add llvm translation tests for groupprivate

Author: skc7

mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp

@@ -6137,13 +6137,13 @@ convertOmpGroupprivate(Operation &opInst, llvm::IRBuilderBase &builder,
 
   if (failed(checkImplementationStatus(opInst)))
     return failure();
-  
+
   bool isTargetDevice = ompBuilder->Config.isTargetDevice();
   auto deviceType = groupprivateOp.getDeviceType();
-  
+
   // skip allocation based on device_type
   bool shouldAllocate = true;
-  if (deviceType.has_value()) {    
+  if (deviceType.has_value()) {
     switch (*deviceType) {
     case mlir::omp::DeclareTargetDeviceType::host:
       // Only allocate on host
@@ -6187,10 +6187,8 @@ convertOmpGroupprivate(Operation &opInst, llvm::IRBuilderBase &builder,
     // Call runtime to allocate shared memory for this group
     llvm::Value *groupPrivatePtr =
         builder.CreateCall(allocSharedFn, {builder.getInt64(typeSize)});
-    resultPtr =
-        builder.CreateBitCast(groupPrivatePtr, globalValue->getType());
-  }
-  else {
+    resultPtr = builder.CreateBitCast(groupPrivatePtr, globalValue->getType());
+  } else {
     // Use original global address when not allocating group-private storage
     resultPtr = moduleTranslation.lookupValue(symAddr);
     if (!resultPtr) {

mlir/test/Target/LLVMIR/openmp-llvm.mlir

@@ -3452,36 +3452,68 @@ llvm.func @nested_task_with_deps() {
 
 // -----
 
-// CHECK: @_QFsubEx = internal global i32 undef
+module attributes {omp.is_target_device = false} {
+llvm.mlir.global internal @any() : i32
+llvm.mlir.global internal @host() : i32
+llvm.mlir.global internal @nohost() : i32
+llvm.func @omp_groupprivate_host() {
+  %0 = llvm.mlir.constant(1 : i32) : i32
+  %1 = llvm.mlir.addressof @any : !llvm.ptr
+  %2 = omp.groupprivate %1 : !llvm.ptr, device_type(any) -> !llvm.ptr
+  llvm.store %0, %2 : i32, !llvm.ptr
+
+  %3 = llvm.mlir.addressof @host : !llvm.ptr
+  %4 = omp.groupprivate %3 : !llvm.ptr, device_type(host) -> !llvm.ptr
+  llvm.store %0, %4 : i32, !llvm.ptr
+
+  %5 = llvm.mlir.addressof @nohost : !llvm.ptr
+  %6 = omp.groupprivate %5 : !llvm.ptr, device_type(nohost) -> !llvm.ptr
+  llvm.store %0, %6 : i32, !llvm.ptr
+  llvm.return
+}
+}
 
-// CHECK-LABEL: @omp_groupprivate
-llvm.func @omp_groupprivate() {
+// CHECK: @any = internal global i32 undef
+// CHECK: @host = internal global i32 undef
+// CHECK: @nohost = internal global i32 undef
+// CHECK-LABEL: @omp_groupprivate_host
 // CHECK:  [[TMP1:%.*]] = call ptr @__kmpc_alloc_shared(i64 4)
 // CHECK:  store i32 1, ptr [[TMP1]], align 4
-
-// CHECK-LABEL: omp.teams.region{{.*}}
 // CHECK:  [[TMP2:%.*]] = call ptr @__kmpc_alloc_shared(i64 4)
-// CHECK:  store i32 2, ptr [[TMP2]], align 4
+// CHECK:  store i32 1, ptr [[TMP2]], align 4
+// CHECK:  store i32 1, ptr @nohost, align 4
 
-  %0 = llvm.mlir.constant(1 : i32) : i32
-  %1 = llvm.mlir.constant(2 : i32) : i32
-  %2 = llvm.mlir.constant(3 : i32) : i32
+// -----
 
-  %3 = llvm.mlir.addressof @_QFsubEx : !llvm.ptr
-  %4 = omp.groupprivate %3 : !llvm.ptr -> !llvm.ptr
+module attributes {omp.is_target_device = true} {
+llvm.mlir.global internal @any() : i32
+llvm.mlir.global internal @host() : i32
+llvm.mlir.global internal @nohost() : i32
+llvm.func @omp_groupprivate_device() {
+  %0 = llvm.mlir.constant(1 : i32) : i32
+  %1 = llvm.mlir.addressof @any : !llvm.ptr
+  %2 = omp.groupprivate %1 : !llvm.ptr, device_type(any) -> !llvm.ptr
+  llvm.store %0, %2 : i32, !llvm.ptr
 
+  %3 = llvm.mlir.addressof @host : !llvm.ptr
+  %4 = omp.groupprivate %3 : !llvm.ptr, device_type(host) -> !llvm.ptr
   llvm.store %0, %4 : i32, !llvm.ptr
 
-  omp.teams  {
-    %5 = omp.groupprivate %3 : !llvm.ptr -> !llvm.ptr
-    llvm.store %1, %5 : i32, !llvm.ptr
-    omp.terminator
-  }
-
-  llvm.store %2, %4 : i32, !llvm.ptr
+  %5 = llvm.mlir.addressof @nohost : !llvm.ptr
+  %6 = omp.groupprivate %5 : !llvm.ptr, device_type(nohost) -> !llvm.ptr
+  llvm.store %0, %6 : i32, !llvm.ptr
   llvm.return
 }
+}
 
-llvm.mlir.global internal @_QFsubEx() : i32
+// CHECK: @any = internal global i32 undef
+// CHECK: @host = internal global i32 undef
+// CHECK: @nohost = internal global i32 undef
+// CHECK-LABEL: @omp_groupprivate_device
+// CHECK:  [[TMP1:%.*]] = call ptr @__kmpc_alloc_shared(i64 4)
+// CHECK:  store i32 1, ptr [[TMP1]], align 4
+// CHECK:  store i32 1, ptr @host, align 4
+// CHECK:  [[TMP2:%.*]] = call ptr @__kmpc_alloc_shared(i64 4)
+// CHECK:  store i32 1, ptr [[TMP2]], align 4
 
 // -----