6b3ba6677d
While https://github.com/llvm/llvm-project/pull/122866 fixed some issues, it introduced a regression in worksharing loops. The new bug comes from the fact that we now conditionally created the `after_alloca` block based on the number of sucessors of the alloca insertion point. This is unneccessary, we can just alway create the block. If we do this, we respect the post condtions expected after calling `allocatePrivateVars` (i.e. that the `afterAlloca` block has a single predecessor.
224 lines
12 KiB
MLIR
224 lines
12 KiB
MLIR
// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
|
|
|
|
// nonesense minimised code simulating the control flow graph generated by flang
|
|
// for array reductions. The important thing here is that we are testing a byref
|
|
// reduction with a cleanup region, and the various regions contain multiple
|
|
// blocks
|
|
omp.declare_reduction @add_reduction_byref_box_Uxf32 : !llvm.ptr alloc {
|
|
%0 = llvm.mlir.constant(1 : i64) : i64
|
|
%1 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr
|
|
omp.yield(%1 : !llvm.ptr)
|
|
} init {
|
|
^bb0(%arg0: !llvm.ptr, %alloc: !llvm.ptr):
|
|
%0 = llvm.mlir.constant(1 : i64) : i64
|
|
llvm.store %0, %alloc : i64, !llvm.ptr
|
|
omp.yield(%alloc : !llvm.ptr)
|
|
} combiner {
|
|
^bb0(%arg0: !llvm.ptr, %arg1: !llvm.ptr):
|
|
%0 = llvm.mlir.constant(0 : i64) : i64
|
|
%1 = llvm.mlir.constant(0 : index) : i64
|
|
%2 = llvm.mlir.constant(1 : index) : i64
|
|
llvm.br ^bb1(%0 : i64)
|
|
^bb1(%3: i64): // 2 preds: ^bb0, ^bb2
|
|
%4 = llvm.icmp "sgt" %3, %1 : i64
|
|
llvm.cond_br %4, ^bb2, ^bb3
|
|
^bb2: // pred: ^bb1
|
|
%5 = llvm.sub %3, %2 : i64
|
|
llvm.br ^bb1(%5 : i64)
|
|
^bb3: // pred: ^bb1
|
|
omp.yield(%arg0 : !llvm.ptr)
|
|
} cleanup {
|
|
^bb0(%arg0: !llvm.ptr):
|
|
%0 = llvm.mlir.constant(0 : i64) : i64
|
|
%1 = llvm.ptrtoint %arg0 : !llvm.ptr to i64
|
|
%2 = llvm.icmp "ne" %1, %0 : i64
|
|
llvm.cond_br %2, ^bb1, ^bb2
|
|
^bb1: // pred: ^bb0
|
|
llvm.br ^bb2
|
|
^bb2: // 2 preds: ^bb0, ^bb1
|
|
omp.yield
|
|
}
|
|
llvm.func @sectionsreduction_(%arg0: !llvm.ptr {fir.bindc_name = "x"}) attributes {fir.internal_name = "_QPsectionsreduction"} {
|
|
%0 = llvm.mlir.constant(1 : i64) : i64
|
|
%1 = llvm.mlir.constant(0 : index) : i64
|
|
%2 = llvm.mlir.constant(1 : index) : i64
|
|
omp.parallel {
|
|
%3 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr
|
|
omp.sections reduction(byref @add_reduction_byref_box_Uxf32 %3 -> %arg1 : !llvm.ptr) {
|
|
omp.section {
|
|
^bb0(%arg2: !llvm.ptr):
|
|
llvm.br ^bb1(%0 : i64)
|
|
^bb1(%4: i64): // 2 preds: ^bb0, ^bb2
|
|
%5 = llvm.icmp "sgt" %4, %1 : i64
|
|
llvm.cond_br %5, ^bb2, ^bb3
|
|
^bb2: // pred: ^bb1
|
|
%6 = llvm.sub %4, %2 : i64
|
|
llvm.br ^bb1(%6 : i64)
|
|
^bb3: // pred: ^bb1
|
|
omp.terminator
|
|
}
|
|
omp.section {
|
|
^bb0(%arg2: !llvm.ptr):
|
|
llvm.br ^bb1(%0 : i64)
|
|
^bb1(%4: i64): // 2 preds: ^bb0, ^bb2
|
|
%5 = llvm.icmp "sgt" %4, %1 : i64
|
|
llvm.cond_br %5, ^bb2, ^bb3
|
|
^bb2: // pred: ^bb1
|
|
%6 = llvm.sub %4, %2 : i64
|
|
llvm.br ^bb1(%6 : i64)
|
|
^bb3: // pred: ^bb1
|
|
omp.terminator
|
|
}
|
|
omp.terminator
|
|
}
|
|
omp.terminator
|
|
}
|
|
llvm.return
|
|
}
|
|
|
|
// CHECK-LABEL: define internal void @sectionsreduction_..omp_par
|
|
// CHECK: [[PAR_ENTRY:omp.par.entry]]:
|
|
// CHECK: %[[VAL_6:.*]] = alloca i32, align 4
|
|
// CHECK: %[[VAL_7:.*]] = alloca i32, align 4
|
|
// CHECK: %[[VAL_8:.*]] = alloca i32, align 4
|
|
// CHECK: %[[VAL_9:.*]] = alloca i32, align 4
|
|
// CHECK: %[[VAL_10:.*]] = alloca i32, align 4
|
|
// CHECK: %[[VAL_11:.*]] = load i32, ptr %[[VAL_12:.*]], align 4
|
|
// CHECK: store i32 %[[VAL_11]], ptr %[[VAL_10]], align 4
|
|
// CHECK: %[[VAL_13:.*]] = load i32, ptr %[[VAL_10]], align 4
|
|
// CHECK: %[[VAL_20:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 8
|
|
// CHECK: %[[VAL_21:.*]] = alloca ptr, align 8
|
|
// CHECK: %[[VAL_14:.*]] = alloca [1 x ptr], align 8
|
|
// CHECK: br label %[[VAL_15:.*]]
|
|
|
|
// CHECK: [[VAL_15]]:
|
|
// CHECK: br label %[[PAR_REG:omp.par.region]]
|
|
|
|
// CHECK: [[PAR_REG]]: ; preds = %[[VAL_15]]
|
|
// CHECK: br label %[[VAL_18:.*]]
|
|
// CHECK: omp.par.region1: ; preds = %[[PAR_REG]]
|
|
// CHECK: %[[VAL_19:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 8
|
|
// CHECK: br label %[[VAL_22:.*]]
|
|
|
|
// CHECK: omp.reduction.init: ; preds = %[[VAL_16:.*]]
|
|
// CHECK: store ptr %[[VAL_20]], ptr %[[VAL_21]], align 8
|
|
// CHECK: br label %[[VAL_17:.*]]
|
|
|
|
// CHECK: omp_section_loop.preheader: ; preds = %[[VAL_22]]
|
|
// CHECK: store i32 0, ptr %[[VAL_7]], align 4
|
|
// CHECK: store i32 1, ptr %[[VAL_8]], align 4
|
|
// CHECK: store i32 1, ptr %[[VAL_9]], align 4
|
|
// CHECK: %[[VAL_23:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)
|
|
// 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)
|
|
// CHECK: %[[VAL_24:.*]] = load i32, ptr %[[VAL_7]], align 4
|
|
// CHECK: %[[VAL_25:.*]] = load i32, ptr %[[VAL_8]], align 4
|
|
// CHECK: %[[VAL_26:.*]] = sub i32 %[[VAL_25]], %[[VAL_24]]
|
|
// CHECK: %[[VAL_27:.*]] = add i32 %[[VAL_26]], 1
|
|
// CHECK: br label %[[VAL_28:.*]]
|
|
// CHECK: omp_section_loop.header: ; preds = %[[VAL_29:.*]], %[[VAL_17]]
|
|
// CHECK: %[[VAL_30:.*]] = phi i32 [ 0, %[[VAL_17]] ], [ %[[VAL_31:.*]], %[[VAL_29]] ]
|
|
// CHECK: br label %[[VAL_32:.*]]
|
|
// CHECK: omp_section_loop.cond: ; preds = %[[VAL_28]]
|
|
// CHECK: %[[VAL_33:.*]] = icmp ult i32 %[[VAL_30]], %[[VAL_27]]
|
|
// CHECK: br i1 %[[VAL_33]], label %[[VAL_34:.*]], label %[[VAL_35:.*]]
|
|
// CHECK: omp_section_loop.exit: ; preds = %[[VAL_32]]
|
|
// CHECK: call void @__kmpc_for_static_fini(ptr @1, i32 %[[VAL_23]])
|
|
// CHECK: %[[VAL_36:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)
|
|
// CHECK: call void @__kmpc_barrier(ptr @2, i32 %[[VAL_36]])
|
|
// CHECK: br label %[[VAL_37:.*]]
|
|
// CHECK: omp_section_loop.after: ; preds = %[[VAL_35]]
|
|
// CHECK: br label %[[VAL_38:.*]]
|
|
// CHECK: omp_section_loop.aftersections.fini: ; preds = %[[VAL_37]]
|
|
// CHECK: %[[VAL_39:.*]] = getelementptr inbounds [1 x ptr], ptr %[[VAL_14]], i64 0, i64 0
|
|
// CHECK: store ptr %[[VAL_21]], ptr %[[VAL_39]], align 8
|
|
// CHECK: %[[VAL_40:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)
|
|
// 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)
|
|
// CHECK: switch i32 %[[VAL_41]], label %[[VAL_42:.*]] [
|
|
// CHECK: i32 1, label %[[VAL_43:.*]]
|
|
// CHECK: i32 2, label %[[VAL_44:.*]]
|
|
// CHECK: ]
|
|
// CHECK: reduce.switch.atomic: ; preds = %[[VAL_38]]
|
|
// CHECK: unreachable
|
|
// CHECK: reduce.switch.nonatomic: ; preds = %[[VAL_38]]
|
|
// CHECK: %[[VAL_45:.*]] = load ptr, ptr %[[VAL_21]], align 8
|
|
// CHECK: br label %[[VAL_46:.*]]
|
|
// CHECK: omp.reduction.nonatomic.body: ; preds = %[[VAL_43]]
|
|
// CHECK: br label %[[VAL_47:.*]]
|
|
// CHECK: omp.reduction.nonatomic.body16: ; preds = %[[VAL_48:.*]], %[[VAL_46]]
|
|
// CHECK: %[[VAL_49:.*]] = phi i64 [ %[[VAL_50:.*]], %[[VAL_48]] ], [ 0, %[[VAL_46]] ]
|
|
// CHECK: %[[VAL_51:.*]] = icmp sgt i64 %[[VAL_49]], 0
|
|
// CHECK: br i1 %[[VAL_51]], label %[[VAL_48]], label %[[VAL_52:.*]]
|
|
// CHECK: omp.reduction.nonatomic.body18: ; preds = %[[VAL_47]]
|
|
// CHECK: br label %[[VAL_53:.*]]
|
|
// CHECK: omp.region.cont15: ; preds = %[[VAL_52]]
|
|
// CHECK: %[[VAL_54:.*]] = phi ptr [ %[[VAL_19]], %[[VAL_52]] ]
|
|
// CHECK: call void @__kmpc_end_reduce(ptr @1, i32 %[[VAL_40]], ptr @.gomp_critical_user_.reduction.var)
|
|
// CHECK: br label %[[VAL_42]]
|
|
// CHECK: omp.reduction.nonatomic.body17: ; preds = %[[VAL_47]]
|
|
// CHECK: %[[VAL_50]] = sub i64 %[[VAL_49]], 1
|
|
// CHECK: br label %[[VAL_47]]
|
|
// CHECK: reduce.finalize: ; preds = %[[VAL_53]], %[[VAL_38]]
|
|
// CHECK: %[[VAL_55:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)
|
|
// CHECK: call void @__kmpc_barrier(ptr @2, i32 %[[VAL_55]])
|
|
// CHECK: %[[VAL_56:.*]] = load ptr, ptr %[[VAL_21]], align 8
|
|
// CHECK: br label %[[VAL_57:.*]]
|
|
// CHECK: omp.reduction.cleanup: ; preds = %[[VAL_42]]
|
|
// CHECK: %[[VAL_58:.*]] = ptrtoint ptr %[[VAL_56]] to i64
|
|
// CHECK: %[[VAL_59:.*]] = icmp ne i64 %[[VAL_58]], 0
|
|
// CHECK: br i1 %[[VAL_59]], label %[[VAL_60:.*]], label %[[VAL_61:.*]]
|
|
// CHECK: omp.reduction.cleanup22: ; preds = %[[VAL_60]], %[[VAL_57]]
|
|
// CHECK: br label %[[VAL_62:.*]]
|
|
// CHECK: omp.region.cont20: ; preds = %[[VAL_61]]
|
|
// CHECK: br label %[[VAL_63:.*]]
|
|
// CHECK: omp.region.cont: ; preds = %[[VAL_62]]
|
|
// CHECK: br label %[[VAL_64:.*]]
|
|
// CHECK: omp.par.pre_finalize: ; preds = %[[VAL_63]]
|
|
// CHECK: br label %[[VAL_65:.*]]
|
|
// CHECK: omp.reduction.cleanup21: ; preds = %[[VAL_57]]
|
|
// CHECK: br label %[[VAL_61]]
|
|
// CHECK: omp_section_loop.body: ; preds = %[[VAL_32]]
|
|
// CHECK: %[[VAL_66:.*]] = add i32 %[[VAL_30]], %[[VAL_24]]
|
|
// CHECK: %[[VAL_67:.*]] = mul i32 %[[VAL_66]], 1
|
|
// CHECK: %[[VAL_68:.*]] = add i32 %[[VAL_67]], 0
|
|
// CHECK: switch i32 %[[VAL_68]], label %[[VAL_69:.*]] [
|
|
// CHECK: i32 0, label %[[VAL_70:.*]]
|
|
// CHECK: i32 1, label %[[VAL_71:.*]]
|
|
// CHECK: ]
|
|
// CHECK: omp_section_loop.body.case6: ; preds = %[[VAL_34]]
|
|
// CHECK: br label %[[VAL_72:.*]]
|
|
// CHECK: omp.section.region8: ; preds = %[[VAL_71]]
|
|
// CHECK: br label %[[VAL_73:.*]]
|
|
// CHECK: omp.section.region9: ; preds = %[[VAL_74:.*]], %[[VAL_72]]
|
|
// CHECK: %[[VAL_75:.*]] = phi i64 [ %[[VAL_76:.*]], %[[VAL_74]] ], [ 1, %[[VAL_72]] ]
|
|
// CHECK: %[[VAL_77:.*]] = icmp sgt i64 %[[VAL_75]], 0
|
|
// CHECK: br i1 %[[VAL_77]], label %[[VAL_74]], label %[[VAL_78:.*]]
|
|
// CHECK: omp.section.region11: ; preds = %[[VAL_73]]
|
|
// CHECK: br label %[[VAL_79:.*]]
|
|
// CHECK: omp.region.cont7: ; preds = %[[VAL_78]]
|
|
// CHECK: br label %[[VAL_69]]
|
|
// CHECK: omp.section.region10: ; preds = %[[VAL_73]]
|
|
// CHECK: %[[VAL_76]] = sub i64 %[[VAL_75]], 1
|
|
// CHECK: br label %[[VAL_73]]
|
|
// CHECK: omp_section_loop.body.case: ; preds = %[[VAL_34]]
|
|
// CHECK: br label %[[VAL_80:.*]]
|
|
// CHECK: omp.section.region: ; preds = %[[VAL_70]]
|
|
// CHECK: br label %[[VAL_81:.*]]
|
|
// CHECK: omp.section.region3: ; preds = %[[VAL_82:.*]], %[[VAL_80]]
|
|
// CHECK: %[[VAL_83:.*]] = phi i64 [ %[[VAL_84:.*]], %[[VAL_82]] ], [ 1, %[[VAL_80]] ]
|
|
// CHECK: %[[VAL_85:.*]] = icmp sgt i64 %[[VAL_83]], 0
|
|
// CHECK: br i1 %[[VAL_85]], label %[[VAL_82]], label %[[VAL_86:.*]]
|
|
// CHECK: omp.section.region5: ; preds = %[[VAL_81]]
|
|
// CHECK: br label %[[VAL_87:.*]]
|
|
// CHECK: omp.region.cont2: ; preds = %[[VAL_86]]
|
|
// CHECK: br label %[[VAL_69]]
|
|
// CHECK: omp.section.region4: ; preds = %[[VAL_81]]
|
|
// CHECK: %[[VAL_84]] = sub i64 %[[VAL_83]], 1
|
|
// CHECK: br label %[[VAL_81]]
|
|
// CHECK: omp_section_loop.body.sections.after: ; preds = %[[VAL_79]], %[[VAL_87]], %[[VAL_34]]
|
|
// CHECK: br label %[[VAL_29]]
|
|
// CHECK: omp_section_loop.inc: ; preds = %[[VAL_69]]
|
|
// CHECK: %[[VAL_31]] = add nuw i32 %[[VAL_30]], 1
|
|
// CHECK: br label %[[VAL_28]]
|
|
// CHECK: omp.par.outlined.exit.exitStub: ; preds = %[[VAL_64]]
|
|
// CHECK: ret void
|