Adding AnyOrder Flag

Change-Id: I6baaef42b98adfbc8cf2605e175ec007e008045f


[ROCm/clr commit: 48690f29e9]

projects/clr/rocclr/device/rocm/rocvirtual.cpp

@@ -2305,9 +2305,16 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
     dispatchPacket.group_segment_size = ldsUsage + sharedMemBytes;
     dispatchPacket.private_segment_size = devKernel->workGroupInfo()->privateMemSize_;
 
+    // Pass the header accordingly
+    auto aqlHeaderWithOrder = aqlHeader_;
+    if (vcmd != nullptr && vcmd->getAnyOrderLaunchFlag()) {
+      constexpr uint32_t kAqlHeaderMask = ~(1 << HSA_PACKET_HEADER_BARRIER);
+      aqlHeaderWithOrder &= kAqlHeaderMask;
+    }
+
     // Dispatch the packet
     if (!dispatchAqlPacket(
-            &dispatchPacket, aqlHeader_,
+            &dispatchPacket, aqlHeaderWithOrder,
             (sizes.dimensions() << HSA_KERNEL_DISPATCH_PACKET_SETUP_DIMENSIONS),
             GPU_FLUSH_ON_EXECUTION)) {
       return false;
@@ -2406,7 +2413,7 @@ void VirtualGPU::submitKernel(amd::NDRangeKernelCommand& vcmd) {
 
     // Submit kernel to HW
     if (!submitKernelInternal(vcmd.sizes(), vcmd.kernel(), vcmd.parameters(),
-      static_cast<void*>(as_cl(&vcmd.event())), vcmd.sharedMemBytes())) {
+      static_cast<void*>(as_cl(&vcmd.event())), vcmd.sharedMemBytes(), &vcmd)) {
       LogError("AQL dispatch failed!");
       vcmd.setStatus(CL_INVALID_OPERATION);
     }

projects/clr/rocclr/platform/command.hpp

@@ -794,6 +794,7 @@ class NDRangeKernelCommand : public Command {
   enum {
     CooperativeGroups = 0x01,
     CooperativeMultiDeviceGroups = 0x02,
+    AnyOrderLaunch = 0x04,
   };
 
   //! Construct an ExecuteKernel command
@@ -827,6 +828,9 @@ class NDRangeKernelCommand : public Command {
     return (extraParam_ & CooperativeMultiDeviceGroups) ? true : false;
   }
 
+  //! Returns extra Param, set when using anyorder launch
+  bool getAnyOrderLaunchFlag() const { return (extraParam_ & AnyOrderLaunch) ? true : false; }
+
   //! Return the current grid ID for multidevice launch
   uint32_t gridId() const { return gridId_; }