Switching vsync to custom thread.

src/xenia/cpu/processor.cc

@@ -144,6 +144,7 @@ bool Processor::Setup() {
   backend_ = std::move(backend);
   frontend_ = std::move(frontend);
 
+  // DEPRECATED: will be removed.
   interrupt_thread_state_ =
       new ThreadState(this, 0, ThreadStackType::kKernelStack, 0, 128 * 1024, 0);
   interrupt_thread_state_->set_name("Interrupt");
@@ -373,17 +374,13 @@ void Processor::LowerIrql(Irql old_value) {
                       reinterpret_cast<volatile uint32_t*>(&irql_));
 }
 
-uint64_t Processor::ExecuteInterrupt(uint32_t cpu, uint32_t address,
-                                     uint64_t args[], size_t arg_count) {
+uint64_t Processor::ExecuteInterrupt(uint32_t address, uint64_t args[],
+                                     size_t arg_count) {
   SCOPE_profile_cpu_f("cpu");
 
   // Acquire lock on interrupt thread (we can only dispatch one at a time).
   std::lock_guard<xe::mutex> lock(interrupt_thread_lock_);
 
-  // Set 0x10C(r13) to the current CPU ID.
-  xe::store_and_swap<uint8_t>(
-      memory_->TranslateVirtual(interrupt_thread_block_ + 0x10C), cpu);
-
   // Execute interrupt.
   uint64_t result = Execute(interrupt_thread_state_, address, args, arg_count);
 

src/xenia/cpu/processor.h

@@ -75,7 +75,8 @@ class Processor {
   Irql RaiseIrql(Irql new_value);
   void LowerIrql(Irql old_value);
 
-  uint64_t ExecuteInterrupt(uint32_t cpu, uint32_t address, uint64_t args[],
+  // DEPRECATED: will be removed.
+  uint64_t ExecuteInterrupt(uint32_t address, uint64_t args[],
                             size_t arg_count);
 
  private:

src/xenia/gpu/gl4/command_processor.cc

@@ -164,7 +164,7 @@ void CommandProcessor::EndTracing() {
 
 void CommandProcessor::CallInThread(std::function<void()> fn) {
   if (pending_fns_.empty() &&
-      worker_thread_.get() == kernel::XThread::GetCurrentThread()) {
+      kernel::XThread::IsInThread(worker_thread_.get())) {
     fn();
   } else {
     pending_fns_.push(std::move(fn));

src/xenia/gpu/gl4/gl4_graphics_system.cc

@@ -12,6 +12,7 @@
 #include "xenia/base/logging.h"
 #include "xenia/base/threading.h"
 #include "xenia/cpu/processor.h"
+#include "xenia/emulator.h"
 #include "xenia/gpu/gl4/gl4_gpu-private.h"
 #include "xenia/gpu/gl4/gl4_profiler_display.h"
 #include "xenia/gpu/gpu-private.h"
@@ -24,7 +25,7 @@ namespace gl4 {
 extern "C" GLEWContext* glewGetContext();
 
 GL4GraphicsSystem::GL4GraphicsSystem(Emulator* emulator)
-    : GraphicsSystem(emulator), timer_queue_(nullptr), vsync_timer_(nullptr) {}
+    : GraphicsSystem(emulator), worker_running_(false) {}
 
 GL4GraphicsSystem::~GL4GraphicsSystem() = default;
 
@@ -80,15 +81,26 @@ X_STATUS GL4GraphicsSystem::Setup(cpu::Processor* processor,
       reinterpret_cast<cpu::MMIOWriteCallback>(MMIOWriteRegisterThunk));
 
   // 60hz vsync timer.
-  DWORD timer_period = 16;
-  if (!FLAGS_vsync) {
-    // DANGER a value too low here will lead to starvation!
-    timer_period = 4;
-  }
-  timer_queue_ = CreateTimerQueue();
-  CreateTimerQueueTimer(&vsync_timer_, timer_queue_,
-                        (WAITORTIMERCALLBACK)VsyncCallbackThunk, this, 16,
-                        timer_period, WT_EXECUTEINPERSISTENTTHREAD);
+  worker_running_ = true;
+  worker_thread_ =
+      kernel::object_ref<kernel::XHostThread>(new kernel::XHostThread(
+          emulator()->kernel_state(), 128 * 1024, 0, [this]() {
+            uint64_t vsync_duration = FLAGS_vsync ? 16 : 1;
+            uint64_t last_frame_time = xe::threading::ticks();
+            while (worker_running_) {
+              uint64_t current_time = xe::threading::ticks();
+              uint64_t elapsed = (current_time - last_frame_time) /
+                                 (xe::threading::ticks_per_second() / 1000);
+              if (elapsed >= vsync_duration) {
+                MarkVblank();
+                last_frame_time = current_time;
+              }
+              Sleep(1);
+            }
+            return 0;
+          }));
+  worker_thread_->set_name("GL4 Vsync");
+  worker_thread_->Create();
 
   if (FLAGS_trace_gpu_stream) {
     BeginTracing();
@@ -100,8 +112,9 @@ X_STATUS GL4GraphicsSystem::Setup(cpu::Processor* processor,
 void GL4GraphicsSystem::Shutdown() {
   EndTracing();
 
-  DeleteTimerQueueTimer(timer_queue_, vsync_timer_, nullptr);
-  DeleteTimerQueue(timer_queue_);
+  worker_running_ = false;
+  worker_thread_->Wait(0, 0, 0, nullptr);
+  worker_thread_.reset();
 
   command_processor_->Shutdown();
 

src/xenia/gpu/gl4/gl4_graphics_system.h

@@ -16,6 +16,7 @@
 #include "xenia/gpu/gl4/wgl_control.h"
 #include "xenia/gpu/graphics_system.h"
 #include "xenia/gpu/register_file.h"
+#include "xenia/kernel/objects/xthread.h"
 
 namespace xe {
 namespace gpu {
@@ -60,16 +61,13 @@ class GL4GraphicsSystem : public GraphicsSystem {
                                      uint64_t value) {
     gs->WriteRegister(addr, value);
   }
-  static void __stdcall VsyncCallbackThunk(GL4GraphicsSystem* gs, BOOLEAN) {
-    gs->MarkVblank();
-  }
 
   RegisterFile register_file_;
   std::unique_ptr<CommandProcessor> command_processor_;
   std::unique_ptr<WGLControl> control_;
 
-  HANDLE timer_queue_;
-  HANDLE vsync_timer_;
+  std::atomic<bool> worker_running_;
+  kernel::object_ref<kernel::XHostThread> worker_thread_;
 };
 
 }  // namespace gl4

src/xenia/gpu/graphics_system.cc

@@ -13,6 +13,7 @@
 #include "xenia/base/math.h"
 #include "xenia/cpu/processor.h"
 #include "xenia/gpu/gpu-private.h"
+#include "xenia/kernel/objects/xthread.h"
 
 namespace xe {
 namespace gpu {
@@ -53,18 +54,21 @@ void GraphicsSystem::DispatchInterruptCallback(uint32_t source, uint32_t cpu) {
     return;
   }
 
+  auto thread = kernel::XThread::GetCurrentThread();
+  assert_not_null(thread);
+
   // Pick a CPU, if needed. We're going to guess 2. Because.
   if (cpu == 0xFFFFFFFF) {
     cpu = 2;
   }
+  thread->SetActiveCpu(cpu);
 
   // XELOGGPU("Dispatching GPU interrupt at %.8X w/ mode %d on cpu %d",
   //         interrupt_callback_, source, cpu);
 
-  // NOTE: we may be executing in some random thread.
   uint64_t args[] = {source, interrupt_callback_data_};
-  processor_->ExecuteInterrupt(cpu, interrupt_callback_, args,
-                               xe::countof(args));
+  processor_->Execute(thread->thread_state(), interrupt_callback_, args,
+                      xe::countof(args));
 }
 
 }  // namespace gpu

src/xenia/kernel/objects/xthread.cc

@@ -35,7 +35,6 @@ using namespace xe::cpu;
 uint32_t next_xthread_id = 0;
 thread_local XThread* current_thread_tls = nullptr;
 xe::mutex critical_region_;
-XThread* shared_kernel_thread_ = 0;
 
 XThread::XThread(KernelState* kernel_state, uint32_t stack_size,
                  uint32_t xapi_thread_startup, uint32_t start_address,
@@ -97,18 +96,14 @@ XThread::~XThread() {
   }
 }
 
+bool XThread::IsInThread(XThread* other) {
+  return current_thread_tls == other;
+}
+
 XThread* XThread::GetCurrentThread() {
   XThread* thread = current_thread_tls;
   if (!thread) {
-    // Assume this is some shared interrupt thread/etc.
-    XThread::EnterCriticalRegion();
-    thread = shared_kernel_thread_;
-    if (!thread) {
-      thread = new XThread(KernelState::shared(), 256 * 1024, 0, 0, 0, 0);
-      shared_kernel_thread_ = thread;
-      current_thread_tls = thread;
-    }
-    XThread::LeaveCriticalRegion();
+    assert_always("Attempting to use kernel stuff from a non-kernel thread");
   }
   return thread;
 }
@@ -496,7 +491,7 @@ void XThread::DeliverAPCs(void* data) {
         apc_address, thread->scratch_address_ + 0, thread->scratch_address_ + 4,
         thread->scratch_address_ + 8, thread->scratch_address_ + 12,
     };
-    processor->ExecuteInterrupt(0, kernel_routine, kernel_args,
+    processor->ExecuteInterrupt(kernel_routine, kernel_args,
                                 xe::countof(kernel_args));
     normal_routine = xe::load_and_swap<uint32_t>(scratch_ptr + 0);
     normal_context = xe::load_and_swap<uint32_t>(scratch_ptr + 4);
@@ -509,7 +504,7 @@ void XThread::DeliverAPCs(void* data) {
       thread->UnlockApc();
       // normal_routine(normal_context, system_arg1, system_arg2)
       uint64_t normal_args[] = {normal_context, system_arg1, system_arg2};
-      processor->ExecuteInterrupt(0, normal_routine, normal_args,
+      processor->ExecuteInterrupt(normal_routine, normal_args,
                                   xe::countof(normal_args));
       thread->LockApc();
     }
@@ -534,7 +529,7 @@ void XThread::RundownAPCs() {
     if (rundown_routine) {
       // rundown_routine(apc)
       uint64_t args[] = {apc_address};
-      kernel_state()->processor()->ExecuteInterrupt(0, rundown_routine, args,
+      kernel_state()->processor()->ExecuteInterrupt(rundown_routine, args,
                                                     xe::countof(args));
     }
   }
@@ -582,6 +577,11 @@ void XThread::SetAffinity(uint32_t affinity) {
   }
 }
 
+void XThread::SetActiveCpu(uint32_t cpu_index) {
+  uint8_t* pcr = memory()->TranslateVirtual(pcr_address_);
+  xe::store_and_swap<uint8_t>(pcr + 0x10C, cpu_index);
+}
+
 X_STATUS XThread::Resume(uint32_t* out_suspend_count) {
   DWORD result = ResumeThread(thread_handle_);
   if (result >= 0) {

src/xenia/kernel/objects/xthread.h

@@ -31,6 +31,7 @@ class XThread : public XObject {
           uint32_t start_context, uint32_t creation_flags);
   virtual ~XThread();
 
+  static bool IsInThread(XThread* other);
   static XThread* GetCurrentThread();
   static uint32_t GetCurrentThreadHandle();
   static uint32_t GetCurrentThreadId(const uint8_t* pcr);
@@ -61,6 +62,7 @@ class XThread : public XObject {
   int32_t QueryPriority();
   void SetPriority(int32_t increment);
   void SetAffinity(uint32_t affinity);
+  void SetActiveCpu(uint32_t cpu_index);
 
   X_STATUS Resume(uint32_t* out_suspend_count = nullptr);
   X_STATUS Suspend(uint32_t* out_suspend_count);

src/xenia/kernel/xboxkrnl_threading.cc

@@ -990,6 +990,7 @@ SHIM_CALL KfAcquireSpinLock_shim(PPCContext* ppc_state, KernelState* state) {
   while (!xe::atomic_cas(0, 1, lock)) {
     // Spin!
     // TODO(benvanik): error on deadlock?
+    YieldProcessor();
   }
 
   // Raise IRQL to DISPATCH.