Move the Vsync logic from VTRenderer into a VsyncSource

app/streaming/video/ffmpeg-renderers/pacer/displaylinkvsyncsource.cpp

@@ -0,0 +1,48 @@
+#include "displaylinkvsyncsource.h"
+
+DisplayLinkVsyncSource::DisplayLinkVsyncSource(Pacer* pacer)
+    : m_Pacer(pacer),
+      m_DisplayLink(nullptr)
+{
+
+}
+
+DisplayLinkVsyncSource::~DisplayLinkVsyncSource()
+{
+    if (m_DisplayLink != nullptr) {
+        CVDisplayLinkStop(m_DisplayLink);
+        CVDisplayLinkRelease(m_DisplayLink);
+    }
+}
+
+CVReturn
+DisplayLinkVsyncSource::displayLinkOutputCallback(
+    CVDisplayLinkRef,
+    const CVTimeStamp* /* now */,
+    const CVTimeStamp* /* vsyncTime */,
+    CVOptionFlags,
+    CVOptionFlags*,
+    void *displayLinkContext)
+{
+    auto me = reinterpret_cast<DisplayLinkVsyncSource*>(displayLinkContext);
+
+    // In my testing on macOS 10.13, this callback is invoked about 24 ms
+    // prior to the specified v-sync time (now - vsyncTime). Since this is
+    // greater than the standard v-sync interval (16 ms = 60 FPS), we will
+    // draw using the current host time, rather than the actual v-sync target
+    // time. Because the CVDisplayLink is in sync with the actual v-sync
+    // interval, even if many ms prior, we can safely use the current host time
+    // and get a consistent callback for each v-sync. This reduces video latency
+    // by at least 1 frame vs. rendering with the actual vsyncTime.
+    me->m_Pacer->vsyncCallback();
+
+    return kCVReturnSuccess;
+}
+
+bool DisplayLinkVsyncSource::initialize(SDL_Window*)
+{
+    CVDisplayLinkCreateWithActiveCGDisplays(&m_DisplayLink);
+    CVDisplayLinkSetOutputCallback(m_DisplayLink, displayLinkOutputCallback, this);
+    CVDisplayLinkStart(m_DisplayLink);
+    return true;
+}