Allow applications to reliably calculate decode unit queue delay

src/Limelight.h

@@ -136,11 +136,17 @@ typedef struct _DECODE_UNIT {
     // Frame type
     int frameType;
 
-    // Receive time of first buffer. This value uses an implementation-defined epoch.
-    // To compute actual latency values, use LiGetMillis() to get a timestamp that
-    // shares the same epoch as this value.
+    // Receive time of first buffer. This value uses an implementation-defined epoch,
+    // but the same epoch as enqueueTimeMs and LiGetMillis().
     uint64_t receiveTimeMs;
 
+    // Time the frame was fully assembled and queued for the video decoder to process.
+    // This is also approximately the same time as the final packet was received, so
+    // enqueueTimeMs - receiveTimeMs is the time taken to receive the frame. At the
+    // time the decode unit is passed to submitDecodeUnit(), the total queue delay
+    // can be calculated by LiGetMillis() - enqueueTimeMs.
+    uint64_t enqueueTimeMs;
+
     // Presentation time in milliseconds with the epoch at the first captured frame.
     // This can be used to aid frame pacing or to drop old frames that were queued too
     // long prior to display.

src/VideoDepacketizer.c

@@ -253,6 +253,7 @@ static void reassembleFrame(int frameNumber) {
             qdu->decodeUnit.frameNumber = frameNumber;
             qdu->decodeUnit.receiveTimeMs = firstPacketReceiveTime;
             qdu->decodeUnit.presentationTimeMs = firstPacketPresentationTime;
+            qdu->decodeUnit.enqueueTimeMs = LiGetMillis();
 
             // IDR frames will have leading CSD buffers
             if (nalChainHead->bufferType != BUFFER_TYPE_PICDATA) {