static u32 InterpretDisplayList(u32 address, u32 size)
{
u8* startAddress;
if (Fifo::g_use_deterministic_gpu_thread)
startAddress = (u8*)Fifo::PopFifoAuxBuffer(size);
else
startAddress = Memory::GetPointer(address);
u32 cycles = 0;
// Avoid the crash if Memory::GetPointer failed ..
if (startAddress != nullptr)
{
// temporarily swap dl and non-dl (small "hack" for the stats)
Statistics::SwapDL();
OpcodeDecoder_Run(DataReader(startAddress, startAddress + size), &cycles, true);
INCSTAT(stats.thisFrame.numDListsCalled);
// un-swap
Statistics::SwapDL();
}
return cycles;
}
void RunGpu()
{
if (SConfig::GetInstance().m_LocalCoreStartupParameter.bCPUThread &&
!g_use_deterministic_gpu_thread)
return;
SCPFifoStruct &fifo = CommandProcessor::fifo;
while (fifo.bFF_GPReadEnable && fifo.CPReadWriteDistance && !AtBreakpoint() )
{
if (g_use_deterministic_gpu_thread)
{
ReadDataFromFifoOnCPU(fifo.CPReadPointer);
}
else
{
FPURoundMode::SaveSIMDState();
FPURoundMode::LoadDefaultSIMDState();
ReadDataFromFifo(fifo.CPReadPointer);
s_video_buffer_read_ptr = OpcodeDecoder_Run(DataReader(s_video_buffer_read_ptr, s_video_buffer_write_ptr), nullptr, false);
FPURoundMode::LoadSIMDState();
}
//DEBUG_LOG(COMMANDPROCESSOR, "Fifo wraps to base");
if (fifo.CPReadPointer == fifo.CPEnd)
fifo.CPReadPointer = fifo.CPBase;
else
fifo.CPReadPointer += 32;
fifo.CPReadWriteDistance -= 32;
}
CommandProcessor::SetCPStatusFromGPU();
}
// 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(SYNC_GPU_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 %lu + new %lu > %lu)", (unsigned long) existing_len, (unsigned long) 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;
}
static void InterpretDisplayListPreprocess(u32 address, u32 size)
{
u8* startAddress = Memory::GetPointer(address);
Fifo::PushFifoAuxBuffer(startAddress, size);
if (startAddress != nullptr)
{
OpcodeDecoder_Run<true>(DataReader(startAddress, startAddress + size), nullptr, true);
}
}
int RunVertices(DataReader src, DataReader dst, int count) override
{
buffer_a.resize(count * a->m_native_vtx_decl.stride + 4);
buffer_b.resize(count * b->m_native_vtx_decl.stride + 4);
int count_a = a->RunVertices(src, DataReader(buffer_a.data(), buffer_a.data()+buffer_a.size()), count);
int count_b = b->RunVertices(src, DataReader(buffer_b.data(), buffer_b.data()+buffer_b.size()), count);
if (count_a != count_b)
ERROR_LOG(VIDEO, "The two vertex loaders have loaded a different amount of vertices (a: %d, b: %d).", count_a, count_b);
if (memcmp(buffer_a.data(), buffer_b.data(), std::min(count_a, count_b) * m_native_vtx_decl.stride))
ERROR_LOG(VIDEO, "The two vertex loaders have loaded different data "
"(guru meditation 0x%016" PRIx64 ", 0x%08x, 0x%08x, 0x%08x).",
m_VtxDesc.Hex, m_vat.g0.Hex, m_vat.g1.Hex, m_vat.g2.Hex);
memcpy(dst.GetPointer(), buffer_a.data(), count_a * m_native_vtx_decl.stride);
m_numLoadedVertices += count;
return count_a;
}
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 RunGpu()
{
SCPFifoStruct &fifo = CommandProcessor::fifo;
const SConfig& param = SConfig::GetInstance();
// execute GPU
if (!param.bCPUThread || g_use_deterministic_gpu_thread)
{
bool reset_simd_state = false;
while (fifo.bFF_GPReadEnable && fifo.CPReadWriteDistance && !AtBreakpoint() )
{
if (g_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);
s_video_buffer_read_ptr = OpcodeDecoder_Run(DataReader(s_video_buffer_read_ptr, s_video_buffer_write_ptr), nullptr, false);
}
//DEBUG_LOG(COMMANDPROCESSOR, "Fifo wraps to base");
if (fifo.CPReadPointer == fifo.CPEnd)
fifo.CPReadPointer = fifo.CPBase;
else
fifo.CPReadPointer += 32;
fifo.CPReadWriteDistance -= 32;
}
CommandProcessor::SetCPStatusFromGPU();
if (reset_simd_state)
{
FPURoundMode::LoadSIMDState();
}
}
// wake up GPU thread
if (param.bCPUThread)
{
s_gpu_mainloop.Wakeup();
}
}
// 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.load())
return;
if (g_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 > 0 && old - (int)cyclesExecuted <= 0)
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 > 0)
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.
VertexManager::Flush();
}
}, 100);
AsyncRequests::GetInstance()->SetEnable(false);
AsyncRequests::GetInstance()->SetPassthrough(true);
}
void ResetPointers()
{
m_src = DataReader(input_memory, input_memory + sizeof(input_memory));
m_dst = DataReader(output_memory, output_memory + sizeof(output_memory));
}
void ResetPointers()
{
m_input_pos = m_output_pos = 0;
src = DataReader(input_memory, input_memory+sizeof(input_memory));
dst = DataReader(output_memory, output_memory+sizeof(output_memory));
}
// Description: Main FIFO update loop
// Purpose: Keep the Core HW updated about the CPU-GPU distance
void RunGpuLoop()
{
std::lock_guard<std::mutex> lk(m_csHWVidOccupied);
GpuRunningState = true;
SCPFifoStruct &fifo = CommandProcessor::fifo;
u32 cyclesExecuted = 0;
// If the host CPU has only two cores, idle loop instead of busy loop
// This allows a system that we are maxing out in dual core mode to do other things
bool yield_cpu = cpu_info.num_cores <= 2;
AsyncRequests::GetInstance()->SetEnable(true);
AsyncRequests::GetInstance()->SetPassthrough(false);
while (GpuRunningState)
{
g_video_backend->PeekMessages();
AsyncRequests::GetInstance()->PullEvents();
if (g_use_deterministic_gpu_thread)
{
// 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);
{
std::lock_guard<std::mutex> vblk(s_video_buffer_lock);
s_video_buffer_seen_ptr = write_ptr;
s_video_buffer_cond.notify_all();
}
}
}
else
{
CommandProcessor::SetCPStatusFromGPU();
Common::AtomicStore(CommandProcessor::VITicks, CommandProcessor::m_cpClockOrigin);
// check if we are able to run this buffer
while (GpuRunningState && EmuRunningState && !CommandProcessor::interruptWaiting && fifo.bFF_GPReadEnable && fifo.CPReadWriteDistance && !AtBreakpoint())
{
fifo.isGpuReadingData = true;
CommandProcessor::isPossibleWaitingSetDrawDone = fifo.bFF_GPLinkEnable ? true : false;
if (!SConfig::GetInstance().m_LocalCoreStartupParameter.bSyncGPU || Common::AtomicLoad(CommandProcessor::VITicks) > CommandProcessor::m_cpClockOrigin)
{
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);
if (SConfig::GetInstance().m_LocalCoreStartupParameter.bSyncGPU && Common::AtomicLoad(CommandProcessor::VITicks) >= cyclesExecuted)
Common::AtomicAdd(CommandProcessor::VITicks, -(s32)cyclesExecuted);
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();
// 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();
CommandProcessor::isPossibleWaitingSetDrawDone = false;
}
fifo.isGpuReadingData = false;
}
if (EmuRunningState)
{
// NOTE(jsd): Calling SwitchToThread() on Windows 7 x64 is a hot spot, according to profiler.
// See https://docs.google.com/spreadsheet/ccc?key=0Ah4nh0yGtjrgdFpDeF9pS3V6RUotRVE3S3J4TGM1NlE#gid=0
// for benchmark details.
if (yield_cpu)
Common::YieldCPU();
}
else
{
// While the emu is paused, we still handle async requests then sleep.
while (!EmuRunningState)
{
g_video_backend->PeekMessages();
m_csHWVidOccupied.unlock();
Common::SleepCurrentThread(1);
m_csHWVidOccupied.lock();
}
}
}
// wake up SyncGPU if we were interrupted
s_video_buffer_cond.notify_all();
AsyncRequests::GetInstance()->SetEnable(false);
AsyncRequests::GetInstance()->SetPassthrough(true);
}
