Optimize waiting scheme in user events scenarios.

- do not wait on map/unmap if zero copy resource.
- do not wait in marker commands at all.

Change-Id: I74cdd8320d13602bf662eed412ed2fcad1504989
Signed-off-by: Mrozek, Michal <michal.mrozek@intel.com>

runtime/helpers/task_information.cpp

@@ -76,9 +76,8 @@ CompletionStamp &CommandMapUnmap::submit(uint32_t taskLevel, bool terminated) {
                                     dispatchFlags,
                                     cmdQ.getDevice());
 
-    cmdQ.waitUntilComplete(completionStamp.taskCount, completionStamp.flushStamp, false);
-
     if (!memObj.isMemObjZeroCopy()) {
+        cmdQ.waitUntilComplete(completionStamp.taskCount, completionStamp.flushStamp, false);
         if (op == MAP) {
             memObj.transferDataToHostPtr(copySize, copyOffset);
         } else if (!readOnly) {
@@ -283,8 +282,6 @@ CompletionStamp &CommandMarker::submit(uint32_t taskLevel, bool terminated) {
                                     dispatchFlags,
                                     cmdQ.getDevice());
 
-    cmdQ.waitUntilComplete(completionStamp.taskCount, completionStamp.flushStamp, false);
-
     return completionStamp;
 }
 } // namespace NEO

unit_tests/command_queue/command_queue_hw_tests.cpp

@@ -177,7 +177,7 @@ HWTEST_F(CommandQueueHwTest, addMapUnmapToWaitlistEventsDoesntAddDependenciesInt
     buffer->decRefInternal();
 }
 
-HWTEST_F(CommandQueueHwTest, givenMapCommandWhenZeroStateCommandIsSubmittedThenTaskCountIsBeingWaited) {
+HWTEST_F(CommandQueueHwTest, givenMapCommandWhenZeroStateCommandIsSubmittedThenTaskCountIsNotBeingWaited) {
     auto buffer = new MockBuffer;
     CommandQueueHw<FamilyType> *pHwQ = reinterpret_cast<CommandQueueHw<FamilyType> *>(pCmdQ);
 
@@ -194,6 +194,28 @@ HWTEST_F(CommandQueueHwTest, givenMapCommandWhenZeroStateCommandIsSubmittedThenT
     EXPECT_NE(nullptr, pHwQ->virtualEvent);
     pHwQ->virtualEvent->setStatus(CL_COMPLETE);
 
+    EXPECT_EQ(std::numeric_limits<uint32_t>::max(), pHwQ->latestTaskCountWaited);
+    buffer->decRefInternal();
+}
+
+HWTEST_F(CommandQueueHwTest, givenMapCommandWhenZeroStateCommandIsSubmittedOnNonZeroCopyBufferThenTaskCountIsBeingWaited) {
+    auto buffer = new MockBuffer;
+    buffer->isZeroCopy = false;
+    CommandQueueHw<FamilyType> *pHwQ = reinterpret_cast<CommandQueueHw<FamilyType> *>(pCmdQ);
+
+    MockEventBuilder eventBuilder;
+    MemObjSizeArray size = {{1, 1, 1}};
+    MemObjOffsetArray offset = {{0, 0, 0}};
+    pHwQ->enqueueBlockedMapUnmapOperation(nullptr,
+                                          0,
+                                          MAP,
+                                          buffer,
+                                          size, offset, false,
+                                          eventBuilder);
+
+    EXPECT_NE(nullptr, pHwQ->virtualEvent);
+    pHwQ->virtualEvent->setStatus(CL_COMPLETE);
+
     EXPECT_EQ(1u, pHwQ->latestTaskCountWaited);
     buffer->decRefInternal();
 }

unit_tests/mocks/mock_buffer.h

@@ -29,6 +29,7 @@ class MockBuffer : public MockBufferStorage, public Buffer {
   public:
     using Buffer::magic;
     using Buffer::offset;
+    using MemObj::isZeroCopy;
     using MockBufferStorage::device;
 
     MockBuffer(GraphicsAllocation &alloc)