SWDEV-305894 - Cooperative groups grid and multi grid sync support for gfx940+

Change-Id: I35d72f1cb50c3a96eee56a612b72d641852b145f


[ROCm/clr commit: 5d7f3f9f3c]

projects/clr/rocclr/platform/command.cpp

@@ -428,6 +428,9 @@ NDRangeKernelCommand::NDRangeKernelCommand(HostQueue& queue, const EventWaitList
   auto devKernel = const_cast<device::Kernel*>(kernel.getDeviceKernel(device));
   profilingInfo_.setCallback(devKernel->getProfilingCallback(
     queue.vdev()), devKernel->getWavesPerSH(queue.vdev()));
+  if (cooperativeGroups()) {
+    setNumWorkgroups();
+  }
   if (forceProfiling) {
     profilingInfo_.enabled_ = true;
     profilingInfo_.clear();

projects/clr/rocclr/platform/command.hpp

@@ -1050,7 +1050,8 @@ class NDRangeKernelCommand : public Command {
   uint32_t numGrids_;       //!< Total number of grids in multi GPU launch
   uint64_t prevGridSum_;    //!< A sum of previous grids to the current launch
   uint64_t allGridSum_;     //!< A sum of all grids in multi GPU launch
-  uint32_t firstDevice_;    //!< Device index of the first device in the grid
+  uint32_t firstDevice_;    //!< Device index of the first device in the gridc
+  uint32_t numWorkgroups_;  //!< Total number of workgroups in the current launch
 
  public:
   enum {
@@ -1118,9 +1119,22 @@ class NDRangeKernelCommand : public Command {
   //! Return the index of the first device in multi GPU launch
   uint64_t firstDevice() const { return firstDevice_; }
 
+  uint32_t numWorkgroups() const { return numWorkgroups_; }
+
   //! Set the local work size.
   void setLocalWorkSize(const NDRange& local) { sizes_.local() = local; }
 
+  //! Set the number of workgroups
+  void setNumWorkgroups() {
+    uint32_t numWorkgroups = 1;
+    for (uint i = 0; i < sizes().dimensions(); i++) {
+      if (sizes().local()[i] != 0) {
+        numWorkgroups *= (sizes().global()[i] / sizes().local()[i]);
+      }
+    }
+    numWorkgroups_ = numWorkgroups;
+  }
+
   int32_t captureAndValidate();
 };