { - glBindProgramPipeline(draw.program_pipeline); - } - // Clip distance - for (std::size_t i = 0; i < clip_distance.size(); ++i) { - if (clip_distance[i] != cur_state.clip_distance[i]) { - if (clip_distance[i]) { - glEnable(GL_CLIP_DISTANCE0 + static_cast<GLenum>(i)); - } else { - glDisable(GL_CLIP_DISTANCE0 + static_cast<GLenum>(i)); - } - } - } - // Point - if (point.size != cur_state.point.size) { - glPointSize(point.size); - } - if (fragment_color_clamp.enabled != cur_state.fragment_color_clamp.enabled) { - glClampColor(GL_CLAMP_FRAGMENT_COLOR_ARB, - fragment_color_clamp.enabled ? GL_TRUE : GL_FALSE); - } - if (multisample_control.alpha_to_coverage != cur_state.multisample_control.alpha_to_coverage) { - if (multisample_control.alpha_to_coverage) { - glEnable(GL_SAMPLE_ALPHA_TO_COVERAGE); - } else { - glDisable(GL_SAMPLE_ALPHA_TO_COVERAGE); - } - } - if (multisample_control.alpha_to_one != cur_state.multisample_control.alpha_to_one) { - if (multisample_control.alpha_to_one) { - glEnable(GL_SAMPLE_ALPHA_TO_ONE); - } else { - glDisable(GL_SAMPLE_ALPHA_TO_ONE); - } - } + ApplyShaderProgram(); + ApplyProgramPipeline(); + ApplyClipDistances(); + ApplyPointSize(); + ApplyFragmentColorClamp(); + ApplyMultisample(); ApplyDepthClamp(); ApplyColorMask(); ApplyViewport(); @@ -574,7 +531,28 @@ void OpenGLState::Apply() const { ApplyTextures(); ApplySamplers(); ApplyPolygonOffset(); - cur_state = *this; +} + +void OpenGLState::EmulateViewportWithScissor() { + auto& current = viewports[0]; + if (current.scissor.enabled) { + const GLint left = std::max(current.x, current.scissor.x); + const GLint right = + std::max(current.x + current.width, current.scissor.x + current.scissor.width); + const GLint bottom = std::max(current.y, current.scissor.y); + const GLint top = + std::max(current.y + current.height, current.scissor.y + current.scissor.height); + current.scissor.x = std::max(left, 0); + current.scissor.y = std::max(bottom, 0); + current.scissor.width = std::max(right - left, 0); + current.scissor.height = std::max(top - bottom, 0); + } else { + current.scissor.enabled = true; + current.scissor.x = current.x; + current.scissor.y = current.y; + current.scissor.width = current.width; + current.scissor.height = current.height; + } } OpenGLState& OpenGLState::UnbindTexture(GLuint handle) { diff --git a/src/video_core/renderer_opengl/gl_state.h b/src/video_core/renderer_opengl/gl_state.h index 9e1eda5b1..41418a7b8 100644 --- a/src/video_core/renderer_opengl/gl_state.h +++ b/src/video_core/renderer_opengl/gl_state.h @@ -54,10 +54,6 @@ public: } depth_clamp; // GL_DEPTH_CLAMP struct { - bool enabled; // viewports arrays are only supported when geometry shaders are enabled. - } geometry_shaders; - - struct { bool enabled; // GL_CULL_FACE GLenum mode; // GL_CULL_FACE_MODE GLenum front_face; // GL_FRONT_FACE @@ -184,34 +180,26 @@ public: static OpenGLState GetCurState() { return cur_state; } + static bool GetsRGBUsed() { return s_rgb_used; } + static void ClearsRGBUsed() { s_rgb_used = false; } + /// Apply this state as the current OpenGL state void Apply() const; - /// Apply only the state affecting the framebuffer + void ApplyFramebufferState() const; - /// Apply only the state affecting the vertex array void ApplyVertexArrayState() const; - /// Set the initial OpenGL state - static void ApplyDefaultState(); - /// Resets any references to the given resource - OpenGLState& UnbindTexture(GLuint handle); - OpenGLState& ResetSampler(GLuint handle); - OpenGLState& ResetProgram(GLuint handle); - OpenGLState& ResetPipeline(GLuint handle); - OpenGLState& ResetVertexArray(GLuint handle); - OpenGLState& ResetFramebuffer(GLuint handle); - void EmulateViewportWithScissor(); - -private: - static OpenGLState cur_state; - // Workaround for sRGB problems caused by - // QT not supporting srgb output - static bool s_rgb_used; + void ApplyShaderProgram() const; + void ApplyProgramPipeline() const; + void ApplyClipDistances() const; + void ApplyPointSize() const; + void ApplyFragmentColorClamp() const; + void ApplyMultisample() const; void ApplySRgb() const; void ApplyCulling() const; void ApplyColorMask() const; @@ -227,6 +215,26 @@ private: void ApplySamplers() const; void ApplyDepthClamp() const; void ApplyPolygonOffset() const; + + /// Set the initial OpenGL state + static void ApplyDefaultState(); + + /// Resets any references to the given resource + OpenGLState& UnbindTexture(GLuint handle); + OpenGLState& ResetSampler(GLuint handle); + OpenGLState& ResetProgram(GLuint handle); + OpenGLState& ResetPipeline(GLuint handle); + OpenGLState& ResetVertexArray(GLuint handle); + OpenGLState& ResetFramebuffer(GLuint handle); + + /// Viewport does not affects glClearBuffer so emulate viewport using scissor test + void EmulateViewportWithScissor(); + +private: + static OpenGLState cur_state; + + // Workaround for sRGB problems caused by QT not supporting srgb output + static bool s_rgb_used; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index a01efeb05..d69cba9c3 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -5,7 +5,6 @@ #include <algorithm> #include <cstddef> #include <cstdlib> -#include <cstring> #include <memory> #include <glad/glad.h> #include "common/assert.h" diff --git a/src/video_core/renderer_opengl/utils.cpp b/src/video_core/renderer_opengl/utils.cpp index d84634cb3..84a987371 100644 --- a/src/video_core/renderer_opengl/utils.cpp +++ b/src/video_core/renderer_opengl/utils.cpp @@ -5,11 +5,39 @@ #include <string> #include <fmt/format.h> #include <glad/glad.h> +#include "common/assert.h" #include "common/common_types.h" #include "video_core/renderer_opengl/utils.h" namespace OpenGL { +BindBuffersRangePushBuffer::BindBuffersRangePushBuffer(GLenum target) : target{target} {} + +BindBuffersRangePushBuffer::~BindBuffersRangePushBuffer() = default; + +void BindBuffersRangePushBuffer::Setup(GLuint first_) { + first = first_; + buffers.clear(); + offsets.clear(); + sizes.clear(); +} + +void BindBuffersRangePushBuffer::Push(GLuint buffer, GLintptr offset, GLsizeiptr size) { + buffers.push_back(buffer); + offsets.push_back(offset); + sizes.push_back(size); +} + +void BindBuffersRangePushBuffer::Bind() const { + const std::size_t count{buffers.size()}; + DEBUG_ASSERT(count == offsets.size() && count == sizes.size()); + if (count == 0) { + return; + } + glBindBuffersRange(target, first, static_cast<GLsizei>(count), buffers.data(), offsets.data(), + sizes.data()); +} + void LabelGLObject(GLenum identifier, GLuint handle, VAddr addr, std::string extra_info) { if (!GLAD_GL_KHR_debug) { return; // We don't need to throw an error as this is just for debugging diff --git a/src/video_core/renderer_opengl/utils.h b/src/video_core/renderer_opengl/utils.h index 1fcb6fc11..aef45c9dc 100644 --- a/src/video_core/renderer_opengl/utils.h +++ b/src/video_core/renderer_opengl/utils.h @@ -5,11 +5,31 @@ #pragma once #include <string> +#include <vector> #include <glad/glad.h> #include "common/common_types.h" namespace OpenGL { +class BindBuffersRangePushBuffer { +public: + BindBuffersRangePushBuffer(GLenum target); + ~BindBuffersRangePushBuffer(); + + void Setup(GLuint first_); + + void Push(GLuint buffer, GLintptr offset, GLsizeiptr size); + + void Bind() const; + +private: + GLenum target; + GLuint first; + std::vector<GLuint> buffers; + std::vector<GLintptr> offsets; + std::vector<GLsizeiptr> sizes; +}; + void LabelGLObject(GLenum identifier, GLuint handle, VAddr addr, std::string extra_info = ""); } // namespace OpenGL
\ No newline at end of file diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 388b5ffd5..02a9f5ecb 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -10,6 +10,7 @@ #include "common/alignment.h" #include "common/assert.h" #include "core/memory.h" +#include "video_core/memory_manager.h" #include "video_core/renderer_vulkan/declarations.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp new file mode 100644 index 000000000..25500f9a3 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp @@ -0,0 +1,1381 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <functional> +#include <map> +#include <set> + +#include <fmt/format.h> + +#include <sirit/sirit.h> + +#include "common/alignment.h" +#include "common/assert.h" +#include "common/common_types.h" +#include "common/logging/log.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/engines/shader_bytecode.h" +#include "video_core/engines/shader_header.h" +#include "video_core/renderer_vulkan/vk_shader_decompiler.h" +#include "video_core/shader/shader_ir.h" + +namespace Vulkan::VKShader { + +using Sirit::Id; +using Tegra::Shader::Attribute; +using Tegra::Shader::AttributeUse; +using Tegra::Shader::Register; +using namespace VideoCommon::Shader; + +using Maxwell = Tegra::Engines::Maxwell3D::Regs; +using ShaderStage = Tegra::Engines::Maxwell3D::Regs::ShaderStage; +using Operation = const OperationNode&; + +// TODO(Rodrigo): Use rasterizer's value +constexpr u32 MAX_CONSTBUFFER_ELEMENTS = 0x1000; +constexpr u32 STAGE_BINDING_STRIDE = 0x100; + +enum class Type { Bool, Bool2, Float, Int, Uint, HalfFloat }; + +struct SamplerImage { + Id image_type; + Id sampled_image_type; + Id sampler; +}; + +namespace { + +spv::Dim GetSamplerDim(const Sampler& sampler) { + switch (sampler.GetType()) { + case Tegra::Shader::TextureType::Texture1D: + return spv::Dim::Dim1D; + case Tegra::Shader::TextureType::Texture2D: + return spv::Dim::Dim2D; + case Tegra::Shader::TextureType::Texture3D: + return spv::Dim::Dim3D; + case Tegra::Shader::TextureType::TextureCube: + return spv::Dim::Cube; + default: + UNIMPLEMENTED_MSG("Unimplemented sampler type={}", static_cast<u32>(sampler.GetType())); + return spv::Dim::Dim2D; + } +} + +/// Returns true if an attribute index is one of the 32 generic attributes +constexpr bool IsGenericAttribute(Attribute::Index attribute) { + return attribute >= Attribute::Index::Attribute_0 && + attribute <= Attribute::Index::Attribute_31; +} + +/// Returns the location of a generic attribute +constexpr u32 GetGenericAttributeLocation(Attribute::Index attribute) { + ASSERT(IsGenericAttribute(attribute)); + return static_cast<u32>(attribute) - static_cast<u32>(Attribute::Index::Attribute_0); +} + +/// Returns true if an object has to be treated as precise +bool IsPrecise(Operation operand) { + const auto& meta = operand.GetMeta(); + + if (std::holds_alternative<MetaArithmetic>(meta)) { + return std::get<MetaArithmetic>(meta).precise; + } + if (std::holds_alternative<MetaHalfArithmetic>(meta)) { + return std::get<MetaHalfArithmetic>(meta).precise; + } + return false; +} + +} // namespace + +class SPIRVDecompiler : public Sirit::Module { +public: + explicit SPIRVDecompiler(const ShaderIR& ir, ShaderStage stage) + : Module(0x00010300), ir{ir}, stage{stage}, header{ir.GetHeader()} { + AddCapability(spv::Capability::Shader); + AddExtension("SPV_KHR_storage_buffer_storage_class"); + AddExtension("SPV_KHR_variable_pointers"); + } + + void Decompile() { + AllocateBindings(); + AllocateLabels(); + + DeclareVertex(); + DeclareGeometry(); + DeclareFragment(); + DeclareRegisters(); + DeclarePredicates(); + DeclareLocalMemory(); + DeclareInternalFlags(); + DeclareInputAttributes(); + DeclareOutputAttributes(); + DeclareConstantBuffers(); + DeclareGlobalBuffers(); + DeclareSamplers(); + + execute_function = + Emit(OpFunction(t_void, spv::FunctionControlMask::Inline, TypeFunction(t_void))); + Emit(OpLabel()); + + const u32 first_address = ir.GetBasicBlocks().begin()->first; + const Id loop_label = OpLabel("loop"); + const Id merge_label = OpLabel("merge"); + const Id dummy_label = OpLabel(); + const Id jump_label = OpLabel(); + continue_label = OpLabel("continue"); + + std::vector<Sirit::Literal> literals; + std::vector<Id> branch_labels; + for (const auto& pair : labels) { + const auto [literal, label] = pair; + literals.push_back(literal); + branch_labels.push_back(label); + } + + // TODO(Rodrigo): Figure out the actual depth of the flow stack, for now it seems unlikely + // that shaders will use 20 nested SSYs and PBKs. + constexpr u32 FLOW_STACK_SIZE = 20; + const Id flow_stack_type = TypeArray(t_uint, Constant(t_uint, FLOW_STACK_SIZE)); + jmp_to = Emit(OpVariable(TypePointer(spv::StorageClass::Function, t_uint), + spv::StorageClass::Function, Constant(t_uint, first_address))); + flow_stack = Emit(OpVariable(TypePointer(spv::StorageClass::Function, flow_stack_type), + spv::StorageClass::Function, ConstantNull(flow_stack_type))); + flow_stack_top = + Emit(OpVariable(t_func_uint, spv::StorageClass::Function, Constant(t_uint, 0))); + + Name(jmp_to, "jmp_to"); + Name(flow_stack, "flow_stack"); + Name(flow_stack_top, "flow_stack_top"); + + Emit(OpBranch(loop_label)); + Emit(loop_label); + Emit(OpLoopMerge(merge_label, continue_label, spv::LoopControlMask::Unroll)); + Emit(OpBranch(dummy_label)); + + Emit(dummy_label); + const Id default_branch = OpLabel(); + const Id jmp_to_load = Emit(OpLoad(t_uint, jmp_to)); + Emit(OpSelectionMerge(jump_label, spv::SelectionControlMask::MaskNone)); + Emit(OpSwitch(jmp_to_load, default_branch, literals, branch_labels)); + + Emit(default_branch); + Emit(OpReturn()); + + for (const auto& pair : ir.GetBasicBlocks()) { + const auto& [address, bb] = pair; + Emit(labels.at(address)); + + VisitBasicBlock(bb); + + const auto next_it = labels.lower_bound(address + 1); + const Id next_label = next_it != labels.end() ? next_it->second : default_branch; + Emit(OpBranch(next_label)); + } + + Emit(jump_label); + Emit(OpBranch(continue_label)); + Emit(continue_label); + Emit(OpBranch(loop_label)); + Emit(merge_label); + Emit(OpReturn()); + Emit(OpFunctionEnd()); + } + + ShaderEntries GetShaderEntries() const { + ShaderEntries entries; + entries.const_buffers_base_binding = const_buffers_base_binding; + entries.global_buffers_base_binding = global_buffers_base_binding; + entries.samplers_base_binding = samplers_base_binding; + for (const auto& cbuf : ir.GetConstantBuffers()) { + entries.const_buffers.emplace_back(cbuf.second, cbuf.first); + } + for (const auto& gmem_pair : ir.GetGlobalMemory()) { + const auto& [base, usage] = gmem_pair; + entries.global_buffers.emplace_back(base.cbuf_index, base.cbuf_offset); + } + for (const auto& sampler : ir.GetSamplers()) { + entries.samplers.emplace_back(sampler); + } + for (const auto& attr : ir.GetInputAttributes()) { + entries.attributes.insert(GetGenericAttributeLocation(attr.first)); + } + entries.clip_distances = ir.GetClipDistances(); + entries.shader_length = ir.GetLength(); + entries.entry_function = execute_function; + entries.interfaces = interfaces; + return entries; + } + +private: + using OperationDecompilerFn = Id (SPIRVDecompiler::*)(Operation); + using OperationDecompilersArray = + std::array<OperationDecompilerFn, static_cast<std::size_t>(OperationCode::Amount)>; + + static constexpr auto INTERNAL_FLAGS_COUNT = static_cast<std::size_t>(InternalFlag::Amount); + static constexpr u32 CBUF_STRIDE = 16; + + void AllocateBindings() { + const u32 binding_base = static_cast<u32>(stage) * STAGE_BINDING_STRIDE; + u32 binding_iterator = binding_base; + + const auto Allocate = [&binding_iterator](std::size_t count) { + const u32 current_binding = binding_iterator; + binding_iterator += static_cast<u32>(count); + return current_binding; + }; + const_buffers_base_binding = Allocate(ir.GetConstantBuffers().size()); + global_buffers_base_binding = Allocate(ir.GetGlobalMemory().size()); + samplers_base_binding = Allocate(ir.GetSamplers().size()); + + ASSERT_MSG(binding_iterator - binding_base < STAGE_BINDING_STRIDE, + "Stage binding stride is too small"); + } + + void AllocateLabels() { + for (const auto& pair : ir.GetBasicBlocks()) { + const u32 address = pair.first; + labels.emplace(address, OpLabel(fmt::format("label_0x{:x}", address))); + } + } + + void DeclareVertex() { + if (stage != ShaderStage::Vertex) + return; + + DeclareVertexRedeclarations(); + } + + void DeclareGeometry() { + if (stage != ShaderStage::Geometry) + return; + + UNIMPLEMENTED(); + } + + void DeclareFragment() { + if (stage != ShaderStage::Fragment) + return; + + for (u32 rt = 0; rt < static_cast<u32>(frag_colors.size()); ++rt) { + if (!IsRenderTargetUsed(rt)) { + continue; + } + + const Id id = AddGlobalVariable(OpVariable(t_out_float4, spv::StorageClass::Output)); + Name(id, fmt::format("frag_color{}", rt)); + Decorate(id, spv::Decoration::Location, rt); + + frag_colors[rt] = id; + interfaces.push_back(id); + } + + if (header.ps.omap.depth) { + frag_depth = AddGlobalVariable(OpVariable(t_out_float, spv::StorageClass::Output)); + Name(frag_depth, "frag_depth"); + Decorate(frag_depth, spv::Decoration::BuiltIn, + static_cast<u32>(spv::BuiltIn::FragDepth)); + + interfaces.push_back(frag_depth); + } + + frag_coord = DeclareBuiltIn(spv::BuiltIn::FragCoord, spv::StorageClass::Input, t_in_float4, + "frag_coord"); + front_facing = DeclareBuiltIn(spv::BuiltIn::FrontFacing, spv::StorageClass::Input, + t_in_bool, "front_facing"); + } + + void DeclareRegisters() { + for (const u32 gpr : ir.GetRegisters()) { + const Id id = OpVariable(t_prv_float, spv::StorageClass::Private, v_float_zero); + Name(id, fmt::format("gpr_{}", gpr)); + registers.emplace(gpr, AddGlobalVariable(id)); + } + } + + void DeclarePredicates() { + for (const auto pred : ir.GetPredicates()) { + const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false); + Name(id, fmt::format("pred_{}", static_cast<u32>(pred))); + predicates.emplace(pred, AddGlobalVariable(id)); + } + } + + void DeclareLocalMemory() { + if (const u64 local_memory_size = header.GetLocalMemorySize(); local_memory_size > 0) { + const auto element_count = static_cast<u32>(Common::AlignUp(local_memory_size, 4) / 4); + const Id type_array = TypeArray(t_float, Constant(t_uint, element_count)); + const Id type_pointer = TypePointer(spv::StorageClass::Private, type_array); + Name(type_pointer, "LocalMemory"); + + local_memory = + OpVariable(type_pointer, spv::StorageClass::Private, ConstantNull(type_array)); + AddGlobalVariable(Name(local_memory, "local_memory")); + } + } + + void DeclareInternalFlags() { + constexpr std::array<const char*, INTERNAL_FLAGS_COUNT> names = {"zero", "sign", "carry", + "overflow"}; + for (std::size_t flag = 0; flag < INTERNAL_FLAGS_COUNT; ++flag) { + const auto flag_code = static_cast<InternalFlag>(flag); + const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false); + internal_flags[flag] = AddGlobalVariable(Name(id, names[flag])); + } + } + + void DeclareInputAttributes() { + for (const auto element : ir.GetInputAttributes()) { + const Attribute::Index index = element.first; + if (!IsGenericAttribute(index)) { + continue; + } + + UNIMPLEMENTED_IF(stage == ShaderStage::Geometry); + + const u32 location = GetGenericAttributeLocation(index); + const Id id = OpVariable(t_in_float4, spv::StorageClass::Input); + Name(AddGlobalVariable(id), fmt::format("in_attr{}", location)); + input_attributes.emplace(index, id); + interfaces.push_back(id); + + Decorate(id, spv::Decoration::Location, location); + + if (stage != ShaderStage::Fragment) { + continue; + } + switch (header.ps.GetAttributeUse(location)) { + case AttributeUse::Constant: + Decorate(id, spv::Decoration::Flat); + break; + case AttributeUse::ScreenLinear: + Decorate(id, spv::Decoration::NoPerspective); + break; + case AttributeUse::Perspective: + // Default + break; + default: + UNREACHABLE_MSG("Unused attribute being fetched"); + } + } + } + + void DeclareOutputAttributes() { + for (const auto index : ir.GetOutputAttributes()) { + if (!IsGenericAttribute(index)) { + continue; + } + const auto location = GetGenericAttributeLocation(index); + const Id id = OpVariable(t_out_float4, spv::StorageClass::Output); + Name(AddGlobalVariable(id), fmt::format("out_attr{}", location)); + output_attributes.emplace(index, id); + interfaces.push_back(id); + + Decorate(id, spv::Decoration::Location, location); + } + } + + void DeclareConstantBuffers() { + u32 binding = const_buffers_base_binding; + for (const auto& entry : ir.GetConstantBuffers()) { + const auto [index, size] = entry; + const Id id = OpVariable(t_cbuf_ubo, spv::StorageClass::Uniform); + AddGlobalVariable(Name(id, fmt::format("cbuf_{}", index))); + + Decorate(id, spv::Decoration::Binding, binding++); + Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); + constant_buffers.emplace(index, id); + } + } + + void DeclareGlobalBuffers() { + u32 binding = global_buffers_base_binding; + for (const auto& entry : ir.GetGlobalMemory()) { + const auto [base, usage] = entry; + const Id id = OpVariable(t_gmem_ssbo, spv::StorageClass::StorageBuffer); + AddGlobalVariable( + Name(id, fmt::format("gmem_{}_{}", base.cbuf_index, base.cbuf_offset))); + + Decorate(id, spv::Decoration::Binding, binding++); + Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); + global_buffers.emplace(base, id); + } + } + + void DeclareSamplers() { + u32 binding = samplers_base_binding; + for (const auto& sampler : ir.GetSamplers()) { + const auto dim = GetSamplerDim(sampler); + const int depth = sampler.IsShadow() ? 1 : 0; + const int arrayed = sampler.IsArray() ? 1 : 0; + // TODO(Rodrigo): Sampled 1 indicates that the image will be used with a sampler. When + // SULD and SUST instructions are implemented, replace this value. + const int sampled = 1; + const Id image_type = + TypeImage(t_float, dim, depth, arrayed, false, sampled, spv::ImageFormat::Unknown); + const Id sampled_image_type = TypeSampledImage(image_type); + const Id pointer_type = + TypePointer(spv::StorageClass::UniformConstant, sampled_image_type); + const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant); + AddGlobalVariable(Name(id, fmt::format("sampler_{}", sampler.GetIndex()))); + + sampler_images.insert( + {static_cast<u32>(sampler.GetIndex()), {image_type, sampled_image_type, id}}); + + Decorate(id, spv::Decoration::Binding, binding++); + Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); + } + } + + void DeclareVertexRedeclarations() { + vertex_index = DeclareBuiltIn(spv::BuiltIn::VertexIndex, spv::StorageClass::Input, + t_in_uint, "vertex_index"); + instance_index = DeclareBuiltIn(spv::BuiltIn::InstanceIndex, spv::StorageClass::Input, + t_in_uint, "instance_index"); + + bool is_point_size_declared = false; + bool is_clip_distances_declared = false; + for (const auto index : ir.GetOutputAttributes()) { + if (index == Attribute::Index::PointSize) { + is_point_size_declared = true; + } else if (index == Attribute::Index::ClipDistances0123 || + index == Attribute::Index::ClipDistances4567) { + is_clip_distances_declared = true; + } + } + + std::vector<Id> members; + members.push_back(t_float4); + if (is_point_size_declared) { + members.push_back(t_float); + } + if (is_clip_distances_declared) { + members.push_back(TypeArray(t_float, Constant(t_uint, 8))); + } + + const Id gl_per_vertex_struct = Name(TypeStruct(members), "PerVertex"); + Decorate(gl_per_vertex_struct, spv::Decoration::Block); + + u32 declaration_index = 0; + const auto MemberDecorateBuiltIn = [&](spv::BuiltIn builtin, std::string name, + bool condition) { + if (!condition) + return u32{}; + MemberName(gl_per_vertex_struct, declaration_index, name); + MemberDecorate(gl_per_vertex_struct, declaration_index, spv::Decoration::BuiltIn, + static_cast<u32>(builtin)); + return declaration_index++; + }; + + position_index = MemberDecorateBuiltIn(spv::BuiltIn::Position, "position", true); + point_size_index = + MemberDecorateBuiltIn(spv::BuiltIn::PointSize, "point_size", is_point_size_declared); + clip_distances_index = MemberDecorateBuiltIn(spv::BuiltIn::ClipDistance, "clip_distances", + is_clip_distances_declared); + + const Id type_pointer = TypePointer(spv::StorageClass::Output, gl_per_vertex_struct); + per_vertex = OpVariable(type_pointer, spv::StorageClass::Output); + AddGlobalVariable(Name(per_vertex, "per_vertex")); + interfaces.push_back(per_vertex); + } + + void VisitBasicBlock(const NodeBlock& bb) { + for (const Node node : bb) { + static_cast<void>(Visit(node)); + } + } + + Id Visit(Node node) { + if (const auto operation = std::get_if<OperationNode>(node)) { + const auto operation_index = static_cast<std::size_t>(operation->GetCode()); + const auto decompiler = operation_decompilers[operation_index]; + if (decompiler == nullptr) { + UNREACHABLE_MSG("Operation decompiler {} not defined", operation_index); + } + return (this->*decompiler)(*operation); + + } else if (const auto gpr = std::get_if<GprNode>(node)) { + const u32 index = gpr->GetIndex(); + if (index == Register::ZeroIndex) { + return Constant(t_float, 0.0f); + } + return Emit(OpLoad(t_float, registers.at(index))); + + } else if (const auto immediate = std::get_if<ImmediateNode>(node)) { + return BitcastTo<Type::Float>(Constant(t_uint, immediate->GetValue())); + + } else if (const auto predicate = std::get_if<PredicateNode>(node)) { + const auto value = [&]() -> Id { + switch (const auto index = predicate->GetIndex(); index) { + case Tegra::Shader::Pred::UnusedIndex: + return v_true; + case Tegra::Shader::Pred::NeverExecute: + return v_false; + default: + return Emit(OpLoad(t_bool, predicates.at(index))); + } + }(); + if (predicate->IsNegated()) { + return Emit(OpLogicalNot(t_bool, value)); + } + return value; + + } else if (const auto abuf = std::get_if<AbufNode>(node)) { + const auto attribute = abuf->GetIndex(); + const auto element = abuf->GetElement(); + + switch (attribute) { + case Attribute::Index::Position: + if (stage != ShaderStage::Fragment) { + UNIMPLEMENTED(); + break; + } else { + if (element == 3) { + return Constant(t_float, 1.0f); + } + return Emit(OpLoad(t_float, AccessElement(t_in_float, frag_coord, element))); + } + case Attribute::Index::TessCoordInstanceIDVertexID: + // TODO(Subv): Find out what the values are for the first two elements when inside a + // vertex shader, and what's the value of the fourth element when inside a Tess Eval + // shader. + ASSERT(stage == ShaderStage::Vertex); + switch (element) { + case 2: + return BitcastFrom<Type::Uint>(Emit(OpLoad(t_uint, instance_index))); + case 3: + return BitcastFrom<Type::Uint>(Emit(OpLoad(t_uint, vertex_index))); + } + UNIMPLEMENTED_MSG("Unmanaged TessCoordInstanceIDVertexID element={}", element); + return Constant(t_float, 0); + case Attribute::Index::FrontFacing: + // TODO(Subv): Find out what the values are for the other elements. + ASSERT(stage == ShaderStage::Fragment); + if (element == 3) { + const Id is_front_facing = Emit(OpLoad(t_bool, front_facing)); + const Id true_value = + BitcastTo<Type::Float>(Constant(t_int, static_cast<s32>(-1))); + const Id false_value = BitcastTo<Type::Float>(Constant(t_int, 0)); + return Emit(OpSelect(t_float, is_front_facing, true_value, false_value)); + } + UNIMPLEMENTED_MSG("Unmanaged FrontFacing element={}", element); + return Constant(t_float, 0.0f); + default: + if (IsGenericAttribute(attribute)) { + const Id pointer = + AccessElement(t_in_float, input_attributes.at(attribute), element); + return Emit(OpLoad(t_float, pointer)); + } + break; + } + UNIMPLEMENTED_MSG("Unhandled input attribute: {}", static_cast<u32>(attribute)); + + } else if (const auto cbuf = std::get_if<CbufNode>(node)) { + const Node offset = cbuf->GetOffset(); + const Id buffer_id = constant_buffers.at(cbuf->GetIndex()); + + Id buffer_index{}; + Id buffer_element{}; + + if (const auto immediate = std::get_if<ImmediateNode>(offset)) { + // Direct access + const u32 offset_imm = immediate->GetValue(); + ASSERT(offset_imm % 4 == 0); + buffer_index = Constant(t_uint, offset_imm / 16); + buffer_element = Constant(t_uint, (offset_imm / 4) % 4); + + } else if (std::holds_alternative<OperationNode>(*offset)) { + // Indirect access + // TODO(Rodrigo): Use a uniform buffer stride of 4 and drop this slow math (which + // emits sub-optimal code on GLSL from my testing). + const Id offset_id = BitcastTo<Type::Uint>(Visit(offset)); + const Id unsafe_offset = Emit(OpUDiv(t_uint, offset_id, Constant(t_uint, 4))); + const Id final_offset = Emit( + OpUMod(t_uint, unsafe_offset, Constant(t_uint, MAX_CONSTBUFFER_ELEMENTS - 1))); + buffer_index = Emit(OpUDiv(t_uint, final_offset, Constant(t_uint, 4))); + buffer_element = Emit(OpUMod(t_uint, final_offset, Constant(t_uint, 4))); + + } else { + UNREACHABLE_MSG("Unmanaged offset node type"); + } + + const Id pointer = Emit(OpAccessChain(t_cbuf_float, buffer_id, Constant(t_uint, 0), + buffer_index, buffer_element)); + return Emit(OpLoad(t_float, pointer)); + + } else if (const auto gmem = std::get_if<GmemNode>(node)) { + const Id gmem_buffer = global_buffers.at(gmem->GetDescriptor()); + const Id real = BitcastTo<Type::Uint>(Visit(gmem->GetRealAddress())); + const Id base = BitcastTo<Type::Uint>(Visit(gmem->GetBaseAddress())); + + Id offset = Emit(OpISub(t_uint, real, base)); + offset = Emit(OpUDiv(t_uint, offset, Constant(t_uint, 4u))); + return Emit(OpLoad(t_float, Emit(OpAccessChain(t_gmem_float, gmem_buffer, + Constant(t_uint, 0u), offset)))); + + } else if (const auto conditional = std::get_if<ConditionalNode>(node)) { + // It's invalid to call conditional on nested nodes, use an operation instead + const Id true_label = OpLabel(); + const Id skip_label = OpLabel(); + Emit(OpBranchConditional(Visit(conditional->GetCondition()), true_label, skip_label)); + Emit(true_label); + + VisitBasicBlock(conditional->GetCode()); + + Emit(OpBranch(skip_label)); + Emit(skip_label); + return {}; + + } else if (const auto comment = std::get_if<CommentNode>(node)) { + Name(Emit(OpUndef(t_void)), comment->GetText()); + return {}; + } + + UNREACHABLE(); + return {}; + } + + template <Id (Module::*func)(Id, Id), Type result_type, Type type_a = result_type> + Id Unary(Operation operation) { + const Id type_def = GetTypeDefinition(result_type); + const Id op_a = VisitOperand<type_a>(operation, 0); + + const Id value = BitcastFrom<result_type>(Emit((this->*func)(type_def, op_a))); + if (IsPrecise(operation)) { + Decorate(value, spv::Decoration::NoContraction); + } + return value; + } + + template <Id (Module::*func)(Id, Id, Id), Type result_type, Type type_a = result_type, + Type type_b = type_a> + Id Binary(Operation operation) { + const Id type_def = GetTypeDefinition(result_type); + const Id op_a = VisitOperand<type_a>(operation, 0); + const Id op_b = VisitOperand<type_b>(operation, 1); + + const Id value = BitcastFrom<result_type>(Emit((this->*func)(type_def, op_a, op_b))); + if (IsPrecise(operation)) { + Decorate(value, spv::Decoration::NoContraction); + } + return value; + } + + template <Id (Module::*func)(Id, Id, Id, Id), Type result_type, Type type_a = result_type, + Type type_b = type_a, Type type_c = type_b> + Id Ternary(Operation operation) { + const Id type_def = GetTypeDefinition(result_type); + const Id op_a = VisitOperand<type_a>(operation, 0); + const Id op_b = VisitOperand<type_b>(operation, 1); + const Id op_c = VisitOperand<type_c>(operation, 2); + + const Id value = BitcastFrom<result_type>(Emit((this->*func)(type_def, op_a, op_b, op_c))); + if (IsPrecise(operation)) { + Decorate(value, spv::Decoration::NoContraction); + } + return value; + } + + template <Id (Module::*func)(Id, Id, Id, Id, Id), Type result_type, Type type_a = result_type, + Type type_b = type_a, Type type_c = type_b, Type type_d = type_c> + Id Quaternary(Operation operation) { + const Id type_def = GetTypeDefinition(result_type); + const Id op_a = VisitOperand<type_a>(operation, 0); + const Id op_b = VisitOperand<type_b>(operation, 1); + const Id op_c = VisitOperand<type_c>(operation, 2); + const Id op_d = VisitOperand<type_d>(operation, 3); + + const Id value = + BitcastFrom<result_type>(Emit((this->*func)(type_def, op_a, op_b, op_c, op_d))); + if (IsPrecise(operation)) { + Decorate(value, spv::Decoration::NoContraction); + } + return value; + } + + Id Assign(Operation operation) { + const Node dest = operation[0]; + const Node src = operation[1]; + + Id target{}; + if (const auto gpr = std::get_if<GprNode>(dest)) { + if (gpr->GetIndex() == Register::ZeroIndex) { + // Writing to Register::ZeroIndex is a no op + return {}; + } + target = registers.at(gpr->GetIndex()); + + } else if (const auto abuf = std::get_if<AbufNode>(dest)) { + target = [&]() -> Id { + switch (const auto attribute = abuf->GetIndex(); attribute) { + case Attribute::Index::Position: + return AccessElement(t_out_float, per_vertex, position_index, + abuf->GetElement()); + case Attribute::Index::PointSize: + return AccessElement(t_out_float, per_vertex, point_size_index); + case Attribute::Index::ClipDistances0123: + return AccessElement(t_out_float, per_vertex, clip_distances_index, + abuf->GetElement()); + case Attribute::Index::ClipDistances4567: + return AccessElement(t_out_float, per_vertex, clip_distances_index, + abuf->GetElement() + 4); + default: + if (IsGenericAttribute(attribute)) { + return AccessElement(t_out_float, output_attributes.at(attribute), + abuf->GetElement()); + } + UNIMPLEMENTED_MSG("Unhandled output attribute: {}", + static_cast<u32>(attribute)); + return {}; + } + }(); + + } else if (const auto lmem = std::get_if<LmemNode>(dest)) { + Id address = BitcastTo<Type::Uint>(Visit(lmem->GetAddress())); + address = Emit(OpUDiv(t_uint, address, Constant(t_uint, 4))); + target = Emit(OpAccessChain(t_prv_float, local_memory, {address})); + } + + Emit(OpStore(target, Visit(src))); + return {}; + } + + Id HNegate(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id HMergeF32(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id HMergeH0(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id HMergeH1(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id HPack2(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id LogicalAssign(Operation operation) { + const Node dest = operation[0]; + const Node src = operation[1]; + + Id target{}; + if (const auto pred = std::get_if<PredicateNode>(dest)) { + ASSERT_MSG(!pred->IsNegated(), "Negating logical assignment"); + + const auto index = pred->GetIndex(); + switch (index) { + case Tegra::Shader::Pred::NeverExecute: + case Tegra::Shader::Pred::UnusedIndex: + // Writing to these predicates is a no-op + return {}; + } + target = predicates.at(index); + + } else if (const auto flag = std::get_if<InternalFlagNode>(dest)) { + target = internal_flags.at(static_cast<u32>(flag->GetFlag())); + } + + Emit(OpStore(target, Visit(src))); + return {}; + } + + Id LogicalPick2(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id LogicalAll2(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id LogicalAny2(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id GetTextureSampler(Operation operation) { + const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); + const auto entry = sampler_images.at(static_cast<u32>(meta->sampler.GetIndex())); + return Emit(OpLoad(entry.sampled_image_type, entry.sampler)); + } + + Id GetTextureImage(Operation operation) { + const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); + const auto entry = sampler_images.at(static_cast<u32>(meta->sampler.GetIndex())); + return Emit(OpImage(entry.image_type, GetTextureSampler(operation))); + } + + Id GetTextureCoordinates(Operation operation) { + const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); + std::vector<Id> coords; + for (std::size_t i = 0; i < operation.GetOperandsCount(); ++i) { + coords.push_back(Visit(operation[i])); + } + if (meta->sampler.IsArray()) { + const Id array_integer = BitcastTo<Type::Int>(Visit(meta->array)); + coords.push_back(Emit(OpConvertSToF(t_float, array_integer))); + } + if (meta->sampler.IsShadow()) { + coords.push_back(Visit(meta->depth_compare)); + } + + const std::array<Id, 4> t_float_lut = {nullptr, t_float2, t_float3, t_float4}; + return coords.size() == 1 + ? coords[0] + : Emit(OpCompositeConstruct(t_float_lut.at(coords.size() - 1), coords)); + } + + Id GetTextureElement(Operation operation, Id sample_value) { + const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); + ASSERT(meta); + return Emit(OpCompositeExtract(t_float, sample_value, meta->element)); + } + + Id Texture(Operation operation) { + const Id texture = Emit(OpImageSampleImplicitLod(t_float4, GetTextureSampler(operation), + GetTextureCoordinates(operation))); + return GetTextureElement(operation, texture); + } + + Id TextureLod(Operation operation) { + const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); + const Id texture = Emit(OpImageSampleExplicitLod( + t_float4, GetTextureSampler(operation), GetTextureCoordinates(operation), + spv::ImageOperandsMask::Lod, Visit(meta->lod))); + return GetTextureElement(operation, texture); + } + + Id TextureGather(Operation operation) { + const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); + const auto coords = GetTextureCoordinates(operation); + + Id texture; + if (meta->sampler.IsShadow()) { + texture = Emit(OpImageDrefGather(t_float4, GetTextureSampler(operation), coords, + Visit(meta->component))); + } else { + u32 component_value = 0; + if (meta->component) { + const auto component = std::get_if<ImmediateNode>(meta->component); + ASSERT_MSG(component, "Component is not an immediate value"); + component_value = component->GetValue(); + } + texture = Emit(OpImageGather(t_float4, GetTextureSampler(operation), coords, + Constant(t_uint, component_value))); + } + + return GetTextureElement(operation, texture); + } + + Id TextureQueryDimensions(Operation operation) { + const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); + const auto image_id = GetTextureImage(operation); + AddCapability(spv::Capability::ImageQuery); + + if (meta->element == 3) { + return BitcastTo<Type::Float>(Emit(OpImageQueryLevels(t_int, image_id))); + } + + const Id lod = VisitOperand<Type::Uint>(operation, 0); + const std::size_t coords_count = [&]() { + switch (const auto type = meta->sampler.GetType(); type) { + case Tegra::Shader::TextureType::Texture1D: + return 1; + case Tegra::Shader::TextureType::Texture2D: + case Tegra::Shader::TextureType::TextureCube: + return 2; + case Tegra::Shader::TextureType::Texture3D: + return 3; + default: + UNREACHABLE_MSG("Invalid texture type={}", static_cast<u32>(type)); + return 2; + } + }(); + + if (meta->element >= coords_count) { + return Constant(t_float, 0.0f); + } + + const std::array<Id, 3> types = {t_int, t_int2, t_int3}; + const Id sizes = Emit(OpImageQuerySizeLod(types.at(coords_count - 1), image_id, lod)); + const Id size = Emit(OpCompositeExtract(t_int, sizes, meta->element)); + return BitcastTo<Type::Float>(size); + } + + Id TextureQueryLod(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id TexelFetch(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id Branch(Operation operation) { + const auto target = std::get_if<ImmediateNode>(operation[0]); + UNIMPLEMENTED_IF(!target); + + Emit(OpStore(jmp_to, Constant(t_uint, target->GetValue()))); + BranchingOp([&]() { Emit(OpBranch(continue_label)); }); + return {}; + } + + Id PushFlowStack(Operation operation) { + const auto target = std::get_if<ImmediateNode>(operation[0]); + ASSERT(target); + + const Id current = Emit(OpLoad(t_uint, flow_stack_top)); + const Id next = Emit(OpIAdd(t_uint, current, Constant(t_uint, 1))); + const Id access = Emit(OpAccessChain(t_func_uint, flow_stack, current)); + + Emit(OpStore(access, Constant(t_uint, target->GetValue()))); + Emit(OpStore(flow_stack_top, next)); + return {}; + } + + Id PopFlowStack(Operation operation) { + const Id current = Emit(OpLoad(t_uint, flow_stack_top)); + const Id previous = Emit(OpISub(t_uint, current, Constant(t_uint, 1))); + const Id access = Emit(OpAccessChain(t_func_uint, flow_stack, previous)); + const Id target = Emit(OpLoad(t_uint, access)); + + Emit(OpStore(flow_stack_top, previous)); + Emit(OpStore(jmp_to, target)); + BranchingOp([&]() { Emit(OpBranch(continue_label)); }); + return {}; + } + + Id Exit(Operation operation) { + switch (stage) { + case ShaderStage::Vertex: { + // TODO(Rodrigo): We should use VK_EXT_depth_range_unrestricted instead, but it doesn't + // seem to be working on Nvidia's drivers and Intel (mesa and blob) doesn't support it. + const Id position = AccessElement(t_float4, per_vertex, position_index); + Id depth = Emit(OpLoad(t_float, AccessElement(t_out_float, position, 2))); + depth = Emit(OpFAdd(t_float, depth, Constant(t_float, 1.0f))); + depth = Emit(OpFMul(t_float, depth, Constant(t_float, 0.5f))); + Emit(OpStore(AccessElement(t_out_float, position, 2), depth)); + break; + } + case ShaderStage::Fragment: { + const auto SafeGetRegister = [&](u32 reg) { + // TODO(Rodrigo): Replace with contains once C++20 releases + if (const auto it = registers.find(reg); it != registers.end()) { + return Emit(OpLoad(t_float, it->second)); + } + return Constant(t_float, 0.0f); + }; + + UNIMPLEMENTED_IF_MSG(header.ps.omap.sample_mask != 0, + "Sample mask write is unimplemented"); + + // TODO(Rodrigo): Alpha testing + + // Write the color outputs using the data in the shader registers, disabled + // rendertargets/components are skipped in the register assignment. + u32 current_reg = 0; + for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) { + // TODO(Subv): Figure out how dual-source blending is configured in the Switch. + for (u32 component = 0; component < 4; ++component) { + if (header.ps.IsColorComponentOutputEnabled(rt, component)) { + Emit(OpStore(AccessElement(t_out_float, frag_colors.at(rt), component), + SafeGetRegister(current_reg))); + ++current_reg; + } + } + } + if (header.ps.omap.depth) { + // The depth output is always 2 registers after the last color output, and + // current_reg already contains one past the last color register. + Emit(OpStore(frag_depth, SafeGetRegister(current_reg + 1))); + } + break; + } + } + + BranchingOp([&]() { Emit(OpReturn()); }); + return {}; + } + + Id Discard(Operation operation) { + BranchingOp([&]() { Emit(OpKill()); }); + return {}; + } + + Id EmitVertex(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id EndPrimitive(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id YNegate(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + + Id DeclareBuiltIn(spv::BuiltIn builtin, spv::StorageClass storage, Id type, + const std::string& name) { + const Id id = OpVariable(type, storage); + Decorate(id, spv::Decoration::BuiltIn, static_cast<u32>(builtin)); + AddGlobalVariable(Name(id, name)); + interfaces.push_back(id); + return id; + } + + bool IsRenderTargetUsed(u32 rt) const { + for (u32 component = 0; component < 4; ++component) { + if (header.ps.IsColorComponentOutputEnabled(rt, component)) { + return true; + } + } + return false; + } + + template <typename... Args> + Id AccessElement(Id pointer_type, Id composite, Args... elements_) { + std::vector<Id> members; + auto elements = {elements_...}; + for (const auto element : elements) { + members.push_back(Constant(t_uint, element)); + } + + return Emit(OpAccessChain(pointer_type, composite, members)); + } + + template <Type type> + Id VisitOperand(Operation operation, std::size_t operand_index) { + const Id value = Visit(operation[operand_index]); + + switch (type) { + case Type::Bool: + case Type::Bool2: + case Type::Float: + return value; + case Type::Int: + return Emit(OpBitcast(t_int, value)); + case Type::Uint: + return Emit(OpBitcast(t_uint, value)); + case Type::HalfFloat: + UNIMPLEMENTED(); + } + UNREACHABLE(); + return value; + } + + template <Type type> + Id BitcastFrom(Id value) { + switch (type) { + case Type::Bool: + case Type::Bool2: + case Type::Float: + return value; + case Type::Int: + case Type::Uint: + return Emit(OpBitcast(t_float, value)); + case Type::HalfFloat: + UNIMPLEMENTED(); + } + UNREACHABLE(); + return value; + } + + template <Type type> + Id BitcastTo(Id value) { + switch (type) { + case Type::Bool: + case Type::Bool2: + UNREACHABLE(); + case Type::Float: + return Emit(OpBitcast(t_float, value)); + case Type::Int: + return Emit(OpBitcast(t_int, value)); + case Type::Uint: + return Emit(OpBitcast(t_uint, value)); + case Type::HalfFloat: + UNIMPLEMENTED(); + } + UNREACHABLE(); + return value; + } + + Id GetTypeDefinition(Type type) { + switch (type) { + case Type::Bool: + return t_bool; + case Type::Bool2: + return t_bool2; + case Type::Float: + return t_float; + case Type::Int: + return t_int; + case Type::Uint: + return t_uint; + case Type::HalfFloat: + UNIMPLEMENTED(); + } + UNREACHABLE(); + return {}; + } + + void BranchingOp(std::function<void()> call) { + const Id true_label = OpLabel(); + const Id skip_label = OpLabel(); + Emit(OpSelectionMerge(skip_label, spv::SelectionControlMask::Flatten)); + Emit(OpBranchConditional(v_true, true_label, skip_label, 1, 0)); + Emit(true_label); + call(); + + Emit(skip_label); + } + + static constexpr OperationDecompilersArray operation_decompilers = { + &SPIRVDecompiler::Assign, + + &SPIRVDecompiler::Ternary<&Module::OpSelect, Type::Float, Type::Bool, Type::Float, + Type::Float>, + + &SPIRVDecompiler::Binary<&Module::OpFAdd, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFMul, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFDiv, Type::Float>, + &SPIRVDecompiler::Ternary<&Module::OpFma, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpFNegate, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::Float>, + &SPIRVDecompiler::Ternary<&Module::OpFClamp, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFMin, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFMax, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpCos, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpSin, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpExp2, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpLog2, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpInverseSqrt, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpSqrt, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpRoundEven, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpFloor, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpCeil, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpTrunc, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpConvertSToF, Type::Float, Type::Int>, + &SPIRVDecompiler::Unary<&Module::OpConvertUToF, Type::Float, Type::Uint>, + + &SPIRVDecompiler::Binary<&Module::OpIAdd, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpIMul, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpSDiv, Type::Int>, + &SPIRVDecompiler::Unary<&Module::OpSNegate, Type::Int>, + &SPIRVDecompiler::Unary<&Module::OpSAbs, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpSMin, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpSMax, Type::Int>, + + &SPIRVDecompiler::Unary<&Module::OpConvertFToS, Type::Int, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpBitcast, Type::Int, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpShiftLeftLogical, Type::Int, Type::Int, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpShiftRightLogical, Type::Int, Type::Int, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpShiftRightArithmetic, Type::Int, Type::Int, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpBitwiseAnd, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpBitwiseOr, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpBitwiseXor, Type::Int>, + &SPIRVDecompiler::Unary<&Module::OpNot, Type::Int>, + &SPIRVDecompiler::Quaternary<&Module::OpBitFieldInsert, Type::Int>, + &SPIRVDecompiler::Ternary<&Module::OpBitFieldSExtract, Type::Int>, + &SPIRVDecompiler::Unary<&Module::OpBitCount, Type::Int>, + + &SPIRVDecompiler::Binary<&Module::OpIAdd, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpIMul, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpUDiv, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpUMin, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpUMax, Type::Uint>, + &SPIRVDecompiler::Unary<&Module::OpConvertFToU, Type::Uint, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpBitcast, Type::Uint, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpShiftLeftLogical, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpShiftRightLogical, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpShiftRightArithmetic, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpBitwiseAnd, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpBitwiseOr, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpBitwiseXor, Type::Uint>, + &SPIRVDecompiler::Unary<&Module::OpNot, Type::Uint>, + &SPIRVDecompiler::Quaternary<&Module::OpBitFieldInsert, Type::Uint>, + &SPIRVDecompiler::Ternary<&Module::OpBitFieldUExtract, Type::Uint>, + &SPIRVDecompiler::Unary<&Module::OpBitCount, Type::Uint>, + + &SPIRVDecompiler::Binary<&Module::OpFAdd, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFMul, Type::HalfFloat>, + &SPIRVDecompiler::Ternary<&Module::OpFma, Type::HalfFloat>, + &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::HalfFloat>, + &SPIRVDecompiler::HNegate, + &SPIRVDecompiler::HMergeF32, + &SPIRVDecompiler::HMergeH0, + &SPIRVDecompiler::HMergeH1, + &SPIRVDecompiler::HPack2, + + &SPIRVDecompiler::LogicalAssign, + &SPIRVDecompiler::Binary<&Module::OpLogicalAnd, Type::Bool>, + &SPIRVDecompiler::Binary<&Module::OpLogicalOr, Type::Bool>, + &SPIRVDecompiler::Binary<&Module::OpLogicalNotEqual, Type::Bool>, + &SPIRVDecompiler::Unary<&Module::OpLogicalNot, Type::Bool>, + &SPIRVDecompiler::LogicalPick2, + &SPIRVDecompiler::LogicalAll2, + &SPIRVDecompiler::LogicalAny2, + + &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThan, Type::Bool, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFOrdNotEqual, Type::Bool, Type::Float>, + &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThanEqual, Type::Bool, Type::Float>, + &SPIRVDecompiler::Unary<&Module::OpIsNan, Type::Bool>, + + &SPIRVDecompiler::Binary<&Module::OpSLessThan, Type::Bool, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpIEqual, Type::Bool, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpSLessThanEqual, Type::Bool, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpSGreaterThan, Type::Bool, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpINotEqual, Type::Bool, Type::Int>, + &SPIRVDecompiler::Binary<&Module::OpSGreaterThanEqual, Type::Bool, Type::Int>, + + &SPIRVDecompiler::Binary<&Module::OpULessThan, Type::Bool, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpIEqual, Type::Bool, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpULessThanEqual, Type::Bool, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpUGreaterThan, Type::Bool, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpINotEqual, Type::Bool, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpUGreaterThanEqual, Type::Bool, Type::Uint>, + + &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThan, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdNotEqual, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThanEqual, Type::Bool, Type::HalfFloat>, + + &SPIRVDecompiler::Texture, + &SPIRVDecompiler::TextureLod, + &SPIRVDecompiler::TextureGather, + &SPIRVDecompiler::TextureQueryDimensions, + &SPIRVDecompiler::TextureQueryLod, + &SPIRVDecompiler::TexelFetch, + + &SPIRVDecompiler::Branch, + &SPIRVDecompiler::PushFlowStack, + &SPIRVDecompiler::PopFlowStack, + &SPIRVDecompiler::Exit, + &SPIRVDecompiler::Discard, + + &SPIRVDecompiler::EmitVertex, + &SPIRVDecompiler::EndPrimitive, + + &SPIRVDecompiler::YNegate, + }; + + const ShaderIR& ir; + const ShaderStage stage; + const Tegra::Shader::Header header; + + const Id t_void = Name(TypeVoid(), "void"); + + const Id t_bool = Name(TypeBool(), "bool"); + const Id t_bool2 = Name(TypeVector(t_bool, 2), "bool2"); + + const Id t_int = Name(TypeInt(32, true), "int"); + const Id t_int2 = Name(TypeVector(t_int, 2), "int2"); + const Id t_int3 = Name(TypeVector(t_int, 3), "int3"); + const Id t_int4 = Name(TypeVector(t_int, 4), "int4"); + + const Id t_uint = Name(TypeInt(32, false), "uint"); + const Id t_uint2 = Name(TypeVector(t_uint, 2), "uint2"); + const Id t_uint3 = Name(TypeVector(t_uint, 3), "uint3"); + const Id t_uint4 = Name(TypeVector(t_uint, 4), "uint4"); + + const Id t_float = Name(TypeFloat(32), "float"); + const Id t_float2 = Name(TypeVector(t_float, 2), "float2"); + const Id t_float3 = Name(TypeVector(t_float, 3), "float3"); + const Id t_float4 = Name(TypeVector(t_float, 4), "float4"); + + const Id t_prv_bool = Name(TypePointer(spv::StorageClass::Private, t_bool), "prv_bool"); + const Id t_prv_float = Name(TypePointer(spv::StorageClass::Private, t_float), "prv_float"); + + const Id t_func_uint = Name(TypePointer(spv::StorageClass::Function, t_uint), "func_uint"); + + const Id t_in_bool = Name(TypePointer(spv::StorageClass::Input, t_bool), "in_bool"); + const Id t_in_uint = Name(TypePointer(spv::StorageClass::Input, t_uint), "in_uint"); + const Id t_in_float = Name(TypePointer(spv::StorageClass::Input, t_float), "in_float"); + const Id t_in_float4 = Name(TypePointer(spv::StorageClass::Input, t_float4), "in_float4"); + + const Id t_out_float = Name(TypePointer(spv::StorageClass::Output, t_float), "out_float"); + const Id t_out_float4 = Name(TypePointer(spv::StorageClass::Output, t_float4), "out_float4"); + + const Id t_cbuf_float = TypePointer(spv::StorageClass::Uniform, t_float); + const Id t_cbuf_array = + Decorate(Name(TypeArray(t_float4, Constant(t_uint, MAX_CONSTBUFFER_ELEMENTS)), "CbufArray"), + spv::Decoration::ArrayStride, CBUF_STRIDE); + const Id t_cbuf_struct = MemberDecorate( + Decorate(TypeStruct(t_cbuf_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0); + const Id t_cbuf_ubo = TypePointer(spv::StorageClass::Uniform, t_cbuf_struct); + + const Id t_gmem_float = TypePointer(spv::StorageClass::StorageBuffer, t_float); + const Id t_gmem_array = + Name(Decorate(TypeRuntimeArray(t_float), spv::Decoration::ArrayStride, 4u), "GmemArray"); + const Id t_gmem_struct = MemberDecorate( + Decorate(TypeStruct(t_gmem_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0); + const Id t_gmem_ssbo = TypePointer(spv::StorageClass::StorageBuffer, t_gmem_struct); + + const Id v_float_zero = Constant(t_float, 0.0f); + const Id v_true = ConstantTrue(t_bool); + const Id v_false = ConstantFalse(t_bool); + + Id per_vertex{}; + std::map<u32, Id> registers; + std::map<Tegra::Shader::Pred, Id> predicates; + Id local_memory{}; + std::array<Id, INTERNAL_FLAGS_COUNT> internal_flags{}; + std::map<Attribute::Index, Id> input_attributes; + std::map<Attribute::Index, Id> output_attributes; + std::map<u32, Id> constant_buffers; + std::map<GlobalMemoryBase, Id> global_buffers; + std::map<u32, SamplerImage> sampler_images; + + Id instance_index{}; + Id vertex_index{}; + std::array<Id, Maxwell::NumRenderTargets> frag_colors{}; + Id frag_depth{}; + Id frag_coord{}; + Id front_facing{}; + + u32 position_index{}; + u32 point_size_index{}; + u32 clip_distances_index{}; + + std::vector<Id> interfaces; + + u32 const_buffers_base_binding{}; + u32 global_buffers_base_binding{}; + u32 samplers_base_binding{}; + + Id execute_function{}; + Id jmp_to{}; + Id flow_stack_top{}; + Id flow_stack{}; + Id continue_label{}; + std::map<u32, Id> labels; +}; + +DecompilerResult Decompile(const VideoCommon::Shader::ShaderIR& ir, Maxwell::ShaderStage stage) { + auto decompiler = std::make_unique<SPIRVDecompiler>(ir, stage); + decompiler->Decompile(); + return {std::move(decompiler), decompiler->GetShaderEntries()}; +} + +} // namespace Vulkan::VKShader diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h new file mode 100644 index 000000000..329d8fa38 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h @@ -0,0 +1,80 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <memory> +#include <set> +#include <utility> +#include <vector> + +#include <sirit/sirit.h> + +#include "common/common_types.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/shader/shader_ir.h" + +namespace VideoCommon::Shader { +class ShaderIR; +} + +namespace Vulkan::VKShader { + +using Maxwell = Tegra::Engines::Maxwell3D::Regs; + +using SamplerEntry = VideoCommon::Shader::Sampler; + +constexpr u32 DESCRIPTOR_SET = 0; + +class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer { +public: + explicit constexpr ConstBufferEntry(const VideoCommon::Shader::ConstBuffer& entry, u32 index) + : VideoCommon::Shader::ConstBuffer{entry}, index{index} {} + + constexpr u32 GetIndex() const { + return index; + } + +private: + u32 index{}; +}; + +class GlobalBufferEntry { +public: + explicit GlobalBufferEntry(u32 cbuf_index, u32 cbuf_offset) + : cbuf_index{cbuf_index}, cbuf_offset{cbuf_offset} {} + + u32 GetCbufIndex() const { + return cbuf_index; + } + + u32 GetCbufOffset() const { + return cbuf_offset; + } + +private: + u32 cbuf_index{}; + u32 cbuf_offset{}; +}; + +struct ShaderEntries { + u32 const_buffers_base_binding{}; + u32 global_buffers_base_binding{}; + u32 samplers_base_binding{}; + std::vector<ConstBufferEntry> const_buffers; + std::vector<GlobalBufferEntry> global_buffers; + std::vector<SamplerEntry> samplers; + std::set<u32> attributes; + std::array<bool, Maxwell::NumClipDistances> clip_distances{}; + std::size_t shader_length{}; + Sirit::Id entry_function{}; + std::vector<Sirit::Id> interfaces; +}; + +using DecompilerResult = std::pair<std::unique_ptr<Sirit::Module>, ShaderEntries>; + +DecompilerResult Decompile(const VideoCommon::Shader::ShaderIR& ir, Maxwell::ShaderStage stage); + +} // namespace Vulkan::VKShader diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp index b2b706cb8..ea1092db1 100644 --- a/src/video_core/shader/decode/memory.cpp +++ b/src/video_core/shader/decode/memory.cpp @@ -19,6 +19,23 @@ using Tegra::Shader::Instruction; using Tegra::Shader::OpCode; using Tegra::Shader::Register; +namespace { +u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) { + switch (uniform_type) { + case Tegra::Shader::UniformType::Single: + return 1; + case Tegra::Shader::UniformType::Double: + return 2; + case Tegra::Shader::UniformType::Quad: + case Tegra::Shader::UniformType::UnsignedQuad: + return 4; + default: + UNIMPLEMENTED_MSG("Unimplemented size={}!", static_cast<u32>(uniform_type)); + return 1; + } +} +} // namespace + u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { const Instruction instr = {program_code[pc]}; const auto opcode = OpCode::Decode(instr); @@ -127,45 +144,15 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { break; } case OpCode::Id::LDG: { - const u32 count = [&]() { - switch (instr.ldg.type) { - case Tegra::Shader::UniformType::Single: - return 1; - case Tegra::Shader::UniformType::Double: - return 2; - case Tegra::Shader::UniformType::Quad: - case Tegra::Shader::UniformType::UnsignedQuad: - return 4; - default: - UNIMPLEMENTED_MSG("Unimplemented LDG size!"); - return 1; - } - }(); - - const Node addr_register = GetRegister(instr.gpr8); - const Node base_address = - TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size())); - const auto cbuf = std::get_if<CbufNode>(base_address); - ASSERT(cbuf != nullptr); - const auto cbuf_offset_imm = std::get_if<ImmediateNode>(cbuf->GetOffset()); - ASSERT(cbuf_offset_imm != nullptr); - const auto cbuf_offset = cbuf_offset_imm->GetValue(); - - bb.push_back(Comment( - fmt::format("Base address is c[0x{:x}][0x{:x}]", cbuf->GetIndex(), cbuf_offset))); - - const GlobalMemoryBase descriptor{cbuf->GetIndex(), cbuf_offset}; - used_global_memory_bases.insert(descriptor); - - const Node immediate_offset = - Immediate(static_cast<u32>(instr.ldg.immediate_offset.Value())); - const Node base_real_address = - Operation(OperationCode::UAdd, NO_PRECISE, immediate_offset, addr_register); + const auto [real_address_base, base_address, descriptor] = + TrackAndGetGlobalMemory(bb, GetRegister(instr.gpr8), + static_cast<u32>(instr.ldg.immediate_offset.Value()), false); + const u32 count = GetUniformTypeElementsCount(instr.ldg.type); for (u32 i = 0; i < count; ++i) { const Node it_offset = Immediate(i * 4); const Node real_address = - Operation(OperationCode::UAdd, NO_PRECISE, base_real_address, it_offset); + Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset); const Node gmem = StoreNode(GmemNode(real_address, base_address, descriptor)); SetTemporal(bb, i, gmem); @@ -175,6 +162,28 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { } break; } + case OpCode::Id::STG: { + const auto [real_address_base, base_address, descriptor] = + TrackAndGetGlobalMemory(bb, GetRegister(instr.gpr8), + static_cast<u32>(instr.stg.immediate_offset.Value()), true); + + // Encode in temporary registers like this: real_base_address, {registers_to_be_written...} + SetTemporal(bb, 0, real_address_base); + + const u32 count = GetUniformTypeElementsCount(instr.stg.type); + for (u32 i = 0; i < count; ++i) { + SetTemporal(bb, i + 1, GetRegister(instr.gpr0.Value() + i)); + } + for (u32 i = 0; i < count; ++i) { + const Node it_offset = Immediate(i * 4); + const Node real_address = + Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset); + const Node gmem = StoreNode(GmemNode(real_address, base_address, descriptor)); + + bb.push_back(Operation(OperationCode::Assign, gmem, GetTemporal(i + 1))); + } + break; + } case OpCode::Id::ST_A: { UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex, "Indirect attribute loads are not supported"); @@ -237,4 +246,34 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { return pc; } +std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackAndGetGlobalMemory(NodeBlock& bb, + Node addr_register, + u32 immediate_offset, + bool is_write) { + const Node base_address{ + TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()))}; + const auto cbuf = std::get_if<CbufNode>(base_address); + ASSERT(cbuf != nullptr); + const auto cbuf_offset_imm = std::get_if<ImmediateNode>(cbuf->GetOffset()); + ASSERT(cbuf_offset_imm != nullptr); + const auto cbuf_offset = cbuf_offset_imm->GetValue(); + + bb.push_back( + Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", cbuf->GetIndex(), cbuf_offset))); + + const GlobalMemoryBase descriptor{cbuf->GetIndex(), cbuf_offset}; + const auto& [entry, is_new] = used_global_memory.try_emplace(descriptor); + auto& usage = entry->second; + if (is_write) { + usage.is_written = true; + } else { + usage.is_read = true; + } + + const auto real_address = + Operation(OperationCode::UAdd, NO_PRECISE, Immediate(immediate_offset), addr_register); + + return {real_address, base_address, descriptor}; +} + } // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp index a99ae19bf..a775b402b 100644 --- a/src/video_core/shader/decode/texture.cpp +++ b/src/video_core/shader/decode/texture.cpp @@ -7,7 +7,9 @@ #include <fmt/format.h> #include "common/assert.h" +#include "common/bit_field.h" #include "common/common_types.h" +#include "common/logging/log.h" #include "video_core/engines/shader_bytecode.h" #include "video_core/shader/shader_ir.h" @@ -41,19 +43,18 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { switch (opcode->get().GetId()) { case OpCode::Id::TEX: { - UNIMPLEMENTED_IF_MSG(instr.tex.UsesMiscMode(TextureMiscMode::AOFFI), - "AOFFI is not implemented"); - if (instr.tex.UsesMiscMode(TextureMiscMode::NODEP)) { LOG_WARNING(HW_GPU, "TEX.NODEP implementation is incomplete"); } const TextureType texture_type{instr.tex.texture_type}; const bool is_array = instr.tex.array != 0; + const bool is_aoffi = instr.tex.UsesMiscMode(TextureMiscMode::AOFFI); const bool depth_compare = instr.tex.UsesMiscMode(TextureMiscMode::DC); const auto process_mode = instr.tex.GetTextureProcessMode(); WriteTexInstructionFloat( - bb, instr, GetTexCode(instr, texture_type, process_mode, depth_compare, is_array)); + bb, instr, + GetTexCode(instr, texture_type, process_mode, depth_compare, is_array, is_aoffi)); break; } case OpCode::Id::TEXS: { @@ -78,8 +79,6 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { } case OpCode::Id::TLD4: { ASSERT(instr.tld4.array == 0); - UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(TextureMiscMode::AOFFI), - "AOFFI is not implemented"); UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(TextureMiscMode::NDV), "NDV is not implemented"); UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(TextureMiscMode::PTP), @@ -92,8 +91,9 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { const auto texture_type = instr.tld4.texture_type.Value(); const bool depth_compare = instr.tld4.UsesMiscMode(TextureMiscMode::DC); const bool is_array = instr.tld4.array != 0; - WriteTexInstructionFloat(bb, instr, - GetTld4Code(instr, texture_type, depth_compare, is_array)); + const bool is_aoffi = instr.tld4.UsesMiscMode(TextureMiscMode::AOFFI); + WriteTexInstructionFloat( + bb, instr, GetTld4Code(instr, texture_type, depth_compare, is_array, is_aoffi)); break; } case OpCode::Id::TLD4S: { @@ -127,7 +127,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { Node4 values; for (u32 element = 0; element < values.size(); ++element) { auto coords_copy = coords; - MetaTexture meta{sampler, {}, {}, {}, {}, component, element}; + MetaTexture meta{sampler, {}, {}, {}, {}, {}, component, element}; values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy)); } @@ -152,7 +152,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { if (!instr.txq.IsComponentEnabled(element)) { continue; } - MetaTexture meta{sampler, {}, {}, {}, {}, {}, element}; + MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, element}; const Node value = Operation(OperationCode::TextureQueryDimensions, meta, GetRegister(instr.gpr8)); SetTemporal(bb, indexer++, value); @@ -202,7 +202,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { for (u32 element = 0; element < 2; ++element) { auto params = coords; - MetaTexture meta{sampler, {}, {}, {}, {}, {}, element}; + MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, element}; const Node value = Operation(OperationCode::TextureQueryLod, meta, std::move(params)); SetTemporal(bb, element, value); } @@ -325,7 +325,8 @@ void ShaderIR::WriteTexsInstructionHalfFloat(NodeBlock& bb, Instruction instr, Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type, TextureProcessMode process_mode, std::vector<Node> coords, - Node array, Node depth_compare, u32 bias_offset) { + Node array, Node depth_compare, u32 bias_offset, + std::vector<Node> aoffi) { const bool is_array = array; const bool is_shadow = depth_compare; @@ -374,7 +375,7 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type, Node4 values; for (u32 element = 0; element < values.size(); ++element) { auto copy_coords = coords; - MetaTexture meta{sampler, array, depth_compare, bias, lod, {}, element}; + MetaTexture meta{sampler, array, depth_compare, aoffi, bias, lod, {}, element}; values[element] = Operation(read_method, meta, std::move(copy_coords)); } @@ -382,9 +383,15 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type, } Node4 ShaderIR::GetTexCode(Instruction instr, TextureType texture_type, - TextureProcessMode process_mode, bool depth_compare, bool is_array) { - const bool lod_bias_enabled = - (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ); + TextureProcessMode process_mode, bool depth_compare, bool is_array, + bool is_aoffi) { + const bool lod_bias_enabled{ + (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ)}; + + u64 parameter_register = instr.gpr20.Value(); + if (lod_bias_enabled) { + ++parameter_register; + } const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement( texture_type, depth_compare, is_array, lod_bias_enabled, 4, 5); @@ -404,15 +411,19 @@ Node4 ShaderIR::GetTexCode(Instruction instr, TextureType texture_type, const Node array = is_array ? GetRegister(array_register) : nullptr; + std::vector<Node> aoffi; + if (is_aoffi) { + aoffi = GetAoffiCoordinates(GetRegister(parameter_register++), coord_count, false); + } + Node dc{}; if (depth_compare) { // Depth is always stored in the register signaled by gpr20 or in the next register if lod // or bias are used - const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0); - dc = GetRegister(depth_register); + dc = GetRegister(parameter_register++); } - return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, 0); + return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, 0, aoffi); } Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type, @@ -448,11 +459,11 @@ Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type, dc = GetRegister(depth_register); } - return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, bias_offset); + return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, bias_offset, {}); } Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool depth_compare, - bool is_array) { + bool is_array, bool is_aoffi) { const std::size_t coord_count = GetCoordCount(texture_type); const std::size_t total_coord_count = coord_count + (is_array ? 1 : 0); const std::size_t total_reg_count = total_coord_count + (depth_compare ? 1 : 0); @@ -463,15 +474,27 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de const u64 coord_register = array_register + (is_array ? 1 : 0); std::vector<Node> coords; - for (size_t i = 0; i < coord_count; ++i) + for (std::size_t i = 0; i < coord_count; ++i) { coords.push_back(GetRegister(coord_register + i)); + } + + u64 parameter_register = instr.gpr20.Value(); + std::vector<Node> aoffi; + if (is_aoffi) { + aoffi = GetAoffiCoordinates(GetRegister(parameter_register++), coord_count, true); + } + + Node dc{}; + if (depth_compare) { + dc = GetRegister(parameter_register++); + } const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, depth_compare); Node4 values; for (u32 element = 0; element < values.size(); ++element) { auto coords_copy = coords; - MetaTexture meta{sampler, GetRegister(array_register), {}, {}, {}, {}, element}; + MetaTexture meta{sampler, GetRegister(array_register), dc, aoffi, {}, {}, {}, element}; values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy)); } @@ -507,7 +530,7 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is Node4 values; for (u32 element = 0; element < values.size(); ++element) { auto coords_copy = coords; - MetaTexture meta{sampler, array, {}, {}, lod, {}, element}; + MetaTexture meta{sampler, array, {}, {}, {}, lod, {}, element}; values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy)); } return values; @@ -531,4 +554,45 @@ std::tuple<std::size_t, std::size_t> ShaderIR::ValidateAndGetCoordinateElement( return {coord_count, total_coord_count}; } -} // namespace VideoCommon::Shader
\ No newline at end of file +std::vector<Node> ShaderIR::GetAoffiCoordinates(Node aoffi_reg, std::size_t coord_count, + bool is_tld4) { + const auto [coord_offsets, size, wrap_value, + diff_value] = [is_tld4]() -> std::tuple<std::array<u32, 3>, u32, s32, s32> { + if (is_tld4) { + return {{0, 8, 16}, 6, 32, 64}; + } else { + return {{0, 4, 8}, 4, 8, 16}; + } + }(); + const u32 mask = (1U << size) - 1; + + std::vector<Node> aoffi; + aoffi.reserve(coord_count); + + const auto aoffi_immediate{ + TrackImmediate(aoffi_reg, global_code, static_cast<s64>(global_code.size()))}; + if (!aoffi_immediate) { + // Variable access, not supported on AMD. + LOG_WARNING(HW_GPU, + "AOFFI constant folding failed, some hardware might have graphical issues"); + for (std::size_t coord = 0; coord < coord_count; ++coord) { + const Node value = BitfieldExtract(aoffi_reg, coord_offsets.at(coord), size); + const Node condition = + Operation(OperationCode::LogicalIGreaterEqual, value, Immediate(wrap_value)); + const Node negative = Operation(OperationCode::IAdd, value, Immediate(-diff_value)); + aoffi.push_back(Operation(OperationCode::Select, condition, negative, value)); + } + return aoffi; + } + + for (std::size_t coord = 0; coord < coord_count; ++coord) { + s32 value = (*aoffi_immediate >> coord_offsets.at(coord)) & mask; + if (value >= wrap_value) { + value -= diff_value; + } + aoffi.push_back(Immediate(value)); + } + return aoffi; +} + +} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/xmad.cpp b/src/video_core/shader/decode/xmad.cpp index c34843307..db15c0718 100644 --- a/src/video_core/shader/decode/xmad.cpp +++ b/src/video_core/shader/decode/xmad.cpp @@ -29,39 +29,55 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) { const bool is_signed_b = instr.xmad.sign_b == 1; const bool is_signed_c = is_signed_a; - auto [is_merge, op_b, op_c] = [&]() -> std::tuple<bool, Node, Node> { + auto [is_merge, is_psl, is_high_b, mode, op_b, + op_c] = [&]() -> std::tuple<bool, bool, bool, Tegra::Shader::XmadMode, Node, Node> { switch (opcode->get().GetId()) { case OpCode::Id::XMAD_CR: return {instr.xmad.merge_56, + instr.xmad.product_shift_left_second, + instr.xmad.high_b, + instr.xmad.mode_cbf, GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()), GetRegister(instr.gpr39)}; case OpCode::Id::XMAD_RR: - return {instr.xmad.merge_37, GetRegister(instr.gpr20), GetRegister(instr.gpr39)}; + return {instr.xmad.merge_37, instr.xmad.product_shift_left, instr.xmad.high_b_rr, + instr.xmad.mode, GetRegister(instr.gpr20), GetRegister(instr.gpr39)}; case OpCode::Id::XMAD_RC: - return {false, GetRegister(instr.gpr39), + return {false, + false, + instr.xmad.high_b, + instr.xmad.mode_cbf, + GetRegister(instr.gpr39), GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; case OpCode::Id::XMAD_IMM: - return {instr.xmad.merge_37, Immediate(static_cast<u32>(instr.xmad.imm20_16)), + return {instr.xmad.merge_37, + instr.xmad.product_shift_left, + false, + instr.xmad.mode, + Immediate(static_cast<u32>(instr.xmad.imm20_16)), GetRegister(instr.gpr39)}; } UNIMPLEMENTED_MSG("Unhandled XMAD instruction: {}", opcode->get().GetName()); - return {false, Immediate(0), Immediate(0)}; + return {false, false, false, Tegra::Shader::XmadMode::None, Immediate(0), Immediate(0)}; }(); op_a = BitfieldExtract(op_a, instr.xmad.high_a ? 16 : 0, 16); const Node original_b = op_b; - op_b = BitfieldExtract(op_b, instr.xmad.high_b ? 16 : 0, 16); + op_b = BitfieldExtract(op_b, is_high_b ? 16 : 0, 16); // TODO(Rodrigo): Use an appropiate sign for this operation Node product = Operation(OperationCode::IMul, NO_PRECISE, op_a, op_b); - if (instr.xmad.product_shift_left) { + if (is_psl) { product = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, product, Immediate(16)); } + SetTemporal(bb, 0, product); + product = GetTemporal(0); const Node original_c = op_c; + const Tegra::Shader::XmadMode set_mode = mode; // Workaround to clang compile error op_c = [&]() { - switch (instr.xmad.mode) { + switch (set_mode) { case Tegra::Shader::XmadMode::None: return original_c; case Tegra::Shader::XmadMode::CLo: @@ -80,8 +96,13 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) { } }(); + SetTemporal(bb, 1, op_c); + op_c = GetTemporal(1); + // TODO(Rodrigo): Use an appropiate sign for this operation Node sum = Operation(OperationCode::IAdd, product, op_c); + SetTemporal(bb, 2, sum); + sum = GetTemporal(2); if (is_merge) { const Node a = BitfieldExtract(sum, 0, 16); const Node b = @@ -95,4 +116,4 @@ u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) { return pc; } -} // namespace VideoCommon::Shader
\ No newline at end of file +} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index 5bc3a3900..1afab08c0 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -7,6 +7,7 @@ #include <array> #include <cstring> #include <map> +#include <optional> #include <set> #include <string> #include <tuple> @@ -275,6 +276,11 @@ struct GlobalMemoryBase { } }; +struct GlobalMemoryUsage { + bool is_read{}; + bool is_written{}; +}; + struct MetaArithmetic { bool precise{}; }; @@ -290,6 +296,7 @@ struct MetaTexture { const Sampler& sampler; Node array{}; Node depth_compare{}; + std::vector<Node> aoffi; Node bias{}; Node lod{}; Node component{}; @@ -576,8 +583,8 @@ public: return used_clip_distances; } - const std::set<GlobalMemoryBase>& GetGlobalMemoryBases() const { - return used_global_memory_bases; + const std::map<GlobalMemoryBase, GlobalMemoryUsage>& GetGlobalMemory() const { + return used_global_memory; } std::size_t GetLength() const { @@ -741,14 +748,14 @@ private: Node4 GetTexCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, Tegra::Shader::TextureProcessMode process_mode, bool depth_compare, - bool is_array); + bool is_array, bool is_aoffi); Node4 GetTexsCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, Tegra::Shader::TextureProcessMode process_mode, bool depth_compare, bool is_array); Node4 GetTld4Code(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, - bool depth_compare, bool is_array); + bool depth_compare, bool is_array, bool is_aoffi); Node4 GetTldsCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, bool is_array); @@ -757,9 +764,11 @@ private: Tegra::Shader::TextureType texture_type, bool depth_compare, bool is_array, bool lod_bias_enabled, std::size_t max_coords, std::size_t max_inputs); + std::vector<Node> GetAoffiCoordinates(Node aoffi_reg, std::size_t coord_count, bool is_tld4); + Node4 GetTextureCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, Tegra::Shader::TextureProcessMode process_mode, std::vector<Node> coords, - Node array, Node depth_compare, u32 bias_offset); + Node array, Node depth_compare, u32 bias_offset, std::vector<Node> aoffi); Node GetVideoOperand(Node op, bool is_chunk, bool is_signed, Tegra::Shader::VideoType type, u64 byte_height); @@ -773,8 +782,15 @@ private: Node TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor); + std::optional<u32> TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor); + std::pair<Node, s64> TrackRegister(const GprNode* tracked, const NodeBlock& code, s64 cursor); + std::tuple<Node, Node, GlobalMemoryBase> TrackAndGetGlobalMemory(NodeBlock& bb, + Node addr_register, + u32 immediate_offset, + bool is_write); + template <typename... T> Node Operation(OperationCode code, const T*... operands) { return StoreNode(OperationNode(code, operands...)); @@ -828,7 +844,7 @@ private: std::map<u32, ConstBuffer> used_cbufs; std::set<Sampler> used_samplers; std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances> used_clip_distances{}; - std::set<GlobalMemoryBase> used_global_memory_bases; + std::map<GlobalMemoryBase, GlobalMemoryUsage> used_global_memory; Tegra::Shader::Header header; }; diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp index 33b071747..4505667ff 100644 --- a/src/video_core/shader/track.cpp +++ b/src/video_core/shader/track.cpp @@ -6,6 +6,7 @@ #include <utility> #include <variant> +#include "common/common_types.h" #include "video_core/shader/shader_ir.h" namespace VideoCommon::Shader { @@ -14,7 +15,7 @@ namespace { std::pair<Node, s64> FindOperation(const NodeBlock& code, s64 cursor, OperationCode operation_code) { for (; cursor >= 0; --cursor) { - const Node node = code[cursor]; + const Node node = code.at(cursor); if (const auto operation = std::get_if<OperationNode>(node)) { if (operation->GetCode() == operation_code) return {node, cursor}; @@ -64,6 +65,20 @@ Node ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) { return nullptr; } +std::optional<u32> ShaderIR::TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) { + // Reduce the cursor in one to avoid infinite loops when the instruction sets the same register + // that it uses as operand + const auto [found, found_cursor] = + TrackRegister(&std::get<GprNode>(*tracked), code, cursor - 1); + if (!found) { + return {}; + } + if (const auto immediate = std::get_if<ImmediateNode>(found)) { + return immediate->GetValue(); + } + return {}; +} + std::pair<Node, s64> ShaderIR::TrackRegister(const GprNode* tracked, const NodeBlock& code, s64 cursor) { for (; cursor >= 0; --cursor) { diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp index a7ac26d71..3b022a456 100644 --- a/src/video_core/surface.cpp +++ b/src/video_core/surface.cpp @@ -294,6 +294,8 @@ PixelFormat PixelFormatFromTextureFormat(Tegra::Texture::TextureFormat format, return PixelFormat::Z16; case Tegra::Texture::TextureFormat::Z24S8: return PixelFormat::Z24S8; + case Tegra::Texture::TextureFormat::ZF32_X24S8: + return PixelFormat::Z32FS8; case Tegra::Texture::TextureFormat::DXT1: return is_srgb ? PixelFormat::DXT1_SRGB : PixelFormat::DXT1; case Tegra::Texture::TextureFormat::DXT23: diff --git a/src/video_core/texture_cache.cpp b/src/video_core/texture_cache.cpp new file mode 100644 index 000000000..e96eba7cc --- /dev/null +++ b/src/video_core/texture_cache.cpp @@ -0,0 +1,386 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/alignment.h" +#include "common/assert.h" +#include "common/cityhash.h" +#include "common/common_types.h" +#include "core/core.h" +#include "video_core/surface.h" +#include "video_core/texture_cache.h" +#include "video_core/textures/decoders.h" +#include "video_core/textures/texture.h" + +namespace VideoCommon { + +using VideoCore::Surface::SurfaceTarget; + +using VideoCore::Surface::ComponentTypeFromDepthFormat; +using VideoCore::Surface::ComponentTypeFromRenderTarget; +using VideoCore::Surface::ComponentTypeFromTexture; +using VideoCore::Surface::PixelFormatFromDepthFormat; +using VideoCore::Surface::PixelFormatFromRenderTargetFormat; +using VideoCore::Surface::PixelFormatFromTextureFormat; +using VideoCore::Surface::SurfaceTargetFromTextureType; + +constexpr u32 GetMipmapSize(bool uncompressed, u32 mip_size, u32 tile) { + return uncompressed ? mip_size : std::max(1U, (mip_size + tile - 1) / tile); +} + +SurfaceParams SurfaceParams::CreateForTexture(Core::System& system, + const Tegra::Texture::FullTextureInfo& config) { + SurfaceParams params; + params.is_tiled = config.tic.IsTiled(); + params.block_width = params.is_tiled ? config.tic.BlockWidth() : 0, + params.block_height = params.is_tiled ? config.tic.BlockHeight() : 0, + params.block_depth = params.is_tiled ? config.tic.BlockDepth() : 0, + params.tile_width_spacing = params.is_tiled ? (1 << config.tic.tile_width_spacing.Value()) : 1; + params.pixel_format = + PixelFormatFromTextureFormat(config.tic.format, config.tic.r_type.Value(), false); + params.component_type = ComponentTypeFromTexture(config.tic.r_type.Value()); + params.type = GetFormatType(params.pixel_format); + params.target = SurfaceTargetFromTextureType(config.tic.texture_type); + params.width = Common::AlignUp(config.tic.Width(), GetCompressionFactor(params.pixel_format)); + params.height = Common::AlignUp(config.tic.Height(), GetCompressionFactor(params.pixel_format)); + params.depth = config.tic.Depth(); + if (params.target == SurfaceTarget::TextureCubemap || + params.target == SurfaceTarget::TextureCubeArray) { + params.depth *= 6; + } + params.pitch = params.is_tiled ? 0 : config.tic.Pitch(); + params.unaligned_height = config.tic.Height(); + params.num_levels = config.tic.max_mip_level + 1; + + params.CalculateCachedValues(); + return params; +} + +SurfaceParams SurfaceParams::CreateForDepthBuffer( + Core::System& system, u32 zeta_width, u32 zeta_height, Tegra::DepthFormat format, + u32 block_width, u32 block_height, u32 block_depth, + Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type) { + SurfaceParams params; + params.is_tiled = type == Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout::BlockLinear; + params.block_width = 1 << std::min(block_width, 5U); + params.block_height = 1 << std::min(block_height, 5U); + params.block_depth = 1 << std::min(block_depth, 5U); + params.tile_width_spacing = 1; + params.pixel_format = PixelFormatFromDepthFormat(format); + params.component_type = ComponentTypeFromDepthFormat(format); + params.type = GetFormatType(params.pixel_format); + params.width = zeta_width; + params.height = zeta_height; + params.unaligned_height = zeta_height; + params.target = SurfaceTarget::Texture2D; + params.depth = 1; + params.num_levels = 1; + + params.CalculateCachedValues(); + return params; +} + +SurfaceParams SurfaceParams::CreateForFramebuffer(Core::System& system, std::size_t index) { + const auto& config{system.GPU().Maxwell3D().regs.rt[index]}; + SurfaceParams params; + params.is_tiled = + config.memory_layout.type == Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout::BlockLinear; + params.block_width = 1 << config.memory_layout.block_width; + params.block_height = 1 << config.memory_layout.block_height; + params.block_depth = 1 << config.memory_layout.block_depth; + params.tile_width_spacing = 1; + params.pixel_format = PixelFormatFromRenderTargetFormat(config.format); + params.component_type = ComponentTypeFromRenderTarget(config.format); + params.type = GetFormatType(params.pixel_format); + if (params.is_tiled) { + params.width = config.width; + } else { + const u32 bpp = GetFormatBpp(params.pixel_format) / CHAR_BIT; + params.pitch = config.width; + params.width = params.pitch / bpp; + } + params.height = config.height; + params.depth = 1; + params.unaligned_height = config.height; + params.target = SurfaceTarget::Texture2D; + params.num_levels = 1; + + params.CalculateCachedValues(); + return params; +} + +SurfaceParams SurfaceParams::CreateForFermiCopySurface( + const Tegra::Engines::Fermi2D::Regs::Surface& config) { + SurfaceParams params{}; + params.is_tiled = !config.linear; + params.block_width = params.is_tiled ? std::min(config.BlockWidth(), 32U) : 0, + params.block_height = params.is_tiled ? std::min(config.BlockHeight(), 32U) : 0, + params.block_depth = params.is_tiled ? std::min(config.BlockDepth(), 32U) : 0, + params.tile_width_spacing = 1; + params.pixel_format = PixelFormatFromRenderTargetFormat(config.format); + params.component_type = ComponentTypeFromRenderTarget(config.format); + params.type = GetFormatType(params.pixel_format); + params.width = config.width; + params.height = config.height; + params.unaligned_height = config.height; + // TODO(Rodrigo): Try to guess the surface target from depth and layer parameters + params.target = SurfaceTarget::Texture2D; + params.depth = 1; + params.num_levels = 1; + + params.CalculateCachedValues(); + return params; +} + +u32 SurfaceParams::GetMipWidth(u32 level) const { + return std::max(1U, width >> level); +} + +u32 SurfaceParams::GetMipHeight(u32 level) const { + return std::max(1U, height >> level); +} + +u32 SurfaceParams::GetMipDepth(u32 level) const { + return IsLayered() ? depth : std::max(1U, depth >> level); +} + +bool SurfaceParams::IsLayered() const { + switch (target) { + case SurfaceTarget::Texture1DArray: + case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubeArray: + case SurfaceTarget::TextureCubemap: + return true; + default: + return false; + } +} + +u32 SurfaceParams::GetMipBlockHeight(u32 level) const { + // Auto block resizing algorithm from: + // https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/nv50/nv50_miptree.c + if (level == 0) { + return block_height; + } + const u32 height{GetMipHeight(level)}; + const u32 default_block_height{GetDefaultBlockHeight(pixel_format)}; + const u32 blocks_in_y{(height + default_block_height - 1) / default_block_height}; + u32 block_height = 16; + while (block_height > 1 && blocks_in_y <= block_height * 4) { + block_height >>= 1; + } + return block_height; +} + +u32 SurfaceParams::GetMipBlockDepth(u32 level) const { + if (level == 0) + return block_depth; + if (target != SurfaceTarget::Texture3D) + return 1; + + const u32 depth{GetMipDepth(level)}; + u32 block_depth = 32; + while (block_depth > 1 && depth * 2 <= block_depth) { + block_depth >>= 1; + } + if (block_depth == 32 && GetMipBlockHeight(level) >= 4) { + return 16; + } + return block_depth; +} + +std::size_t SurfaceParams::GetGuestMipmapLevelOffset(u32 level) const { + std::size_t offset = 0; + for (u32 i = 0; i < level; i++) { + offset += GetInnerMipmapMemorySize(i, false, IsLayered(), false); + } + return offset; +} + +std::size_t SurfaceParams::GetHostMipmapLevelOffset(u32 level) const { + std::size_t offset = 0; + for (u32 i = 0; i < level; i++) { + offset += GetInnerMipmapMemorySize(i, true, false, false); + } + return offset; +} + +std::size_t SurfaceParams::GetGuestLayerSize() const { + return GetInnerMemorySize(false, true, false); +} + +std::size_t SurfaceParams::GetHostLayerSize(u32 level) const { + return GetInnerMipmapMemorySize(level, true, IsLayered(), false); +} + +bool SurfaceParams::IsFamiliar(const SurfaceParams& view_params) const { + if (std::tie(is_tiled, tile_width_spacing, pixel_format, component_type, type) != + std::tie(view_params.is_tiled, view_params.tile_width_spacing, view_params.pixel_format, + view_params.component_type, view_params.type)) { + return false; + } + + const SurfaceTarget view_target{view_params.target}; + if (view_target == target) { + return true; + } + + switch (target) { + case SurfaceTarget::Texture1D: + case SurfaceTarget::Texture2D: + case SurfaceTarget::Texture3D: + return false; + case SurfaceTarget::Texture1DArray: + return view_target == SurfaceTarget::Texture1D; + case SurfaceTarget::Texture2DArray: + return view_target == SurfaceTarget::Texture2D; + case SurfaceTarget::TextureCubemap: + return view_target == SurfaceTarget::Texture2D || + view_target == SurfaceTarget::Texture2DArray; + case SurfaceTarget::TextureCubeArray: + return view_target == SurfaceTarget::Texture2D || + view_target == SurfaceTarget::Texture2DArray || + view_target == SurfaceTarget::TextureCubemap; + default: + UNIMPLEMENTED_MSG("Unimplemented texture family={}", static_cast<u32>(target)); + return false; + } +} + +bool SurfaceParams::IsPixelFormatZeta() const { + return pixel_format >= VideoCore::Surface::PixelFormat::MaxColorFormat && + pixel_format < VideoCore::Surface::PixelFormat::MaxDepthStencilFormat; +} + +void SurfaceParams::CalculateCachedValues() { + guest_size_in_bytes = GetInnerMemorySize(false, false, false); + + // ASTC is uncompressed in software, in emulated as RGBA8 + if (IsPixelFormatASTC(pixel_format)) { + host_size_in_bytes = width * height * depth * 4; + } else { + host_size_in_bytes = GetInnerMemorySize(true, false, false); + } + + switch (target) { + case SurfaceTarget::Texture1D: + case SurfaceTarget::Texture2D: + case SurfaceTarget::Texture3D: + num_layers = 1; + break; + case SurfaceTarget::Texture1DArray: + case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubemap: + case SurfaceTarget::TextureCubeArray: + num_layers = depth; + break; + default: + UNREACHABLE(); + } +} + +std::size_t SurfaceParams::GetInnerMipmapMemorySize(u32 level, bool as_host_size, bool layer_only, + bool uncompressed) const { + const bool tiled{as_host_size ? false : is_tiled}; + const u32 tile_x{GetDefaultBlockWidth(pixel_format)}; + const u32 tile_y{GetDefaultBlockHeight(pixel_format)}; + const u32 width{GetMipmapSize(uncompressed, GetMipWidth(level), tile_x)}; + const u32 height{GetMipmapSize(uncompressed, GetMipHeight(level), tile_y)}; + const u32 depth{layer_only ? 1U : GetMipDepth(level)}; + return Tegra::Texture::CalculateSize(tiled, GetBytesPerPixel(pixel_format), width, height, + depth, GetMipBlockHeight(level), GetMipBlockDepth(level)); +} + +std::size_t SurfaceParams::GetInnerMemorySize(bool as_host_size, bool layer_only, + bool uncompressed) const { + std::size_t size = 0; + for (u32 level = 0; level < num_levels; ++level) { + size += GetInnerMipmapMemorySize(level, as_host_size, layer_only, uncompressed); + } + if (!as_host_size && is_tiled) { + size = Common::AlignUp(size, Tegra::Texture::GetGOBSize() * block_height * block_depth); + } + return size; +} + +std::map<u64, std::pair<u32, u32>> SurfaceParams::CreateViewOffsetMap() const { + std::map<u64, std::pair<u32, u32>> view_offset_map; + switch (target) { + case SurfaceTarget::Texture1D: + case SurfaceTarget::Texture2D: + case SurfaceTarget::Texture3D: { + constexpr u32 layer = 0; + for (u32 level = 0; level < num_levels; ++level) { + const std::size_t offset{GetGuestMipmapLevelOffset(level)}; + view_offset_map.insert({offset, {layer, level}}); + } + break; + } + case SurfaceTarget::Texture1DArray: + case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubemap: + case SurfaceTarget::TextureCubeArray: { + const std::size_t layer_size{GetGuestLayerSize()}; + for (u32 level = 0; level < num_levels; ++level) { + const std::size_t level_offset{GetGuestMipmapLevelOffset(level)}; + for (u32 layer = 0; layer < num_layers; ++layer) { + const auto layer_offset{static_cast<std::size_t>(layer_size * layer)}; + const std::size_t offset{level_offset + layer_offset}; + view_offset_map.insert({offset, {layer, level}}); + } + } + break; + } + default: + UNIMPLEMENTED_MSG("Unimplemented surface target {}", static_cast<u32>(target)); + } + return view_offset_map; +} + +bool SurfaceParams::IsViewValid(const SurfaceParams& view_params, u32 layer, u32 level) const { + return IsDimensionValid(view_params, level) && IsDepthValid(view_params, level) && + IsInBounds(view_params, layer, level); +} + +bool SurfaceParams::IsDimensionValid(const SurfaceParams& view_params, u32 level) const { + return view_params.width == GetMipWidth(level) && view_params.height == GetMipHeight(level); +} + +bool SurfaceParams::IsDepthValid(const SurfaceParams& view_params, u32 level) const { + if (view_params.target != SurfaceTarget::Texture3D) { + return true; + } + return view_params.depth == GetMipDepth(level); +} + +bool SurfaceParams::IsInBounds(const SurfaceParams& view_params, u32 layer, u32 level) const { + return layer + view_params.num_layers <= num_layers && + level + view_params.num_levels <= num_levels; +} + +std::size_t HasheableSurfaceParams::Hash() const { + return static_cast<std::size_t>( + Common::CityHash64(reinterpret_cast<const char*>(this), sizeof(*this))); +} + +bool HasheableSurfaceParams::operator==(const HasheableSurfaceParams& rhs) const { + return std::tie(is_tiled, block_width, block_height, block_depth, tile_width_spacing, width, + height, depth, pitch, unaligned_height, num_levels, pixel_format, + component_type, type, target) == + std::tie(rhs.is_tiled, rhs.block_width, rhs.block_height, rhs.block_depth, + rhs.tile_width_spacing, rhs.width, rhs.height, rhs.depth, rhs.pitch, + rhs.unaligned_height, rhs.num_levels, rhs.pixel_format, rhs.component_type, + rhs.type, rhs.target); +} + +std::size_t ViewKey::Hash() const { + return static_cast<std::size_t>( + Common::CityHash64(reinterpret_cast<const char*>(this), sizeof(*this))); +} + +bool ViewKey::operator==(const ViewKey& rhs) const { + return std::tie(base_layer, num_layers, base_level, num_levels) == + std::tie(rhs.base_layer, rhs.num_layers, rhs.base_level, rhs.num_levels); +} + +} // namespace VideoCommon diff --git a/src/video_core/texture_cache.h b/src/video_core/texture_cache.h new file mode 100644 index 000000000..041551691 --- /dev/null +++ b/src/video_core/texture_cache.h @@ -0,0 +1,586 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <list> +#include <memory> +#include <set> +#include <tuple> +#include <type_traits> +#include <unordered_map> + +#include <boost/icl/interval_map.hpp> +#include <boost/range/iterator_range.hpp> + +#include "common/assert.h" +#include "common/common_types.h" +#include "core/memory.h" +#include "video_core/engines/fermi_2d.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/gpu.h" +#include "video_core/rasterizer_interface.h" +#include "video_core/surface.h" + +namespace Core { +class System; +} + +namespace Tegra::Texture { +struct FullTextureInfo; +} + +namespace VideoCore { +class RasterizerInterface; +} + +namespace VideoCommon { + +class HasheableSurfaceParams { +public: + std::size_t Hash() const; + + bool operator==(const HasheableSurfaceParams& rhs) const; + +protected: + // Avoid creation outside of a managed environment. + HasheableSurfaceParams() = default; + + bool is_tiled; + u32 block_width; + u32 block_height; + u32 block_depth; + u32 tile_width_spacing; + u32 width; + u32 height; + u32 depth; + u32 pitch; + u32 unaligned_height; + u32 num_levels; + VideoCore::Surface::PixelFormat pixel_format; + VideoCore::Surface::ComponentType component_type; + VideoCore::Surface::SurfaceType type; + VideoCore::Surface::SurfaceTarget target; +}; + +class SurfaceParams final : public HasheableSurfaceParams { +public: + /// Creates SurfaceCachedParams from a texture configuration. + static SurfaceParams CreateForTexture(Core::System& system, + const Tegra::Texture::FullTextureInfo& config); + + /// Creates SurfaceCachedParams for a depth buffer configuration. + static SurfaceParams CreateForDepthBuffer( + Core::System& system, u32 zeta_width, u32 zeta_height, Tegra::DepthFormat format, + u32 block_width, u32 block_height, u32 block_depth, + Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type); + + /// Creates SurfaceCachedParams from a framebuffer configuration. + static SurfaceParams CreateForFramebuffer(Core::System& system, std::size_t index); + + /// Creates SurfaceCachedParams from a Fermi2D surface configuration. + static SurfaceParams CreateForFermiCopySurface( + const Tegra::Engines::Fermi2D::Regs::Surface& config); + + bool IsTiled() const { + return is_tiled; + } + + u32 GetBlockWidth() const { + return block_width; + } + + u32 GetTileWidthSpacing() const { + return tile_width_spacing; + } + + u32 GetWidth() const { + return width; + } + + u32 GetHeight() const { + return height; + } + + u32 GetDepth() const { + return depth; + } + + u32 GetPitch() const { + return pitch; + } + + u32 GetNumLevels() const { + return num_levels; + } + + VideoCore::Surface::PixelFormat GetPixelFormat() const { + return pixel_format; + } + + VideoCore::Surface::ComponentType GetComponentType() const { + return component_type; + } + + VideoCore::Surface::SurfaceTarget GetTarget() const { + return target; + } + + VideoCore::Surface::SurfaceType GetType() const { + return type; + } + + std::size_t GetGuestSizeInBytes() const { + return guest_size_in_bytes; + } + + std::size_t GetHostSizeInBytes() const { + return host_size_in_bytes; + } + + u32 GetNumLayers() const { + return num_layers; + } + + /// Returns the width of a given mipmap level. + u32 GetMipWidth(u32 level) const; + + /// Returns the height of a given mipmap level. + u32 GetMipHeight(u32 level) const; + + /// Returns the depth of a given mipmap level. + u32 GetMipDepth(u32 level) const; + + /// Returns true if these parameters are from a layered surface. + bool IsLayered() const; + + /// Returns the block height of a given mipmap level. + u32 GetMipBlockHeight(u32 level) const; + + /// Returns the block depth of a given mipmap level. + u32 GetMipBlockDepth(u32 level) const; + + /// Returns the offset in bytes in guest memory of a given mipmap level. + std::size_t GetGuestMipmapLevelOffset(u32 level) const; + + /// Returns the offset in bytes in host memory (linear) of a given mipmap level. + std::size_t GetHostMipmapLevelOffset(u32 level) const; + + /// Returns the size of a layer in bytes in guest memory. + std::size_t GetGuestLayerSize() const; + + /// Returns the size of a layer in bytes in host memory for a given mipmap level. + std::size_t GetHostLayerSize(u32 level) const; + + /// Returns true if another surface can be familiar with this. This is a loosely defined term + /// that reflects the possibility of these two surface parameters potentially being part of a + /// bigger superset. + bool IsFamiliar(const SurfaceParams& view_params) const; + + /// Returns true if the pixel format is a depth and/or stencil format. + bool IsPixelFormatZeta() const; + + /// Creates a map that redirects an address difference to a layer and mipmap level. + std::map<u64, std::pair<u32, u32>> CreateViewOffsetMap() const; + + /// Returns true if the passed surface view parameters is equal or a valid subset of this. + bool IsViewValid(const SurfaceParams& view_params, u32 layer, u32 level) const; + +private: + /// Calculates values that can be deduced from HasheableSurfaceParams. + void CalculateCachedValues(); + + /// Returns the size of a given mipmap level. + std::size_t GetInnerMipmapMemorySize(u32 level, bool as_host_size, bool layer_only, + bool uncompressed) const; + + /// Returns the size of all mipmap levels and aligns as needed. + std::size_t GetInnerMemorySize(bool as_host_size, bool layer_only, bool uncompressed) const; + + /// Returns true if the passed view width and height match the size of this params in a given + /// mipmap level. + bool IsDimensionValid(const SurfaceParams& view_params, u32 level) const; + + /// Returns true if the passed view depth match the size of this params in a given mipmap level. + bool IsDepthValid(const SurfaceParams& view_params, u32 level) const; + + /// Returns true if the passed view layers and mipmap levels are in bounds. + bool IsInBounds(const SurfaceParams& view_params, u32 layer, u32 level) const; + + std::size_t guest_size_in_bytes; + std::size_t host_size_in_bytes; + u32 num_layers; +}; + +struct ViewKey { + std::size_t Hash() const; + + bool operator==(const ViewKey& rhs) const; + + u32 base_layer{}; + u32 num_layers{}; + u32 base_level{}; + u32 num_levels{}; +}; + +} // namespace VideoCommon + +namespace std { + +template <> +struct hash<VideoCommon::SurfaceParams> { + std::size_t operator()(const VideoCommon::SurfaceParams& k) const noexcept { + return k.Hash(); + } +}; + +template <> +struct hash<VideoCommon::ViewKey> { + std::size_t operator()(const VideoCommon::ViewKey& k) const noexcept { + return k.Hash(); + } +}; + +} // namespace std + +namespace VideoCommon { + +template <typename TView, typename TExecutionContext> +class SurfaceBase { + static_assert(std::is_trivially_copyable_v<TExecutionContext>); + +public: + virtual void LoadBuffer() = 0; + + virtual TExecutionContext FlushBuffer(TExecutionContext exctx) = 0; + + virtual TExecutionContext UploadTexture(TExecutionContext exctx) = 0; + + TView* TryGetView(VAddr view_addr, const SurfaceParams& view_params) { + if (view_addr < cpu_addr || !params.IsFamiliar(view_params)) { + // It can't be a view if it's in a prior address. + return {}; + } + + const auto relative_offset{static_cast<u64>(view_addr - cpu_addr)}; + const auto it{view_offset_map.find(relative_offset)}; + if (it == view_offset_map.end()) { + // Couldn't find an aligned view. + return {}; + } + const auto [layer, level] = it->second; + + if (!params.IsViewValid(view_params, layer, level)) { + return {}; + } + + return GetView(layer, view_params.GetNumLayers(), level, view_params.GetNumLevels()); + } + + VAddr GetCpuAddr() const { + ASSERT(is_registered); + return cpu_addr; + } + + u8* GetHostPtr() const { + ASSERT(is_registered); + return host_ptr; + } + + CacheAddr GetCacheAddr() const { + ASSERT(is_registered); + return cache_addr; + } + + std::size_t GetSizeInBytes() const { + return params.GetGuestSizeInBytes(); + } + + void MarkAsModified(bool is_modified_) { + is_modified = is_modified_; + } + + const SurfaceParams& GetSurfaceParams() const { + return params; + } + + TView* GetView(VAddr view_addr, const SurfaceParams& view_params) { + TView* view{TryGetView(view_addr, view_params)}; + ASSERT(view != nullptr); + return view; + } + + void Register(VAddr cpu_addr_, u8* host_ptr_) { + ASSERT(!is_registered); + is_registered = true; + cpu_addr = cpu_addr_; + host_ptr = host_ptr_; + cache_addr = ToCacheAddr(host_ptr_); + } + + void Register(VAddr cpu_addr_) { + Register(cpu_addr_, Memory::GetPointer(cpu_addr_)); + } + + void Unregister() { + ASSERT(is_registered); + is_registered = false; + } + + bool IsRegistered() const { + return is_registered; + } + +protected: + explicit SurfaceBase(const SurfaceParams& params) + : params{params}, view_offset_map{params.CreateViewOffsetMap()} {} + + ~SurfaceBase() = default; + + virtual std::unique_ptr<TView> CreateView(const ViewKey& view_key) = 0; + + bool IsModified() const { + return is_modified; + } + + const SurfaceParams params; + +private: + TView* GetView(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels) { + const ViewKey key{base_layer, num_layers, base_level, num_levels}; + const auto [entry, is_cache_miss] = views.try_emplace(key); + auto& view{entry->second}; + if (is_cache_miss) { + view = CreateView(key); + } + return view.get(); + } + + const std::map<u64, std::pair<u32, u32>> view_offset_map; + + VAddr cpu_addr{}; + u8* host_ptr{}; + CacheAddr cache_addr{}; + bool is_modified{}; + bool is_registered{}; + std::unordered_map<ViewKey, std::unique_ptr<TView>> views; +}; + +template <typename TSurface, typename TView, typename TExecutionContext> +class TextureCache { + static_assert(std::is_trivially_copyable_v<TExecutionContext>); + using ResultType = std::tuple<TView*, TExecutionContext>; + using IntervalMap = boost::icl::interval_map<CacheAddr, std::set<TSurface*>>; + using IntervalType = typename IntervalMap::interval_type; + +public: + void InvalidateRegion(CacheAddr addr, std::size_t size) { + for (TSurface* surface : GetSurfacesInRegion(addr, size)) { + if (!surface->IsRegistered()) { + // Skip duplicates + continue; + } + Unregister(surface); + } + } + + ResultType GetTextureSurface(TExecutionContext exctx, + const Tegra::Texture::FullTextureInfo& config) { + auto& memory_manager{system.GPU().MemoryManager()}; + const auto cpu_addr{memory_manager.GpuToCpuAddress(config.tic.Address())}; + if (!cpu_addr) { + return {{}, exctx}; + } + const auto params{SurfaceParams::CreateForTexture(system, config)}; + return GetSurfaceView(exctx, *cpu_addr, params, true); + } + + ResultType GetDepthBufferSurface(TExecutionContext exctx, bool preserve_contents) { + const auto& regs{system.GPU().Maxwell3D().regs}; + if (!regs.zeta.Address() || !regs.zeta_enable) { + return {{}, exctx}; + } + + auto& memory_manager{system.GPU().MemoryManager()}; + const auto cpu_addr{memory_manager.GpuToCpuAddress(regs.zeta.Address())}; + if (!cpu_addr) { + return {{}, exctx}; + } + + const auto depth_params{SurfaceParams::CreateForDepthBuffer( + system, regs.zeta_width, regs.zeta_height, regs.zeta.format, + regs.zeta.memory_layout.block_width, regs.zeta.memory_layout.block_height, + regs.zeta.memory_layout.block_depth, regs.zeta.memory_layout.type)}; + return GetSurfaceView(exctx, *cpu_addr, depth_params, preserve_contents); + } + + ResultType GetColorBufferSurface(TExecutionContext exctx, std::size_t index, + bool preserve_contents) { + ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets); + + const auto& regs{system.GPU().Maxwell3D().regs}; + if (index >= regs.rt_control.count || regs.rt[index].Address() == 0 || + regs.rt[index].format == Tegra::RenderTargetFormat::NONE) { + return {{}, exctx}; + } + + auto& memory_manager{system.GPU().MemoryManager()}; + const auto& config{system.GPU().Maxwell3D().regs.rt[index]}; + const auto cpu_addr{memory_manager.GpuToCpuAddress( + config.Address() + config.base_layer * config.layer_stride * sizeof(u32))}; + if (!cpu_addr) { + return {{}, exctx}; + } + + return GetSurfaceView(exctx, *cpu_addr, SurfaceParams::CreateForFramebuffer(system, index), + preserve_contents); + } + + ResultType GetFermiSurface(TExecutionContext exctx, + const Tegra::Engines::Fermi2D::Regs::Surface& config) { + const auto cpu_addr{system.GPU().MemoryManager().GpuToCpuAddress(config.Address())}; + ASSERT(cpu_addr); + return GetSurfaceView(exctx, *cpu_addr, SurfaceParams::CreateForFermiCopySurface(config), + true); + } + + TSurface* TryFindFramebufferSurface(const u8* host_ptr) const { + const auto it{registered_surfaces.find(ToCacheAddr(host_ptr))}; + return it != registered_surfaces.end() ? *it->second.begin() : nullptr; + } + +protected: + TextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer) + : system{system}, rasterizer{rasterizer} {} + + ~TextureCache() = default; + + virtual ResultType TryFastGetSurfaceView(TExecutionContext exctx, VAddr cpu_addr, u8* host_ptr, + const SurfaceParams& params, bool preserve_contents, + const std::vector<TSurface*>& overlaps) = 0; + + virtual std::unique_ptr<TSurface> CreateSurface(const SurfaceParams& params) = 0; + + void Register(TSurface* surface, VAddr cpu_addr, u8* host_ptr) { + surface->Register(cpu_addr, host_ptr); + registered_surfaces.add({GetSurfaceInterval(surface), {surface}}); + rasterizer.UpdatePagesCachedCount(surface->GetCpuAddr(), surface->GetSizeInBytes(), 1); + } + + void Unregister(TSurface* surface) { + registered_surfaces.subtract({GetSurfaceInterval(surface), {surface}}); + rasterizer.UpdatePagesCachedCount(surface->GetCpuAddr(), surface->GetSizeInBytes(), -1); + surface->Unregister(); + } + + TSurface* GetUncachedSurface(const SurfaceParams& params) { + if (TSurface* surface = TryGetReservedSurface(params); surface) + return surface; + // No reserved surface available, create a new one and reserve it + auto new_surface{CreateSurface(params)}; + TSurface* surface{new_surface.get()}; + ReserveSurface(params, std::move(new_surface)); + return surface; + } + + Core::System& system; + +private: + ResultType GetSurfaceView(TExecutionContext exctx, VAddr cpu_addr, const SurfaceParams& params, + bool preserve_contents) { + const auto host_ptr{Memory::GetPointer(cpu_addr)}; + const auto cache_addr{ToCacheAddr(host_ptr)}; + const auto overlaps{GetSurfacesInRegion(cache_addr, params.GetGuestSizeInBytes())}; + if (overlaps.empty()) { + return LoadSurfaceView(exctx, cpu_addr, host_ptr, params, preserve_contents); + } + + if (overlaps.size() == 1) { + if (TView* view = overlaps[0]->TryGetView(cpu_addr, params); view) + return {view, exctx}; + } + + TView* fast_view; + std::tie(fast_view, exctx) = + TryFastGetSurfaceView(exctx, cpu_addr, host_ptr, params, preserve_contents, overlaps); + + for (TSurface* surface : overlaps) { + if (!fast_view) { + // Flush even when we don't care about the contents, to preserve memory not written + // by the new surface. + exctx = surface->FlushBuffer(exctx); + } + Unregister(surface); + } + + if (fast_view) { + return {fast_view, exctx}; + } + + return LoadSurfaceView(exctx, cpu_addr, host_ptr, params, preserve_contents); + } + + ResultType LoadSurfaceView(TExecutionContext exctx, VAddr cpu_addr, u8* host_ptr, + const SurfaceParams& params, bool preserve_contents) { + TSurface* new_surface{GetUncachedSurface(params)}; + Register(new_surface, cpu_addr, host_ptr); + if (preserve_contents) { + exctx = LoadSurface(exctx, new_surface); + } + return {new_surface->GetView(cpu_addr, params), exctx}; + } + + TExecutionContext LoadSurface(TExecutionContext exctx, TSurface* surface) { + surface->LoadBuffer(); + exctx = surface->UploadTexture(exctx); + surface->MarkAsModified(false); + return exctx; + } + + std::vector<TSurface*> GetSurfacesInRegion(CacheAddr cache_addr, std::size_t size) const { + if (size == 0) { + return {}; + } + const IntervalType interval{cache_addr, cache_addr + size}; + + std::vector<TSurface*> surfaces; + for (auto& pair : boost::make_iterator_range(registered_surfaces.equal_range(interval))) { + surfaces.push_back(*pair.second.begin()); + } + return surfaces; + } + + void ReserveSurface(const SurfaceParams& params, std::unique_ptr<TSurface> surface) { + surface_reserve[params].push_back(std::move(surface)); + } + + TSurface* TryGetReservedSurface(const SurfaceParams& params) { + auto search{surface_reserve.find(params)}; + if (search == surface_reserve.end()) { + return {}; + } + for (auto& surface : search->second) { + if (!surface->IsRegistered()) { + return surface.get(); + } + } + return {}; + } + + IntervalType GetSurfaceInterval(TSurface* surface) const { + return IntervalType::right_open(surface->GetCacheAddr(), + surface->GetCacheAddr() + surface->GetSizeInBytes()); + } + + VideoCore::RasterizerInterface& rasterizer; + + IntervalMap registered_surfaces; + + /// The surface reserve is a "backup" cache, this is where we put unique surfaces that have + /// previously been used. This is to prevent surfaces from being constantly created and + /// destroyed when used with different surface parameters. + std::unordered_map<SurfaceParams, std::list<std::unique_ptr<TSurface>>> surface_reserve; +}; + +} // namespace VideoCommon diff --git a/src/video_core/textures/convert.cpp b/src/video_core/textures/convert.cpp index 5e439f036..82050bd51 100644 --- a/src/video_core/textures/convert.cpp +++ b/src/video_core/textures/convert.cpp @@ -10,6 +10,7 @@ #include "common/assert.h" #include "common/common_types.h" #include "common/logging/log.h" +#include "video_core/surface.h" #include "video_core/textures/astc.h" #include "video_core/textures/convert.h" diff --git a/src/video_core/textures/convert.h b/src/video_core/textures/convert.h index 07cd8b5da..12542e71c 100644 --- a/src/video_core/textures/convert.h +++ b/src/video_core/textures/convert.h @@ -5,7 +5,10 @@ #pragma once #include "common/common_types.h" -#include "video_core/surface.h" + +namespace VideoCore::Surface { +enum class PixelFormat; +} namespace Tegra::Texture { diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h index 93ecc6e31..bea0d5bc2 100644 --- a/src/video_core/textures/texture.h +++ b/src/video_core/textures/texture.h @@ -7,9 +7,7 @@ #include <array> #include "common/assert.h" #include "common/bit_field.h" -#include "common/common_funcs.h" #include "common/common_types.h" -#include "video_core/memory_manager.h" namespace Tegra::Texture { |