[flang] Use integer arith.max/min operations for max/min lowering. - #186466
Merged
Conversation
arith.maxsi/maxui/minsi/minui are more concise than cmp+select and probably allow more folding, so we should use it in Flang lowering.
llvmbot
added
flang
Member
|
@llvm/pr-subscribers-flang-openmp @llvm/pr-subscribers-openacc Author: Slava Zakharin (vzakhari) Changesarith.maxsi/maxui/minsi/minui are more concise than cmp+select Patch is 44.49 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/186466.diff 21 Files Affected:
diff --git a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
index d67eebdd8c93c..f421bddebea8a 100644
--- a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
+++ b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
@@ -8736,12 +8736,6 @@ static mlir::Value genExtremumResult(mlir::Location loc,
fir::FirOpBuilder &builder,
mlir::Value left, mlir::Value right) {
mlir::Type type = left.getType();
- mlir::arith::CmpIPredicate integerPredicate =
- type.isUnsignedInteger() ? isMax ? mlir::arith::CmpIPredicate::ugt
- : mlir::arith::CmpIPredicate::ult
- : isMax ? mlir::arith::CmpIPredicate::sgt
- : mlir::arith::CmpIPredicate::slt;
- mlir::Value pred;
if (fir::isa_real(type)) {
switch (builder.getFPMaxminBehavior()) {
case Fortran::common::FPMaxminBehavior::Portable:
@@ -8782,26 +8776,31 @@ static mlir::Value genExtremumResult(mlir::Location loc,
llvm_unreachable("unsupported FPMaxminBehavior");
} else if (fir::isa_integer(type)) {
- mlir::Value cmpLeft = left;
- mlir::Value cmpRight = right;
+ // It is probably okay to use signed index.maxs/mins, but
+ // maybe the caller needs to specify signedness.
+ // There are currently no callers that pass values of index
+ // type, so just emit a TODO.
+ if (mlir::isa<mlir::IndexType>(type))
+ TODO(loc, "extremum for index type");
+
if (type.isUnsignedInteger()) {
- mlir::Type signlessType = mlir::IntegerType::get(
- builder.getContext(), type.getIntOrFloatBitWidth(),
- mlir::IntegerType::SignednessSemantics::Signless);
- cmpLeft = builder.createConvert(loc, signlessType, left);
- cmpRight = builder.createConvert(loc, signlessType, right);
+ if constexpr (isMax)
+ return mlir::arith::MaxUIOp::create(builder, loc, left, right);
+ else
+ return mlir::arith::MinUIOp::create(builder, loc, left, right);
+ } else {
+ if constexpr (isMax)
+ return mlir::arith::MaxSIOp::create(builder, loc, left, right);
+ else
+ return mlir::arith::MinSIOp::create(builder, loc, left, right);
}
- pred = mlir::arith::CmpIOp::create(builder, loc, integerPredicate, cmpLeft,
- cmpRight);
} else if (fir::isa_char(type) || fir::isa_char(fir::unwrapRefType(type))) {
// TODO: ! character min and max is tricky because the result
// length is the length of the longest argument!
// So we may need a temp.
TODO(loc, "intrinsic: min and max for CHARACTER");
}
- assert(pred && "pred must be defined");
-
- return mlir::arith::SelectOp::create(builder, loc, pred, left, right);
+ llvm_unreachable("unsupported extremum");
}
// UNLINK
diff --git a/flang/test/Lower/HLFIR/array-ctor-as-elemental.f90 b/flang/test/Lower/HLFIR/array-ctor-as-elemental.f90
index 5c4f079c86e20..f52216d2f558d 100644
--- a/flang/test/Lower/HLFIR/array-ctor-as-elemental.f90
+++ b/flang/test/Lower/HLFIR/array-ctor-as-elemental.f90
@@ -54,8 +54,7 @@ subroutine test_as_strided_elemental(lb, ub, stride)
! CHECK: %[[VAL_12:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i64>
! CHECK: %[[VAL_13:.*]] = arith.divsi %[[VAL_11]], %[[VAL_12]] : i64
! CHECK: %[[VAL_14:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_15:.*]] = arith.cmpi sgt, %[[VAL_13]], %[[VAL_14]] : i64
-! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_15]], %[[VAL_13]], %[[VAL_14]] : i64
+! CHECK: %[[VAL_16:.*]] = arith.maxsi %[[VAL_13]], %[[VAL_14]] : i64
! CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_6]], %[[VAL_16]] : i64
! CHECK: %[[VAL_18:.*]] = fir.convert %[[VAL_17]] : (i64) -> index
! CHECK: %[[VAL_19:.*]] = fir.shape %[[VAL_18]] : (index) -> !fir.shape<1>
diff --git a/flang/test/Lower/HLFIR/array-ctor-as-inlined-temp.f90 b/flang/test/Lower/HLFIR/array-ctor-as-inlined-temp.f90
index 415a9df5bf838..6fcd942f4a1e5 100644
--- a/flang/test/Lower/HLFIR/array-ctor-as-inlined-temp.f90
+++ b/flang/test/Lower/HLFIR/array-ctor-as-inlined-temp.f90
@@ -127,8 +127,7 @@ subroutine test_implied_do(n)
! CHECK: %[[VAL_10:.*]] = arith.constant 1 : i64
! CHECK: %[[VAL_11:.*]] = arith.divsi %[[VAL_9]], %[[VAL_10]] : i64
! CHECK: %[[VAL_12:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_13:.*]] = arith.cmpi sgt, %[[VAL_11]], %[[VAL_12]] : i64
-! CHECK: %[[VAL_14:.*]] = arith.select %[[VAL_13]], %[[VAL_11]], %[[VAL_12]] : i64
+! CHECK: %[[VAL_14:.*]] = arith.maxsi %[[VAL_11]], %[[VAL_12]] : i64
! CHECK: %[[VAL_15:.*]] = arith.muli %[[VAL_4]], %[[VAL_14]] : i64
! CHECK: %[[VAL_16:.*]] = arith.addi %[[VAL_3]], %[[VAL_15]] : i64
! CHECK: %[[VAL_17:.*]] = fir.convert %[[VAL_16]] : (i64) -> index
@@ -191,8 +190,7 @@ subroutine test_strided_implied_do(lb, ub, stride)
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i64>
! CHECK: %[[VAL_15:.*]] = arith.divsi %[[VAL_13]], %[[VAL_14]] : i64
! CHECK: %[[VAL_16:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_17:.*]] = arith.cmpi sgt, %[[VAL_15]], %[[VAL_16]] : i64
-! CHECK: %[[VAL_18:.*]] = arith.select %[[VAL_17]], %[[VAL_15]], %[[VAL_16]] : i64
+! CHECK: %[[VAL_18:.*]] = arith.maxsi %[[VAL_15]], %[[VAL_16]] : i64
! CHECK: %[[VAL_19:.*]] = arith.muli %[[VAL_8]], %[[VAL_18]] : i64
! CHECK: %[[VAL_20:.*]] = arith.addi %[[VAL_7]], %[[VAL_19]] : i64
! CHECK: %[[VAL_21:.*]] = fir.convert %[[VAL_20]] : (i64) -> index
@@ -253,8 +251,7 @@ subroutine test_nested_implied_do(n, m)
! CHECK: %[[VAL_12:.*]] = arith.constant 1 : i64
! CHECK: %[[VAL_13:.*]] = arith.divsi %[[VAL_11]], %[[VAL_12]] : i64
! CHECK: %[[VAL_14:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_15:.*]] = arith.cmpi sgt, %[[VAL_13]], %[[VAL_14]] : i64
-! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_15]], %[[VAL_13]], %[[VAL_14]] : i64
+! CHECK: %[[VAL_16:.*]] = arith.maxsi %[[VAL_13]], %[[VAL_14]] : i64
! CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_6]], %[[VAL_16]] : i64
! CHECK: %[[VAL_18:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i64>
! CHECK: %[[VAL_19:.*]] = arith.constant 1 : i64
@@ -264,8 +261,7 @@ subroutine test_nested_implied_do(n, m)
! CHECK: %[[VAL_23:.*]] = arith.constant 1 : i64
! CHECK: %[[VAL_24:.*]] = arith.divsi %[[VAL_22]], %[[VAL_23]] : i64
! CHECK: %[[VAL_25:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_26:.*]] = arith.cmpi sgt, %[[VAL_24]], %[[VAL_25]] : i64
-! CHECK: %[[VAL_27:.*]] = arith.select %[[VAL_26]], %[[VAL_24]], %[[VAL_25]] : i64
+! CHECK: %[[VAL_27:.*]] = arith.maxsi %[[VAL_24]], %[[VAL_25]] : i64
! CHECK: %[[VAL_28:.*]] = arith.muli %[[VAL_17]], %[[VAL_27]] : i64
! CHECK: %[[VAL_29:.*]] = arith.addi %[[VAL_5]], %[[VAL_28]] : i64
! CHECK: %[[VAL_30:.*]] = fir.convert %[[VAL_29]] : (i64) -> index
diff --git a/flang/test/Lower/HLFIR/binary-ops.f90 b/flang/test/Lower/HLFIR/binary-ops.f90
index f4e1643dd78b6..3a83cd44d3c2e 100644
--- a/flang/test/Lower/HLFIR/binary-ops.f90
+++ b/flang/test/Lower/HLFIR/binary-ops.f90
@@ -206,8 +206,7 @@ subroutine extremum(c, n, l)
! CHECK: hlfir.declare {{.*}}c"}
! CHECK: %[[VAL_11:.*]] = arith.constant 0 : i64
! CHECK: %[[VAL_12:.*]] = fir.load %{{.*}} : !fir.ref<i64>
-! CHECK: %[[VAL_13:.*]] = arith.cmpi sgt, %[[VAL_11]], %[[VAL_12]] : i64
-! CHECK: arith.select %[[VAL_13]], %[[VAL_11]], %[[VAL_12]] : i64
+! CHECK: arith.maxsi %[[VAL_11]], %[[VAL_12]] : i64
subroutine cmp_int(l, x, y)
logical :: l
diff --git a/flang/test/Lower/HLFIR/custom-intrinsic.f90 b/flang/test/Lower/HLFIR/custom-intrinsic.f90
index 4999eebf376e7..294077105ed7b 100644
--- a/flang/test/Lower/HLFIR/custom-intrinsic.f90
+++ b/flang/test/Lower/HLFIR/custom-intrinsic.f90
@@ -14,8 +14,7 @@ function max_simple(a, b)
! CHECK-NEXT: %[[RES_DECL:.*]]:2 = hlfir.declare %[[RES_ALLOC]] {uniq_name = "_QFmax_simpleEmax_simple"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK-NEXT: %[[A_LD:.*]] = fir.load %[[A_DECL]]#0 : !fir.ref<i32>
! CHECK-NEXT: %[[B_LD:.*]] = fir.load %[[B_DECL]]#0 : !fir.ref<i32>
-! CHECK-NEXT: %[[A_GT_B:.*]] = arith.cmpi sgt, %[[A_LD]], %[[B_LD]] : i32
-! CHECK-NEXT: %[[SELECT:.*]] = arith.select %[[A_GT_B]], %[[A_LD]], %[[B_LD]] : i32
+! CHECK-NEXT: %[[SELECT:.*]] = arith.maxsi %[[A_LD]], %[[B_LD]] : i32
! CHECK-NEXT: hlfir.assign %[[SELECT]] to %[[RES_DECL]]#0 : i32, !fir.ref<i32>
! CHECK-NEXT: %[[RES_LD:.*]] = fir.load %[[RES_DECL]]#0 : !fir.ref<i32>
! CHECK-NEXT: return %[[RES_LD]] : i32
@@ -38,12 +37,10 @@ function max_dynamic_optional_scalar(a, b, c)
! CHECK: %[[VAL_8:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_9:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_10:.*]] = fir.is_present %[[VAL_5]]#0 : (!fir.ref<i32>) -> i1
-! CHECK: %[[VAL_11:.*]] = arith.cmpi sgt, %[[VAL_8]], %[[VAL_9]] : i32
-! CHECK: %[[VAL_12:.*]] = arith.select %[[VAL_11]], %[[VAL_8]], %[[VAL_9]] : i32
+! CHECK: %[[VAL_12:.*]] = arith.maxsi %[[VAL_8]], %[[VAL_9]] : i32
! CHECK: %[[VAL_13:.*]] = fir.if %[[VAL_10]] -> (i32) {
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_15:.*]] = arith.cmpi sgt, %[[VAL_12]], %[[VAL_14]] : i32
-! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_15]], %[[VAL_12]], %[[VAL_14]] : i32
+! CHECK: %[[VAL_16:.*]] = arith.maxsi %[[VAL_12]], %[[VAL_14]] : i32
! CHECK: fir.result %[[VAL_16]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_12]] : i32
@@ -73,20 +70,17 @@ function max_dynamic_optional_scalar2(a, b, c, d)
! CHECK: %[[VAL_11:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_12:.*]] = fir.is_present %[[VAL_6]]#0 : (!fir.ref<i32>) -> i1
! CHECK: %[[VAL_13:.*]] = fir.is_present %[[VAL_7]]#0 : (!fir.ref<i32>) -> i1
-! CHECK: %[[VAL_14:.*]] = arith.cmpi sgt, %[[VAL_10]], %[[VAL_11]] : i32
-! CHECK: %[[VAL_15:.*]] = arith.select %[[VAL_14]], %[[VAL_10]], %[[VAL_11]] : i32
+! CHECK: %[[VAL_15:.*]] = arith.maxsi %[[VAL_10]], %[[VAL_11]] : i32
! CHECK: %[[VAL_16:.*]] = fir.if %[[VAL_12]] -> (i32) {
! CHECK: %[[VAL_17:.*]] = fir.load %[[VAL_6]]#0 : !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: %[[VAL_19:.*]] = arith.maxsi %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: fir.result %[[VAL_19]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_15]] : i32
! CHECK: }
! CHECK: %[[VAL_20:.*]] = fir.if %[[VAL_13]] -> (i32) {
! CHECK: %[[VAL_21:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_22:.*]] = arith.cmpi sgt, %[[VAL_23:.*]], %[[VAL_21]] : i32
-! CHECK: %[[VAL_24:.*]] = arith.select %[[VAL_22]], %[[VAL_23]], %[[VAL_21]] : i32
+! CHECK: %[[VAL_24:.*]] = arith.maxsi %[[VAL_23:.*]], %[[VAL_21]] : i32
! CHECK: fir.result %[[VAL_24]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_25:.*]] : i32
@@ -119,8 +113,7 @@ function max_array(a, b)
! 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: %[[VAL_19:.*]] = arith.maxsi %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: hlfir.yield_element %[[VAL_19]] : i32
! CHECK: }
! CHECK: hlfir.assign %[[VAL_20:.*]] to %[[VAL_11]]#0 : !hlfir.expr<42xi32>, !fir.ref<!fir.array<42xi32>>
@@ -155,13 +148,11 @@ function max_dynamic_optional_array(a, b, c)
! CHECK: ^bb0(%[[VAL_17:.*]]: index):
! CHECK: %[[VAL_18:.*]] = hlfir.designate %[[VAL_6]]#0 (%[[VAL_17]]) : (!fir.ref<!fir.array<10xi32>>, index) -> !fir.ref<i32>
! CHECK: %[[VAL_19:.*]] = fir.load %[[VAL_18]] : !fir.ref<i32>
-! CHECK: %[[VAL_20:.*]] = arith.cmpi sgt, %[[VAL_15]], %[[VAL_19]] : i32
-! CHECK: %[[VAL_21:.*]] = arith.select %[[VAL_20]], %[[VAL_15]], %[[VAL_19]] : i32
+! CHECK: %[[VAL_21:.*]] = arith.maxsi %[[VAL_15]], %[[VAL_19]] : i32
! CHECK: %[[VAL_22:.*]] = fir.if %[[VAL_14]] -> (i32) {
! CHECK: %[[VAL_23:.*]] = hlfir.designate %[[VAL_9]]#0 (%[[VAL_17]]) : (!fir.ref<!fir.array<10xi32>>, index) -> !fir.ref<i32>
! CHECK: %[[VAL_24:.*]] = fir.load %[[VAL_23]] : !fir.ref<i32>
-! CHECK: %[[VAL_25:.*]] = arith.cmpi sgt, %[[VAL_21]], %[[VAL_24]] : i32
-! CHECK: %[[VAL_26:.*]] = arith.select %[[VAL_25]], %[[VAL_21]], %[[VAL_24]] : i32
+! CHECK: %[[VAL_26:.*]] = arith.maxsi %[[VAL_21]], %[[VAL_24]] : i32
! CHECK: fir.result %[[VAL_26]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_21]] : i32
@@ -187,8 +178,7 @@ function min_simple(a, b)
! CHECK: %[[VAL_5:.*]]:2 = hlfir.declare %[[VAL_4]] {uniq_name = "_QFmin_simpleEmin_simple"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: %[[VAL_6:.*]] = fir.load %[[VAL_2]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_7:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_8:.*]] = arith.cmpi slt, %[[VAL_6]], %[[VAL_7]] : i32
-! CHECK: %[[VAL_9:.*]] = arith.select %[[VAL_8]], %[[VAL_6]], %[[VAL_7]] : i32
+! CHECK: %[[VAL_9:.*]] = arith.minsi %[[VAL_6]], %[[VAL_7]] : i32
! CHECK: hlfir.assign %[[VAL_9]] to %[[VAL_5]]#0 : i32, !fir.ref<i32>
! CHECK: %[[VAL_10:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
! CHECK: return %[[VAL_10]] : i32
@@ -211,12 +201,10 @@ function min_dynamic_optional_scalar(a, b, c)
! CHECK: %[[VAL_8:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_9:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_10:.*]] = fir.is_present %[[VAL_5]]#0 : (!fir.ref<i32>) -> i1
-! CHECK: %[[VAL_11:.*]] = arith.cmpi slt, %[[VAL_8]], %[[VAL_9]] : i32
-! CHECK: %[[VAL_12:.*]] = arith.select %[[VAL_11]], %[[VAL_8]], %[[VAL_9]] : i32
+! CHECK: %[[VAL_12:.*]] = arith.minsi %[[VAL_8]], %[[VAL_9]] : i32
! CHECK: %[[VAL_13:.*]] = fir.if %[[VAL_10]] -> (i32) {
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_15:.*]] = arith.cmpi slt, %[[VAL_12]], %[[VAL_14]] : i32
-! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_15]], %[[VAL_12]], %[[VAL_14]] : i32
+! CHECK: %[[VAL_16:.*]] = arith.minsi %[[VAL_12]], %[[VAL_14]] : i32
! CHECK: fir.result %[[VAL_16]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_12]] : i32
@@ -246,20 +234,17 @@ function min_dynamic_optional_scalar2(a, b, c, d)
! CHECK: %[[VAL_11:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_12:.*]] = fir.is_present %[[VAL_6]]#0 : (!fir.ref<i32>) -> i1
! CHECK: %[[VAL_13:.*]] = fir.is_present %[[VAL_7]]#0 : (!fir.ref<i32>) -> i1
-! CHECK: %[[VAL_14:.*]] = arith.cmpi slt, %[[VAL_10]], %[[VAL_11]] : i32
-! CHECK: %[[VAL_15:.*]] = arith.select %[[VAL_14]], %[[VAL_10]], %[[VAL_11]] : i32
+! CHECK: %[[VAL_15:.*]] = arith.minsi %[[VAL_10]], %[[VAL_11]] : i32
! CHECK: %[[VAL_16:.*]] = fir.if %[[VAL_12]] -> (i32) {
! CHECK: %[[VAL_17:.*]] = fir.load %[[VAL_6]]#0 : !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: %[[VAL_19:.*]] = arith.minsi %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: fir.result %[[VAL_19]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_15]] : i32
! CHECK: }
! CHECK: %[[VAL_20:.*]] = fir.if %[[VAL_13]] -> (i32) {
! CHECK: %[[VAL_21:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_22:.*]] = arith.cmpi slt, %[[VAL_23:.*]], %[[VAL_21]] : i32
-! CHECK: %[[VAL_24:.*]] = arith.select %[[VAL_22]], %[[VAL_23]], %[[VAL_21]] : i32
+! CHECK: %[[VAL_24:.*]] = arith.minsi %[[VAL_23:.*]], %[[VAL_21]] : i32
! CHECK: fir.result %[[VAL_24]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_25:.*]] : i32
@@ -292,8 +277,7 @@ function min_array(a, b)
! 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: %[[VAL_19:.*]] = arith.minsi %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: hlfir.yield_element %[[VAL_19]] : i32
! CHECK: }
! CHECK: hlfir.assign %[[VAL_20:.*]] to %[[VAL_11]]#0 : !hlfir.expr<42xi32>, !fir.ref<!fir.array<42xi32>>
@@ -328,13 +312,11 @@ function min_dynamic_optional_array(a, b, c)
! CHECK: ^bb0(%[[VAL_17:.*]]: index):
! CHECK: %[[VAL_18:.*]] = hlfir.designate %[[VAL_6]]#0 (%[[VAL_17]]) : (!fir.ref<!fir.array<10xi32>>, index) -> !fir.ref<i32>
! CHECK: %[[VAL_19:.*]] = fir.load %[[VAL_18]] : !fir.ref<i32>
-! CHECK: %[[VAL_20:.*]] = arith.cmpi slt, %[[VAL_15]], %[[VAL_19]] : i32
-! CHECK: %[[VAL_21:.*]] = arith.select %[[VAL_20]], %[[VAL_15]], %[[VAL_19]] : i32
+! CHECK: %[[VAL_21:.*]] = arith.minsi %[[VAL_15]], %[[VAL_19]] : i32
! CHECK: %[[VAL_22:.*]] = fir.if %[[VAL_14]] -> (i32) {
! CHECK: %[[VAL_23:.*]] = hlfir.designate %[[VAL_9]]#0 (%[[VAL_17]]) : (!fir.ref<!fir.array<10xi32>>, index) -> !fir.ref<i32>
! CHECK: %[[VAL_24:.*]] = fir.load %[[VAL_23]] : !fir.ref<i32>
-! CHECK: %[[VAL_25:.*]] = arith.cmpi slt, %[[VAL_21]], %[[VAL_24]] : i32
-! CHECK: %[[VAL_26:.*]] = arith.select %[[VAL_25]], %[[VAL_21]], %[[VAL_24]] : i32
+! CHECK: %[[VAL_26:.*]] = arith.minsi %[[VAL_21]], %[[VAL_24]] : i32
! CHECK: fir.result %[[VAL_26]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_21]] : i32
@@ -812,8 +794,7 @@ subroutine allocatables_test(a, b, c)
! CHECK: %[[VAL_112:.*]] = arith.addi %[[VAL_84]], %[[VAL_111]] : index
! CHECK: %[[VAL_113:.*]] = hlfir.designate %[[VAL_79]] (%[[VAL_108]], %[[VAL_110]], %[[VAL_112]]) : (!fir.box<!fir.heap<!fir.array<?x?x?xi32>>>, index, index, index) -> !fir.ref<i32>
! CHECK: %[[VAL_114:.*]] = fir.load %[[VAL_113]] : !fir.ref<i32>
-! CHECK: %[[VAL_115:.*]] = arith.cmpi slt, %[[VAL_99]], %[[VAL_114]] : i32
-! CHECK: %[[VAL_116:.*]] = arith.select %[[VAL_115]], %[[VAL_99]], %[[VAL_114]] : i32
+! CHECK: %[[VAL_116:.*]] = arith.minsi %[[VAL_99]], %[[VAL_114]] : i32
! CHECK: %[[VAL_117:.*]] = fir.if %[[VAL_70]] -> (i32) {
! CHECK: %[[VAL_118:.*]] = arith.constant 0 : index
! CHECK: %[[VAL_119:.*]]:3 = fir.box_dims %[[VAL_80]], %[[VAL_118]] : (!fir.box<!fir.heap<!fir.array<?x?x?xi32>>>, inde...
[truncated]
|
Member
|
@llvm/pr-subscribers-flang-fir-hlfir Author: Slava Zakharin (vzakhari) Changesarith.maxsi/maxui/minsi/minui are more concise than cmp+select Patch is 44.49 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/186466.diff 21 Files Affected:
diff --git a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
index d67eebdd8c93c..f421bddebea8a 100644
--- a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
+++ b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
@@ -8736,12 +8736,6 @@ static mlir::Value genExtremumResult(mlir::Location loc,
fir::FirOpBuilder &builder,
mlir::Value left, mlir::Value right) {
mlir::Type type = left.getType();
- mlir::arith::CmpIPredicate integerPredicate =
- type.isUnsignedInteger() ? isMax ? mlir::arith::CmpIPredicate::ugt
- : mlir::arith::CmpIPredicate::ult
- : isMax ? mlir::arith::CmpIPredicate::sgt
- : mlir::arith::CmpIPredicate::slt;
- mlir::Value pred;
if (fir::isa_real(type)) {
switch (builder.getFPMaxminBehavior()) {
case Fortran::common::FPMaxminBehavior::Portable:
@@ -8782,26 +8776,31 @@ static mlir::Value genExtremumResult(mlir::Location loc,
llvm_unreachable("unsupported FPMaxminBehavior");
} else if (fir::isa_integer(type)) {
- mlir::Value cmpLeft = left;
- mlir::Value cmpRight = right;
+ // It is probably okay to use signed index.maxs/mins, but
+ // maybe the caller needs to specify signedness.
+ // There are currently no callers that pass values of index
+ // type, so just emit a TODO.
+ if (mlir::isa<mlir::IndexType>(type))
+ TODO(loc, "extremum for index type");
+
if (type.isUnsignedInteger()) {
- mlir::Type signlessType = mlir::IntegerType::get(
- builder.getContext(), type.getIntOrFloatBitWidth(),
- mlir::IntegerType::SignednessSemantics::Signless);
- cmpLeft = builder.createConvert(loc, signlessType, left);
- cmpRight = builder.createConvert(loc, signlessType, right);
+ if constexpr (isMax)
+ return mlir::arith::MaxUIOp::create(builder, loc, left, right);
+ else
+ return mlir::arith::MinUIOp::create(builder, loc, left, right);
+ } else {
+ if constexpr (isMax)
+ return mlir::arith::MaxSIOp::create(builder, loc, left, right);
+ else
+ return mlir::arith::MinSIOp::create(builder, loc, left, right);
}
- pred = mlir::arith::CmpIOp::create(builder, loc, integerPredicate, cmpLeft,
- cmpRight);
} else if (fir::isa_char(type) || fir::isa_char(fir::unwrapRefType(type))) {
// TODO: ! character min and max is tricky because the result
// length is the length of the longest argument!
// So we may need a temp.
TODO(loc, "intrinsic: min and max for CHARACTER");
}
- assert(pred && "pred must be defined");
-
- return mlir::arith::SelectOp::create(builder, loc, pred, left, right);
+ llvm_unreachable("unsupported extremum");
}
// UNLINK
diff --git a/flang/test/Lower/HLFIR/array-ctor-as-elemental.f90 b/flang/test/Lower/HLFIR/array-ctor-as-elemental.f90
index 5c4f079c86e20..f52216d2f558d 100644
--- a/flang/test/Lower/HLFIR/array-ctor-as-elemental.f90
+++ b/flang/test/Lower/HLFIR/array-ctor-as-elemental.f90
@@ -54,8 +54,7 @@ subroutine test_as_strided_elemental(lb, ub, stride)
! CHECK: %[[VAL_12:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i64>
! CHECK: %[[VAL_13:.*]] = arith.divsi %[[VAL_11]], %[[VAL_12]] : i64
! CHECK: %[[VAL_14:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_15:.*]] = arith.cmpi sgt, %[[VAL_13]], %[[VAL_14]] : i64
-! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_15]], %[[VAL_13]], %[[VAL_14]] : i64
+! CHECK: %[[VAL_16:.*]] = arith.maxsi %[[VAL_13]], %[[VAL_14]] : i64
! CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_6]], %[[VAL_16]] : i64
! CHECK: %[[VAL_18:.*]] = fir.convert %[[VAL_17]] : (i64) -> index
! CHECK: %[[VAL_19:.*]] = fir.shape %[[VAL_18]] : (index) -> !fir.shape<1>
diff --git a/flang/test/Lower/HLFIR/array-ctor-as-inlined-temp.f90 b/flang/test/Lower/HLFIR/array-ctor-as-inlined-temp.f90
index 415a9df5bf838..6fcd942f4a1e5 100644
--- a/flang/test/Lower/HLFIR/array-ctor-as-inlined-temp.f90
+++ b/flang/test/Lower/HLFIR/array-ctor-as-inlined-temp.f90
@@ -127,8 +127,7 @@ subroutine test_implied_do(n)
! CHECK: %[[VAL_10:.*]] = arith.constant 1 : i64
! CHECK: %[[VAL_11:.*]] = arith.divsi %[[VAL_9]], %[[VAL_10]] : i64
! CHECK: %[[VAL_12:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_13:.*]] = arith.cmpi sgt, %[[VAL_11]], %[[VAL_12]] : i64
-! CHECK: %[[VAL_14:.*]] = arith.select %[[VAL_13]], %[[VAL_11]], %[[VAL_12]] : i64
+! CHECK: %[[VAL_14:.*]] = arith.maxsi %[[VAL_11]], %[[VAL_12]] : i64
! CHECK: %[[VAL_15:.*]] = arith.muli %[[VAL_4]], %[[VAL_14]] : i64
! CHECK: %[[VAL_16:.*]] = arith.addi %[[VAL_3]], %[[VAL_15]] : i64
! CHECK: %[[VAL_17:.*]] = fir.convert %[[VAL_16]] : (i64) -> index
@@ -191,8 +190,7 @@ subroutine test_strided_implied_do(lb, ub, stride)
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i64>
! CHECK: %[[VAL_15:.*]] = arith.divsi %[[VAL_13]], %[[VAL_14]] : i64
! CHECK: %[[VAL_16:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_17:.*]] = arith.cmpi sgt, %[[VAL_15]], %[[VAL_16]] : i64
-! CHECK: %[[VAL_18:.*]] = arith.select %[[VAL_17]], %[[VAL_15]], %[[VAL_16]] : i64
+! CHECK: %[[VAL_18:.*]] = arith.maxsi %[[VAL_15]], %[[VAL_16]] : i64
! CHECK: %[[VAL_19:.*]] = arith.muli %[[VAL_8]], %[[VAL_18]] : i64
! CHECK: %[[VAL_20:.*]] = arith.addi %[[VAL_7]], %[[VAL_19]] : i64
! CHECK: %[[VAL_21:.*]] = fir.convert %[[VAL_20]] : (i64) -> index
@@ -253,8 +251,7 @@ subroutine test_nested_implied_do(n, m)
! CHECK: %[[VAL_12:.*]] = arith.constant 1 : i64
! CHECK: %[[VAL_13:.*]] = arith.divsi %[[VAL_11]], %[[VAL_12]] : i64
! CHECK: %[[VAL_14:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_15:.*]] = arith.cmpi sgt, %[[VAL_13]], %[[VAL_14]] : i64
-! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_15]], %[[VAL_13]], %[[VAL_14]] : i64
+! CHECK: %[[VAL_16:.*]] = arith.maxsi %[[VAL_13]], %[[VAL_14]] : i64
! CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_6]], %[[VAL_16]] : i64
! CHECK: %[[VAL_18:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i64>
! CHECK: %[[VAL_19:.*]] = arith.constant 1 : i64
@@ -264,8 +261,7 @@ subroutine test_nested_implied_do(n, m)
! CHECK: %[[VAL_23:.*]] = arith.constant 1 : i64
! CHECK: %[[VAL_24:.*]] = arith.divsi %[[VAL_22]], %[[VAL_23]] : i64
! CHECK: %[[VAL_25:.*]] = arith.constant 0 : i64
-! CHECK: %[[VAL_26:.*]] = arith.cmpi sgt, %[[VAL_24]], %[[VAL_25]] : i64
-! CHECK: %[[VAL_27:.*]] = arith.select %[[VAL_26]], %[[VAL_24]], %[[VAL_25]] : i64
+! CHECK: %[[VAL_27:.*]] = arith.maxsi %[[VAL_24]], %[[VAL_25]] : i64
! CHECK: %[[VAL_28:.*]] = arith.muli %[[VAL_17]], %[[VAL_27]] : i64
! CHECK: %[[VAL_29:.*]] = arith.addi %[[VAL_5]], %[[VAL_28]] : i64
! CHECK: %[[VAL_30:.*]] = fir.convert %[[VAL_29]] : (i64) -> index
diff --git a/flang/test/Lower/HLFIR/binary-ops.f90 b/flang/test/Lower/HLFIR/binary-ops.f90
index f4e1643dd78b6..3a83cd44d3c2e 100644
--- a/flang/test/Lower/HLFIR/binary-ops.f90
+++ b/flang/test/Lower/HLFIR/binary-ops.f90
@@ -206,8 +206,7 @@ subroutine extremum(c, n, l)
! CHECK: hlfir.declare {{.*}}c"}
! CHECK: %[[VAL_11:.*]] = arith.constant 0 : i64
! CHECK: %[[VAL_12:.*]] = fir.load %{{.*}} : !fir.ref<i64>
-! CHECK: %[[VAL_13:.*]] = arith.cmpi sgt, %[[VAL_11]], %[[VAL_12]] : i64
-! CHECK: arith.select %[[VAL_13]], %[[VAL_11]], %[[VAL_12]] : i64
+! CHECK: arith.maxsi %[[VAL_11]], %[[VAL_12]] : i64
subroutine cmp_int(l, x, y)
logical :: l
diff --git a/flang/test/Lower/HLFIR/custom-intrinsic.f90 b/flang/test/Lower/HLFIR/custom-intrinsic.f90
index 4999eebf376e7..294077105ed7b 100644
--- a/flang/test/Lower/HLFIR/custom-intrinsic.f90
+++ b/flang/test/Lower/HLFIR/custom-intrinsic.f90
@@ -14,8 +14,7 @@ function max_simple(a, b)
! CHECK-NEXT: %[[RES_DECL:.*]]:2 = hlfir.declare %[[RES_ALLOC]] {uniq_name = "_QFmax_simpleEmax_simple"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK-NEXT: %[[A_LD:.*]] = fir.load %[[A_DECL]]#0 : !fir.ref<i32>
! CHECK-NEXT: %[[B_LD:.*]] = fir.load %[[B_DECL]]#0 : !fir.ref<i32>
-! CHECK-NEXT: %[[A_GT_B:.*]] = arith.cmpi sgt, %[[A_LD]], %[[B_LD]] : i32
-! CHECK-NEXT: %[[SELECT:.*]] = arith.select %[[A_GT_B]], %[[A_LD]], %[[B_LD]] : i32
+! CHECK-NEXT: %[[SELECT:.*]] = arith.maxsi %[[A_LD]], %[[B_LD]] : i32
! CHECK-NEXT: hlfir.assign %[[SELECT]] to %[[RES_DECL]]#0 : i32, !fir.ref<i32>
! CHECK-NEXT: %[[RES_LD:.*]] = fir.load %[[RES_DECL]]#0 : !fir.ref<i32>
! CHECK-NEXT: return %[[RES_LD]] : i32
@@ -38,12 +37,10 @@ function max_dynamic_optional_scalar(a, b, c)
! CHECK: %[[VAL_8:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_9:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_10:.*]] = fir.is_present %[[VAL_5]]#0 : (!fir.ref<i32>) -> i1
-! CHECK: %[[VAL_11:.*]] = arith.cmpi sgt, %[[VAL_8]], %[[VAL_9]] : i32
-! CHECK: %[[VAL_12:.*]] = arith.select %[[VAL_11]], %[[VAL_8]], %[[VAL_9]] : i32
+! CHECK: %[[VAL_12:.*]] = arith.maxsi %[[VAL_8]], %[[VAL_9]] : i32
! CHECK: %[[VAL_13:.*]] = fir.if %[[VAL_10]] -> (i32) {
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_15:.*]] = arith.cmpi sgt, %[[VAL_12]], %[[VAL_14]] : i32
-! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_15]], %[[VAL_12]], %[[VAL_14]] : i32
+! CHECK: %[[VAL_16:.*]] = arith.maxsi %[[VAL_12]], %[[VAL_14]] : i32
! CHECK: fir.result %[[VAL_16]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_12]] : i32
@@ -73,20 +70,17 @@ function max_dynamic_optional_scalar2(a, b, c, d)
! CHECK: %[[VAL_11:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_12:.*]] = fir.is_present %[[VAL_6]]#0 : (!fir.ref<i32>) -> i1
! CHECK: %[[VAL_13:.*]] = fir.is_present %[[VAL_7]]#0 : (!fir.ref<i32>) -> i1
-! CHECK: %[[VAL_14:.*]] = arith.cmpi sgt, %[[VAL_10]], %[[VAL_11]] : i32
-! CHECK: %[[VAL_15:.*]] = arith.select %[[VAL_14]], %[[VAL_10]], %[[VAL_11]] : i32
+! CHECK: %[[VAL_15:.*]] = arith.maxsi %[[VAL_10]], %[[VAL_11]] : i32
! CHECK: %[[VAL_16:.*]] = fir.if %[[VAL_12]] -> (i32) {
! CHECK: %[[VAL_17:.*]] = fir.load %[[VAL_6]]#0 : !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: %[[VAL_19:.*]] = arith.maxsi %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: fir.result %[[VAL_19]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_15]] : i32
! CHECK: }
! CHECK: %[[VAL_20:.*]] = fir.if %[[VAL_13]] -> (i32) {
! CHECK: %[[VAL_21:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_22:.*]] = arith.cmpi sgt, %[[VAL_23:.*]], %[[VAL_21]] : i32
-! CHECK: %[[VAL_24:.*]] = arith.select %[[VAL_22]], %[[VAL_23]], %[[VAL_21]] : i32
+! CHECK: %[[VAL_24:.*]] = arith.maxsi %[[VAL_23:.*]], %[[VAL_21]] : i32
! CHECK: fir.result %[[VAL_24]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_25:.*]] : i32
@@ -119,8 +113,7 @@ function max_array(a, b)
! 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: %[[VAL_19:.*]] = arith.maxsi %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: hlfir.yield_element %[[VAL_19]] : i32
! CHECK: }
! CHECK: hlfir.assign %[[VAL_20:.*]] to %[[VAL_11]]#0 : !hlfir.expr<42xi32>, !fir.ref<!fir.array<42xi32>>
@@ -155,13 +148,11 @@ function max_dynamic_optional_array(a, b, c)
! CHECK: ^bb0(%[[VAL_17:.*]]: index):
! CHECK: %[[VAL_18:.*]] = hlfir.designate %[[VAL_6]]#0 (%[[VAL_17]]) : (!fir.ref<!fir.array<10xi32>>, index) -> !fir.ref<i32>
! CHECK: %[[VAL_19:.*]] = fir.load %[[VAL_18]] : !fir.ref<i32>
-! CHECK: %[[VAL_20:.*]] = arith.cmpi sgt, %[[VAL_15]], %[[VAL_19]] : i32
-! CHECK: %[[VAL_21:.*]] = arith.select %[[VAL_20]], %[[VAL_15]], %[[VAL_19]] : i32
+! CHECK: %[[VAL_21:.*]] = arith.maxsi %[[VAL_15]], %[[VAL_19]] : i32
! CHECK: %[[VAL_22:.*]] = fir.if %[[VAL_14]] -> (i32) {
! CHECK: %[[VAL_23:.*]] = hlfir.designate %[[VAL_9]]#0 (%[[VAL_17]]) : (!fir.ref<!fir.array<10xi32>>, index) -> !fir.ref<i32>
! CHECK: %[[VAL_24:.*]] = fir.load %[[VAL_23]] : !fir.ref<i32>
-! CHECK: %[[VAL_25:.*]] = arith.cmpi sgt, %[[VAL_21]], %[[VAL_24]] : i32
-! CHECK: %[[VAL_26:.*]] = arith.select %[[VAL_25]], %[[VAL_21]], %[[VAL_24]] : i32
+! CHECK: %[[VAL_26:.*]] = arith.maxsi %[[VAL_21]], %[[VAL_24]] : i32
! CHECK: fir.result %[[VAL_26]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_21]] : i32
@@ -187,8 +178,7 @@ function min_simple(a, b)
! CHECK: %[[VAL_5:.*]]:2 = hlfir.declare %[[VAL_4]] {uniq_name = "_QFmin_simpleEmin_simple"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: %[[VAL_6:.*]] = fir.load %[[VAL_2]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_7:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_8:.*]] = arith.cmpi slt, %[[VAL_6]], %[[VAL_7]] : i32
-! CHECK: %[[VAL_9:.*]] = arith.select %[[VAL_8]], %[[VAL_6]], %[[VAL_7]] : i32
+! CHECK: %[[VAL_9:.*]] = arith.minsi %[[VAL_6]], %[[VAL_7]] : i32
! CHECK: hlfir.assign %[[VAL_9]] to %[[VAL_5]]#0 : i32, !fir.ref<i32>
! CHECK: %[[VAL_10:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
! CHECK: return %[[VAL_10]] : i32
@@ -211,12 +201,10 @@ function min_dynamic_optional_scalar(a, b, c)
! CHECK: %[[VAL_8:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_9:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_10:.*]] = fir.is_present %[[VAL_5]]#0 : (!fir.ref<i32>) -> i1
-! CHECK: %[[VAL_11:.*]] = arith.cmpi slt, %[[VAL_8]], %[[VAL_9]] : i32
-! CHECK: %[[VAL_12:.*]] = arith.select %[[VAL_11]], %[[VAL_8]], %[[VAL_9]] : i32
+! CHECK: %[[VAL_12:.*]] = arith.minsi %[[VAL_8]], %[[VAL_9]] : i32
! CHECK: %[[VAL_13:.*]] = fir.if %[[VAL_10]] -> (i32) {
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_15:.*]] = arith.cmpi slt, %[[VAL_12]], %[[VAL_14]] : i32
-! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_15]], %[[VAL_12]], %[[VAL_14]] : i32
+! CHECK: %[[VAL_16:.*]] = arith.minsi %[[VAL_12]], %[[VAL_14]] : i32
! CHECK: fir.result %[[VAL_16]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_12]] : i32
@@ -246,20 +234,17 @@ function min_dynamic_optional_scalar2(a, b, c, d)
! CHECK: %[[VAL_11:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
! CHECK: %[[VAL_12:.*]] = fir.is_present %[[VAL_6]]#0 : (!fir.ref<i32>) -> i1
! CHECK: %[[VAL_13:.*]] = fir.is_present %[[VAL_7]]#0 : (!fir.ref<i32>) -> i1
-! CHECK: %[[VAL_14:.*]] = arith.cmpi slt, %[[VAL_10]], %[[VAL_11]] : i32
-! CHECK: %[[VAL_15:.*]] = arith.select %[[VAL_14]], %[[VAL_10]], %[[VAL_11]] : i32
+! CHECK: %[[VAL_15:.*]] = arith.minsi %[[VAL_10]], %[[VAL_11]] : i32
! CHECK: %[[VAL_16:.*]] = fir.if %[[VAL_12]] -> (i32) {
! CHECK: %[[VAL_17:.*]] = fir.load %[[VAL_6]]#0 : !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: %[[VAL_19:.*]] = arith.minsi %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: fir.result %[[VAL_19]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_15]] : i32
! CHECK: }
! CHECK: %[[VAL_20:.*]] = fir.if %[[VAL_13]] -> (i32) {
! CHECK: %[[VAL_21:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<i32>
-! CHECK: %[[VAL_22:.*]] = arith.cmpi slt, %[[VAL_23:.*]], %[[VAL_21]] : i32
-! CHECK: %[[VAL_24:.*]] = arith.select %[[VAL_22]], %[[VAL_23]], %[[VAL_21]] : i32
+! CHECK: %[[VAL_24:.*]] = arith.minsi %[[VAL_23:.*]], %[[VAL_21]] : i32
! CHECK: fir.result %[[VAL_24]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_25:.*]] : i32
@@ -292,8 +277,7 @@ function min_array(a, b)
! 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: %[[VAL_19:.*]] = arith.minsi %[[VAL_15]], %[[VAL_17]] : i32
! CHECK: hlfir.yield_element %[[VAL_19]] : i32
! CHECK: }
! CHECK: hlfir.assign %[[VAL_20:.*]] to %[[VAL_11]]#0 : !hlfir.expr<42xi32>, !fir.ref<!fir.array<42xi32>>
@@ -328,13 +312,11 @@ function min_dynamic_optional_array(a, b, c)
! CHECK: ^bb0(%[[VAL_17:.*]]: index):
! CHECK: %[[VAL_18:.*]] = hlfir.designate %[[VAL_6]]#0 (%[[VAL_17]]) : (!fir.ref<!fir.array<10xi32>>, index) -> !fir.ref<i32>
! CHECK: %[[VAL_19:.*]] = fir.load %[[VAL_18]] : !fir.ref<i32>
-! CHECK: %[[VAL_20:.*]] = arith.cmpi slt, %[[VAL_15]], %[[VAL_19]] : i32
-! CHECK: %[[VAL_21:.*]] = arith.select %[[VAL_20]], %[[VAL_15]], %[[VAL_19]] : i32
+! CHECK: %[[VAL_21:.*]] = arith.minsi %[[VAL_15]], %[[VAL_19]] : i32
! CHECK: %[[VAL_22:.*]] = fir.if %[[VAL_14]] -> (i32) {
! CHECK: %[[VAL_23:.*]] = hlfir.designate %[[VAL_9]]#0 (%[[VAL_17]]) : (!fir.ref<!fir.array<10xi32>>, index) -> !fir.ref<i32>
! CHECK: %[[VAL_24:.*]] = fir.load %[[VAL_23]] : !fir.ref<i32>
-! CHECK: %[[VAL_25:.*]] = arith.cmpi slt, %[[VAL_21]], %[[VAL_24]] : i32
-! CHECK: %[[VAL_26:.*]] = arith.select %[[VAL_25]], %[[VAL_21]], %[[VAL_24]] : i32
+! CHECK: %[[VAL_26:.*]] = arith.minsi %[[VAL_21]], %[[VAL_24]] : i32
! CHECK: fir.result %[[VAL_26]] : i32
! CHECK: } else {
! CHECK: fir.result %[[VAL_21]] : i32
@@ -812,8 +794,7 @@ subroutine allocatables_test(a, b, c)
! CHECK: %[[VAL_112:.*]] = arith.addi %[[VAL_84]], %[[VAL_111]] : index
! CHECK: %[[VAL_113:.*]] = hlfir.designate %[[VAL_79]] (%[[VAL_108]], %[[VAL_110]], %[[VAL_112]]) : (!fir.box<!fir.heap<!fir.array<?x?x?xi32>>>, index, index, index) -> !fir.ref<i32>
! CHECK: %[[VAL_114:.*]] = fir.load %[[VAL_113]] : !fir.ref<i32>
-! CHECK: %[[VAL_115:.*]] = arith.cmpi slt, %[[VAL_99]], %[[VAL_114]] : i32
-! CHECK: %[[VAL_116:.*]] = arith.select %[[VAL_115]], %[[VAL_99]], %[[VAL_114]] : i32
+! CHECK: %[[VAL_116:.*]] = arith.minsi %[[VAL_99]], %[[VAL_114]] : i32
! CHECK: %[[VAL_117:.*]] = fir.if %[[VAL_70]] -> (i32) {
! CHECK: %[[VAL_118:.*]] = arith.constant 0 : index
! CHECK: %[[VAL_119:.*]]:3 = fir.box_dims %[[VAL_80]], %[[VAL_118]] : (!fir.box<!fir.heap<!fir.array<?x?x?xi32>>>, inde...
[truncated]
|
| cmpLeft = builder.createConvert(loc, signlessType, left); | ||
| cmpRight = builder.createConvert(loc, signlessType, right); | ||
| if constexpr (isMax) | ||
| return mlir::arith::MaxUIOp::create(builder, loc, left, right); |
Contributor
There was a problem hiding this comment.
According to https://mlir.llvm.org/docs/Dialects/ArithOps/#arithmaxui-arithmaxuiop , maxUIOp requires signless integers. It seems that the original code was creating signless integers for left and right, but the new code doesn't.
I quickly checked the tests modifications and it seems all existing tests work with signed args. Perhaps add some unsigned tests to see if not having explicitly signless integers is a problem?
Contributor
Author
There was a problem hiding this comment.
Good catch, Eugene! I need to bring back those converts. Unfortunately, I do not know a way to wrtie such a test, because unsigned argument of max/min is rejected by semantics.
Contributor
There was a problem hiding this comment.
Unfortunately, I do not know a way to write such a test, because unsigned argument of max/min is rejected by semantics.
I'll look into this. Maybe I'll at least give you a temporary patch to get past the error, so the unsigned code could be tested.
Contributor
There was a problem hiding this comment.
Maybe you could test it via some FIR to FIR tests starting with hlfir.maxval on unsigned (if we allow it).
Contributor
There was a problem hiding this comment.
@vzakhari , the following worked for me with upstream flang:
implicit none
unsigned a, b, c
a = unsigned(1)
b = unsigned(2)
c = unsigned(3)
print *, "min =", min(a, b, c), ", max =", max(a, b, c)
end
Compile with -funsigned.
Contributor
Author
There was a problem hiding this comment.
Indeed, it works. Thank you, Eugene. I did see some semantics errors, but I must have done some mistake in the code. I will add the tests!
🐧 Linux x64 Test Results
✅ The build succeeded and all tests passed. |
Unified code between max/min and maxval/minval.
Contributor
There was a problem hiding this comment.
What about here? Should this code be also updated?
Contributor
Author
There was a problem hiding this comment.
This would require a bit more restructuring, because this function generates just the comparison for the select, but not the whole max/min. I would prefer to keep it as-is for the time being.
Contributor
There was a problem hiding this comment.
Sounds good. Maybe add a short comment note for the future developer?
Comment on lines
785
to
790
Contributor
There was a problem hiding this comment.
Is this check still relevant? (I mean, you start pulling on threads, and now you have to unravel the whole yarn :-) )
Contributor
Author
There was a problem hiding this comment.
Should be relevant as I was not planning to enable inlining UNSIGNED maxval/minval in this PR. That would be an unrelated change. So we will continue keeping UNSIGNED maxval/minval as the hlfir operation in this pass.
eugeneepshteyn
approved these changes
Mar 17, 2026
eugeneepshteyn
left a comment
eugeneepshteyn
left a comment
Contributor
There was a problem hiding this comment.
LGTM. Thanks, Slava!
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
arith.maxsi/maxui/minsi/minui are more concise than cmp+select
and probably allow more folding, so we should use it in Flang lowering.