[flang] lower RANK intrinsic #93694
Merged
[flang] lower RANK intrinsic #93694
Conversation
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
llvmbot
added
flang
|
@llvm/pr-subscribers-flang-codegen @llvm/pr-subscribers-flang-fir-hlfir Author: None (jeanPerier) ChangesFirst commit is reviewed in #93682. Lower RANK using fir.box_rank. This patches updates fir.box_rank to accept box reference, this avoids the need of generating an assumed-rank fir.load just for the sake of reading ALLOCATABLE/POINTER rank. The fir.load would generate a "dynamic" memcpy that is hard to optimize without further knowledge. A read effect is conditionally given to the operation. Full diff: https://github.com/llvm/llvm-project/pull/93694.diff 10 Files Affected:
diff --git a/flang/include/flang/Optimizer/Builder/HLFIRTools.h b/flang/include/flang/Optimizer/Builder/HLFIRTools.h
index 43aa1661550ec..6dce89d2ecabb 100644
--- a/flang/include/flang/Optimizer/Builder/HLFIRTools.h
+++ b/flang/include/flang/Optimizer/Builder/HLFIRTools.h
@@ -334,6 +334,9 @@ void genLengthParameters(mlir::Location loc, fir::FirOpBuilder &builder,
mlir::Value genCharLength(mlir::Location loc, fir::FirOpBuilder &builder,
Entity entity);
+mlir::Value genRank(mlir::Location loc, fir::FirOpBuilder &builder,
+ Entity entity, mlir::Type resultType);
+
/// Return the fir base, shape, and type parameters for a variable. Note that
/// type parameters are only added if the entity is not a box and the type
/// parameters is not a constant in the base type. This matches the arguments
diff --git a/flang/include/flang/Optimizer/Dialect/FIROps.td b/flang/include/flang/Optimizer/Dialect/FIROps.td
index 584b7e82bf27a..3afc97475db11 100644
--- a/flang/include/flang/Optimizer/Dialect/FIROps.td
+++ b/flang/include/flang/Optimizer/Dialect/FIROps.td
@@ -62,6 +62,12 @@ class fir_SimpleOneResultOp<string mnemonic, list<Trait> traits = []> :
let builders = [fir_OneResultOpBuilder];
}
+// Whether a type is a BaseBoxType or a reference to a BaseBoxType.
+def IsBoxAddressOrValueTypePred
+ : CPred<"::fir::isBoxAddressOrValue($_self)">;
+def fir_BoxAddressOrValueType : Type<IsBoxAddressOrValueTypePred,
+ "fir.box or fir.class type or reference">;
+
//===----------------------------------------------------------------------===//
// Memory SSA operations
//===----------------------------------------------------------------------===//
@@ -1205,7 +1211,8 @@ def fir_BoxProcHostOp : fir_SimpleOp<"boxproc_host", [NoMemoryEffect]> {
let results = (outs fir_ReferenceType);
}
-def fir_BoxRankOp : fir_SimpleOneResultOp<"box_rank", [NoMemoryEffect]> {
+def fir_BoxRankOp : fir_SimpleOneResultOp<"box_rank",
+ [DeclareOpInterfaceMethods<MemoryEffectsOpInterface>]> {
let summary = "return the number of dimensions for the boxed value";
let description = [{
@@ -1222,7 +1229,7 @@ def fir_BoxRankOp : fir_SimpleOneResultOp<"box_rank", [NoMemoryEffect]> {
descriptor may be either an array or a scalar, so the value is nonnegative.
}];
- let arguments = (ins fir_BoxType:$val);
+ let arguments = (ins fir_BoxAddressOrValueType:$box);
let results = (outs AnyIntegerType);
}
diff --git a/flang/lib/Lower/Allocatable.cpp b/flang/lib/Lower/Allocatable.cpp
index 61f4bbd856a8a..068f5d25967c9 100644
--- a/flang/lib/Lower/Allocatable.cpp
+++ b/flang/lib/Lower/Allocatable.cpp
@@ -831,7 +831,7 @@ genDeallocate(fir::FirOpBuilder &builder,
const Fortran::semantics::Symbol *symbol = nullptr) {
bool isCudaSymbol = symbol && Fortran::semantics::HasCUDAAttr(*symbol);
// Deallocate intrinsic types inline.
- if (!box.isDerived() && !box.isPolymorphic() &&
+ if (!box.isDerived() && !box.isPolymorphic() && !box.hasAssumedRank() &&
!box.isUnlimitedPolymorphic() && !errorManager.hasStatSpec() &&
!useAllocateRuntime && !box.isPointer() && !isCudaSymbol) {
// Pointers must use PointerDeallocate so that their deallocations
diff --git a/flang/lib/Lower/ConvertExprToHLFIR.cpp b/flang/lib/Lower/ConvertExprToHLFIR.cpp
index 3c305955520e2..a4352ee3359e5 100644
--- a/flang/lib/Lower/ConvertExprToHLFIR.cpp
+++ b/flang/lib/Lower/ConvertExprToHLFIR.cpp
@@ -1626,7 +1626,7 @@ class HlfirBuilder {
return castResult(
hlfir::genExtent(loc, builder, entity, desc.dimension()));
case Fortran::evaluate::DescriptorInquiry::Field::Rank:
- TODO(loc, "rank inquiry on assumed rank");
+ return castResult(hlfir::genRank(loc, builder, entity, resultType));
case Fortran::evaluate::DescriptorInquiry::Field::Stride:
// So far the front end does not generate this inquiry.
TODO(loc, "stride inquiry");
diff --git a/flang/lib/Lower/ConvertVariable.cpp b/flang/lib/Lower/ConvertVariable.cpp
index 8e9c1d640c330..c15d6b682bdb6 100644
--- a/flang/lib/Lower/ConvertVariable.cpp
+++ b/flang/lib/Lower/ConvertVariable.cpp
@@ -1901,8 +1901,6 @@ void Fortran::lower::mapSymbolAttributes(
// First deal with pointers and allocatables, because their handling here
// is the same regardless of their rank.
if (Fortran::semantics::IsAllocatableOrPointer(sym)) {
- if (isAssumedRank)
- TODO(loc, "assumed-rank pointer or allocatable");
// Get address of fir.box describing the entity.
// global
mlir::Value boxAlloc = preAlloc;
@@ -1910,6 +1908,7 @@ void Fortran::lower::mapSymbolAttributes(
if (!boxAlloc)
if (Fortran::lower::SymbolBox symbox = symMap.lookupSymbol(sym))
boxAlloc = symbox.getAddr();
+ assert((boxAlloc || !isAssumedRank) && "assumed-ranks cannot be local");
// local
if (!boxAlloc)
boxAlloc = createNewLocal(converter, loc, var, preAlloc);
diff --git a/flang/lib/Optimizer/Builder/HLFIRTools.cpp b/flang/lib/Optimizer/Builder/HLFIRTools.cpp
index 511585dc76894..e14e635a7f46b 100644
--- a/flang/lib/Optimizer/Builder/HLFIRTools.cpp
+++ b/flang/lib/Optimizer/Builder/HLFIRTools.cpp
@@ -642,6 +642,15 @@ mlir::Value hlfir::genCharLength(mlir::Location loc, fir::FirOpBuilder &builder,
return lenParams[0];
}
+mlir::Value hlfir::genRank(mlir::Location loc, fir::FirOpBuilder &builder,
+ hlfir::Entity entity, mlir::Type resultType) {
+ if (!entity.isAssumedRank())
+ return builder.createIntegerConstant(loc, resultType, entity.getRank());
+ assert(entity.isBoxAddressOrValue() &&
+ "assumed-ranks are box addresses or values");
+ return builder.create<fir::BoxRankOp>(loc, resultType, entity);
+}
+
// Return a "shape" that can be used in fir.embox/fir.rebox with \p exv base.
static mlir::Value asEmboxShape(mlir::Location loc, fir::FirOpBuilder &builder,
const fir::ExtendedValue &exv,
diff --git a/flang/lib/Optimizer/CodeGen/CodeGen.cpp b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
index 664453ebaf2f7..0ec667fc8314c 100644
--- a/flang/lib/Optimizer/CodeGen/CodeGen.cpp
+++ b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
@@ -428,7 +428,8 @@ struct BoxRankOpConversion : public fir::FIROpConversion<fir::BoxRankOp> {
mlir::Value a = adaptor.getOperands()[0];
auto loc = boxrank.getLoc();
mlir::Type ty = convertType(boxrank.getType());
- TypePair boxTyPair = getBoxTypePair(boxrank.getVal().getType());
+ TypePair boxTyPair =
+ getBoxTypePair(fir::unwrapRefType(boxrank.getBox().getType()));
mlir::Value rank = getRankFromBox(loc, boxTyPair, a, rewriter);
mlir::Value result = integerCast(loc, rewriter, ty, rank);
rewriter.replaceOp(boxrank, result);
diff --git a/flang/lib/Optimizer/Dialect/FIROps.cpp b/flang/lib/Optimizer/Dialect/FIROps.cpp
index 998e9535582cb..b541b7cdc7a5b 100644
--- a/flang/lib/Optimizer/Dialect/FIROps.cpp
+++ b/flang/lib/Optimizer/Dialect/FIROps.cpp
@@ -675,6 +675,20 @@ mlir::Type fir::BoxDimsOp::getTupleType() {
return mlir::TupleType::get(getContext(), triple);
}
+//===----------------------------------------------------------------------===//
+// BoxRankOp
+//===----------------------------------------------------------------------===//
+
+void fir::BoxRankOp::getEffects(
+ llvm::SmallVectorImpl<
+ mlir::SideEffects::EffectInstance<mlir::MemoryEffects::Effect>>
+ &effects) {
+ mlir::Value inputBox = getBox();
+ if (fir::isBoxAddress(inputBox.getType()))
+ effects.emplace_back(mlir::MemoryEffects::Read::get(), inputBox,
+ mlir::SideEffects::DefaultResource::get());
+}
+
//===----------------------------------------------------------------------===//
// CallOp
//===----------------------------------------------------------------------===//
diff --git a/flang/test/Fir/convert-to-llvm.fir b/flang/test/Fir/convert-to-llvm.fir
index 70cb0443e9a64..2a927b647b88a 100644
--- a/flang/test/Fir/convert-to-llvm.fir
+++ b/flang/test/Fir/convert-to-llvm.fir
@@ -945,6 +945,18 @@ func.func @extract_rank(%arg0: !fir.box<!fir.array<*:f64>>) -> i32 {
// CHECK: %[[RANK:.*]] = llvm.sext %[[RAW_RANK]] : i8 to i32
// CHECK: llvm.return %[[RANK]] : i32
+func.func @extract_rank2(%arg0: !fir.ref<!fir.box<!fir.array<*:f64>>>) -> i32 {
+ %0 = fir.box_rank %arg0 : (!fir.ref<!fir.box<!fir.array<*:f64>>>) -> i32
+ return %0 : i32
+}
+
+// CHECK-LABEL: llvm.func @extract_rank2(
+// CHECK-SAME: %[[ARG0:.*]]: !llvm.ptr) -> i32
+// CHECK: %[[GEP:.*]] = llvm.getelementptr %[[ARG0]][0, 3] : (!llvm.ptr) -> !llvm.ptr, !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}})>
+// CHECK: %[[RAW_RANK:.*]] = llvm.load %[[GEP]] : !llvm.ptr -> i8
+// CHECK: %[[RANK:.*]] = llvm.sext %[[RAW_RANK]] : i8 to i32
+// CHECK: llvm.return %[[RANK]] : i32
+
// -----
// Test `fir.box_addr` conversion.
diff --git a/flang/test/Lower/HLFIR/convert-variable-assumed-rank.f90 b/flang/test/Lower/HLFIR/convert-variable-assumed-rank.f90
index 748c15be84496..cd65696ca5ede 100644
--- a/flang/test/Lower/HLFIR/convert-variable-assumed-rank.f90
+++ b/flang/test/Lower/HLFIR/convert-variable-assumed-rank.f90
@@ -32,6 +32,23 @@ subroutine test_with_attrs(x)
real, target, optional :: x(..)
call takes_real(x)
end subroutine
+
+subroutine test_simple_allocatable(x)
+ real, allocatable :: x(..)
+end subroutine
+
+subroutine test_simple_pointer(x)
+ real, pointer :: x(..)
+end subroutine
+
+subroutine test_intentout(x)
+ real, intent(out), allocatable :: x(..)
+end subroutine
+
+subroutine test_assumed_length_alloc(x)
+ character(*), allocatable :: x(..)
+end subroutine
+
! CHECK-LABEL: func.func @_QMassumed_rank_testsPtest_intrinsic(
! CHECK-SAME: %[[VAL_0:.*]]: !fir.box<!fir.array<*:f32>> {fir.bindc_name = "x"}) {
! CHECK: %[[VAL_1:.*]] = fir.dummy_scope : !fir.dscope
@@ -67,4 +84,45 @@ subroutine test_with_attrs(x)
! CHECK: %[[VAL_1:.*]] = fir.dummy_scope : !fir.dscope
! CHECK: %[[VAL_2:.*]]:2 = hlfir.declare %[[VAL_0]] dummy_scope %[[VAL_1]] {fortran_attrs = #fir.var_attrs<optional, target>, uniq_name = "_QMassumed_rank_testsFtest_with_attrsEx"} : (!fir.box<!fir.array<*:f32>>, !fir.dscope) -> (!fir.box<!fir.array<*:f32>>, !fir.box<!fir.array<*:f32>>)
! CHECK: fir.call @_QPtakes_real(%[[VAL_2]]#0) fastmath<contract> : (!fir.box<!fir.array<*:f32>>) -> ()
+
+! CHECK-LABEL: func.func @_QMassumed_rank_testsPtest_simple_allocatable(
+! CHECK-SAME: %[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>> {fir.bindc_name = "x"}) {
+! CHECK: %[[VAL_1:.*]] = fir.dummy_scope : !fir.dscope
+! CHECK: %[[VAL_2:.*]]:2 = hlfir.declare %[[VAL_0]] dummy_scope %[[VAL_1]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMassumed_rank_testsFtest_simple_allocatableEx"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>>, !fir.dscope) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>>)
+! CHECK: return
+! CHECK: }
+
+! CHECK-LABEL: func.func @_QMassumed_rank_testsPtest_simple_pointer(
+! CHECK-SAME: %[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.ptr<!fir.array<*:f32>>>> {fir.bindc_name = "x"}) {
+! CHECK: %[[VAL_1:.*]] = fir.dummy_scope : !fir.dscope
+! CHECK: %[[VAL_2:.*]]:2 = hlfir.declare %[[VAL_0]] dummy_scope %[[VAL_1]] {fortran_attrs = #fir.var_attrs<pointer>, uniq_name = "_QMassumed_rank_testsFtest_simple_pointerEx"} : (!fir.ref<!fir.box<!fir.ptr<!fir.array<*:f32>>>>, !fir.dscope) -> (!fir.ref<!fir.box<!fir.ptr<!fir.array<*:f32>>>>, !fir.ref<!fir.box<!fir.ptr<!fir.array<*:f32>>>>)
+! CHECK: return
+! CHECK: }
+
+! CHECK-LABEL: func.func @_QMassumed_rank_testsPtest_intentout(
+! CHECK-SAME: %[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>> {fir.bindc_name = "x"}) {
+! CHECK: %[[VAL_1:.*]] = fir.dummy_scope : !fir.dscope
+! CHECK: %[[VAL_2:.*]]:2 = hlfir.declare %[[VAL_0]] dummy_scope %[[VAL_1]] {fortran_attrs = #fir.var_attrs<allocatable, intent_out>, uniq_name = "_QMassumed_rank_testsFtest_intentoutEx"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>>, !fir.dscope) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>>)
+! CHECK: %[[VAL_3:.*]] = fir.load %[[VAL_2]]#1 : !fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>>
+! CHECK: %[[VAL_4:.*]] = fir.box_addr %[[VAL_3]] : (!fir.box<!fir.heap<!fir.array<*:f32>>>) -> !fir.heap<!fir.array<*:f32>>
+! CHECK: %[[VAL_5:.*]] = fir.convert %[[VAL_4]] : (!fir.heap<!fir.array<*:f32>>) -> i64
+! CHECK: %[[VAL_6:.*]] = arith.constant 0 : i64
+! CHECK: %[[VAL_7:.*]] = arith.cmpi ne, %[[VAL_5]], %[[VAL_6]] : i64
+! CHECK: fir.if %[[VAL_7]] {
+! CHECK: %[[VAL_8:.*]] = arith.constant false
+! CHECK: %[[VAL_9:.*]] = fir.absent !fir.box<none>
+! CHECK: %[[VAL_12:.*]] = fir.convert %[[VAL_2]]#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<*:f32>>>>) -> !fir.ref<!fir.box<none>>
+! CHECK: %[[VAL_14:.*]] = fir.call @_FortranAAllocatableDeallocate(%[[VAL_12]], %[[VAL_8]], %[[VAL_9]], %{{.*}}, %{{.*}}) fastmath<contract> : (!fir.ref<!fir.box<none>>, i1, !fir.box<none>, !fir.ref<i8>, i32) -> i32
+! CHECK: }
+! CHECK: return
+! CHECK: }
+
+! CHECK-LABEL: func.func @_QMassumed_rank_testsPtest_assumed_length_alloc(
+! CHECK-SAME: %[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.heap<!fir.array<*:!fir.char<1,?>>>>> {fir.bindc_name = "x"}) {
+! CHECK: %[[VAL_1:.*]] = fir.dummy_scope : !fir.dscope
+! CHECK: %[[VAL_2:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.heap<!fir.array<*:!fir.char<1,?>>>>>
+! CHECK: %[[VAL_3:.*]] = fir.box_elesize %[[VAL_2]] : (!fir.box<!fir.heap<!fir.array<*:!fir.char<1,?>>>>) -> index
+! CHECK: %[[VAL_4:.*]]:2 = hlfir.declare %[[VAL_0]] typeparams %[[VAL_3]] dummy_scope %[[VAL_1]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMassumed_rank_testsFtest_assumed_length_allocEx"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<*:!fir.char<1,?>>>>>, index, !fir.dscope) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<*:!fir.char<1,?>>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<*:!fir.char<1,?>>>>>)
+! CHECK: return
+! CHECK: }
end module
|
jeanPerier
force-pushed
the
jp-assumed-rank-rank
branch
from
235ded0 to
77b5bdb
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
First commit is reviewed in #93682.
Lower RANK using fir.box_rank. This patches updates fir.box_rank to accept box reference, this avoids the need of generating an assumed-rank fir.load just for the sake of reading ALLOCATABLE/POINTER rank. The fir.load would generate a "dynamic" memcpy that is hard to optimize without further knowledge. A read effect is conditionally given to the operation.