diff options
Diffstat (limited to 'src/video_core')
38 files changed, 5413 insertions, 346 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index ed87f8ff1..a710c4bc5 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -1,24 +1,44 @@ add_library(video_core STATIC command_processor.cpp command_processor.h + debug_utils/debug_utils.cpp + debug_utils/debug_utils.h engines/fermi_2d.cpp engines/fermi_2d.h engines/maxwell_3d.cpp engines/maxwell_3d.h engines/maxwell_compute.cpp engines/maxwell_compute.h + gpu.cpp gpu.h + macro_interpreter.cpp + macro_interpreter.h memory_manager.cpp memory_manager.h + rasterizer_interface.h renderer_base.cpp renderer_base.h + renderer_opengl/gl_rasterizer.cpp + renderer_opengl/gl_rasterizer.h + renderer_opengl/gl_rasterizer_cache.cpp + renderer_opengl/gl_rasterizer_cache.h renderer_opengl/gl_resource_manager.h + renderer_opengl/gl_shader_decompiler.cpp + renderer_opengl/gl_shader_decompiler.h + renderer_opengl/gl_shader_gen.cpp + renderer_opengl/gl_shader_gen.h renderer_opengl/gl_shader_util.cpp renderer_opengl/gl_shader_util.h renderer_opengl/gl_state.cpp renderer_opengl/gl_state.h + renderer_opengl/gl_stream_buffer.cpp + renderer_opengl/gl_stream_buffer.h + renderer_opengl/maxwell_to_gl.h renderer_opengl/renderer_opengl.cpp renderer_opengl/renderer_opengl.h + textures/decoders.cpp + textures/decoders.h + textures/texture.h utils.h video_core.cpp video_core.h diff --git a/src/video_core/command_processor.cpp b/src/video_core/command_processor.cpp index 21d672085..d4cdb4ab2 100644 --- a/src/video_core/command_processor.cpp +++ b/src/video_core/command_processor.cpp @@ -24,12 +24,37 @@ namespace Tegra { enum class BufferMethods { BindObject = 0, + SetGraphMacroCode = 0x45, + SetGraphMacroCodeArg = 0x46, + SetGraphMacroEntry = 0x47, CountBufferMethods = 0x100, }; -void GPU::WriteReg(u32 method, u32 subchannel, u32 value) { - LOG_WARNING(HW_GPU, "Processing method %08X on subchannel %u value %08X", method, subchannel, - value); +void GPU::WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params) { + LOG_WARNING(HW_GPU, "Processing method %08X on subchannel %u value %08X remaining params %u", + method, subchannel, value, remaining_params); + + if (method == static_cast<u32>(BufferMethods::SetGraphMacroEntry)) { + // Prepare to upload a new macro, reset the upload counter. + LOG_DEBUG(HW_GPU, "Uploading GPU macro %08X", value); + current_macro_entry = value; + current_macro_code.clear(); + return; + } + + if (method == static_cast<u32>(BufferMethods::SetGraphMacroCodeArg)) { + // Append a new code word to the current macro. + current_macro_code.push_back(value); + + // There are no more params remaining, submit the code to the 3D engine. + if (remaining_params == 0) { + maxwell_3d->SubmitMacroCode(current_macro_entry, std::move(current_macro_code)); + current_macro_entry = InvalidGraphMacroEntry; + current_macro_code.clear(); + } + + return; + } if (method == static_cast<u32>(BufferMethods::BindObject)) { // Bind the current subchannel to the desired engine id. @@ -54,7 +79,7 @@ void GPU::WriteReg(u32 method, u32 subchannel, u32 value) { fermi_2d->WriteReg(method, value); break; case EngineID::MAXWELL_B: - maxwell_3d->WriteReg(method, value); + maxwell_3d->WriteReg(method, value, remaining_params); break; case EngineID::MAXWELL_COMPUTE_B: maxwell_compute->WriteReg(method, value); @@ -78,7 +103,8 @@ void GPU::ProcessCommandList(GPUVAddr address, u32 size) { case SubmissionMode::Increasing: { // Increase the method value with each argument. for (unsigned i = 0; i < header.arg_count; ++i) { - WriteReg(header.method + i, header.subchannel, Memory::Read32(current_addr)); + WriteReg(header.method + i, header.subchannel, Memory::Read32(current_addr), + header.arg_count - i - 1); current_addr += sizeof(u32); } break; @@ -87,27 +113,31 @@ void GPU::ProcessCommandList(GPUVAddr address, u32 size) { case SubmissionMode::NonIncreasing: { // Use the same method value for all arguments. for (unsigned i = 0; i < header.arg_count; ++i) { - WriteReg(header.method, header.subchannel, Memory::Read32(current_addr)); + WriteReg(header.method, header.subchannel, Memory::Read32(current_addr), + header.arg_count - i - 1); current_addr += sizeof(u32); } break; } case SubmissionMode::IncreaseOnce: { ASSERT(header.arg_count.Value() >= 1); + // Use the original method for the first argument and then the next method for all other // arguments. - WriteReg(header.method, header.subchannel, Memory::Read32(current_addr)); + WriteReg(header.method, header.subchannel, Memory::Read32(current_addr), + header.arg_count - 1); current_addr += sizeof(u32); - // Use the same method value for all arguments. + for (unsigned i = 1; i < header.arg_count; ++i) { - WriteReg(header.method + 1, header.subchannel, Memory::Read32(current_addr)); + WriteReg(header.method + 1, header.subchannel, Memory::Read32(current_addr), + header.arg_count - i - 1); current_addr += sizeof(u32); } break; } case SubmissionMode::Inline: { // The register value is stored in the bits 16-28 as an immediate - WriteReg(header.method, header.subchannel, header.inline_data); + WriteReg(header.method, header.subchannel, header.inline_data, 0); break; } default: diff --git a/src/video_core/command_processor.h b/src/video_core/command_processor.h index b511bfcf7..f7214ffec 100644 --- a/src/video_core/command_processor.h +++ b/src/video_core/command_processor.h @@ -34,6 +34,4 @@ static_assert(std::is_standard_layout<CommandHeader>::value == true, "CommandHeader does not use standard layout"); static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!"); -void ProcessCommandList(VAddr address, u32 size); - } // namespace Tegra diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp new file mode 100644 index 000000000..22d44aab2 --- /dev/null +++ b/src/video_core/debug_utils/debug_utils.cpp @@ -0,0 +1,64 @@ +// Copyright 2014 Citra Emulator Project +// Licensed under GPLv2 +// Refer to the license.txt file included. + +#include <algorithm> +#include <condition_variable> +#include <cstdint> +#include <cstring> +#include <fstream> +#include <map> +#include <mutex> +#include <string> + +#include "common/assert.h" +#include "common/bit_field.h" +#include "common/color.h" +#include "common/common_types.h" +#include "common/file_util.h" +#include "common/logging/log.h" +#include "common/math_util.h" +#include "common/vector_math.h" +#include "video_core/debug_utils/debug_utils.h" + +namespace Tegra { + +void DebugContext::DoOnEvent(Event event, void* data) { + { + std::unique_lock<std::mutex> lock(breakpoint_mutex); + + // TODO(Subv): Commit the rasterizer's caches so framebuffers, render targets, etc. will + // show on debug widgets + + // TODO: Should stop the CPU thread here once we multithread emulation. + + active_breakpoint = event; + at_breakpoint = true; + + // Tell all observers that we hit a breakpoint + for (auto& breakpoint_observer : breakpoint_observers) { + breakpoint_observer->OnMaxwellBreakPointHit(event, data); + } + + // Wait until another thread tells us to Resume() + resume_from_breakpoint.wait(lock, [&] { return !at_breakpoint; }); + } +} + +void DebugContext::Resume() { + { + std::lock_guard<std::mutex> lock(breakpoint_mutex); + + // Tell all observers that we are about to resume + for (auto& breakpoint_observer : breakpoint_observers) { + breakpoint_observer->OnMaxwellResume(); + } + + // Resume the waiting thread (i.e. OnEvent()) + at_breakpoint = false; + } + + resume_from_breakpoint.notify_one(); +} + +} // namespace Tegra diff --git a/src/video_core/debug_utils/debug_utils.h b/src/video_core/debug_utils/debug_utils.h new file mode 100644 index 000000000..bbba8e380 --- /dev/null +++ b/src/video_core/debug_utils/debug_utils.h @@ -0,0 +1,163 @@ +// Copyright 2014 Citra Emulator Project +// Licensed under GPLv2 +// Refer to the license.txt file included. + +#pragma once + +#include <algorithm> +#include <array> +#include <condition_variable> +#include <iterator> +#include <list> +#include <map> +#include <memory> +#include <mutex> +#include <string> +#include <utility> +#include <vector> +#include "common/common_types.h" +#include "common/vector_math.h" + +namespace Tegra { + +class DebugContext { +public: + enum class Event { + FirstEvent = 0, + + MaxwellCommandLoaded = FirstEvent, + MaxwellCommandProcessed, + IncomingPrimitiveBatch, + FinishedPrimitiveBatch, + + NumEvents + }; + + /** + * Inherit from this class to be notified of events registered to some debug context. + * Most importantly this is used for our debugger GUI. + * + * To implement event handling, override the OnMaxwellBreakPointHit and OnMaxwellResume methods. + * @warning All BreakPointObservers need to be on the same thread to guarantee thread-safe state + * access + * @todo Evaluate an alternative interface, in which there is only one managing observer and + * multiple child observers running (by design) on the same thread. + */ + class BreakPointObserver { + public: + /// Constructs the object such that it observes events of the given DebugContext. + BreakPointObserver(std::shared_ptr<DebugContext> debug_context) + : context_weak(debug_context) { + std::unique_lock<std::mutex> lock(debug_context->breakpoint_mutex); + debug_context->breakpoint_observers.push_back(this); + } + + virtual ~BreakPointObserver() { + auto context = context_weak.lock(); + if (context) { + std::unique_lock<std::mutex> lock(context->breakpoint_mutex); + context->breakpoint_observers.remove(this); + + // If we are the last observer to be destroyed, tell the debugger context that + // it is free to continue. In particular, this is required for a proper yuzu + // shutdown, when the emulation thread is waiting at a breakpoint. + if (context->breakpoint_observers.empty()) + context->Resume(); + } + } + + /** + * Action to perform when a breakpoint was reached. + * @param event Type of event which triggered the breakpoint + * @param data Optional data pointer (if unused, this is a nullptr) + * @note This function will perform nothing unless it is overridden in the child class. + */ + virtual void OnMaxwellBreakPointHit(Event event, void* data) {} + + /** + * Action to perform when emulation is resumed from a breakpoint. + * @note This function will perform nothing unless it is overridden in the child class. + */ + virtual void OnMaxwellResume() {} + + protected: + /** + * Weak context pointer. This need not be valid, so when requesting a shared_ptr via + * context_weak.lock(), always compare the result against nullptr. + */ + std::weak_ptr<DebugContext> context_weak; + }; + + /** + * Simple structure defining a breakpoint state + */ + struct BreakPoint { + bool enabled = false; + }; + + /** + * Static constructor used to create a shared_ptr of a DebugContext. + */ + static std::shared_ptr<DebugContext> Construct() { + return std::shared_ptr<DebugContext>(new DebugContext); + } + + /** + * Used by the emulation core when a given event has happened. If a breakpoint has been set + * for this event, OnEvent calls the event handlers of the registered breakpoint observers. + * The current thread then is halted until Resume() is called from another thread (or until + * emulation is stopped). + * @param event Event which has happened + * @param data Optional data pointer (pass nullptr if unused). Needs to remain valid until + * Resume() is called. + */ + void OnEvent(Event event, void* data) { + // This check is left in the header to allow the compiler to inline it. + if (!breakpoints[(int)event].enabled) + return; + // For the rest of event handling, call a separate function. + DoOnEvent(event, data); + } + + void DoOnEvent(Event event, void* data); + + /** + * Resume from the current breakpoint. + * @warning Calling this from the same thread that OnEvent was called in will cause a deadlock. + * Calling from any other thread is safe. + */ + void Resume(); + + /** + * Delete all set breakpoints and resume emulation. + */ + void ClearBreakpoints() { + for (auto& bp : breakpoints) { + bp.enabled = false; + } + Resume(); + } + + // TODO: Evaluate if access to these members should be hidden behind a public interface. + std::array<BreakPoint, (int)Event::NumEvents> breakpoints; + Event active_breakpoint; + bool at_breakpoint = false; + +private: + /** + * Private default constructor to make sure people always construct this through Construct() + * instead. + */ + DebugContext() = default; + + /// Mutex protecting current breakpoint state and the observer list. + std::mutex breakpoint_mutex; + + /// Used by OnEvent to wait for resumption. + std::condition_variable resume_from_breakpoint; + + /// List of registered observers + std::list<BreakPointObserver*> breakpoint_observers; +}; + +} // namespace Tegra diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 9f699399f..124753032 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -2,23 +2,132 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <cinttypes> #include "common/assert.h" +#include "core/core.h" +#include "video_core/debug_utils/debug_utils.h" #include "video_core/engines/maxwell_3d.h" +#include "video_core/rasterizer_interface.h" +#include "video_core/renderer_base.h" +#include "video_core/textures/decoders.h" +#include "video_core/textures/texture.h" +#include "video_core/video_core.h" namespace Tegra { namespace Engines { -Maxwell3D::Maxwell3D(MemoryManager& memory_manager) : memory_manager(memory_manager) {} +/// First register id that is actually a Macro call. +constexpr u32 MacroRegistersStart = 0xE00; -void Maxwell3D::WriteReg(u32 method, u32 value) { +Maxwell3D::Maxwell3D(MemoryManager& memory_manager) + : memory_manager(memory_manager), macro_interpreter(*this) {} + +void Maxwell3D::SubmitMacroCode(u32 entry, std::vector<u32> code) { + uploaded_macros[entry * 2 + MacroRegistersStart] = std::move(code); +} + +void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) { + auto macro_code = uploaded_macros.find(method); + // The requested macro must have been uploaded already. + ASSERT_MSG(macro_code != uploaded_macros.end(), "Macro %08X was not uploaded", method); + + // Reset the current macro and execute it. + executing_macro = 0; + macro_interpreter.Execute(macro_code->second, std::move(parameters)); +} + +void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) { ASSERT_MSG(method < Regs::NUM_REGS, "Invalid Maxwell3D register, increase the size of the Regs structure"); + auto debug_context = Core::System::GetInstance().GetGPUDebugContext(); + + // It is an error to write to a register other than the current macro's ARG register before it + // has finished execution. + if (executing_macro != 0) { + ASSERT(method == executing_macro + 1); + } + + // Methods after 0xE00 are special, they're actually triggers for some microcode that was + // uploaded to the GPU during initialization. + if (method >= MacroRegistersStart) { + // We're trying to execute a macro + if (executing_macro == 0) { + // A macro call must begin by writing the macro method's register, not its argument. + ASSERT_MSG((method % 2) == 0, + "Can't start macro execution by writing to the ARGS register"); + executing_macro = method; + } + + macro_params.push_back(value); + + // Call the macro when there are no more parameters in the command buffer + if (remaining_params == 0) { + CallMacroMethod(executing_macro, std::move(macro_params)); + } + return; + } + + if (debug_context) { + debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandLoaded, nullptr); + } + regs.reg_array[method] = value; #define MAXWELL3D_REG_INDEX(field_name) (offsetof(Regs, field_name) / sizeof(u32)) switch (method) { + case MAXWELL3D_REG_INDEX(code_address.code_address_high): + case MAXWELL3D_REG_INDEX(code_address.code_address_low): { + // Note: For some reason games (like Puyo Puyo Tetris) seem to write 0 to the CODE_ADDRESS + // register, we do not currently know if that's intended or a bug, so we assert it lest + // stuff breaks in other places (like the shader address calculation). + ASSERT_MSG(regs.code_address.CodeAddress() == 0, "Unexpected CODE_ADDRESS register value."); + break; + } + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): { + ProcessCBData(value); + break; + } + case MAXWELL3D_REG_INDEX(cb_bind[0].raw_config): { + ProcessCBBind(Regs::ShaderStage::Vertex); + break; + } + case MAXWELL3D_REG_INDEX(cb_bind[1].raw_config): { + ProcessCBBind(Regs::ShaderStage::TesselationControl); + break; + } + case MAXWELL3D_REG_INDEX(cb_bind[2].raw_config): { + ProcessCBBind(Regs::ShaderStage::TesselationEval); + break; + } + case MAXWELL3D_REG_INDEX(cb_bind[3].raw_config): { + ProcessCBBind(Regs::ShaderStage::Geometry); + break; + } + case MAXWELL3D_REG_INDEX(cb_bind[4].raw_config): { + ProcessCBBind(Regs::ShaderStage::Fragment); + break; + } + case MAXWELL3D_REG_INDEX(draw.vertex_end_gl): { + DrawArrays(); + break; + } case MAXWELL3D_REG_INDEX(query.query_get): { ProcessQueryGet(); break; @@ -28,6 +137,10 @@ void Maxwell3D::WriteReg(u32 method, u32 value) { } #undef MAXWELL3D_REG_INDEX + + if (debug_context) { + debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandProcessed, nullptr); + } } void Maxwell3D::ProcessQueryGet() { @@ -44,8 +157,147 @@ void Maxwell3D::ProcessQueryGet() { break; } default: - UNIMPLEMENTED_MSG("Query mode %u not implemented", regs.query.query_get.mode.Value()); + UNIMPLEMENTED_MSG("Query mode %u not implemented", + static_cast<u32>(regs.query.query_get.mode.Value())); + } +} + +void Maxwell3D::DrawArrays() { + LOG_DEBUG(HW_GPU, "called, topology=%d, count=%d", regs.draw.topology.Value(), + regs.vertex_buffer.count); + + auto debug_context = Core::System::GetInstance().GetGPUDebugContext(); + + if (debug_context) { + debug_context->OnEvent(Tegra::DebugContext::Event::IncomingPrimitiveBatch, nullptr); + } + + if (debug_context) { + debug_context->OnEvent(Tegra::DebugContext::Event::FinishedPrimitiveBatch, nullptr); + } + + VideoCore::g_renderer->Rasterizer()->AccelerateDrawBatch(false /*is_indexed*/); +} + +void Maxwell3D::ProcessCBBind(Regs::ShaderStage stage) { + // Bind the buffer currently in CB_ADDRESS to the specified index in the desired shader stage. + auto& shader = state.shader_stages[static_cast<size_t>(stage)]; + auto& bind_data = regs.cb_bind[static_cast<size_t>(stage)]; + + auto& buffer = shader.const_buffers[bind_data.index]; + + buffer.enabled = bind_data.valid.Value() != 0; + buffer.index = bind_data.index; + buffer.address = regs.const_buffer.BufferAddress(); + buffer.size = regs.const_buffer.cb_size; +} + +void Maxwell3D::ProcessCBData(u32 value) { + // Write the input value to the current const buffer at the current position. + GPUVAddr buffer_address = regs.const_buffer.BufferAddress(); + ASSERT(buffer_address != 0); + + // Don't allow writing past the end of the buffer. + ASSERT(regs.const_buffer.cb_pos + sizeof(u32) <= regs.const_buffer.cb_size); + + VAddr address = + memory_manager.PhysicalToVirtualAddress(buffer_address + regs.const_buffer.cb_pos); + + Memory::Write32(address, value); + + // Increment the current buffer position. + regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4; +} + +Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const { + GPUVAddr tic_base_address = regs.tic.TICAddress(); + + GPUVAddr tic_address_gpu = tic_base_address + tic_index * sizeof(Texture::TICEntry); + VAddr tic_address_cpu = memory_manager.PhysicalToVirtualAddress(tic_address_gpu); + + Texture::TICEntry tic_entry; + Memory::ReadBlock(tic_address_cpu, &tic_entry, sizeof(Texture::TICEntry)); + + ASSERT_MSG(tic_entry.header_version == Texture::TICHeaderVersion::BlockLinear, + "TIC versions other than BlockLinear are unimplemented"); + + ASSERT_MSG(tic_entry.texture_type == Texture::TextureType::Texture2D, + "Texture types other than Texture2D are unimplemented"); + + auto r_type = tic_entry.r_type.Value(); + auto g_type = tic_entry.g_type.Value(); + auto b_type = tic_entry.b_type.Value(); + auto a_type = tic_entry.a_type.Value(); + + // TODO(Subv): Different data types for separate components are not supported + ASSERT(r_type == g_type && r_type == b_type && r_type == a_type); + + return tic_entry; +} + +Texture::TSCEntry Maxwell3D::GetTSCEntry(u32 tsc_index) const { + GPUVAddr tsc_base_address = regs.tsc.TSCAddress(); + + GPUVAddr tsc_address_gpu = tsc_base_address + tsc_index * sizeof(Texture::TSCEntry); + VAddr tsc_address_cpu = memory_manager.PhysicalToVirtualAddress(tsc_address_gpu); + + Texture::TSCEntry tsc_entry; + Memory::ReadBlock(tsc_address_cpu, &tsc_entry, sizeof(Texture::TSCEntry)); + return tsc_entry; +} + +std::vector<Texture::FullTextureInfo> Maxwell3D::GetStageTextures(Regs::ShaderStage stage) const { + std::vector<Texture::FullTextureInfo> textures; + + auto& fragment_shader = state.shader_stages[static_cast<size_t>(stage)]; + auto& tex_info_buffer = fragment_shader.const_buffers[regs.tex_cb_index]; + ASSERT(tex_info_buffer.enabled && tex_info_buffer.address != 0); + + GPUVAddr tic_base_address = regs.tic.TICAddress(); + + GPUVAddr tex_info_buffer_end = tex_info_buffer.address + tex_info_buffer.size; + + // Offset into the texture constbuffer where the texture info begins. + static constexpr size_t TextureInfoOffset = 0x20; + + for (GPUVAddr current_texture = tex_info_buffer.address + TextureInfoOffset; + current_texture < tex_info_buffer_end; current_texture += sizeof(Texture::TextureHandle)) { + + Texture::TextureHandle tex_handle{ + Memory::Read32(memory_manager.PhysicalToVirtualAddress(current_texture))}; + + Texture::FullTextureInfo tex_info{}; + // TODO(Subv): Use the shader to determine which textures are actually accessed. + tex_info.index = (current_texture - tex_info_buffer.address - TextureInfoOffset) / + sizeof(Texture::TextureHandle); + + // Load the TIC data. + if (tex_handle.tic_id != 0) { + tex_info.enabled = true; + + auto tic_entry = GetTICEntry(tex_handle.tic_id); + // TODO(Subv): Workaround for BitField's move constructor being deleted. + std::memcpy(&tex_info.tic, &tic_entry, sizeof(tic_entry)); + } + + // Load the TSC data + if (tex_handle.tsc_id != 0) { + auto tsc_entry = GetTSCEntry(tex_handle.tsc_id); + // TODO(Subv): Workaround for BitField's move constructor being deleted. + std::memcpy(&tex_info.tsc, &tsc_entry, sizeof(tsc_entry)); + } + + if (tex_info.enabled) + textures.push_back(tex_info); } + + return textures; } + +u32 Maxwell3D::GetRegisterValue(u32 method) const { + ASSERT_MSG(method < Regs::NUM_REGS, "Invalid Maxwell3D register"); + return regs.reg_array[method]; +} + } // namespace Engines } // namespace Tegra diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 1eeef6857..98b39b2ff 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -4,10 +4,18 @@ #pragma once +#include <array> +#include <unordered_map> +#include <vector> +#include "common/assert.h" #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" +#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 Tegra { namespace Engines { @@ -17,22 +25,357 @@ public: explicit Maxwell3D(MemoryManager& memory_manager); ~Maxwell3D() = default; - /// Write the value to the register identified by method. - void WriteReg(u32 method, u32 value); - /// Register structure of the Maxwell3D engine. /// TODO(Subv): This structure will need to be made bigger as more registers are discovered. struct Regs { static constexpr size_t NUM_REGS = 0xE36; + static constexpr size_t NumRenderTargets = 8; + static constexpr size_t NumViewports = 16; + static constexpr size_t NumCBData = 16; + static constexpr size_t NumVertexArrays = 32; + static constexpr size_t NumVertexAttributes = 32; + static constexpr size_t MaxShaderProgram = 6; + static constexpr size_t MaxShaderStage = 5; + // Maximum number of const buffers per shader stage. + static constexpr size_t MaxConstBuffers = 16; + enum class QueryMode : u32 { Write = 0, Sync = 1, }; + enum class ShaderProgram : u32 { + VertexA = 0, + VertexB = 1, + TesselationControl = 2, + TesselationEval = 3, + Geometry = 4, + Fragment = 5, + }; + + enum class ShaderStage : u32 { + Vertex = 0, + TesselationControl = 1, + TesselationEval = 2, + Geometry = 3, + Fragment = 4, + }; + + struct VertexAttribute { + enum class Size : u32 { + Size_32_32_32_32 = 0x01, + Size_32_32_32 = 0x02, + Size_16_16_16_16 = 0x03, + Size_32_32 = 0x04, + Size_16_16_16 = 0x05, + Size_8_8_8_8 = 0x0a, + Size_16_16 = 0x0f, + Size_32 = 0x12, + Size_8_8_8 = 0x13, + Size_8_8 = 0x18, + Size_16 = 0x1b, + Size_8 = 0x1d, + Size_10_10_10_2 = 0x30, + Size_11_11_10 = 0x31, + }; + + enum class Type : u32 { + SignedNorm = 1, + UnsignedNorm = 2, + SignedInt = 3, + UnsignedInt = 4, + UnsignedScaled = 5, + SignedScaled = 6, + Float = 7, + }; + + union { + BitField<0, 5, u32> buffer; + BitField<6, 1, u32> constant; + BitField<7, 14, u32> offset; + BitField<21, 6, Size> size; + BitField<27, 3, Type> type; + BitField<31, 1, u32> bgra; + }; + + u32 ComponentCount() const { + switch (size) { + case Size::Size_32_32_32_32: + return 4; + case Size::Size_32_32_32: + return 3; + case Size::Size_16_16_16_16: + return 4; + case Size::Size_32_32: + return 2; + case Size::Size_16_16_16: + return 3; + case Size::Size_8_8_8_8: + return 4; + case Size::Size_16_16: + return 2; + case Size::Size_32: + return 1; + case Size::Size_8_8_8: + return 3; + case Size::Size_8_8: + return 2; + case Size::Size_16: + return 1; + case Size::Size_8: + return 1; + case Size::Size_10_10_10_2: + return 4; + case Size::Size_11_11_10: + return 3; + default: + UNREACHABLE(); + } + } + + u32 SizeInBytes() const { + switch (size) { + case Size::Size_32_32_32_32: + return 16; + case Size::Size_32_32_32: + return 12; + case Size::Size_16_16_16_16: + return 8; + case Size::Size_32_32: + return 8; + case Size::Size_16_16_16: + return 6; + case Size::Size_8_8_8_8: + return 4; + case Size::Size_16_16: + return 4; + case Size::Size_32: + return 4; + case Size::Size_8_8_8: + return 3; + case Size::Size_8_8: + return 2; + case Size::Size_16: + return 2; + case Size::Size_8: + return 1; + case Size::Size_10_10_10_2: + return 4; + case Size::Size_11_11_10: + return 4; + default: + UNREACHABLE(); + } + } + + std::string SizeString() const { + switch (size) { + case Size::Size_32_32_32_32: + return "32_32_32_32"; + case Size::Size_32_32_32: + return "32_32_32"; + case Size::Size_16_16_16_16: + return "16_16_16_16"; + case Size::Size_32_32: + return "32_32"; + case Size::Size_16_16_16: + return "16_16_16"; + case Size::Size_8_8_8_8: + return "8_8_8_8"; + case Size::Size_16_16: + return "16_16"; + case Size::Size_32: + return "32"; + case Size::Size_8_8_8: + return "8_8_8"; + case Size::Size_8_8: + return "8_8"; + case Size::Size_16: + return "16"; + case Size::Size_8: + return "8"; + case Size::Size_10_10_10_2: + return "10_10_10_2"; + case Size::Size_11_11_10: + return "11_11_10"; + } + UNREACHABLE(); + return {}; + } + + std::string TypeString() const { + switch (type) { + case Type::SignedNorm: + return "SNORM"; + case Type::UnsignedNorm: + return "UNORM"; + case Type::SignedInt: + return "SINT"; + case Type::UnsignedInt: + return "UINT"; + case Type::UnsignedScaled: + return "USCALED"; + case Type::SignedScaled: + return "SSCALED"; + case Type::Float: + return "FLOAT"; + } + UNREACHABLE(); + return {}; + } + + bool IsNormalized() const { + return (type == Type::SignedNorm) || (type == Type::UnsignedNorm); + } + }; + + enum class PrimitiveTopology : u32 { + Points = 0x0, + Lines = 0x1, + LineLoop = 0x2, + LineStrip = 0x3, + Triangles = 0x4, + TriangleStrip = 0x5, + TriangleFan = 0x6, + Quads = 0x7, + QuadStrip = 0x8, + Polygon = 0x9, + LinesAdjacency = 0xa, + LineStripAdjacency = 0xb, + TrianglesAdjacency = 0xc, + TriangleStripAdjacency = 0xd, + Patches = 0xe, + }; + union { struct { - INSERT_PADDING_WORDS(0x6C0); + INSERT_PADDING_WORDS(0x200); + + struct { + u32 address_high; + u32 address_low; + u32 width; + u32 height; + Tegra::RenderTargetFormat format; + u32 block_dimensions; + u32 array_mode; + u32 layer_stride; + u32 base_layer; + INSERT_PADDING_WORDS(7); + + GPUVAddr Address() const { + return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | + address_low); + } + } rt[NumRenderTargets]; + + INSERT_PADDING_WORDS(0x80); + + struct { + union { + BitField<0, 16, u32> x; + BitField<16, 16, u32> width; + }; + union { + BitField<0, 16, u32> y; + BitField<16, 16, u32> height; + }; + float depth_range_near; + float depth_range_far; + + MathUtil::Rectangle<s32> GetRect() const { + return { + static_cast<s32>(x), // left + static_cast<s32>(y + height), // top + static_cast<s32>(x + width), // right + static_cast<s32>(y) // bottom + }; + }; + } viewport[NumViewports]; + + INSERT_PADDING_WORDS(0x1D); + + struct { + u32 first; + u32 count; + } vertex_buffer; + + INSERT_PADDING_WORDS(0x99); + + struct { + u32 address_high; + u32 address_low; + u32 format; + u32 block_dimensions; + u32 layer_stride; + + GPUVAddr Address() const { + return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | + address_low); + } + } zeta; + + INSERT_PADDING_WORDS(0x5B); + + VertexAttribute vertex_attrib_format[NumVertexAttributes]; + + INSERT_PADDING_WORDS(0xF); + + struct { + union { + BitField<0, 4, u32> count; + }; + } rt_control; + + INSERT_PADDING_WORDS(0xCF); + + struct { + u32 tsc_address_high; + u32 tsc_address_low; + u32 tsc_limit; + + GPUVAddr TSCAddress() const { + return static_cast<GPUVAddr>( + (static_cast<GPUVAddr>(tsc_address_high) << 32) | tsc_address_low); + } + } tsc; + + INSERT_PADDING_WORDS(0x3); + + struct { + u32 tic_address_high; + u32 tic_address_low; + u32 tic_limit; + + GPUVAddr TICAddress() const { + return static_cast<GPUVAddr>( + (static_cast<GPUVAddr>(tic_address_high) << 32) | tic_address_low); + } + } tic; + + INSERT_PADDING_WORDS(0x22); + + struct { + u32 code_address_high; + u32 code_address_low; + + GPUVAddr CodeAddress() const { + return static_cast<GPUVAddr>( + (static_cast<GPUVAddr>(code_address_high) << 32) | code_address_low); + } + } code_address; + INSERT_PADDING_WORDS(1); + + struct { + u32 vertex_end_gl; + union { + u32 vertex_begin_gl; + BitField<0, 16, PrimitiveTopology> topology; + }; + } draw; + + INSERT_PADDING_WORDS(0x139); struct { u32 query_address_high; u32 query_address_low; @@ -49,7 +392,98 @@ public: (static_cast<GPUVAddr>(query_address_high) << 32) | query_address_low); } } query; - INSERT_PADDING_WORDS(0x772); + + INSERT_PADDING_WORDS(0x3C); + + struct { + union { + BitField<0, 12, u32> stride; + BitField<12, 1, u32> enable; + }; + u32 start_high; + u32 start_low; + u32 divisor; + + GPUVAddr StartAddress() const { + return static_cast<GPUVAddr>((static_cast<GPUVAddr>(start_high) << 32) | + start_low); + } + } vertex_array[NumVertexArrays]; + + INSERT_PADDING_WORDS(0x40); + + struct { + u32 limit_high; + u32 limit_low; + + GPUVAddr LimitAddress() const { + return static_cast<GPUVAddr>((static_cast<GPUVAddr>(limit_high) << 32) | + limit_low); + } + } vertex_array_limit[NumVertexArrays]; + + struct { + union { + BitField<0, 1, u32> enable; + BitField<4, 4, ShaderProgram> program; + }; + u32 start_id; + INSERT_PADDING_WORDS(1); + u32 gpr_alloc; + ShaderStage type; + INSERT_PADDING_WORDS(9); + } shader_config[MaxShaderProgram]; + + INSERT_PADDING_WORDS(0x8C); + + struct { + u32 cb_size; + u32 cb_address_high; + u32 cb_address_low; + u32 cb_pos; + u32 cb_data[NumCBData]; + + GPUVAddr BufferAddress() const { + return static_cast<GPUVAddr>( + (static_cast<GPUVAddr>(cb_address_high) << 32) | cb_address_low); + } + } const_buffer; + + INSERT_PADDING_WORDS(0x10); + + struct { + union { + u32 raw_config; + BitField<0, 1, u32> valid; + BitField<4, 5, u32> index; + }; + INSERT_PADDING_WORDS(7); + } cb_bind[MaxShaderStage]; + + INSERT_PADDING_WORDS(0x56); + + u32 tex_cb_index; + + INSERT_PADDING_WORDS(0x395); + + struct { + /// Compressed address of a buffer that holds information about bound SSBOs. + /// This address is usually bound to c0 in the shaders. + u32 buffer_address; + + GPUVAddr BufferAddress() const { + return static_cast<GPUVAddr>(buffer_address) << 8; + } + } ssbo_info; + + INSERT_PADDING_WORDS(0x11); + + struct { + u32 address[MaxShaderStage]; + u32 size[MaxShaderStage]; + } tex_info_buffers; + + INSERT_PADDING_WORDS(0x102); }; std::array<u32, NUM_REGS> reg_array; }; @@ -57,18 +491,98 @@ public: static_assert(sizeof(Regs) == Regs::NUM_REGS * sizeof(u32), "Maxwell3D Regs has wrong size"); + struct State { + struct ConstBufferInfo { + GPUVAddr address; + u32 index; + u32 size; + bool enabled; + }; + + struct ShaderStageInfo { + std::array<ConstBufferInfo, Regs::MaxConstBuffers> const_buffers; + }; + + std::array<ShaderStageInfo, Regs::MaxShaderStage> shader_stages; + }; + + State state{}; + + /// Reads a register value located at the input method address + u32 GetRegisterValue(u32 method) const; + + /// Write the value to the register identified by method. + void WriteReg(u32 method, u32 value, u32 remaining_params); + + /// Uploads the code for a GPU macro program associated with the specified entry. + void SubmitMacroCode(u32 entry, std::vector<u32> code); + + /// Returns a list of enabled textures for the specified shader stage. + std::vector<Texture::FullTextureInfo> GetStageTextures(Regs::ShaderStage stage) const; + private: + MemoryManager& memory_manager; + + std::unordered_map<u32, std::vector<u32>> uploaded_macros; + + /// Macro method that is currently being executed / being fed parameters. + u32 executing_macro = 0; + /// Parameters that have been submitted to the macro call so far. + std::vector<u32> macro_params; + + /// Interpreter for the macro codes uploaded to the GPU. + MacroInterpreter macro_interpreter; + + /// Retrieves information about a specific TIC entry from the TIC buffer. + Texture::TICEntry GetTICEntry(u32 tic_index) const; + + /// Retrieves information about a specific TSC entry from the TSC buffer. + Texture::TSCEntry GetTSCEntry(u32 tsc_index) const; + + /** + * Call a macro on this engine. + * @param method Method to call + * @param parameters Arguments to the method call + */ + void CallMacroMethod(u32 method, std::vector<u32> parameters); + /// Handles a write to the QUERY_GET register. void ProcessQueryGet(); - MemoryManager& memory_manager; + /// Handles a write to the CB_DATA[i] register. + void ProcessCBData(u32 value); + + /// Handles a write to the CB_BIND register. + void ProcessCBBind(Regs::ShaderStage stage); + + /// Handles a write to the VERTEX_END_GL register, triggering a draw. + void DrawArrays(); }; #define ASSERT_REG_POSITION(field_name, position) \ static_assert(offsetof(Maxwell3D::Regs, field_name) == position * 4, \ "Field " #field_name " has invalid position") +ASSERT_REG_POSITION(rt, 0x200); +ASSERT_REG_POSITION(viewport, 0x300); +ASSERT_REG_POSITION(vertex_buffer, 0x35D); +ASSERT_REG_POSITION(zeta, 0x3F8); +ASSERT_REG_POSITION(vertex_attrib_format[0], 0x458); +ASSERT_REG_POSITION(rt_control, 0x487); +ASSERT_REG_POSITION(tsc, 0x557); +ASSERT_REG_POSITION(tic, 0x55D); +ASSERT_REG_POSITION(code_address, 0x582); +ASSERT_REG_POSITION(draw, 0x585); ASSERT_REG_POSITION(query, 0x6C0); +ASSERT_REG_POSITION(vertex_array[0], 0x700); +ASSERT_REG_POSITION(vertex_array_limit[0], 0x7C0); +ASSERT_REG_POSITION(shader_config[0], 0x800); +ASSERT_REG_POSITION(const_buffer, 0x8E0); +ASSERT_REG_POSITION(cb_bind[0], 0x904); +ASSERT_REG_POSITION(tex_cb_index, 0x982); +ASSERT_REG_POSITION(ssbo_info, 0xD18); +ASSERT_REG_POSITION(tex_info_buffers.address[0], 0xD2A); +ASSERT_REG_POSITION(tex_info_buffers.size[0], 0xD2F); #undef ASSERT_REG_POSITION diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp new file mode 100644 index 000000000..9463cd5d6 --- /dev/null +++ b/src/video_core/gpu.cpp @@ -0,0 +1,25 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "video_core/engines/fermi_2d.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/engines/maxwell_compute.h" +#include "video_core/gpu.h" + +namespace Tegra { + +GPU::GPU() { + memory_manager = std::make_unique<MemoryManager>(); + maxwell_3d = std::make_unique<Engines::Maxwell3D>(*memory_manager); + fermi_2d = std::make_unique<Engines::Fermi2D>(); + maxwell_compute = std::make_unique<Engines::MaxwellCompute>(); +} + +GPU::~GPU() = default; + +const Tegra::Engines::Maxwell3D& GPU::Get3DEngine() const { + return *maxwell_3d; +} + +} // namespace Tegra diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index ba7781756..71a8661b4 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -6,14 +6,57 @@ #include <memory> #include <unordered_map> +#include <vector> #include "common/common_types.h" -#include "video_core/engines/fermi_2d.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/maxwell_compute.h" +#include "core/hle/service/nvflinger/buffer_queue.h" #include "video_core/memory_manager.h" namespace Tegra { +enum class RenderTargetFormat : u32 { + NONE = 0x0, + RGBA8_UNORM = 0xD5, +}; + +class DebugContext; + +/** + * Struct describing framebuffer configuration + */ +struct FramebufferConfig { + enum class PixelFormat : u32 { + ABGR8 = 1, + }; + + /** + * Returns the number of bytes per pixel. + */ + static u32 BytesPerPixel(PixelFormat format) { + switch (format) { + case PixelFormat::ABGR8: + return 4; + } + + UNREACHABLE(); + } + + VAddr address; + u32 offset; + u32 width; + u32 height; + u32 stride; + PixelFormat pixel_format; + + using TransformFlags = Service::NVFlinger::BufferQueue::BufferTransformFlags; + TransformFlags transform_flags; +}; + +namespace Engines { +class Fermi2D; +class Maxwell3D; +class MaxwellCompute; +} // namespace Engines + enum class EngineID { FERMI_TWOD_A = 0x902D, // 2D Engine MAXWELL_B = 0xB197, // 3D Engine @@ -24,22 +67,26 @@ enum class EngineID { class GPU final { public: - GPU() { - memory_manager = std::make_unique<MemoryManager>(); - maxwell_3d = std::make_unique<Engines::Maxwell3D>(*memory_manager); - fermi_2d = std::make_unique<Engines::Fermi2D>(); - maxwell_compute = std::make_unique<Engines::MaxwellCompute>(); - } - ~GPU() = default; + GPU(); + ~GPU(); /// Processes a command list stored at the specified address in GPU memory. void ProcessCommandList(GPUVAddr address, u32 size); + /// Returns a reference to the Maxwell3D GPU engine. + const Engines::Maxwell3D& Get3DEngine() const; + std::unique_ptr<MemoryManager> memory_manager; + Engines::Maxwell3D& Maxwell3D() { + return *maxwell_3d; + } + private: + static constexpr u32 InvalidGraphMacroEntry = 0xFFFFFFFF; + /// Writes a single register in the engine bound to the specified subchannel - void WriteReg(u32 method, u32 subchannel, u32 value); + void WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params); /// Mapping of command subchannels to their bound engine ids. std::unordered_map<u32, EngineID> bound_engines; @@ -50,6 +97,11 @@ private: std::unique_ptr<Engines::Fermi2D> fermi_2d; /// Compute engine std::unique_ptr<Engines::MaxwellCompute> maxwell_compute; + + /// Entry of the macro that is currently being uploaded + u32 current_macro_entry = InvalidGraphMacroEntry; + /// Code being uploaded for the current macro + std::vector<u32> current_macro_code; }; } // namespace Tegra diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro_interpreter.cpp new file mode 100644 index 000000000..993a67746 --- /dev/null +++ b/src/video_core/macro_interpreter.cpp @@ -0,0 +1,257 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/assert.h" +#include "common/logging/log.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/macro_interpreter.h" + +namespace Tegra { + +MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d) : maxwell3d(maxwell3d) {} + +void MacroInterpreter::Execute(const std::vector<u32>& code, std::vector<u32> parameters) { + Reset(); + registers[1] = parameters[0]; + this->parameters = std::move(parameters); + + // Execute the code until we hit an exit condition. + bool keep_executing = true; + while (keep_executing) { + keep_executing = Step(code, false); + } + + // Assert the the macro used all the input parameters + ASSERT(next_parameter_index == this->parameters.size()); +} + +void MacroInterpreter::Reset() { + registers = {}; + pc = 0; + delayed_pc = boost::none; + method_address.raw = 0; + parameters.clear(); + // The next parameter index starts at 1, because $r1 already has the value of the first + // parameter. + next_parameter_index = 1; +} + +bool MacroInterpreter::Step(const std::vector<u32>& code, bool is_delay_slot) { + u32 base_address = pc; + + Opcode opcode = GetOpcode(code); + pc += 4; + + // Update the program counter if we were delayed + if (delayed_pc != boost::none) { + ASSERT(is_delay_slot); + pc = *delayed_pc; + delayed_pc = boost::none; + } + + switch (opcode.operation) { + case Operation::ALU: { + u32 result = GetALUResult(opcode.alu_operation, GetRegister(opcode.src_a), + GetRegister(opcode.src_b)); + ProcessResult(opcode.result_operation, opcode.dst, result); + break; + } + case Operation::AddImmediate: { + ProcessResult(opcode.result_operation, opcode.dst, + GetRegister(opcode.src_a) + opcode.immediate); + break; + } + case Operation::ExtractInsert: { + u32 dst = GetRegister(opcode.src_a); + u32 src = GetRegister(opcode.src_b); + + src = (src >> opcode.bf_src_bit) & opcode.GetBitfieldMask(); + dst &= ~(opcode.GetBitfieldMask() << opcode.bf_dst_bit); + dst |= src << opcode.bf_dst_bit; + ProcessResult(opcode.result_operation, opcode.dst, dst); + break; + } + case Operation::ExtractShiftLeftImmediate: { + u32 dst = GetRegister(opcode.src_a); + u32 src = GetRegister(opcode.src_b); + + u32 result = ((src >> dst) & opcode.GetBitfieldMask()) << opcode.bf_dst_bit; + + ProcessResult(opcode.result_operation, opcode.dst, result); + break; + } + case Operation::ExtractShiftLeftRegister: { + u32 dst = GetRegister(opcode.src_a); + u32 src = GetRegister(opcode.src_b); + + u32 result = ((src >> opcode.bf_src_bit) & opcode.GetBitfieldMask()) << dst; + + ProcessResult(opcode.result_operation, opcode.dst, result); + break; + } + case Operation::Read: { + u32 result = Read(GetRegister(opcode.src_a) + opcode.immediate); + ProcessResult(opcode.result_operation, opcode.dst, result); + break; + } + case Operation::Branch: { + ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid"); + u32 value = GetRegister(opcode.src_a); + bool taken = EvaluateBranchCondition(opcode.branch_condition, value); + if (taken) { + // Ignore the delay slot if the branch has the annul bit. + if (opcode.branch_annul) { + pc = base_address + (opcode.immediate << 2); + return true; + } + + delayed_pc = base_address + (opcode.immediate << 2); + // Execute one more instruction due to the delay slot. + return Step(code, true); + } + break; + } + default: + UNIMPLEMENTED_MSG("Unimplemented macro operation %u", + static_cast<u32>(opcode.operation.Value())); + } + + if (opcode.is_exit) { + // Exit has a delay slot, execute the next instruction + // Note: Executing an exit during a branch delay slot will cause the instruction at the + // branch target to be executed before exiting. + Step(code, true); + return false; + } + + return true; +} + +MacroInterpreter::Opcode MacroInterpreter::GetOpcode(const std::vector<u32>& code) const { + ASSERT((pc % sizeof(u32)) == 0); + ASSERT(pc < code.size() * sizeof(u32)); + return {code[pc / sizeof(u32)]}; +} + +u32 MacroInterpreter::GetALUResult(ALUOperation operation, u32 src_a, u32 src_b) const { + switch (operation) { + case ALUOperation::Add: + return src_a + src_b; + // TODO(Subv): Implement AddWithCarry + case ALUOperation::Subtract: + return src_a - src_b; + // TODO(Subv): Implement SubtractWithBorrow + case ALUOperation::Xor: + return src_a ^ src_b; + case ALUOperation::Or: + return src_a | src_b; + case ALUOperation::And: + return src_a & src_b; + case ALUOperation::AndNot: + return src_a & ~src_b; + case ALUOperation::Nand: + return ~(src_a & src_b); + + default: + UNIMPLEMENTED_MSG("Unimplemented ALU operation %u", static_cast<u32>(operation)); + } +} + +void MacroInterpreter::ProcessResult(ResultOperation operation, u32 reg, u32 result) { + switch (operation) { + case ResultOperation::IgnoreAndFetch: + // Fetch parameter and ignore result. + SetRegister(reg, FetchParameter()); + break; + case ResultOperation::Move: + // Move result. + SetRegister(reg, result); + break; + case ResultOperation::MoveAndSetMethod: + // Move result and use as Method Address. + SetRegister(reg, result); + SetMethodAddress(result); + break; + case ResultOperation::FetchAndSend: + // Fetch parameter and send result. + SetRegister(reg, FetchParameter()); + Send(result); + break; + case ResultOperation::MoveAndSend: + // Move and send result. + SetRegister(reg, result); + Send(result); + break; + case ResultOperation::FetchAndSetMethod: + // Fetch parameter and use result as Method Address. + SetRegister(reg, FetchParameter()); + SetMethodAddress(result); + break; + case ResultOperation::MoveAndSetMethodFetchAndSend: + // Move result and use as Method Address, then fetch and send parameter. + SetRegister(reg, result); + SetMethodAddress(result); + Send(FetchParameter()); + break; + case ResultOperation::MoveAndSetMethodSend: + // Move result and use as Method Address, then send bits 12:17 of result. + SetRegister(reg, result); + SetMethodAddress(result); + Send((result >> 12) & 0b111111); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented result operation %u", static_cast<u32>(operation)); + } +} + +u32 MacroInterpreter::FetchParameter() { + ASSERT(next_parameter_index < parameters.size()); + return parameters[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]; +} + +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) + return; + + ASSERT(register_id < registers.size()); + registers[register_id] = value; +} + +void MacroInterpreter::SetMethodAddress(u32 address) { + method_address.raw = address; +} + +void MacroInterpreter::Send(u32 value) { + maxwell3d.WriteReg(method_address.address, value, 0); + // Increment the method address by the method increment. + method_address.address.Assign(method_address.address.Value() + + method_address.increment.Value()); +} + +u32 MacroInterpreter::Read(u32 method) const { + return maxwell3d.GetRegisterValue(method); +} + +bool MacroInterpreter::EvaluateBranchCondition(BranchCondition cond, u32 value) const { + switch (cond) { + case BranchCondition::Zero: + return value == 0; + case BranchCondition::NotZero: + return value != 0; + } + UNREACHABLE(); +} + +} // namespace Tegra diff --git a/src/video_core/macro_interpreter.h b/src/video_core/macro_interpreter.h new file mode 100644 index 000000000..a71e359d8 --- /dev/null +++ b/src/video_core/macro_interpreter.h @@ -0,0 +1,164 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <vector> +#include <boost/optional.hpp> +#include "common/bit_field.h" +#include "common/common_types.h" + +namespace Tegra { +namespace Engines { +class Maxwell3D; +} + +class MacroInterpreter final { +public: + explicit MacroInterpreter(Engines::Maxwell3D& maxwell3d); + + /** + * Executes the macro code with the specified input parameters. + * @param code The macro byte code to execute + * @param parameters The parameters of the macro + */ + void Execute(const std::vector<u32>& code, std::vector<u32> parameters); + +private: + enum class Operation : u32 { + ALU = 0, + AddImmediate = 1, + ExtractInsert = 2, + ExtractShiftLeftImmediate = 3, + ExtractShiftLeftRegister = 4, + Read = 5, + Unused = 6, // This operation doesn't seem to be a valid encoding. + Branch = 7, + }; + + enum class ALUOperation : u32 { + Add = 0, + AddWithCarry = 1, + Subtract = 2, + SubtractWithBorrow = 3, + // Operations 4-7 don't seem to be valid encodings. + Xor = 8, + Or = 9, + And = 10, + AndNot = 11, + Nand = 12 + }; + + enum class ResultOperation : u32 { + IgnoreAndFetch = 0, + Move = 1, + MoveAndSetMethod = 2, + FetchAndSend = 3, + MoveAndSend = 4, + FetchAndSetMethod = 5, + MoveAndSetMethodFetchAndSend = 6, + MoveAndSetMethodSend = 7 + }; + + enum class BranchCondition : u32 { + Zero = 0, + NotZero = 1, + }; + + union Opcode { + u32 raw; + BitField<0, 3, Operation> operation; + BitField<4, 3, ResultOperation> result_operation; + BitField<4, 1, BranchCondition> branch_condition; + BitField<5, 1, u32> + branch_annul; // If set on a branch, then the branch doesn't have a delay slot. + BitField<7, 1, u32> is_exit; + BitField<8, 3, u32> dst; + BitField<11, 3, u32> src_a; + BitField<14, 3, u32> src_b; + // The signed immediate overlaps the second source operand and the alu operation. + BitField<14, 18, s32> immediate; + + BitField<17, 5, ALUOperation> alu_operation; + + // Bitfield instructions data + BitField<17, 5, u32> bf_src_bit; + BitField<22, 5, u32> bf_size; + BitField<27, 5, u32> bf_dst_bit; + + u32 GetBitfieldMask() const { + return (1 << bf_size) - 1; + } + }; + + union MethodAddress { + u32 raw; + BitField<0, 12, u32> address; + BitField<12, 6, u32> increment; + }; + + /// Resets the execution engine state, zeroing registers, etc. + void Reset(); + + /** + * Executes a single macro instruction located at the current program counter. Returns whether + * the interpreter should keep running. + * @param code The macro code to execute. + * @param is_delay_slot Whether the current step is being executed due to a delay slot in a + * previous instruction. + */ + bool Step(const std::vector<u32>& code, bool is_delay_slot); + + /// Calculates the result of an ALU operation. src_a OP src_b; + u32 GetALUResult(ALUOperation operation, u32 src_a, u32 src_b) const; + + /// Performs the result operation on the input result and stores it in the specified register + /// (if necessary). + void ProcessResult(ResultOperation operation, u32 reg, u32 result); + + /// Evaluates the branch condition and returns whether the branch should be taken or not. + bool EvaluateBranchCondition(BranchCondition cond, u32 value) const; + + /// Reads an opcode at the current program counter location. + Opcode GetOpcode(const std::vector<u32>& code) const; + + /// Returns the specified register's value. Register 0 is hardcoded to always return 0. + u32 GetRegister(u32 register_id) const; + + /// Sets the register to the input value. + void SetRegister(u32 register_id, u32 value); + + /// Sets the method address to use for the next Send instruction. + void SetMethodAddress(u32 address); + + /// Calls a GPU Engine method with the input parameter. + void Send(u32 value); + + /// Reads a GPU register located at the method address. + u32 Read(u32 method) const; + + /// Returns the next parameter in the parameter queue. + u32 FetchParameter(); + + Engines::Maxwell3D& maxwell3d; + + u32 pc; ///< Current program counter + boost::optional<u32> + delayed_pc; ///< Program counter to execute at after the delay slot is executed. + + static constexpr size_t NumMacroRegisters = 8; + + /// General purpose macro registers. + std::array<u32, NumMacroRegisters> registers = {}; + + /// Method address to use for the next Send instruction. + MethodAddress method_address = {}; + + /// Input parameters of the current macro. + std::vector<u32> parameters; + /// Index of the next parameter that will be fetched by the 'parm' instruction. + u32 next_parameter_index = 0; +}; +} // namespace Tegra diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h new file mode 100644 index 000000000..8239f9aad --- /dev/null +++ b/src/video_core/rasterizer_interface.h @@ -0,0 +1,63 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include "common/common_types.h" +#include "video_core/gpu.h" + +struct ScreenInfo; + +namespace VideoCore { + +class RasterizerInterface { +public: + virtual ~RasterizerInterface() {} + + /// Draw the current batch of vertex arrays + virtual void DrawArrays() = 0; + + /// Notify rasterizer that the specified Maxwell register has been changed + virtual void NotifyMaxwellRegisterChanged(u32 id) = 0; + + /// Notify rasterizer that all caches should be flushed to 3DS memory + virtual void FlushAll() = 0; + + /// Notify rasterizer that any caches of the specified region should be flushed to 3DS memory + virtual void FlushRegion(VAddr addr, u64 size) = 0; + + /// Notify rasterizer that any caches of the specified region should be invalidated + virtual void InvalidateRegion(VAddr addr, u64 size) = 0; + + /// Notify rasterizer that any caches of the specified region should be flushed to 3DS memory + /// and invalidated + virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0; + + /// Attempt to use a faster method to perform a display transfer with is_texture_copy = 0 + virtual bool AccelerateDisplayTransfer(const void* config) { + return false; + } + + /// Attempt to use a faster method to perform a display transfer with is_texture_copy = 1 + virtual bool AccelerateTextureCopy(const void* config) { + return false; + } + + /// Attempt to use a faster method to fill a region + virtual bool AccelerateFill(const void* config) { + return false; + } + + /// Attempt to use a faster method to display the framebuffer to screen + virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& framebuffer, + VAddr framebuffer_addr, u32 pixel_stride, + ScreenInfo& screen_info) { + return false; + } + + virtual bool AccelerateDrawBatch(bool is_indexed) { + return false; + } +}; +} // namespace VideoCore diff --git a/src/video_core/renderer_base.cpp b/src/video_core/renderer_base.cpp index 51e1d45f9..30075b23c 100644 --- a/src/video_core/renderer_base.cpp +++ b/src/video_core/renderer_base.cpp @@ -5,6 +5,11 @@ #include <atomic> #include <memory> #include "video_core/renderer_base.h" +#include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/video_core.h" -void RendererBase::RefreshRasterizerSetting() {} +void RendererBase::RefreshRasterizerSetting() { + if (rasterizer == nullptr) { + rasterizer = std::make_unique<RasterizerOpenGL>(); + } +} diff --git a/src/video_core/renderer_base.h b/src/video_core/renderer_base.h index 2aba50eda..89a960eaf 100644 --- a/src/video_core/renderer_base.h +++ b/src/video_core/renderer_base.h @@ -8,6 +8,8 @@ #include <boost/optional.hpp> #include "common/assert.h" #include "common/common_types.h" +#include "video_core/gpu.h" +#include "video_core/rasterizer_interface.h" class EmuWindow; @@ -16,40 +18,10 @@ public: /// Used to reference a framebuffer enum kFramebuffer { kFramebuffer_VirtualXFB = 0, kFramebuffer_EFB, kFramebuffer_Texture }; - /** - * Struct describing framebuffer metadata - * TODO(bunnei): This struct belongs in the GPU code, but we don't have a good place for it yet. - */ - struct FramebufferInfo { - enum class PixelFormat : u32 { - ABGR8 = 1, - }; - - /** - * Returns the number of bytes per pixel. - */ - static u32 BytesPerPixel(PixelFormat format) { - switch (format) { - case PixelFormat::ABGR8: - return 4; - } - - UNREACHABLE(); - } - - VAddr address; - u32 offset; - u32 width; - u32 height; - u32 stride; - PixelFormat pixel_format; - bool flip_vertical; - }; - virtual ~RendererBase() {} /// Swap buffers (render frame) - virtual void SwapBuffers(boost::optional<const FramebufferInfo&> framebuffer_info) = 0; + virtual void SwapBuffers(boost::optional<const Tegra::FramebufferConfig&> framebuffer) = 0; /** * Set the emulator window to use for renderer @@ -74,12 +46,16 @@ public: return m_current_frame; } + VideoCore::RasterizerInterface* Rasterizer() const { + return rasterizer.get(); + } + void RefreshRasterizerSetting(); protected: + std::unique_ptr<VideoCore::RasterizerInterface> rasterizer; f32 m_current_fps = 0.0f; ///< Current framerate, should be set by the renderer int m_current_frame = 0; ///< Current frame, should be set by the renderer private: - bool opengl_rasterizer_active = false; }; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp new file mode 100644 index 000000000..911890f16 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -0,0 +1,578 @@ +// Copyright 2015 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <memory> +#include <string> +#include <tuple> +#include <utility> +#include <glad/glad.h> +#include "common/alignment.h" +#include "common/assert.h" +#include "common/logging/log.h" +#include "common/math_util.h" +#include "common/microprofile.h" +#include "common/scope_exit.h" +#include "common/vector_math.h" +#include "core/core.h" +#include "core/hle/kernel/process.h" +#include "core/settings.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/renderer_opengl/gl_rasterizer.h" +#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" + +using Maxwell = Tegra::Engines::Maxwell3D::Regs; +using PixelFormat = SurfaceParams::PixelFormat; +using SurfaceType = SurfaceParams::SurfaceType; + +MICROPROFILE_DEFINE(OpenGL_VAO, "OpenGL", "Vertex Array Setup", MP_RGB(128, 128, 192)); +MICROPROFILE_DEFINE(OpenGL_VS, "OpenGL", "Vertex Shader Setup", MP_RGB(128, 128, 192)); +MICROPROFILE_DEFINE(OpenGL_FS, "OpenGL", "Fragment Shader Setup", MP_RGB(128, 128, 192)); +MICROPROFILE_DEFINE(OpenGL_Drawing, "OpenGL", "Drawing", MP_RGB(128, 128, 192)); +MICROPROFILE_DEFINE(OpenGL_Blits, "OpenGL", "Blits", MP_RGB(100, 100, 255)); +MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Mgmt", MP_RGB(100, 255, 100)); + +enum class UniformBindings : GLuint { Common, VS, FS }; + +static void SetShaderUniformBlockBinding(GLuint shader, const char* name, UniformBindings binding, + size_t expected_size) { + GLuint ub_index = glGetUniformBlockIndex(shader, name); + if (ub_index != GL_INVALID_INDEX) { + GLint ub_size = 0; + glGetActiveUniformBlockiv(shader, ub_index, GL_UNIFORM_BLOCK_DATA_SIZE, &ub_size); + ASSERT_MSG(ub_size == expected_size, + "Uniform block size did not match! Got %d, expected %zu", + static_cast<int>(ub_size), expected_size); + glUniformBlockBinding(shader, ub_index, static_cast<GLuint>(binding)); + } +} + +static void SetShaderUniformBlockBindings(GLuint shader) { + SetShaderUniformBlockBinding(shader, "shader_data", UniformBindings::Common, + sizeof(RasterizerOpenGL::UniformData)); + SetShaderUniformBlockBinding(shader, "vs_config", UniformBindings::VS, + sizeof(RasterizerOpenGL::VSUniformData)); + SetShaderUniformBlockBinding(shader, "fs_config", UniformBindings::FS, + sizeof(RasterizerOpenGL::FSUniformData)); +} + +RasterizerOpenGL::RasterizerOpenGL() { + shader_dirty = true; + + has_ARB_buffer_storage = false; + has_ARB_direct_state_access = false; + has_ARB_separate_shader_objects = false; + has_ARB_vertex_attrib_binding = false; + + GLint ext_num; + glGetIntegerv(GL_NUM_EXTENSIONS, &ext_num); + for (GLint i = 0; i < ext_num; i++) { + std::string extension{reinterpret_cast<const char*>(glGetStringi(GL_EXTENSIONS, i))}; + + if (extension == "GL_ARB_buffer_storage") { + has_ARB_buffer_storage = true; + } else if (extension == "GL_ARB_direct_state_access") { + has_ARB_direct_state_access = true; + } else if (extension == "GL_ARB_separate_shader_objects") { + has_ARB_separate_shader_objects = true; + } else if (extension == "GL_ARB_vertex_attrib_binding") { + has_ARB_vertex_attrib_binding = true; + } + } + + // Clipping plane 0 is always enabled for PICA fixed clip plane z <= 0 + state.clip_distance[0] = true; + + // Generate VBO, VAO and UBO + vertex_buffer = OGLStreamBuffer::MakeBuffer(GLAD_GL_ARB_buffer_storage, GL_ARRAY_BUFFER); + vertex_buffer->Create(VERTEX_BUFFER_SIZE, VERTEX_BUFFER_SIZE / 2); + sw_vao.Create(); + uniform_buffer.Create(); + + state.draw.vertex_array = sw_vao.handle; + state.draw.vertex_buffer = vertex_buffer->GetHandle(); + state.draw.uniform_buffer = uniform_buffer.handle; + state.Apply(); + + glBufferData(GL_UNIFORM_BUFFER, sizeof(UniformData), nullptr, GL_STATIC_DRAW); + glBindBufferBase(GL_UNIFORM_BUFFER, 0, uniform_buffer.handle); + + uniform_block_data.dirty = true; + + // Create render framebuffer + framebuffer.Create(); + + if (has_ARB_separate_shader_objects) { + hw_vao.Create(); + hw_vao_enabled_attributes.fill(false); + + stream_buffer = OGLStreamBuffer::MakeBuffer(has_ARB_buffer_storage, GL_ARRAY_BUFFER); + stream_buffer->Create(STREAM_BUFFER_SIZE, STREAM_BUFFER_SIZE / 2); + state.draw.vertex_buffer = stream_buffer->GetHandle(); + + pipeline.Create(); + state.draw.program_pipeline = pipeline.handle; + state.draw.shader_program = 0; + state.draw.vertex_array = hw_vao.handle; + state.Apply(); + + glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, stream_buffer->GetHandle()); + + vs_uniform_buffer.Create(); + glBindBuffer(GL_UNIFORM_BUFFER, vs_uniform_buffer.handle); + glBufferData(GL_UNIFORM_BUFFER, sizeof(VSUniformData), nullptr, GL_STREAM_COPY); + glBindBufferBase(GL_UNIFORM_BUFFER, 1, vs_uniform_buffer.handle); + } else { + UNREACHABLE(); + } + + accelerate_draw = AccelDraw::Disabled; + + glEnable(GL_BLEND); + + LOG_CRITICAL(Render_OpenGL, "Sync fixed function OpenGL state here!"); +} + +RasterizerOpenGL::~RasterizerOpenGL() { + if (stream_buffer != nullptr) { + state.draw.vertex_buffer = stream_buffer->GetHandle(); + state.Apply(); + stream_buffer->Release(); + } +} + +void RasterizerOpenGL::AnalyzeVertexArray(bool is_indexed) { + const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs; + + if (is_indexed) { + UNREACHABLE(); + } + + // TODO(bunnei): Add support for 1+ vertex arrays + vs_input_size = regs.vertex_buffer.count * regs.vertex_array[0].stride; +} + +void RasterizerOpenGL::SetupVertexArray(u8* array_ptr, GLintptr buffer_offset) { + MICROPROFILE_SCOPE(OpenGL_VAO); + const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs; + const auto& memory_manager = Core::System().GetInstance().GPU().memory_manager; + + state.draw.vertex_array = hw_vao.handle; + state.draw.vertex_buffer = stream_buffer->GetHandle(); + state.Apply(); + + // TODO(bunnei): Add support for 1+ vertex arrays + const auto& vertex_array{regs.vertex_array[0]}; + ASSERT_MSG(vertex_array.enable, "vertex array 0 is disabled?"); + ASSERT_MSG(!vertex_array.divisor, "vertex array 0 divisor is unimplemented!"); + for (unsigned index = 1; index < Maxwell::NumVertexArrays; ++index) { + ASSERT_MSG(!regs.vertex_array[index].enable, "vertex array %d is unimplemented!", index); + } + + // Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL. + // Enables the first 16 vertex attributes always, as we don't know which ones are actually used + // until shader time. Note, Tegra technically supports 32, but we're cappinig this to 16 for now + // to avoid OpenGL errors. + for (unsigned index = 0; index < 16; ++index) { + auto& attrib = regs.vertex_attrib_format[index]; + glVertexAttribPointer(index, attrib.ComponentCount(), MaxwellToGL::VertexType(attrib), + attrib.IsNormalized() ? GL_TRUE : GL_FALSE, vertex_array.stride, + reinterpret_cast<GLvoid*>(buffer_offset + attrib.offset)); + glEnableVertexAttribArray(index); + hw_vao_enabled_attributes[index] = true; + } + + // Copy vertex array data + const u32 data_size{vertex_array.stride * regs.vertex_buffer.count}; + const VAddr data_addr{memory_manager->PhysicalToVirtualAddress(vertex_array.StartAddress())}; + res_cache.FlushRegion(data_addr, data_size, nullptr); + Memory::ReadBlock(data_addr, array_ptr, data_size); + + array_ptr += data_size; + buffer_offset += data_size; +} + +void RasterizerOpenGL::SetupVertexShader(VSUniformData* ub_ptr, GLintptr buffer_offset) { + MICROPROFILE_SCOPE(OpenGL_VS); + LOG_CRITICAL(Render_OpenGL, "Emulated shaders are not supported! Using a passthrough shader."); + glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, current_shader->shader.handle); +} + +void RasterizerOpenGL::SetupFragmentShader(FSUniformData* ub_ptr, GLintptr buffer_offset) { + MICROPROFILE_SCOPE(OpenGL_FS); + UNREACHABLE(); +} + +bool RasterizerOpenGL::AccelerateDrawBatch(bool is_indexed) { + if (!has_ARB_separate_shader_objects) { + UNREACHABLE(); + return false; + } + + accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays; + DrawArrays(); + + return true; +} + +void RasterizerOpenGL::DrawArrays() { + if (accelerate_draw == AccelDraw::Disabled) + return; + + MICROPROFILE_SCOPE(OpenGL_Drawing); + const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs; + + // TODO(bunnei): Implement these + const bool has_stencil = false; + const bool using_color_fb = true; + const bool using_depth_fb = false; + const MathUtil::Rectangle<s32> viewport_rect{regs.viewport[0].GetRect()}; + + const bool write_color_fb = + state.color_mask.red_enabled == GL_TRUE || state.color_mask.green_enabled == GL_TRUE || + state.color_mask.blue_enabled == GL_TRUE || state.color_mask.alpha_enabled == GL_TRUE; + + const bool write_depth_fb = + (state.depth.test_enabled && state.depth.write_mask == GL_TRUE) || + (has_stencil && state.stencil.test_enabled && state.stencil.write_mask != 0); + + Surface color_surface; + Surface depth_surface; + MathUtil::Rectangle<u32> surfaces_rect; + std::tie(color_surface, depth_surface, surfaces_rect) = + res_cache.GetFramebufferSurfaces(using_color_fb, using_depth_fb, viewport_rect); + + const u16 res_scale = color_surface != nullptr + ? color_surface->res_scale + : (depth_surface == nullptr ? 1u : depth_surface->res_scale); + + MathUtil::Rectangle<u32> draw_rect{ + static_cast<u32>(MathUtil::Clamp<s32>(static_cast<s32>(surfaces_rect.left) + + viewport_rect.left * res_scale, + surfaces_rect.left, surfaces_rect.right)), // Left + static_cast<u32>(MathUtil::Clamp<s32>(static_cast<s32>(surfaces_rect.bottom) + + viewport_rect.top * res_scale, + surfaces_rect.bottom, surfaces_rect.top)), // Top + static_cast<u32>(MathUtil::Clamp<s32>(static_cast<s32>(surfaces_rect.left) + + viewport_rect.right * res_scale, + surfaces_rect.left, surfaces_rect.right)), // Right + static_cast<u32>(MathUtil::Clamp<s32>(static_cast<s32>(surfaces_rect.bottom) + + viewport_rect.bottom * res_scale, + surfaces_rect.bottom, surfaces_rect.top))}; // Bottom + + // Bind the framebuffer surfaces + BindFramebufferSurfaces(color_surface, depth_surface, has_stencil); + + // Sync the viewport + SyncViewport(surfaces_rect, res_scale); + + // TODO(bunnei): Sync framebuffer_scale uniform here + // TODO(bunnei): Sync scissorbox uniform(s) here + // TODO(bunnei): Sync and bind the texture surfaces + + // Sync and bind the shader + if (shader_dirty) { + SetShader(); + shader_dirty = false; + } + + // Sync the uniform data + if (uniform_block_data.dirty) { + glBufferSubData(GL_UNIFORM_BUFFER, 0, sizeof(UniformData), &uniform_block_data.data); + uniform_block_data.dirty = false; + } + + // Viewport can have negative offsets or larger dimensions than our framebuffer sub-rect. Enable + // scissor test to prevent drawing outside of the framebuffer region + state.scissor.enabled = true; + state.scissor.x = draw_rect.left; + state.scissor.y = draw_rect.bottom; + state.scissor.width = draw_rect.GetWidth(); + state.scissor.height = draw_rect.GetHeight(); + state.Apply(); + + // Draw the vertex batch + const bool is_indexed = accelerate_draw == AccelDraw::Indexed; + AnalyzeVertexArray(is_indexed); + state.draw.vertex_buffer = stream_buffer->GetHandle(); + state.Apply(); + + size_t buffer_size = static_cast<size_t>(vs_input_size); + if (is_indexed) { + UNREACHABLE(); + } + buffer_size += sizeof(VSUniformData); + + size_t ptr_pos = 0; + u8* buffer_ptr; + GLintptr buffer_offset; + std::tie(buffer_ptr, buffer_offset) = + stream_buffer->Map(static_cast<GLsizeiptr>(buffer_size), 4); + + SetupVertexArray(buffer_ptr, buffer_offset); + ptr_pos += vs_input_size; + + GLintptr index_buffer_offset = 0; + if (is_indexed) { + UNREACHABLE(); + } + + SetupVertexShader(reinterpret_cast<VSUniformData*>(&buffer_ptr[ptr_pos]), + buffer_offset + static_cast<GLintptr>(ptr_pos)); + const GLintptr vs_ubo_offset = buffer_offset + static_cast<GLintptr>(ptr_pos); + ptr_pos += sizeof(VSUniformData); + + stream_buffer->Unmap(); + + const auto copy_buffer = [&](GLuint handle, GLintptr offset, GLsizeiptr size) { + if (has_ARB_direct_state_access) { + glCopyNamedBufferSubData(stream_buffer->GetHandle(), handle, offset, 0, size); + } else { + glBindBuffer(GL_COPY_WRITE_BUFFER, handle); + glCopyBufferSubData(GL_ARRAY_BUFFER, GL_COPY_WRITE_BUFFER, offset, 0, size); + } + }; + + copy_buffer(vs_uniform_buffer.handle, vs_ubo_offset, sizeof(VSUniformData)); + + glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, current_shader->shader.handle); + + if (is_indexed) { + UNREACHABLE(); + } else { + glDrawArrays(MaxwellToGL::PrimitiveTopology(regs.draw.topology), 0, + regs.vertex_buffer.count); + } + + // Disable scissor test + state.scissor.enabled = false; + + accelerate_draw = AccelDraw::Disabled; + + // Unbind textures for potential future use as framebuffer attachments + for (auto& texture_unit : state.texture_units) { + texture_unit.texture_2d = 0; + } + state.Apply(); + + // Mark framebuffer surfaces as dirty + MathUtil::Rectangle<u32> draw_rect_unscaled{ + draw_rect.left / res_scale, draw_rect.top / res_scale, draw_rect.right / res_scale, + draw_rect.bottom / res_scale}; + + if (color_surface != nullptr && write_color_fb) { + auto interval = color_surface->GetSubRectInterval(draw_rect_unscaled); + res_cache.InvalidateRegion(boost::icl::first(interval), boost::icl::length(interval), + color_surface); + } + if (depth_surface != nullptr && write_depth_fb) { + auto interval = depth_surface->GetSubRectInterval(draw_rect_unscaled); + res_cache.InvalidateRegion(boost::icl::first(interval), boost::icl::length(interval), + depth_surface); + } +} + +void RasterizerOpenGL::NotifyMaxwellRegisterChanged(u32 id) {} + +void RasterizerOpenGL::FlushAll() { + MICROPROFILE_SCOPE(OpenGL_CacheManagement); + res_cache.FlushAll(); +} + +void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) { + MICROPROFILE_SCOPE(OpenGL_CacheManagement); + res_cache.FlushRegion(addr, size); +} + +void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) { + MICROPROFILE_SCOPE(OpenGL_CacheManagement); + res_cache.InvalidateRegion(addr, size, nullptr); +} + +void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) { + MICROPROFILE_SCOPE(OpenGL_CacheManagement); + res_cache.FlushRegion(addr, size); + res_cache.InvalidateRegion(addr, size, nullptr); +} + +bool RasterizerOpenGL::AccelerateDisplayTransfer(const void* config) { + MICROPROFILE_SCOPE(OpenGL_Blits); + UNREACHABLE(); + return true; +} + +bool RasterizerOpenGL::AccelerateTextureCopy(const void* config) { + UNREACHABLE(); + return true; +} + +bool RasterizerOpenGL::AccelerateFill(const void* config) { + UNREACHABLE(); + return true; +} + +bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& framebuffer, + VAddr framebuffer_addr, u32 pixel_stride, + ScreenInfo& screen_info) { + if (framebuffer_addr == 0) { + return false; + } + MICROPROFILE_SCOPE(OpenGL_CacheManagement); + + SurfaceParams src_params; + src_params.addr = framebuffer_addr; + src_params.width = std::min(framebuffer.width, pixel_stride); + src_params.height = framebuffer.height; + src_params.stride = pixel_stride; + src_params.is_tiled = false; + src_params.pixel_format = + SurfaceParams::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format); + src_params.UpdateParams(); + + MathUtil::Rectangle<u32> src_rect; + Surface src_surface; + std::tie(src_surface, src_rect) = + res_cache.GetSurfaceSubRect(src_params, ScaleMatch::Ignore, true); + + if (src_surface == nullptr) { + return false; + } + + u32 scaled_width = src_surface->GetScaledWidth(); + u32 scaled_height = src_surface->GetScaledHeight(); + + screen_info.display_texcoords = MathUtil::Rectangle<float>( + (float)src_rect.bottom / (float)scaled_height, (float)src_rect.left / (float)scaled_width, + (float)src_rect.top / (float)scaled_height, (float)src_rect.right / (float)scaled_width); + + screen_info.display_texture = src_surface->texture.handle; + + return true; +} + +void RasterizerOpenGL::SetShader() { + // TODO(bunnei): The below sets up a static test shader for passing untransformed vertices to + // OpenGL for rendering. This should be removed/replaced when we start emulating Maxwell + // shaders. + + static constexpr char vertex_shader[] = R"( +#version 150 core + +in vec2 vert_position; +in vec2 vert_tex_coord; +out vec2 frag_tex_coord; + +void main() { + // Multiply input position by the rotscale part of the matrix and then manually translate by + // the last column. This is equivalent to using a full 3x3 matrix and expanding the vector + // to `vec3(vert_position.xy, 1.0)` + gl_Position = vec4(mat2(mat3x2(0.0015625f, 0.0, 0.0, -0.0027778, -1.0, 1.0)) * vert_position + mat3x2(0.0015625f, 0.0, 0.0, -0.0027778, -1.0, 1.0)[2], 0.0, 1.0); + frag_tex_coord = vert_tex_coord; +} +)"; + + static constexpr char fragment_shader[] = R"( +#version 150 core + +in vec2 frag_tex_coord; +out vec4 color; + +uniform sampler2D color_texture; + +void main() { + color = vec4(1.0, 0.0, 1.0, 0.0); +} +)"; + + if (current_shader) { + return; + } + + LOG_CRITICAL(Render_OpenGL, "Emulated shaders are not supported! Using a passthrough shader."); + + current_shader = &test_shader; + if (has_ARB_separate_shader_objects) { + test_shader.shader.Create(vertex_shader, nullptr, fragment_shader, {}, true); + glActiveShaderProgram(pipeline.handle, test_shader.shader.handle); + } else { + UNREACHABLE(); + } + + state.draw.shader_program = test_shader.shader.handle; + state.Apply(); + + if (has_ARB_separate_shader_objects) { + state.draw.shader_program = 0; + state.Apply(); + } +} + +void RasterizerOpenGL::BindFramebufferSurfaces(const Surface& color_surface, + const Surface& depth_surface, bool has_stencil) { + state.draw.draw_framebuffer = framebuffer.handle; + state.Apply(); + + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, + color_surface != nullptr ? color_surface->texture.handle : 0, 0); + if (depth_surface != nullptr) { + if (has_stencil) { + // attach both depth and stencil + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, + depth_surface->texture.handle, 0); + } else { + // attach depth + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, + depth_surface->texture.handle, 0); + // clear stencil attachment + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); + } + } else { + // clear both depth and stencil attachment + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, + 0); + } +} + +void RasterizerOpenGL::SyncViewport(const MathUtil::Rectangle<u32>& surfaces_rect, u16 res_scale) { + const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs; + const MathUtil::Rectangle<s32> viewport_rect{regs.viewport[0].GetRect()}; + + state.viewport.x = static_cast<GLint>(surfaces_rect.left) + viewport_rect.left * res_scale; + state.viewport.y = static_cast<GLint>(surfaces_rect.bottom) + viewport_rect.bottom * res_scale; + state.viewport.width = static_cast<GLsizei>(viewport_rect.GetWidth() * res_scale); + state.viewport.height = static_cast<GLsizei>(viewport_rect.GetHeight() * res_scale); +} + +void RasterizerOpenGL::SyncClipEnabled() { + UNREACHABLE(); +} + +void RasterizerOpenGL::SyncClipCoef() { + UNREACHABLE(); +} + +void RasterizerOpenGL::SyncCullMode() { + UNREACHABLE(); +} + +void RasterizerOpenGL::SyncDepthScale() { + UNREACHABLE(); +} + +void RasterizerOpenGL::SyncDepthOffset() { + UNREACHABLE(); +} + +void RasterizerOpenGL::SyncBlendEnabled() { + UNREACHABLE(); +} + +void RasterizerOpenGL::SyncBlendFuncs() { + UNREACHABLE(); +} + +void RasterizerOpenGL::SyncBlendColor() { + UNREACHABLE(); +} diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h new file mode 100644 index 000000000..fd53e94cd --- /dev/null +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -0,0 +1,179 @@ +// Copyright 2015 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <cstddef> +#include <cstring> +#include <memory> +#include <unordered_map> +#include <vector> +#include <glad/glad.h> +#include "common/bit_field.h" +#include "common/common_types.h" +#include "common/hash.h" +#include "common/vector_math.h" +#include "video_core/rasterizer_interface.h" +#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_gen.h" +#include "video_core/renderer_opengl/gl_state.h" +#include "video_core/renderer_opengl/gl_stream_buffer.h" + +struct ScreenInfo; + +class RasterizerOpenGL : public VideoCore::RasterizerInterface { +public: + RasterizerOpenGL(); + ~RasterizerOpenGL() override; + + void DrawArrays() override; + void NotifyMaxwellRegisterChanged(u32 id) override; + void FlushAll() override; + void FlushRegion(VAddr addr, u64 size) override; + void InvalidateRegion(VAddr addr, u64 size) override; + void FlushAndInvalidateRegion(VAddr addr, u64 size) override; + bool AccelerateDisplayTransfer(const void* config) override; + bool AccelerateTextureCopy(const void* config) override; + bool AccelerateFill(const void* config) override; + bool AccelerateDisplay(const Tegra::FramebufferConfig& framebuffer, VAddr framebuffer_addr, + u32 pixel_stride, ScreenInfo& screen_info) override; + bool AccelerateDrawBatch(bool is_indexed) override; + + /// OpenGL shader generated for a given Maxwell register state + struct MaxwellShader { + /// OpenGL shader resource + OGLShader shader; + }; + + struct VertexShader { + OGLShader shader; + }; + + struct FragmentShader { + OGLShader shader; + }; + + /// 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 wether 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 UniformData {}; + + // static_assert( + // sizeof(UniformData) == 0x460, + // "The size of the UniformData structure has changed, update the structure in the shader"); + static_assert(sizeof(UniformData) < 16384, + "UniformData structure must be less than 16kb as per the OpenGL spec"); + + struct VSUniformData {}; + // static_assert( + // sizeof(VSUniformData) == 1856, + // "The size of the VSUniformData structure has changed, update the structure in the + // shader"); + static_assert(sizeof(VSUniformData) < 16384, + "VSUniformData structure must be less than 16kb as per the OpenGL spec"); + + struct FSUniformData {}; + // static_assert( + // sizeof(FSUniformData) == 1856, + // "The size of the FSUniformData structure has changed, update the structure in the + // shader"); + static_assert(sizeof(FSUniformData) < 16384, + "FSUniformData structure must be less than 16kb as per the OpenGL spec"); + +private: + struct SamplerInfo {}; + + /// Binds the framebuffer color and depth surface + void BindFramebufferSurfaces(const Surface& color_surface, const Surface& depth_surface, + bool has_stencil); + + /// Syncs the viewport to match the guest state + void SyncViewport(const MathUtil::Rectangle<u32>& surfaces_rect, u16 res_scale); + + /// Syncs the clip enabled status to match the guest state + void SyncClipEnabled(); + + /// Syncs the clip coefficients to match the guest state + void SyncClipCoef(); + + /// Sets the OpenGL shader in accordance with the current guest state + void SetShader(); + + /// Syncs the cull mode to match the guest state + void SyncCullMode(); + + /// Syncs the depth scale to match the guest state + void SyncDepthScale(); + + /// Syncs the depth offset to match the guest state + void SyncDepthOffset(); + + /// Syncs the blend enabled status to match the guest state + void SyncBlendEnabled(); + + /// Syncs the blend functions to match the guest state + void SyncBlendFuncs(); + + /// Syncs the blend color to match the guest state + void SyncBlendColor(); + + bool has_ARB_buffer_storage; + bool has_ARB_direct_state_access; + bool has_ARB_separate_shader_objects; + bool has_ARB_vertex_attrib_binding; + + OpenGLState state; + + RasterizerCacheOpenGL res_cache; + + /// Shader used for test renderering - to be removed once we have emulated shaders + MaxwellShader test_shader{}; + + const MaxwellShader* current_shader{}; + bool shader_dirty{}; + + struct { + UniformData data; + bool dirty; + } uniform_block_data = {}; + + OGLPipeline pipeline; + OGLVertexArray sw_vao; + OGLVertexArray hw_vao; + std::array<bool, 16> hw_vao_enabled_attributes; + + std::array<SamplerInfo, 32> texture_samplers; + static constexpr size_t VERTEX_BUFFER_SIZE = 128 * 1024 * 1024; + std::unique_ptr<OGLStreamBuffer> vertex_buffer; + OGLBuffer uniform_buffer; + OGLFramebuffer framebuffer; + + static constexpr size_t STREAM_BUFFER_SIZE = 4 * 1024 * 1024; + std::unique_ptr<OGLStreamBuffer> stream_buffer; + + GLsizeiptr vs_input_size; + + void AnalyzeVertexArray(bool is_indexed); + void SetupVertexArray(u8* array_ptr, GLintptr buffer_offset); + + OGLBuffer vs_uniform_buffer; + std::unordered_map<GLShader::MaxwellVSConfig, VertexShader*> vs_shader_map; + std::unordered_map<std::string, VertexShader> vs_shader_cache; + OGLShader vs_default_shader; + + void SetupVertexShader(VSUniformData* ub_ptr, GLintptr buffer_offset); + + OGLBuffer fs_uniform_buffer; + std::unordered_map<GLShader::MaxwellFSConfig, FragmentShader*> fs_shader_map; + std::unordered_map<std::string, FragmentShader> fs_shader_cache; + OGLShader fs_default_shader; + + void SetupFragmentShader(FSUniformData* ub_ptr, GLintptr buffer_offset); + + enum class AccelDraw { Disabled, Arrays, Indexed }; + AccelDraw accelerate_draw; +}; diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp new file mode 100644 index 000000000..2ffbd3bab --- /dev/null +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -0,0 +1,1432 @@ +// Copyright 2015 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <algorithm> +#include <atomic> +#include <cstring> +#include <iterator> +#include <memory> +#include <unordered_set> +#include <utility> +#include <vector> +#include <boost/optional.hpp> +#include <boost/range/iterator_range.hpp> +#include <glad/glad.h> +#include "common/alignment.h" +#include "common/bit_field.h" +#include "common/color.h" +#include "common/logging/log.h" +#include "common/math_util.h" +#include "common/microprofile.h" +#include "common/scope_exit.h" +#include "common/vector_math.h" +#include "core/core.h" +#include "core/frontend/emu_window.h" +#include "core/hle/kernel/process.h" +#include "core/hle/kernel/vm_manager.h" +#include "core/memory.h" +#include "core/settings.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/renderer_opengl/gl_rasterizer_cache.h" +#include "video_core/renderer_opengl/gl_state.h" +#include "video_core/utils.h" +#include "video_core/video_core.h" + +using SurfaceType = SurfaceParams::SurfaceType; +using PixelFormat = SurfaceParams::PixelFormat; + +struct FormatTuple { + GLint internal_format; + GLenum format; + GLenum type; +}; + +static constexpr std::array<FormatTuple, 5> fb_format_tuples = {{ + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8}, // RGBA8 + {GL_RGB8, GL_BGR, GL_UNSIGNED_BYTE}, // RGB8 + {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_5_5_5_1}, // RGB5A1 + {GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5}, // RGB565 + {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4}, // RGBA4 +}}; + +static constexpr std::array<FormatTuple, 4> depth_format_tuples = {{ + {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16 + {}, + {GL_DEPTH_COMPONENT24, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT}, // D24 + {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24S8 +}}; + +static constexpr FormatTuple tex_tuple = {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE}; + +static const FormatTuple& GetFormatTuple(PixelFormat pixel_format) { + const SurfaceType type = SurfaceParams::GetFormatType(pixel_format); + if (type == SurfaceType::Color) { + ASSERT(static_cast<size_t>(pixel_format) < fb_format_tuples.size()); + return fb_format_tuples[static_cast<unsigned int>(pixel_format)]; + } else if (type == SurfaceType::Depth || type == SurfaceType::DepthStencil) { + size_t tuple_idx = static_cast<size_t>(pixel_format) - 14; + ASSERT(tuple_idx < depth_format_tuples.size()); + return depth_format_tuples[tuple_idx]; + } + return tex_tuple; +} + +template <typename Map, typename Interval> +constexpr auto RangeFromInterval(Map& map, const Interval& interval) { + return boost::make_iterator_range(map.equal_range(interval)); +} + +static u16 GetResolutionScaleFactor() { + return static_cast<u16>(!Settings::values.resolution_factor + ? VideoCore::g_emu_window->GetFramebufferLayout().GetScalingRatio() + : Settings::values.resolution_factor); +} + +template <bool morton_to_gl, PixelFormat format> +static void MortonCopyTile(u32 stride, u8* tile_buffer, u8* gl_buffer) { + constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / 8; + constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format); + for (u32 y = 0; y < 8; ++y) { + for (u32 x = 0; x < 8; ++x) { + u8* tile_ptr = tile_buffer + VideoCore::MortonInterleave(x, y) * bytes_per_pixel; + u8* gl_ptr = gl_buffer + ((7 - y) * stride + x) * gl_bytes_per_pixel; + if (morton_to_gl) { + if (format == PixelFormat::D24S8) { + gl_ptr[0] = tile_ptr[3]; + std::memcpy(gl_ptr + 1, tile_ptr, 3); + } else { + std::memcpy(gl_ptr, tile_ptr, bytes_per_pixel); + } + } else { + if (format == PixelFormat::D24S8) { + std::memcpy(tile_ptr, gl_ptr + 1, 3); + tile_ptr[3] = gl_ptr[0]; + } else { + std::memcpy(tile_ptr, gl_ptr, bytes_per_pixel); + } + } + } + } +} + +template <bool morton_to_gl, PixelFormat format> +static void MortonCopy(u32 stride, u32 height, u8* gl_buffer, VAddr base, VAddr start, VAddr end) { + constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / 8; + constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format); + + // TODO(bunnei): Assumes the default rendering GOB size of 16 (128 lines). We should check the + // configuration for this and perform more generic un/swizzle + LOG_WARNING(Render_OpenGL, "need to use correct swizzle/GOB parameters!"); + VideoCore::MortonCopyPixels128(stride, height, bytes_per_pixel, gl_bytes_per_pixel, + Memory::GetPointer(base), gl_buffer, morton_to_gl); +} + +static constexpr std::array<void (*)(u32, u32, u8*, VAddr, VAddr, VAddr), 18> morton_to_gl_fns = { + MortonCopy<true, PixelFormat::RGBA8>, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, +}; + +static constexpr std::array<void (*)(u32, u32, u8*, VAddr, VAddr, VAddr), 18> gl_to_morton_fns = { + MortonCopy<false, PixelFormat::RGBA8>, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, + nullptr, +}; + +// Allocate an uninitialized texture of appropriate size and format for the surface +static void AllocateSurfaceTexture(GLuint texture, const FormatTuple& format_tuple, u32 width, + u32 height) { + OpenGLState cur_state = OpenGLState::GetCurState(); + + // Keep track of previous texture bindings + GLuint old_tex = cur_state.texture_units[0].texture_2d; + cur_state.texture_units[0].texture_2d = texture; + cur_state.Apply(); + glActiveTexture(GL_TEXTURE0); + + glTexImage2D(GL_TEXTURE_2D, 0, format_tuple.internal_format, width, height, 0, + format_tuple.format, format_tuple.type, nullptr); + + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 0); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); + + // Restore previous texture bindings + cur_state.texture_units[0].texture_2d = old_tex; + cur_state.Apply(); +} + +static bool BlitTextures(GLuint src_tex, const MathUtil::Rectangle<u32>& src_rect, GLuint dst_tex, + const MathUtil::Rectangle<u32>& dst_rect, SurfaceType type, + GLuint read_fb_handle, GLuint draw_fb_handle) { + OpenGLState state = OpenGLState::GetCurState(); + + OpenGLState prev_state = state; + SCOPE_EXIT({ prev_state.Apply(); }); + + // Make sure textures aren't bound to texture units, since going to bind them to framebuffer + // components + state.ResetTexture(src_tex); + state.ResetTexture(dst_tex); + + state.draw.read_framebuffer = read_fb_handle; + state.draw.draw_framebuffer = draw_fb_handle; + state.Apply(); + + u32 buffers = 0; + + if (type == SurfaceType::Color || type == SurfaceType::Texture) { + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, src_tex, + 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, + 0); + + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, dst_tex, + 0); + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, + 0); + + buffers = GL_COLOR_BUFFER_BIT; + } else if (type == SurfaceType::Depth) { + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, src_tex, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); + + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, dst_tex, 0); + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); + + buffers = GL_DEPTH_BUFFER_BIT; + } else if (type == SurfaceType::DepthStencil) { + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, + src_tex, 0); + + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, + dst_tex, 0); + + buffers = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT; + } + + glBlitFramebuffer(src_rect.left, src_rect.bottom, src_rect.right, src_rect.top, dst_rect.left, + dst_rect.bottom, dst_rect.right, dst_rect.top, buffers, + buffers == GL_COLOR_BUFFER_BIT ? GL_LINEAR : GL_NEAREST); + + return true; +} + +static bool FillSurface(const Surface& surface, const u8* fill_data, + const MathUtil::Rectangle<u32>& fill_rect, GLuint draw_fb_handle) { + UNREACHABLE(); + return {}; +} + +SurfaceParams SurfaceParams::FromInterval(SurfaceInterval interval) const { + SurfaceParams params = *this; + const u32 tiled_size = is_tiled ? 8 : 1; + const u64 stride_tiled_bytes = BytesInPixels(stride * tiled_size); + VAddr aligned_start = + addr + Common::AlignDown(boost::icl::first(interval) - addr, stride_tiled_bytes); + VAddr aligned_end = + addr + Common::AlignUp(boost::icl::last_next(interval) - addr, stride_tiled_bytes); + + if (aligned_end - aligned_start > stride_tiled_bytes) { + params.addr = aligned_start; + params.height = static_cast<u32>((aligned_end - aligned_start) / BytesInPixels(stride)); + } else { + // 1 row + ASSERT(aligned_end - aligned_start == stride_tiled_bytes); + const u64 tiled_alignment = BytesInPixels(is_tiled ? 8 * 8 : 1); + aligned_start = + addr + Common::AlignDown(boost::icl::first(interval) - addr, tiled_alignment); + aligned_end = + addr + Common::AlignUp(boost::icl::last_next(interval) - addr, tiled_alignment); + params.addr = aligned_start; + params.width = static_cast<u32>(PixelsInBytes(aligned_end - aligned_start) / tiled_size); + params.stride = params.width; + params.height = tiled_size; + } + params.UpdateParams(); + + return params; +} + +SurfaceInterval SurfaceParams::GetSubRectInterval(MathUtil::Rectangle<u32> unscaled_rect) const { + if (unscaled_rect.GetHeight() == 0 || unscaled_rect.GetWidth() == 0) { + return {}; + } + + if (is_tiled) { + unscaled_rect.left = Common::AlignDown(unscaled_rect.left, 8) * 8; + unscaled_rect.bottom = Common::AlignDown(unscaled_rect.bottom, 8) / 8; + unscaled_rect.right = Common::AlignUp(unscaled_rect.right, 8) * 8; + unscaled_rect.top = Common::AlignUp(unscaled_rect.top, 8) / 8; + } + + const u32 stride_tiled = !is_tiled ? stride : stride * 8; + + const u32 pixel_offset = + stride_tiled * (!is_tiled ? unscaled_rect.bottom : (height / 8) - unscaled_rect.top) + + unscaled_rect.left; + + const u32 pixels = (unscaled_rect.GetHeight() - 1) * stride_tiled + unscaled_rect.GetWidth(); + + return {addr + BytesInPixels(pixel_offset), addr + BytesInPixels(pixel_offset + pixels)}; +} + +MathUtil::Rectangle<u32> SurfaceParams::GetSubRect(const SurfaceParams& sub_surface) const { + const u32 begin_pixel_index = static_cast<u32>(PixelsInBytes(sub_surface.addr - addr)); + + if (is_tiled) { + const int x0 = (begin_pixel_index % (stride * 8)) / 8; + const int y0 = (begin_pixel_index / (stride * 8)) * 8; + // Top to bottom + return MathUtil::Rectangle<u32>(x0, height - y0, x0 + sub_surface.width, + height - (y0 + sub_surface.height)); + } + + const int x0 = begin_pixel_index % stride; + const int y0 = begin_pixel_index / stride; + // Bottom to top + return MathUtil::Rectangle<u32>(x0, y0 + sub_surface.height, x0 + sub_surface.width, y0); +} + +MathUtil::Rectangle<u32> SurfaceParams::GetScaledSubRect(const SurfaceParams& sub_surface) const { + auto rect = GetSubRect(sub_surface); + rect.left = rect.left * res_scale; + rect.right = rect.right * res_scale; + rect.top = rect.top * res_scale; + rect.bottom = rect.bottom * res_scale; + return rect; +} + +bool SurfaceParams::ExactMatch(const SurfaceParams& other_surface) const { + return std::tie(other_surface.addr, other_surface.width, other_surface.height, + other_surface.stride, other_surface.pixel_format, other_surface.is_tiled) == + std::tie(addr, width, height, stride, pixel_format, is_tiled) && + pixel_format != PixelFormat::Invalid; +} + +bool SurfaceParams::CanSubRect(const SurfaceParams& sub_surface) const { + return sub_surface.addr >= addr && sub_surface.end <= end && + sub_surface.pixel_format == pixel_format && pixel_format != PixelFormat::Invalid && + sub_surface.is_tiled == is_tiled && + (sub_surface.addr - addr) % BytesInPixels(is_tiled ? 64 : 1) == 0 && + (sub_surface.stride == stride || sub_surface.height <= (is_tiled ? 8u : 1u)) && + GetSubRect(sub_surface).left + sub_surface.width <= stride; +} + +bool SurfaceParams::CanExpand(const SurfaceParams& expanded_surface) const { + return pixel_format != PixelFormat::Invalid && pixel_format == expanded_surface.pixel_format && + addr <= expanded_surface.end && expanded_surface.addr <= end && + is_tiled == expanded_surface.is_tiled && stride == expanded_surface.stride && + (std::max(expanded_surface.addr, addr) - std::min(expanded_surface.addr, addr)) % + BytesInPixels(stride * (is_tiled ? 8 : 1)) == + 0; +} + +bool SurfaceParams::CanTexCopy(const SurfaceParams& texcopy_params) const { + if (pixel_format == PixelFormat::Invalid || addr > texcopy_params.addr || + end < texcopy_params.end) { + return false; + } + if (texcopy_params.width != texcopy_params.stride) { + const u32 tile_stride = static_cast<u32>(BytesInPixels(stride * (is_tiled ? 8 : 1))); + return (texcopy_params.addr - addr) % BytesInPixels(is_tiled ? 64 : 1) == 0 && + texcopy_params.width % BytesInPixels(is_tiled ? 64 : 1) == 0 && + (texcopy_params.height == 1 || texcopy_params.stride == tile_stride) && + ((texcopy_params.addr - addr) % tile_stride) + texcopy_params.width <= tile_stride; + } + return FromInterval(texcopy_params.GetInterval()).GetInterval() == texcopy_params.GetInterval(); +} + +bool CachedSurface::CanFill(const SurfaceParams& dest_surface, + SurfaceInterval fill_interval) const { + if (type == SurfaceType::Fill && IsRegionValid(fill_interval) && + boost::icl::first(fill_interval) >= addr && + boost::icl::last_next(fill_interval) <= end && // dest_surface is within our fill range + dest_surface.FromInterval(fill_interval).GetInterval() == + fill_interval) { // make sure interval is a rectangle in dest surface + if (fill_size * 8 != dest_surface.GetFormatBpp()) { + // Check if bits repeat for our fill_size + const u32 dest_bytes_per_pixel = std::max(dest_surface.GetFormatBpp() / 8, 1u); + std::vector<u8> fill_test(fill_size * dest_bytes_per_pixel); + + for (u32 i = 0; i < dest_bytes_per_pixel; ++i) + std::memcpy(&fill_test[i * fill_size], &fill_data[0], fill_size); + + for (u32 i = 0; i < fill_size; ++i) + if (std::memcmp(&fill_test[dest_bytes_per_pixel * i], &fill_test[0], + dest_bytes_per_pixel) != 0) + return false; + + if (dest_surface.GetFormatBpp() == 4 && (fill_test[0] & 0xF) != (fill_test[0] >> 4)) + return false; + } + return true; + } + return false; +} + +bool CachedSurface::CanCopy(const SurfaceParams& dest_surface, + SurfaceInterval copy_interval) const { + SurfaceParams subrect_params = dest_surface.FromInterval(copy_interval); + ASSERT(subrect_params.GetInterval() == copy_interval); + if (CanSubRect(subrect_params)) + return true; + + if (CanFill(dest_surface, copy_interval)) + return true; + + return false; +} + +SurfaceInterval SurfaceParams::GetCopyableInterval(const Surface& src_surface) const { + SurfaceInterval result{}; + const auto valid_regions = + SurfaceRegions(GetInterval() & src_surface->GetInterval()) - src_surface->invalid_regions; + for (auto& valid_interval : valid_regions) { + const SurfaceInterval aligned_interval{ + addr + Common::AlignUp(boost::icl::first(valid_interval) - addr, + BytesInPixels(is_tiled ? 8 * 8 : 1)), + addr + Common::AlignDown(boost::icl::last_next(valid_interval) - addr, + BytesInPixels(is_tiled ? 8 * 8 : 1))}; + + if (BytesInPixels(is_tiled ? 8 * 8 : 1) > boost::icl::length(valid_interval) || + boost::icl::length(aligned_interval) == 0) { + continue; + } + + // Get the rectangle within aligned_interval + const u32 stride_bytes = static_cast<u32>(BytesInPixels(stride)) * (is_tiled ? 8 : 1); + SurfaceInterval rect_interval{ + addr + Common::AlignUp(boost::icl::first(aligned_interval) - addr, stride_bytes), + addr + Common::AlignDown(boost::icl::last_next(aligned_interval) - addr, stride_bytes), + }; + if (boost::icl::first(rect_interval) > boost::icl::last_next(rect_interval)) { + // 1 row + rect_interval = aligned_interval; + } else if (boost::icl::length(rect_interval) == 0) { + // 2 rows that do not make a rectangle, return the larger one + const SurfaceInterval row1{boost::icl::first(aligned_interval), + boost::icl::first(rect_interval)}; + const SurfaceInterval row2{boost::icl::first(rect_interval), + boost::icl::last_next(aligned_interval)}; + rect_interval = (boost::icl::length(row1) > boost::icl::length(row2)) ? row1 : row2; + } + + if (boost::icl::length(rect_interval) > boost::icl::length(result)) { + result = rect_interval; + } + } + return result; +} + +void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surface& dst_surface, + SurfaceInterval copy_interval) { + SurfaceParams subrect_params = dst_surface->FromInterval(copy_interval); + ASSERT(subrect_params.GetInterval() == copy_interval); + + ASSERT(src_surface != dst_surface); + + // This is only called when CanCopy is true, no need to run checks here + if (src_surface->type == SurfaceType::Fill) { + // FillSurface needs a 4 bytes buffer + const u64 fill_offset = + (boost::icl::first(copy_interval) - src_surface->addr) % src_surface->fill_size; + std::array<u8, 4> fill_buffer; + + u64 fill_buff_pos = fill_offset; + for (int i : {0, 1, 2, 3}) + fill_buffer[i] = src_surface->fill_data[fill_buff_pos++ % src_surface->fill_size]; + + FillSurface(dst_surface, &fill_buffer[0], dst_surface->GetScaledSubRect(subrect_params), + draw_framebuffer.handle); + return; + } + if (src_surface->CanSubRect(subrect_params)) { + BlitTextures(src_surface->texture.handle, src_surface->GetScaledSubRect(subrect_params), + dst_surface->texture.handle, dst_surface->GetScaledSubRect(subrect_params), + src_surface->type, read_framebuffer.handle, draw_framebuffer.handle); + return; + } + UNREACHABLE(); +} + +MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 64, 192)); +void CachedSurface::LoadGLBuffer(VAddr load_start, VAddr load_end) { + ASSERT(type != SurfaceType::Fill); + + u8* const texture_src_data = Memory::GetPointer(addr); + if (texture_src_data == nullptr) + return; + + if (gl_buffer == nullptr) { + gl_buffer_size = width * height * GetGLBytesPerPixel(pixel_format); + gl_buffer.reset(new u8[gl_buffer_size]); + } + + MICROPROFILE_SCOPE(OpenGL_SurfaceLoad); + + ASSERT(load_start >= addr && load_end <= end); + const u64 start_offset = load_start - addr; + + if (!is_tiled) { + ASSERT(type == SurfaceType::Color); + const u32 bytes_per_pixel{GetFormatBpp() >> 3}; + + // TODO(bunnei): Assumes the default rendering GOB size of 16 (128 lines). We should check + // the configuration for this and perform more generic un/swizzle + LOG_WARNING(Render_OpenGL, "need to use correct swizzle/GOB parameters!"); + VideoCore::MortonCopyPixels128(width, height, bytes_per_pixel, 4, + texture_src_data + start_offset, &gl_buffer[start_offset], + true); + } else { + morton_to_gl_fns[static_cast<size_t>(pixel_format)](stride, height, &gl_buffer[0], addr, + load_start, load_end); + } +} + +MICROPROFILE_DEFINE(OpenGL_SurfaceFlush, "OpenGL", "Surface Flush", MP_RGB(128, 192, 64)); +void CachedSurface::FlushGLBuffer(VAddr flush_start, VAddr flush_end) { + u8* const dst_buffer = Memory::GetPointer(addr); + if (dst_buffer == nullptr) + return; + + ASSERT(gl_buffer_size == width * height * GetGLBytesPerPixel(pixel_format)); + + // TODO: Should probably be done in ::Memory:: and check for other regions too + // same as loadglbuffer() + if (flush_start < Memory::VRAM_VADDR_END && flush_end > Memory::VRAM_VADDR_END) + flush_end = Memory::VRAM_VADDR_END; + + if (flush_start < Memory::VRAM_VADDR && flush_end > Memory::VRAM_VADDR) + flush_start = Memory::VRAM_VADDR; + + MICROPROFILE_SCOPE(OpenGL_SurfaceFlush); + + ASSERT(flush_start >= addr && flush_end <= end); + const u64 start_offset = flush_start - addr; + const u64 end_offset = flush_end - addr; + + if (type == SurfaceType::Fill) { + const u64 coarse_start_offset = start_offset - (start_offset % fill_size); + const u64 backup_bytes = start_offset % fill_size; + std::array<u8, 4> backup_data; + if (backup_bytes) + std::memcpy(&backup_data[0], &dst_buffer[coarse_start_offset], backup_bytes); + + for (u64 offset = coarse_start_offset; offset < end_offset; offset += fill_size) { + std::memcpy(&dst_buffer[offset], &fill_data[0], + std::min(fill_size, end_offset - offset)); + } + + if (backup_bytes) + std::memcpy(&dst_buffer[coarse_start_offset], &backup_data[0], backup_bytes); + } else if (!is_tiled) { + ASSERT(type == SurfaceType::Color); + std::memcpy(dst_buffer + start_offset, &gl_buffer[start_offset], flush_end - flush_start); + } else { + gl_to_morton_fns[static_cast<size_t>(pixel_format)](stride, height, &gl_buffer[0], addr, + flush_start, flush_end); + } +} + +MICROPROFILE_DEFINE(OpenGL_TextureUL, "OpenGL", "Texture Upload", MP_RGB(128, 64, 192)); +void CachedSurface::UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle, + GLuint draw_fb_handle) { + if (type == SurfaceType::Fill) + return; + + MICROPROFILE_SCOPE(OpenGL_TextureUL); + + ASSERT(gl_buffer_size == width * height * GetGLBytesPerPixel(pixel_format)); + + // Load data from memory to the surface + GLint x0 = static_cast<GLint>(rect.left); + GLint y0 = static_cast<GLint>(rect.bottom); + size_t buffer_offset = (y0 * stride + x0) * GetGLBytesPerPixel(pixel_format); + + const FormatTuple& tuple = GetFormatTuple(pixel_format); + GLuint target_tex = texture.handle; + + // If not 1x scale, create 1x texture that we will blit from to replace texture subrect in + // surface + OGLTexture unscaled_tex; + if (res_scale != 1) { + x0 = 0; + y0 = 0; + + unscaled_tex.Create(); + AllocateSurfaceTexture(unscaled_tex.handle, tuple, rect.GetWidth(), rect.GetHeight()); + target_tex = unscaled_tex.handle; + } + + OpenGLState cur_state = OpenGLState::GetCurState(); + + GLuint old_tex = cur_state.texture_units[0].texture_2d; + cur_state.texture_units[0].texture_2d = target_tex; + cur_state.Apply(); + + // Ensure no bad interactions with GL_UNPACK_ALIGNMENT + ASSERT(stride * GetGLBytesPerPixel(pixel_format) % 4 == 0); + glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(stride)); + + glActiveTexture(GL_TEXTURE0); + glTexSubImage2D(GL_TEXTURE_2D, 0, x0, y0, static_cast<GLsizei>(rect.GetWidth()), + static_cast<GLsizei>(rect.GetHeight()), tuple.format, tuple.type, + &gl_buffer[buffer_offset]); + + glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); + + cur_state.texture_units[0].texture_2d = old_tex; + cur_state.Apply(); + + if (res_scale != 1) { + auto scaled_rect = rect; + scaled_rect.left *= res_scale; + scaled_rect.top *= res_scale; + scaled_rect.right *= res_scale; + scaled_rect.bottom *= res_scale; + + BlitTextures(unscaled_tex.handle, {0, rect.GetHeight(), rect.GetWidth(), 0}, texture.handle, + scaled_rect, type, read_fb_handle, draw_fb_handle); + } +} + +MICROPROFILE_DEFINE(OpenGL_TextureDL, "OpenGL", "Texture Download", MP_RGB(128, 192, 64)); +void CachedSurface::DownloadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle, + GLuint draw_fb_handle) { + if (type == SurfaceType::Fill) + return; + + MICROPROFILE_SCOPE(OpenGL_TextureDL); + + if (gl_buffer == nullptr) { + gl_buffer_size = width * height * GetGLBytesPerPixel(pixel_format); + gl_buffer.reset(new u8[gl_buffer_size]); + } + + OpenGLState state = OpenGLState::GetCurState(); + OpenGLState prev_state = state; + SCOPE_EXIT({ prev_state.Apply(); }); + + const FormatTuple& tuple = GetFormatTuple(pixel_format); + + // Ensure no bad interactions with GL_PACK_ALIGNMENT + ASSERT(stride * GetGLBytesPerPixel(pixel_format) % 4 == 0); + glPixelStorei(GL_PACK_ROW_LENGTH, static_cast<GLint>(stride)); + size_t buffer_offset = (rect.bottom * stride + rect.left) * GetGLBytesPerPixel(pixel_format); + + // If not 1x scale, blit scaled texture to a new 1x texture and use that to flush + if (res_scale != 1) { + auto scaled_rect = rect; + scaled_rect.left *= res_scale; + scaled_rect.top *= res_scale; + scaled_rect.right *= res_scale; + scaled_rect.bottom *= res_scale; + + OGLTexture unscaled_tex; + unscaled_tex.Create(); + + MathUtil::Rectangle<u32> unscaled_tex_rect{0, rect.GetHeight(), rect.GetWidth(), 0}; + AllocateSurfaceTexture(unscaled_tex.handle, tuple, rect.GetWidth(), rect.GetHeight()); + BlitTextures(texture.handle, scaled_rect, unscaled_tex.handle, unscaled_tex_rect, type, + read_fb_handle, draw_fb_handle); + + state.texture_units[0].texture_2d = unscaled_tex.handle; + state.Apply(); + + glActiveTexture(GL_TEXTURE0); + glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, &gl_buffer[buffer_offset]); + } else { + state.ResetTexture(texture.handle); + state.draw.read_framebuffer = read_fb_handle; + state.Apply(); + + if (type == SurfaceType::Color || type == SurfaceType::Texture) { + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, + texture.handle, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, + 0, 0); + } else if (type == SurfaceType::Depth) { + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, + texture.handle, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); + } else { + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, + texture.handle, 0); + } + glReadPixels(static_cast<GLint>(rect.left), static_cast<GLint>(rect.bottom), + static_cast<GLsizei>(rect.GetWidth()), static_cast<GLsizei>(rect.GetHeight()), + tuple.format, tuple.type, &gl_buffer[buffer_offset]); + } + + glPixelStorei(GL_PACK_ROW_LENGTH, 0); +} + +enum MatchFlags { + Invalid = 1, // Flag that can be applied to other match types, invalid matches require + // validation before they can be used + Exact = 1 << 1, // Surfaces perfectly match + SubRect = 1 << 2, // Surface encompasses params + Copy = 1 << 3, // Surface we can copy from + Expand = 1 << 4, // Surface that can expand params + TexCopy = 1 << 5 // Surface that will match a display transfer "texture copy" parameters +}; + +constexpr MatchFlags operator|(MatchFlags lhs, MatchFlags rhs) { + return static_cast<MatchFlags>(static_cast<int>(lhs) | static_cast<int>(rhs)); +} + +/// Get the best surface match (and its match type) for the given flags +template <MatchFlags find_flags> +Surface FindMatch(const SurfaceCache& surface_cache, const SurfaceParams& params, + ScaleMatch match_scale_type, + boost::optional<SurfaceInterval> validate_interval = boost::none) { + Surface match_surface = nullptr; + bool match_valid = false; + u32 match_scale = 0; + SurfaceInterval match_interval{}; + + for (auto& pair : RangeFromInterval(surface_cache, params.GetInterval())) { + for (auto& surface : pair.second) { + bool res_scale_matched = match_scale_type == ScaleMatch::Exact + ? (params.res_scale == surface->res_scale) + : (params.res_scale <= surface->res_scale); + // validity will be checked in GetCopyableInterval + bool is_valid = + find_flags & MatchFlags::Copy + ? true + : surface->IsRegionValid(validate_interval.value_or(params.GetInterval())); + + if (!(find_flags & MatchFlags::Invalid) && !is_valid) + continue; + + auto IsMatch_Helper = [&](auto check_type, auto match_fn) { + if (!(find_flags & check_type)) + return; + + bool matched; + SurfaceInterval surface_interval; + std::tie(matched, surface_interval) = match_fn(); + if (!matched) + return; + + if (!res_scale_matched && match_scale_type != ScaleMatch::Ignore && + surface->type != SurfaceType::Fill) + return; + + // Found a match, update only if this is better than the previous one + auto UpdateMatch = [&] { + match_surface = surface; + match_valid = is_valid; + match_scale = surface->res_scale; + match_interval = surface_interval; + }; + + if (surface->res_scale > match_scale) { + UpdateMatch(); + return; + } else if (surface->res_scale < match_scale) { + return; + } + + if (is_valid && !match_valid) { + UpdateMatch(); + return; + } else if (is_valid != match_valid) { + return; + } + + if (boost::icl::length(surface_interval) > boost::icl::length(match_interval)) { + UpdateMatch(); + } + }; + IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::Exact>{}, [&] { + return std::make_pair(surface->ExactMatch(params), surface->GetInterval()); + }); + IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::SubRect>{}, [&] { + return std::make_pair(surface->CanSubRect(params), surface->GetInterval()); + }); + IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::Copy>{}, [&] { + auto copy_interval = + params.FromInterval(*validate_interval).GetCopyableInterval(surface); + bool matched = boost::icl::length(copy_interval & *validate_interval) != 0 && + surface->CanCopy(params, copy_interval); + return std::make_pair(matched, copy_interval); + }); + IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::Expand>{}, [&] { + return std::make_pair(surface->CanExpand(params), surface->GetInterval()); + }); + IsMatch_Helper(std::integral_constant<MatchFlags, MatchFlags::TexCopy>{}, [&] { + return std::make_pair(surface->CanTexCopy(params), surface->GetInterval()); + }); + } + } + return match_surface; +} + +RasterizerCacheOpenGL::RasterizerCacheOpenGL() { + read_framebuffer.Create(); + draw_framebuffer.Create(); + + attributeless_vao.Create(); + + d24s8_abgr_buffer.Create(); + d24s8_abgr_buffer_size = 0; + + const char* vs_source = R"( +#version 330 core +const vec2 vertices[4] = vec2[4](vec2(-1.0, -1.0), vec2(1.0, -1.0), vec2(-1.0, 1.0), vec2(1.0, 1.0)); +void main() { + gl_Position = vec4(vertices[gl_VertexID], 0.0, 1.0); +} +)"; + const char* fs_source = R"( +#version 330 core + +uniform samplerBuffer tbo; +uniform vec2 tbo_size; +uniform vec4 viewport; + +out vec4 color; + +void main() { + vec2 tbo_coord = (gl_FragCoord.xy - viewport.xy) * tbo_size / viewport.zw; + int tbo_offset = int(tbo_coord.y) * int(tbo_size.x) + int(tbo_coord.x); + color = texelFetch(tbo, tbo_offset).rabg; +} +)"; + d24s8_abgr_shader.Create(vs_source, nullptr, fs_source); + + OpenGLState state = OpenGLState::GetCurState(); + GLuint old_program = state.draw.shader_program; + state.draw.shader_program = d24s8_abgr_shader.handle; + state.Apply(); + + GLint tbo_u_id = glGetUniformLocation(d24s8_abgr_shader.handle, "tbo"); + ASSERT(tbo_u_id != -1); + glUniform1i(tbo_u_id, 0); + + state.draw.shader_program = old_program; + state.Apply(); + + d24s8_abgr_tbo_size_u_id = glGetUniformLocation(d24s8_abgr_shader.handle, "tbo_size"); + ASSERT(d24s8_abgr_tbo_size_u_id != -1); + d24s8_abgr_viewport_u_id = glGetUniformLocation(d24s8_abgr_shader.handle, "viewport"); + ASSERT(d24s8_abgr_viewport_u_id != -1); +} + +RasterizerCacheOpenGL::~RasterizerCacheOpenGL() { + FlushAll(); + while (!surface_cache.empty()) + UnregisterSurface(*surface_cache.begin()->second.begin()); +} + +bool RasterizerCacheOpenGL::BlitSurfaces(const Surface& src_surface, + const MathUtil::Rectangle<u32>& src_rect, + const Surface& dst_surface, + const MathUtil::Rectangle<u32>& dst_rect) { + if (!SurfaceParams::CheckFormatsBlittable(src_surface->pixel_format, dst_surface->pixel_format)) + return false; + + return BlitTextures(src_surface->texture.handle, src_rect, dst_surface->texture.handle, + dst_rect, src_surface->type, read_framebuffer.handle, + draw_framebuffer.handle); +} + +void RasterizerCacheOpenGL::ConvertD24S8toABGR(GLuint src_tex, + const MathUtil::Rectangle<u32>& src_rect, + GLuint dst_tex, + const MathUtil::Rectangle<u32>& dst_rect) { + OpenGLState prev_state = OpenGLState::GetCurState(); + SCOPE_EXIT({ prev_state.Apply(); }); + + OpenGLState state; + state.draw.read_framebuffer = read_framebuffer.handle; + state.draw.draw_framebuffer = draw_framebuffer.handle; + state.Apply(); + + glBindBuffer(GL_PIXEL_PACK_BUFFER, d24s8_abgr_buffer.handle); + + GLsizeiptr target_pbo_size = src_rect.GetWidth() * src_rect.GetHeight() * 4; + if (target_pbo_size > d24s8_abgr_buffer_size) { + d24s8_abgr_buffer_size = target_pbo_size * 2; + glBufferData(GL_PIXEL_PACK_BUFFER, d24s8_abgr_buffer_size, nullptr, GL_STREAM_COPY); + } + + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, src_tex, + 0); + glReadPixels(static_cast<GLint>(src_rect.left), static_cast<GLint>(src_rect.bottom), + static_cast<GLsizei>(src_rect.GetWidth()), + static_cast<GLsizei>(src_rect.GetHeight()), GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8, + 0); + + glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); + + // PBO now contains src_tex in RABG format + state.draw.shader_program = d24s8_abgr_shader.handle; + state.draw.vertex_array = attributeless_vao.handle; + state.viewport.x = static_cast<GLint>(dst_rect.left); + state.viewport.y = static_cast<GLint>(dst_rect.bottom); + state.viewport.width = static_cast<GLsizei>(dst_rect.GetWidth()); + state.viewport.height = static_cast<GLsizei>(dst_rect.GetHeight()); + state.Apply(); + + OGLTexture tbo; + tbo.Create(); + glActiveTexture(GL_TEXTURE0); + glBindTexture(GL_TEXTURE_BUFFER, tbo.handle); + glTexBuffer(GL_TEXTURE_BUFFER, GL_RGBA8, d24s8_abgr_buffer.handle); + + glUniform2f(d24s8_abgr_tbo_size_u_id, static_cast<GLfloat>(src_rect.GetWidth()), + static_cast<GLfloat>(src_rect.GetHeight())); + glUniform4f(d24s8_abgr_viewport_u_id, static_cast<GLfloat>(state.viewport.x), + static_cast<GLfloat>(state.viewport.y), static_cast<GLfloat>(state.viewport.width), + static_cast<GLfloat>(state.viewport.height)); + + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, dst_tex, 0); + glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); + glDrawArrays(GL_TRIANGLE_STRIP, 0, 4); + + glBindTexture(GL_TEXTURE_BUFFER, 0); +} + +Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, ScaleMatch match_res_scale, + bool load_if_create) { + if (params.addr == 0 || params.height * params.width == 0) { + return nullptr; + } + // Use GetSurfaceSubRect instead + ASSERT(params.width == params.stride); + + ASSERT(!params.is_tiled || (params.width % 8 == 0 && params.height % 8 == 0)); + + // Check for an exact match in existing surfaces + Surface surface = + FindMatch<MatchFlags::Exact | MatchFlags::Invalid>(surface_cache, params, match_res_scale); + + if (surface == nullptr) { + u16 target_res_scale = params.res_scale; + if (match_res_scale != ScaleMatch::Exact) { + // This surface may have a subrect of another surface with a higher res_scale, find it + // to adjust our params + SurfaceParams find_params = params; + Surface expandable = FindMatch<MatchFlags::Expand | MatchFlags::Invalid>( + surface_cache, find_params, match_res_scale); + if (expandable != nullptr && expandable->res_scale > target_res_scale) { + target_res_scale = expandable->res_scale; + } + // Keep res_scale when reinterpreting d24s8 -> rgba8 + if (params.pixel_format == PixelFormat::RGBA8) { + find_params.pixel_format = PixelFormat::D24S8; + expandable = FindMatch<MatchFlags::Expand | MatchFlags::Invalid>( + surface_cache, find_params, match_res_scale); + if (expandable != nullptr && expandable->res_scale > target_res_scale) { + target_res_scale = expandable->res_scale; + } + } + } + SurfaceParams new_params = params; + new_params.res_scale = target_res_scale; + surface = CreateSurface(new_params); + RegisterSurface(surface); + } + + if (load_if_create) { + ValidateSurface(surface, params.addr, params.size); + } + + return surface; +} + +SurfaceRect_Tuple RasterizerCacheOpenGL::GetSurfaceSubRect(const SurfaceParams& params, + ScaleMatch match_res_scale, + bool load_if_create) { + if (params.addr == 0 || params.height * params.width == 0) { + return std::make_tuple(nullptr, MathUtil::Rectangle<u32>{}); + } + + // Attempt to find encompassing surface + Surface surface = FindMatch<MatchFlags::SubRect | MatchFlags::Invalid>(surface_cache, params, + match_res_scale); + + // Check if FindMatch failed because of res scaling + // If that's the case create a new surface with + // the dimensions of the lower res_scale surface + // to suggest it should not be used again + if (surface == nullptr && match_res_scale != ScaleMatch::Ignore) { + surface = FindMatch<MatchFlags::SubRect | MatchFlags::Invalid>(surface_cache, params, + ScaleMatch::Ignore); + if (surface != nullptr) { + ASSERT(surface->res_scale < params.res_scale); + SurfaceParams new_params = *surface; + new_params.res_scale = params.res_scale; + + surface = CreateSurface(new_params); + RegisterSurface(surface); + } + } + + SurfaceParams aligned_params = params; + if (params.is_tiled) { + aligned_params.height = Common::AlignUp(params.height, 8); + aligned_params.width = Common::AlignUp(params.width, 8); + aligned_params.stride = Common::AlignUp(params.stride, 8); + aligned_params.UpdateParams(); + } + + // Check for a surface we can expand before creating a new one + if (surface == nullptr) { + surface = FindMatch<MatchFlags::Expand | MatchFlags::Invalid>(surface_cache, aligned_params, + match_res_scale); + if (surface != nullptr) { + aligned_params.width = aligned_params.stride; + aligned_params.UpdateParams(); + + SurfaceParams new_params = *surface; + new_params.addr = std::min(aligned_params.addr, surface->addr); + new_params.end = std::max(aligned_params.end, surface->end); + new_params.size = new_params.end - new_params.addr; + new_params.height = static_cast<u32>( + new_params.size / aligned_params.BytesInPixels(aligned_params.stride)); + ASSERT(new_params.size % aligned_params.BytesInPixels(aligned_params.stride) == 0); + + Surface new_surface = CreateSurface(new_params); + DuplicateSurface(surface, new_surface); + + // Delete the expanded surface, this can't be done safely yet + // because it may still be in use + remove_surfaces.emplace(surface); + + surface = new_surface; + RegisterSurface(new_surface); + } + } + + // No subrect found - create and return a new surface + if (surface == nullptr) { + SurfaceParams new_params = aligned_params; + // Can't have gaps in a surface + new_params.width = aligned_params.stride; + new_params.UpdateParams(); + // GetSurface will create the new surface and possibly adjust res_scale if necessary + surface = GetSurface(new_params, match_res_scale, load_if_create); + } else if (load_if_create) { + ValidateSurface(surface, aligned_params.addr, aligned_params.size); + } + + return std::make_tuple(surface, surface->GetScaledSubRect(params)); +} + +Surface RasterizerCacheOpenGL::GetTextureSurface(const void* config) { + UNREACHABLE(); + return {}; +} + +SurfaceSurfaceRect_Tuple RasterizerCacheOpenGL::GetFramebufferSurfaces( + bool using_color_fb, bool using_depth_fb, const MathUtil::Rectangle<s32>& viewport) { + const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs; + const auto& memory_manager = Core::System().GetInstance().GPU().memory_manager; + const auto& config = regs.rt[0]; + + // TODO(bunnei): This is hard corded to use just the first render buffer + LOG_WARNING(Render_OpenGL, "hard-coded for render target 0!"); + + // update resolution_scale_factor and reset cache if changed + // TODO (bunnei): This code was ported as-is from Citra, and is technically not thread-safe. We + // need to fix this before making the renderer multi-threaded. + static u16 resolution_scale_factor = GetResolutionScaleFactor(); + if (resolution_scale_factor != GetResolutionScaleFactor()) { + resolution_scale_factor = GetResolutionScaleFactor(); + FlushAll(); + while (!surface_cache.empty()) + UnregisterSurface(*surface_cache.begin()->second.begin()); + } + + MathUtil::Rectangle<u32> viewport_clamped{ + static_cast<u32>(MathUtil::Clamp(viewport.left, 0, static_cast<s32>(config.width))), + static_cast<u32>(MathUtil::Clamp(viewport.top, 0, static_cast<s32>(config.height))), + static_cast<u32>(MathUtil::Clamp(viewport.right, 0, static_cast<s32>(config.width))), + static_cast<u32>(MathUtil::Clamp(viewport.bottom, 0, static_cast<s32>(config.height)))}; + + // get color and depth surfaces + SurfaceParams color_params; + color_params.is_tiled = true; + color_params.res_scale = resolution_scale_factor; + color_params.width = config.width; + color_params.height = config.height; + SurfaceParams depth_params = color_params; + + color_params.addr = memory_manager->PhysicalToVirtualAddress(config.Address()); + color_params.pixel_format = SurfaceParams::PixelFormatFromRenderTargetFormat(config.format); + color_params.UpdateParams(); + + ASSERT_MSG(!using_depth_fb, "depth buffer is unimplemented"); + // depth_params.addr = config.GetDepthBufferPhysicalAddress(); + // depth_params.pixel_format = SurfaceParams::PixelFormatFromDepthFormat(config.depth_format); + // depth_params.UpdateParams(); + + auto color_vp_interval = color_params.GetSubRectInterval(viewport_clamped); + auto depth_vp_interval = depth_params.GetSubRectInterval(viewport_clamped); + + // Make sure that framebuffers don't overlap if both color and depth are being used + if (using_color_fb && using_depth_fb && + boost::icl::length(color_vp_interval & depth_vp_interval)) { + LOG_CRITICAL(Render_OpenGL, "Color and depth framebuffer memory regions overlap; " + "overlapping framebuffers not supported!"); + using_depth_fb = false; + } + + MathUtil::Rectangle<u32> color_rect{}; + Surface color_surface = nullptr; + if (using_color_fb) + std::tie(color_surface, color_rect) = + GetSurfaceSubRect(color_params, ScaleMatch::Exact, false); + + MathUtil::Rectangle<u32> depth_rect{}; + Surface depth_surface = nullptr; + if (using_depth_fb) + std::tie(depth_surface, depth_rect) = + GetSurfaceSubRect(depth_params, ScaleMatch::Exact, false); + + MathUtil::Rectangle<u32> fb_rect{}; + if (color_surface != nullptr && depth_surface != nullptr) { + fb_rect = color_rect; + // Color and Depth surfaces must have the same dimensions and offsets + if (color_rect.bottom != depth_rect.bottom || color_rect.top != depth_rect.top || + color_rect.left != depth_rect.left || color_rect.right != depth_rect.right) { + color_surface = GetSurface(color_params, ScaleMatch::Exact, false); + depth_surface = GetSurface(depth_params, ScaleMatch::Exact, false); + fb_rect = color_surface->GetScaledRect(); + } + } else if (color_surface != nullptr) { + fb_rect = color_rect; + } else if (depth_surface != nullptr) { + fb_rect = depth_rect; + } + + if (color_surface != nullptr) { + ValidateSurface(color_surface, boost::icl::first(color_vp_interval), + boost::icl::length(color_vp_interval)); + } + if (depth_surface != nullptr) { + ValidateSurface(depth_surface, boost::icl::first(depth_vp_interval), + boost::icl::length(depth_vp_interval)); + } + + return std::make_tuple(color_surface, depth_surface, fb_rect); +} + +Surface RasterizerCacheOpenGL::GetFillSurface(const void* config) { + UNREACHABLE(); + return {}; +} + +SurfaceRect_Tuple RasterizerCacheOpenGL::GetTexCopySurface(const SurfaceParams& params) { + MathUtil::Rectangle<u32> rect{}; + + Surface match_surface = FindMatch<MatchFlags::TexCopy | MatchFlags::Invalid>( + surface_cache, params, ScaleMatch::Ignore); + + if (match_surface != nullptr) { + ValidateSurface(match_surface, params.addr, params.size); + + SurfaceParams match_subrect; + if (params.width != params.stride) { + const u32 tiled_size = match_surface->is_tiled ? 8 : 1; + match_subrect = params; + match_subrect.width = + static_cast<u32>(match_surface->PixelsInBytes(params.width) / tiled_size); + match_subrect.stride = + static_cast<u32>(match_surface->PixelsInBytes(params.stride) / tiled_size); + match_subrect.height *= tiled_size; + } else { + match_subrect = match_surface->FromInterval(params.GetInterval()); + ASSERT(match_subrect.GetInterval() == params.GetInterval()); + } + + rect = match_surface->GetScaledSubRect(match_subrect); + } + + return std::make_tuple(match_surface, rect); +} + +void RasterizerCacheOpenGL::DuplicateSurface(const Surface& src_surface, + const Surface& dest_surface) { + ASSERT(dest_surface->addr <= src_surface->addr && dest_surface->end >= src_surface->end); + + BlitSurfaces(src_surface, src_surface->GetScaledRect(), dest_surface, + dest_surface->GetScaledSubRect(*src_surface)); + + dest_surface->invalid_regions -= src_surface->GetInterval(); + dest_surface->invalid_regions += src_surface->invalid_regions; + + SurfaceRegions regions; + for (auto& pair : RangeFromInterval(dirty_regions, src_surface->GetInterval())) { + if (pair.second == src_surface) { + regions += pair.first; + } + } + for (auto& interval : regions) { + dirty_regions.set({interval, dest_surface}); + } +} + +void RasterizerCacheOpenGL::ValidateSurface(const Surface& surface, VAddr addr, u64 size) { + if (size == 0) + return; + + const SurfaceInterval validate_interval(addr, addr + size); + + if (surface->type == SurfaceType::Fill) { + // Sanity check, fill surfaces will always be valid when used + ASSERT(surface->IsRegionValid(validate_interval)); + return; + } + + while (true) { + const auto it = surface->invalid_regions.find(validate_interval); + if (it == surface->invalid_regions.end()) + break; + + const auto interval = *it & validate_interval; + // Look for a valid surface to copy from + SurfaceParams params = surface->FromInterval(interval); + + Surface copy_surface = + FindMatch<MatchFlags::Copy>(surface_cache, params, ScaleMatch::Ignore, interval); + if (copy_surface != nullptr) { + SurfaceInterval copy_interval = params.GetCopyableInterval(copy_surface); + CopySurface(copy_surface, surface, copy_interval); + surface->invalid_regions.erase(copy_interval); + continue; + } + + // D24S8 to RGBA8 + if (surface->pixel_format == PixelFormat::RGBA8) { + params.pixel_format = PixelFormat::D24S8; + Surface reinterpret_surface = + FindMatch<MatchFlags::Copy>(surface_cache, params, ScaleMatch::Ignore, interval); + if (reinterpret_surface != nullptr) { + ASSERT(reinterpret_surface->pixel_format == PixelFormat::D24S8); + + SurfaceInterval convert_interval = params.GetCopyableInterval(reinterpret_surface); + SurfaceParams convert_params = surface->FromInterval(convert_interval); + auto src_rect = reinterpret_surface->GetScaledSubRect(convert_params); + auto dest_rect = surface->GetScaledSubRect(convert_params); + + ConvertD24S8toABGR(reinterpret_surface->texture.handle, src_rect, + surface->texture.handle, dest_rect); + + surface->invalid_regions.erase(convert_interval); + continue; + } + } + + // Load data from 3DS memory + FlushRegion(params.addr, params.size); + surface->LoadGLBuffer(params.addr, params.end); + surface->UploadGLTexture(surface->GetSubRect(params), read_framebuffer.handle, + draw_framebuffer.handle); + surface->invalid_regions.erase(params.GetInterval()); + } +} + +void RasterizerCacheOpenGL::FlushRegion(VAddr addr, u64 size, Surface flush_surface) { + if (size == 0) + return; + + const SurfaceInterval flush_interval(addr, addr + size); + SurfaceRegions flushed_intervals; + + for (auto& pair : RangeFromInterval(dirty_regions, flush_interval)) { + // small sizes imply that this most likely comes from the cpu, flush the entire region + // the point is to avoid thousands of small writes every frame if the cpu decides to access + // that region, anything higher than 8 you're guaranteed it comes from a service + const auto interval = size <= 8 ? pair.first : pair.first & flush_interval; + auto& surface = pair.second; + + if (flush_surface != nullptr && surface != flush_surface) + continue; + + // Sanity check, this surface is the last one that marked this region dirty + ASSERT(surface->IsRegionValid(interval)); + + if (surface->type != SurfaceType::Fill) { + SurfaceParams params = surface->FromInterval(interval); + surface->DownloadGLTexture(surface->GetSubRect(params), read_framebuffer.handle, + draw_framebuffer.handle); + } + surface->FlushGLBuffer(boost::icl::first(interval), boost::icl::last_next(interval)); + flushed_intervals += interval; + } + // Reset dirty regions + dirty_regions -= flushed_intervals; +} + +void RasterizerCacheOpenGL::FlushAll() { + FlushRegion(0, Kernel::VMManager::MAX_ADDRESS); +} + +void RasterizerCacheOpenGL::InvalidateRegion(VAddr addr, u64 size, const Surface& region_owner) { + if (size == 0) + return; + + const SurfaceInterval invalid_interval(addr, addr + size); + + if (region_owner != nullptr) { + ASSERT(region_owner->type != SurfaceType::Texture); + ASSERT(addr >= region_owner->addr && addr + size <= region_owner->end); + // Surfaces can't have a gap + ASSERT(region_owner->width == region_owner->stride); + region_owner->invalid_regions.erase(invalid_interval); + } + + for (auto& pair : RangeFromInterval(surface_cache, invalid_interval)) { + for (auto& cached_surface : pair.second) { + if (cached_surface == region_owner) + continue; + + // If cpu is invalidating this region we want to remove it + // to (likely) mark the memory pages as uncached + if (region_owner == nullptr && size <= 8) { + FlushRegion(cached_surface->addr, cached_surface->size, cached_surface); + remove_surfaces.emplace(cached_surface); + continue; + } + + const auto interval = cached_surface->GetInterval() & invalid_interval; + cached_surface->invalid_regions.insert(interval); + + // Remove only "empty" fill surfaces to avoid destroying and recreating OGL textures + if (cached_surface->type == SurfaceType::Fill && + cached_surface->IsSurfaceFullyInvalid()) { + remove_surfaces.emplace(cached_surface); + } + } + } + + if (region_owner != nullptr) + dirty_regions.set({invalid_interval, region_owner}); + else + dirty_regions.erase(invalid_interval); + + for (auto& remove_surface : remove_surfaces) { + if (remove_surface == region_owner) { + Surface expanded_surface = FindMatch<MatchFlags::SubRect | MatchFlags::Invalid>( + surface_cache, *region_owner, ScaleMatch::Ignore); + ASSERT(expanded_surface); + + if ((region_owner->invalid_regions - expanded_surface->invalid_regions).empty()) { + DuplicateSurface(region_owner, expanded_surface); + } else { + continue; + } + } + UnregisterSurface(remove_surface); + } + + remove_surfaces.clear(); +} + +Surface RasterizerCacheOpenGL::CreateSurface(const SurfaceParams& params) { + Surface surface = std::make_shared<CachedSurface>(); + static_cast<SurfaceParams&>(*surface) = params; + + surface->texture.Create(); + + surface->gl_buffer_size = 0; + surface->invalid_regions.insert(surface->GetInterval()); + AllocateSurfaceTexture(surface->texture.handle, GetFormatTuple(surface->pixel_format), + surface->GetScaledWidth(), surface->GetScaledHeight()); + + return surface; +} + +void RasterizerCacheOpenGL::RegisterSurface(const Surface& surface) { + if (surface->registered) { + return; + } + surface->registered = true; + surface_cache.add({surface->GetInterval(), SurfaceSet{surface}}); + UpdatePagesCachedCount(surface->addr, surface->size, 1); +} + +void RasterizerCacheOpenGL::UnregisterSurface(const Surface& surface) { + if (!surface->registered) { + return; + } + surface->registered = false; + UpdatePagesCachedCount(surface->addr, surface->size, -1); + surface_cache.subtract({surface->GetInterval(), SurfaceSet{surface}}); +} + +void RasterizerCacheOpenGL::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) { + const u64 num_pages = + ((addr + size - 1) >> Memory::PAGE_BITS) - (addr >> Memory::PAGE_BITS) + 1; + const u64 page_start = addr >> Memory::PAGE_BITS; + const u64 page_end = page_start + num_pages; + + // Interval maps will erase segments if count reaches 0, so if delta is negative we have to + // subtract after iterating + const auto pages_interval = PageMap::interval_type::right_open(page_start, page_end); + if (delta > 0) + cached_pages.add({pages_interval, delta}); + + for (const auto& pair : RangeFromInterval(cached_pages, pages_interval)) { + const auto interval = pair.first & pages_interval; + const int count = pair.second; + + const VAddr interval_start_addr = boost::icl::first(interval) << Memory::PAGE_BITS; + const VAddr interval_end_addr = boost::icl::last_next(interval) << Memory::PAGE_BITS; + const u64 interval_size = interval_end_addr - interval_start_addr; + + if (delta > 0 && count == delta) + Memory::RasterizerMarkRegionCached(interval_start_addr, interval_size, true); + else if (delta < 0 && count == -delta) + Memory::RasterizerMarkRegionCached(interval_start_addr, interval_size, false); + else + ASSERT(count >= 0); + } + + if (delta < 0) + cached_pages.add({pages_interval, delta}); +} diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h new file mode 100644 index 000000000..1f660d30c --- /dev/null +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -0,0 +1,369 @@ +// Copyright 2015 Citra 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 <tuple> +#ifdef __GNUC__ +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wunused-local-typedefs" +#endif +#include <boost/icl/interval_map.hpp> +#include <boost/icl/interval_set.hpp> +#ifdef __GNUC__ +#pragma GCC diagnostic pop +#endif +#include <glad/glad.h> +#include "common/assert.h" +#include "common/common_funcs.h" +#include "common/common_types.h" +#include "common/math_util.h" +#include "video_core/gpu.h" +#include "video_core/renderer_opengl/gl_resource_manager.h" + +struct CachedSurface; +using Surface = std::shared_ptr<CachedSurface>; +using SurfaceSet = std::set<Surface>; + +using SurfaceRegions = boost::icl::interval_set<VAddr>; +using SurfaceMap = boost::icl::interval_map<VAddr, Surface>; +using SurfaceCache = boost::icl::interval_map<VAddr, SurfaceSet>; + +using SurfaceInterval = SurfaceCache::interval_type; +static_assert(std::is_same<SurfaceRegions::interval_type, SurfaceCache::interval_type>() && + std::is_same<SurfaceMap::interval_type, SurfaceCache::interval_type>(), + "incorrect interval types"); + +using SurfaceRect_Tuple = std::tuple<Surface, MathUtil::Rectangle<u32>>; +using SurfaceSurfaceRect_Tuple = std::tuple<Surface, Surface, MathUtil::Rectangle<u32>>; + +using PageMap = boost::icl::interval_map<u64, int>; + +enum class ScaleMatch { + Exact, // only accept same res scale + Upscale, // only allow higher scale than params + Ignore // accept every scaled res +}; + +struct SurfaceParams { + enum class PixelFormat { + // First 5 formats are shared between textures and color buffers + RGBA8 = 0, + RGB8 = 1, + RGB5A1 = 2, + RGB565 = 3, + RGBA4 = 4, + + // Texture-only formats + IA8 = 5, + RG8 = 6, + I8 = 7, + A8 = 8, + IA4 = 9, + I4 = 10, + A4 = 11, + ETC1 = 12, + ETC1A4 = 13, + + // Depth buffer-only formats + D16 = 14, + // gap + D24 = 16, + D24S8 = 17, + + Invalid = 255, + }; + + enum class SurfaceType { + Color = 0, + Texture = 1, + Depth = 2, + DepthStencil = 3, + Fill = 4, + Invalid = 5 + }; + + static constexpr unsigned int GetFormatBpp(PixelFormat format) { + constexpr std::array<unsigned int, 18> bpp_table = { + 32, // RGBA8 + 24, // RGB8 + 16, // RGB5A1 + 16, // RGB565 + 16, // RGBA4 + 16, // IA8 + 16, // RG8 + 8, // I8 + 8, // A8 + 8, // IA4 + 4, // I4 + 4, // A4 + 4, // ETC1 + 8, // ETC1A4 + 16, // D16 + 0, + 24, // D24 + 32, // D24S8 + }; + + assert(static_cast<size_t>(format) < bpp_table.size()); + return bpp_table[static_cast<size_t>(format)]; + } + unsigned int GetFormatBpp() const { + return GetFormatBpp(pixel_format); + } + + static PixelFormat PixelFormatFromRenderTargetFormat(Tegra::RenderTargetFormat format) { + switch (format) { + case Tegra::RenderTargetFormat::RGBA8_UNORM: + return PixelFormat::RGBA8; + default: + UNREACHABLE(); + } + } + + static PixelFormat PixelFormatFromGPUPixelFormat(Tegra::FramebufferConfig::PixelFormat format) { + switch (format) { + case Tegra::FramebufferConfig::PixelFormat::ABGR8: + return PixelFormat::RGBA8; + default: + UNREACHABLE(); + } + } + + static bool CheckFormatsBlittable(PixelFormat pixel_format_a, PixelFormat pixel_format_b) { + SurfaceType a_type = GetFormatType(pixel_format_a); + SurfaceType b_type = GetFormatType(pixel_format_b); + + if ((a_type == SurfaceType::Color || a_type == SurfaceType::Texture) && + (b_type == SurfaceType::Color || b_type == SurfaceType::Texture)) { + return true; + } + + if (a_type == SurfaceType::Depth && b_type == SurfaceType::Depth) { + return true; + } + + if (a_type == SurfaceType::DepthStencil && b_type == SurfaceType::DepthStencil) { + return true; + } + + return false; + } + + static constexpr SurfaceType GetFormatType(PixelFormat pixel_format) { + if ((unsigned int)pixel_format < 5) { + return SurfaceType::Color; + } + + if ((unsigned int)pixel_format < 14) { + return SurfaceType::Texture; + } + + if (pixel_format == PixelFormat::D16 || pixel_format == PixelFormat::D24) { + return SurfaceType::Depth; + } + + if (pixel_format == PixelFormat::D24S8) { + return SurfaceType::DepthStencil; + } + + return SurfaceType::Invalid; + } + + /// Update the params "size", "end" and "type" from the already set "addr", "width", "height" + /// and "pixel_format" + void UpdateParams() { + if (stride == 0) { + stride = width; + } + type = GetFormatType(pixel_format); + size = !is_tiled ? BytesInPixels(stride * (height - 1) + width) + : BytesInPixels(stride * 8 * (height / 8 - 1) + width * 8); + end = addr + size; + } + + SurfaceInterval GetInterval() const { + return SurfaceInterval::right_open(addr, end); + } + + // Returns the outer rectangle containing "interval" + SurfaceParams FromInterval(SurfaceInterval interval) const; + + SurfaceInterval GetSubRectInterval(MathUtil::Rectangle<u32> unscaled_rect) const; + + // Returns the region of the biggest valid rectange within interval + SurfaceInterval GetCopyableInterval(const Surface& src_surface) const; + + u32 GetScaledWidth() const { + return width * res_scale; + } + + u32 GetScaledHeight() const { + return height * res_scale; + } + + MathUtil::Rectangle<u32> GetRect() const { + return {0, height, width, 0}; + } + + MathUtil::Rectangle<u32> GetScaledRect() const { + return {0, GetScaledHeight(), GetScaledWidth(), 0}; + } + + u64 PixelsInBytes(u64 size) const { + return size * CHAR_BIT / GetFormatBpp(pixel_format); + } + + u64 BytesInPixels(u64 pixels) const { + return pixels * GetFormatBpp(pixel_format) / CHAR_BIT; + } + + bool ExactMatch(const SurfaceParams& other_surface) const; + bool CanSubRect(const SurfaceParams& sub_surface) const; + bool CanExpand(const SurfaceParams& expanded_surface) const; + bool CanTexCopy(const SurfaceParams& texcopy_params) const; + + MathUtil::Rectangle<u32> GetSubRect(const SurfaceParams& sub_surface) const; + MathUtil::Rectangle<u32> GetScaledSubRect(const SurfaceParams& sub_surface) const; + + VAddr addr = 0; + VAddr end = 0; + u64 size = 0; + + u32 width = 0; + u32 height = 0; + u32 stride = 0; + u16 res_scale = 1; + + bool is_tiled = false; + PixelFormat pixel_format = PixelFormat::Invalid; + SurfaceType type = SurfaceType::Invalid; +}; + +struct CachedSurface : SurfaceParams { + bool CanFill(const SurfaceParams& dest_surface, SurfaceInterval fill_interval) const; + bool CanCopy(const SurfaceParams& dest_surface, SurfaceInterval copy_interval) const; + + bool IsRegionValid(SurfaceInterval interval) const { + return (invalid_regions.find(interval) == invalid_regions.end()); + } + + bool IsSurfaceFullyInvalid() const { + return (invalid_regions & GetInterval()) == SurfaceRegions(GetInterval()); + } + + bool registered = false; + SurfaceRegions invalid_regions; + + u64 fill_size = 0; /// Number of bytes to read from fill_data + std::array<u8, 4> fill_data; + + OGLTexture texture; + + static constexpr unsigned int GetGLBytesPerPixel(PixelFormat format) { + // OpenGL needs 4 bpp alignment for D24 since using GL_UNSIGNED_INT as type + return format == PixelFormat::Invalid + ? 0 + : (format == PixelFormat::D24 || GetFormatType(format) == SurfaceType::Texture) + ? 4 + : SurfaceParams::GetFormatBpp(format) / 8; + } + + std::unique_ptr<u8[]> gl_buffer; + size_t gl_buffer_size = 0; + + // Read/Write data in Switch memory to/from gl_buffer + void LoadGLBuffer(VAddr load_start, VAddr load_end); + void FlushGLBuffer(VAddr flush_start, VAddr flush_end); + + // Upload/Download data in gl_buffer in/to this surface's texture + void UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle, + GLuint draw_fb_handle); + void DownloadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle, + GLuint draw_fb_handle); +}; + +class RasterizerCacheOpenGL : NonCopyable { +public: + RasterizerCacheOpenGL(); + ~RasterizerCacheOpenGL(); + + /// Blit one surface's texture to another + bool BlitSurfaces(const Surface& src_surface, const MathUtil::Rectangle<u32>& src_rect, + const Surface& dst_surface, const MathUtil::Rectangle<u32>& dst_rect); + + void ConvertD24S8toABGR(GLuint src_tex, const MathUtil::Rectangle<u32>& src_rect, + GLuint dst_tex, const MathUtil::Rectangle<u32>& dst_rect); + + /// Copy one surface's region to another + void CopySurface(const Surface& src_surface, const Surface& dst_surface, + SurfaceInterval copy_interval); + + /// Load a texture from 3DS memory to OpenGL and cache it (if not already cached) + Surface GetSurface(const SurfaceParams& params, ScaleMatch match_res_scale, + bool load_if_create); + + /// Attempt to find a subrect (resolution scaled) of a surface, otherwise loads a texture from + /// 3DS memory to OpenGL and caches it (if not already cached) + SurfaceRect_Tuple GetSurfaceSubRect(const SurfaceParams& params, ScaleMatch match_res_scale, + bool load_if_create); + + /// Get a surface based on the texture configuration + Surface GetTextureSurface(const void* config); + + /// Get the color and depth surfaces based on the framebuffer configuration + SurfaceSurfaceRect_Tuple GetFramebufferSurfaces(bool using_color_fb, bool using_depth_fb, + const MathUtil::Rectangle<s32>& viewport); + + /// Get a surface that matches the fill config + Surface GetFillSurface(const void* config); + + /// Get a surface that matches a "texture copy" display transfer config + SurfaceRect_Tuple GetTexCopySurface(const SurfaceParams& params); + + /// Write any cached resources overlapping the region back to memory (if dirty) + void FlushRegion(VAddr addr, u64 size, Surface flush_surface = nullptr); + + /// Mark region as being invalidated by region_owner (nullptr if 3DS memory) + void InvalidateRegion(VAddr addr, u64 size, const Surface& region_owner); + + /// Flush all cached resources tracked by this cache manager + void FlushAll(); + +private: + void DuplicateSurface(const Surface& src_surface, const Surface& dest_surface); + + /// Update surface's texture for given region when necessary + void ValidateSurface(const Surface& surface, VAddr addr, u64 size); + + /// Create a new surface + Surface CreateSurface(const SurfaceParams& params); + + /// Register surface into the cache + void RegisterSurface(const Surface& surface); + + /// Remove surface from the cache + void UnregisterSurface(const Surface& surface); + + /// Increase/decrease the number of surface in pages touching the specified region + void UpdatePagesCachedCount(VAddr addr, u64 size, int delta); + + SurfaceCache surface_cache; + PageMap cached_pages; + SurfaceMap dirty_regions; + SurfaceSet remove_surfaces; + + OGLFramebuffer read_framebuffer; + OGLFramebuffer draw_framebuffer; + + OGLVertexArray attributeless_vao; + OGLBuffer d24s8_abgr_buffer; + GLsizeiptr d24s8_abgr_buffer_size; + OGLShader d24s8_abgr_shader; + GLint d24s8_abgr_tbo_size_u_id; + GLint d24s8_abgr_viewport_u_id; +}; diff --git a/src/video_core/renderer_opengl/gl_resource_manager.h b/src/video_core/renderer_opengl/gl_resource_manager.h index 13301ec9f..7da5e74d1 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.h +++ b/src/video_core/renderer_opengl/gl_resource_manager.h @@ -36,7 +36,7 @@ public: if (handle == 0) return; glDeleteTextures(1, &handle); - OpenGLState::ResetTexture(handle); + OpenGLState::GetCurState().ResetTexture(handle).Apply(); handle = 0; } @@ -69,7 +69,7 @@ public: if (handle == 0) return; glDeleteSamplers(1, &handle); - OpenGLState::ResetSampler(handle); + OpenGLState::GetCurState().ResetSampler(handle).Apply(); handle = 0; } @@ -91,10 +91,13 @@ public: } /// Creates a new internal OpenGL resource and stores the handle - void Create(const char* vert_shader, const char* frag_shader) { + void Create(const char* vert_shader, const char* geo_shader, const char* frag_shader, + const std::vector<const char*>& feedback_vars = {}, + bool separable_program = false) { if (handle != 0) return; - handle = GLShader::LoadProgram(vert_shader, frag_shader); + handle = GLShader::LoadProgram(vert_shader, geo_shader, frag_shader, feedback_vars, + separable_program); } /// Deletes the internal OpenGL resource @@ -102,7 +105,40 @@ public: if (handle == 0) return; glDeleteProgram(handle); - OpenGLState::ResetProgram(handle); + OpenGLState::GetCurState().ResetProgram(handle).Apply(); + handle = 0; + } + + GLuint handle = 0; +}; + +class OGLPipeline : private NonCopyable { +public: + OGLPipeline() = default; + OGLPipeline(OGLPipeline&& o) { + handle = std::exchange<GLuint>(o.handle, 0); + } + ~OGLPipeline() { + Release(); + } + OGLPipeline& operator=(OGLPipeline&& o) { + handle = std::exchange<GLuint>(o.handle, 0); + return *this; + } + + /// Creates a new internal OpenGL resource and stores the handle + void Create() { + if (handle != 0) + return; + glGenProgramPipelines(1, &handle); + } + + /// Deletes the internal OpenGL resource + void Release() { + if (handle == 0) + return; + glDeleteProgramPipelines(1, &handle); + OpenGLState::GetCurState().ResetPipeline(handle).Apply(); handle = 0; } @@ -135,13 +171,46 @@ public: if (handle == 0) return; glDeleteBuffers(1, &handle); - OpenGLState::ResetBuffer(handle); + OpenGLState::GetCurState().ResetBuffer(handle).Apply(); handle = 0; } GLuint handle = 0; }; +class OGLSync : private NonCopyable { +public: + OGLSync() = default; + + OGLSync(OGLSync&& o) : handle(std::exchange(o.handle, nullptr)) {} + + ~OGLSync() { + Release(); + } + OGLSync& operator=(OGLSync&& o) { + Release(); + handle = std::exchange(o.handle, nullptr); + return *this; + } + + /// Creates a new internal OpenGL resource and stores the handle + void Create() { + if (handle != 0) + return; + handle = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); + } + + /// Deletes the internal OpenGL resource + void Release() { + if (handle == 0) + return; + glDeleteSync(handle); + handle = 0; + } + + GLsync handle = 0; +}; + class OGLVertexArray : private NonCopyable { public: OGLVertexArray() = default; @@ -168,7 +237,7 @@ public: if (handle == 0) return; glDeleteVertexArrays(1, &handle); - OpenGLState::ResetVertexArray(handle); + OpenGLState::GetCurState().ResetVertexArray(handle).Apply(); handle = 0; } @@ -201,7 +270,7 @@ public: if (handle == 0) return; glDeleteFramebuffers(1, &handle); - OpenGLState::ResetFramebuffer(handle); + OpenGLState::GetCurState().ResetFramebuffer(handle).Apply(); handle = 0; } diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp new file mode 100644 index 000000000..564ea8f9e --- /dev/null +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -0,0 +1,58 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <string> +#include <queue> +#include "common/assert.h" +#include "common/common_types.h" +#include "video_core/renderer_opengl/gl_shader_decompiler.h" + +namespace Maxwell3D { +namespace Shader { +namespace Decompiler { + +constexpr u32 PROGRAM_END = MAX_PROGRAM_CODE_LENGTH; + +class Impl { +public: + Impl(const std::array<u32, MAX_PROGRAM_CODE_LENGTH>& program_code, + const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>& swizzle_data, u32 main_offset, + const std::function<std::string(u32)>& inputreg_getter, + const std::function<std::string(u32)>& outputreg_getter, bool sanitize_mul, + const std::string& emit_cb, const std::string& setemit_cb) + : program_code(program_code), swizzle_data(swizzle_data), main_offset(main_offset), + inputreg_getter(inputreg_getter), outputreg_getter(outputreg_getter), + sanitize_mul(sanitize_mul), emit_cb(emit_cb), setemit_cb(setemit_cb) {} + + std::string Decompile() { + UNREACHABLE(); + return {}; + } + +private: + const std::array<u32, MAX_PROGRAM_CODE_LENGTH>& program_code; + const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>& swizzle_data; + u32 main_offset; + const std::function<std::string(u32)>& inputreg_getter; + const std::function<std::string(u32)>& outputreg_getter; + bool sanitize_mul; + const std::string& emit_cb; + const std::string& setemit_cb; +}; + +std::string DecompileProgram(const std::array<u32, MAX_PROGRAM_CODE_LENGTH>& program_code, + const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>& swizzle_data, + u32 main_offset, + const std::function<std::string(u32)>& inputreg_getter, + const std::function<std::string(u32)>& outputreg_getter, + bool sanitize_mul, const std::string& emit_cb, + const std::string& setemit_cb) { + Impl impl(program_code, swizzle_data, main_offset, inputreg_getter, outputreg_getter, + sanitize_mul, emit_cb, setemit_cb); + return impl.Decompile(); +} + +} // namespace Decompiler +} // namespace Shader +} // namespace Maxwell3D diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h new file mode 100644 index 000000000..02ebfcbe8 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h @@ -0,0 +1,27 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <array> +#include <functional> +#include <string> +#include "common/common_types.h" + +namespace Maxwell3D { +namespace Shader { +namespace Decompiler { + +constexpr size_t MAX_PROGRAM_CODE_LENGTH{0x100000}; +constexpr size_t MAX_SWIZZLE_DATA_LENGTH{0x100000}; + +std::string DecompileProgram(const std::array<u32, MAX_PROGRAM_CODE_LENGTH>& program_code, + const std::array<u32, MAX_SWIZZLE_DATA_LENGTH>& swizzle_data, + u32 main_offset, + const std::function<std::string(u32)>& inputreg_getter, + const std::function<std::string(u32)>& outputreg_getter, + bool sanitize_mul, const std::string& emit_cb = "", + const std::string& setemit_cb = ""); + +} // namespace Decompiler +} // namespace Shader +} // namespace Maxwell3D diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp new file mode 100644 index 000000000..8f3c98800 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp @@ -0,0 +1,20 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/assert.h" +#include "video_core/renderer_opengl/gl_shader_gen.h" + +namespace GLShader { + +std::string GenerateVertexShader(const MaxwellVSConfig& config) { + UNREACHABLE(); + return {}; +} + +std::string GenerateFragmentShader(const MaxwellFSConfig& config) { + UNREACHABLE(); + return {}; +} + +} // namespace GLShader diff --git a/src/video_core/renderer_opengl/gl_shader_gen.h b/src/video_core/renderer_opengl/gl_shader_gen.h new file mode 100644 index 000000000..5101e7d30 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_shader_gen.h @@ -0,0 +1,66 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <cstring> +#include <string> +#include <type_traits> +#include "common/hash.h" + +namespace GLShader { + +enum Attributes { + ATTRIBUTE_POSITION, + ATTRIBUTE_COLOR, + ATTRIBUTE_TEXCOORD0, + ATTRIBUTE_TEXCOORD1, + ATTRIBUTE_TEXCOORD2, + ATTRIBUTE_TEXCOORD0_W, + ATTRIBUTE_NORMQUAT, + ATTRIBUTE_VIEW, +}; + +struct MaxwellShaderConfigCommon { + explicit MaxwellShaderConfigCommon(){}; +}; + +struct MaxwellVSConfig : MaxwellShaderConfigCommon { + explicit MaxwellVSConfig() : MaxwellShaderConfigCommon() {} + + bool operator==(const MaxwellVSConfig& o) const { + return std::memcmp(this, &o, sizeof(MaxwellVSConfig)) == 0; + }; +}; + +struct MaxwellFSConfig : MaxwellShaderConfigCommon { + explicit MaxwellFSConfig() : MaxwellShaderConfigCommon() {} + + bool operator==(const MaxwellFSConfig& o) const { + return std::memcmp(this, &o, sizeof(MaxwellFSConfig)) == 0; + }; +}; + +std::string GenerateVertexShader(const MaxwellVSConfig& config); +std::string GenerateFragmentShader(const MaxwellFSConfig& config); + +} // namespace GLShader + +namespace std { + +template <> +struct hash<GLShader::MaxwellVSConfig> { + size_t operator()(const GLShader::MaxwellVSConfig& k) const { + return Common::ComputeHash64(&k, sizeof(GLShader::MaxwellVSConfig)); + } +}; + +template <> +struct hash<GLShader::MaxwellFSConfig> { + size_t operator()(const GLShader::MaxwellFSConfig& k) const { + return Common::ComputeHash64(&k, sizeof(GLShader::MaxwellFSConfig)); + } +}; + +} // namespace std diff --git a/src/video_core/renderer_opengl/gl_shader_util.cpp b/src/video_core/renderer_opengl/gl_shader_util.cpp index 4da241d83..a6c6204d5 100644 --- a/src/video_core/renderer_opengl/gl_shader_util.cpp +++ b/src/video_core/renderer_opengl/gl_shader_util.cpp @@ -10,53 +10,85 @@ namespace GLShader { -GLuint LoadProgram(const char* vertex_shader, const char* fragment_shader) { - +GLuint LoadProgram(const char* vertex_shader, const char* geometry_shader, + const char* fragment_shader, const std::vector<const char*>& feedback_vars, + bool separable_program) { // Create the shaders - GLuint vertex_shader_id = glCreateShader(GL_VERTEX_SHADER); - GLuint fragment_shader_id = glCreateShader(GL_FRAGMENT_SHADER); + GLuint vertex_shader_id = vertex_shader ? glCreateShader(GL_VERTEX_SHADER) : 0; + GLuint geometry_shader_id = geometry_shader ? glCreateShader(GL_GEOMETRY_SHADER) : 0; + GLuint fragment_shader_id = fragment_shader ? glCreateShader(GL_FRAGMENT_SHADER) : 0; GLint result = GL_FALSE; int info_log_length; - // Compile Vertex Shader - LOG_DEBUG(Render_OpenGL, "Compiling vertex shader..."); - - glShaderSource(vertex_shader_id, 1, &vertex_shader, nullptr); - glCompileShader(vertex_shader_id); - - // Check Vertex Shader - glGetShaderiv(vertex_shader_id, GL_COMPILE_STATUS, &result); - glGetShaderiv(vertex_shader_id, GL_INFO_LOG_LENGTH, &info_log_length); - - if (info_log_length > 1) { - std::vector<char> vertex_shader_error(info_log_length); - glGetShaderInfoLog(vertex_shader_id, info_log_length, nullptr, &vertex_shader_error[0]); - if (result == GL_TRUE) { - LOG_DEBUG(Render_OpenGL, "%s", &vertex_shader_error[0]); - } else { - LOG_ERROR(Render_OpenGL, "Error compiling vertex shader:\n%s", &vertex_shader_error[0]); + if (vertex_shader) { + // Compile Vertex Shader + LOG_DEBUG(Render_OpenGL, "Compiling vertex shader..."); + + glShaderSource(vertex_shader_id, 1, &vertex_shader, nullptr); + glCompileShader(vertex_shader_id); + + // Check Vertex Shader + glGetShaderiv(vertex_shader_id, GL_COMPILE_STATUS, &result); + glGetShaderiv(vertex_shader_id, GL_INFO_LOG_LENGTH, &info_log_length); + + if (info_log_length > 1) { + std::vector<char> vertex_shader_error(info_log_length); + glGetShaderInfoLog(vertex_shader_id, info_log_length, nullptr, &vertex_shader_error[0]); + if (result == GL_TRUE) { + LOG_DEBUG(Render_OpenGL, "%s", &vertex_shader_error[0]); + } else { + LOG_CRITICAL(Render_OpenGL, "Error compiling vertex shader:\n%s", + &vertex_shader_error[0]); + } } } - // Compile Fragment Shader - LOG_DEBUG(Render_OpenGL, "Compiling fragment shader..."); - - glShaderSource(fragment_shader_id, 1, &fragment_shader, nullptr); - glCompileShader(fragment_shader_id); - - // Check Fragment Shader - glGetShaderiv(fragment_shader_id, GL_COMPILE_STATUS, &result); - glGetShaderiv(fragment_shader_id, GL_INFO_LOG_LENGTH, &info_log_length); + if (geometry_shader) { + // Compile Geometry Shader + LOG_DEBUG(Render_OpenGL, "Compiling geometry shader..."); + + glShaderSource(geometry_shader_id, 1, &geometry_shader, nullptr); + glCompileShader(geometry_shader_id); + + // Check Geometry Shader + glGetShaderiv(geometry_shader_id, GL_COMPILE_STATUS, &result); + glGetShaderiv(geometry_shader_id, GL_INFO_LOG_LENGTH, &info_log_length); + + if (info_log_length > 1) { + std::vector<char> geometry_shader_error(info_log_length); + glGetShaderInfoLog(geometry_shader_id, info_log_length, nullptr, + &geometry_shader_error[0]); + if (result == GL_TRUE) { + LOG_DEBUG(Render_OpenGL, "%s", &geometry_shader_error[0]); + } else { + LOG_CRITICAL(Render_OpenGL, "Error compiling geometry shader:\n%s", + &geometry_shader_error[0]); + } + } + } - if (info_log_length > 1) { - std::vector<char> fragment_shader_error(info_log_length); - glGetShaderInfoLog(fragment_shader_id, info_log_length, nullptr, &fragment_shader_error[0]); - if (result == GL_TRUE) { - LOG_DEBUG(Render_OpenGL, "%s", &fragment_shader_error[0]); - } else { - LOG_ERROR(Render_OpenGL, "Error compiling fragment shader:\n%s", - &fragment_shader_error[0]); + if (fragment_shader) { + // Compile Fragment Shader + LOG_DEBUG(Render_OpenGL, "Compiling fragment shader..."); + + glShaderSource(fragment_shader_id, 1, &fragment_shader, nullptr); + glCompileShader(fragment_shader_id); + + // Check Fragment Shader + glGetShaderiv(fragment_shader_id, GL_COMPILE_STATUS, &result); + glGetShaderiv(fragment_shader_id, GL_INFO_LOG_LENGTH, &info_log_length); + + if (info_log_length > 1) { + std::vector<char> fragment_shader_error(info_log_length); + glGetShaderInfoLog(fragment_shader_id, info_log_length, nullptr, + &fragment_shader_error[0]); + if (result == GL_TRUE) { + LOG_DEBUG(Render_OpenGL, "%s", &fragment_shader_error[0]); + } else { + LOG_CRITICAL(Render_OpenGL, "Error compiling fragment shader:\n%s", + &fragment_shader_error[0]); + } } } @@ -64,8 +96,25 @@ GLuint LoadProgram(const char* vertex_shader, const char* fragment_shader) { LOG_DEBUG(Render_OpenGL, "Linking program..."); GLuint program_id = glCreateProgram(); - glAttachShader(program_id, vertex_shader_id); - glAttachShader(program_id, fragment_shader_id); + if (vertex_shader) { + glAttachShader(program_id, vertex_shader_id); + } + if (geometry_shader) { + glAttachShader(program_id, geometry_shader_id); + } + if (fragment_shader) { + glAttachShader(program_id, fragment_shader_id); + } + + if (!feedback_vars.empty()) { + auto varyings = feedback_vars; + glTransformFeedbackVaryings(program_id, static_cast<GLsizei>(feedback_vars.size()), + &varyings[0], GL_INTERLEAVED_ATTRIBS); + } + + if (separable_program) { + glProgramParameteri(program_id, GL_PROGRAM_SEPARABLE, GL_TRUE); + } glLinkProgram(program_id); @@ -79,19 +128,36 @@ GLuint LoadProgram(const char* vertex_shader, const char* fragment_shader) { if (result == GL_TRUE) { LOG_DEBUG(Render_OpenGL, "%s", &program_error[0]); } else { - LOG_ERROR(Render_OpenGL, "Error linking shader:\n%s", &program_error[0]); + LOG_CRITICAL(Render_OpenGL, "Error linking shader:\n%s", &program_error[0]); } } // If the program linking failed at least one of the shaders was probably bad if (result == GL_FALSE) { - LOG_ERROR(Render_OpenGL, "Vertex shader:\n%s", vertex_shader); - LOG_ERROR(Render_OpenGL, "Fragment shader:\n%s", fragment_shader); + if (vertex_shader) { + LOG_CRITICAL(Render_OpenGL, "Vertex shader:\n%s", vertex_shader); + } + if (geometry_shader) { + LOG_CRITICAL(Render_OpenGL, "Geometry shader:\n%s", geometry_shader); + } + if (fragment_shader) { + LOG_CRITICAL(Render_OpenGL, "Fragment shader:\n%s", fragment_shader); + } } ASSERT_MSG(result == GL_TRUE, "Shader not linked"); - glDeleteShader(vertex_shader_id); - glDeleteShader(fragment_shader_id); + if (vertex_shader) { + glDetachShader(program_id, vertex_shader_id); + glDeleteShader(vertex_shader_id); + } + if (geometry_shader) { + glDetachShader(program_id, geometry_shader_id); + glDeleteShader(geometry_shader_id); + } + if (fragment_shader) { + glDetachShader(program_id, fragment_shader_id); + glDeleteShader(fragment_shader_id); + } return program_id; } diff --git a/src/video_core/renderer_opengl/gl_shader_util.h b/src/video_core/renderer_opengl/gl_shader_util.h index a4bcffdfa..fc7b5e080 100644 --- a/src/video_core/renderer_opengl/gl_shader_util.h +++ b/src/video_core/renderer_opengl/gl_shader_util.h @@ -4,6 +4,7 @@ #pragma once +#include <vector> #include <glad/glad.h> namespace GLShader { @@ -11,9 +12,12 @@ namespace GLShader { /** * Utility function to create and compile an OpenGL GLSL shader program (vertex + fragment shader) * @param vertex_shader String of the GLSL vertex shader program + * @param geometry_shader String of the GLSL geometry shader program * @param fragment_shader String of the GLSL fragment shader program * @returns Handle of the newly created OpenGL shader object */ -GLuint LoadProgram(const char* vertex_shader, const char* fragment_shader); +GLuint LoadProgram(const char* vertex_shader, const char* geometry_shader, + const char* fragment_shader, const std::vector<const char*>& feedback_vars = {}, + bool separable_program = false); } // namespace GLShader diff --git a/src/video_core/renderer_opengl/gl_state.cpp b/src/video_core/renderer_opengl/gl_state.cpp index 5770ae08f..1d396728b 100644 --- a/src/video_core/renderer_opengl/gl_state.cpp +++ b/src/video_core/renderer_opengl/gl_state.cpp @@ -33,7 +33,7 @@ OpenGLState::OpenGLState() { stencil.action_depth_pass = GL_KEEP; stencil.action_stencil_fail = GL_KEEP; - blend.enabled = false; + blend.enabled = true; blend.rgb_equation = GL_FUNC_ADD; blend.a_equation = GL_FUNC_ADD; blend.src_rgb_func = GL_ONE; @@ -68,6 +68,18 @@ OpenGLState::OpenGLState() { draw.vertex_buffer = 0; draw.uniform_buffer = 0; draw.shader_program = 0; + draw.program_pipeline = 0; + + scissor.enabled = false; + scissor.x = 0; + scissor.y = 0; + scissor.width = 0; + scissor.height = 0; + + viewport.x = 0; + viewport.y = 0; + viewport.width = 0; + viewport.height = 0; clip_distance = {}; } @@ -148,9 +160,6 @@ void OpenGLState::Apply() const { if (blend.enabled != cur_state.blend.enabled) { if (blend.enabled) { glEnable(GL_BLEND); - - cur_state.logic_op = GL_COPY; - glLogicOp(cur_state.logic_op); glDisable(GL_COLOR_LOGIC_OP); } else { glDisable(GL_BLEND); @@ -196,7 +205,7 @@ void OpenGLState::Apply() const { // Lighting LUTs if (lighting_lut.texture_buffer != cur_state.lighting_lut.texture_buffer) { glActiveTexture(TextureUnits::LightingLUT.Enum()); - glBindTexture(GL_TEXTURE_BUFFER, cur_state.lighting_lut.texture_buffer); + glBindTexture(GL_TEXTURE_BUFFER, lighting_lut.texture_buffer); } // Fog LUT @@ -263,6 +272,31 @@ void OpenGLState::Apply() const { glUseProgram(draw.shader_program); } + // Program pipeline + if (draw.program_pipeline != cur_state.draw.program_pipeline) { + glBindProgramPipeline(draw.program_pipeline); + } + + // Scissor test + if (scissor.enabled != cur_state.scissor.enabled) { + if (scissor.enabled) { + glEnable(GL_SCISSOR_TEST); + } else { + glDisable(GL_SCISSOR_TEST); + } + } + + if (scissor.x != cur_state.scissor.x || scissor.y != cur_state.scissor.y || + scissor.width != cur_state.scissor.width || scissor.height != cur_state.scissor.height) { + glScissor(scissor.x, scissor.y, scissor.width, scissor.height); + } + + if (viewport.x != cur_state.viewport.x || viewport.y != cur_state.viewport.y || + viewport.width != cur_state.viewport.width || + viewport.height != cur_state.viewport.height) { + glViewport(viewport.x, viewport.y, viewport.width, viewport.height); + } + // Clip distance for (size_t i = 0; i < clip_distance.size(); ++i) { if (clip_distance[i] != cur_state.clip_distance[i]) { @@ -277,62 +311,75 @@ void OpenGLState::Apply() const { cur_state = *this; } -void OpenGLState::ResetTexture(GLuint handle) { - for (auto& unit : cur_state.texture_units) { +OpenGLState& OpenGLState::ResetTexture(GLuint handle) { + for (auto& unit : texture_units) { if (unit.texture_2d == handle) { unit.texture_2d = 0; } } - if (cur_state.lighting_lut.texture_buffer == handle) - cur_state.lighting_lut.texture_buffer = 0; - if (cur_state.fog_lut.texture_buffer == handle) - cur_state.fog_lut.texture_buffer = 0; - if (cur_state.proctex_noise_lut.texture_buffer == handle) - cur_state.proctex_noise_lut.texture_buffer = 0; - if (cur_state.proctex_color_map.texture_buffer == handle) - cur_state.proctex_color_map.texture_buffer = 0; - if (cur_state.proctex_alpha_map.texture_buffer == handle) - cur_state.proctex_alpha_map.texture_buffer = 0; - if (cur_state.proctex_lut.texture_buffer == handle) - cur_state.proctex_lut.texture_buffer = 0; - if (cur_state.proctex_diff_lut.texture_buffer == handle) - cur_state.proctex_diff_lut.texture_buffer = 0; + if (lighting_lut.texture_buffer == handle) + lighting_lut.texture_buffer = 0; + if (fog_lut.texture_buffer == handle) + fog_lut.texture_buffer = 0; + if (proctex_noise_lut.texture_buffer == handle) + proctex_noise_lut.texture_buffer = 0; + if (proctex_color_map.texture_buffer == handle) + proctex_color_map.texture_buffer = 0; + if (proctex_alpha_map.texture_buffer == handle) + proctex_alpha_map.texture_buffer = 0; + if (proctex_lut.texture_buffer == handle) + proctex_lut.texture_buffer = 0; + if (proctex_diff_lut.texture_buffer == handle) + proctex_diff_lut.texture_buffer = 0; + return *this; } -void OpenGLState::ResetSampler(GLuint handle) { - for (auto& unit : cur_state.texture_units) { +OpenGLState& OpenGLState::ResetSampler(GLuint handle) { + for (auto& unit : texture_units) { if (unit.sampler == handle) { unit.sampler = 0; } } + return *this; +} + +OpenGLState& OpenGLState::ResetProgram(GLuint handle) { + if (draw.shader_program == handle) { + draw.shader_program = 0; + } + return *this; } -void OpenGLState::ResetProgram(GLuint handle) { - if (cur_state.draw.shader_program == handle) { - cur_state.draw.shader_program = 0; +OpenGLState& OpenGLState::ResetPipeline(GLuint handle) { + if (draw.program_pipeline == handle) { + draw.program_pipeline = 0; } + return *this; } -void OpenGLState::ResetBuffer(GLuint handle) { - if (cur_state.draw.vertex_buffer == handle) { - cur_state.draw.vertex_buffer = 0; +OpenGLState& OpenGLState::ResetBuffer(GLuint handle) { + if (draw.vertex_buffer == handle) { + draw.vertex_buffer = 0; } - if (cur_state.draw.uniform_buffer == handle) { - cur_state.draw.uniform_buffer = 0; + if (draw.uniform_buffer == handle) { + draw.uniform_buffer = 0; } + return *this; } -void OpenGLState::ResetVertexArray(GLuint handle) { - if (cur_state.draw.vertex_array == handle) { - cur_state.draw.vertex_array = 0; +OpenGLState& OpenGLState::ResetVertexArray(GLuint handle) { + if (draw.vertex_array == handle) { + draw.vertex_array = 0; } + return *this; } -void OpenGLState::ResetFramebuffer(GLuint handle) { - if (cur_state.draw.read_framebuffer == handle) { - cur_state.draw.read_framebuffer = 0; +OpenGLState& OpenGLState::ResetFramebuffer(GLuint handle) { + if (draw.read_framebuffer == handle) { + draw.read_framebuffer = 0; } - if (cur_state.draw.draw_framebuffer == handle) { - cur_state.draw.draw_framebuffer = 0; + if (draw.draw_framebuffer == handle) { + draw.draw_framebuffer = 0; } + return *this; } diff --git a/src/video_core/renderer_opengl/gl_state.h b/src/video_core/renderer_opengl/gl_state.h index 437fe34c4..c1f4efc8c 100644 --- a/src/video_core/renderer_opengl/gl_state.h +++ b/src/video_core/renderer_opengl/gl_state.h @@ -85,7 +85,7 @@ public: struct { GLuint texture_2d; // GL_TEXTURE_BINDING_2D GLuint sampler; // GL_SAMPLER_BINDING - } texture_units[3]; + } texture_units[32]; struct { GLuint texture_buffer; // GL_TEXTURE_BINDING_BUFFER @@ -122,27 +122,44 @@ public: GLuint vertex_buffer; // GL_ARRAY_BUFFER_BINDING GLuint uniform_buffer; // GL_UNIFORM_BUFFER_BINDING GLuint shader_program; // GL_CURRENT_PROGRAM + GLuint program_pipeline; // GL_PROGRAM_PIPELINE_BINDING } draw; + struct { + bool enabled; // GL_SCISSOR_TEST + GLint x; + GLint y; + GLsizei width; + GLsizei height; + } scissor; + + struct { + GLint x; + GLint y; + GLsizei width; + GLsizei height; + } viewport; + std::array<bool, 2> clip_distance; // GL_CLIP_DISTANCE OpenGLState(); /// Get the currently active OpenGL state - static const OpenGLState& GetCurState() { + static OpenGLState GetCurState() { return cur_state; } /// Apply this state as the current OpenGL state void Apply() const; - /// Resets and unbinds any references to the given resource in the current OpenGL state - static void ResetTexture(GLuint handle); - static void ResetSampler(GLuint handle); - static void ResetProgram(GLuint handle); - static void ResetBuffer(GLuint handle); - static void ResetVertexArray(GLuint handle); - static void ResetFramebuffer(GLuint handle); + /// Resets any references to the given resource + OpenGLState& ResetTexture(GLuint handle); + OpenGLState& ResetSampler(GLuint handle); + OpenGLState& ResetProgram(GLuint handle); + OpenGLState& ResetPipeline(GLuint handle); + OpenGLState& ResetBuffer(GLuint handle); + OpenGLState& ResetVertexArray(GLuint handle); + OpenGLState& ResetFramebuffer(GLuint handle); private: static OpenGLState cur_state; diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.cpp b/src/video_core/renderer_opengl/gl_stream_buffer.cpp new file mode 100644 index 000000000..a2713e9f0 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_stream_buffer.cpp @@ -0,0 +1,182 @@ +// Copyright 2018 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <deque> +#include <vector> +#include "common/alignment.h" +#include "common/assert.h" +#include "video_core/renderer_opengl/gl_state.h" +#include "video_core/renderer_opengl/gl_stream_buffer.h" + +class OrphanBuffer : public OGLStreamBuffer { +public: + explicit OrphanBuffer(GLenum target) : OGLStreamBuffer(target) {} + ~OrphanBuffer() override; + +private: + void Create(size_t size, size_t sync_subdivide) override; + void Release() override; + + std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) override; + void Unmap() override; + + std::vector<u8> data; +}; + +class StorageBuffer : public OGLStreamBuffer { +public: + explicit StorageBuffer(GLenum target) : OGLStreamBuffer(target) {} + ~StorageBuffer() override; + +private: + void Create(size_t size, size_t sync_subdivide) override; + void Release() override; + + std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) override; + void Unmap() override; + + struct Fence { + OGLSync sync; + size_t offset; + }; + std::deque<Fence> head; + std::deque<Fence> tail; + + u8* mapped_ptr; +}; + +OGLStreamBuffer::OGLStreamBuffer(GLenum target) { + gl_target = target; +} + +GLuint OGLStreamBuffer::GetHandle() const { + return gl_buffer.handle; +} + +std::unique_ptr<OGLStreamBuffer> OGLStreamBuffer::MakeBuffer(bool storage_buffer, GLenum target) { + if (storage_buffer) { + return std::make_unique<StorageBuffer>(target); + } + return std::make_unique<OrphanBuffer>(target); +} + +OrphanBuffer::~OrphanBuffer() { + Release(); +} + +void OrphanBuffer::Create(size_t size, size_t /*sync_subdivide*/) { + buffer_pos = 0; + buffer_size = size; + data.resize(buffer_size); + + if (gl_buffer.handle == 0) { + gl_buffer.Create(); + glBindBuffer(gl_target, gl_buffer.handle); + } + + glBufferData(gl_target, static_cast<GLsizeiptr>(buffer_size), nullptr, GL_STREAM_DRAW); +} + +void OrphanBuffer::Release() { + gl_buffer.Release(); +} + +std::pair<u8*, GLintptr> OrphanBuffer::Map(size_t size, size_t alignment) { + buffer_pos = Common::AlignUp(buffer_pos, alignment); + + if (buffer_pos + size > buffer_size) { + Create(std::max(buffer_size, size), 0); + } + + mapped_size = size; + return std::make_pair(&data[buffer_pos], static_cast<GLintptr>(buffer_pos)); +} + +void OrphanBuffer::Unmap() { + glBufferSubData(gl_target, static_cast<GLintptr>(buffer_pos), + static_cast<GLsizeiptr>(mapped_size), &data[buffer_pos]); + buffer_pos += mapped_size; +} + +StorageBuffer::~StorageBuffer() { + Release(); +} + +void StorageBuffer::Create(size_t size, size_t sync_subdivide) { + if (gl_buffer.handle != 0) + return; + + buffer_pos = 0; + buffer_size = size; + buffer_sync_subdivide = std::max<size_t>(sync_subdivide, 1); + + gl_buffer.Create(); + glBindBuffer(gl_target, gl_buffer.handle); + + glBufferStorage(gl_target, static_cast<GLsizeiptr>(buffer_size), nullptr, + GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT); + mapped_ptr = reinterpret_cast<u8*>( + glMapBufferRange(gl_target, 0, static_cast<GLsizeiptr>(buffer_size), + GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | GL_MAP_FLUSH_EXPLICIT_BIT)); +} + +void StorageBuffer::Release() { + if (gl_buffer.handle == 0) + return; + + glUnmapBuffer(gl_target); + + gl_buffer.Release(); + head.clear(); + tail.clear(); +} + +std::pair<u8*, GLintptr> StorageBuffer::Map(size_t size, size_t alignment) { + ASSERT(size <= buffer_size); + + OGLSync sync; + + buffer_pos = Common::AlignUp(buffer_pos, alignment); + size_t effective_offset = Common::AlignDown(buffer_pos, buffer_sync_subdivide); + + if (!head.empty() && + (effective_offset > head.back().offset || buffer_pos + size > buffer_size)) { + ASSERT(head.back().sync.handle == 0); + head.back().sync.Create(); + } + + if (buffer_pos + size > buffer_size) { + if (!tail.empty()) { + std::swap(sync, tail.back().sync); + tail.clear(); + } + std::swap(tail, head); + buffer_pos = 0; + effective_offset = 0; + } + + while (!tail.empty() && buffer_pos + size > tail.front().offset) { + std::swap(sync, tail.front().sync); + tail.pop_front(); + } + + if (sync.handle != 0) { + glClientWaitSync(sync.handle, GL_SYNC_FLUSH_COMMANDS_BIT, GL_TIMEOUT_IGNORED); + sync.Release(); + } + + if (head.empty() || effective_offset > head.back().offset) { + head.emplace_back(); + head.back().offset = effective_offset; + } + + mapped_size = size; + return std::make_pair(&mapped_ptr[buffer_pos], static_cast<GLintptr>(buffer_pos)); +} + +void StorageBuffer::Unmap() { + glFlushMappedBufferRange(gl_target, static_cast<GLintptr>(buffer_pos), + static_cast<GLsizeiptr>(mapped_size)); + buffer_pos += mapped_size; +} diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.h b/src/video_core/renderer_opengl/gl_stream_buffer.h new file mode 100644 index 000000000..4bc2f52e0 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_stream_buffer.h @@ -0,0 +1,34 @@ +// Copyright 2018 Citra Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <memory> +#include <glad/glad.h> +#include "common/common_types.h" +#include "video_core/renderer_opengl/gl_resource_manager.h" + +class OGLStreamBuffer : private NonCopyable { +public: + explicit OGLStreamBuffer(GLenum target); + virtual ~OGLStreamBuffer() = default; + +public: + static std::unique_ptr<OGLStreamBuffer> MakeBuffer(bool storage_buffer, GLenum target); + + virtual void Create(size_t size, size_t sync_subdivide) = 0; + virtual void Release() {} + + GLuint GetHandle() const; + + virtual std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) = 0; + virtual void Unmap() = 0; + +protected: + OGLBuffer gl_buffer; + GLenum gl_target; + + size_t buffer_pos = 0; + size_t buffer_size = 0; + size_t buffer_sync_subdivide = 0; + size_t mapped_size = 0; +}; diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h new file mode 100644 index 000000000..d847317ac --- /dev/null +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -0,0 +1,50 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <glad/glad.h> +#include "common/common_types.h" +#include "common/logging/log.h" +#include "video_core/engines/maxwell_3d.h" + +namespace MaxwellToGL { + +using Maxwell = Tegra::Engines::Maxwell3D::Regs; + +inline GLenum VertexType(Maxwell::VertexAttribute attrib) { + switch (attrib.type) { + case Maxwell::VertexAttribute::Type::UnsignedNorm: { + + switch (attrib.size) { + case Maxwell::VertexAttribute::Size::Size_8_8_8_8: + return GL_UNSIGNED_BYTE; + } + + LOG_CRITICAL(Render_OpenGL, "Unimplemented vertex size=%s", attrib.SizeString().c_str()); + UNREACHABLE(); + return {}; + } + + case Maxwell::VertexAttribute::Type::Float: + return GL_FLOAT; + } + + LOG_CRITICAL(Render_OpenGL, "Unimplemented vertex type=%s", attrib.TypeString().c_str()); + UNREACHABLE(); + return {}; +} + +inline GLenum PrimitiveTopology(Maxwell::PrimitiveTopology topology) { + switch (topology) { + case Maxwell::PrimitiveTopology::TriangleStrip: + return GL_TRIANGLE_STRIP; + } + LOG_CRITICAL(Render_OpenGL, "Unimplemented primitive topology=%d", topology); + UNREACHABLE(); + return {}; +} + +} // namespace MaxwellToGL diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index 7f921fa32..78b50b227 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -20,6 +20,7 @@ #include "core/settings.h" #include "core/tracer/recorder.h" #include "video_core/renderer_opengl/renderer_opengl.h" +#include "video_core/utils.h" #include "video_core/video_core.h" static const char vertex_shader[] = R"( @@ -98,197 +99,88 @@ RendererOpenGL::RendererOpenGL() = default; RendererOpenGL::~RendererOpenGL() = default; /// Swap buffers (render frame) -void RendererOpenGL::SwapBuffers(boost::optional<const FramebufferInfo&> framebuffer_info) { +void RendererOpenGL::SwapBuffers(boost::optional<const Tegra::FramebufferConfig&> framebuffer) { + Core::System::GetInstance().perf_stats.EndSystemFrame(); + // Maintain the rasterizer's state as a priority OpenGLState prev_state = OpenGLState::GetCurState(); state.Apply(); - if (framebuffer_info != boost::none) { - // If framebuffer_info is provided, reload it from memory to a texture - if (screen_info.texture.width != (GLsizei)framebuffer_info->width || - screen_info.texture.height != (GLsizei)framebuffer_info->height || - screen_info.texture.pixel_format != framebuffer_info->pixel_format) { + if (framebuffer != boost::none) { + // If framebuffer is provided, reload it from memory to a texture + if (screen_info.texture.width != (GLsizei)framebuffer->width || + screen_info.texture.height != (GLsizei)framebuffer->height || + screen_info.texture.pixel_format != framebuffer->pixel_format) { // Reallocate texture if the framebuffer size has changed. // This is expected to not happen very often and hence should not be a // performance problem. - ConfigureFramebufferTexture(screen_info.texture, *framebuffer_info); + ConfigureFramebufferTexture(screen_info.texture, *framebuffer); } - LoadFBToScreenInfo(*framebuffer_info, screen_info); - } - - DrawScreens(); - Core::System::GetInstance().perf_stats.EndSystemFrame(); + // Load the framebuffer from memory, draw it to the screen, and swap buffers + LoadFBToScreenInfo(*framebuffer, screen_info); + DrawScreen(); + render_window->SwapBuffers(); + } - // Swap buffers render_window->PollEvents(); - render_window->SwapBuffers(); Core::System::GetInstance().frame_limiter.DoFrameLimiting(CoreTiming::GetGlobalTimeUs()); Core::System::GetInstance().perf_stats.BeginSystemFrame(); + // Restore the rasterizer state prev_state.Apply(); RefreshRasterizerSetting(); } -static inline u32 MortonInterleave128(u32 x, u32 y) { - // 128x128 Z-Order coordinate from 2D coordinates - static constexpr u32 xlut[] = { - 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, - 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, - 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, - 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, - 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, - 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, - 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, - 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, - 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, - 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, - 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, - 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, - 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, - 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, - 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, - 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, - 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, - 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, - 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, - 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, - 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, - 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, - 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, - 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, - 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, - 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, - 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, - 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, - 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, - 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, - 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, - 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, - 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, - 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, - 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, - }; - static constexpr u32 ylut[] = { - 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, - 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, - 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, - 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, - 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, - 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, - 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, - 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, - 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, - 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, - 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, - 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, - 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, - 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, - 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, - 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, - 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, - 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, - 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, - 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, - 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, - 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, - 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, - 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, - 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, - 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, - 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, - 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, - 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, - 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, - 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, - 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, - 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, - 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, - 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, - }; - return xlut[x % 128] + ylut[y % 128]; -} - -static inline u32 GetMortonOffset128(u32 x, u32 y, u32 bytes_per_pixel) { - // Calculates the offset of the position of the pixel in Morton order - // Framebuffer images are split into 128x128 tiles. - - const unsigned int block_height = 128; - const unsigned int coarse_x = x & ~127; - - u32 i = MortonInterleave128(x, y); - - const unsigned int offset = coarse_x * block_height; - - return (i + offset) * bytes_per_pixel; -} - -static void MortonCopyPixels128(u32 width, u32 height, u32 bytes_per_pixel, u32 gl_bytes_per_pixel, - u8* morton_data, u8* gl_data, bool morton_to_gl) { - u8* data_ptrs[2]; - for (unsigned y = 0; y < height; ++y) { - for (unsigned x = 0; x < width; ++x) { - const u32 coarse_y = y & ~127; - u32 morton_offset = - GetMortonOffset128(x, y, bytes_per_pixel) + coarse_y * width * bytes_per_pixel; - u32 gl_pixel_index = (x + (height - 1 - y) * width) * gl_bytes_per_pixel; - - data_ptrs[morton_to_gl] = morton_data + morton_offset; - data_ptrs[!morton_to_gl] = &gl_data[gl_pixel_index]; - - memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); - } - } -} - /** * Loads framebuffer from emulated memory into the active OpenGL texture. */ -void RendererOpenGL::LoadFBToScreenInfo(const FramebufferInfo& framebuffer_info, +void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuffer, ScreenInfo& screen_info) { - const u32 bpp{FramebufferInfo::BytesPerPixel(framebuffer_info.pixel_format)}; - const u32 size_in_bytes{framebuffer_info.stride * framebuffer_info.height * bpp}; + const u32 bytes_per_pixel{Tegra::FramebufferConfig::BytesPerPixel(framebuffer.pixel_format)}; + const u64 size_in_bytes{framebuffer.stride * framebuffer.height * bytes_per_pixel}; + const VAddr framebuffer_addr{framebuffer.address + framebuffer.offset}; - MortonCopyPixels128(framebuffer_info.width, framebuffer_info.height, bpp, 4, - Memory::GetPointer(framebuffer_info.address), gl_framebuffer_data.data(), - true); - - LOG_TRACE(Render_OpenGL, "0x%08x bytes from 0x%llx(%dx%d), fmt %x", size_in_bytes, - framebuffer_info.address, framebuffer_info.width, framebuffer_info.height, - (int)framebuffer_info.pixel_format); + // Framebuffer orientation handling + framebuffer_transform_flags = framebuffer.transform_flags; // Ensure no bad interactions with GL_UNPACK_ALIGNMENT, which by default // only allows rows to have a memory alignement of 4. - ASSERT(framebuffer_info.stride % 4 == 0); + ASSERT(framebuffer.stride % 4 == 0); - framebuffer_flip_vertical = framebuffer_info.flip_vertical; + if (!Rasterizer()->AccelerateDisplay(framebuffer, framebuffer_addr, framebuffer.stride, + screen_info)) { + // Reset the screen info's display texture to its own permanent texture + screen_info.display_texture = screen_info.texture.resource.handle; + screen_info.display_texcoords = MathUtil::Rectangle<float>(0.f, 0.f, 1.f, 1.f); - // Reset the screen info's display texture to its own permanent texture - screen_info.display_texture = screen_info.texture.resource.handle; - screen_info.display_texcoords = MathUtil::Rectangle<float>(0.f, 0.f, 1.f, 1.f); + Rasterizer()->FlushRegion(framebuffer_addr, size_in_bytes); - // Memory::RasterizerFlushRegion(framebuffer_info.address, size_in_bytes); + VideoCore::MortonCopyPixels128(framebuffer.width, framebuffer.height, bytes_per_pixel, 4, + Memory::GetPointer(framebuffer_addr), + gl_framebuffer_data.data(), true); - state.texture_units[0].texture_2d = screen_info.texture.resource.handle; - state.Apply(); + state.texture_units[0].texture_2d = screen_info.texture.resource.handle; + state.Apply(); - glActiveTexture(GL_TEXTURE0); - glPixelStorei(GL_UNPACK_ROW_LENGTH, (GLint)framebuffer_info.stride); + glActiveTexture(GL_TEXTURE0); + glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(framebuffer.stride)); - // Update existing texture - // TODO: Test what happens on hardware when you change the framebuffer dimensions so that - // they differ from the LCD resolution. - // TODO: Applications could theoretically crash Citra here by specifying too large - // framebuffer sizes. We should make sure that this cannot happen. - glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, framebuffer_info.width, framebuffer_info.height, - screen_info.texture.gl_format, screen_info.texture.gl_type, - gl_framebuffer_data.data()); + // Update existing texture + // TODO: Test what happens on hardware when you change the framebuffer dimensions so that + // they differ from the LCD resolution. + // TODO: Applications could theoretically crash yuzu here by specifying too large + // framebuffer sizes. We should make sure that this cannot happen. + glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, framebuffer.width, framebuffer.height, + screen_info.texture.gl_format, screen_info.texture.gl_type, + gl_framebuffer_data.data()); - glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); + glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); - state.texture_units[0].texture_2d = 0; - state.Apply(); + state.texture_units[0].texture_2d = 0; + state.Apply(); + } } /** @@ -318,7 +210,7 @@ void RendererOpenGL::InitOpenGLObjects() { 0.0f); // Link shaders and get variable locations - shader.Create(vertex_shader, fragment_shader); + shader.Create(vertex_shader, nullptr, fragment_shader); state.draw.shader_program = shader.handle; state.Apply(); uniform_modelview_matrix = glGetUniformLocation(shader.handle, "modelview_matrix"); @@ -372,14 +264,14 @@ void RendererOpenGL::InitOpenGLObjects() { } void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, - const FramebufferInfo& framebuffer_info) { + const Tegra::FramebufferConfig& framebuffer) { - texture.width = framebuffer_info.width; - texture.height = framebuffer_info.height; + texture.width = framebuffer.width; + texture.height = framebuffer.height; GLint internal_format; - switch (framebuffer_info.pixel_format) { - case FramebufferInfo::PixelFormat::ABGR8: + switch (framebuffer.pixel_format) { + case Tegra::FramebufferConfig::PixelFormat::ABGR8: // Use RGBA8 and swap in the fragment shader internal_format = GL_RGBA; texture.gl_format = GL_RGBA; @@ -387,7 +279,7 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, gl_framebuffer_data.resize(texture.width * texture.height * 4); break; default: - UNIMPLEMENTED(); + UNREACHABLE(); } state.texture_units[0].texture_2d = texture.resource.handle; @@ -401,11 +293,22 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, state.Apply(); } -void RendererOpenGL::DrawSingleScreen(const ScreenInfo& screen_info, float x, float y, float w, - float h) { +void RendererOpenGL::DrawScreenTriangles(const ScreenInfo& screen_info, float x, float y, float w, + float h) { const auto& texcoords = screen_info.display_texcoords; - const auto& left = framebuffer_flip_vertical ? texcoords.right : texcoords.left; - const auto& right = framebuffer_flip_vertical ? texcoords.left : texcoords.right; + auto left = texcoords.left; + auto right = texcoords.right; + if (framebuffer_transform_flags != Tegra::FramebufferConfig::TransformFlags::Unset) + if (framebuffer_transform_flags == Tegra::FramebufferConfig::TransformFlags::FlipV) { + // Flip the framebuffer vertically + left = texcoords.right; + right = texcoords.left; + } else { + // Other transformations are unsupported + LOG_CRITICAL(Render_OpenGL, "Unsupported framebuffer_transform_flags=%d", + framebuffer_transform_flags); + UNIMPLEMENTED(); + } std::array<ScreenRectVertex, 4> vertices = {{ ScreenRectVertex(x, y, texcoords.top, right), @@ -427,7 +330,7 @@ void RendererOpenGL::DrawSingleScreen(const ScreenInfo& screen_info, float x, fl /** * Draws the emulated screens to the emulator window. */ -void RendererOpenGL::DrawScreens() { +void RendererOpenGL::DrawScreen() { const auto& layout = render_window->GetFramebufferLayout(); const auto& screen = layout.screen; @@ -443,8 +346,8 @@ void RendererOpenGL::DrawScreens() { glActiveTexture(GL_TEXTURE0); glUniform1i(uniform_color_texture, 0); - DrawSingleScreen(screen_info, (float)screen.left, (float)screen.top, (float)screen.GetWidth(), - (float)screen.GetHeight()); + DrawScreenTriangles(screen_info, (float)screen.left, (float)screen.top, + (float)screen.GetWidth(), (float)screen.GetHeight()); m_current_frame++; } diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h index 05bb3c5cf..fffd0f9f4 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.h +++ b/src/video_core/renderer_opengl/renderer_opengl.h @@ -21,7 +21,7 @@ struct TextureInfo { GLsizei height; GLenum gl_format; GLenum gl_type; - RendererBase::FramebufferInfo::PixelFormat pixel_format; + Tegra::FramebufferConfig::PixelFormat pixel_format; }; /// Structure used for storing information about the display target for each 3DS screen @@ -37,7 +37,7 @@ public: ~RendererOpenGL() override; /// Swap buffers (render frame) - void SwapBuffers(boost::optional<const FramebufferInfo&> framebuffer_info) override; + void SwapBuffers(boost::optional<const Tegra::FramebufferConfig&> framebuffer) override; /** * Set the emulator window to use for renderer @@ -53,13 +53,14 @@ public: private: void InitOpenGLObjects(); - void ConfigureFramebufferTexture(TextureInfo& texture, const FramebufferInfo& framebuffer_info); - void DrawScreens(); - void DrawSingleScreen(const ScreenInfo& screen_info, float x, float y, float w, float h); + void ConfigureFramebufferTexture(TextureInfo& texture, + const Tegra::FramebufferConfig& framebuffer); + void DrawScreen(); + void DrawScreenTriangles(const ScreenInfo& screen_info, float x, float y, float w, float h); void UpdateFramerate(); // Loads framebuffer from emulated memory into the display information structure - void LoadFBToScreenInfo(const FramebufferInfo& framebuffer_info, ScreenInfo& screen_info); + void LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuffer, ScreenInfo& screen_info); // Fills active OpenGL texture with the given RGBA color. void LoadColorToActiveGLTexture(u8 color_r, u8 color_g, u8 color_b, u8 color_a, const TextureInfo& texture); @@ -87,6 +88,6 @@ private: GLuint attrib_position; GLuint attrib_tex_coord; - /// Flips the framebuffer vertically when true - bool framebuffer_flip_vertical; + /// Used for transforming the framebuffer orientation + Tegra::FramebufferConfig::TransformFlags framebuffer_transform_flags; }; diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp new file mode 100644 index 000000000..2e87281eb --- /dev/null +++ b/src/video_core/textures/decoders.cpp @@ -0,0 +1,105 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <cstring> +#include "common/assert.h" +#include "video_core/textures/decoders.h" +#include "video_core/textures/texture.h" + +namespace Tegra { +namespace Texture { + +/** + * Calculates the offset of an (x, y) position within a swizzled texture. + * Taken from the Tegra X1 TRM. + */ +static u32 GetSwizzleOffset(u32 x, u32 y, u32 image_width, u32 bytes_per_pixel, u32 block_height) { + u32 image_width_in_gobs = image_width * bytes_per_pixel / 64; + u32 GOB_address = 0 + (y / (8 * block_height)) * 512 * block_height * image_width_in_gobs + + (x * bytes_per_pixel / 64) * 512 * block_height + + (y % (8 * block_height) / 8) * 512; + x *= bytes_per_pixel; + u32 address = GOB_address + ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 + + (y % 2) * 16 + (x % 16); + + return address; +} + +static void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, + u8* swizzled_data, u8* unswizzled_data, bool unswizzle, + u32 block_height) { + u8* data_ptrs[2]; + for (unsigned y = 0; y < height; ++y) { + for (unsigned x = 0; x < width; ++x) { + u32 swizzle_offset = GetSwizzleOffset(x, y, width, bytes_per_pixel, block_height); + u32 pixel_index = (x + y * width) * out_bytes_per_pixel; + + data_ptrs[unswizzle] = swizzled_data + swizzle_offset; + data_ptrs[!unswizzle] = &unswizzled_data[pixel_index]; + + std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); + } + } +} + +u32 BytesPerPixel(TextureFormat format) { + switch (format) { + case TextureFormat::DXT1: + // In this case a 'pixel' actually refers to a 4x4 tile. + return 8; + case TextureFormat::A8R8G8B8: + return 4; + default: + UNIMPLEMENTED_MSG("Format not implemented"); + break; + } +} + +std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width, u32 height) { + u8* data = Memory::GetPointer(address); + u32 bytes_per_pixel = BytesPerPixel(format); + + static constexpr u32 DefaultBlockHeight = 16; + + std::vector<u8> unswizzled_data(width * height * bytes_per_pixel); + + switch (format) { + case TextureFormat::DXT1: + // In the DXT1 format, each 4x4 tile is swizzled instead of just individual pixel values. + CopySwizzledData(width / 4, height / 4, bytes_per_pixel, bytes_per_pixel, data, + unswizzled_data.data(), true, DefaultBlockHeight); + break; + case TextureFormat::A8R8G8B8: + CopySwizzledData(width, height, bytes_per_pixel, bytes_per_pixel, data, + unswizzled_data.data(), true, DefaultBlockHeight); + break; + default: + UNIMPLEMENTED_MSG("Format not implemented"); + break; + } + + return unswizzled_data; +} + +std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat format, u32 width, + u32 height) { + std::vector<u8> rgba_data; + + // TODO(Subv): Implement. + switch (format) { + case TextureFormat::DXT1: + case TextureFormat::A8R8G8B8: + // TODO(Subv): For the time being just forward the same data without any decoding. + rgba_data = texture_data; + break; + default: + UNIMPLEMENTED_MSG("Format not implemented"); + break; + } + + return rgba_data; +} + +} // namespace Texture +} // namespace Tegra diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h new file mode 100644 index 000000000..0c21694ff --- /dev/null +++ b/src/video_core/textures/decoders.h @@ -0,0 +1,26 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <vector> +#include "common/common_types.h" +#include "video_core/textures/texture.h" + +namespace Tegra { +namespace Texture { + +/** + * Unswizzles a swizzled texture without changing its format. + */ +std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width, u32 height); + +/** + * Decodes an unswizzled texture into a A8R8G8B8 texture. + */ +std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat format, u32 width, + u32 height); + +} // namespace Texture +} // namespace Tegra diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h new file mode 100644 index 000000000..07936f8a3 --- /dev/null +++ b/src/video_core/textures/texture.h @@ -0,0 +1,137 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include "common/bit_field.h" +#include "common/common_funcs.h" +#include "common/common_types.h" +#include "video_core/memory_manager.h" + +namespace Tegra { +namespace Texture { + +enum class TextureFormat : u32 { + A8R8G8B8 = 8, + DXT1 = 0x24, +}; + +enum class TextureType : u32 { + Texture1D = 0, + Texture2D = 1, + Texture3D = 2, + TextureCubemap = 3, + Texture1DArray = 4, + Texture2DArray = 5, + Texture1DBuffer = 6, + Texture2DNoMipmap = 7, + TextureCubeArray = 8, +}; + +enum class TICHeaderVersion : u32 { + OneDBuffer = 0, + PitchColorKey = 1, + Pitch = 2, + BlockLinear = 3, + BlockLinearColorKey = 4, +}; + +union TextureHandle { + u32 raw; + BitField<0, 20, u32> tic_id; + BitField<20, 12, u32> tsc_id; +}; +static_assert(sizeof(TextureHandle) == 4, "TextureHandle has wrong size"); + +struct TICEntry { + union { + u32 raw; + BitField<0, 7, TextureFormat> format; + BitField<7, 3, u32> r_type; + BitField<10, 3, u32> g_type; + BitField<13, 3, u32> b_type; + BitField<16, 3, u32> a_type; + }; + u32 address_low; + union { + BitField<0, 16, u32> address_high; + BitField<21, 3, TICHeaderVersion> header_version; + }; + INSERT_PADDING_BYTES(4); + union { + BitField<0, 16, u32> width_minus_1; + BitField<23, 4, TextureType> texture_type; + }; + u16 height_minus_1; + INSERT_PADDING_BYTES(10); + + GPUVAddr Address() const { + return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | address_low); + } + + u32 Width() const { + return width_minus_1 + 1; + } + + u32 Height() const { + return height_minus_1 + 1; + } +}; +static_assert(sizeof(TICEntry) == 0x20, "TICEntry has wrong size"); + +enum class WrapMode : u32 { + Wrap = 0, + Mirror = 1, + ClampToEdge = 2, + Border = 3, + ClampOGL = 4, + MirrorOnceClampToEdge = 5, + MirrorOnceBorder = 6, + MirrorOnceClampOGL = 7, +}; + +enum class TextureFilter : u32 { + Nearest = 1, + Linear = 2, +}; + +enum class TextureMipmapFilter : u32 { + None = 1, + Nearest = 2, + Linear = 3, +}; + +struct TSCEntry { + union { + BitField<0, 3, WrapMode> wrap_u; + BitField<3, 3, WrapMode> wrap_v; + BitField<6, 3, WrapMode> wrap_p; + BitField<9, 1, u32> depth_compare_enabled; + BitField<10, 3, u32> depth_compare_func; + }; + union { + BitField<0, 2, TextureFilter> mag_filter; + BitField<4, 2, TextureFilter> min_filter; + BitField<6, 2, TextureMipmapFilter> mip_filter; + }; + INSERT_PADDING_BYTES(8); + u32 border_color_r; + u32 border_color_g; + u32 border_color_b; + u32 border_color_a; +}; +static_assert(sizeof(TSCEntry) == 0x20, "TSCEntry has wrong size"); + +struct FullTextureInfo { + u32 index; + TICEntry tic; + TSCEntry tsc; + bool enabled; +}; + +/// Returns the number of bytes per pixel of the input texture format. +u32 BytesPerPixel(TextureFormat format); + +} // namespace Texture +} // namespace Tegra diff --git a/src/video_core/utils.h b/src/video_core/utils.h index d94a10417..be0f7e22b 100644 --- a/src/video_core/utils.h +++ b/src/video_core/utils.h @@ -49,4 +49,116 @@ static inline u32 GetMortonOffset(u32 x, u32 y, u32 bytes_per_pixel) { return (i + offset) * bytes_per_pixel; } +static inline u32 MortonInterleave128(u32 x, u32 y) { + // 128x128 Z-Order coordinate from 2D coordinates + static constexpr u32 xlut[] = { + 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, + 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, + 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, + 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, + 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, + 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, + 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, + 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, + 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, + 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, + 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, + 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, + 0x0008, 0x0009, 0x000a, 0x000b, 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, + 0x004b, 0x0800, 0x0801, 0x0802, 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, + 0x0842, 0x0843, 0x0848, 0x0849, 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, + 0x1009, 0x100a, 0x100b, 0x1040, 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, + 0x1800, 0x1801, 0x1802, 0x1803, 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, + 0x1843, 0x1848, 0x1849, 0x184a, 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, + 0x200a, 0x200b, 0x2040, 0x2041, 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, + 0x2801, 0x2802, 0x2803, 0x2808, 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, + 0x2848, 0x2849, 0x284a, 0x284b, 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, + 0x300b, 0x3040, 0x3041, 0x3042, 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, + 0x3802, 0x3803, 0x3808, 0x3809, 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, + 0x3849, 0x384a, 0x384b, 0x0000, 0x0001, 0x0002, 0x0003, 0x0008, 0x0009, 0x000a, 0x000b, + 0x0040, 0x0041, 0x0042, 0x0043, 0x0048, 0x0049, 0x004a, 0x004b, 0x0800, 0x0801, 0x0802, + 0x0803, 0x0808, 0x0809, 0x080a, 0x080b, 0x0840, 0x0841, 0x0842, 0x0843, 0x0848, 0x0849, + 0x084a, 0x084b, 0x1000, 0x1001, 0x1002, 0x1003, 0x1008, 0x1009, 0x100a, 0x100b, 0x1040, + 0x1041, 0x1042, 0x1043, 0x1048, 0x1049, 0x104a, 0x104b, 0x1800, 0x1801, 0x1802, 0x1803, + 0x1808, 0x1809, 0x180a, 0x180b, 0x1840, 0x1841, 0x1842, 0x1843, 0x1848, 0x1849, 0x184a, + 0x184b, 0x2000, 0x2001, 0x2002, 0x2003, 0x2008, 0x2009, 0x200a, 0x200b, 0x2040, 0x2041, + 0x2042, 0x2043, 0x2048, 0x2049, 0x204a, 0x204b, 0x2800, 0x2801, 0x2802, 0x2803, 0x2808, + 0x2809, 0x280a, 0x280b, 0x2840, 0x2841, 0x2842, 0x2843, 0x2848, 0x2849, 0x284a, 0x284b, + 0x3000, 0x3001, 0x3002, 0x3003, 0x3008, 0x3009, 0x300a, 0x300b, 0x3040, 0x3041, 0x3042, + 0x3043, 0x3048, 0x3049, 0x304a, 0x304b, 0x3800, 0x3801, 0x3802, 0x3803, 0x3808, 0x3809, + 0x380a, 0x380b, 0x3840, 0x3841, 0x3842, 0x3843, 0x3848, 0x3849, 0x384a, 0x384b, + }; + static constexpr u32 ylut[] = { + 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, + 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, + 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, + 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, + 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, + 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, + 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, + 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, + 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, + 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, + 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, + 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, + 0x0020, 0x0024, 0x0030, 0x0034, 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, + 0x00b4, 0x0100, 0x0104, 0x0110, 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, + 0x0190, 0x0194, 0x01a0, 0x01a4, 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, + 0x0224, 0x0230, 0x0234, 0x0280, 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, + 0x0300, 0x0304, 0x0310, 0x0314, 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, + 0x0394, 0x03a0, 0x03a4, 0x03b0, 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, + 0x0430, 0x0434, 0x0480, 0x0484, 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, + 0x0504, 0x0510, 0x0514, 0x0520, 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, + 0x05a0, 0x05a4, 0x05b0, 0x05b4, 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, + 0x0634, 0x0680, 0x0684, 0x0690, 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, + 0x0710, 0x0714, 0x0720, 0x0724, 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, + 0x07a4, 0x07b0, 0x07b4, 0x0000, 0x0004, 0x0010, 0x0014, 0x0020, 0x0024, 0x0030, 0x0034, + 0x0080, 0x0084, 0x0090, 0x0094, 0x00a0, 0x00a4, 0x00b0, 0x00b4, 0x0100, 0x0104, 0x0110, + 0x0114, 0x0120, 0x0124, 0x0130, 0x0134, 0x0180, 0x0184, 0x0190, 0x0194, 0x01a0, 0x01a4, + 0x01b0, 0x01b4, 0x0200, 0x0204, 0x0210, 0x0214, 0x0220, 0x0224, 0x0230, 0x0234, 0x0280, + 0x0284, 0x0290, 0x0294, 0x02a0, 0x02a4, 0x02b0, 0x02b4, 0x0300, 0x0304, 0x0310, 0x0314, + 0x0320, 0x0324, 0x0330, 0x0334, 0x0380, 0x0384, 0x0390, 0x0394, 0x03a0, 0x03a4, 0x03b0, + 0x03b4, 0x0400, 0x0404, 0x0410, 0x0414, 0x0420, 0x0424, 0x0430, 0x0434, 0x0480, 0x0484, + 0x0490, 0x0494, 0x04a0, 0x04a4, 0x04b0, 0x04b4, 0x0500, 0x0504, 0x0510, 0x0514, 0x0520, + 0x0524, 0x0530, 0x0534, 0x0580, 0x0584, 0x0590, 0x0594, 0x05a0, 0x05a4, 0x05b0, 0x05b4, + 0x0600, 0x0604, 0x0610, 0x0614, 0x0620, 0x0624, 0x0630, 0x0634, 0x0680, 0x0684, 0x0690, + 0x0694, 0x06a0, 0x06a4, 0x06b0, 0x06b4, 0x0700, 0x0704, 0x0710, 0x0714, 0x0720, 0x0724, + 0x0730, 0x0734, 0x0780, 0x0784, 0x0790, 0x0794, 0x07a0, 0x07a4, 0x07b0, 0x07b4, + }; + return xlut[x % 128] + ylut[y % 128]; +} + +static inline u32 GetMortonOffset128(u32 x, u32 y, u32 bytes_per_pixel) { + // Calculates the offset of the position of the pixel in Morton order + // Framebuffer images are split into 128x128 tiles. + + const unsigned int block_height = 128; + const unsigned int coarse_x = x & ~127; + + u32 i = MortonInterleave128(x, y); + + const unsigned int offset = coarse_x * block_height; + + return (i + offset) * bytes_per_pixel; +} + +static inline void MortonCopyPixels128(u32 width, u32 height, u32 bytes_per_pixel, + u32 gl_bytes_per_pixel, u8* morton_data, u8* gl_data, + bool morton_to_gl) { + u8* data_ptrs[2]; + for (unsigned y = 0; y < height; ++y) { + for (unsigned x = 0; x < width; ++x) { + const u32 coarse_y = y & ~127; + u32 morton_offset = + GetMortonOffset128(x, y, bytes_per_pixel) + coarse_y * width * bytes_per_pixel; + u32 gl_pixel_index = (x + (height - 1 - y) * width) * gl_bytes_per_pixel; + + data_ptrs[morton_to_gl] = morton_data + morton_offset; + data_ptrs[!morton_to_gl] = &gl_data[gl_pixel_index]; + + memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); + } + } +} + } // namespace VideoCore diff --git a/src/video_core/video_core.cpp b/src/video_core/video_core.cpp index 864691baa..289140f31 100644 --- a/src/video_core/video_core.cpp +++ b/src/video_core/video_core.cpp @@ -26,7 +26,7 @@ bool Init(EmuWindow* emu_window) { if (g_renderer->Init()) { LOG_DEBUG(Render, "initialized OK"); } else { - LOG_ERROR(Render, "initialization failed !"); + LOG_CRITICAL(Render, "initialization failed !"); return false; } return true; diff --git a/src/video_core/video_core.h b/src/video_core/video_core.h index 1fd90b9d0..37da62436 100644 --- a/src/video_core/video_core.h +++ b/src/video_core/video_core.h @@ -15,6 +15,8 @@ class RendererBase; namespace VideoCore { +enum class Renderer { Software, OpenGL }; + extern std::unique_ptr<RendererBase> g_renderer; ///< Renderer plugin extern EmuWindow* g_emu_window; ///< Emu window |