u32 VideoBackendHardware::Video_AccessEFB(EFBAccessType type, u32 x, u32 y, u32 InputData)
{
if (s_BackendInitialized && g_ActiveConfig.bEFBAccessEnable)
{
s_accessEFBArgs.type = type;
s_accessEFBArgs.x = x;
s_accessEFBArgs.y = y;
s_accessEFBArgs.Data = InputData;
s_efbAccessRequested.Set();
if (SConfig::GetInstance().m_LocalCoreStartupParameter.bCPUThread)
{
s_efbAccessReadyEvent.Reset();
if (s_FifoShuttingDown.IsSet())
return 0;
s_efbAccessRequested.Set();
s_efbAccessReadyEvent.Wait();
}
else
VideoFifo_CheckEFBAccess();
return s_AccessEFBResult;
}
return 0;
}
static void DVDThread()
{
Common::SetCurrentThreadName("DVD thread");
while (true)
{
s_request_queue_expanded.Wait();
if (s_dvd_thread_exiting.IsSet())
return;
ReadRequest request;
while (s_request_queue.Pop(request))
{
FileMonitor::Log(*s_disc, request.partition, request.dvd_offset);
std::vector<u8> buffer(request.length);
if (!s_disc->Read(request.dvd_offset, request.length, buffer.data(), request.partition))
buffer.resize(0);
request.realtime_done_us = Common::Timer::GetTimeUs();
s_result_queue.Push(ReadResult(std::move(request), std::move(buffer)));
s_result_queue_expanded.Set();
if (s_dvd_thread_exiting.IsSet())
return;
}
}
}
static void VideoFifo_CheckPerfQueryRequest()
{
if (s_perfQueryRequested.IsSet())
{
g_perf_query->FlushResults();
s_perfQueryRequested.Clear();
s_perfQueryReadyEvent.Set();
}
}
void VideoFifo_CheckEFBAccess()
{
if (s_efbAccessRequested.IsSet())
{
s_AccessEFBResult = g_renderer->AccessEFB(s_accessEFBArgs.type, s_accessEFBArgs.x, s_accessEFBArgs.y, s_accessEFBArgs.Data);
s_efbAccessRequested.Clear();
s_efbAccessReadyEvent.Set();
}
}
// Run from the graphics thread (from Fifo.cpp)
static void VideoFifo_CheckSwapRequest()
{
if (g_ActiveConfig.bUseXFB)
{
if (s_swapRequested.IsSet())
{
EFBRectangle rc;
Renderer::Swap(s_beginFieldArgs.xfbAddr, s_beginFieldArgs.fbWidth, s_beginFieldArgs.fbHeight,rc);
s_swapRequested.Clear();
}
}
}
void VideoBackendHardware::InitializeShared()
{
VideoCommon_Init();
s_swapRequested.Clear();
s_efbAccessRequested.Clear();
s_perfQueryRequested.Clear();
s_FifoShuttingDown.Clear();
memset((void*)&s_beginFieldArgs, 0, sizeof(s_beginFieldArgs));
memset(&s_accessEFBArgs, 0, sizeof(s_accessEFBArgs));
s_AccessEFBResult = 0;
m_invalid = false;
}
void VideoBackendHardware::Video_ExitLoop()
{
ExitGpuLoop();
s_FifoShuttingDown.Set();
s_efbAccessReadyEvent.Set();
s_perfQueryReadyEvent.Set();
}
// Run from the CPU thread (from VideoInterface.cpp)
void VideoBackendHardware::Video_EndField()
{
if (s_BackendInitialized)
{
s_swapRequested.Set();
}
}
void Host_Message(int Id)
{
if (Id == WM_USER_STOP)
{
s_running.Clear();
updateMainFrameEvent.Set();
}
}
virtual void MainLoop()
{
while (s_running.IsSet())
{
Core::HostDispatchJobs();
std::this_thread::sleep_for(std::chrono::milliseconds(100));
}
}
void VideoBackendHardware::InitializeShared()
{
VideoCommon_Init();
s_FifoShuttingDown.Clear();
memset((void*)&s_beginFieldArgs, 0, sizeof(s_beginFieldArgs));
m_invalid = false;
}
static void signal_handler(int)
{
const char message[] = "A signal was received. A second signal will force Dolphin to stop.\n";
if (write(STDERR_FILENO, message, sizeof(message)) < 0)
{
}
s_shutdown_requested.Set();
}
void EmulatorState(bool running)
{
s_emu_running_state.Set(running);
if (running)
s_gpu_mainloop.Wakeup();
else
s_gpu_mainloop.AllowSleep();
}
// May be executed from any thread, even the graphics thread.
// Created to allow for self shutdown.
void ExitGpuLoop()
{
// This should break the wait loop in CPU thread
CommandProcessor::fifo.bFF_GPReadEnable = false;
FlushGpu();
// Terminate GPU thread loop
s_emu_running_state.Set();
s_gpu_mainloop.Stop(false);
}
void HiresTexture::Shutdown()
{
if (s_prefetcher.joinable())
{
s_textureCacheAbortLoading.Set();
s_prefetcher.join();
}
s_textureMap.clear();
s_textureCache.clear();
}
static void StopDVDThread()
{
ASSERT(s_dvd_thread.joinable());
// By setting s_DVD_thread_exiting, we ask the DVD thread to cleanly exit.
// In case the request queue is empty, we need to set s_request_queue_expanded
// so that the DVD thread will wake up and check s_DVD_thread_exiting.
s_dvd_thread_exiting.Set();
s_request_queue_expanded.Set();
s_dvd_thread.join();
}
u32 VideoBackendHardware::Video_GetQueryResult(PerfQueryType type)
{
if (!g_perf_query->ShouldEmulate())
{
return 0;
}
// TODO: Is this check sane?
if (!g_perf_query->IsFlushed())
{
if (SConfig::GetInstance().m_LocalCoreStartupParameter.bCPUThread)
{
s_perfQueryReadyEvent.Reset();
if (s_FifoShuttingDown.IsSet())
return 0;
s_perfQueryRequested.Set();
s_perfQueryReadyEvent.Wait();
}
else
g_perf_query->FlushResults();
}
return g_perf_query->GetQueryResult(type);
}
// Run from the graphics thread (from Fifo.cpp)
void VideoFifo_CheckSwapRequestAt(u32 xfbAddr, u32 fbWidth, u32 fbHeight)
{
if (g_ActiveConfig.bUseXFB)
{
if (s_swapRequested.IsSet())
{
u32 aLower = xfbAddr;
u32 aUpper = xfbAddr + 2 * fbWidth * fbHeight;
u32 bLower = s_beginFieldArgs.xfbAddr;
u32 bUpper = s_beginFieldArgs.xfbAddr + 2 * s_beginFieldArgs.fbWidth * s_beginFieldArgs.fbHeight;
if (addrRangesOverlap(aLower, aUpper, bLower, bUpper))
VideoFifo_CheckSwapRequest();
}
}
}
void HiresTexture::Prefetch()
{
Common::SetCurrentThreadName("Prefetcher");
u32 starttime = Common::Timer::GetTimeMs();
for (const auto& entry : s_textureMap)
{
const std::string& base_filename = entry.first;
std::unique_lock<std::mutex> lk(s_textureCacheMutex);
auto iter = s_textureCache.find(base_filename);
if (iter == s_textureCache.end())
{
lk.unlock();
HiresTexture* ptr = Load(base_filename, [](size_t requested_size)
{
return new u8[requested_size];
}, true);
lk.lock();
if (ptr != nullptr)
{
size_sum.fetch_add(ptr->m_cached_data_size);
iter = s_textureCache.insert(iter, std::make_pair(base_filename, std::shared_ptr<HiresTexture>(ptr)));
}
}
if (s_textureCacheAbortLoading.IsSet())
{
return;
}
if (size_sum.load() > max_mem)
{
g_Config.bCacheHiresTextures = false;
OSD::AddMessage(StringFromFormat("Custom Textures prefetching after %.1f MB aborted, not enough RAM available", size_sum / (1024.0 * 1024.0)), 10000);
return;
}
}
u32 stoptime = Common::Timer::GetTimeMs();
OSD::AddMessage(StringFromFormat("Custom Textures loaded, %.1f MB in %.1f s", size_sum / (1024.0 * 1024.0), (stoptime - starttime) / 1000.0), 10000);
}
void VideoBackendBase::InitializeShared()
{
memset(&g_main_cp_state, 0, sizeof(g_main_cp_state));
memset(&g_preprocess_cp_state, 0, sizeof(g_preprocess_cp_state));
memset(texMem, 0, TMEM_SIZE);
// Do our OSD callbacks
OSD::DoCallbacks(OSD::CallbackType::Initialization);
// do not initialize again for the config window
m_initialized = true;
s_FifoShuttingDown.Clear();
memset((void*)&s_beginFieldArgs, 0, sizeof(s_beginFieldArgs));
m_invalid = false;
frameCount = 0;
CommandProcessor::Init();
Fifo::Init();
OpcodeDecoder::Init();
PixelEngine::Init();
BPInit();
VertexLoaderManager::Init();
IndexGenerator::Init();
VertexShaderManager::Init();
GeometryShaderManager::Init();
PixelShaderManager::Init();
g_Config.Load(File::GetUserPath(D_CONFIG_IDX) + "GFX.ini");
g_Config.GameIniLoad();
g_Config.UpdateProjectionHack();
g_Config.VerifyValidity();
UpdateActiveConfig();
// Notify the core that the video backend is ready
Host_Message(WM_USER_CREATE);
}
void HostDispatchJobs()
{
// WARNING: This should only run on the Host Thread.
// NOTE: This function is potentially re-entrant. If a job calls
// Core::Stop for instance then we'll enter this a second time.
std::unique_lock<std::mutex> guard(s_host_jobs_lock);
while (!s_host_jobs_queue.empty())
{
HostJob job = std::move(s_host_jobs_queue.front());
s_host_jobs_queue.pop();
// NOTE: Memory ordering is important. The booting flag needs to be
// checked first because the state transition is:
// CORE_UNINITIALIZED: s_is_booting -> s_hardware_initialized
// We need to check variables in the same order as the state
// transition, otherwise we race and get transient failures.
if (!job.run_after_stop && !s_is_booting.IsSet() && !IsRunning())
continue;
guard.unlock();
job.job();
guard.lock();
}
}
void HiresTexture::Update()
{
bool BuildMaterialMaps = g_ActiveConfig.bHiresMaterialMapsBuild;
if (s_prefetcher.joinable())
{
s_textureCacheAbortLoading.Set();
s_prefetcher.join();
}
if (!g_ActiveConfig.bHiresTextures)
{
s_textureMap.clear();
s_textureCache.clear();
size_sum.store(0);
return;
}
if (!g_ActiveConfig.bCacheHiresTextures)
{
s_textureCache.clear();
size_sum.store(0);
}
s_textureMap.clear();
const std::string& game_id = SConfig::GetInstance().GetGameID();
const std::string texture_directory = GetTextureDirectory(game_id);
std::string ddscode(".dds");
std::string cddscode(".DDS");
std::vector<std::string> Extensions;
Extensions.push_back(".png");
if (!BuildMaterialMaps)
{
Extensions.push_back(".dds");
}
std::vector<std::string> filenames =
Common::DoFileSearch({texture_directory}, Extensions, /*recursive*/ true);
const std::string miptag = "mip";
for (const std::string& fileitem : filenames)
{
std::string filename;
std::string Extension;
SplitPath(fileitem, nullptr, &filename, &Extension);
if (filename.substr(0, s_format_prefix.length()) != s_format_prefix)
{
// Discard wrong files
continue;
}
size_t map_index = 0;
size_t max_type = BuildMaterialMaps ? MapType::specular : MapType::normal;
bool arbitrary_mips = false;
for (size_t tag = 1; tag <= MapType::specular; tag++)
{
if (StringEndsWith(filename, s_maps_tags[tag]))
{
map_index = tag;
filename = filename.substr(0, filename.size() - s_maps_tags[tag].size());
break;
}
}
if (map_index > max_type)
{
continue;
}
if (BuildMaterialMaps && map_index == MapType::material)
{
continue;
}
else if (!BuildMaterialMaps && map_index == MapType::color)
{
const size_t arb_index = filename.rfind("_arb");
arbitrary_mips = arb_index != std::string::npos;
if (arbitrary_mips)
filename.erase(arb_index, 4);
}
const bool is_compressed = Extension.compare(ddscode) == 0 || Extension.compare(cddscode) == 0;
hires_mip_level mip_level_detail(fileitem, Extension, is_compressed);
u32 level = 0;
size_t idx = filename.find_last_of('_');
std::string miplevel = filename.substr(idx + 1, std::string::npos);
if (miplevel.substr(0, miptag.length()) == miptag)
{
sscanf(miplevel.substr(3, std::string::npos).c_str(), "%i", &level);
filename = filename.substr(0, idx);
}
HiresTextureCache::iterator iter = s_textureMap.find(filename);
u32 min_item_size = level + 1;
if (iter == s_textureMap.end())
{
HiresTextureCacheItem item(min_item_size);
if (arbitrary_mips)
{
item.has_arbitrary_mips = true;
}
item.maps[map_index].resize(min_item_size);
std::vector<hires_mip_level>& dst = item.maps[map_index];
dst[level] = mip_level_detail;
s_textureMap.emplace(filename, item);
}
else
{
std::vector<hires_mip_level>& dst = iter->second.maps[map_index];
if (arbitrary_mips)
{
iter->second.has_arbitrary_mips = true;
}
if (dst.size() < min_item_size)
{
dst.resize(min_item_size);
}
dst[level] = mip_level_detail;
}
}
if (g_ActiveConfig.bCacheHiresTextures && s_textureMap.size() > 0)
{
// remove cached but deleted textures
auto iter = s_textureCache.begin();
while (iter != s_textureCache.end())
{
if (s_textureMap.find(iter->first) == s_textureMap.end())
{
size_sum.fetch_sub(iter->second->m_cached_data_size);
iter = s_textureCache.erase(iter);
}
else
{
iter++;
}
}
s_textureCacheAbortLoading.Clear();
s_prefetcher = std::thread(Prefetch);
if (g_ActiveConfig.bWaitForCacheHiresTextures && s_prefetcher.joinable())
{
s_prefetcher.join();
}
}
}
void HiresTexture::Prefetch()
{
Common::SetCurrentThreadName("Prefetcher");
const size_t total = s_textureMap.size();
size_t count = 0;
size_t notification = 10;
u32 starttime = Common::Timer::GetTimeMs();
for (const auto& entry : s_textureMap)
{
const std::string& base_filename = entry.first;
std::unique_lock<std::mutex> lk(s_textureCacheMutex);
auto iter = s_textureCache.find(base_filename);
if (iter == s_textureCache.end())
{
lk.unlock();
HiresTexture* ptr =
Load(base_filename, [](size_t requested_size) { return new u8[requested_size]; }, true);
lk.lock();
if (ptr != nullptr)
{
size_sum.fetch_add(ptr->m_cached_data_size);
iter = s_textureCache.insert(
iter, std::make_pair(base_filename, std::shared_ptr<HiresTexture>(ptr)));
}
}
if (s_textureCacheAbortLoading.IsSet())
{
if (g_ActiveConfig.bWaitForCacheHiresTextures)
{
Host_UpdateProgressDialog("", -1, -1);
}
return;
}
if (size_sum.load() > max_mem)
{
Config::SetCurrent(Config::GFX_HIRES_TEXTURES, false);
OSD::AddMessage(
StringFromFormat(
"Custom Textures prefetching after %.1f MB aborted, not enough RAM available",
size_sum / (1024.0 * 1024.0)),
10000);
if (g_ActiveConfig.bWaitForCacheHiresTextures)
{
Host_UpdateProgressDialog("", -1, -1);
}
return;
}
count++;
size_t percent = (count * 100) / total;
if (percent >= notification)
{
if (g_ActiveConfig.bWaitForCacheHiresTextures)
{
Host_UpdateProgressDialog(GetStringT("Prefetching Custom Textures...").c_str(),
static_cast<int>(count), static_cast<int>(total));
}
else
{
OSD::AddMessage(StringFromFormat("Custom Textures prefetching %.1f MB %zu %% finished",
size_sum / (1024.0 * 1024.0), percent),
2000);
}
notification += 10;
}
}
if (g_ActiveConfig.bWaitForCacheHiresTextures)
{
Host_UpdateProgressDialog("", -1, -1);
}
u32 stoptime = Common::Timer::GetTimeMs();
OSD::AddMessage(StringFromFormat("Custom Textures loaded, %.1f MB in %.1f s",
size_sum / (1024.0 * 1024.0), (stoptime - starttime) / 1000.0),
10000);
}
static void StartDVDThread()
{
ASSERT(!s_dvd_thread.joinable());
s_dvd_thread_exiting.Clear();
s_dvd_thread = std::thread(DVDThread);
}
void HiresTexture::Update()
{
s_check_native_format = false;
s_check_new_format = false;
if (s_prefetcher.joinable())
{
s_textureCacheAbortLoading.Set();
s_prefetcher.join();
}
if (!g_ActiveConfig.bHiresTextures)
{
s_textureMap.clear();
s_textureCache.clear();
size_sum.store(0);
return;
}
if (!g_ActiveConfig.bCacheHiresTextures)
{
s_textureCache.clear();
size_sum.store(0);
}
s_textureMap.clear();
const std::string& gameCode = SConfig::GetInstance().m_strUniqueID;
std::string szDir = StringFromFormat("%s%s", File::GetUserPath(D_HIRESTEXTURES_IDX).c_str(), gameCode.c_str());
std::string ddscode(".dds");
std::string cddscode(".DDS");
std::vector<std::string> Extensions = {
".png",
".dds"
};
auto rFilenames = DoFileSearch(Extensions, { szDir }, /*recursive*/ true);
const std::string code = StringFromFormat("%s_", gameCode.c_str());
const std::string miptag = "mip";
const std::string normaltag = ".nrm";
for (u32 i = 0; i < rFilenames.size(); i++)
{
std::string FileName;
std::string Extension;
SplitPath(rFilenames[i], nullptr, &FileName, &Extension);
if (FileName.substr(0, code.length()) == code)
{
s_check_native_format = true;
}
else if (FileName.substr(0, s_format_prefix.length()) == s_format_prefix)
{
s_check_new_format = true;
}
else
{
// Discard wrong files
continue;
}
const bool is_compressed = Extension.compare(ddscode) == 0 || Extension.compare(cddscode) == 0;
const bool is_normal_map = hasEnding(FileName, normaltag);
if (is_normal_map)
{
FileName = FileName.substr(0, FileName.size() - normaltag.size());
}
hires_mip_level mip_level_detail(rFilenames[i], Extension, is_compressed);
u32 level = 0;
size_t idx = FileName.find_last_of('_');
std::string miplevel = FileName.substr(idx + 1, std::string::npos);
if (miplevel.substr(0, miptag.length()) == miptag)
{
sscanf(miplevel.substr(3, std::string::npos).c_str(), "%i", &level);
FileName = FileName.substr(0, idx);
}
HiresTextureCache::iterator iter = s_textureMap.find(FileName);
u32 min_item_size = level + 1;
if (iter == s_textureMap.end())
{
HiresTextureCacheItem item(min_item_size);
if (is_normal_map)
{
item.normal_map.resize(min_item_size);
}
std::vector<hires_mip_level> &dst = is_normal_map ? item.normal_map : item.color_map;
dst[level] = mip_level_detail;
s_textureMap.emplace(FileName, item);
}
else
{
std::vector<hires_mip_level> &dst = is_normal_map ? iter->second.normal_map : iter->second.color_map;
if (dst.size() < min_item_size)
{
dst.resize(min_item_size);
}
dst[level] = mip_level_detail;
}
}
if (g_ActiveConfig.bCacheHiresTextures && s_textureMap.size() > 0)
{
// remove cached but deleted textures
auto iter = s_textureCache.begin();
while (iter != s_textureCache.end())
{
if (s_textureMap.find(iter->first) == s_textureMap.end())
{
size_sum.fetch_sub(iter->second->m_cached_data_size);
iter = s_textureCache.erase(iter);
}
else
{
iter++;
}
}
s_textureCacheAbortLoading.Clear();
s_prefetcher = std::thread(Prefetch);
}
}
namespace DVDThread
{
struct ReadRequest
{
bool copy_to_ram;
u32 output_address;
u64 dvd_offset;
u32 length;
DiscIO::Partition partition;
// This determines which code DVDInterface will run to reply
// to the emulated software. We can't use callbacks,
// because function pointers can't be stored in savestates.
DVDInterface::ReplyType reply_type;
// IDs are used to uniquely identify a request. They must not be
// identical to IDs of any other requests that currently exist, but
// it's fine to re-use IDs of requests that have existed in the past.
u64 id;
// Only used for logging
u64 time_started_ticks;
u64 realtime_started_us;
u64 realtime_done_us;
};
using ReadResult = std::pair<ReadRequest, std::vector<u8>>;
static void StartDVDThread();
static void StopDVDThread();
static void DVDThread();
static void WaitUntilIdle();
static void StartReadInternal(bool copy_to_ram, u32 output_address, u64 dvd_offset, u32 length,
const DiscIO::Partition& partition,
DVDInterface::ReplyType reply_type, s64 ticks_until_completion);
static void FinishRead(u64 id, s64 cycles_late);
static CoreTiming::EventType* s_finish_read;
static u64 s_next_id = 0;
static std::thread s_dvd_thread;
static Common::Event s_request_queue_expanded; // Is set by CPU thread
static Common::Event s_result_queue_expanded; // Is set by DVD thread
static Common::Flag s_dvd_thread_exiting(false); // Is set by CPU thread
static Common::SPSCQueue<ReadRequest, false> s_request_queue;
static Common::SPSCQueue<ReadResult, false> s_result_queue;
static std::map<u64, ReadResult> s_result_map;
static std::unique_ptr<DiscIO::Volume> s_disc;
void Start()
{
s_finish_read = CoreTiming::RegisterEvent("FinishReadDVDThread", FinishRead);
s_request_queue_expanded.Reset();
s_result_queue_expanded.Reset();
s_request_queue.Clear();
s_result_queue.Clear();
// This is reset on every launch for determinism, but it doesn't matter
// much, because this will never get exposed to the emulated game.
s_next_id = 0;
StartDVDThread();
}
static void StartDVDThread()
{
ASSERT(!s_dvd_thread.joinable());
s_dvd_thread_exiting.Clear();
s_dvd_thread = std::thread(DVDThread);
}
void Stop()
{
StopDVDThread();
s_disc.reset();
}
static void StopDVDThread()
{
ASSERT(s_dvd_thread.joinable());
// By setting s_DVD_thread_exiting, we ask the DVD thread to cleanly exit.
// In case the request queue is empty, we need to set s_request_queue_expanded
// so that the DVD thread will wake up and check s_DVD_thread_exiting.
s_dvd_thread_exiting.Set();
s_request_queue_expanded.Set();
s_dvd_thread.join();
}
void DoState(PointerWrap& p)
{
// By waiting for the DVD thread to be done working, we ensure
// that s_request_queue will be empty and that the DVD thread
// won't be touching anything while this function runs.
WaitUntilIdle();
// Move all results from s_result_queue to s_result_map because
// PointerWrap::Do supports std::map but not Common::SPSCQueue.
// This won't affect the behavior of FinishRead.
ReadResult result;
while (s_result_queue.Pop(result))
s_result_map.emplace(result.first.id, std::move(result));
// Both queues are now empty, so we don't need to savestate them.
p.Do(s_result_map);
p.Do(s_next_id);
// s_disc isn't savestated (because it points to files on the
// local system). Instead, we check that the status of the disc
// is the same as when the savestate was made. This won't catch
// cases of having the wrong disc inserted, though.
// TODO: Check the game ID, disc number, revision?
bool had_disc = HasDisc();
p.Do(had_disc);
if (had_disc != HasDisc())
{
if (had_disc)
PanicAlertT("An inserted disc was expected but not found.");
else
s_disc.reset();
}
// TODO: Savestates can be smaller if the buffers of results aren't saved,
// but instead get re-read from the disc when loading the savestate.
// TODO: It would be possible to create a savestate faster by stopping
// the DVD thread regardless of whether there are pending requests.
// After loading a savestate, the debug log in FinishRead will report
// screwed up times for requests that were submitted before the savestate
// was made. Handling that properly may be more effort than it's worth.
}
void SetDisc(std::unique_ptr<DiscIO::Volume> disc)
{
WaitUntilIdle();
s_disc = std::move(disc);
}
bool HasDisc()
{
return s_disc != nullptr;
}
bool IsEncryptedAndHashed()
{
// IsEncryptedAndHashed is thread-safe, so calling WaitUntilIdle isn't necessary.
return s_disc->IsEncryptedAndHashed();
}
DiscIO::Platform GetDiscType()
{
// GetVolumeType is thread-safe, so calling WaitUntilIdle isn't necessary.
return s_disc->GetVolumeType();
}
u64 PartitionOffsetToRawOffset(u64 offset, const DiscIO::Partition& partition)
{
// PartitionOffsetToRawOffset is thread-safe, so calling WaitUntilIdle isn't necessary.
return s_disc->PartitionOffsetToRawOffset(offset, partition);
}
IOS::ES::TMDReader GetTMD(const DiscIO::Partition& partition)
{
WaitUntilIdle();
return s_disc->GetTMD(partition);
}
IOS::ES::TicketReader GetTicket(const DiscIO::Partition& partition)
{
WaitUntilIdle();
return s_disc->GetTicket(partition);
}
bool IsInsertedDiscRunning()
{
if (!s_disc)
return false;
WaitUntilIdle();
return SConfig::GetInstance().GetGameID() == s_disc->GetGameID();
}
bool UpdateRunningGameMetadata(const DiscIO::Partition& partition, std::optional<u64> title_id)
{
if (!s_disc)
return false;
WaitUntilIdle();
if (title_id)
{
const std::optional<u64> volume_title_id = s_disc->GetTitleID(partition);
if (!volume_title_id || *volume_title_id != *title_id)
return false;
}
SConfig::GetInstance().SetRunningGameMetadata(*s_disc, partition);
return true;
}
void WaitUntilIdle()
{
ASSERT(Core::IsCPUThread());
while (!s_request_queue.Empty())
s_result_queue_expanded.Wait();
StopDVDThread();
StartDVDThread();
}
void StartRead(u64 dvd_offset, u32 length, const DiscIO::Partition& partition,
DVDInterface::ReplyType reply_type, s64 ticks_until_completion)
{
StartReadInternal(false, 0, dvd_offset, length, partition, reply_type, ticks_until_completion);
}
void StartReadToEmulatedRAM(u32 output_address, u64 dvd_offset, u32 length,
const DiscIO::Partition& partition, DVDInterface::ReplyType reply_type,
s64 ticks_until_completion)
{
StartReadInternal(true, output_address, dvd_offset, length, partition, reply_type,
ticks_until_completion);
}
static void StartReadInternal(bool copy_to_ram, u32 output_address, u64 dvd_offset, u32 length,
const DiscIO::Partition& partition,
DVDInterface::ReplyType reply_type, s64 ticks_until_completion)
{
ASSERT(Core::IsCPUThread());
ReadRequest request;
request.copy_to_ram = copy_to_ram;
request.output_address = output_address;
request.dvd_offset = dvd_offset;
request.length = length;
request.partition = partition;
request.reply_type = reply_type;
u64 id = s_next_id++;
request.id = id;
request.time_started_ticks = CoreTiming::GetTicks();
request.realtime_started_us = Common::Timer::GetTimeUs();
s_request_queue.Push(std::move(request));
s_request_queue_expanded.Set();
CoreTiming::ScheduleEvent(ticks_until_completion, s_finish_read, id);
}
static void FinishRead(u64 id, s64 cycles_late)
{
// We can't simply pop s_result_queue and always get the ReadResult
// we want, because the DVD thread may add ReadResults to the queue
// in a different order than we want to get them. What we do instead
// is to pop the queue until we find the ReadResult we want (the one
// whose ID matches userdata), which means we may end up popping
// ReadResults that we don't want. We can't add those unwanted results
// back to the queue, because the queue can only have one writer.
// Instead, we add them to a map that only is used by the CPU thread.
// When this function is called again later, it will check the map for
// the wanted ReadResult before it starts searching through the queue.
ReadResult result;
auto it = s_result_map.find(id);
if (it != s_result_map.end())
{
result = std::move(it->second);
s_result_map.erase(it);
}
else
{
while (true)
{
while (!s_result_queue.Pop(result))
s_result_queue_expanded.Wait();
if (result.first.id == id)
break;
else
s_result_map.emplace(result.first.id, std::move(result));
}
}
// We have now obtained the right ReadResult.
const ReadRequest& request = result.first;
const std::vector<u8>& buffer = result.second;
DEBUG_LOG(DVDINTERFACE,
"Disc has been read. Real time: %" PRIu64 " us. "
"Real time including delay: %" PRIu64 " us. "
"Emulated time including delay: %" PRIu64 " us.",
request.realtime_done_us - request.realtime_started_us,
Common::Timer::GetTimeUs() - request.realtime_started_us,
(CoreTiming::GetTicks() - request.time_started_ticks) /
(SystemTimers::GetTicksPerSecond() / 1000000));
if (buffer.size() != request.length)
{
PanicAlertT("The disc could not be read (at 0x%" PRIx64 " - 0x%" PRIx64 ").",
request.dvd_offset, request.dvd_offset + request.length);
}
else
{
if (request.copy_to_ram)
Memory::CopyToEmu(request.output_address, buffer.data(), request.length);
}
// Notify the emulated software that the command has been executed
DVDInterface::FinishExecutingCommand(request.reply_type, DVDInterface::INT_TCINT, cycles_late,
buffer);
}
static void DVDThread()
{
Common::SetCurrentThreadName("DVD thread");
while (true)
{
s_request_queue_expanded.Wait();
if (s_dvd_thread_exiting.IsSet())
return;
ReadRequest request;
while (s_request_queue.Pop(request))
{
FileMonitor::Log(*s_disc, request.partition, request.dvd_offset);
std::vector<u8> buffer(request.length);
if (!s_disc->Read(request.dvd_offset, request.length, buffer.data(), request.partition))
buffer.resize(0);
request.realtime_done_us = Common::Timer::GetTimeUs();
s_result_queue.Push(ReadResult(std::move(request), std::move(buffer)));
s_result_queue_expanded.Set();
if (s_dvd_thread_exiting.IsSet())
return;
}
}
}
} // namespace DVDThread
// Description: Main FIFO update loop
// Purpose: Keep the Core HW updated about the CPU-GPU distance
void RunGpuLoop()
{
AsyncRequests::GetInstance()->SetEnable(true);
AsyncRequests::GetInstance()->SetPassthrough(false);
s_gpu_mainloop.Run(
[] {
const SConfig& param = SConfig::GetInstance();
g_video_backend->PeekMessages();
// Do nothing while paused
if (!s_emu_running_state.IsSet())
return;
if (s_use_deterministic_gpu_thread)
{
AsyncRequests::GetInstance()->PullEvents();
// All the fifo/CP stuff is on the CPU. We just need to run the opcode decoder.
u8* seen_ptr = s_video_buffer_seen_ptr;
u8* write_ptr = s_video_buffer_write_ptr;
// See comment in SyncGPU
if (write_ptr > seen_ptr)
{
s_video_buffer_read_ptr =
OpcodeDecoder::Run(DataReader(s_video_buffer_read_ptr, write_ptr), nullptr, false);
s_video_buffer_seen_ptr = write_ptr;
}
}
else
{
SCPFifoStruct& fifo = CommandProcessor::fifo;
AsyncRequests::GetInstance()->PullEvents();
CommandProcessor::SetCPStatusFromGPU();
// check if we are able to run this buffer
while (!CommandProcessor::IsInterruptWaiting() && fifo.bFF_GPReadEnable &&
fifo.CPReadWriteDistance && !AtBreakpoint())
{
if (param.bSyncGPU && s_sync_ticks.load() < param.iSyncGpuMinDistance)
break;
u32 cyclesExecuted = 0;
u32 readPtr = fifo.CPReadPointer;
ReadDataFromFifo(readPtr);
if (readPtr == fifo.CPEnd)
readPtr = fifo.CPBase;
else
readPtr += 32;
_assert_msg_(COMMANDPROCESSOR, (s32)fifo.CPReadWriteDistance - 32 >= 0,
"Negative fifo.CPReadWriteDistance = %i in FIFO Loop !\nThat can produce "
"instability in the game. Please report it.",
fifo.CPReadWriteDistance - 32);
u8* write_ptr = s_video_buffer_write_ptr;
s_video_buffer_read_ptr = OpcodeDecoder::Run(
DataReader(s_video_buffer_read_ptr, write_ptr), &cyclesExecuted, false);
Common::AtomicStore(fifo.CPReadPointer, readPtr);
Common::AtomicAdd(fifo.CPReadWriteDistance, -32);
if ((write_ptr - s_video_buffer_read_ptr) == 0)
Common::AtomicStore(fifo.SafeCPReadPointer, fifo.CPReadPointer);
CommandProcessor::SetCPStatusFromGPU();
if (param.bSyncGPU)
{
cyclesExecuted = (int)(cyclesExecuted / param.fSyncGpuOverclock);
int old = s_sync_ticks.fetch_sub(cyclesExecuted);
if (old >= param.iSyncGpuMaxDistance &&
old - (int)cyclesExecuted < param.iSyncGpuMaxDistance)
s_sync_wakeup_event.Set();
}
// This call is pretty important in DualCore mode and must be called in the FIFO Loop.
// If we don't, s_swapRequested or s_efbAccessRequested won't be set to false
// leading the CPU thread to wait in Video_BeginField or Video_AccessEFB thus slowing
// things down.
AsyncRequests::GetInstance()->PullEvents();
}
// fast skip remaining GPU time if fifo is empty
if (s_sync_ticks.load() > 0)
{
int old = s_sync_ticks.exchange(0);
if (old >= param.iSyncGpuMaxDistance)
s_sync_wakeup_event.Set();
}
// The fifo is empty and it's unlikely we will get any more work in the near future.
// Make sure VertexManager finishes drawing any primitives it has stored in it's buffer.
g_vertex_manager->Flush();
}
},
100);
AsyncRequests::GetInstance()->SetEnable(false);
AsyncRequests::GetInstance()->SetPassthrough(true);
}
int main(int argc, char* argv[])
{
auto parser = CommandLineParse::CreateParser(CommandLineParse::ParserOptions::OmitGUIOptions);
optparse::Values& options = CommandLineParse::ParseArguments(parser.get(), argc, argv);
std::vector<std::string> args = parser->args();
std::string boot_filename;
if (options.is_set("exec"))
{
boot_filename = static_cast<const char*>(options.get("exec"));
}
else if (args.size())
{
boot_filename = args.front();
args.erase(args.begin());
}
else
{
parser->print_help();
return 0;
}
std::string user_directory;
if (options.is_set("user"))
{
user_directory = static_cast<const char*>(options.get("user"));
}
platform = GetPlatform();
if (!platform)
{
fprintf(stderr, "No platform found\n");
return 1;
}
UICommon::SetUserDirectory(user_directory);
UICommon::Init();
Core::SetOnStoppedCallback([]() { s_running.Clear(); });
platform->Init();
// Shut down cleanly on SIGINT and SIGTERM
struct sigaction sa;
sa.sa_handler = signal_handler;
sigemptyset(&sa.sa_mask);
sa.sa_flags = SA_RESETHAND;
sigaction(SIGINT, &sa, nullptr);
sigaction(SIGTERM, &sa, nullptr);
DolphinAnalytics::Instance()->ReportDolphinStart("nogui");
if (!BootManager::BootCore(BootParameters::GenerateFromFile(boot_filename)))
{
fprintf(stderr, "Could not boot %s\n", boot_filename.c_str());
return 1;
}
while (!Core::IsRunning() && s_running.IsSet())
{
Core::HostDispatchJobs();
updateMainFrameEvent.Wait();
}
if (s_running.IsSet())
platform->MainLoop();
Core::Stop();
Core::Shutdown();
platform->Shutdown();
UICommon::Shutdown();
delete platform;
return 0;
}
void MainLoop() override
{
bool fullscreen = SConfig::GetInstance().bFullscreen;
int last_window_width = SConfig::GetInstance().iRenderWindowWidth;
int last_window_height = SConfig::GetInstance().iRenderWindowHeight;
if (fullscreen)
{
rendererIsFullscreen = X11Utils::ToggleFullscreen(dpy, win);
#if defined(HAVE_XRANDR) && HAVE_XRANDR
XRRConfig->ToggleDisplayMode(True);
#endif
}
// The actual loop
while (s_running.IsSet())
{
if (s_shutdown_requested.TestAndClear())
{
const auto ios = IOS::HLE::GetIOS();
const auto stm = ios ? ios->GetDeviceByName("/dev/stm/eventhook") : nullptr;
if (!s_tried_graceful_shutdown.IsSet() && stm &&
std::static_pointer_cast<IOS::HLE::Device::STMEventHook>(stm)->HasHookInstalled())
{
ProcessorInterface::PowerButton_Tap();
s_tried_graceful_shutdown.Set();
}
else
{
s_running.Clear();
}
}
XEvent event;
KeySym key;
for (int num_events = XPending(dpy); num_events > 0; num_events--)
{
XNextEvent(dpy, &event);
switch (event.type)
{
case KeyPress:
key = XLookupKeysym((XKeyEvent*)&event, 0);
if (key == XK_Escape)
{
if (Core::GetState() == Core::State::Running)
{
if (SConfig::GetInstance().bHideCursor)
XUndefineCursor(dpy, win);
Core::SetState(Core::State::Paused);
}
else
{
if (SConfig::GetInstance().bHideCursor)
XDefineCursor(dpy, win, blankCursor);
Core::SetState(Core::State::Running);
}
}
else if ((key == XK_Return) && (event.xkey.state & Mod1Mask))
{
fullscreen = !fullscreen;
X11Utils::ToggleFullscreen(dpy, win);
#if defined(HAVE_XRANDR) && HAVE_XRANDR
XRRConfig->ToggleDisplayMode(fullscreen);
#endif
}
else if (key >= XK_F1 && key <= XK_F8)
{
int slot_number = key - XK_F1 + 1;
if (event.xkey.state & ShiftMask)
State::Save(slot_number);
else
State::Load(slot_number);
}
else if (key == XK_F9)
Core::SaveScreenShot();
else if (key == XK_F11)
State::LoadLastSaved();
else if (key == XK_F12)
{
if (event.xkey.state & ShiftMask)
State::UndoLoadState();
else
State::UndoSaveState();
}
break;
case FocusIn:
rendererHasFocus = true;
if (SConfig::GetInstance().bHideCursor && Core::GetState() != Core::State::Paused)
XDefineCursor(dpy, win, blankCursor);
break;
case FocusOut:
rendererHasFocus = false;
if (SConfig::GetInstance().bHideCursor)
XUndefineCursor(dpy, win);
break;
case ClientMessage:
if ((unsigned long)event.xclient.data.l[0] == XInternAtom(dpy, "WM_DELETE_WINDOW", False))
s_shutdown_requested.Set();
break;
case ConfigureNotify:
{
if (last_window_width != event.xconfigure.width ||
last_window_height != event.xconfigure.height)
{
last_window_width = event.xconfigure.width;
last_window_height = event.xconfigure.height;
// We call Renderer::ChangeSurface here to indicate the size has changed,
// but pass the same window handle. This is needed for the Vulkan backend,
// otherwise it cannot tell that the window has been resized on some drivers.
if (g_renderer)
g_renderer->ChangeSurface(s_window_handle);
}
}
break;
}
}
if (!fullscreen)
{
Window winDummy;
unsigned int borderDummy, depthDummy;
XGetGeometry(dpy, win, &winDummy, &SConfig::GetInstance().iRenderWindowXPos,
&SConfig::GetInstance().iRenderWindowYPos,
(unsigned int*)&SConfig::GetInstance().iRenderWindowWidth,
(unsigned int*)&SConfig::GetInstance().iRenderWindowHeight, &borderDummy,
&depthDummy);
rendererIsFullscreen = false;
}
Core::HostDispatchJobs();
usleep(100000);
}
}
namespace Fifo
{
static constexpr u32 FIFO_SIZE = 2 * 1024 * 1024;
static constexpr int GPU_TIME_SLOT_SIZE = 1000;
static Common::BlockingLoop s_gpu_mainloop;
static Common::Flag s_emu_running_state;
// Most of this array is unlikely to be faulted in...
static u8 s_fifo_aux_data[FIFO_SIZE];
static u8* s_fifo_aux_write_ptr;
static u8* s_fifo_aux_read_ptr;
// This could be in SConfig, but it depends on multiple settings
// and can change at runtime.
static bool s_use_deterministic_gpu_thread;
static CoreTiming::EventType* s_event_sync_gpu;
// STATE_TO_SAVE
static u8* s_video_buffer;
static u8* s_video_buffer_read_ptr;
static std::atomic<u8*> s_video_buffer_write_ptr;
static std::atomic<u8*> s_video_buffer_seen_ptr;
static u8* s_video_buffer_pp_read_ptr;
// The read_ptr is always owned by the GPU thread. In normal mode, so is the
// write_ptr, despite it being atomic. In deterministic GPU thread mode,
// things get a bit more complicated:
// - The seen_ptr is written by the GPU thread, and points to what it's already
// processed as much of as possible - in the case of a partial command which
// caused it to stop, not the same as the read ptr. It's written by the GPU,
// under the lock, and updating the cond.
// - The write_ptr is written by the CPU thread after it copies data from the
// FIFO. Maybe someday it will be under the lock. For now, because RunGpuLoop
// polls, it's just atomic.
// - The pp_read_ptr is the CPU preprocessing version of the read_ptr.
static std::atomic<int> s_sync_ticks;
static bool s_syncing_suspended;
static Common::Event s_sync_wakeup_event;
void DoState(PointerWrap& p)
{
p.DoArray(s_video_buffer, FIFO_SIZE);
u8* write_ptr = s_video_buffer_write_ptr;
p.DoPointer(write_ptr, s_video_buffer);
s_video_buffer_write_ptr = write_ptr;
p.DoPointer(s_video_buffer_read_ptr, s_video_buffer);
if (p.mode == PointerWrap::MODE_READ && s_use_deterministic_gpu_thread)
{
// We're good and paused, right?
s_video_buffer_seen_ptr = s_video_buffer_pp_read_ptr = s_video_buffer_read_ptr;
}
p.Do(s_sync_ticks);
p.Do(s_syncing_suspended);
}
void PauseAndLock(bool doLock, bool unpauseOnUnlock)
{
if (doLock)
{
SyncGPU(SyncGPUReason::Other);
EmulatorState(false);
const SConfig& param = SConfig::GetInstance();
if (!param.bCPUThread || s_use_deterministic_gpu_thread)
return;
s_gpu_mainloop.WaitYield(std::chrono::milliseconds(100), Host_YieldToUI);
}
else
{
if (unpauseOnUnlock)
EmulatorState(true);
}
}
void Init()
{
// Padded so that SIMD overreads in the vertex loader are safe
s_video_buffer = static_cast<u8*>(Common::AllocateMemoryPages(FIFO_SIZE + 4));
ResetVideoBuffer();
if (SConfig::GetInstance().bCPUThread)
s_gpu_mainloop.Prepare();
s_sync_ticks.store(0);
}
void Shutdown()
{
if (s_gpu_mainloop.IsRunning())
PanicAlert("Fifo shutting down while active");
Common::FreeMemoryPages(s_video_buffer, FIFO_SIZE + 4);
s_video_buffer = nullptr;
s_video_buffer_write_ptr = nullptr;
s_video_buffer_pp_read_ptr = nullptr;
s_video_buffer_read_ptr = nullptr;
s_video_buffer_seen_ptr = nullptr;
s_fifo_aux_write_ptr = nullptr;
s_fifo_aux_read_ptr = nullptr;
}
// May be executed from any thread, even the graphics thread.
// Created to allow for self shutdown.
void ExitGpuLoop()
{
// This should break the wait loop in CPU thread
CommandProcessor::fifo.bFF_GPReadEnable = false;
FlushGpu();
// Terminate GPU thread loop
s_emu_running_state.Set();
s_gpu_mainloop.Stop(false);
}
void EmulatorState(bool running)
{
s_emu_running_state.Set(running);
if (running)
s_gpu_mainloop.Wakeup();
else
s_gpu_mainloop.AllowSleep();
}
void SyncGPU(SyncGPUReason reason, bool may_move_read_ptr)
{
if (s_use_deterministic_gpu_thread)
{
s_gpu_mainloop.Wait();
if (!s_gpu_mainloop.IsRunning())
return;
// Opportunistically reset FIFOs so we don't wrap around.
if (may_move_read_ptr && s_fifo_aux_write_ptr != s_fifo_aux_read_ptr)
PanicAlert("aux fifo not synced (%p, %p)", s_fifo_aux_write_ptr, s_fifo_aux_read_ptr);
memmove(s_fifo_aux_data, s_fifo_aux_read_ptr, s_fifo_aux_write_ptr - s_fifo_aux_read_ptr);
s_fifo_aux_write_ptr -= (s_fifo_aux_read_ptr - s_fifo_aux_data);
s_fifo_aux_read_ptr = s_fifo_aux_data;
if (may_move_read_ptr)
{
u8* write_ptr = s_video_buffer_write_ptr;
// what's left over in the buffer
size_t size = write_ptr - s_video_buffer_pp_read_ptr;
memmove(s_video_buffer, s_video_buffer_pp_read_ptr, size);
// This change always decreases the pointers. We write seen_ptr
// after write_ptr here, and read it before in RunGpuLoop, so
// 'write_ptr > seen_ptr' there cannot become spuriously true.
s_video_buffer_write_ptr = write_ptr = s_video_buffer + size;
s_video_buffer_pp_read_ptr = s_video_buffer;
s_video_buffer_read_ptr = s_video_buffer;
s_video_buffer_seen_ptr = write_ptr;
}
}
}
void PushFifoAuxBuffer(void* ptr, size_t size)
{
if (size > (size_t)(s_fifo_aux_data + FIFO_SIZE - s_fifo_aux_write_ptr))
{
SyncGPU(SyncGPUReason::AuxSpace, /* may_move_read_ptr */ false);
if (!s_gpu_mainloop.IsRunning())
{
// GPU is shutting down
return;
}
if (size > (size_t)(s_fifo_aux_data + FIFO_SIZE - s_fifo_aux_write_ptr))
{
// That will sync us up to the last 32 bytes, so this short region
// of FIFO would have to point to a 2MB display list or something.
PanicAlert("absurdly large aux buffer");
return;
}
}
memcpy(s_fifo_aux_write_ptr, ptr, size);
s_fifo_aux_write_ptr += size;
}
void* PopFifoAuxBuffer(size_t size)
{
void* ret = s_fifo_aux_read_ptr;
s_fifo_aux_read_ptr += size;
return ret;
}
// Description: RunGpuLoop() sends data through this function.
static void ReadDataFromFifo(u32 readPtr)
{
size_t len = 32;
if (len > (size_t)(s_video_buffer + FIFO_SIZE - s_video_buffer_write_ptr))
{
size_t existing_len = s_video_buffer_write_ptr - s_video_buffer_read_ptr;
if (len > (size_t)(FIFO_SIZE - existing_len))
{
PanicAlert("FIFO out of bounds (existing %zu + new %zu > %lu)", existing_len, len,
(unsigned long)FIFO_SIZE);
return;
}
memmove(s_video_buffer, s_video_buffer_read_ptr, existing_len);
s_video_buffer_write_ptr = s_video_buffer + existing_len;
s_video_buffer_read_ptr = s_video_buffer;
}
// Copy new video instructions to s_video_buffer for future use in rendering the new picture
Memory::CopyFromEmu(s_video_buffer_write_ptr, readPtr, len);
s_video_buffer_write_ptr += len;
}
// The deterministic_gpu_thread version.
static void ReadDataFromFifoOnCPU(u32 readPtr)
{
size_t len = 32;
u8* write_ptr = s_video_buffer_write_ptr;
if (len > (size_t)(s_video_buffer + FIFO_SIZE - write_ptr))
{
// We can't wrap around while the GPU is working on the data.
// This should be very rare due to the reset in SyncGPU.
SyncGPU(SyncGPUReason::Wraparound);
if (!s_gpu_mainloop.IsRunning())
{
// GPU is shutting down, so the next asserts may fail
return;
}
if (s_video_buffer_pp_read_ptr != s_video_buffer_read_ptr)
{
PanicAlert("desynced read pointers");
return;
}
write_ptr = s_video_buffer_write_ptr;
size_t existing_len = write_ptr - s_video_buffer_pp_read_ptr;
if (len > (size_t)(FIFO_SIZE - existing_len))
{
PanicAlert("FIFO out of bounds (existing %zu + new %zu > %lu)", existing_len, len,
(unsigned long)FIFO_SIZE);
return;
}
}
Memory::CopyFromEmu(s_video_buffer_write_ptr, readPtr, len);
s_video_buffer_pp_read_ptr = OpcodeDecoder::Run<true>(
DataReader(s_video_buffer_pp_read_ptr, write_ptr + len), nullptr, false);
// This would have to be locked if the GPU thread didn't spin.
s_video_buffer_write_ptr = write_ptr + len;
}
void ResetVideoBuffer()
{
s_video_buffer_read_ptr = s_video_buffer;
s_video_buffer_write_ptr = s_video_buffer;
s_video_buffer_seen_ptr = s_video_buffer;
s_video_buffer_pp_read_ptr = s_video_buffer;
s_fifo_aux_write_ptr = s_fifo_aux_data;
s_fifo_aux_read_ptr = s_fifo_aux_data;
}
// Description: Main FIFO update loop
// Purpose: Keep the Core HW updated about the CPU-GPU distance
void RunGpuLoop()
{
AsyncRequests::GetInstance()->SetEnable(true);
AsyncRequests::GetInstance()->SetPassthrough(false);
s_gpu_mainloop.Run(
[] {
const SConfig& param = SConfig::GetInstance();
g_video_backend->PeekMessages();
// Do nothing while paused
if (!s_emu_running_state.IsSet())
return;
if (s_use_deterministic_gpu_thread)
{
AsyncRequests::GetInstance()->PullEvents();
// All the fifo/CP stuff is on the CPU. We just need to run the opcode decoder.
u8* seen_ptr = s_video_buffer_seen_ptr;
u8* write_ptr = s_video_buffer_write_ptr;
// See comment in SyncGPU
if (write_ptr > seen_ptr)
{
s_video_buffer_read_ptr =
OpcodeDecoder::Run(DataReader(s_video_buffer_read_ptr, write_ptr), nullptr, false);
s_video_buffer_seen_ptr = write_ptr;
}
}
else
{
SCPFifoStruct& fifo = CommandProcessor::fifo;
AsyncRequests::GetInstance()->PullEvents();
CommandProcessor::SetCPStatusFromGPU();
// check if we are able to run this buffer
while (!CommandProcessor::IsInterruptWaiting() && fifo.bFF_GPReadEnable &&
fifo.CPReadWriteDistance && !AtBreakpoint())
{
if (param.bSyncGPU && s_sync_ticks.load() < param.iSyncGpuMinDistance)
break;
u32 cyclesExecuted = 0;
u32 readPtr = fifo.CPReadPointer;
ReadDataFromFifo(readPtr);
if (readPtr == fifo.CPEnd)
readPtr = fifo.CPBase;
else
readPtr += 32;
_assert_msg_(COMMANDPROCESSOR, (s32)fifo.CPReadWriteDistance - 32 >= 0,
"Negative fifo.CPReadWriteDistance = %i in FIFO Loop !\nThat can produce "
"instability in the game. Please report it.",
fifo.CPReadWriteDistance - 32);
u8* write_ptr = s_video_buffer_write_ptr;
s_video_buffer_read_ptr = OpcodeDecoder::Run(
DataReader(s_video_buffer_read_ptr, write_ptr), &cyclesExecuted, false);
Common::AtomicStore(fifo.CPReadPointer, readPtr);
Common::AtomicAdd(fifo.CPReadWriteDistance, -32);
if ((write_ptr - s_video_buffer_read_ptr) == 0)
Common::AtomicStore(fifo.SafeCPReadPointer, fifo.CPReadPointer);
CommandProcessor::SetCPStatusFromGPU();
if (param.bSyncGPU)
{
cyclesExecuted = (int)(cyclesExecuted / param.fSyncGpuOverclock);
int old = s_sync_ticks.fetch_sub(cyclesExecuted);
if (old >= param.iSyncGpuMaxDistance &&
old - (int)cyclesExecuted < param.iSyncGpuMaxDistance)
s_sync_wakeup_event.Set();
}
// This call is pretty important in DualCore mode and must be called in the FIFO Loop.
// If we don't, s_swapRequested or s_efbAccessRequested won't be set to false
// leading the CPU thread to wait in Video_BeginField or Video_AccessEFB thus slowing
// things down.
AsyncRequests::GetInstance()->PullEvents();
}
// fast skip remaining GPU time if fifo is empty
if (s_sync_ticks.load() > 0)
{
int old = s_sync_ticks.exchange(0);
if (old >= param.iSyncGpuMaxDistance)
s_sync_wakeup_event.Set();
}
// The fifo is empty and it's unlikely we will get any more work in the near future.
// Make sure VertexManager finishes drawing any primitives it has stored in it's buffer.
g_vertex_manager->Flush();
}
},
100);
AsyncRequests::GetInstance()->SetEnable(false);
AsyncRequests::GetInstance()->SetPassthrough(true);
}
void FlushGpu()
{
const SConfig& param = SConfig::GetInstance();
if (!param.bCPUThread || s_use_deterministic_gpu_thread)
return;
s_gpu_mainloop.Wait();
}
void GpuMaySleep()
{
s_gpu_mainloop.AllowSleep();
}
bool AtBreakpoint()
{
SCPFifoStruct& fifo = CommandProcessor::fifo;
return fifo.bFF_BPEnable && (fifo.CPReadPointer == fifo.CPBreakpoint);
}
void RunGpu()
{
const SConfig& param = SConfig::GetInstance();
// wake up GPU thread
if (param.bCPUThread && !s_use_deterministic_gpu_thread)
{
s_gpu_mainloop.Wakeup();
}
// if the sync GPU callback is suspended, wake it up.
if (!SConfig::GetInstance().bCPUThread || s_use_deterministic_gpu_thread ||
SConfig::GetInstance().bSyncGPU)
{
if (s_syncing_suspended)
{
s_syncing_suspended = false;
CoreTiming::ScheduleEvent(GPU_TIME_SLOT_SIZE, s_event_sync_gpu, GPU_TIME_SLOT_SIZE);
}
}
}
static int RunGpuOnCpu(int ticks)
{
SCPFifoStruct& fifo = CommandProcessor::fifo;
bool reset_simd_state = false;
int available_ticks = int(ticks * SConfig::GetInstance().fSyncGpuOverclock) + s_sync_ticks.load();
while (fifo.bFF_GPReadEnable && fifo.CPReadWriteDistance && !AtBreakpoint() &&
available_ticks >= 0)
{
if (s_use_deterministic_gpu_thread)
{
ReadDataFromFifoOnCPU(fifo.CPReadPointer);
s_gpu_mainloop.Wakeup();
}
else
{
if (!reset_simd_state)
{
FPURoundMode::SaveSIMDState();
FPURoundMode::LoadDefaultSIMDState();
reset_simd_state = true;
}
ReadDataFromFifo(fifo.CPReadPointer);
u32 cycles = 0;
s_video_buffer_read_ptr = OpcodeDecoder::Run(
DataReader(s_video_buffer_read_ptr, s_video_buffer_write_ptr), &cycles, false);
available_ticks -= cycles;
}
if (fifo.CPReadPointer == fifo.CPEnd)
fifo.CPReadPointer = fifo.CPBase;
else
fifo.CPReadPointer += 32;
fifo.CPReadWriteDistance -= 32;
}
CommandProcessor::SetCPStatusFromGPU();
if (reset_simd_state)
{
FPURoundMode::LoadSIMDState();
}
// Discard all available ticks as there is nothing to do any more.
s_sync_ticks.store(std::min(available_ticks, 0));
// If the GPU is idle, drop the handler.
if (available_ticks >= 0)
return -1;
// Always wait at least for GPU_TIME_SLOT_SIZE cycles.
return -available_ticks + GPU_TIME_SLOT_SIZE;
}
void UpdateWantDeterminism(bool want)
{
// We are paused (or not running at all yet), so
// it should be safe to change this.
const SConfig& param = SConfig::GetInstance();
bool gpu_thread = false;
switch (param.m_GPUDeterminismMode)
{
case GPU_DETERMINISM_AUTO:
gpu_thread = want;
// Hack: For now movies are an exception to this being on (but not
// to wanting determinism in general). Once vertex arrays are
// fixed, there should be no reason to want this off for movies by
// default, so this can be removed.
if (!NetPlay::IsNetPlayRunning())
gpu_thread = false;
break;
case GPU_DETERMINISM_NONE:
gpu_thread = false;
break;
case GPU_DETERMINISM_FAKE_COMPLETION:
gpu_thread = true;
break;
}
gpu_thread = gpu_thread && param.bCPUThread;
if (s_use_deterministic_gpu_thread != gpu_thread)
{
s_use_deterministic_gpu_thread = gpu_thread;
if (gpu_thread)
{
// These haven't been updated in non-deterministic mode.
s_video_buffer_seen_ptr = s_video_buffer_pp_read_ptr = s_video_buffer_read_ptr;
CopyPreprocessCPStateFromMain();
VertexLoaderManager::MarkAllDirty();
}
}
}
bool UseDeterministicGPUThread()
{
return s_use_deterministic_gpu_thread;
}
/* This function checks the emulated CPU - GPU distance and may wake up the GPU,
* or block the CPU if required. It should be called by the CPU thread regularly.
* @ticks The gone emulated CPU time.
* @return A good time to call WaitForGpuThread() next.
*/
static int WaitForGpuThread(int ticks)
{
const SConfig& param = SConfig::GetInstance();
int old = s_sync_ticks.fetch_add(ticks);
int now = old + ticks;
// GPU is idle, so stop polling.
if (old >= 0 && s_gpu_mainloop.IsDone())
return -1;
// Wakeup GPU
if (old < param.iSyncGpuMinDistance && now >= param.iSyncGpuMinDistance)
RunGpu();
// If the GPU is still sleeping, wait for a longer time
if (now < param.iSyncGpuMinDistance)
return GPU_TIME_SLOT_SIZE + param.iSyncGpuMinDistance - now;
// Wait for GPU
if (now >= param.iSyncGpuMaxDistance)
s_sync_wakeup_event.Wait();
return GPU_TIME_SLOT_SIZE;
}
static void SyncGPUCallback(u64 ticks, s64 cyclesLate)
{
ticks += cyclesLate;
int next = -1;
if (!SConfig::GetInstance().bCPUThread || s_use_deterministic_gpu_thread)
{
next = RunGpuOnCpu((int)ticks);
}
else if (SConfig::GetInstance().bSyncGPU)
{
next = WaitForGpuThread((int)ticks);
}
s_syncing_suspended = next < 0;
if (!s_syncing_suspended)
CoreTiming::ScheduleEvent(next, s_event_sync_gpu, next);
}
// Initialize GPU - CPU thread syncing, this gives us a deterministic way to start the GPU thread.
void Prepare()
{
s_event_sync_gpu = CoreTiming::RegisterEvent("SyncGPUCallback", SyncGPUCallback);
s_syncing_suspended = true;
}
}