32 lines · plain
1// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s2 3// CHECK: define void @main(i32 %[[NUM_TEAMS_ARG:.*]])4// CHECK: %[[KERNEL_ARGS:.*]] = alloca %struct.__tgt_kernel_arguments5// CHECK: %[[NUM_TEAMS:.*]] = insertvalue [3 x i32] zeroinitializer, i32 %[[NUM_TEAMS_ARG]], 06 7// CHECK: %[[NUM_TEAMS_KARG:.*]] = getelementptr inbounds nuw %struct.__tgt_kernel_arguments, ptr %[[KERNEL_ARGS]], i32 0, i32 108// CHECK: store [3 x i32] %[[NUM_TEAMS]], ptr %[[NUM_TEAMS_KARG]], align 49 10// CHECK: %[[NUM_THREADS_ARG:.*]] = getelementptr inbounds nuw %struct.__tgt_kernel_arguments, ptr %[[KERNEL_ARGS]], i32 0, i32 1111// CHECK: store [3 x i32] [i32 10, i32 0, i32 0], ptr %[[NUM_THREADS_ARG]], align 412 13// CHECK: %{{.*}} = call i32 @__tgt_target_kernel(ptr {{.*}}, i64 -1, i32 %[[NUM_TEAMS_ARG]], i32 [[NUM_THREADS:10]], ptr @.[[OUTLINED_FN:.*]].region_id, ptr %[[KERNEL_ARGS]])14// CHECK: call void @[[OUTLINED_FN]](i32 %[[NUM_TEAMS_ARG]])15 16// CHECK: define internal void @[[OUTLINED_FN]](i32 %[[NUM_TEAMS_OUTLINED:.*]])17// CHECK: call void @__kmpc_push_num_teams_51(ptr {{.*}}, i32 {{.*}}, i32 %[[NUM_TEAMS_OUTLINED]], i32 %[[NUM_TEAMS_OUTLINED]], i32 [[NUM_THREADS]])18module attributes {omp.is_target_device = false, omp.target_triples = ["amdgcn-amd-amdhsa"]} {19 llvm.func @main(%num_teams : i32) {20 %target_threads = llvm.mlir.constant(20) : i3221 %teams_threads = llvm.mlir.constant(10) : i3222 omp.target thread_limit(%target_threads : i32)23 host_eval(%num_teams -> %arg_teams, %teams_threads -> %arg_teams_threads : i32, i32) {24 omp.teams num_teams(to %arg_teams : i32) thread_limit(%arg_teams_threads : i32) {25 omp.terminator26 }27 omp.terminator28 }29 llvm.return30 }31}32