Merge pull request #7214 from stenzek/cp-access-sync

Fifo: Run/sync with the GPU on command processor register access

Source/Core/VideoCommon/CommandProcessor.cpp

@@ -234,6 +234,7 @@ void RegisterMMIO(MMIO::Mapping* mmio, u32 base)
   }
 
   mmio->Register(base | STATUS_REGISTER, MMIO::ComplexRead<u16>([](u32) {
+                   Fifo::SyncGPUForRegisterAccess();
                    SetCpStatusRegister();
                    return m_CPStatusReg.Hex;
                  }),
@@ -271,18 +272,21 @@ void RegisterMMIO(MMIO::Mapping* mmio, u32 base)
                  MMIO::DirectWrite<u16>(MMIO::Utils::LowPart(&fifo.CPReadWriteDistance),
                                         WMASK_LO_ALIGN_32BIT));
   mmio->Register(base | FIFO_RW_DISTANCE_HI,
-                 IsOnThread() ?
-                     MMIO::ComplexRead<u16>([](u32) {
+                 IsOnThread() ? MMIO::ComplexRead<u16>([](u32) {
+                   Fifo::SyncGPUForRegisterAccess();
                    if (fifo.CPWritePointer >= fifo.SafeCPReadPointer)
                      return ReadHigh(fifo.CPWritePointer - fifo.SafeCPReadPointer);
                    else
                      return ReadHigh(fifo.CPEnd - fifo.SafeCPReadPointer + fifo.CPWritePointer -
                                      fifo.CPBase + 32);
                  }) :
-                     MMIO::DirectRead<u16>(MMIO::Utils::HighPart(&fifo.CPReadWriteDistance)),
+                                MMIO::ComplexRead<u16>([](u32) {
+                                  Fifo::SyncGPUForRegisterAccess();
+                                  return ReadHigh(fifo.CPReadWriteDistance);
+                                }),
                  MMIO::ComplexWrite<u16>([WMASK_HI_RESTRICT](u32, u16 val) {
+                   Fifo::SyncGPUForRegisterAccess();
                    WriteHigh(fifo.CPReadWriteDistance, val & WMASK_HI_RESTRICT);
-                   Fifo::SyncGPU(Fifo::SyncGPUReason::Other);
                    Fifo::RunGpu();
                  }));
   mmio->Register(
@@ -290,16 +294,24 @@ void RegisterMMIO(MMIO::Mapping* mmio, u32 base)
       IsOnThread() ? MMIO::DirectRead<u16>(MMIO::Utils::LowPart(&fifo.SafeCPReadPointer)) :
                      MMIO::DirectRead<u16>(MMIO::Utils::LowPart(&fifo.CPReadPointer)),
       MMIO::DirectWrite<u16>(MMIO::Utils::LowPart(&fifo.CPReadPointer), WMASK_LO_ALIGN_32BIT));
-  mmio->Register(
-      base | FIFO_READ_POINTER_HI,
-      IsOnThread() ? MMIO::DirectRead<u16>(MMIO::Utils::HighPart(&fifo.SafeCPReadPointer)) :
-                     MMIO::DirectRead<u16>(MMIO::Utils::HighPart(&fifo.CPReadPointer)),
-      IsOnThread() ?
-          MMIO::ComplexWrite<u16>([WMASK_HI_RESTRICT](u32, u16 val) {
+  mmio->Register(base | FIFO_READ_POINTER_HI,
+                 IsOnThread() ? MMIO::ComplexRead<u16>([](u32) {
+                   Fifo::SyncGPUForRegisterAccess();
+                   return ReadHigh(fifo.SafeCPReadPointer);
+                 }) :
+                                MMIO::ComplexRead<u16>([](u32) {
+                                  Fifo::SyncGPUForRegisterAccess();
+                                  return ReadHigh(fifo.CPReadPointer);
+                                }),
+                 IsOnThread() ? MMIO::ComplexWrite<u16>([WMASK_HI_RESTRICT](u32, u16 val) {
+                   Fifo::SyncGPUForRegisterAccess();
                    WriteHigh(fifo.CPReadPointer, val & WMASK_HI_RESTRICT);
                    fifo.SafeCPReadPointer = fifo.CPReadPointer;
                  }) :
-          MMIO::DirectWrite<u16>(MMIO::Utils::HighPart(&fifo.CPReadPointer), WMASK_HI_RESTRICT));
+                                MMIO::ComplexWrite<u16>([WMASK_HI_RESTRICT](u32, u16 val) {
+                                  Fifo::SyncGPUForRegisterAccess();
+                                  WriteHigh(fifo.CPReadPointer, val & WMASK_HI_RESTRICT);
+                                }));
 }
 
 void GatherPipeBursted()

Source/Core/VideoCommon/Fifo.cpp

@@ -580,6 +580,16 @@ static void SyncGPUCallback(u64 ticks, s64 cyclesLate)
     CoreTiming::ScheduleEvent(next, s_event_sync_gpu, next);
 }
 
+void SyncGPUForRegisterAccess()
+{
+  SyncGPU(SyncGPUReason::Other);
+
+  if (!SConfig::GetInstance().bCPUThread || s_use_deterministic_gpu_thread)
+    RunGpuOnCpu(GPU_TIME_SLOT_SIZE);
+  else if (SConfig::GetInstance().bSyncGPU)
+    WaitForGpuThread(GPU_TIME_SLOT_SIZE);
+}
+
 // Initialize GPU - CPU thread syncing, this gives us a deterministic way to start the GPU thread.
 void Prepare()
 {

Source/Core/VideoCommon/Fifo.h

@@ -33,6 +33,10 @@ enum class SyncGPUReason
 // In deterministic GPU thread mode this waits for the GPU to be done with pending work.
 void SyncGPU(SyncGPUReason reason, bool may_move_read_ptr = true);
 
+// In single core mode, this runs the GPU for a single slice.
+// In dual core mode, this synchronizes with the GPU thread.
+void SyncGPUForRegisterAccess();
+
 void PushFifoAuxBuffer(const void* ptr, size_t size);
 void* PopFifoAuxBuffer(size_t size);