pipeline: Use a high-precision timer for Thread::sleep() on Win32

This is mainly used for clock mode M_limited, which is currently suffering from terrible imprecision on Windows.  This fixes that, making it possible to lower the value of sleep-precision down to 1ms.

panda/src/pipeline/threadWin32Impl.I

@@ -63,14 +63,6 @@ is_simple_threads() {
   return false;
 }
 
-/**
- *
- */
-INLINE void ThreadWin32Impl::
-sleep(double seconds) {
-  Sleep((int)(seconds * 1000));
-}
-
 /**
  *
  */

panda/src/pipeline/threadWin32Impl.cxx

@@ -78,6 +78,11 @@ ThreadWin32Impl::
   }
 
   CloseHandle(_thread);
+
+  if (_timer != nullptr) {
+    CloseHandle(_timer);
+    _timer = nullptr;
+  }
 }
 
 /**
@@ -200,6 +205,27 @@ bind_thread(Thread *thread) {
   return thread;
 }
 
+
+/**
+ *
+ */
+void ThreadWin32Impl::
+sleep(double seconds) {
+  Thread *thread = get_current_thread();
+  ThreadWin32Impl *self = &thread->_impl;
+
+  HANDLE timer = self->_timer;
+  if (timer == nullptr) {
+    timer = CreateWaitableTimerExW(nullptr, nullptr, CREATE_WAITABLE_TIMER_MANUAL_RESET | CREATE_WAITABLE_TIMER_HIGH_RESOLUTION, TIMER_ALL_ACCESS);
+    self->_timer = timer;
+  }
+
+  LARGE_INTEGER ft;
+  ft.QuadPart = seconds * -10000000LL;
+  SetWaitableTimer(timer, &ft, 0, nullptr, nullptr, 0);
+  WaitForSingleObject(timer, INFINITE);
+}
+
 /**
  * Returns the number of context switches that occurred on the current thread.
  * The first number is the total number of context switches reported by the OS,

panda/src/pipeline/threadWin32Impl.h

@@ -48,7 +48,7 @@ public:
   INLINE static bool is_threading_supported();
   INLINE static bool is_true_threads();
   INLINE static bool is_simple_threads();
-  INLINE static void sleep(double seconds);
+  static void sleep(double seconds);
   INLINE static void yield();
   INLINE static void consider_yield();
 
@@ -72,6 +72,7 @@ private:
   bool _joinable;
   Status _status;
   HANDLE _profiling;
+  HANDLE _timer = nullptr;
 };
 
 #include "threadWin32Impl.I"