Add dynamic shared memory arg for olLaunchKernel

Author: callumfare

offload/liboffload/API/Kernel.td

@@ -34,7 +34,8 @@ def : Struct {
         StructMember<"size_t", "NumGroupsZ", "Number of work groups on the Z dimension">,
         StructMember<"size_t", "GroupSizeX", "Size of a work group on the X dimension.">,
         StructMember<"size_t", "GroupSizeY", "Size of a work group on the Y dimension.">,
-        StructMember<"size_t", "GroupSizeZ", "Size of a work group on the Z dimension.">
+        StructMember<"size_t", "GroupSizeZ", "Size of a work group on the Z dimension.">,
+        StructMember<"size_t", "DynSharedMemory", "Size of dynamic shared memory in bytes.">
     ];
 }
 

offload/liboffload/include/generated/OffloadAPI.h

@@ -736,13 +736,14 @@ OL_APIEXPORT ol_result_t OL_APICALL olGetKernel(
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Size-related arguments for a kernel launch.
 typedef struct ol_kernel_launch_size_args_t {
-  size_t Dimensions; /// Number of work dimensions
-  size_t NumGroupsX; /// Number of work groups on the X dimension
-  size_t NumGroupsY; /// Number of work groups on the Y dimension
-  size_t NumGroupsZ; /// Number of work groups on the Z dimension
-  size_t GroupSizeX; /// Size of a work group on the X dimension.
-  size_t GroupSizeY; /// Size of a work group on the Y dimension.
-  size_t GroupSizeZ; /// Size of a work group on the Z dimension.
+  size_t Dimensions;      /// Number of work dimensions
+  size_t NumGroupsX;      /// Number of work groups on the X dimension
+  size_t NumGroupsY;      /// Number of work groups on the Y dimension
+  size_t NumGroupsZ;      /// Number of work groups on the Z dimension
+  size_t GroupSizeX;      /// Size of a work group on the X dimension.
+  size_t GroupSizeY;      /// Size of a work group on the Y dimension.
+  size_t GroupSizeZ;      /// Size of a work group on the Z dimension.
+  size_t DynSharedMemory; /// Size of dynamic shared memory in bytes.
 } ol_kernel_launch_size_args_t;
 
 ///////////////////////////////////////////////////////////////////////////////

offload/liboffload/include/generated/OffloadPrint.hpp

@@ -362,6 +362,9 @@ operator<<(llvm::raw_ostream &os,
   os << ", ";
   os << ".GroupSizeZ = ";
   os << params.GroupSizeZ;
+  os << ", ";
+  os << ".DynSharedMemory = ";
+  os << params.DynSharedMemory;
   os << "}";
   return os;
 }

offload/liboffload/src/OffloadImpl.cpp

@@ -524,6 +524,7 @@ olLaunchKernel_impl(ol_queue_handle_t Queue, ol_device_handle_t Device,
   LaunchArgs.ThreadLimit[0] = LaunchSizeArgs->GroupSizeX;
   LaunchArgs.ThreadLimit[1] = LaunchSizeArgs->GroupSizeY;
   LaunchArgs.ThreadLimit[2] = LaunchSizeArgs->GroupSizeZ;
+  LaunchArgs.DynCGroupMem = LaunchSizeArgs->DynSharedMemory;
 
   KernelLaunchParamsTy Params;
   Params.Data = const_cast<void *>(ArgumentsData);

offload/unittests/OffloadAPI/kernel/olLaunchKernel.cpp

@@ -26,6 +26,8 @@ struct olLaunchKernelTest : OffloadQueueTest {
     LaunchArgs.NumGroupsX = 1;
     LaunchArgs.NumGroupsY = 1;
     LaunchArgs.NumGroupsZ = 1;
+
+    LaunchArgs.DynSharedMemory = 0;
   }
 
   void TearDown() override {