[SYCL][PI][L0] Close and submit batch immediately when Queue is empty. (#3552)

This change causes executeCommandList to close a command list and
immediately execute it when it has determined that the last command
submitted to the queue has already been completed.

Updates code to only consider whether the Queue is empty when
doing dynamic batching.  Fixed size batching will not consider this
and give user the exact batching specified.  This change-set also
updates the dynamic batching unit test.

Author: kbsmith-intel

sycl/plugins/level_zero/pi_level_zero.cpp

@@ -843,9 +843,23 @@ pi_result _pi_queue::executeCommandList(ze_command_list_handle_t ZeCommandList,
                                         ze_fence_handle_t ZeFence,
                                         pi_event Event, bool IsBlocking,
                                         bool OKToBatchCommand) {
+  // If the current LastCommandEvent is the nullptr, then it means
+  // either that no command has ever been issued to the queue
+  // or it means that the LastCommandEvent has been signalled and
+  // therefore that this Queue is idle.
+  bool CurrentlyEmpty = this->LastCommandEvent == nullptr;
+
   this->LastCommandEvent = Event;
 
-  if (OKToBatchCommand && this->isBatchingAllowed()) {
+  // Batch if allowed to, but don't batch if we know there are no kernels
+  // from this queue that are currently executing.  This is intended to gets
+  // kernels started as soon as possible when there are no kernels from this
+  // queue awaiting execution, while allowing batching to occur when there
+  // are kernels already executing. Also, if we are using fixed size batching,
+  // as indicated by !UseDynamicBatching, then just ignore CurrentlyEmpty
+  // as we want to strictly follow the batching the user specified.
+  if (OKToBatchCommand && this->isBatchingAllowed() &&
+      (!UseDynamicBatching || !CurrentlyEmpty)) {
     if (this->ZeOpenCommandList != nullptr &&
         this->ZeOpenCommandList != ZeCommandList)
       die("executeCommandList: ZeOpenCommandList should be equal to"

sycl/test/on-device/plugins/level_zero_dynamic_batch_test.cpp

@@ -28,7 +28,7 @@
 // CKALL: Test Pass
 // CKALL: Test Pass
 // CKALL: Test Pass
-// CKDYN: Lowering QueueBatchSize to 3
+// CKDYN: Lowering QueueBatchSize to 2
 // CKDYN-NOT: Lowering QueueBatchSize
 // CKALL: Test Pass
 // CKALL: Test Pass