-
Notifications
You must be signed in to change notification settings - Fork 14.3k
[flang][semantics] fix issue with equality of min/max in module files #145824
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Conversation
@llvm/pr-subscribers-flang-semantics @llvm/pr-subscribers-flang-fir-hlfir Author: Andre Kuhlenschmidt (akuhlens) ChangesConvert all binary calls of min/max to extremum operations, so that extremums generated by the compiler compare equal, and user min/max calls also compare equal. Fixes #133646 Originally opened as #144162 but I accidentally pushed a merge in such a way that a bunch of code owners got added to the review. This is just rebasing the original work on main and fixing the failing tests. Patch is 28.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/145824.diff 10 Files Affected:
diff --git a/flang/include/flang/Evaluate/tools.h b/flang/include/flang/Evaluate/tools.h
index e04621f71f9a7..d26a477ddded1 100644
--- a/flang/include/flang/Evaluate/tools.h
+++ b/flang/include/flang/Evaluate/tools.h
@@ -1510,6 +1510,16 @@ Operator OperationCode(
}
}
+template <typename T, typename... Ts>
+Operator OperationCode(
+ const evaluate::Operation<evaluate::Extremum<T>, Ts...> &op) {
+ if (op.derived().ordering == evaluate::Ordering::Greater) {
+ return Operator::Max;
+ } else {
+ return Operator::Min;
+ }
+}
+
template <typename T> Operator OperationCode(const evaluate::Constant<T> &x) {
return Operator::Constant;
}
diff --git a/flang/lib/Evaluate/fold-implementation.h b/flang/lib/Evaluate/fold-implementation.h
index b0f39e63d0941..0154833baaf17 100644
--- a/flang/lib/Evaluate/fold-implementation.h
+++ b/flang/lib/Evaluate/fold-implementation.h
@@ -1102,6 +1102,8 @@ template <typename T> Expr<T> Folder<T>::TRANSFER(FunctionRef<T> &&funcRef) {
}
}
+// TODO: Once the backend supports character extremums we could support
+// min/max with non-optional arguments to trees of extremum operations.
template <typename T>
Expr<T> FoldMINorMAX(
FoldingContext &context, FunctionRef<T> &&funcRef, Ordering order) {
@@ -1109,42 +1111,76 @@ Expr<T> FoldMINorMAX(
T::category == TypeCategory::Unsigned ||
T::category == TypeCategory::Real ||
T::category == TypeCategory::Character);
+
+ // Lots of constraints:
+ // - We want Extremum<T> generated by semantics to compare equal to
+ // Extremum<T> written out to module files as max or min calls.
+ // - Users can also write min/max calls that must also compare equal
+ // to min/max calls that wind up being written to module files.
+ // - Extremeum<T> is binary and can't currently handle processing
+ // optional arguments that may show up in 3rd + argument.
+ // - The code below only accepts more than 2 arguments if all the
+ // arguments are constant (and hence known to be present).
+ // - ConvertExprToHLFIR can't currently handle Extremum<Character>
+ // - Semantics doesn't currently generate Extremum<Character>
+ // The original code did the folding of arguments and the overall extremum
+ // operation in a single pass. This was shorter code-wise, but took me
+ // a while to tease out all the logic and was doing redundant work.
+ // So I split it into two passes:
+ // 1) fold the arguments and check if they are constant,
+ // 2) Decide if we:
+ // - can constant-fold the min/max operation, or
+ // - need to generate an extremum anyway,
+ // and do it if so.
+ // Otherwise, return the original call.
auto &args{funcRef.arguments()};
- bool ok{true};
- std::optional<Expr<T>> result;
- Folder<T> folder{context};
- for (std::optional<ActualArgument> &arg : args) {
- // Call Folding on all arguments to make operand promotion explicit.
- if (!folder.Folding(arg)) {
- // TODO: Lowering can't handle having every FunctionRef for max and min
- // being converted into Extremum<T>. That needs fixing. Until that
- // is corrected, however, it is important that max and min references
- // in module files be converted into Extremum<T> even when not constant;
- // the Extremum<SubscriptInteger> operations created to normalize the
- // values of array bounds are formatted as max operations in the
- // declarations in modules, and need to be read back in as such in
- // order for expression comparison to not produce false inequalities
- // when checking function results for procedure interface compatibility.
- if (!context.moduleFileName()) {
- ok = false;
+ std::size_t nargs{args.size()};
+ bool allArgsConstant{true};
+ bool extremumAnyway{nargs == 2 && T::category != TypeCategory::Character};
+ // 1a)Fold the first two arguments.
+ {
+ Folder<T> folder{context, /*forOptionalArgument=*/false};
+ if (!folder.Folding(args[0])) {
+ allArgsConstant = false;
+ }
+ if (!folder.Folding(args[1])) {
+ allArgsConstant = false;
+ }
+ }
+ // 1b) Fold any optional arguments.
+ if (nargs > 2) {
+ Folder<T> folder{context, /*forOptionalArgument=*/true};
+ for (std::size_t i{2}; i < nargs; ++i) {
+ if (args[i]) {
+ if (!folder.Folding(args[i])) {
+ allArgsConstant = false;
+ }
}
}
- Expr<SomeType> *argExpr{arg ? arg->UnwrapExpr() : nullptr};
- if (argExpr) {
- *argExpr = Fold(context, std::move(*argExpr));
- }
- if (Expr<T> * tExpr{UnwrapExpr<Expr<T>>(argExpr)}) {
- if (result) {
- result = FoldOperation(
- context, Extremum<T>{order, std::move(*result), Expr<T>{*tExpr}});
- } else {
- result = Expr<T>{*tExpr};
+ }
+ // 2) If we can fold the result or the call to min/max may compare equal to
+ // an extremum generated by semantics go ahead and convert to an extremum,
+ // and try to fold the result.
+ if (allArgsConstant || extremumAnyway) {
+ // Folding updates the argument expressions in place, no need to call
+ // Fold() on each argument again.
+ if (const auto *resultp{UnwrapExpr<Expr<T>>(args[0])}) {
+ Expr<T> result{*resultp};
+ for (std::size_t i{1}; i < nargs; ++i) {
+ if (const auto *tExpr{UnwrapExpr<Expr<T>>(args[i])}) {
+ result = FoldOperation(
+ context, Extremum<T>{order, std::move(result), *tExpr});
+ } else {
+ // This should never happen, but here is a value to return.
+ return Expr<T>{std::move(funcRef)};
+ }
}
- } else {
- ok = false;
+ return result;
}
}
- return ok && result ? std::move(*result) : Expr<T>{std::move(funcRef)};
+ // If we decided to not generate an extremum just return the original call,
+ // with the arguments folded.
+ return Expr<T>{std::move(funcRef)};
}
// For AMAX0, AMIN0, AMAX1, AMIN1, DMAX1, DMIN1, MAX0, MIN0, MAX1, and MIN1
diff --git a/flang/test/Lower/HLFIR/custom-intrinsic.f90 b/flang/test/Lower/HLFIR/custom-intrinsic.f90
index 161a2ab75b7c8..5ec6e0a17e9ac 100644
--- a/flang/test/Lower/HLFIR/custom-intrinsic.f90
+++ b/flang/test/Lower/HLFIR/custom-intrinsic.f90
@@ -115,10 +115,10 @@ function max_array(a, b)
! CHECK: %[[VAL_11:.*]]:2 = hlfir.declare %[[VAL_9]](%[[VAL_10]]) {uniq_name = "_QFmax_arrayEmax_array"} : (!fir.ref<!fir.array<42xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<42xi32>>, !fir.ref<!fir.array<42xi32>>)
! CHECK: %[[VAL_12:.*]] = hlfir.elemental %[[VAL_3]] unordered : (!fir.shape<1>) -> !hlfir.expr<42xi32> {
! CHECK: ^bb0(%[[VAL_13:.*]]: index):
-! CHECK: %[[VAL_14:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<42xi32>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_15:.*]] = fir.load %[[VAL_14]] : !fir.ref<i32>
-! CHECK: %[[VAL_16:.*]] = hlfir.designate %[[VAL_7]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<42xi32>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_17:.*]] = fir.load %[[VAL_16]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_14:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<42xi32>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_15:.*]] = fir.load %[[VAL_14]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_16:.*]] = hlfir.designate %[[VAL_7]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<42xi32>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_17:.*]] = fir.load %[[VAL_16]] : !fir.ref<i32>
! CHECK: %[[VAL_18:.*]] = arith.cmpi sgt, %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: %[[VAL_19:.*]] = arith.select %[[VAL_18]], %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: hlfir.yield_element %[[VAL_19]] : i32
@@ -288,10 +288,10 @@ function min_array(a, b)
! CHECK: %[[VAL_11:.*]]:2 = hlfir.declare %[[VAL_9]](%[[VAL_10]]) {uniq_name = "_QFmin_arrayEmin_array"} : (!fir.ref<!fir.array<42xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<42xi32>>, !fir.ref<!fir.array<42xi32>>)
! CHECK: %[[VAL_12:.*]] = hlfir.elemental %[[VAL_3]] unordered : (!fir.shape<1>) -> !hlfir.expr<42xi32> {
! CHECK: ^bb0(%[[VAL_13:.*]]: index):
-! CHECK: %[[VAL_14:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<42xi32>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_15:.*]] = fir.load %[[VAL_14]] : !fir.ref<i32>
-! CHECK: %[[VAL_16:.*]] = hlfir.designate %[[VAL_7]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<42xi32>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_17:.*]] = fir.load %[[VAL_16]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_14:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<42xi32>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_15:.*]] = fir.load %[[VAL_14]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_16:.*]] = hlfir.designate %[[VAL_7]]#0 (%[[VAL_13]]) : (!fir.ref<!fir.array<42xi32>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_17:.*]] = fir.load %[[VAL_16]] : !fir.ref<i32>
! CHECK: %[[VAL_18:.*]] = arith.cmpi slt, %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: %[[VAL_19:.*]] = arith.select %[[VAL_18]], %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: hlfir.yield_element %[[VAL_19]] : i32
diff --git a/flang/test/Lower/OpenMP/reduction-array-intrinsic.f90 b/flang/test/Lower/OpenMP/reduction-array-intrinsic.f90
index 8b4f37278185e..0cf88cf889868 100644
--- a/flang/test/Lower/OpenMP/reduction-array-intrinsic.f90
+++ b/flang/test/Lower/OpenMP/reduction-array-intrinsic.f90
@@ -82,10 +82,10 @@ subroutine max_array_reduction(l, r)
! CHECK: %[[VAL_16:.*]] = arith.constant 1 : index
! CHECK: %[[VAL_17:.*]] = arith.subi %[[VAL_15]]#0, %[[VAL_16]] : index
! CHECK: %[[VAL_18:.*]] = arith.addi %[[VAL_13]], %[[VAL_17]] : index
-! CHECK: %[[VAL_19:.*]] = hlfir.designate %[[VAL_8]] (%[[VAL_18]]) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_20:.*]] = fir.load %[[VAL_19]] : !fir.ref<i32>
-! CHECK: %[[VAL_21:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_13]]) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_22:.*]] = fir.load %[[VAL_21]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_19:.*]] = hlfir.designate %[[VAL_8]] (%[[VAL_18]]) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_20:.*]] = fir.load %[[VAL_19]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_21:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_13]]) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_22:.*]] = fir.load %[[VAL_21]] : !fir.ref<i32>
! CHECK: %[[VAL_23:.*]] = arith.cmpi sgt, %[[VAL_20]], %[[VAL_22]] : i32
! CHECK: %[[VAL_24:.*]] = arith.select %[[VAL_23]], %[[VAL_20]], %[[VAL_22]] : i32
! CHECK: hlfir.yield_element %[[VAL_24]] : i32
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-allocatable-array-minmax.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-allocatable-array-minmax.f90
index 5b4c5e65ffccc..58b68e5ec4cfd 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-allocatable-array-minmax.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-allocatable-array-minmax.f90
@@ -228,11 +228,11 @@ program reduce15
! CHECK: %[[VAL_56:.*]]:2 = hlfir.declare %[[VAL_55]] {uniq_name = "_QFEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: %[[VAL_62:.*]]:2 = hlfir.declare %[[VAL_60]] {fortran_attrs = {{.*}}<allocatable>, uniq_name = "_QFEmaxes"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>)
! CHECK: hlfir.assign %[[VAL_61]] to %[[VAL_56]]#0 : i32, !fir.ref<i32>
-! CHECK: %[[VAL_63:.*]] = fir.load %[[VAL_1]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
-! CHECK: %[[VAL_64:.*]] = arith.constant 0 : index
-! CHECK: %[[VAL_65:.*]]:3 = fir.box_dims %[[VAL_63]], %[[VAL_64]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
-! CHECK: %[[VAL_66:.*]] = fir.shape %[[VAL_65]]#1 : (index) -> !fir.shape<1>
-! CHECK: %[[VAL_67:.*]] = fir.load %[[VAL_62]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK-DAG: %[[VAL_63:.*]] = fir.load %[[VAL_1]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK-DAG: %[[VAL_64:.*]] = arith.constant 0 : index
+! CHECK-DAG: %[[VAL_65:.*]]:3 = fir.box_dims %[[VAL_63]], %[[VAL_64]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
+! CHECK-DAG: %[[VAL_66:.*]] = fir.shape %[[VAL_65]]#1 : (index) -> !fir.shape<1>
+! CHECK-DAG: %[[VAL_67:.*]] = fir.load %[[VAL_62]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
! CHECK: %[[VAL_68:.*]] = hlfir.elemental %[[VAL_66]] unordered : (!fir.shape<1>) -> !hlfir.expr<?xi32> {
! CHECK: ^bb0(%[[VAL_69:.*]]: index):
! CHECK: %[[VAL_70:.*]] = arith.constant 0 : index
@@ -240,15 +240,15 @@ program reduce15
! CHECK: %[[VAL_72:.*]] = arith.constant 1 : index
! CHECK: %[[VAL_73:.*]] = arith.subi %[[VAL_71]]#0, %[[VAL_72]] : index
! CHECK: %[[VAL_74:.*]] = arith.addi %[[VAL_69]], %[[VAL_73]] : index
-! CHECK: %[[VAL_75:.*]] = hlfir.designate %[[VAL_63]] (%[[VAL_74]]) : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_76:.*]] = fir.load %[[VAL_75]] : !fir.ref<i32>
-! CHECK: %[[VAL_77:.*]] = arith.constant 0 : index
-! CHECK: %[[VAL_78:.*]]:3 = fir.box_dims %[[VAL_67]], %[[VAL_77]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
-! CHECK: %[[VAL_79:.*]] = arith.constant 1 : index
-! CHECK: %[[VAL_80:.*]] = arith.subi %[[VAL_78]]#0, %[[VAL_79]] : index
-! CHECK: %[[VAL_81:.*]] = arith.addi %[[VAL_69]], %[[VAL_80]] : index
-! CHECK: %[[VAL_82:.*]] = hlfir.designate %[[VAL_67]] (%[[VAL_81]]) : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_83:.*]] = fir.load %[[VAL_82]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_75:.*]] = hlfir.designate %[[VAL_63]] (%[[VAL_74]]) : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_76:.*]] = fir.load %[[VAL_75]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_77:.*]] = arith.constant 0 : index
+! CHECK-DAG: %[[VAL_78:.*]]:3 = fir.box_dims %[[VAL_67]], %[[VAL_77]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
+! CHECK-DAG: %[[VAL_79:.*]] = arith.constant 1 : index
+! CHECK-DAG: %[[VAL_80:.*]] = arith.subi %[[VAL_78]]#0, %[[VAL_79]] : index
+! CHECK-DAG: %[[VAL_81:.*]] = arith.addi %[[VAL_69]], %[[VAL_80]] : index
+! CHECK-DAG: %[[VAL_82:.*]] = hlfir.designate %[[VAL_67]] (%[[VAL_81]]) : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_83:.*]] = fir.load %[[VAL_82]] : !fir.ref<i32>
! CHECK: %[[VAL_84:.*]] = arith.cmpi sgt, %[[VAL_76]], %[[VAL_83]] : i32
! CHECK: %[[VAL_85:.*]] = arith.select %[[VAL_84]], %[[VAL_76]], %[[VAL_83]] : i32
! CHECK: hlfir.yield_element %[[VAL_85]] : i32
@@ -269,27 +269,27 @@ program reduce15
! CHECK: %[[VAL_88:.*]]:2 = hlfir.declare %[[VAL_87]] {uniq_name = "_QFEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: %[[VAL_94:.*]]:2 = hlfir.declare %[[VAL_92]] {fortran_attrs = {{.*}}<allocatable>, uniq_name = "_QFEmins"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>)
! CHECK: hlfir.assign %[[VAL_93]] to %[[VAL_88]]#0 : i32, !fir.ref<i32>
-! CHECK: %[[VAL_95:.*]] = fir.load %[[VAL_1]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
-! CHECK: %[[VAL_96:.*]] = arith.constant 0 : index
-! CHECK: %[[VAL_97:.*]]:3 = fir.box_dims %[[VAL_95]], %[[VAL_96]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
-! CHECK: %[[VAL_98:.*]] = fir.shape %[[VAL_97]]#1 : (index) -> !fir.shape<1>
-! CHECK: %[[VAL_99:.*]] = fir.load %[[VAL_94]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
-! CHECK: %[[VAL_100:.*]] = hlfir.elemental %[[VAL_98]] unordered : (!fir.shape<1>) -> !hlfir.expr<?xi32> {
+! CHECK-DAG: %[[VAL_95:.*]] = fir.load %[[VAL_1]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK-DAG: %[[VAL_96:.*]] = arith.constant 0 : index
+! CHECK-DAG: %[[VAL_97:.*]]:3 = fir.box_dims %[[VAL_95]], %[[VAL_96]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
+! CHECK-DAG: %[[VAL_98:.*]] = fir.shape %[[VAL_97]]#1 : (index) -> !fir.shape<1>
+! CHECK-DAG: %[[VAL_99:.*]] = fir.load %[[VAL_94]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK-DAG: %[[VAL_100:.*]] = hlfir.elemental %[[VAL_98]] unordered : (!fir.shape<1>) -> !hlfir.expr<?xi32> {
! CHECK: ^bb0(%[[VAL_101:.*]]: index):
! CHECK: %[[VAL_102:.*]] = arith.constant 0 : index
! CHECK: %[[VAL_103:.*]]:3 = fir.box_dims %[[VAL_95]], %[[VAL_102]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
! CHECK: %[[VAL_104:.*]] = arith.constant 1 : index
! CHECK: %[[VAL_105:.*]] = arith.subi %[[VAL_103]]#0, %[[VAL_104]] : index
! CHECK: %[[VAL_106:.*]] = arith.addi %[[VAL_101]], %[[VAL_105]] : index
-! CHECK: %[[VAL_107:.*]] = hlfir.designate %[[VAL_95]] (%[[VAL_106]]) : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_108:.*]] = fir.load %[[VAL_107]] : !fir.ref<i32>
-! CHECK: %[[VAL_109:.*]] = arith.constant 0 : index
-! CHECK: %[[VAL_110:.*]]:3 = fir.box_dims %[[VAL_99]], %[[VAL_109]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
-! CHECK: %[[VAL_111:.*]] = arith.constant 1 : index
-! CHECK: %[[VAL_112:.*]] = arith.subi %[[VAL_110]]#0, %[[VAL_111]] : index
-! CHECK: %[[VAL_113:.*]] = arith.addi %[[VAL_101]], %[[VAL_112]] : index
-! CHECK: %[[VAL_114:.*]] = hlfir.designate %[[VAL_99]] (%[[VAL_113]]) : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
-! CHECK: %[[VAL_115:.*]] = fir.load %[[VAL_114]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_107:.*]] = hlfir.designate %[[VAL_95]] (%[[VAL_106]]) : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_108:.*]] = fir.load %[[VAL_107]] : !fir.ref<i32>
+! CHECK-DAG: %[[VAL_109:.*]] = arith.constant 0 : index
+! CHECK-DAG: %[[VAL_110:.*]]:3 = fir.box_dims %[[VAL_99]], %[[VAL_109]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
+! CHECK-DAG: %[[VAL_111:.*]] = arith.constant 1 : index
+! CHECK-DAG: %[[VAL_112:.*]] = arith.subi %[[VAL_110]]#0, %[[VAL_111]] : index
+! CHECK-DAG: %[[VAL_113:.*]] = arith.addi %[[VAL_101]], %[[VAL_112]] : index
+! CHECK-DAG: %[[VAL_114:.*]] = hlfir.designate %[[VAL_99]] (%[[VAL_113]]) : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
+! CHECK-DAG: %[[VAL_115:.*]] = fir.load %[[VAL_114]] : !fir.ref<i32>
! CHECK: %[[VAL_116:.*]] = arith.cmpi slt, %[[VAL_108]], %[[VAL_115]] : i32
! CHECK: %[[VAL_117:.*]] = arith.select %[[VAL_116]], %[[VAL_108]], %[[VAL_115]] : i32
...
[truncated]
|
Convert all binary calls of min/max to extremum operations, so that extremums generated by the compiler compare equal, and user min/max calls also compare equal.
Fixes #133646
Originally opened as #144162 but I accidentally pushed a merge in such a way that a bunch of code owners got added to the review. This is just rebasing the original work on main and fixing the failing tests.