1// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s 2 3// nonesense minimised code simulating the control flow graph generated by flang 4// for array reductions. The important thing here is that we are testing a byref 5// reduction with a cleanup region, and the various regions contain multiple 6// blocks 7omp.declare_reduction @add_reduction_byref_box_Uxf32 : !llvm.ptr alloc { 8 %0 = llvm.mlir.constant(1 : i64) : i64 9 %1 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr 10 omp.yield(%1 : !llvm.ptr) 11} init { 12^bb0(%arg0: !llvm.ptr, %alloc: !llvm.ptr): 13 %0 = llvm.mlir.constant(1 : i64) : i64 14 llvm.store %0, %alloc : i64, !llvm.ptr 15 omp.yield(%alloc : !llvm.ptr) 16} combiner { 17^bb0(%arg0: !llvm.ptr, %arg1: !llvm.ptr): 18 %0 = llvm.mlir.constant(0 : i64) : i64 19 %1 = llvm.mlir.constant(0 : index) : i64 20 %2 = llvm.mlir.constant(1 : index) : i64 21 llvm.br ^bb1(%0 : i64) 22^bb1(%3: i64): // 2 preds: ^bb0, ^bb2 23 %4 = llvm.icmp "sgt" %3, %1 : i64 24 llvm.cond_br %4, ^bb2, ^bb3 25^bb2: // pred: ^bb1 26 %5 = llvm.sub %3, %2 : i64 27 llvm.br ^bb1(%5 : i64) 28^bb3: // pred: ^bb1 29 omp.yield(%arg0 : !llvm.ptr) 30} cleanup { 31^bb0(%arg0: !llvm.ptr): 32 %0 = llvm.mlir.constant(0 : i64) : i64 33 %1 = llvm.ptrtoint %arg0 : !llvm.ptr to i64 34 %2 = llvm.icmp "ne" %1, %0 : i64 35 llvm.cond_br %2, ^bb1, ^bb2 36^bb1: // pred: ^bb0 37 llvm.br ^bb2 38^bb2: // 2 preds: ^bb0, ^bb1 39 omp.yield 40} 41llvm.func @sectionsreduction_(%arg0: !llvm.ptr {fir.bindc_name = "x"}) attributes {fir.internal_name = "_QPsectionsreduction"} { 42 %0 = llvm.mlir.constant(1 : i64) : i64 43 %1 = llvm.mlir.constant(0 : index) : i64 44 %2 = llvm.mlir.constant(1 : index) : i64 45 omp.parallel { 46 %3 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr 47 omp.sections reduction(byref @add_reduction_byref_box_Uxf32 %3 -> %arg1 : !llvm.ptr) { 48 omp.section { 49 ^bb0(%arg2: !llvm.ptr): 50 llvm.br ^bb1(%0 : i64) 51 ^bb1(%4: i64): // 2 preds: ^bb0, ^bb2 52 %5 = llvm.icmp "sgt" %4, %1 : i64 53 llvm.cond_br %5, ^bb2, ^bb3 54 ^bb2: // pred: ^bb1 55 %6 = llvm.sub %4, %2 : i64 56 llvm.br ^bb1(%6 : i64) 57 ^bb3: // pred: ^bb1 58 omp.terminator 59 } 60 omp.section { 61 ^bb0(%arg2: !llvm.ptr): 62 llvm.br ^bb1(%0 : i64) 63 ^bb1(%4: i64): // 2 preds: ^bb0, ^bb2 64 %5 = llvm.icmp "sgt" %4, %1 : i64 65 llvm.cond_br %5, ^bb2, ^bb3 66 ^bb2: // pred: ^bb1 67 %6 = llvm.sub %4, %2 : i64 68 llvm.br ^bb1(%6 : i64) 69 ^bb3: // pred: ^bb1 70 omp.terminator 71 } 72 omp.terminator 73 } 74 omp.terminator 75 } 76 llvm.return 77} 78 79// CHECK-LABEL: define internal void @sectionsreduction_..omp_par 80// CHECK: [[PAR_ENTRY:omp.par.entry]]: 81// CHECK: %[[VAL_6:.*]] = alloca i32, align 4 82// CHECK: %[[VAL_7:.*]] = alloca i32, align 4 83// CHECK: %[[VAL_8:.*]] = alloca i32, align 4 84// CHECK: %[[VAL_9:.*]] = alloca i32, align 4 85// CHECK: %[[VAL_10:.*]] = alloca i32, align 4 86// CHECK: %[[VAL_11:.*]] = load i32, ptr %[[VAL_12:.*]], align 4 87// CHECK: store i32 %[[VAL_11]], ptr %[[VAL_10]], align 4 88// CHECK: %[[VAL_13:.*]] = load i32, ptr %[[VAL_10]], align 4 89// CHECK: %[[VAL_20:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 8 90// CHECK: %[[VAL_21:.*]] = alloca ptr, align 8 91// CHECK: %[[VAL_14:.*]] = alloca [1 x ptr], align 8 92// CHECK: br label %[[VAL_15:.*]] 93 94// CHECK: [[VAL_15]]: 95// CHECK: br label %[[PAR_REG:omp.par.region]] 96 97// CHECK: [[PAR_REG]]: ; preds = %[[VAL_15]] 98// CHECK: br label %[[VAL_18:.*]] 99// CHECK: omp.par.region1: ; preds = %[[PAR_REG]] 100// CHECK: %[[VAL_19:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 8 101// CHECK: br label %[[VAL_22:.*]] 102 103// CHECK: omp.reduction.init: ; preds = %[[VAL_16:.*]] 104// CHECK: store ptr %[[VAL_20]], ptr %[[VAL_21]], align 8 105// CHECK: br label %[[VAL_17:.*]] 106 107// CHECK: omp_section_loop.preheader: ; preds = %[[VAL_22]] 108// CHECK: store i32 0, ptr %[[VAL_7]], align 4 109// CHECK: store i32 1, ptr %[[VAL_8]], align 4 110// CHECK: store i32 1, ptr %[[VAL_9]], align 4 111// CHECK: %[[VAL_23:.*]] = call i32 @__kmpc_global_thread_num(ptr @1) 112// CHECK: call void @__kmpc_for_static_init_4u(ptr @1, i32 %[[VAL_23]], i32 34, ptr %[[VAL_6]], ptr %[[VAL_7]], ptr %[[VAL_8]], ptr %[[VAL_9]], i32 1, i32 0) 113// CHECK: %[[VAL_24:.*]] = load i32, ptr %[[VAL_7]], align 4 114// CHECK: %[[VAL_25:.*]] = load i32, ptr %[[VAL_8]], align 4 115// CHECK: %[[VAL_26:.*]] = sub i32 %[[VAL_25]], %[[VAL_24]] 116// CHECK: %[[VAL_27:.*]] = add i32 %[[VAL_26]], 1 117// CHECK: br label %[[VAL_28:.*]] 118// CHECK: omp_section_loop.header: ; preds = %[[VAL_29:.*]], %[[VAL_17]] 119// CHECK: %[[VAL_30:.*]] = phi i32 [ 0, %[[VAL_17]] ], [ %[[VAL_31:.*]], %[[VAL_29]] ] 120// CHECK: br label %[[VAL_32:.*]] 121// CHECK: omp_section_loop.cond: ; preds = %[[VAL_28]] 122// CHECK: %[[VAL_33:.*]] = icmp ult i32 %[[VAL_30]], %[[VAL_27]] 123// CHECK: br i1 %[[VAL_33]], label %[[VAL_34:.*]], label %[[VAL_35:.*]] 124// CHECK: omp_section_loop.exit: ; preds = %[[VAL_32]] 125// CHECK: call void @__kmpc_for_static_fini(ptr @1, i32 %[[VAL_23]]) 126// CHECK: %[[VAL_36:.*]] = call i32 @__kmpc_global_thread_num(ptr @1) 127// CHECK: call void @__kmpc_barrier(ptr @2, i32 %[[VAL_36]]) 128// CHECK: br label %[[VAL_37:.*]] 129// CHECK: omp_section_loop.after: ; preds = %[[VAL_35]] 130// CHECK: br label %[[VAL_38:.*]] 131// CHECK: omp_section_loop.aftersections.fini: ; preds = %[[VAL_37]] 132// CHECK: %[[VAL_39:.*]] = getelementptr inbounds [1 x ptr], ptr %[[VAL_14]], i64 0, i64 0 133// CHECK: store ptr %[[VAL_21]], ptr %[[VAL_39]], align 8 134// CHECK: %[[VAL_40:.*]] = call i32 @__kmpc_global_thread_num(ptr @1) 135// CHECK: %[[VAL_41:.*]] = call i32 @__kmpc_reduce(ptr @1, i32 %[[VAL_40]], i32 1, i64 8, ptr %[[VAL_14]], ptr @.omp.reduction.func, ptr @.gomp_critical_user_.reduction.var) 136// CHECK: switch i32 %[[VAL_41]], label %[[VAL_42:.*]] [ 137// CHECK: i32 1, label %[[VAL_43:.*]] 138// CHECK: i32 2, label %[[VAL_44:.*]] 139// CHECK: ] 140// CHECK: reduce.switch.atomic: ; preds = %[[VAL_38]] 141// CHECK: unreachable 142// CHECK: reduce.switch.nonatomic: ; preds = %[[VAL_38]] 143// CHECK: %[[VAL_45:.*]] = load ptr, ptr %[[VAL_21]], align 8 144// CHECK: br label %[[VAL_46:.*]] 145// CHECK: omp.reduction.nonatomic.body: ; preds = %[[VAL_43]] 146// CHECK: br label %[[VAL_47:.*]] 147// CHECK: omp.reduction.nonatomic.body16: ; preds = %[[VAL_48:.*]], %[[VAL_46]] 148// CHECK: %[[VAL_49:.*]] = phi i64 [ %[[VAL_50:.*]], %[[VAL_48]] ], [ 0, %[[VAL_46]] ] 149// CHECK: %[[VAL_51:.*]] = icmp sgt i64 %[[VAL_49]], 0 150// CHECK: br i1 %[[VAL_51]], label %[[VAL_48]], label %[[VAL_52:.*]] 151// CHECK: omp.reduction.nonatomic.body18: ; preds = %[[VAL_47]] 152// CHECK: br label %[[VAL_53:.*]] 153// CHECK: omp.region.cont15: ; preds = %[[VAL_52]] 154// CHECK: %[[VAL_54:.*]] = phi ptr [ %[[VAL_19]], %[[VAL_52]] ] 155// CHECK: call void @__kmpc_end_reduce(ptr @1, i32 %[[VAL_40]], ptr @.gomp_critical_user_.reduction.var) 156// CHECK: br label %[[VAL_42]] 157// CHECK: omp.reduction.nonatomic.body17: ; preds = %[[VAL_47]] 158// CHECK: %[[VAL_50]] = sub i64 %[[VAL_49]], 1 159// CHECK: br label %[[VAL_47]] 160// CHECK: reduce.finalize: ; preds = %[[VAL_53]], %[[VAL_38]] 161// CHECK: %[[VAL_55:.*]] = call i32 @__kmpc_global_thread_num(ptr @1) 162// CHECK: call void @__kmpc_barrier(ptr @2, i32 %[[VAL_55]]) 163// CHECK: %[[VAL_56:.*]] = load ptr, ptr %[[VAL_21]], align 8 164// CHECK: br label %[[VAL_57:.*]] 165// CHECK: omp.reduction.cleanup: ; preds = %[[VAL_42]] 166// CHECK: %[[VAL_58:.*]] = ptrtoint ptr %[[VAL_56]] to i64 167// CHECK: %[[VAL_59:.*]] = icmp ne i64 %[[VAL_58]], 0 168// CHECK: br i1 %[[VAL_59]], label %[[VAL_60:.*]], label %[[VAL_61:.*]] 169// CHECK: omp.reduction.cleanup22: ; preds = %[[VAL_60]], %[[VAL_57]] 170// CHECK: br label %[[VAL_62:.*]] 171// CHECK: omp.region.cont20: ; preds = %[[VAL_61]] 172// CHECK: br label %[[VAL_63:.*]] 173// CHECK: omp.region.cont: ; preds = %[[VAL_62]] 174// CHECK: br label %[[VAL_64:.*]] 175// CHECK: omp.par.pre_finalize: ; preds = %[[VAL_63]] 176// CHECK: br label %[[VAL_65:.*]] 177// CHECK: omp.reduction.cleanup21: ; preds = %[[VAL_57]] 178// CHECK: br label %[[VAL_61]] 179// CHECK: omp_section_loop.body: ; preds = %[[VAL_32]] 180// CHECK: %[[VAL_66:.*]] = add i32 %[[VAL_30]], %[[VAL_24]] 181// CHECK: %[[VAL_67:.*]] = mul i32 %[[VAL_66]], 1 182// CHECK: %[[VAL_68:.*]] = add i32 %[[VAL_67]], 0 183// CHECK: switch i32 %[[VAL_68]], label %[[VAL_69:.*]] [ 184// CHECK: i32 0, label %[[VAL_70:.*]] 185// CHECK: i32 1, label %[[VAL_71:.*]] 186// CHECK: ] 187// CHECK: omp_section_loop.body.case6: ; preds = %[[VAL_34]] 188// CHECK: br label %[[VAL_72:.*]] 189// CHECK: omp.section.region8: ; preds = %[[VAL_71]] 190// CHECK: br label %[[VAL_73:.*]] 191// CHECK: omp.section.region9: ; preds = %[[VAL_74:.*]], %[[VAL_72]] 192// CHECK: %[[VAL_75:.*]] = phi i64 [ %[[VAL_76:.*]], %[[VAL_74]] ], [ 1, %[[VAL_72]] ] 193// CHECK: %[[VAL_77:.*]] = icmp sgt i64 %[[VAL_75]], 0 194// CHECK: br i1 %[[VAL_77]], label %[[VAL_74]], label %[[VAL_78:.*]] 195// CHECK: omp.section.region11: ; preds = %[[VAL_73]] 196// CHECK: br label %[[VAL_79:.*]] 197// CHECK: omp.region.cont7: ; preds = %[[VAL_78]] 198// CHECK: br label %[[VAL_69]] 199// CHECK: omp.section.region10: ; preds = %[[VAL_73]] 200// CHECK: %[[VAL_76]] = sub i64 %[[VAL_75]], 1 201// CHECK: br label %[[VAL_73]] 202// CHECK: omp_section_loop.body.case: ; preds = %[[VAL_34]] 203// CHECK: br label %[[VAL_80:.*]] 204// CHECK: omp.section.region: ; preds = %[[VAL_70]] 205// CHECK: br label %[[VAL_81:.*]] 206// CHECK: omp.section.region3: ; preds = %[[VAL_82:.*]], %[[VAL_80]] 207// CHECK: %[[VAL_83:.*]] = phi i64 [ %[[VAL_84:.*]], %[[VAL_82]] ], [ 1, %[[VAL_80]] ] 208// CHECK: %[[VAL_85:.*]] = icmp sgt i64 %[[VAL_83]], 0 209// CHECK: br i1 %[[VAL_85]], label %[[VAL_82]], label %[[VAL_86:.*]] 210// CHECK: omp.section.region5: ; preds = %[[VAL_81]] 211// CHECK: br label %[[VAL_87:.*]] 212// CHECK: omp.region.cont2: ; preds = %[[VAL_86]] 213// CHECK: br label %[[VAL_69]] 214// CHECK: omp.section.region4: ; preds = %[[VAL_81]] 215// CHECK: %[[VAL_84]] = sub i64 %[[VAL_83]], 1 216// CHECK: br label %[[VAL_81]] 217// CHECK: omp_section_loop.body.sections.after: ; preds = %[[VAL_79]], %[[VAL_87]], %[[VAL_34]] 218// CHECK: br label %[[VAL_29]] 219// CHECK: omp_section_loop.inc: ; preds = %[[VAL_69]] 220// CHECK: %[[VAL_31]] = add nuw i32 %[[VAL_30]], 1 221// CHECK: br label %[[VAL_28]] 222// CHECK: omp.par.outlined.exit.exitStub: ; preds = %[[VAL_64]] 223// CHECK: ret void 224