[flang-commits] [flang] [flang][openacc] Use the array section for assumed shape array reduction (PR #68147)
Valentin Clement バレンタイン クレメン via flang-commits
flang-commits at lists.llvm.org
Tue Oct 3 13:21:30 PDT 2023
https://github.com/clementval updated https://github.com/llvm/llvm-project/pull/68147
>From 3582da45384e68e746b696808673526bd8cd8a3a Mon Sep 17 00:00:00 2001
From: Valentin Clement <clementval at gmail.com>
Date: Fri, 29 Sep 2023 10:36:47 -0700
Subject: [PATCH 1/2] [flang][openacc] Use the array section for assumed shape
array reduction
---
.../include/flang/Optimizer/Dialect/FIRType.h | 3 +
flang/lib/Lower/OpenACC.cpp | 126 ++++++++++++++----
flang/lib/Optimizer/Dialect/FIRType.cpp | 7 +
flang/test/Lower/OpenACC/acc-reduction.f90 | 64 +++++++--
4 files changed, 162 insertions(+), 38 deletions(-)
diff --git a/flang/include/flang/Optimizer/Dialect/FIRType.h b/flang/include/flang/Optimizer/Dialect/FIRType.h
index 77807ea2a308c67..e93026ea2b4f075 100644
--- a/flang/include/flang/Optimizer/Dialect/FIRType.h
+++ b/flang/include/flang/Optimizer/Dialect/FIRType.h
@@ -333,6 +333,9 @@ bool isUnlimitedPolymorphicType(mlir::Type ty);
/// Return true iff `ty` is the type of an assumed type.
bool isAssumedType(mlir::Type ty);
+/// Return true iff `ty` is the type of an assumed shape array.
+bool isAssumedShape(mlir::Type ty);
+
/// Return true iff `boxTy` wraps a record type or an unlimited polymorphic
/// entity. Polymorphic entities with intrinsic type spec do not have addendum
inline bool boxHasAddendum(fir::BaseBoxType boxTy) {
diff --git a/flang/lib/Lower/OpenACC.cpp b/flang/lib/Lower/OpenACC.cpp
index 57dd0fab2b9c69a..84ce4cde94200f9 100644
--- a/flang/lib/Lower/OpenACC.cpp
+++ b/flang/lib/Lower/OpenACC.cpp
@@ -492,8 +492,8 @@ std::string getBoundsString(llvm::SmallVector<mlir::Value> &bounds) {
fir::getIntIfConstant(boundsOp.getLowerbound()) &&
boundsOp.getUpperbound() &&
fir::getIntIfConstant(boundsOp.getUpperbound())) {
- boundStr << "lb" << *fir::getIntIfConstant(boundsOp.getUpperbound())
- << ".ub" << *fir::getIntIfConstant(boundsOp.getLowerbound());
+ boundStr << "lb" << *fir::getIntIfConstant(boundsOp.getLowerbound())
+ << ".ub" << *fir::getIntIfConstant(boundsOp.getUpperbound());
} else if (boundsOp.getExtent() &&
fir::getIntIfConstant(boundsOp.getExtent())) {
boundStr << "ext" << *fir::getIntIfConstant(boundsOp.getExtent());
@@ -890,6 +890,52 @@ static mlir::Value genScalarCombiner(fir::FirOpBuilder &builder,
TODO(loc, "reduction operator");
}
+static fir::ShapeOp
+genShapeFromBounds(mlir::Location loc, fir::FirOpBuilder &builder,
+ const llvm::SmallVector<mlir::Value> &args) {
+ assert(args.size() % 3 == 0 && "Triplets must be a multiple of 3");
+ llvm::SmallVector<mlir::Value> extents;
+ mlir::Type idxTy = builder.getIndexType();
+ mlir::Value one = builder.createIntegerConstant(loc, idxTy, 1);
+ mlir::Value zero = builder.createIntegerConstant(loc, idxTy, 0);
+ for (unsigned i = 0; i < args.size(); i += 3) {
+ mlir::Value s1 =
+ builder.create<mlir::arith::SubIOp>(loc, args[i + 1], args[0]);
+ mlir::Value s2 = builder.create<mlir::arith::AddIOp>(loc, s1, one);
+ mlir::Value s3 = builder.create<mlir::arith::DivSIOp>(loc, s2, args[i + 2]);
+ mlir::Value cmp = builder.create<mlir::arith::CmpIOp>(
+ loc, mlir::arith::CmpIPredicate::sgt, s3, zero);
+ mlir::Value ext = builder.create<mlir::arith::SelectOp>(loc, cmp, s3, zero);
+ extents.push_back(ext);
+ }
+ return builder.create<fir::ShapeOp>(loc, extents);
+}
+
+static llvm::SmallVector<mlir::Value>
+genConstantBounds(fir::FirOpBuilder &builder, mlir::Location loc,
+ mlir::acc::DataBoundsOp &dataBound) {
+ mlir::Type idxTy = builder.getIndexType();
+ mlir::Value lb, ub, step;
+ if (dataBound.getLowerbound() &&
+ fir::getIntIfConstant(dataBound.getLowerbound()) &&
+ dataBound.getUpperbound() &&
+ fir::getIntIfConstant(dataBound.getUpperbound())) {
+ lb = builder.createIntegerConstant(
+ loc, idxTy, *fir::getIntIfConstant(dataBound.getLowerbound()));
+ ub = builder.createIntegerConstant(
+ loc, idxTy, *fir::getIntIfConstant(dataBound.getUpperbound()));
+ step = builder.createIntegerConstant(loc, idxTy, 1);
+ } else if (dataBound.getExtent()) {
+ lb = builder.createIntegerConstant(loc, idxTy, 0);
+ ub = builder.createIntegerConstant(
+ loc, idxTy, *fir::getIntIfConstant(dataBound.getExtent()) - 1);
+ step = builder.createIntegerConstant(loc, idxTy, 1);
+ } else {
+ llvm::report_fatal_error("Expect constant lb/ub or extent");
+ }
+ return {lb, ub, step};
+}
+
static void genCombiner(fir::FirOpBuilder &builder, mlir::Location loc,
mlir::acc::ReductionOperator op, mlir::Type ty,
mlir::Value value1, mlir::Value value2,
@@ -907,30 +953,14 @@ static void genCombiner(fir::FirOpBuilder &builder, mlir::Location loc,
if (allConstantBound) {
// Use the constant bound directly in the combiner region so they do not
// need to be passed as block argument.
- mlir::Type idxTy = builder.getIndexType();
for (auto bound : llvm::reverse(bounds)) {
auto dataBound =
mlir::dyn_cast<mlir::acc::DataBoundsOp>(bound.getDefiningOp());
- mlir::Value lb, ub, step;
- if (dataBound.getLowerbound() &&
- fir::getIntIfConstant(dataBound.getLowerbound()) &&
- dataBound.getUpperbound() &&
- fir::getIntIfConstant(dataBound.getUpperbound())) {
- lb = builder.createIntegerConstant(
- loc, idxTy, *fir::getIntIfConstant(dataBound.getLowerbound()));
- ub = builder.createIntegerConstant(
- loc, idxTy, *fir::getIntIfConstant(dataBound.getUpperbound()));
- step = builder.createIntegerConstant(loc, idxTy, 1);
- } else if (dataBound.getExtent()) {
- lb = builder.createIntegerConstant(loc, idxTy, 0);
- ub = builder.createIntegerConstant(
- loc, idxTy, *fir::getIntIfConstant(dataBound.getExtent()) - 1);
- step = builder.createIntegerConstant(loc, idxTy, 1);
- } else {
- llvm::report_fatal_error("Expect constant lb/ub or extent");
- }
- auto loop = builder.create<fir::DoLoopOp>(loc, lb, ub, step,
- /*unordered=*/false);
+ llvm::SmallVector<mlir::Value> values =
+ genConstantBounds(builder, loc, dataBound);
+ auto loop =
+ builder.create<fir::DoLoopOp>(loc, values[0], values[1], values[2],
+ /*unordered=*/false);
builder.setInsertionPointToStart(loop.getBody());
loops.push_back(loop);
ivs.push_back(loop.getInductionVar());
@@ -962,13 +992,55 @@ static void genCombiner(fir::FirOpBuilder &builder, mlir::Location loc,
builder.create<fir::StoreOp>(loc, res, addr1);
builder.setInsertionPointAfter(loops[0]);
} else if (auto boxTy = mlir::dyn_cast<fir::BaseBoxType>(ty)) {
+ llvm::SmallVector<mlir::Value> tripletArgs;
fir::SequenceType seqTy =
mlir::dyn_cast_or_null<fir::SequenceType>(boxTy.getEleTy());
if (!seqTy)
TODO(loc, "Unsupported boxed type in OpenACC reduction");
- hlfir::Entity left = hlfir::Entity{value1};
- hlfir::Entity right = hlfir::Entity{value2};
- auto shape = hlfir::genShape(loc, builder, left);
+
+ if (allConstantBound) {
+ for (auto bound : llvm::reverse(bounds)) {
+ auto dataBound =
+ mlir::cast<mlir::acc::DataBoundsOp>(bound.getDefiningOp());
+ tripletArgs.append(genConstantBounds(builder, loc, dataBound));
+ }
+ } else {
+ assert(((recipe.getCombinerRegion().getArguments().size() - 2) / 3 ==
+ seqTy.getDimension()) &&
+ "Expect 3 block arguments per dimension");
+ for (auto arg : recipe.getCombinerRegion().getArguments().drop_front(2))
+ tripletArgs.push_back(arg);
+ }
+ auto shape = genShapeFromBounds(loc, builder, tripletArgs);
+
+ hlfir::DesignateOp::Subscripts triplets;
+ for (unsigned i = 2; i < recipe.getCombinerRegion().getArguments().size();
+ i += 3)
+ triplets.emplace_back(hlfir::DesignateOp::Triplet{
+ recipe.getCombinerRegion().getArgument(i),
+ recipe.getCombinerRegion().getArgument(i + 1),
+ recipe.getCombinerRegion().getArgument(i + 2)});
+
+ llvm::SmallVector<mlir::Value> lenParamsLeft;
+ auto leftEntity = hlfir::Entity{value1};
+ hlfir::genLengthParameters(loc, builder, leftEntity, lenParamsLeft);
+ auto leftDesignate = builder.create<hlfir::DesignateOp>(
+ loc, value1.getType(), leftEntity, /*component=*/"",
+ /*componentShape=*/mlir::Value{}, triplets,
+ /*substring=*/mlir::ValueRange{}, /*complexPartAttr=*/std::nullopt,
+ shape, lenParamsLeft);
+ auto left = hlfir::Entity{leftDesignate.getResult()};
+
+ llvm::SmallVector<mlir::Value> lenParamsRight;
+ auto rightEntity = hlfir::Entity{value2};
+ hlfir::genLengthParameters(loc, builder, rightEntity, lenParamsRight);
+ auto rightDesignate = builder.create<hlfir::DesignateOp>(
+ loc, value2.getType(), rightEntity, /*component=*/"",
+ /*componentShape=*/mlir::Value{}, triplets,
+ /*substring=*/mlir::ValueRange{}, /*complexPartAttr=*/std::nullopt,
+ shape, lenParamsRight);
+ auto right = hlfir::Entity{rightDesignate.getResult()};
+
llvm::SmallVector<mlir::Value, 1> typeParams;
auto genKernel = [&builder, &loc, op, seqTy, &left, &right](
mlir::Location l, fir::FirOpBuilder &b,
@@ -1079,7 +1151,7 @@ genReductions(const Fortran::parser::AccObjectListWithReduction &objectList,
std::string recipeName = fir::getTypeAsString(
ty, converter.getKindMap(),
("reduction_" + stringifyReductionOperator(mlirOp)).str() + suffix);
- if (!areAllBoundConstant(bounds))
+ if (!areAllBoundConstant(bounds) || fir::isAssumedShape(baseAddr.getType()))
ty = baseAddr.getType();
mlir::acc::ReductionRecipeOp recipe =
Fortran::lower::createOrGetReductionRecipe(
diff --git a/flang/lib/Optimizer/Dialect/FIRType.cpp b/flang/lib/Optimizer/Dialect/FIRType.cpp
index dd79aa27645452f..9301dd590b8b937 100644
--- a/flang/lib/Optimizer/Dialect/FIRType.cpp
+++ b/flang/lib/Optimizer/Dialect/FIRType.cpp
@@ -311,6 +311,13 @@ bool isAssumedType(mlir::Type ty) {
return false;
}
+bool isAssumedShape(mlir::Type ty) {
+ if (auto boxTy = mlir::dyn_cast<fir::BoxType>(ty))
+ if (auto seqTy = mlir::dyn_cast<fir::SequenceType>(boxTy.getEleTy()))
+ return seqTy.hasDynamicExtents();
+ return false;
+}
+
bool isPolymorphicType(mlir::Type ty) {
if (auto refTy = fir::dyn_cast_ptrEleTy(ty))
ty = refTy;
diff --git a/flang/test/Lower/OpenACC/acc-reduction.f90 b/flang/test/Lower/OpenACC/acc-reduction.f90
index 20231c0f6c7bd43..b3281587beb3778 100644
--- a/flang/test/Lower/OpenACC/acc-reduction.f90
+++ b/flang/test/Lower/OpenACC/acc-reduction.f90
@@ -3,6 +3,32 @@
! RUN: bbc -fopenacc -emit-fir %s -o - | FileCheck %s --check-prefixes=CHECK,FIR
! RUN: bbc -fopenacc -emit-hlfir %s -o - | FileCheck %s --check-prefixes=CHECK,HLFIR
+! CHECK-LABEL: acc.reduction.recipe @"reduction_add_section_lb1.ub3_ref_?xi32" : !fir.box<!fir.array<?xi32>> reduction_operator <add> init {
+! CHECK: ^bb0(%[[ARG0:.*]]: !fir.box<!fir.array<?xi32>>):
+! HLFIR: %[[BOX_DIMS:.*]]:3 = fir.box_dims %[[ARG0]], %c0{{.*}} : (!fir.box<!fir.array<?xi32>>, index) -> (index, index, index)
+! HLFIR: %[[SHAPE:.*]] = fir.shape %[[BOX_DIMS]]#1 : (index) -> !fir.shape<1>
+! HLFIR: %[[TEMP:.*]] = fir.allocmem !fir.array<?xi32>, %0#1 {bindc_name = ".tmp", uniq_name = ""}
+! HLFIR: %[[DECLARE:.*]]:2 = hlfir.declare %[[TEMP]](%[[SHAPE]]) {uniq_name = ".tmp"} : (!fir.heap<!fir.array<?xi32>>, !fir.shape<1>) -> (!fir.box<!fir.array<?xi32>>, !fir.heap<!fir.array<?xi32>>)
+! HLFIR: hlfir.assign %c0{{.*}} to %[[DECLARE]]#0 : i32, !fir.box<!fir.array<?xi32>>
+! HLFIR: acc.yield %[[DECLARE]]#0 : !fir.box<!fir.array<?xi32>>
+! CHECK: } combiner {
+! CHECK: ^bb0(%[[ARG0:.*]]: !fir.box<!fir.array<?xi32>>, %[[ARG1:.*]]: !fir.box<!fir.array<?xi32>>):
+! HLFIR: %[[SHAPE:.*]] = fir.shape %{{.*}} : (index) -> !fir.shape<1>
+! HLFIR: %[[DES1:.*]] = hlfir.designate %[[ARG0]] shape %[[SHAPE]] : (!fir.box<!fir.array<?xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<?xi32>>
+! HLFIR: %[[DES2:.*]] = hlfir.designate %[[ARG1]] shape %[[SHAPE]] : (!fir.box<!fir.array<?xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<?xi32>>
+! HLFIR: %[[ELEMENTAL:.*]] = hlfir.elemental %[[SHAPE]] unordered : (!fir.shape<1>) -> !hlfir.expr<?xi32> {
+! HLFIR: ^bb0(%[[IV:.*]]: index):
+! HLFIR: %[[DES_V1:.*]] = hlfir.designate %[[DES1]] (%[[IV]]) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
+! HLFIR: %[[DES_V2:.*]] = hlfir.designate %[[DES2]] (%[[IV]]) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
+! HLFIR: %[[LOAD_V1:.*]] = fir.load %[[DES_V1]] : !fir.ref<i32>
+! HLFIR: %[[LOAD_V2:.*]] = fir.load %[[DES_V2]] : !fir.ref<i32>
+! HLFIR: %[[COMBINED:.*]] = arith.addi %[[LOAD_V1]], %[[LOAD_V2]] : i32
+! HLFIR: hlfir.yield_element %[[COMBINED]] : i32
+! HLFIR: }
+! HLFIR: hlfir.assign %[[ELEMENTAL]] to %[[ARG0]] : !hlfir.expr<?xi32>, !fir.box<!fir.array<?xi32>>
+! HLFIR: acc.yield %[[ARG0]] : !fir.box<!fir.array<?xi32>>
+! HLFIR: }
+
! CHECK-LABEL: acc.reduction.recipe @"reduction_max_ref_?xf32" : !fir.box<!fir.array<?xf32>> reduction_operator <max> init {
! CHECK: ^bb0(%[[ARG0:.*]]: !fir.box<!fir.array<?xf32>>):
! CHECK: %[[INIT_VALUE:.*]] = arith.constant -1.401300e-45 : f32
@@ -15,12 +41,12 @@
! HLFIR: acc.yield %[[DECLARE]]#0 : !fir.box<!fir.array<?xf32>>
! CHECK: } combiner {
! CHECK: ^bb0(%[[ARG0:.*]]: !fir.box<!fir.array<?xf32>>, %[[ARG1:.*]]: !fir.box<!fir.array<?xf32>>
-! HLFIR: %[[BOX_DIMS:.*]]:3 = fir.box_dims %[[ARG0]], %{{.*}} : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
-! HLFIR: %[[SHAPE:.*]] = fir.shape %[[BOX_DIMS]]#1 : (index) -> !fir.shape<1>
-! HLFIR: %[[ELEMENTAL:.*]] = hlfir.elemental %[[SHAPE]] unordered : (!fir.shape<1>) -> !hlfir.expr<?xf32> {
+! HLFIR: %[[LEFT:.*]] = hlfir.designate %[[ARG0]] (%{{.*}}:%{{.*}}:%{{.*}}) shape %{{.*}} : (!fir.box<!fir.array<?xf32>>, index, index, index, !fir.shape<1>) -> !fir.box<!fir.array<?xf32>>
+! HLFIR: %[[RIGHT:.*]] = hlfir.designate %[[ARG1]] (%{{.*}}:%{{.*}}:%{{.*}}) shape %{{.*}} : (!fir.box<!fir.array<?xf32>>, index, index, index, !fir.shape<1>) -> !fir.box<!fir.array<?xf32>>
+! HLFIR: %[[ELEMENTAL:.*]] = hlfir.elemental %{{.*}} unordered : (!fir.shape<1>) -> !hlfir.expr<?xf32> {
! HLFIR: ^bb0(%{{.*}}: index):
-! HLFIR: %[[DES_V1:.*]] = hlfir.designate %[[ARG0]] (%{{.*}}) : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
-! HLFIR: %[[DES_V2:.*]] = hlfir.designate %[[ARG1]] (%{{.*}}) : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
+! HLFIR: %[[DES_V1:.*]] = hlfir.designate %[[LEFT]] (%{{.*}}) : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
+! HLFIR: %[[DES_V2:.*]] = hlfir.designate %[[RIGHT]] (%{{.*}}) : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
! HLFIR: %[[LOAD_V1:.*]] = fir.load %[[DES_V1]] : !fir.ref<f32>
! HLFIR: %[[LOAD_V2:.*]] = fir.load %[[DES_V2]] : !fir.ref<f32>
! HLFIR: %[[CMPF:.*]] = arith.cmpf ogt, %[[LOAD_V1]], %[[LOAD_V2]] : f32
@@ -43,12 +69,12 @@
! HLFIR: acc.yield %[[DECLARE]]#0 : !fir.box<!fir.array<?xi32>>
! CHECK: } combiner {
! CHECK: ^bb0(%[[V1:.*]]: !fir.box<!fir.array<?xi32>>, %[[V2:.*]]: !fir.box<!fir.array<?xi32>>
-! HLFIR: %[[BOX_DIMS]]:3 = fir.box_dims %[[V1]], %{{.*}} : (!fir.box<!fir.array<?xi32>>, index) -> (index, index, index)
-! HLFIR: %[[SHAPE:.*]] = fir.shape %[[BOX_DIMS]]#1 : (index) -> !fir.shape<1>
-! HLFIR: %[[ELEMENTAL:.*]] = hlfir.elemental %[[SHAPE]] unordered : (!fir.shape<1>) -> !hlfir.expr<?xi32> {
+! HLFIR: %[[LEFT:.*]] = hlfir.designate %[[ARG0]] (%{{.*}}:%{{.*}}:%{{.*}}) shape %{{.*}} : (!fir.box<!fir.array<?xi32>>, index, index, index, !fir.shape<1>) -> !fir.box<!fir.array<?xi32>>
+! HLFIR: %[[RIGHT:.*]] = hlfir.designate %[[ARG1]] (%{{.*}}:%{{.*}}:%{{.*}}) shape %{{.*}} : (!fir.box<!fir.array<?xi32>>, index, index, index, !fir.shape<1>) -> !fir.box<!fir.array<?xi32>>
+! HLFIR: %[[ELEMENTAL:.*]] = hlfir.elemental %{{.*}} unordered : (!fir.shape<1>) -> !hlfir.expr<?xi32> {
! HLFIR: ^bb0(%{{.*}}: index):
-! HLFIR: %[[DES_V1:.*]] = hlfir.designate %[[V1]] (%{{.*}}) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
-! HLFIR: %[[DES_V2:.*]] = hlfir.designate %[[V2]] (%{{.*}}) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
+! HLFIR: %[[DES_V1:.*]] = hlfir.designate %[[LEFT]] (%{{.*}}) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
+! HLFIR: %[[DES_V2:.*]] = hlfir.designate %[[RIGHT]] (%{{.*}}) : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
! HLFIR: %[[LOAD_V1:.*]] = fir.load %[[DES_V1]] : !fir.ref<i32>
! HLFIR: %[[LOAD_V2:.*]] = fir.load %[[DES_V2]] : !fir.ref<i32>
! HLFIR: %[[COMBINED:.*]] = arith.addi %[[LOAD_V1]], %[[LOAD_V2]] : i32
@@ -1059,7 +1085,7 @@ subroutine acc_reduction_add_static_slice(a)
! CHECK: %[[BOUND:.*]] = acc.bounds lowerbound(%[[LB]] : index) upperbound(%[[UB]] : index) stride(%[[C1]] : index) startIdx(%[[C1]] : index)
! FIR: %[[RED:.*]] = acc.reduction varPtr(%[[ARG0]] : !fir.ref<!fir.array<100xi32>>) bounds(%[[BOUND]]) -> !fir.ref<!fir.array<100xi32>> {name = "a(11:20)"}
! HLFIR: %[[RED:.*]] = acc.reduction varPtr(%[[DECLARG0]]#1 : !fir.ref<!fir.array<100xi32>>) bounds(%[[BOUND]]) -> !fir.ref<!fir.array<100xi32>> {name = "a(11:20)"}
-! CHECK: acc.parallel reduction(@reduction_add_section_lb19.ub10_ref_100xi32 -> %[[RED]] : !fir.ref<!fir.array<100xi32>>)
+! CHECK: acc.parallel reduction(@reduction_add_section_lb10.ub19_ref_100xi32 -> %[[RED]] : !fir.ref<!fir.array<100xi32>>)
subroutine acc_reduction_add_dynamic_extent_add(a)
integer :: a(:)
@@ -1084,3 +1110,19 @@ subroutine acc_reduction_add_dynamic_extent_max(a)
! HLFIR: %[[DECLARG0:.*]]:2 = hlfir.declare %[[ARG0]]
! HLFIR: %[[RED:.*]] = acc.reduction varPtr(%{{.*}} : !fir.ref<!fir.array<?xf32>>) bounds(%{{.*}}) -> !fir.ref<!fir.array<?xf32>> {name = "a"}
! HLFIR: acc.parallel reduction(@"reduction_max_ref_?xf32" -> %[[RED]] : !fir.ref<!fir.array<?xf32>>) {
+
+subroutine acc_reduction_add_dynamic_extent_add_with_section(a)
+ integer :: a(:)
+ !$acc parallel reduction(+:a(2:4))
+ !$acc end parallel
+end subroutine
+
+! CHECK-LABEL: func.func @_QPacc_reduction_add_dynamic_extent_add_with_section(
+! CHECK-SAME: %[[ARG0:.*]]: !fir.box<!fir.array<?xi32>> {fir.bindc_name = "a"})
+! HLFIR: %[[DECL:.*]]:2 = hlfir.declare %[[[ARG0]]] {uniq_name = "_QFacc_reduction_add_dynamic_extent_add_with_sectionEa"} : (!fir.box<!fir.array<?xi32>>) -> (!fir.box<!fir.array<?xi32>>, !fir.box<!fir.array<?xi32>>)
+! HLFIR: %[[C1:.*]] = arith.constant 1 : index
+! HLFIR: %[[C3:.*]] = arith.constant 3 : index
+! HLFIR: %[[BOUND:.*]] = acc.bounds lowerbound(%[[C1]] : index) upperbound(%[[C3]] : index) stride(%{{.*}}#2 : index) startIdx(%{{.*}} : index) {strideInBytes = true}
+! HLFIR: %[[BOX_ADDR:.*]] = fir.box_addr %[[DECL]]#1 : (!fir.box<!fir.array<?xi32>>) -> !fir.ref<!fir.array<?xi32>>
+! HLFIR: %[[RED:.*]] = acc.reduction varPtr(%[[BOX_ADDR]] : !fir.ref<!fir.array<?xi32>>) bounds(%[[BOUND]]) -> !fir.ref<!fir.array<?xi32>> {name = "a(2:4)"}
+! HLFIR: acc.parallel reduction(@"reduction_add_section_lb3.ub1_ref_?xi32" -> %[[RED]] : !fir.ref<!fir.array<?xi32>>)
>From 3427b72dafa925a4135ff1a1fa47166b6b6b3ff6 Mon Sep 17 00:00:00 2001
From: Valentin Clement <clementval at gmail.com>
Date: Tue, 3 Oct 2023 13:21:16 -0700
Subject: [PATCH 2/2] Fix test
---
flang/test/Lower/OpenACC/acc-reduction.f90 | 8 +++-----
1 file changed, 3 insertions(+), 5 deletions(-)
diff --git a/flang/test/Lower/OpenACC/acc-reduction.f90 b/flang/test/Lower/OpenACC/acc-reduction.f90
index b3281587beb3778..381388c29e658fe 100644
--- a/flang/test/Lower/OpenACC/acc-reduction.f90
+++ b/flang/test/Lower/OpenACC/acc-reduction.f90
@@ -1119,10 +1119,8 @@ subroutine acc_reduction_add_dynamic_extent_add_with_section(a)
! CHECK-LABEL: func.func @_QPacc_reduction_add_dynamic_extent_add_with_section(
! CHECK-SAME: %[[ARG0:.*]]: !fir.box<!fir.array<?xi32>> {fir.bindc_name = "a"})
-! HLFIR: %[[DECL:.*]]:2 = hlfir.declare %[[[ARG0]]] {uniq_name = "_QFacc_reduction_add_dynamic_extent_add_with_sectionEa"} : (!fir.box<!fir.array<?xi32>>) -> (!fir.box<!fir.array<?xi32>>, !fir.box<!fir.array<?xi32>>)
-! HLFIR: %[[C1:.*]] = arith.constant 1 : index
-! HLFIR: %[[C3:.*]] = arith.constant 3 : index
-! HLFIR: %[[BOUND:.*]] = acc.bounds lowerbound(%[[C1]] : index) upperbound(%[[C3]] : index) stride(%{{.*}}#2 : index) startIdx(%{{.*}} : index) {strideInBytes = true}
+! HLFIR: %[[DECL:.*]]:2 = hlfir.declare %[[ARG0]] {uniq_name = "_QFacc_reduction_add_dynamic_extent_add_with_sectionEa"} : (!fir.box<!fir.array<?xi32>>) -> (!fir.box<!fir.array<?xi32>>, !fir.box<!fir.array<?xi32>>)
+! HLFIR: %[[BOUND:.*]] = acc.bounds lowerbound(%c1{{.*}} : index) upperbound(%c3{{.*}} : index) stride(%{{.*}}#2 : index) startIdx(%{{.*}} : index) {strideInBytes = true}
! HLFIR: %[[BOX_ADDR:.*]] = fir.box_addr %[[DECL]]#1 : (!fir.box<!fir.array<?xi32>>) -> !fir.ref<!fir.array<?xi32>>
! HLFIR: %[[RED:.*]] = acc.reduction varPtr(%[[BOX_ADDR]] : !fir.ref<!fir.array<?xi32>>) bounds(%[[BOUND]]) -> !fir.ref<!fir.array<?xi32>> {name = "a(2:4)"}
-! HLFIR: acc.parallel reduction(@"reduction_add_section_lb3.ub1_ref_?xi32" -> %[[RED]] : !fir.ref<!fir.array<?xi32>>)
+! HLFIR: acc.parallel reduction(@"reduction_add_section_lb1.ub3_ref_?xi32" -> %[[RED]] : !fir.ref<!fir.array<?xi32>>)
More information about the flang-commits
mailing list