add test for target groupprivate usage

Author: skc7

mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td

@@ -2236,9 +2236,9 @@ def GroupprivateOp : OpenMP_Op<"groupprivate",
     each group having its own copy.
 
     This operation takes in the address of a symbol that represents the original
-    variable and returns the address of its groupprivate copy. All occurrences of
-    groupprivate variables in a parallel region should use the groupprivate copy
-    returned by this operation.
+    variable, optional DeviceTypeAttr and returns the address of its groupprivate copy.
+    All occurrences of groupprivate variables in a parallel region should
+    use the groupprivate copy returned by this operation.
 
     The `sym_addr` refers to the address of the symbol, which is a pointer to
     the original variable.

mlir/test/Dialect/OpenMP/ops.mlir

@@ -3392,3 +3392,5 @@ func.func @omp_groupprivate_device_type() {
 
   return
 }
+
+llvm.mlir.global internal @_QFgpEx() : i32

mlir/test/Target/LLVMIR/omptarget-groupprivate.mlir

@@ -0,0 +1,41 @@
+// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
+
+module attributes {omp.is_target_device = true, llvm.target_triple = "amdgcn-amd-amdhsa",
+                    dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<"dlti.alloca_memory_space", 5 : ui32>>} {
+  llvm.func @_QQmain() attributes {fir.bindc_name = "main"} {
+
+    %ga = llvm.mlir.addressof @global_a : !llvm.ptr
+    %map_a = omp.map.info var_ptr(%ga : !llvm.ptr, i32) map_clauses(tofrom) capture(ByCopy) -> !llvm.ptr {name = "i"}
+    omp.target map_entries(%map_a -> %arg1 : !llvm.ptr) {
+      %loaded = llvm.load %arg1 : !llvm.ptr -> i32
+
+      %any_addr = llvm.mlir.addressof @global_any : !llvm.ptr
+      %any_gp = omp.groupprivate %any_addr : !llvm.ptr, device_type(any) -> !llvm.ptr
+      llvm.store %loaded, %any_gp : i32, !llvm.ptr
+
+      %host_addr = llvm.mlir.addressof @global_host : !llvm.ptr
+      %host_gp = omp.groupprivate %host_addr : !llvm.ptr, device_type(host) -> !llvm.ptr
+      llvm.store %loaded, %host_gp : i32, !llvm.ptr
+
+      %nohost_addr = llvm.mlir.addressof @global_nohost : !llvm.ptr
+      %nohost_gp = omp.groupprivate %nohost_addr : !llvm.ptr, device_type(nohost) -> !llvm.ptr
+      llvm.store %loaded, %nohost_gp : i32, !llvm.ptr
+
+      omp.terminator
+    }
+    llvm.return
+  }
+  llvm.mlir.global internal @global_a() : i32
+  llvm.mlir.global internal @global_any() : i32
+  llvm.mlir.global internal @global_host() : i32
+  llvm.mlir.global internal @global_nohost() : i32
+}
+
+// CHECK: define {{.*}} amdgpu_kernel void @__omp_offloading_{{.*}}_{{.*}}__QQmain_{{.*}}(ptr %{{.*}}, ptr %{{.*}}) #{{[0-9]+}} {
+// CHECK-LABEL:  omp.target:
+// CHECK-NEXT :    %[[LOAD:.*]] = load i32, ptr %3, align 4
+// CHECK-NEXT :    %[[ALLOC_any:.*]] = call ptr @__kmpc_alloc_shared(i64 4)
+// CHECK-NEXT :    store i32 %[[LOAD]], ptr %[[ALLOC_any]], align 4
+// CHECK-NEXT :    store i32 %[[LOAD]], ptr @global_host, align 4
+// CHECK-NEXT :    %[[ALLOC_NOHOST:.*]] = call ptr @__kmpc_alloc_shared(i64 4)
+// CHECK-NEXT :    store i32 %[[LOAD]], ptr %[[ALLOC_NOHOST]], align 4