Add direct dispatch simple hack for testing

The hack dosn't really track the commands status. It may be not
necessary for HIP, but will cause early resource release.

Change-Id: I791ad36dd8abd3b6b3d2c9b16a210a555c08ca64


[ROCm/clr commit: 532f0ae951]

projects/clr/rocclr/platform/command.cpp

@@ -258,8 +258,22 @@ void Command::enqueue() {
   }
 
   ClPrint(LOG_DEBUG, LOG_CMD, "command is enqueued: %p", this);
-  queue_->append(*this);
-  queue_->flush();
+  if (AMD_DIRECT_DISPATCH) {
+    if (type() == CL_COMMAND_MARKER || type() == 0) {
+      setStatus(CL_SUBMITTED);
+      queue_->vdev()->flush();
+      retain();
+      setStatus(CL_COMPLETE);
+    } else {
+      setStatus(CL_SUBMITTED);
+      submit(*queue_->vdev());
+      retain();
+      setStatus(CL_COMPLETE);
+    }
+  } else {
+    queue_->append(*this);
+    queue_->flush();
+  }
   if ((queue_->device().settings().waitCommand_ && (type_ != 0)) ||
       ((commandWaitBits_ & 0x2) != 0)) {
     awaitCompletion();

projects/clr/rocclr/utils/flags.hpp

@@ -231,6 +231,8 @@ release(uint, HIP_HOST_COHERENT, 0,                                           \
 release(uint, AMD_OPT_FLUSH, 1,                                               \
         "Kernel flush option , 0x0 = Use system-scope fence operations."      \
         "0x1 = Use device-scope fence operations when possible.")             \
+release(bool, AMD_DIRECT_DISPATCH, false,                                     \
+        "Enable direct kernel dispatch.")                                     \
 release(uint, HIP_HIDDEN_FREE_MEM, 0,                                         \
         "Reserve free mem reporting in Mb"                                    \
         "0 = Disable")                                                        \