Dangerous --vsync=false flag.

src/xenia/gpu/gl4/command_processor.cc

@@ -123,22 +123,23 @@ void CommandProcessor::WorkerMain() {
 
   while (worker_running_) {
     uint32_t write_ptr_index = write_ptr_index_.load();
-    while (write_ptr_index == 0xBAADF00D ||
-           read_ptr_index_ == write_ptr_index) {
+    if (write_ptr_index == 0xBAADF00D || read_ptr_index_ == write_ptr_index) {
       SCOPE_profile_cpu_i("gpu", "xe::gpu::gl4::CommandProcessor::Stall");
-      // Check if the pointer has moved.
-      // We wait a short bit here to yield time. Since we are also running the
-      // main window display we don't want to pause too long, though.
-      // YieldProcessor();
-      PrepareForWait();
-      const int wait_time_ms = 5;
-      if (WaitForSingleObject(write_ptr_index_event_, wait_time_ms) ==
-          WAIT_TIMEOUT) {
-        ReturnFromWait();
+      // We've run out of commands to execute.
+      // We spin here waiting for new ones, as the overhead of waiting on our
+      // event is too high.
+      //PrepareForWait();
+      do {
+        // TODO(benvanik): if we go longer than Nms, switch to waiting?
+        // It'll keep us from burning power.
+        // const int wait_time_ms = 5;
+        // WaitForSingleObject(write_ptr_index_event_, wait_time_ms);
+        SwitchToThread();
+        MemoryBarrier();
         write_ptr_index = write_ptr_index_.load();
-        continue;
-      }
-      ReturnFromWait();
+      } while (write_ptr_index == 0xBAADF00D ||
+               read_ptr_index_ == write_ptr_index);
+      //ReturnFromWait();
     }
     assert_true(read_ptr_index_ != write_ptr_index);
 
@@ -925,7 +926,13 @@ bool CommandProcessor::ExecutePacketType3_WAIT_REG_MEM(RingbufferReader* reader,
       // Wait.
       if (wait >= 0x100) {
         PrepareForWait();
-        Sleep(wait / 0x100);
+        if (!FLAGS_vsync) {
+          // User wants it fast and dangerous.
+          SwitchToThread();
+        } else {
+          Sleep(wait / 0x100);
+        }
+        MemoryBarrier();
         ReturnFromWait();
       } else {
         SwitchToThread();

src/xenia/gpu/gl4/gl4_graphics_system.cc

@@ -77,10 +77,15 @@ X_STATUS GL4GraphicsSystem::Setup() {
       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, 16,
-                        WT_EXECUTEINTIMERTHREAD);
+                        (WAITORTIMERCALLBACK)VsyncCallbackThunk, this, 16,
+                        timer_period, WT_EXECUTEINPERSISTENTTHREAD);
 
   return X_STATUS_SUCCESS;
 }
@@ -138,9 +143,6 @@ void GL4GraphicsSystem::SwapHandler(const SwapParameters& swap_params) {
                            control_->width(), control_->height(),
                            GL_COLOR_BUFFER_BIT, GL_LINEAR);
   });
-
-  // Roll over vblank.
-  MarkVblank();
 }
 
 uint64_t GL4GraphicsSystem::ReadRegister(uint64_t addr) {

src/xenia/gpu/gpu-private.h

@@ -17,4 +17,6 @@ DECLARE_string(gpu);
 DECLARE_bool(trace_ring_buffer);
 DECLARE_string(dump_shaders);
 
+DECLARE_bool(vsync);
+
 #endif  // XENIA_GPU_PRIVATE_H_

src/xenia/gpu/gpu.cc

@@ -19,6 +19,8 @@ DEFINE_bool(trace_ring_buffer, false, "Trace GPU ring buffer packets.");
 DEFINE_string(dump_shaders, "",
               "Path to write GPU shaders to as they are compiled.");
 
+DEFINE_bool(vsync, true, "Enable VSYNC.");
+
 namespace xe {
 namespace gpu {
 

src/xenia/profiling.cc

@@ -15,6 +15,7 @@
 #define MICROPROFILEUI_IMPL 1
 #define MICROPROFILE_PER_THREAD_BUFFER_SIZE (1024 * 1024 * 10)
 #define MICROPROFILE_USE_THREAD_NAME_CALLBACK 1
+#define MICROPROFILE_WEBSERVER_MAXFRAMES 10
 #define MICROPROFILE_PRINTF PLOGI
 #define MICROPROFILE_WEBSERVER 1
 #define MICROPROFILE_DEBUG 0