Forks/citra
1
0
Fork 0
mirror of https://github.com/PabloMK7/citra.git synced 2025-09-10 12:50:04 +00:00
Code Issues Projects Releases Packages Wiki Activity

video_core: Refactor GPU interface (#7272)

Browse source 
* video_core: Refactor GPU interface

* citra_qt: Better debug widget lifetime
This commit is contained in:
GPUCode 2023-12-28 12:46:57 +02:00 • committed by GitHub
parent 602f4f60d8
commit 2bb7f89c30
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
167 changed files with 4172 additions and 4866 deletions
Download patch file Download diff file Expand all files Collapse all files

8
src/core/CMakeLists.txt
View file

@ -299,8 +299,10 @@ add_library(citra_core STATIC
hle/service/fs/fs_user.h
hle/service/gsp/gsp.cpp
hle/service/gsp/gsp.h
hle/service/gsp/gsp_command.h
hle/service/gsp/gsp_gpu.cpp
hle/service/gsp/gsp_gpu.h
hle/service/gsp/gsp_interrupt.h
hle/service/gsp/gsp_lcd.cpp
hle/service/gsp/gsp_lcd.h
hle/service/hid/hid.cpp
@ -433,12 +435,6 @@ add_library(citra_core STATIC
hw/aes/ccm.h
hw/aes/key.cpp
hw/aes/key.h
hw/gpu.cpp
hw/gpu.h
hw/hw.cpp
hw/hw.h
hw/lcd.cpp
hw/lcd.h
hw/rsa/rsa.cpp
hw/rsa/rsa.h
hw/y2r.cpp

1
src/core/cheats/cheats.cpp
View file

@ -11,7 +11,6 @@
#include "core/core.h"
#include "core/core_timing.h"
#include "core/hle/kernel/process.h"
#include "core/hw/gpu.h"
namespace Cheats {

56
src/core/core.cpp
View file

@ -35,14 +35,13 @@
#include "core/hle/service/cam/cam.h"
#include "core/hle/service/fs/archive.h"
#include "core/hle/service/gsp/gsp.h"
#include "core/hle/service/gsp/gsp_gpu.h"
#include "core/hle/service/ir/ir_rst.h"
#include "core/hle/service/mic/mic_u.h"
#include "core/hle/service/plgldr/plgldr.h"
#include "core/hle/service/service.h"
#include "core/hle/service/sm/sm.h"
#include "core/hw/gpu.h"
#include "core/hw/hw.h"
#include "core/hw/lcd.h"
#include "core/hw/aes/key.h"
#include "core/loader/loader.h"
#include "core/movie.h"
#ifdef ENABLE_SCRIPTING
@ -51,8 +50,8 @@
#include "core/telemetry_session.h"
#include "network/network.h"
#include "video_core/custom_textures/custom_tex_manager.h"
#include "video_core/gpu.h"
#include "video_core/renderer_base.h"
#include "video_core/video_core.h"
namespace Core {
@ -235,7 +234,6 @@ System::ResultStatus System::RunLoop(bool tight_loop) {
GDBStub::SetCpuStepFlag(false);
}
HW::Update();
Reschedule();
return status;
@ -433,7 +431,7 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window,
service_manager = std::make_unique<Service::SM::ServiceManager>(*this);
archive_manager = std::make_unique<Service::FS::ArchiveManager>(*this);
HW::Init(*memory);
HW::AES::InitKeys();
Service::Init(*this);
GDBStub::DeferStart();
@ -443,7 +441,10 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window,
custom_tex_manager = std::make_unique<VideoCore::CustomTexManager>(*this);
VideoCore::Init(emu_window, secondary_window, *this);
auto gsp = service_manager->GetService<Service::GSP::GSP_GPU>("gsp::Gpu");
gpu = std::make_unique<VideoCore::GPU>(*this, emu_window, secondary_window);
gpu->SetInterruptHandler(
[gsp](Service::GSP::InterruptId interrupt_id) { gsp->SignalInterrupt(interrupt_id); });
LOG_DEBUG(Core, "Initialized OK");
@ -452,8 +453,8 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window,
return ResultStatus::Success;
}
VideoCore::RendererBase& System::Renderer() {
return *VideoCore::g_renderer;
VideoCore::GPU& System::GPU() {
return *gpu;
}
Service::SM::ServiceManager& System::ServiceManager() {
@ -555,8 +556,7 @@ void System::Shutdown(bool is_deserializing) {
// Shutdown emulation session
is_powered_on = false;
VideoCore::Shutdown();
HW::Shutdown();
gpu.reset();
if (!is_deserializing) {
GDBStub::Shutdown();
perf_stats.reset();
@ -626,18 +626,9 @@ void System::ApplySettings() {
GDBStub::SetServerPort(Settings::values.gdbstub_port.GetValue());
GDBStub::ToggleServer(Settings::values.use_gdbstub.GetValue());
VideoCore::g_shader_jit_enabled = Settings::values.use_shader_jit.GetValue();
VideoCore::g_hw_shader_enabled = Settings::values.use_hw_shader.GetValue();
VideoCore::g_hw_shader_accurate_mul = Settings::values.shaders_accurate_mul.GetValue();
#ifndef ANDROID
if (VideoCore::g_renderer) {
VideoCore::g_renderer->UpdateCurrentFramebufferLayout();
}
#endif
if (VideoCore::g_renderer) {
auto& settings = VideoCore::g_renderer->Settings();
if (gpu) {
gpu->Renderer().UpdateCurrentFramebufferLayout();
auto& settings = gpu->Renderer().Settings();
settings.bg_color_update_requested = true;
settings.shader_update_requested = true;
}
@ -699,17 +690,15 @@ void System::serialize(Archive& ar, const unsigned int file_version) {
*m_emu_window, m_secondary_window, *memory_mode.first, *n3ds_hw_caps.first, num_cores);
}
// flush on save, don't flush on load
bool should_flush = !Archive::is_loading::value;
Memory::RasterizerClearAll(should_flush);
// Flush on save, don't flush on load
const bool should_flush = !Archive::is_loading::value;
gpu->ClearAll(should_flush);
ar&* timing.get();
for (u32 i = 0; i < num_cores; i++) {
ar&* cpu_cores[i].get();
}
ar&* service_manager.get();
ar&* archive_manager.get();
ar& GPU::g_regs;
ar& LCD::g_regs;
// NOTE: DSP doesn't like being destroyed and recreated. So instead we do an inline
// serialization; this means that the DSP Settings need to match for loading to work.
@ -722,16 +711,21 @@ void System::serialize(Archive& ar, const unsigned int file_version) {
ar&* memory.get();
ar&* kernel.get();
VideoCore::serialize(ar, file_version);
ar&* gpu.get();
ar& movie;
// This needs to be set from somewhere - might as well be here!
if (Archive::is_loading::value) {
timing->UnlockEventQueue();
Service::GSP::SetGlobalModule(*this);
memory->SetDSP(*dsp_core);
cheat_engine->Connect();
VideoCore::g_renderer->Sync();
gpu->Sync();
// Re-register gpu callback, because gsp service changed after service_manager got
// serialized
auto gsp = service_manager->GetService<Service::GSP::GSP_GPU>("gsp::Gpu");
gpu->SetInterruptHandler(
[gsp](Service::GSP::InterruptId interrupt_id) { gsp->SignalInterrupt(interrupt_id); });
}
}

10
src/core/core.h
View file

@ -58,9 +58,13 @@ class Backend;
namespace VideoCore {
class CustomTexManager;
class RendererBase;
class GPU;
} // namespace VideoCore
namespace Pica {
class DebugContext;
}
namespace Loader {
class AppLoader;
}
@ -217,7 +221,7 @@ public:
return *dsp_core;
}
[[nodiscard]] VideoCore::RendererBase& Renderer();
[[nodiscard]] VideoCore::GPU& GPU();
/**
* Gets a reference to the service manager.
@ -384,6 +388,8 @@ private:
/// Telemetry session for this emulation session
std::unique_ptr<Core::TelemetrySession> telemetry_session;
std::unique_ptr<VideoCore::GPU> gpu;
/// Service manager
std::unique_ptr<Service::SM::ServiceManager> service_manager;

4
src/core/core_timing.h
View file

@ -37,6 +37,10 @@
constexpr u64 BASE_CLOCK_RATE_ARM11 = 268111856;
constexpr u64 MAX_VALUE_TO_MULTIPLY = std::numeric_limits<s64>::max() / BASE_CLOCK_RATE_ARM11;
/// Refresh rate defined by ratio of ARM11 frequency to ARM11 ticks per frame
/// (268,111,856) / (4,481,136) = 59.83122493939037Hz
constexpr double SCREEN_REFRESH_RATE = BASE_CLOCK_RATE_ARM11 / static_cast<double>(4481136ull);
constexpr s64 msToCycles(int ms) {
// since ms is int there is no way to overflow
return BASE_CLOCK_RATE_ARM11 * static_cast<s64>(ms) / 1000;

12
src/core/dumping/ffmpeg_backend.cpp
View file

@ -11,10 +11,10 @@
#include "common/scope_exit.h"
#include "common/settings.h"
#include "common/string_util.h"
#include "core/core_timing.h"
#include "core/dumping/ffmpeg_backend.h"
#include "core/hw/gpu.h"
#include "video_core/gpu.h"
#include "video_core/renderer_base.h"
#include "video_core/video_core.h"
using namespace DynamicLibrary;
@ -381,7 +381,7 @@ bool FFmpegVideoStream::InitFilters() {
}
// Configure buffer source
static constexpr AVRational src_time_base{static_cast<int>(GPU::frame_ticks),
static constexpr AVRational src_time_base{static_cast<int>(VideoCore::FRAME_TICKS),
static_cast<int>(BASE_CLOCK_RATE_ARM11)};
const std::string in_args =
fmt::format("video_size={}x{}:pix_fmt={}:time_base={}/{}:pixel_aspect=1", layout.width,
@ -732,7 +732,7 @@ void FFmpegMuxer::WriteTrailer() {
FFmpeg::av_write_trailer(format_context.get());
}
FFmpegBackend::FFmpegBackend() = default;
FFmpegBackend::FFmpegBackend(VideoCore::RendererBase& renderer_) : renderer{renderer_} {}
FFmpegBackend::~FFmpegBackend() {
ASSERT_MSG(!IsDumping(), "Dumping must be stopped first");
@ -796,7 +796,7 @@ bool FFmpegBackend::StartDumping(const std::string& path, const Layout::Framebuf
}
});
VideoCore::g_renderer->PrepareVideoDumping();
renderer.PrepareVideoDumping();
is_dumping = true;
return true;
@ -829,7 +829,7 @@ void FFmpegBackend::AddAudioSample(const std::array<s16, 2>& sample) {
void FFmpegBackend::StopDumping() {
is_dumping = false;
VideoCore::g_renderer->CleanupVideoDumping();
renderer.CleanupVideoDumping();
// Flush the video processing queue
AddVideoFrame(VideoFrame());

7
src/core/dumping/ffmpeg_backend.h
View file

@ -18,6 +18,10 @@
#include "common/threadsafe_queue.h"
#include "core/dumping/backend.h"
namespace VideoCore {
class RendererBase;
}
namespace VideoDumper {
using VariableAudioFrame = std::vector<s16>;
@ -181,7 +185,7 @@ private:
*/
class FFmpegBackend : public Backend {
public:
FFmpegBackend();
FFmpegBackend(VideoCore::RendererBase& renderer);
~FFmpegBackend() override;
bool StartDumping(const std::string& path, const Layout::FramebufferLayout& layout) override;
void AddVideoFrame(VideoFrame frame) override;
@ -194,6 +198,7 @@ public:
private:
void EndDumping();
VideoCore::RendererBase& renderer;
std::atomic_bool is_dumping = false; ///< Whether the backend is currently dumping
FFmpegMuxer ffmpeg{};

22
src/core/hle/service/gsp/gsp.cpp
View file

@ -2,33 +2,17 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <vector>
#include "core/core.h"
#include "core/hle/kernel/event.h"
#include "core/hle/kernel/shared_memory.h"
#include "core/hle/service/gsp/gsp.h"
#include "core/hle/service/gsp/gsp_gpu.h"
#include "core/hle/service/gsp/gsp_lcd.h"
namespace Service::GSP {
static std::weak_ptr<GSP_GPU> gsp_gpu;
void SignalInterrupt(InterruptId interrupt_id) {
auto gpu = gsp_gpu.lock();
ASSERT(gpu != nullptr);
return gpu->SignalInterrupt(interrupt_id);
}
void InstallInterfaces(Core::System& system) {
auto& service_manager = system.ServiceManager();
auto gpu = std::make_shared<GSP_GPU>(system);
gpu->InstallAsService(service_manager);
gsp_gpu = gpu;
std::make_shared<GSP_GPU>(system)->InstallAsService(service_manager);
std::make_shared<GSP_LCD>()->InstallAsService(service_manager);
}
void SetGlobalModule(Core::System& system) {
gsp_gpu = system.ServiceManager().GetService<GSP_GPU>("gsp::Gpu");
}
} // namespace Service::GSP

13
src/core/hle/service/gsp/gsp.h
View file

@ -4,25 +4,12 @@
#pragma once
#include <cstddef>
#include <string>
#include "common/common_types.h"
#include "core/hle/result.h"
#include "core/hle/service/gsp/gsp_gpu.h"
#include "core/hle/service/gsp/gsp_lcd.h"
namespace Core {
class System;
}
namespace Service::GSP {
/**
* Signals that the specified interrupt type has occurred to userland code
* @param interrupt_id ID of interrupt that is being signalled
*/
void SignalInterrupt(InterruptId interrupt_id);
void InstallInterfaces(Core::System& system);
void SetGlobalModule(Core::System& system);
} // namespace Service::GSP

110
src/core/hle/service/gsp/gsp_command.h Normal file
View file

@ -0,0 +1,110 @@
// Copyright 2023 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/bit_field.h"
namespace Service::GSP {
/// GSP command ID
enum class CommandId : u32 {
RequestDma = 0x00,
SubmitCmdList = 0x01,
MemoryFill = 0x02,
DisplayTransfer = 0x03,
TextureCopy = 0x04,
CacheFlush = 0x05,
};
struct DmaCommand {
u32 source_address;
u32 dest_address;
u32 size;
};
struct SubmitCmdListCommand {
u32 address;
u32 size;
u32 flags;
u32 unused[3];
u32 do_flush;
};
struct MemoryFillCommand {
u32 start1;
u32 value1;
u32 end1;
u32 start2;
u32 value2;
u32 end2;
u16 control1;
u16 control2;
};
struct DisplayTransferCommand {
u32 in_buffer_address;
u32 out_buffer_address;
u32 in_buffer_size;
u32 out_buffer_size;
u32 flags;
};
struct TextureCopyCommand {
u32 in_buffer_address;
u32 out_buffer_address;
u32 size;
u32 in_width_gap;
u32 out_width_gap;
u32 flags;
};
struct CacheFlushCommand {
struct {
u32 address;
u32 size;
} regions[3];
};
/// GSP command
struct Command {
BitField<0, 8, CommandId> id;
union {
DmaCommand dma_request;
SubmitCmdListCommand submit_gpu_cmdlist;
MemoryFillCommand memory_fill;
DisplayTransferCommand display_transfer;
TextureCopyCommand texture_copy;
CacheFlushCommand cache_flush;
std::array<u8, 0x1C> raw_data;
};
};
static_assert(sizeof(Command) == 0x20, "Command struct has incorrect size");
/// GSP shared memory GX command buffer header
struct CommandBuffer {
union {
u32 hex;
// Current command index. This index is updated by GSP module after loading the command
// data, right before the command is processed. When this index is updated by GSP module,
// the total commands field is decreased by one as well.
BitField<0, 8, u32> index;
// Total commands to process, must not be value 0 when GSP module handles commands. This
// must be <=15 when writing a command to shared memory. This is incremented by the
// application when writing a command to shared memory, after increasing this value
// TriggerCmdReqQueue is only used if this field is value 1.
BitField<8, 8, u32> number_commands;
};
u32 unk[7];
Command commands[0xF];
};
static_assert(sizeof(CommandBuffer) == 0x200, "CommandBuffer struct has incorrect size");
} // namespace Service::GSP

436
src/core/hle/service/gsp/gsp_gpu.cpp
View file

@ -9,30 +9,21 @@
#include <boost/serialization/shared_ptr.hpp>
#include "common/archives.h"
#include "common/bit_field.h"
#include "common/microprofile.h"
#include "common/swap.h"
#include "core/core.h"
#include "core/file_sys/plugin_3gx.h"
#include "core/hle/ipc.h"
#include "core/hle/ipc_helpers.h"
#include "core/hle/kernel/shared_memory.h"
#include "core/hle/kernel/shared_page.h"
#include "core/hle/result.h"
#include "core/hle/service/gsp/gsp_gpu.h"
#include "core/hw/gpu.h"
#include "core/hw/hw.h"
#include "core/hw/lcd.h"
#include "core/memory.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/gpu.h"
#include "video_core/gpu_debugger.h"
#include "video_core/pica/regs_lcd.h"
SERIALIZE_EXPORT_IMPL(Service::GSP::SessionData)
SERIALIZE_EXPORT_IMPL(Service::GSP::GSP_GPU)
SERVICE_CONSTRUCT_IMPL(Service::GSP::GSP_GPU)
// Main graphics debugger object - TODO: Here is probably not the best place for this
GraphicsDebugger g_debugger;
namespace Service::GSP {
// Beginning address of HW regs
@ -59,60 +50,32 @@ constexpr ResultCode ERR_REGS_INVALID_SIZE(ErrorDescription::InvalidSize, ErrorM
ErrorSummary::InvalidArgument,
ErrorLevel::Usage); // 0xE0E02BEC
static PAddr VirtualToPhysicalAddress(VAddr addr) {
if (addr == 0) {
return 0;
}
// Note: the region end check is inclusive because the game can pass in an address that
// represents an open right boundary
if (addr >= Memory::VRAM_VADDR && addr <= Memory::VRAM_VADDR_END) {
return addr - Memory::VRAM_VADDR + Memory::VRAM_PADDR;
}
if (addr >= Memory::LINEAR_HEAP_VADDR && addr <= Memory::LINEAR_HEAP_VADDR_END) {
return addr - Memory::LINEAR_HEAP_VADDR + Memory::FCRAM_PADDR;
}
if (addr >= Memory::NEW_LINEAR_HEAP_VADDR && addr <= Memory::NEW_LINEAR_HEAP_VADDR_END) {
return addr - Memory::NEW_LINEAR_HEAP_VADDR + Memory::FCRAM_PADDR;
}
if (addr >= Memory::PLUGIN_3GX_FB_VADDR && addr <= Memory::PLUGIN_3GX_FB_VADDR_END) {
return addr - Memory::PLUGIN_3GX_FB_VADDR + Service::PLGLDR::PLG_LDR::GetPluginFBAddr();
}
LOG_ERROR(HW_Memory, "Unknown virtual address @ 0x{:08X}", addr);
// To help with debugging, set bit on address so that it's obviously invalid.
// TODO: find the correct way to handle this error
return addr | 0x80000000;
}
u32 GSP_GPU::GetUnusedThreadId() const {
for (u32 id = 0; id < MaxGSPThreads; ++id) {
if (!used_thread_ids[id])
if (!used_thread_ids[id]) {
return id;
}
}
UNREACHABLE_MSG("All GSP threads are in use");
return 0;
}
/// Gets a pointer to a thread command buffer in GSP shared memory
static inline u8* GetCommandBuffer(std::shared_ptr<Kernel::SharedMemory> shared_memory,
u32 thread_id) {
return shared_memory->GetPointer(0x800 + (thread_id * sizeof(CommandBuffer)));
CommandBuffer* GSP_GPU::GetCommandBuffer(u32 thread_id) {
auto* ptr = shared_memory->GetPointer(0x800 + (thread_id * sizeof(CommandBuffer)));
return reinterpret_cast<CommandBuffer*>(ptr);
}
FrameBufferUpdate* GSP_GPU::GetFrameBufferInfo(u32 thread_id, u32 screen_index) {
DEBUG_ASSERT_MSG(screen_index < 2, "Invalid screen index");
// For each thread there are two FrameBufferUpdate fields
u32 offset = 0x200 + (2 * thread_id + screen_index) * sizeof(FrameBufferUpdate);
const u32 offset = 0x200 + (2 * thread_id + screen_index) * sizeof(FrameBufferUpdate);
u8* ptr = shared_memory->GetPointer(offset);
return reinterpret_cast<FrameBufferUpdate*>(ptr);
}
/// Gets a pointer to the interrupt relay queue for a given thread index
static inline InterruptRelayQueue* GetInterruptRelayQueue(
std::shared_ptr<Kernel::SharedMemory> shared_memory, u32 thread_id) {
InterruptRelayQueue* GSP_GPU::GetInterruptRelayQueue(u32 thread_id) {
u8* ptr = shared_memory->GetPointer(sizeof(InterruptRelayQueue) * thread_id);
return reinterpret_cast<InterruptRelayQueue*>(ptr);
}
@ -125,19 +88,6 @@ void GSP_GPU::ClientDisconnected(std::shared_ptr<Kernel::ServerSession> server_s
SessionRequestHandler::ClientDisconnected(server_session);
}
/**
* Writes a single GSP GPU hardware registers with a single u32 value
* (For internal use.)
*
* @param base_address The address of the register in question
* @param data Data to be written
*/
static void WriteSingleHWReg(u32 base_address, u32 data) {
DEBUG_ASSERT_MSG((base_address & 3) == 0 && base_address < 0x420000,
"Write address out of range or misaligned");
HW::Write<u32>(base_address + REGS_BEGIN, data);
}
/**
* Writes sequential GSP GPU hardware registers using an array of source data
*
@ -146,7 +96,8 @@ static void WriteSingleHWReg(u32 base_address, u32 data) {
* @param data A vector containing the source data
* @return RESULT_SUCCESS if the parameters are valid, error code otherwise
*/
static ResultCode WriteHWRegs(u32 base_address, u32 size_in_bytes, std::span<const u8> data) {
static ResultCode WriteHWRegs(u32 base_address, u32 size_in_bytes, std::span<const u8> data,
VideoCore::GPU& gpu) {
// This magic number is verified to be done by the gsp module
const u32 max_size_in_bytes = 0x80;
@ -155,28 +106,30 @@ static ResultCode WriteHWRegs(u32 base_address, u32 size_in_bytes, std::span<con
"Write address was out of range or misaligned! (address=0x{:08x}, size=0x{:08x})",
base_address, size_in_bytes);
return ERR_REGS_OUTOFRANGE_OR_MISALIGNED;
} else if (size_in_bytes <= max_size_in_bytes) {
if (size_in_bytes & 3) {
LOG_ERROR(Service_GSP, "Misaligned size 0x{:08x}", size_in_bytes);
return ERR_REGS_MISALIGNED;
} else {
std::size_t offset = 0;
while (size_in_bytes > 0) {
u32 value;
std::memcpy(&value, &data[offset], sizeof(u32));
WriteSingleHWReg(base_address, value);
}
size_in_bytes -= 4;
offset += 4;
base_address += 4;
}
return RESULT_SUCCESS;
}
} else {
if (size_in_bytes > max_size_in_bytes) {
LOG_ERROR(Service_GSP, "Out of range size 0x{:08x}", size_in_bytes);
return ERR_REGS_INVALID_SIZE;
}
if (size_in_bytes & 3) {
LOG_ERROR(Service_GSP, "Misaligned size 0x{:08x}", size_in_bytes);
return ERR_REGS_MISALIGNED;
}
std::size_t offset = 0;
while (size_in_bytes > 0) {
u32 value;
std::memcpy(&value, &data[offset], sizeof(u32));
gpu.WriteReg(REGS_BEGIN + base_address, value);
size_in_bytes -= 4;
offset += 4;
base_address += 4;
}
return RESULT_SUCCESS;
}
/**
@ -190,7 +143,7 @@ static ResultCode WriteHWRegs(u32 base_address, u32 size_in_bytes, std::span<con
* @return RESULT_SUCCESS if the parameters are valid, error code otherwise
*/
static ResultCode WriteHWRegsWithMask(u32 base_address, u32 size_in_bytes, std::span<const u8> data,
std::span<const u8> masks) {
std::span<const u8> masks, VideoCore::GPU& gpu) {
// This magic number is verified to be done by the gsp module
const u32 max_size_in_bytes = 0x80;
@ -199,60 +152,58 @@ static ResultCode WriteHWRegsWithMask(u32 base_address, u32 size_in_bytes, std::
"Write address was out of range or misaligned! (address=0x{:08x}, size=0x{:08x})",
base_address, size_in_bytes);
return ERR_REGS_OUTOFRANGE_OR_MISALIGNED;
} else if (size_in_bytes <= max_size_in_bytes) {
if (size_in_bytes & 3) {
LOG_ERROR(Service_GSP, "Misaligned size 0x{:08x}", size_in_bytes);
return ERR_REGS_MISALIGNED;
} else {
std::size_t offset = 0;
while (size_in_bytes > 0) {
const u32 reg_address = base_address + REGS_BEGIN;
}
u32 reg_value;
HW::Read<u32>(reg_value, reg_address);
u32 value, mask;
std::memcpy(&value, &data[offset], sizeof(u32));
std::memcpy(&mask, &masks[offset], sizeof(u32));
// Update the current value of the register only for set mask bits
reg_value = (reg_value & ~mask) | (value & mask);
WriteSingleHWReg(base_address, reg_value);
size_in_bytes -= 4;
offset += 4;
base_address += 4;
}
return RESULT_SUCCESS;
}
} else {
if (size_in_bytes > max_size_in_bytes) {
LOG_ERROR(Service_GSP, "Out of range size 0x{:08x}", size_in_bytes);
return ERR_REGS_INVALID_SIZE;
}
if (size_in_bytes & 3) {
LOG_ERROR(Service_GSP, "Misaligned size 0x{:08x}", size_in_bytes);
return ERR_REGS_MISALIGNED;
}
std::size_t offset = 0;
while (size_in_bytes > 0) {
const u32 reg_address = base_address + REGS_BEGIN;
u32 reg_value = gpu.ReadReg(reg_address);
u32 value, mask;
std::memcpy(&value, &data[offset], sizeof(u32));
std::memcpy(&mask, &masks[offset], sizeof(u32));
// Update the current value of the register only for set mask bits
reg_value = (reg_value & ~mask) | (value & mask);
gpu.WriteReg(reg_address, reg_value);
size_in_bytes -= 4;
offset += 4;
base_address += 4;
}
return RESULT_SUCCESS;
}
void GSP_GPU::WriteHWRegs(Kernel::HLERequestContext& ctx) {
IPC::RequestParser rp(ctx);
u32 reg_addr = rp.Pop<u32>();
u32 size = rp.Pop<u32>();
std::vector<u8> src_data = rp.PopStaticBuffer();
const u32 reg_addr = rp.Pop<u32>();
const u32 size = rp.Pop<u32>();
const auto src_data = rp.PopStaticBuffer();
IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
rb.Push(GSP::WriteHWRegs(reg_addr, size, src_data));
rb.Push(GSP::WriteHWRegs(reg_addr, size, src_data, system.GPU()));
}
void GSP_GPU::WriteHWRegsWithMask(Kernel::HLERequestContext& ctx) {
IPC::RequestParser rp(ctx);
u32 reg_addr = rp.Pop<u32>();
u32 size = rp.Pop<u32>();
std::vector<u8> src_data = rp.PopStaticBuffer();
std::vector<u8> mask_data = rp.PopStaticBuffer();
const u32 reg_addr = rp.Pop<u32>();
const u32 size = rp.Pop<u32>();
const auto src_data = rp.PopStaticBuffer();
const auto mask_data = rp.PopStaticBuffer();
IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
rb.Push(GSP::WriteHWRegsWithMask(reg_addr, size, src_data, mask_data));
rb.Push(GSP::WriteHWRegsWithMask(reg_addr, size, src_data, mask_data, system.GPU()));
}
void GSP_GPU::ReadHWRegs(Kernel::HLERequestContext& ctx) {
@ -270,7 +221,7 @@ void GSP_GPU::ReadHWRegs(Kernel::HLERequestContext& ctx) {
return;
}
// size should be word-aligned
// Size should be word-aligned
if ((size % 4) != 0) {
IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
rb.Push(ERR_REGS_MISALIGNED);
@ -279,8 +230,9 @@ void GSP_GPU::ReadHWRegs(Kernel::HLERequestContext& ctx) {
}
std::vector<u8> buffer(size);
for (u32 offset = 0; offset < size; ++offset) {
HW::Read<u8>(buffer[offset], REGS_BEGIN + reg_addr + offset);
for (u32 word = 0; word < size / sizeof(u32); ++word) {
const u32 data = system.GPU().ReadReg(REGS_BEGIN + reg_addr + word * sizeof(u32));
std::memcpy(buffer.data() + word * sizeof(u32), &data, sizeof(u32));
}
IPC::RequestBuilder rb = rp.MakeBuilder(1, 2);
@ -288,53 +240,15 @@ void GSP_GPU::ReadHWRegs(Kernel::HLERequestContext& ctx) {
rb.PushStaticBuffer(std::move(buffer), 0);
}
ResultCode SetBufferSwap(u32 screen_id, const FrameBufferInfo& info) {
u32 base_address = 0x400000;
PAddr phys_address_left = VirtualToPhysicalAddress(info.address_left);
PAddr phys_address_right = VirtualToPhysicalAddress(info.address_right);
if (info.active_fb == 0) {
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(
screen_id, address_left1)),
phys_address_left);
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(
screen_id, address_right1)),
phys_address_right);
} else {
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(
screen_id, address_left2)),
phys_address_left);
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(
screen_id, address_right2)),
phys_address_right);
}
WriteSingleHWReg(base_address +
4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(screen_id, stride)),
info.stride);
WriteSingleHWReg(base_address +
4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(screen_id, color_format)),
info.format);
WriteSingleHWReg(base_address +
4 * static_cast<u32>(GPU_FRAMEBUFFER_REG_INDEX(screen_id, active_fb)),
info.shown_fb);
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::BufferSwapped, nullptr);
if (screen_id == 0) {
MicroProfileFlip();
Core::System::GetInstance().perf_stats->EndGameFrame();
}
return RESULT_SUCCESS;
}
void GSP_GPU::SetBufferSwap(Kernel::HLERequestContext& ctx) {
IPC::RequestParser rp(ctx);
u32 screen_id = rp.Pop<u32>();
auto fb_info = rp.PopRaw<FrameBufferInfo>();
system.GPU().SetBufferSwap(screen_id, fb_info);
IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
rb.Push(GSP::SetBufferSwap(screen_id, fb_info));
rb.Push(RESULT_SUCCESS);
}
void GSP_GPU::FlushDataCache(Kernel::HLERequestContext& ctx) {
@ -382,10 +296,9 @@ void GSP_GPU::RegisterInterruptRelayQueue(Kernel::HLERequestContext& ctx) {
u32 flags = rp.Pop<u32>();
auto interrupt_event = rp.PopObject<Kernel::Event>();
// TODO(mailwl): return right error code instead assert
ASSERT_MSG((interrupt_event != nullptr), "handle is not valid!");
ASSERT_MSG(interrupt_event, "handle is not valid!");
interrupt_event->SetName("GSP_GSP_GPU::interrupt_event");
interrupt_event->SetName("GSP_GPU::interrupt_event");
SessionData* session_data = GetSessionData(ctx.Session());
session_data->interrupt_event = std::move(interrupt_event);
@ -422,15 +335,17 @@ void GSP_GPU::UnregisterInterruptRelayQueue(Kernel::HLERequestContext& ctx) {
void GSP_GPU::SignalInterruptForThread(InterruptId interrupt_id, u32 thread_id) {
SessionData* session_data = FindRegisteredThreadData(thread_id);
if (session_data == nullptr)
if (!session_data) {
return;
}
auto interrupt_event = session_data->interrupt_event;
if (interrupt_event == nullptr) {
LOG_WARNING(Service_GSP, "cannot synchronize until GSP event has been created!");
return;
}
InterruptRelayQueue* interrupt_relay_queue = GetInterruptRelayQueue(shared_memory, thread_id);
auto* interrupt_relay_queue = GetInterruptRelayQueue(thread_id);
u8 next = interrupt_relay_queue->index;
next += interrupt_relay_queue->number_interrupts;
next = next % 0x34; // 0x34 is the number of interrupt slots
@ -441,29 +356,20 @@ void GSP_GPU::SignalInterruptForThread(InterruptId interrupt_id, u32 thread_id)
interrupt_relay_queue->error_code = 0x0; // No error
// Update framebuffer information if requested
// TODO(yuriks): Confirm where this code should be called. It is definitely updated without
// executing any GSP commands, only waiting on the event.
// TODO(Subv): The real GSP module triggers PDC0 after updating both the top and bottom
// screen, it is currently unknown what PDC1 does.
int screen_id = (interrupt_id == InterruptId::PDC0) ? 0
: (interrupt_id == InterruptId::PDC1) ? 1
: -1;
const s32 screen_id = (interrupt_id == InterruptId::PDC0) ? 0
: (interrupt_id == InterruptId::PDC1) ? 1
: -1;
if (screen_id != -1) {
FrameBufferUpdate* info = GetFrameBufferInfo(thread_id, screen_id);
auto* info = GetFrameBufferInfo(thread_id, screen_id);
if (info->is_dirty) {
GSP::SetBufferSwap(screen_id, info->framebuffer_info[info->index]);
system.GPU().SetBufferSwap(screen_id, info->framebuffer_info[info->index]);
info->is_dirty.Assign(false);
}
}
interrupt_event->Signal();
}
/**
* Signals that the specified interrupt type has occurred to userland code
* @param interrupt_id ID of interrupt that is being signalled
* @todo This should probably take a thread_id parameter and only signal this thread?
* @todo This probably does not belong in the GSP module, instead move to video_core
*/
void GSP_GPU::SignalInterrupt(InterruptId interrupt_id) {
if (nullptr == shared_memory) {
LOG_WARNING(Service_GSP, "cannot synchronize until GSP shared memory has been created!");
@ -488,154 +394,13 @@ void GSP_GPU::SignalInterrupt(InterruptId interrupt_id) {
SignalInterruptForThread(interrupt_id, active_thread_id);
}
MICROPROFILE_DEFINE(GPU_GSP_DMA, "GPU", "GSP DMA", MP_RGB(100, 0, 255));
/// Executes the next GSP command
static void ExecuteCommand(const Command& command, u32 thread_id) {
// Utility function to convert register ID to address
static auto WriteGPURegister = [](u32 id, u32 data) {
GPU::Write<u32>(0x1EF00000 + 4 * id, data);
};
switch (command.id) {
// GX request DMA - typically used for copying memory from GSP heap to VRAM
case CommandId::REQUEST_DMA: {
MICROPROFILE_SCOPE(GPU_GSP_DMA);
Memory::MemorySystem& memory = Core::System::GetInstance().Memory();
// TODO: Consider attempting rasterizer-accelerated surface blit if that usage is ever
// possible/likely
Memory::RasterizerFlushVirtualRegion(command.dma_request.source_address,
command.dma_request.size, Memory::FlushMode::Flush);
Memory::RasterizerFlushVirtualRegion(command.dma_request.dest_address,
command.dma_request.size,
Memory::FlushMode::Invalidate);
// TODO(Subv): These memory accesses should not go through the application's memory mapping.
// They should go through the GSP module's memory mapping.
memory.CopyBlock(*Core::System::GetInstance().Kernel().GetCurrentProcess(),
command.dma_request.dest_address, command.dma_request.source_address,
command.dma_request.size);
SignalInterrupt(InterruptId::DMA);
break;
}
// TODO: This will need some rework in the future. (why?)
case CommandId::SUBMIT_GPU_CMDLIST: {
auto& params = command.submit_gpu_cmdlist;
if (params.do_flush) {
// This flag flushes the command list (params.address, params.size) from the cache.
// Command lists are not processed by the hardware renderer, so we don't need to
// actually flush them in Citra.
}
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.address)),
VirtualToPhysicalAddress(params.address) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.size)),
params.size);
// TODO: Not sure if we are supposed to always write this .. seems to trigger processing
// though
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.trigger)), 1);
// TODO(yuriks): Figure out the meaning of the `flags` field.
break;
}
// It's assumed that the two "blocks" behave equivalently.
// Presumably this is done simply to allow two memory fills to run in parallel.
case CommandId::SET_MEMORY_FILL: {
auto& params = command.memory_fill;
if (params.start1 != 0) {
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_start)),
VirtualToPhysicalAddress(params.start1) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_end)),
VirtualToPhysicalAddress(params.end1) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].value_32bit)),
params.value1);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].control)),
params.control1);
}
if (params.start2 != 0) {
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_start)),
VirtualToPhysicalAddress(params.start2) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_end)),
VirtualToPhysicalAddress(params.end2) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].value_32bit)),
params.value2);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].control)),
params.control2);
}
break;
}
case CommandId::SET_DISPLAY_TRANSFER: {
auto& params = command.display_transfer;
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.input_address)),
VirtualToPhysicalAddress(params.in_buffer_address) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.output_address)),
VirtualToPhysicalAddress(params.out_buffer_address) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.input_size)),
params.in_buffer_size);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.output_size)),
params.out_buffer_size);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.flags)),
params.flags);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(display_transfer_config.trigger)), 1);
break;
}
case CommandId::SET_TEXTURE_COPY: {
auto& params = command.texture_copy;
WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.input_address),
VirtualToPhysicalAddress(params.in_buffer_address) >> 3);
WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.output_address),
VirtualToPhysicalAddress(params.out_buffer_address) >> 3);
WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.texture_copy.size),
params.size);
WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.texture_copy.input_size),
params.in_width_gap);
WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.texture_copy.output_size),
params.out_width_gap);
WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.flags), params.flags);
// NOTE: Actual GSP ORs 1 with current register instead of overwriting. Doesn't seem to
// matter.
WriteGPURegister((u32)GPU_REG_INDEX(display_transfer_config.trigger), 1);
break;
}
case CommandId::CACHE_FLUSH: {
// NOTE: Rasterizer flushing handled elsewhere in CPU read/write and other GPU handlers
// Use command.cache_flush.regions to implement this handler
break;
}
default:
LOG_ERROR(Service_GSP, "unknown command 0x{:08X}", (int)command.id.Value());
}
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::GSPCommandProcessed,
(void*)&command);
}
void GSP_GPU::SetLcdForceBlack(Kernel::HLERequestContext& ctx) {
IPC::RequestParser rp(ctx);
const bool enable_black = rp.Pop<bool>();
bool enable_black = rp.Pop<bool>();
LCD::Regs::ColorFill data = {0};
// Since data is already zeroed, there is no need to explicitly set
// the color to black (all zero).
Pica::ColorFill data{};
data.is_enabled.Assign(enable_black);
LCD::Write(HW::VADDR_LCD + 4 * LCD_REG_INDEX(color_fill_top), data.raw); // Top LCD
LCD::Write(HW::VADDR_LCD + 4 * LCD_REG_INDEX(color_fill_bottom), data.raw); // Bottom LCD
system.GPU().SetColorFill(data);
IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);
rb.Push(RESULT_SUCCESS);
@ -644,20 +409,17 @@ void GSP_GPU::SetLcdForceBlack(Kernel::HLERequestContext& ctx) {
void GSP_GPU::TriggerCmdReqQueue(Kernel::HLERequestContext& ctx) {
IPC::RequestParser rp(ctx);
// Iterate through each thread's command queue...
for (unsigned thread_id = 0; thread_id < 0x4; ++thread_id) {
CommandBuffer* command_buffer = (CommandBuffer*)GetCommandBuffer(shared_memory, thread_id);
// Iterate through each command.
auto* command_buffer = GetCommandBuffer(active_thread_id);
auto& gpu = system.GPU();
for (u32 i = 0; i < command_buffer->number_commands; i++) {
gpu.Debugger().GXCommandProcessed(command_buffer->commands[i]);
// Iterate through each command...
for (unsigned i = 0; i < command_buffer->number_commands; ++i) {
g_debugger.GXCommandProcessed((u8*)&command_buffer->commands[i]);
// Decode and execute command
gpu.Execute(command_buffer->commands[i]);
// Decode and execute command
ExecuteCommand(command_buffer->commands[i], thread_id);
// Indicates that command has completed
command_buffer->number_commands.Assign(command_buffer->number_commands - 1);
}
// Indicates that command has completed
command_buffer->number_commands.Assign(command_buffer->number_commands - 1);
}
IPC::RequestBuilder rb = rp.MakeBuilder(1, 0);

146
src/core/hle/service/gsp/gsp_gpu.h
View file

@ -13,7 +13,8 @@
#include "common/common_types.h"
#include "core/hle/kernel/event.h"
#include "core/hle/kernel/hle_ipc.h"
#include "core/hle/result.h"
#include "core/hle/service/gsp/gsp_command.h"
#include "core/hle/service/gsp/gsp_interrupt.h"
#include "core/hle/service/service.h"
namespace Core {
@ -28,53 +29,6 @@ class SharedMemory;
namespace Service::GSP {
/// GSP interrupt ID
enum class InterruptId : u8 {
PSC0 = 0x00,
PSC1 = 0x01,
PDC0 = 0x02, // Seems called every vertical screen line
PDC1 = 0x03, // Seems called every frame
PPF = 0x04,
P3D = 0x05,
DMA = 0x06,
};
/// GSP command ID
enum class CommandId : u32 {
REQUEST_DMA = 0x00,
/// Submits a commandlist for execution by the GPU.
SUBMIT_GPU_CMDLIST = 0x01,
// Fills a given memory range with a particular value
SET_MEMORY_FILL = 0x02,
// Copies an image and optionally performs color-conversion or scaling.
// This is highly similar to the GameCube's EFB copy feature
SET_DISPLAY_TRANSFER = 0x03,
// Conceptionally similar to SET_DISPLAY_TRANSFER and presumable uses the same hardware path
SET_TEXTURE_COPY = 0x04,
/// Flushes up to 3 cache regions in a single command.
CACHE_FLUSH = 0x05,
};
/// GSP thread interrupt relay queue
struct InterruptRelayQueue {
// Index of last interrupt in the queue
u8 index;
// Number of interrupts remaining to be processed by the userland code
u8 number_interrupts;
// Error code - zero on success, otherwise an error has occurred
u8 error_code;
u8 padding1;
u32 missed_PDC0;
u32 missed_PDC1;
InterruptId slot[0x34]; ///< Interrupt ID slots
};
static_assert(sizeof(InterruptRelayQueue) == 0x40, "InterruptRelayQueue struct has incorrect size");
struct FrameBufferInfo {
u32 active_fb; // 0 = first, 1 = second
u32 address_left;
@ -96,95 +50,9 @@ struct FrameBufferUpdate {
u32 pad2;
};
static_assert(sizeof(FrameBufferUpdate) == 0x40, "Struct has incorrect size");
// TODO: Not sure if this padding is correct.
// Chances are the second block is stored at offset 0x24 rather than 0x20.
static_assert(offsetof(FrameBufferUpdate, framebuffer_info[1]) == 0x20,
"FrameBufferInfo element has incorrect alignment");
/// GSP command
struct Command {
BitField<0, 8, CommandId> id;
union {
struct {
u32 source_address;
u32 dest_address;
u32 size;
} dma_request;
struct {
u32 address;
u32 size;
u32 flags;
u32 unused[3];
u32 do_flush;
} submit_gpu_cmdlist;
struct {
u32 start1;
u32 value1;
u32 end1;
u32 start2;
u32 value2;
u32 end2;
u16 control1;
u16 control2;
} memory_fill;
struct {
u32 in_buffer_address;
u32 out_buffer_address;
u32 in_buffer_size;
u32 out_buffer_size;
u32 flags;
} display_transfer;
struct {
u32 in_buffer_address;
u32 out_buffer_address;
u32 size;
u32 in_width_gap;
u32 out_width_gap;
u32 flags;
} texture_copy;
struct {
struct {
u32 address;
u32 size;
} regions[3];
} cache_flush;
u8 raw_data[0x1C];
};
};
static_assert(sizeof(Command) == 0x20, "Command struct has incorrect size");
/// GSP shared memory GX command buffer header
struct CommandBuffer {
union {
u32 hex;
// Current command index. This index is updated by GSP module after loading the command
// data, right before the command is processed. When this index is updated by GSP module,
// the total commands field is decreased by one as well.
BitField<0, 8, u32> index;
// Total commands to process, must not be value 0 when GSP module handles commands. This
// must be <=15 when writing a command to shared memory. This is incremented by the
// application when writing a command to shared memory, after increasing this value
// TriggerCmdReqQueue is only used if this field is value 1.
BitField<8, 8, u32> number_commands;
};
u32 unk[7];
Command commands[0xF];
};
static_assert(sizeof(CommandBuffer) == 0x200, "CommandBuffer struct has incorrect size");
constexpr u32 FRAMEBUFFER_WIDTH = 240;
constexpr u32 FRAMEBUFFER_WIDTH_POW2 = 256;
constexpr u32 TOP_FRAMEBUFFER_HEIGHT = 400;
@ -242,6 +110,12 @@ public:
*/
FrameBufferUpdate* GetFrameBufferInfo(u32 thread_id, u32 screen_index);
/// Gets a pointer to a thread command buffer in GSP shared memory
CommandBuffer* GetCommandBuffer(u32 thread_id);
/// Gets a pointer to the interrupt relay queue for a given thread index
InterruptRelayQueue* GetInterruptRelayQueue(u32 thread_id);
/**
* Retreives the ID of the thread with GPU rights.
*/
@ -513,7 +387,7 @@ private:
static constexpr u32 MaxGSPThreads = 4;
/// Thread ids currently in use by the sessions connected to the GSPGPU service.
std::array<bool, MaxGSPThreads> used_thread_ids = {false, false, false, false};
std::array<bool, MaxGSPThreads> used_thread_ids{};
friend class SessionData;
@ -522,8 +396,6 @@ private:
friend class boost::serialization::access;
};
ResultCode SetBufferSwap(u32 screen_id, const FrameBufferInfo& info);
} // namespace Service::GSP
BOOST_CLASS_EXPORT_KEY(Service::GSP::SessionData)

42
src/core/hle/service/gsp/gsp_interrupt.h Normal file
View file

@ -0,0 +1,42 @@
// Copyright 2023 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <functional>
#include "common/common_types.h"
namespace Service::GSP {
/// GSP interrupt ID
enum class InterruptId : u8 {
PSC0 = 0x00,
PSC1 = 0x01,
PDC0 = 0x02,
PDC1 = 0x03,
PPF = 0x04,
P3D = 0x05,
DMA = 0x06,
};
/// GSP thread interrupt relay queue
struct InterruptRelayQueue {
// Index of last interrupt in the queue
u8 index;
// Number of interrupts remaining to be processed by the userland code
u8 number_interrupts;
// Error code - zero on success, otherwise an error has occurred
u8 error_code;
u8 padding1;
u32 missed_PDC0;
u32 missed_PDC1;
InterruptId slot[0x34]; ///< Interrupt ID slots
};
static_assert(sizeof(InterruptRelayQueue) == 0x40, "InterruptRelayQueue struct has incorrect size");
using InterruptHandler = std::function<void(InterruptId)>;
} // namespace Service::GSP

1
src/core/hle/service/hid/hid.cpp
View file

@ -22,7 +22,6 @@
#include "core/hle/service/hid/hid_user.h"
#include "core/hle/service/service.h"
#include "core/movie.h"
#include "video_core/video_core.h"
SERVICE_CONSTRUCT_IMPL(Service::HID::Module)
SERIALIZE_EXPORT_IMPL(Service::HID::Module)

572
src/core/hw/gpu.cpp
View file

@ -1,572 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include <numeric>
#include <type_traits>
#include "common/alignment.h"
#include "common/color.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "common/microprofile.h"
#include "common/vector_math.h"
#include "core/core.h"
#include "core/core_timing.h"
#include "core/hle/service/gsp/gsp.h"
#include "core/hw/gpu.h"
#include "core/hw/hw.h"
#include "core/memory.h"
#include "core/tracer/recorder.h"
#include "video_core/command_processor.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_base.h"
#include "video_core/utils.h"
#include "video_core/video_core.h"
namespace GPU {
Regs g_regs;
Memory::MemorySystem* g_memory;
/// Event id for CoreTiming
static Core::TimingEventType* vblank_event;
template <typename T>
inline void Read(T& var, const u32 raw_addr) {
u32 addr = raw_addr - HW::VADDR_GPU;
u32 index = addr / 4;
// Reads other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Read{} @ {:#010X}", sizeof(var) * 8, addr);
return;
}
var = g_regs[addr / 4];
}
static Common::Vec4<u8> DecodePixel(Regs::PixelFormat input_format, const u8* src_pixel) {
switch (input_format) {
case Regs::PixelFormat::RGBA8:
return Common::Color::DecodeRGBA8(src_pixel);
case Regs::PixelFormat::RGB8:
return Common::Color::DecodeRGB8(src_pixel);
case Regs::PixelFormat::RGB565:
return Common::Color::DecodeRGB565(src_pixel);
case Regs::PixelFormat::RGB5A1:
return Common::Color::DecodeRGB5A1(src_pixel);
case Regs::PixelFormat::RGBA4:
return Common::Color::DecodeRGBA4(src_pixel);
default:
LOG_ERROR(HW_GPU, "Unknown source framebuffer format {:x}", input_format);
return {0, 0, 0, 0};
}
}
MICROPROFILE_DEFINE(GPU_DisplayTransfer, "GPU", "DisplayTransfer", MP_RGB(100, 100, 255));
MICROPROFILE_DEFINE(GPU_CmdlistProcessing, "GPU", "Cmdlist Processing", MP_RGB(100, 255, 100));
static void MemoryFill(const Regs::MemoryFillConfig& config) {
const PAddr start_addr = config.GetStartAddress();
const PAddr end_addr = config.GetEndAddress();
// TODO: do hwtest with these cases
if (!g_memory->IsValidPhysicalAddress(start_addr)) {
LOG_CRITICAL(HW_GPU, "invalid start address {:#010X}", start_addr);
return;
}
if (!g_memory->IsValidPhysicalAddress(end_addr)) {
LOG_CRITICAL(HW_GPU, "invalid end address {:#010X}", end_addr);
return;
}
if (end_addr <= start_addr) {
LOG_CRITICAL(HW_GPU, "invalid memory range from {:#010X} to {:#010X}", start_addr,
end_addr);
return;
}
u8* start = g_memory->GetPhysicalPointer(start_addr);
u8* end = g_memory->GetPhysicalPointer(end_addr);
if (VideoCore::g_renderer->Rasterizer()->AccelerateFill(config))
return;
Memory::RasterizerInvalidateRegion(config.GetStartAddress(),
config.GetEndAddress() - config.GetStartAddress());
if (config.fill_24bit) {
// fill with 24-bit values
for (u8* ptr = start; ptr < end; ptr += 3) {
ptr[0] = config.value_24bit_r;
ptr[1] = config.value_24bit_g;
ptr[2] = config.value_24bit_b;
}
} else if (config.fill_32bit) {
// fill with 32-bit values
if (end > start) {
u32 value = config.value_32bit;
std::size_t len = (end - start) / sizeof(u32);
for (std::size_t i = 0; i < len; ++i)
std::memcpy(&start[i * sizeof(u32)], &value, sizeof(u32));
}
} else {
// fill with 16-bit values
u16 value_16bit = config.value_16bit.Value();
for (u8* ptr = start; ptr < end; ptr += sizeof(u16))
std::memcpy(ptr, &value_16bit, sizeof(u16));
}
}
static void DisplayTransfer(const Regs::DisplayTransferConfig& config) {
const PAddr src_addr = config.GetPhysicalInputAddress();
PAddr dst_addr = config.GetPhysicalOutputAddress();
// TODO: do hwtest with these cases
if (!g_memory->IsValidPhysicalAddress(src_addr)) {
LOG_CRITICAL(HW_GPU, "invalid input address {:#010X}", src_addr);
return;
}
if (!g_memory->IsValidPhysicalAddress(dst_addr)) {
LOG_CRITICAL(HW_GPU, "invalid output address {:#010X}", dst_addr);
return;
}
if (config.input_width == 0) {
LOG_CRITICAL(HW_GPU, "zero input width");
return;
}
if (config.input_height == 0) {
LOG_CRITICAL(HW_GPU, "zero input height");
return;
}
if (config.output_width == 0) {
LOG_CRITICAL(HW_GPU, "zero output width");
return;
}
if (config.output_height == 0) {
LOG_CRITICAL(HW_GPU, "zero output height");
return;
}
if (VideoCore::g_renderer->Rasterizer()->AccelerateDisplayTransfer(config))
return;
// Using flip_vertically alongside crop_input_lines produces skewed output on hardware.
// We have to emulate this because some games rely on this behaviour to render correctly.
if (config.flip_vertically && config.crop_input_lines &&
config.input_width > config.output_width) {
dst_addr += (config.input_width - config.output_width) * (config.output_height - 1) *
GPU::Regs::BytesPerPixel(config.output_format);
}
u8* src_pointer = g_memory->GetPhysicalPointer(src_addr);
u8* dst_pointer = g_memory->GetPhysicalPointer(dst_addr);
if (config.scaling > config.ScaleXY) {
LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode {}",
config.scaling.Value());
UNIMPLEMENTED();
return;
}
if (config.input_linear && config.scaling != config.NoScale) {
LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
UNIMPLEMENTED();
return;
}
int horizontal_scale = config.scaling != config.NoScale ? 1 : 0;
int vertical_scale = config.scaling == config.ScaleXY ? 1 : 0;
u32 output_width = config.output_width >> horizontal_scale;
u32 output_height = config.output_height >> vertical_scale;
u32 input_size =
config.input_width * config.input_height * GPU::Regs::BytesPerPixel(config.input_format);
u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), input_size);
Memory::RasterizerInvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
Common::Vec4<u8> src_color;
// Calculate the [x,y] position of the input image
// based on the current output position and the scale
u32 input_x = x << horizontal_scale;
u32 input_y = y << vertical_scale;
u32 output_y;
if (config.flip_vertically) {
// Flip the y value of the output data,
// we do this after calculating the [x,y] position of the input image
// to account for the scaling options.
output_y = output_height - y - 1;
} else {
output_y = y;
}
u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
u32 src_offset;
u32 dst_offset;
if (config.input_linear) {
if (!config.dont_swizzle) {
// Interpret the input as linear and the output as tiled
u32 coarse_y = output_y & ~7;
u32 stride = output_width * dst_bytes_per_pixel;
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = VideoCore::GetMortonOffset(x, output_y, dst_bytes_per_pixel) +
coarse_y * stride;
} else {
// Both input and output are linear
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = (x + output_y * output_width) * dst_bytes_per_pixel;
}
} else {
if (!config.dont_swizzle) {
// Interpret the input as tiled and the output as linear
u32 coarse_y = input_y & ~7;
u32 stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) +
coarse_y * stride;
dst_offset = (x + output_y * output_width) * dst_bytes_per_pixel;
} else {
// Both input and output are tiled
u32 out_coarse_y = output_y & ~7;
u32 out_stride = output_width * dst_bytes_per_pixel;
u32 in_coarse_y = input_y & ~7;
u32 in_stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) +
in_coarse_y * in_stride;
dst_offset = VideoCore::GetMortonOffset(x, output_y, dst_bytes_per_pixel) +
out_coarse_y * out_stride;
}
}
const u8* src_pixel = src_pointer + src_offset;
src_color = DecodePixel(config.input_format, src_pixel);
if (config.scaling == config.ScaleX) {
Common::Vec4<u8> pixel =
DecodePixel(config.input_format, src_pixel + src_bytes_per_pixel);
src_color = ((src_color + pixel) / 2).Cast<u8>();
} else if (config.scaling == config.ScaleXY) {
Common::Vec4<u8> pixel1 =
DecodePixel(config.input_format, src_pixel + 1 * src_bytes_per_pixel);
Common::Vec4<u8> pixel2 =
DecodePixel(config.input_format, src_pixel + 2 * src_bytes_per_pixel);
Common::Vec4<u8> pixel3 =
DecodePixel(config.input_format, src_pixel + 3 * src_bytes_per_pixel);
src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
}
u8* dst_pixel = dst_pointer + dst_offset;
switch (config.output_format) {
case Regs::PixelFormat::RGBA8:
Common::Color::EncodeRGBA8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB8:
Common::Color::EncodeRGB8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB565:
Common::Color::EncodeRGB565(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB5A1:
Common::Color::EncodeRGB5A1(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGBA4:
Common::Color::EncodeRGBA4(src_color, dst_pixel);
break;
default:
LOG_ERROR(HW_GPU, "Unknown destination framebuffer format {:x}",
static_cast<u32>(config.output_format.Value()));
break;
}
}
}
}
static void TextureCopy(const Regs::DisplayTransferConfig& config) {
const PAddr src_addr = config.GetPhysicalInputAddress();
const PAddr dst_addr = config.GetPhysicalOutputAddress();
// TODO: do hwtest with invalid addresses
if (!g_memory->IsValidPhysicalAddress(src_addr)) {
LOG_CRITICAL(HW_GPU, "invalid input address {:#010X}", src_addr);
return;
}
if (!g_memory->IsValidPhysicalAddress(dst_addr)) {
LOG_CRITICAL(HW_GPU, "invalid output address {:#010X}", dst_addr);
return;
}
if (VideoCore::g_renderer->Rasterizer()->AccelerateTextureCopy(config))
return;
u8* src_pointer = g_memory->GetPhysicalPointer(src_addr);
u8* dst_pointer = g_memory->GetPhysicalPointer(dst_addr);
u32 remaining_size = Common::AlignDown(config.texture_copy.size, 16);
if (remaining_size == 0) {
LOG_CRITICAL(HW_GPU, "zero size. Real hardware freezes on this.");
return;
}
u32 input_gap = config.texture_copy.input_gap * 16;
u32 output_gap = config.texture_copy.output_gap * 16;
// Zero gap means contiguous input/output even if width = 0. To avoid infinite loop below, width
// is assigned with the total size if gap = 0.
u32 input_width = input_gap == 0 ? remaining_size : config.texture_copy.input_width * 16;
u32 output_width = output_gap == 0 ? remaining_size : config.texture_copy.output_width * 16;
if (input_width == 0) {
LOG_CRITICAL(HW_GPU, "zero input width. Real hardware freezes on this.");
return;
}
if (output_width == 0) {
LOG_CRITICAL(HW_GPU, "zero output width. Real hardware freezes on this.");
return;
}
std::size_t contiguous_input_size =
config.texture_copy.size / input_width * (input_width + input_gap);
Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(),
static_cast<u32>(contiguous_input_size));
std::size_t contiguous_output_size =
config.texture_copy.size / output_width * (output_width + output_gap);
// Only need to flush output if it has a gap
const auto FlushInvalidate_fn = (output_gap != 0) ? Memory::RasterizerFlushAndInvalidateRegion
: Memory::RasterizerInvalidateRegion;
FlushInvalidate_fn(config.GetPhysicalOutputAddress(), static_cast<u32>(contiguous_output_size));
u32 remaining_input = input_width;
u32 remaining_output = output_width;
while (remaining_size > 0) {
u32 copy_size = std::min({remaining_input, remaining_output, remaining_size});
std::memcpy(dst_pointer, src_pointer, copy_size);
src_pointer += copy_size;
dst_pointer += copy_size;
remaining_input -= copy_size;
remaining_output -= copy_size;
remaining_size -= copy_size;
if (remaining_input == 0) {
remaining_input = input_width;
src_pointer += input_gap;
}
if (remaining_output == 0) {
remaining_output = output_width;
dst_pointer += output_gap;
}
}
}
template <typename T>
inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_GPU;
u32 index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Write{} {:#010X} @ {:#010X}", sizeof(data) * 8, (u32)data, addr);
return;
}
g_regs[index] = static_cast<u32>(data);
switch (index) {
// Memory fills are triggered once the fill value is written.
case GPU_REG_INDEX(memory_fill_config[0].trigger):
case GPU_REG_INDEX(memory_fill_config[1].trigger): {
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.trigger) {
MemoryFill(config);
LOG_TRACE(HW_GPU, "MemoryFill from {:#010X} to {:#010X}", config.GetStartAddress(),
config.GetEndAddress());
// It seems that it won't signal interrupt if "address_start" is zero.
// TODO: hwtest this
if (config.GetStartAddress() != 0) {
if (!is_second_filler) {
Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PSC0);
} else {
Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PSC1);
}
}
// Reset "trigger" flag and set the "finish" flag
// NOTE: This was confirmed to happen on hardware even if "address_start" is zero.
config.trigger.Assign(0);
config.finished.Assign(1);
}
break;
}
case GPU_REG_INDEX(display_transfer_config.trigger): {
MICROPROFILE_SCOPE(GPU_DisplayTransfer);
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer,
nullptr);
if (config.is_texture_copy) {
TextureCopy(config);
LOG_TRACE(HW_GPU,
"TextureCopy: {:#X} bytes from {:#010X}({}+{})-> "
"{:#010X}({}+{}), flags {:#010X}",
config.texture_copy.size, config.GetPhysicalInputAddress(),
config.texture_copy.input_width * 16, config.texture_copy.input_gap * 16,
config.GetPhysicalOutputAddress(), config.texture_copy.output_width * 16,
config.texture_copy.output_gap * 16, config.flags);
} else {
DisplayTransfer(config);
LOG_TRACE(HW_GPU,
"DisplayTransfer: {:#010X}({}x{})-> "
"{:#010X}({}x{}), dst format {:x}, flags {:#010X}",
config.GetPhysicalInputAddress(), config.input_width.Value(),
config.input_height.Value(), config.GetPhysicalOutputAddress(),
config.output_width.Value(), config.output_height.Value(),
static_cast<u32>(config.output_format.Value()), config.flags);
}
g_regs.display_transfer_config.trigger = 0;
Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PPF);
}
break;
}
// Seems like writing to this register triggers processing
case GPU_REG_INDEX(command_processor_config.trigger): {
const auto& config = g_regs.command_processor_config;
if (config.trigger & 1) {
MICROPROFILE_SCOPE(GPU_CmdlistProcessing);
Pica::CommandProcessor::ProcessCommandList(config.GetPhysicalAddress(), config.size);
g_regs.command_processor_config.trigger = 0;
}
break;
}
default:
break;
}
// Notify tracer about the register write
// This is happening *after* handling the write to make sure we properly catch all memory reads.
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
// addr + GPU VBase - IO VBase + IO PBase
Pica::g_debug_context->recorder->RegisterWritten<T>(
addr + 0x1EF00000 - 0x1EC00000 + 0x10100000, data);
}
}
// Explicitly instantiate template functions because we aren't defining this in the header:
template void Read<u64>(u64& var, const u32 addr);
template void Read<u32>(u32& var, const u32 addr);
template void Read<u16>(u16& var, const u32 addr);
template void Read<u8>(u8& var, const u32 addr);
template void Write<u64>(u32 addr, const u64 data);
template void Write<u32>(u32 addr, const u32 data);
template void Write<u16>(u32 addr, const u16 data);
template void Write<u8>(u32 addr, const u8 data);
/// Update hardware
static void VBlankCallback(std::uintptr_t user_data, s64 cycles_late) {
VideoCore::g_renderer->SwapBuffers();
// Signal to GSP that GPU interrupt has occurred
// TODO(yuriks): hwtest to determine if PDC0 is for the Top screen and PDC1 for the Sub
// screen, or if both use the same interrupts and these two instead determine the
// beginning and end of the VBlank period. If needed, split the interrupt firing into
// two different intervals.
Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PDC0);
Service::GSP::SignalInterrupt(Service::GSP::InterruptId::PDC1);
// Reschedule recurrent event
Core::System::GetInstance().CoreTiming().ScheduleEvent(frame_ticks - cycles_late, vblank_event);
}
/// Initialize hardware
void Init(Memory::MemorySystem& memory) {
g_memory = &memory;
std::memset(&g_regs, 0, sizeof(g_regs));
auto& framebuffer_top = g_regs.framebuffer_config[0];
auto& framebuffer_sub = g_regs.framebuffer_config[1];
// Setup default framebuffer addresses (located in VRAM)
// .. or at least these are the ones used by system applets.
// There's probably a smarter way to come up with addresses
// like this which does not require hardcoding.
framebuffer_top.address_left1 = 0x181E6000;
framebuffer_top.address_left2 = 0x1822C800;
framebuffer_top.address_right1 = 0x18273000;
framebuffer_top.address_right2 = 0x182B9800;
framebuffer_sub.address_left1 = 0x1848F000;
framebuffer_sub.address_left2 = 0x184C7800;
framebuffer_top.width.Assign(240);
framebuffer_top.height.Assign(400);
framebuffer_top.stride = 3 * 240;
framebuffer_top.color_format.Assign(Regs::PixelFormat::RGB8);
framebuffer_top.active_fb = 0;
framebuffer_sub.width.Assign(240);
framebuffer_sub.height.Assign(320);
framebuffer_sub.stride = 3 * 240;
framebuffer_sub.color_format.Assign(Regs::PixelFormat::RGB8);
framebuffer_sub.active_fb = 0;
Core::Timing& timing = Core::System::GetInstance().CoreTiming();
vblank_event = timing.RegisterEvent("GPU::VBlankCallback", VBlankCallback);
timing.ScheduleEvent(frame_ticks, vblank_event);
LOG_DEBUG(HW_GPU, "initialized OK");
}
/// Shutdown hardware
void Shutdown() {
LOG_DEBUG(HW_GPU, "shutdown OK");
}
} // namespace GPU

344
src/core/hw/gpu.h
View file

@ -1,344 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cstddef>
#include <type_traits>
#include <boost/serialization/access.hpp>
#include <boost/serialization/binary_object.hpp>
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "core/core_timing.h"
namespace Memory {
class MemorySystem;
}
namespace GPU {
// Measured on hardware to be 2240568 timer cycles or 4481136 ARM11 cycles
constexpr u64 frame_ticks = 4481136ull;
// Refresh rate defined by ratio of ARM11 frequency to ARM11 ticks per frame
// (268,111,856) / (4,481,136) = 59.83122493939037Hz
constexpr double SCREEN_REFRESH_RATE = BASE_CLOCK_RATE_ARM11 / static_cast<double>(frame_ticks);
// Returns index corresponding to the Regs member labeled by field_name
#define GPU_REG_INDEX(field_name) (offsetof(GPU::Regs, field_name) / sizeof(u32))
// Returns index corresponding to the Regs::FramebufferConfig labeled by field_name
// screen_id is a subscript for Regs::framebuffer_config
#define GPU_FRAMEBUFFER_REG_INDEX(screen_id, field_name) \
((offsetof(GPU::Regs, framebuffer_config) + \
sizeof(GPU::Regs::FramebufferConfig) * (screen_id) + \
offsetof(GPU::Regs::FramebufferConfig, field_name)) / \
sizeof(u32))
// MMIO region 0x1EFxxxxx
struct Regs {
// helper macro to make sure the defined structures are of the expected size.
#define ASSERT_MEMBER_SIZE(name, size_in_bytes) \
static_assert(sizeof(name) == size_in_bytes, \
"Structure size and register block length don't match")
// Components are laid out in reverse byte order, most significant bits first.
enum class PixelFormat : u32 {
RGBA8 = 0,
RGB8 = 1,
RGB565 = 2,
RGB5A1 = 3,
RGBA4 = 4,
};
/**
* Returns the number of bytes per pixel.
*/
static int BytesPerPixel(PixelFormat format) {
switch (format) {
case PixelFormat::RGBA8:
return 4;
case PixelFormat::RGB8:
return 3;
case PixelFormat::RGB565:
case PixelFormat::RGB5A1:
case PixelFormat::RGBA4:
return 2;
default:
UNREACHABLE();
}
return 0;
}
INSERT_PADDING_WORDS(0x4);
struct MemoryFillConfig {
u32 address_start;
u32 address_end;
union {
u32 value_32bit;
BitField<0, 16, u32> value_16bit;
// TODO: Verify component order
BitField<0, 8, u32> value_24bit_r;
BitField<8, 8, u32> value_24bit_g;
BitField<16, 8, u32> value_24bit_b;
};
union {
u32 control;
// Setting this field to 1 triggers the memory fill.
// This field also acts as a status flag, and gets reset to 0 upon completion.
BitField<0, 1, u32> trigger;
// Set to 1 upon completion.
BitField<1, 1, u32> finished;
// If both of these bits are unset, then it will fill the memory with a 16 bit value
// 1: fill with 24-bit wide values
BitField<8, 1, u32> fill_24bit;
// 1: fill with 32-bit wide values
BitField<9, 1, u32> fill_32bit;
};
inline u32 GetStartAddress() const {
return DecodeAddressRegister(address_start);
}
inline u32 GetEndAddress() const {
return DecodeAddressRegister(address_end);
}
inline std::string DebugName() const {
return fmt::format("from {:#X} to {:#X} with {}-bit value {:#X}", GetStartAddress(),
GetEndAddress(), fill_32bit ? "32" : (fill_24bit ? "24" : "16"),
value_32bit);
}
} memory_fill_config[2];
ASSERT_MEMBER_SIZE(memory_fill_config[0], 0x10);
INSERT_PADDING_WORDS(0x10b);
struct FramebufferConfig {
union {
u32 size;
BitField<0, 16, u32> width;
BitField<16, 16, u32> height;
};
INSERT_PADDING_WORDS(0x2);
u32 address_left1;
u32 address_left2;
union {
u32 format;
BitField<0, 3, PixelFormat> color_format;
};
INSERT_PADDING_WORDS(0x1);
union {
u32 active_fb;
// 0: Use parameters ending with "1"
// 1: Use parameters ending with "2"
BitField<0, 1, u32> second_fb_active;
};
INSERT_PADDING_WORDS(0x5);
// Distance between two pixel rows, in bytes
u32 stride;
u32 address_right1;
u32 address_right2;
INSERT_PADDING_WORDS(0x30);
} framebuffer_config[2];
ASSERT_MEMBER_SIZE(framebuffer_config[0], 0x100);
INSERT_PADDING_WORDS(0x169);
struct DisplayTransferConfig {
u32 input_address;
u32 output_address;
inline u32 GetPhysicalInputAddress() const {
return DecodeAddressRegister(input_address);
}
inline u32 GetPhysicalOutputAddress() const {
return DecodeAddressRegister(output_address);
}
inline std::string DebugName() const noexcept {
return fmt::format("from {:#x} to {:#x} with {} scaling and stride {}, width {}",
GetPhysicalInputAddress(), GetPhysicalOutputAddress(),
scaling == NoScale ? "no" : (scaling == ScaleX ? "X" : "XY"),
input_width.Value(), output_width.Value());
}
union {
u32 output_size;
BitField<0, 16, u32> output_width;
BitField<16, 16, u32> output_height;
};
union {
u32 input_size;
BitField<0, 16, u32> input_width;
BitField<16, 16, u32> input_height;
};
enum ScalingMode : u32 {
NoScale = 0, // Doesn't scale the image
ScaleX = 1, // Downscales the image in half in the X axis and applies a box filter
ScaleXY =
2, // Downscales the image in half in both the X and Y axes and applies a box filter
};
union {
u32 flags;
BitField<0, 1, u32> flip_vertically; // flips input data vertically
BitField<1, 1, u32> input_linear; // Converts from linear to tiled format
BitField<2, 1, u32> crop_input_lines;
BitField<3, 1, u32> is_texture_copy; // Copies the data without performing any
// processing and respecting texture copy fields
BitField<5, 1, u32> dont_swizzle;
BitField<8, 3, PixelFormat> input_format;
BitField<12, 3, PixelFormat> output_format;
/// Uses some kind of 32x32 block swizzling mode, instead of the usual 8x8 one.
BitField<16, 1, u32> block_32; // TODO(yuriks): unimplemented
BitField<24, 2, ScalingMode> scaling; // Determines the scaling mode of the transfer
};
INSERT_PADDING_WORDS(0x1);
// it seems that writing to this field triggers the display transfer
u32 trigger;
INSERT_PADDING_WORDS(0x1);
struct {
u32 size; // The lower 4 bits are ignored
union {
u32 input_size;
BitField<0, 16, u32> input_width;
BitField<16, 16, u32> input_gap;
};
union {
u32 output_size;
BitField<0, 16, u32> output_width;
BitField<16, 16, u32> output_gap;
};
} texture_copy;
} display_transfer_config;
ASSERT_MEMBER_SIZE(display_transfer_config, 0x2c);
INSERT_PADDING_WORDS(0x32D);
struct {
// command list size (in bytes)
u32 size;
INSERT_PADDING_WORDS(0x1);
// command list address
u32 address;
INSERT_PADDING_WORDS(0x1);
// it seems that writing to this field triggers command list processing
u32 trigger;
inline u32 GetPhysicalAddress() const {
return DecodeAddressRegister(address);
}
} command_processor_config;
ASSERT_MEMBER_SIZE(command_processor_config, 0x14);
INSERT_PADDING_WORDS(0x9c3);
static constexpr std::size_t NumIds() {
return sizeof(Regs) / sizeof(u32);
}
const u32& operator[](int index) const {
const u32* content = reinterpret_cast<const u32*>(this);
return content[index];
}
u32& operator[](int index) {
u32* content = reinterpret_cast<u32*>(this);
return content[index];
}
#undef ASSERT_MEMBER_SIZE
private:
/*
* Most physical addresses which GPU registers refer to are 8-byte aligned.
* This function should be used to get the address from a raw register value.
*/
static inline u32 DecodeAddressRegister(u32 register_value) {
return register_value * 8;
}
template <class Archive>
void serialize(Archive& ar, const unsigned int) {
ar& boost::serialization::make_binary_object(this, sizeof(Regs));
}
friend class boost::serialization::access;
};
static_assert(std::is_standard_layout<Regs>::value, "Structure does not use standard layout");
#define ASSERT_REG_POSITION(field_name, position) \
static_assert(offsetof(Regs, field_name) == position * 4, \
"Field " #field_name " has invalid position")
ASSERT_REG_POSITION(memory_fill_config[0], 0x00004);
ASSERT_REG_POSITION(memory_fill_config[1], 0x00008);
ASSERT_REG_POSITION(framebuffer_config[0], 0x00117);
ASSERT_REG_POSITION(framebuffer_config[1], 0x00157);
ASSERT_REG_POSITION(display_transfer_config, 0x00300);
ASSERT_REG_POSITION(command_processor_config, 0x00638);
#undef ASSERT_REG_POSITION
// The total number of registers is chosen arbitrarily, but let's make sure it's not some odd value
// anyway.
static_assert(sizeof(Regs) == 0x1000 * sizeof(u32), "Invalid total size of register set");
extern Regs g_regs;
template <typename T>
void Read(T& var, const u32 addr);
template <typename T>
void Write(u32 addr, const T data);
/// Initialize hardware
void Init(Memory::MemorySystem& memory);
/// Shutdown hardware
void Shutdown();
} // namespace GPU

102
src/core/hw/hw.cpp
View file

@ -1,102 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/common_types.h"
#include "common/logging/log.h"
#include "core/hw/aes/key.h"
#include "core/hw/gpu.h"
#include "core/hw/hw.h"
#include "core/hw/lcd.h"
namespace HW {
template <typename T>
inline void Read(T& var, const u32 addr) {
switch (addr & 0xFFFFF000) {
case VADDR_GPU:
case VADDR_GPU + 0x1000:
case VADDR_GPU + 0x2000:
case VADDR_GPU + 0x3000:
case VADDR_GPU + 0x4000:
case VADDR_GPU + 0x5000:
case VADDR_GPU + 0x6000:
case VADDR_GPU + 0x7000:
case VADDR_GPU + 0x8000:
case VADDR_GPU + 0x9000:
case VADDR_GPU + 0xA000:
case VADDR_GPU + 0xB000:
case VADDR_GPU + 0xC000:
case VADDR_GPU + 0xD000:
case VADDR_GPU + 0xE000:
case VADDR_GPU + 0xF000:
GPU::Read(var, addr);
break;
case VADDR_LCD:
LCD::Read(var, addr);
break;
default:
LOG_ERROR(HW_Memory, "unknown Read{} @ {:#010X}", sizeof(var) * 8, addr);
}
}
template <typename T>
inline void Write(u32 addr, const T data) {
switch (addr & 0xFFFFF000) {
case VADDR_GPU:
case VADDR_GPU + 0x1000:
case VADDR_GPU + 0x2000:
case VADDR_GPU + 0x3000:
case VADDR_GPU + 0x4000:
case VADDR_GPU + 0x5000:
case VADDR_GPU + 0x6000:
case VADDR_GPU + 0x7000:
case VADDR_GPU + 0x8000:
case VADDR_GPU + 0x9000:
case VADDR_GPU + 0xA000:
case VADDR_GPU + 0xB000:
case VADDR_GPU + 0xC000:
case VADDR_GPU + 0xD000:
case VADDR_GPU + 0xE000:
case VADDR_GPU + 0xF000:
GPU::Write(addr, data);
break;
case VADDR_LCD:
LCD::Write(addr, data);
break;
default:
LOG_ERROR(HW_Memory, "unknown Write{} {:#010X} @ {:#010X}", sizeof(data) * 8, (u32)data,
addr);
}
}
// Explicitly instantiate template functions because we aren't defining this in the header:
template void Read<u64>(u64& var, const u32 addr);
template void Read<u32>(u32& var, const u32 addr);
template void Read<u16>(u16& var, const u32 addr);
template void Read<u8>(u8& var, const u32 addr);
template void Write<u64>(u32 addr, const u64 data);
template void Write<u32>(u32 addr, const u32 data);
template void Write<u16>(u32 addr, const u16 data);
template void Write<u8>(u32 addr, const u8 data);
/// Update hardware
void Update() {}
/// Initialize hardware
void Init(Memory::MemorySystem& memory) {
AES::InitKeys();
GPU::Init(memory);
LCD::Init();
LOG_DEBUG(HW, "initialized OK");
}
/// Shutdown hardware
void Shutdown() {
GPU::Shutdown();
LCD::Shutdown();
LOG_DEBUG(HW, "shutdown OK");
}
} // namespace HW

54
src/core/hw/hw.h
View file

@ -1,54 +0,0 @@
// Copyright 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/common_types.h"
namespace Memory {
class MemorySystem;
}
namespace HW {
/// Beginnings of IO register regions, in the user VA space.
enum : u32 {
VADDR_HASH = 0x1EC01000,
VADDR_CSND = 0x1EC03000,
VADDR_DSP = 0x1EC40000,
VADDR_PDN = 0x1EC41000,
VADDR_CODEC = 0x1EC41000,
VADDR_SPI = 0x1EC42000,
VADDR_SPI_2 = 0x1EC43000, // Only used under TWL_FIRM?
VADDR_I2C = 0x1EC44000,
VADDR_CODEC_2 = 0x1EC45000,
VADDR_HID = 0x1EC46000,
VADDR_GPIO = 0x1EC47000,
VADDR_I2C_2 = 0x1EC48000,
VADDR_SPI_3 = 0x1EC60000,
VADDR_I2C_3 = 0x1EC61000,
VADDR_MIC = 0x1EC62000,
VADDR_PXI = 0x1EC63000,
VADDR_LCD = 0x1ED02000,
VADDR_DSP_2 = 0x1ED03000,
VADDR_HASH_2 = 0x1EE01000,
VADDR_GPU = 0x1EF00000,
};
template <typename T>
void Read(T& var, const u32 addr);
template <typename T>
void Write(u32 addr, const T data);
/// Update hardware
void Update();
/// Initialize hardware
void Init(Memory::MemorySystem& memory);
/// Shutdown hardware
void Shutdown();
} // namespace HW

76
src/core/hw/lcd.cpp
View file

@ -1,76 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <cstring>
#include "common/common_types.h"
#include "common/logging/log.h"
#include "core/hw/hw.h"
#include "core/hw/lcd.h"
#include "core/tracer/recorder.h"
#include "video_core/debug_utils/debug_utils.h"
namespace LCD {
Regs g_regs;
template <typename T>
inline void Read(T& var, const u32 raw_addr) {
u32 addr = raw_addr - HW::VADDR_LCD;
u32 index = addr / 4;
// Reads other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= 0x400 || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_LCD, "unknown Read{} @ {:#010X}", sizeof(var) * 8, addr);
return;
}
var = g_regs[index];
}
template <typename T>
inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_LCD;
u32 index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= 0x400 || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_LCD, "unknown Write{} {:#010X} @ {:#010X}", sizeof(data) * 8, (u32)data, addr);
return;
}
g_regs[index] = static_cast<u32>(data);
// Notify tracer about the register write
// This is happening *after* handling the write to make sure we properly catch all memory reads.
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
// addr + GPU VBase - IO VBase + IO PBase
Pica::g_debug_context->recorder->RegisterWritten<T>(
addr + HW::VADDR_LCD - 0x1EC00000 + 0x10100000, data);
}
}
// Explicitly instantiate template functions because we aren't defining this in the header:
template void Read<u64>(u64& var, const u32 addr);
template void Read<u32>(u32& var, const u32 addr);
template void Read<u16>(u16& var, const u32 addr);
template void Read<u8>(u8& var, const u32 addr);
template void Write<u64>(u32 addr, const u64 data);
template void Write<u32>(u32 addr, const u32 data);
template void Write<u16>(u32 addr, const u16 data);
template void Write<u8>(u32 addr, const u8 data);
/// Initialize hardware
void Init() {
std::memset(&g_regs, 0, sizeof(g_regs));
LOG_DEBUG(HW_LCD, "initialized OK");
}
/// Shutdown hardware
void Shutdown() {
LOG_DEBUG(HW_LCD, "shutdown OK");
}
} // namespace LCD

89
src/core/hw/lcd.h
View file

@ -1,89 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <cstddef>
#include <type_traits>
#include <boost/serialization/access.hpp>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#define LCD_REG_INDEX(field_name) (offsetof(LCD::Regs, field_name) / sizeof(u32))
namespace LCD {
struct Regs {
union ColorFill {
u32 raw;
BitField<0, 8, u32> color_r;
BitField<8, 8, u32> color_g;
BitField<16, 8, u32> color_b;
BitField<24, 1, u32> is_enabled;
};
INSERT_PADDING_WORDS(0x81);
ColorFill color_fill_top;
INSERT_PADDING_WORDS(0xE);
u32 backlight_top;
INSERT_PADDING_WORDS(0x1F0);
ColorFill color_fill_bottom;
INSERT_PADDING_WORDS(0xE);
u32 backlight_bottom;
INSERT_PADDING_WORDS(0x16F);
static constexpr std::size_t NumIds() {
return sizeof(Regs) / sizeof(u32);
}
const u32& operator[](int index) const {
const u32* content = reinterpret_cast<const u32*>(this);
return content[index];
}
u32& operator[](int index) {
u32* content = reinterpret_cast<u32*>(this);
return content[index];
}
private:
template <class Archive>
void serialize(Archive& ar, const unsigned int) {
ar& color_fill_top.raw;
ar& backlight_top;
ar& color_fill_bottom.raw;
ar& backlight_bottom;
}
friend class boost::serialization::access;
};
static_assert(std::is_standard_layout<Regs>::value, "Structure does not use standard layout");
#define ASSERT_REG_POSITION(field_name, position) \
static_assert(offsetof(Regs, field_name) == position * 4, \
"Field " #field_name " has invalid position")
ASSERT_REG_POSITION(color_fill_top, 0x81);
ASSERT_REG_POSITION(backlight_top, 0x90);
ASSERT_REG_POSITION(color_fill_bottom, 0x281);
ASSERT_REG_POSITION(backlight_bottom, 0x290);
extern Regs g_regs;
template <typename T>
void Read(T& var, const u32 addr);
template <typename T>
void Write(u32 addr, const T data);
/// Initialize hardware
void Init();
/// Shutdown hardware
void Shutdown();
} // namespace LCD

2
src/core/hw/y2r.cpp
View file

@ -9,7 +9,7 @@
#include "common/assert.h"
#include "common/color.h"
#include "common/common_types.h"
#include "common/microprofileui.h"
#include "common/microprofile.h"
#include "common/vector_math.h"
#include "core/core.h"
#include "core/hle/service/cam/y2r_u.h"

134
src/core/memory.cpp
View file

@ -19,10 +19,9 @@
#include "core/global.h"
#include "core/hle/kernel/process.h"
#include "core/hle/service/plgldr/plgldr.h"
#include "core/hw/hw.h"
#include "core/memory.h"
#include "video_core/gpu.h"
#include "video_core/renderer_base.h"
#include "video_core/video_core.h"
SERIALIZE_EXPORT_IMPL(Memory::MemorySystem::BackingMemImpl<Memory::Region::FCRAM>)
SERIALIZE_EXPORT_IMPL(Memory::MemorySystem::BackingMemImpl<Memory::Region::VRAM>)
@ -346,13 +345,52 @@ std::shared_ptr<PageTable> MemorySystem::GetCurrentPageTable() const {
return impl->current_page_table;
}
void RasterizerFlushVirtualRegion(VAddr start, u32 size, FlushMode mode) {
const VAddr end = start + size;
auto CheckRegion = [&](VAddr region_start, VAddr region_end, PAddr paddr_region_start) {
if (start >= region_end || end <= region_start) {
// No overlap with region
return;
}
auto& renderer = Core::System::GetInstance().GPU().Renderer();
VAddr overlap_start = std::max(start, region_start);
VAddr overlap_end = std::min(end, region_end);
PAddr physical_start = paddr_region_start + (overlap_start - region_start);
u32 overlap_size = overlap_end - overlap_start;
auto* rasterizer = renderer.Rasterizer();
switch (mode) {
case FlushMode::Flush:
rasterizer->FlushRegion(physical_start, overlap_size);
break;
case FlushMode::Invalidate:
rasterizer->InvalidateRegion(physical_start, overlap_size);
break;
case FlushMode::FlushAndInvalidate:
rasterizer->FlushAndInvalidateRegion(physical_start, overlap_size);
break;
}
};
CheckRegion(LINEAR_HEAP_VADDR, LINEAR_HEAP_VADDR_END, FCRAM_PADDR);
CheckRegion(NEW_LINEAR_HEAP_VADDR, NEW_LINEAR_HEAP_VADDR_END, FCRAM_PADDR);
CheckRegion(VRAM_VADDR, VRAM_VADDR_END, VRAM_PADDR);
if (Service::PLGLDR::PLG_LDR::GetPluginFBAddr())
CheckRegion(PLUGIN_3GX_FB_VADDR, PLUGIN_3GX_FB_VADDR_END,
Service::PLGLDR::PLG_LDR::GetPluginFBAddr());
}
void MemorySystem::MapPages(PageTable& page_table, u32 base, u32 size, MemoryRef memory,
PageType type) {
LOG_DEBUG(HW_Memory, "Mapping {} onto {:08X}-{:08X}", (void*)memory.GetPtr(),
base * CITRA_PAGE_SIZE, (base + size) * CITRA_PAGE_SIZE);
RasterizerFlushVirtualRegion(base << CITRA_PAGE_BITS, size * CITRA_PAGE_SIZE,
FlushMode::FlushAndInvalidate);
if (impl->system.IsPoweredOn()) {
RasterizerFlushVirtualRegion(base << CITRA_PAGE_BITS, size * CITRA_PAGE_SIZE,
FlushMode::FlushAndInvalidate);
}
u32 end = base + size;
while (base != end) {
@ -421,9 +459,8 @@ T MemorySystem::Read(const VAddr vaddr) {
return value;
} else if ((paddr & 0xF0000000) == 0x10000000 &&
paddr >= Memory::IO_AREA_PADDR) { // Check MMIO region
T ret;
HW::Read<T>(ret, static_cast<VAddr>(paddr) - Memory::IO_AREA_PADDR + 0x1EC00000);
return ret;
return impl->system.GPU().ReadReg(static_cast<VAddr>(paddr) - Memory::IO_AREA_PADDR +
0x1EC00000);
}
}
@ -468,7 +505,10 @@ void MemorySystem::Write(const VAddr vaddr, const T data) {
return;
} else if ((paddr & 0xF0000000) == 0x10000000 &&
paddr >= Memory::IO_AREA_PADDR) { // Check MMIO region
HW::Write<T>(static_cast<VAddr>(paddr) - Memory::IO_AREA_PADDR + 0x1EC00000, data);
ASSERT(sizeof(data) == sizeof(u32));
impl->system.GPU().WriteReg(static_cast<VAddr>(paddr) - Memory::IO_AREA_PADDR +
0x1EC00000,
static_cast<u32>(data));
return;
}
}
@ -713,84 +753,6 @@ void MemorySystem::RasterizerMarkRegionCached(PAddr start, u32 size, bool cached
}
}
void RasterizerFlushRegion(PAddr start, u32 size) {
if (VideoCore::g_renderer == nullptr) {
return;
}
VideoCore::g_renderer->Rasterizer()->FlushRegion(start, size);
}
void RasterizerInvalidateRegion(PAddr start, u32 size) {
if (VideoCore::g_renderer == nullptr) {
return;
}
VideoCore::g_renderer->Rasterizer()->InvalidateRegion(start, size);
}
void RasterizerFlushAndInvalidateRegion(PAddr start, u32 size) {
// Since pages are unmapped on shutdown after video core is shutdown, the renderer may be
// null here
if (VideoCore::g_renderer == nullptr) {
return;
}
VideoCore::g_renderer->Rasterizer()->FlushAndInvalidateRegion(start, size);
}
void RasterizerClearAll(bool flush) {
// Since pages are unmapped on shutdown after video core is shutdown, the renderer may be
// null here
if (VideoCore::g_renderer == nullptr) {
return;
}
VideoCore::g_renderer->Rasterizer()->ClearAll(flush);
}
void RasterizerFlushVirtualRegion(VAddr start, u32 size, FlushMode mode) {
// Since pages are unmapped on shutdown after video core is shutdown, the renderer may be
// null here
if (VideoCore::g_renderer == nullptr) {
return;
}
VAddr end = start + size;
auto CheckRegion = [&](VAddr region_start, VAddr region_end, PAddr paddr_region_start) {
if (start >= region_end || end <= region_start) {
// No overlap with region
return;
}
VAddr overlap_start = std::max(start, region_start);
VAddr overlap_end = std::min(end, region_end);
PAddr physical_start = paddr_region_start + (overlap_start - region_start);
u32 overlap_size = overlap_end - overlap_start;
auto* rasterizer = VideoCore::g_renderer->Rasterizer();
switch (mode) {
case FlushMode::Flush:
rasterizer->FlushRegion(physical_start, overlap_size);
break;
case FlushMode::Invalidate:
rasterizer->InvalidateRegion(physical_start, overlap_size);
break;
case FlushMode::FlushAndInvalidate:
rasterizer->FlushAndInvalidateRegion(physical_start, overlap_size);
break;
}
};
CheckRegion(LINEAR_HEAP_VADDR, LINEAR_HEAP_VADDR_END, FCRAM_PADDR);
CheckRegion(NEW_LINEAR_HEAP_VADDR, NEW_LINEAR_HEAP_VADDR_END, FCRAM_PADDR);
CheckRegion(VRAM_VADDR, VRAM_VADDR_END, VRAM_PADDR);
if (Service::PLGLDR::PLG_LDR::GetPluginFBAddr())
CheckRegion(PLUGIN_3GX_FB_VADDR, PLUGIN_3GX_FB_VADDR_END,
Service::PLGLDR::PLG_LDR::GetPluginFBAddr());
}
u8 MemorySystem::Read8(const VAddr addr) {
return Read<u8>(addr);
}

25
src/core/memory.h
View file

@ -226,21 +226,6 @@ enum : VAddr {
PLUGIN_3GX_FB_VADDR_END = PLUGIN_3GX_FB_VADDR + PLUGIN_3GX_FB_SIZE
};
/**
* Flushes any externally cached rasterizer resources touching the given region.
*/
void RasterizerFlushRegion(PAddr start, u32 size);
/**
* Invalidates any externally cached rasterizer resources touching the given region.
*/
void RasterizerInvalidateRegion(PAddr start, u32 size);
/**
* Flushes and invalidates any externally cached rasterizer resources touching the given region.
*/
void RasterizerFlushAndInvalidateRegion(PAddr start, u32 size);
enum class FlushMode {
/// Write back modified surfaces to RAM
Flush,
@ -250,16 +235,6 @@ enum class FlushMode {
FlushAndInvalidate,
};
/**
* Flushes and invalidates all memory in the rasterizer cache and removes any leftover state
* If flush is true, the rasterizer should flush any cached resources to RAM before clearing
*/
void RasterizerClearAll(bool flush);
/**
* Flushes and invalidates any externally cached rasterizer resources touching the given virtual
* address region.
*/
void RasterizerFlushVirtualRegion(VAddr start, u32 size, FlushMode mode);
class MemorySystem {

5
src/core/movie.cpp
View file

@ -21,7 +21,6 @@
#include "core/hle/service/hid/hid.h"
#include "core/hle/service/ir/extra_hid.h"
#include "core/hle/service/ir/ir_rst.h"
#include "core/hw/gpu.h"
#include "core/loader/loader.h"
#include "core/movie.h"
@ -218,10 +217,10 @@ Movie::PlayMode Movie::GetPlayMode() const {
}
u64 Movie::GetCurrentInputIndex() const {
return static_cast<u64>(std::nearbyint(current_input / 234.0 * GPU::SCREEN_REFRESH_RATE));
return static_cast<u64>(std::nearbyint(current_input / 234.0 * SCREEN_REFRESH_RATE));
}
u64 Movie::GetTotalInputCount() const {
return static_cast<u64>(std::nearbyint(total_input / 234.0 * GPU::SCREEN_REFRESH_RATE));
return static_cast<u64>(std::nearbyint(total_input / 234.0 * SCREEN_REFRESH_RATE));
}
void Movie::CheckInputEnd() {

5
src/core/perf_stats.cpp
View file

@ -13,8 +13,9 @@
#include <fmt/format.h>
#include "common/file_util.h"
#include "common/settings.h"
#include "core/hw/gpu.h"
#include "core/core_timing.h"
#include "core/perf_stats.h"
#include "video_core/gpu.h"
using namespace std::chrono_literals;
using DoubleSecs = std::chrono::duration<double, std::chrono::seconds::period>;
@ -120,7 +121,7 @@ PerfStats::Results PerfStats::GetLastStats() {
double PerfStats::GetLastFrameTimeScale() const {
std::scoped_lock lock{object_mutex};
constexpr double FRAME_LENGTH = 1.0 / GPU::SCREEN_REFRESH_RATE;
constexpr double FRAME_LENGTH = 1.0 / SCREEN_REFRESH_RATE;
return duration_cast<DoubleSecs>(previous_frame_length).count() / FRAME_LENGTH;
}

11
src/core/tracer/citrace.h
View file

@ -75,16 +75,7 @@ struct CTMemoryLoad {
struct CTRegisterWrite {
u32 physical_address;
enum : u32 {
SIZE_8 = 0xD1,
SIZE_16 = 0xD2,
SIZE_32 = 0xD3,
SIZE_64 = 0xD4,
} size;
// TODO: Make it clearer which bits of this member are used for sizes other than 32 bits
u64 value;
u32 value;
};
struct CTStreamElement {

29
src/core/tracer/recorder.cpp
View file

@ -22,9 +22,8 @@ void Recorder::Finish(const std::string& filename) {
// Calculate file offsets
auto& initial = header.initial_state_offsets;
initial.gpu_registers_size = static_cast<u32>(initial_state.gpu_registers.size());
initial.lcd_registers_size = static_cast<u32>(initial_state.lcd_registers.size());
initial.pica_registers_size = static_cast<u32>(initial_state.pica_registers.size());
initial.lcd_registers_size = static_cast<u32>(initial_state.lcd_registers.size());
initial.default_attributes_size = static_cast<u32>(initial_state.default_attributes.size());
initial.vs_program_binary_size = static_cast<u32>(initial_state.vs_program_binary.size());
initial.vs_swizzle_data_size = static_cast<u32>(initial_state.vs_swizzle_data.size());
@ -81,22 +80,17 @@ void Recorder::Finish(const std::string& filename) {
throw "Failed to write header";
// Write initial state
written =
file.WriteArray(initial_state.gpu_registers.data(), initial_state.gpu_registers.size());
if (written != initial_state.gpu_registers.size() || file.Tell() != initial.lcd_registers)
throw "Failed to write GPU registers";
written =
file.WriteArray(initial_state.lcd_registers.data(), initial_state.lcd_registers.size());
if (written != initial_state.lcd_registers.size() || file.Tell() != initial.pica_registers)
throw "Failed to write LCD registers";
written = file.WriteArray(initial_state.pica_registers.data(),
initial_state.pica_registers.size());
if (written != initial_state.pica_registers.size() ||
file.Tell() != initial.default_attributes)
throw "Failed to write Pica registers";
written =
file.WriteArray(initial_state.lcd_registers.data(), initial_state.lcd_registers.size());
if (written != initial_state.lcd_registers.size() || file.Tell() != initial.pica_registers)
throw "Failed to write LCD registers";
written = file.WriteArray(initial_state.default_attributes.data(),
initial_state.default_attributes.size());
if (written != initial_state.default_attributes.size() ||
@ -187,21 +181,12 @@ void Recorder::MemoryAccessed(const u8* data, u32 size, u32 physical_address) {
stream.push_back(element);
}
template <typename T>
void Recorder::RegisterWritten(u32 physical_address, T value) {
void Recorder::RegisterWritten(u32 physical_address, u32 value) {
StreamElement element = {{RegisterWrite}};
element.data.register_write.size = (sizeof(T) == 1) ? CTRegisterWrite::SIZE_8
: (sizeof(T) == 2) ? CTRegisterWrite::SIZE_16
: (sizeof(T) == 4) ? CTRegisterWrite::SIZE_32
: CTRegisterWrite::SIZE_64;
element.data.register_write.physical_address = physical_address;
element.data.register_write.value = value;
stream.push_back(element);
}
template void Recorder::RegisterWritten(u32, u8);
template void Recorder::RegisterWritten(u32, u16);
template void Recorder::RegisterWritten(u32, u32);
template void Recorder::RegisterWritten(u32, u64);
} // namespace CiTrace

5
src/core/tracer/recorder.h
View file

@ -4,7 +4,6 @@
#pragma once
#include <span>
#include <string>
#include <unordered_map>
#include <vector>
@ -17,7 +16,6 @@ namespace CiTrace {
class Recorder {
public:
struct InitialState {
std::vector<u32> gpu_registers;
std::vector<u32> lcd_registers;
std::vector<u32> pica_registers;
std::vector<u32> default_attributes;
@ -52,8 +50,7 @@ public:
* Record a register write.
* @note Use this whenever a GPU-related MMIO register has been written to.
*/
template <typename T>
void RegisterWritten(u32 physical_address, T value);
void RegisterWritten(u32 physical_address, u32 value);
private:
// Initial state of recording start