146 lines · plain
1// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s2 3module attributes {omp.is_target_device = false, omp.target_triples = ["amdgcn-amd-amdhsa"]} {4 llvm.func @_QQmain() attributes {fir.bindc_name = "main"} {5 %0 = llvm.mlir.constant(39 : index) : i646 %1 = llvm.mlir.constant(0 : index) : i647 %2 = llvm.mlir.constant(1 : index) : i648 %3 = llvm.mlir.constant(40 : index) : i649 %4 = llvm.mlir.addressof @_QFEa : !llvm.ptr10 %5 = llvm.mlir.addressof @_QFEb : !llvm.ptr11 %6 = llvm.mlir.constant(1 : i64) : i6412 %7 = llvm.alloca %6 x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr13 %8 = llvm.mlir.addressof @_QFEn : !llvm.ptr14 omp.task {15 %14 = llvm.mlir.constant(1 : i64) : i6416 %15 = llvm.alloca %14 x i32 {bindc_name = "i", pinned} : (i64) -> !llvm.ptr17 %16 = llvm.load %8 : !llvm.ptr -> i3218 %17 = llvm.sext %16 : i32 to i6419 %18 = llvm.trunc %2 : i64 to i3220 llvm.br ^bb1(%18, %17 : i32, i64)21 ^bb1(%19: i32, %20: i64): // 2 preds: ^bb0, ^bb222 %21 = llvm.icmp "sgt" %20, %1 : i6423 llvm.cond_br %21, ^bb2, ^bb324 ^bb2: // pred: ^bb125 llvm.store %19, %15 : i32, !llvm.ptr26 %22 = llvm.load %15 : !llvm.ptr -> i3227 %23 = llvm.sext %22 : i32 to i6428 %24 = llvm.mlir.constant(1 : i64) : i6429 %25 = llvm.mlir.constant(0 : i64) : i6430 %26 = llvm.sub %23, %24 overflow<nsw> : i6431 %27 = llvm.mul %26, %24 overflow<nsw> : i6432 %28 = llvm.mul %27, %24 overflow<nsw> : i6433 %29 = llvm.add %28, %25 overflow<nsw> : i6434 %30 = llvm.mul %24, %3 overflow<nsw> : i6435 %31 = llvm.getelementptr %4[%29] : (!llvm.ptr, i64) -> !llvm.ptr, i3236 llvm.store %22, %31 : i32, !llvm.ptr37 %32 = llvm.load %15 : !llvm.ptr -> i3238 %33 = llvm.add %32, %18 : i3239 %34 = llvm.sub %20, %2 : i6440 llvm.br ^bb1(%33, %34 : i32, i64)41 ^bb3: // pred: ^bb142 llvm.store %19, %15 : i32, !llvm.ptr43 omp.terminator44 }45 %9 = omp.map.bounds lower_bound(%1 : i64) upper_bound(%0 : i64) extent(%3 : i64) stride(%2 : i64) start_idx(%2 : i64)46 %10 = omp.map.info var_ptr(%4 : !llvm.ptr, !llvm.array<40 x i32>) map_clauses(to) capture(ByRef) bounds(%9) -> !llvm.ptr {name = "a"}47 %11 = omp.map.info var_ptr(%5 : !llvm.ptr, !llvm.array<40 x i32>) map_clauses(from) capture(ByRef) bounds(%9) -> !llvm.ptr {name = "b"}48 %12 = omp.map.info var_ptr(%7 : !llvm.ptr, i32) map_clauses(implicit, exit_release_or_enter_alloc) capture(ByCopy) -> !llvm.ptr {name = "i"}49 %13 = omp.map.info var_ptr(%8 : !llvm.ptr, i32) map_clauses(implicit, exit_release_or_enter_alloc) capture(ByCopy) -> !llvm.ptr {name = "n"}50 omp.target depend(taskdependin -> %4 : !llvm.ptr) map_entries(%10 -> %arg0, %11 -> %arg1, %12 -> %arg2, %13 -> %arg3 : !llvm.ptr, !llvm.ptr, !llvm.ptr, !llvm.ptr) {51 %14 = llvm.mlir.constant(0 : index) : i6452 %15 = llvm.mlir.constant(10 : i32) : i3253 %16 = llvm.mlir.constant(1 : index) : i6454 %17 = llvm.mlir.constant(40 : index) : i6455 %18 = llvm.load %arg3 : !llvm.ptr -> i3256 %19 = llvm.sext %18 : i32 to i6457 %20 = llvm.trunc %16 : i64 to i3258 llvm.br ^bb1(%20, %19 : i32, i64)59 ^bb1(%21: i32, %22: i64): // 2 preds: ^bb0, ^bb260 %23 = llvm.icmp "sgt" %22, %14 : i6461 llvm.cond_br %23, ^bb2, ^bb362 ^bb2: // pred: ^bb163 llvm.store %21, %arg2 : i32, !llvm.ptr64 %24 = llvm.load %arg2 : !llvm.ptr -> i3265 %25 = llvm.sext %24 : i32 to i6466 %26 = llvm.mlir.constant(1 : i64) : i6467 %27 = llvm.mlir.constant(0 : i64) : i6468 %28 = llvm.sub %25, %26 overflow<nsw> : i6469 %29 = llvm.mul %28, %26 overflow<nsw> : i6470 %30 = llvm.mul %29, %26 overflow<nsw> : i6471 %31 = llvm.add %30, %27 overflow<nsw> : i6472 %32 = llvm.mul %26, %17 overflow<nsw> : i6473 %33 = llvm.getelementptr %arg0[%31] : (!llvm.ptr, i64) -> !llvm.ptr, i3274 %34 = llvm.load %33 : !llvm.ptr -> i3275 %35 = llvm.add %34, %15 : i3276 %36 = llvm.mlir.constant(1 : i64) : i6477 %37 = llvm.mlir.constant(0 : i64) : i6478 %38 = llvm.sub %25, %36 overflow<nsw> : i6479 %39 = llvm.mul %38, %36 overflow<nsw> : i6480 %40 = llvm.mul %39, %36 overflow<nsw> : i6481 %41 = llvm.add %40, %37 overflow<nsw> : i6482 %42 = llvm.mul %36, %17 overflow<nsw> : i6483 %43 = llvm.getelementptr %arg1[%41] : (!llvm.ptr, i64) -> !llvm.ptr, i3284 llvm.store %35, %43 : i32, !llvm.ptr85 %44 = llvm.load %arg2 : !llvm.ptr -> i3286 %45 = llvm.add %44, %20 : i3287 %46 = llvm.sub %22, %16 : i6488 llvm.br ^bb1(%45, %46 : i32, i64)89 ^bb3: // pred: ^bb190 llvm.store %21, %arg2 : i32, !llvm.ptr91 omp.terminator92 }93 llvm.return94 }95 llvm.mlir.global internal @_QFEa() {addr_space = 0 : i32} : !llvm.array<40 x i32> {96 %0 = llvm.mlir.zero : !llvm.array<40 x i32>97 llvm.return %0 : !llvm.array<40 x i32>98 }99 llvm.mlir.global internal @_QFEb() {addr_space = 0 : i32} : !llvm.array<40 x i32> {100 %0 = llvm.mlir.zero : !llvm.array<40 x i32>101 llvm.return %0 : !llvm.array<40 x i32>102 }103 llvm.mlir.global internal @_QFEc() {addr_space = 0 : i32} : !llvm.array<40 x i32> {104 %0 = llvm.mlir.zero : !llvm.array<40 x i32>105 llvm.return %0 : !llvm.array<40 x i32>106 }107 llvm.mlir.global internal @_QFEn() {addr_space = 0 : i32} : i32 {108 %0 = llvm.mlir.constant(40 : i32) : i32109 llvm.return %0 : i32110 }111 llvm.func @_FortranAProgramStart(i32, !llvm.ptr, !llvm.ptr, !llvm.ptr) attributes {sym_visibility = "private"}112 llvm.func @_FortranAProgramEndStatement() attributes {sym_visibility = "private"}113 llvm.func @main(%arg0: i32, %arg1: !llvm.ptr, %arg2: !llvm.ptr) -> i32 {114 %0 = llvm.mlir.constant(0 : i32) : i32115 %1 = llvm.mlir.zero : !llvm.ptr116 llvm.call @_FortranAProgramStart(%arg0, %arg1, %arg2, %1) {fastmathFlags = #llvm.fastmath<contract>} : (i32, !llvm.ptr, !llvm.ptr, !llvm.ptr) -> ()117 llvm.call @_QQmain() {fastmathFlags = #llvm.fastmath<contract>} : () -> ()118 llvm.call @_FortranAProgramEndStatement() {fastmathFlags = #llvm.fastmath<contract>} : () -> ()119 llvm.return %0 : i32120 }121}122 123// %strucArg holds pointers to shared data.124// CHECK: define void @_QQmain() {125// CHECK-DAG: %[[STRUCTARG:.+]] = alloca { ptr, ptr, ptr }, align 8126// CHECK-DAG: %[[DEP_ARRAY:.+]] = alloca [1 x %struct.kmp_dep_info], align 8127 128// CHECK: %[[TASKDATA:.+]] = call ptr @__kmpc_omp_task_alloc({{.+}}, ptr @.omp_target_task_proxy_func)129// CHECK: %[[SHARED_PTR:.+]] = getelementptr inbounds nuw %struct.kmp_task_ompbuilder_t, ptr %[[TASKDATA]], i32 0, i32 0130// CHECK: %[[SHARED_DATA:.+]] = load ptr, ptr %[[SHARED_PTR]], align 8131// CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align 1 %[[SHARED_DATA]], ptr align 1 %[[STRUCTARG]], i64 24, i1 false)132 133// CHECK: %[[DEP_INFO:.+]] = getelementptr inbounds [1 x %struct.kmp_dep_info], ptr %[[DEP_ARRAY]], i64 0, i64 0134// CHECK: %[[PTR0:.+]] = getelementptr inbounds nuw %struct.kmp_dep_info, ptr %[[DEP_INFO]], i32 0, i32 0135// CHECK: store i64 ptrtoint (ptr @_QFEa to i64), ptr %[[PTR0]], align 4136// CHECK: %[[PTR1:.+]] = getelementptr inbounds nuw %struct.kmp_dep_info, ptr %[[DEP_INFO]], i32 0, i32 1137// CHECK: store i64 8, ptr %[[PTR1]], align 4138// CHECK: %[[PTR2:.+]] = getelementptr inbounds nuw %struct.kmp_dep_info, ptr %[[DEP_INFO]], i32 0, i32 2139// CHECK: store i8 1, ptr %[[PTR2]], align 1140 141// CHECK: call void @__kmpc_omp_wait_deps({{.+}}, i32 1, ptr %[[DEP_ARRAY]], i32 0, ptr null)142// CHECK: call void @__kmpc_omp_task_begin_if0({{.+}}, ptr %[[TASKDATA]])143// CHECK: call void @.omp_target_task_proxy_func({{.+}}, ptr %[[TASKDATA]])144// CHECK: call void @__kmpc_omp_task_complete_if0({{.+}}, ptr %[[TASKDATA]])145 146