[llvm-branch-commits] [Flang][OpenMP] Derived type explicit allocatable member mapping (PR #111192)
via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Fri Oct 4 11:38:04 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-flang-fir-hlfir
Author: None (agozillon)
<details>
<summary>Changes</summary>
This PR is one of 3 in a PR stack, this is the primary change set which seeks
to extend the current derived type explicit member mapping support to
handle descriptor member mapping at arbitrary levels of nesting. The PR
stack seems to do this reasonably (from testing so far) but as you can
create quite complex mappings with derived types (in particular when adding
allocatable derived types or arrays of allocatable derived types) I imagine
there will be hiccups, which I am more than happy to address. There will
also be further extensions to this work to handle the implicit auto-magical
mapping of descriptor members in derived types and a few other changes
planned for the future (with some ideas on optimizing things).
The changes in this PR primarily occur in the OpenMP lowering and
the OMPMapInfoFinalization pass.
In the OpenMP lowering several utility functions were added or extended
to support the generation of appropriate intermediate member mappings
which are currently required when the parent (or multiple parents) of a
mapped member are descriptor types. We need to map the entirety of
these types or do a "deep copy" for lack of a better term, where we map
both the base address and the descriptor as without the copying of both
of these we lack the information in the case of the descriptor to access the
member or attach the pointers data to the pointer and in the latter case we
require the base address to map the chunk of data. Currently we do not
segment descriptor based derived types as we do with regular
non-descriptor derived types, we effectively map their entirety in all
cases at the moment, I hope to address this at some point in the future as
it adds a fair bit of a performance penalty to having nestings of allocatable
derived types as an example. The process of mapping all intermediate
descriptor members in a members path only occurs if a member has
an allocatable or object parent in its symbol path or the member itself
is a member or allocatable. This occurs in the
createParentSymAndGenIntermediateMaps function, which will also
generate the appropriate address for the allocatable member
within the derived type to use as a the varPtr field of the map (for
intermediate allocatable maps and final allocatable mappings). In
this case it's necessary as we can't utilise the usual Fortran::lower
functionality such as gatherDataOperandAddrAndBounds without
causing issues later in the lowering due to extra allocas being spawned
which seem to affect the pointer attachment (at least this is my
current assumption, it results in memory access errors on the device
due to incorrect map information generation). This is similar
to why we do not use the MLIR value generated for this and utilise
the original symbol provided when mapping descriptor types external
to derived types. Hopefully this can be rectified in the future so this
function can be simplified and more closely aligned to the other type
mappings. We also make use of fir::CoordinateOp as opposed to the
HLFIR version as the HLFIR version doesn't support the appropriate
lowering to FIR necessary at the moment, we also cannot use a
single CoordinateOp (similarly to a single GEP) as when we index
through a descriptor operation (BoxType) we encounter issues later
in the lowering, however in either case we need access to intermediate
descriptors so individual CoordinateOp's aid this (although, being
able to compress them into a smaller amount of CoordinateOp's may
simplify the IR and perhaps result in a better end product, something
to consider for the future).
The other large change area was in the OMPMapInfoFinalization pass,
where the pass had to be extended to support the expansion of box
types (or multiple nestings of box types) within derived types, or box
type derived types. This requires expanding each BoxType mapping
from one into two maps and then modifying all of the existing
member indices of the overarching parent mapping to account for
the addition of these new members alongside adjusting the existing
member indices to support the addition of these new maps which
extend the original member indices (as a base address of a box type
is currently considered a member of the box type at a position of
0 as when lowered to LLVM-IR it's a pointer contained at this position
in the descriptor type, however, this means extending mapped children
of this expanded descriptor type to additionally incorporate the new
member index in the correct location in its own index list). I believe
there is a reasonable amount of comments that should aid in
understanding this better, alongside the test alterations for the pass.
A subset of the changes were also aimed at making some of the utilities
for packing and unpacking the DenseIntElementsAttr
containing the member indices shareable across the lowering and
OMPMapInfoFinalization, this required moving some functions to the
Lower/Support/Utils.h header, and transforming the lowering structure
containing the member index data into something more similar to the
version used in OMPMapInfoFinalization. There we also some other
attempts at tidying things up in relation to the member index data
generation in the lowering, some of which required creating a logical
operator for the OpenMP ID class so it can be utilised as a map key
(it simply utilises the symbol address for the moment as ordering
isn't particularly important).
Otherwise I have added a set of new tests encompassing some of
the mappings currently supported by this PR (unfortunately as
you can have arbitrary nestings of all shapes and types it's not
very feasible to cover them all).
---
Patch is 271.45 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/111192.diff
18 Files Affected:
- (modified) flang/include/flang/Optimizer/Builder/FIRBuilder.h (+5)
- (modified) flang/lib/Lower/OpenMP/ClauseProcessor.cpp (+44-32)
- (modified) flang/lib/Lower/OpenMP/ClauseProcessor.h (+1-2)
- (modified) flang/lib/Lower/OpenMP/Clauses.h (+11)
- (modified) flang/lib/Lower/OpenMP/OpenMP.cpp (+2-2)
- (modified) flang/lib/Lower/OpenMP/Utils.cpp (+248-130)
- (modified) flang/lib/Lower/OpenMP/Utils.h (+81-24)
- (modified) flang/lib/Optimizer/Builder/FIRBuilder.cpp (+20)
- (modified) flang/lib/Optimizer/OpenMP/MapInfoFinalization.cpp (+310-121)
- (modified) flang/test/Fir/convert-to-llvm-openmp-and-fir.fir (+170-4)
- (modified) flang/test/Integration/OpenMP/map-types-and-sizes.f90 (+444-27)
- (modified) flang/test/Lower/OpenMP/allocatable-array-bounds.f90 (+3-3)
- (modified) flang/test/Lower/OpenMP/allocatable-map.f90 (+1-1)
- (modified) flang/test/Lower/OpenMP/array-bounds.f90 (+1-1)
- (modified) flang/test/Lower/OpenMP/declare-target-link-tarop-cap.f90 (+2-2)
- (added) flang/test/Lower/OpenMP/derived-type-allocatable-map.f90 (+161)
- (modified) flang/test/Lower/OpenMP/target.f90 (+2-2)
- (modified) flang/test/Transforms/omp-map-info-finalization.fir (+218-5)
``````````diff
diff --git a/flang/include/flang/Optimizer/Builder/FIRBuilder.h b/flang/include/flang/Optimizer/Builder/FIRBuilder.h
index 09f7b892f1ecbe..b772c523caba49 100644
--- a/flang/include/flang/Optimizer/Builder/FIRBuilder.h
+++ b/flang/include/flang/Optimizer/Builder/FIRBuilder.h
@@ -215,6 +215,11 @@ class FirOpBuilder : public mlir::OpBuilder, public mlir::OpBuilder::Listener {
llvm::ArrayRef<mlir::Value> lenParams,
bool asTarget = false);
+ /// Create a two dimensional ArrayAttr containing integer data as
+ /// IntegerAttrs, effectively: ArrayAttr<ArrayAttr<IntegerAttr>>>.
+ mlir::ArrayAttr create2DI64ArrayAttr(
+ llvm::SmallVectorImpl<llvm::SmallVector<int64_t>> &intData);
+
/// Create a temporary using `fir.alloca`. This function does not hoist.
/// It is the callers responsibility to set the insertion point if
/// hoisting is required.
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index a4d2524bccf5c3..209e79e5182634 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -889,16 +889,17 @@ void ClauseProcessor::processMapObjects(
lower::StatementContext &stmtCtx, mlir::Location clauseLocation,
const omp::ObjectList &objects,
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits,
- std::map<const semantics::Symbol *,
- llvm::SmallVector<OmpMapMemberIndicesData>> &parentMemberIndices,
+ std::map<Object, OmpMapParentAndMemberData> &parentMemberIndices,
llvm::SmallVectorImpl<mlir::Value> &mapVars,
llvm::SmallVectorImpl<const semantics::Symbol *> *mapSyms,
llvm::SmallVectorImpl<mlir::Location> *mapSymLocs,
llvm::SmallVectorImpl<mlir::Type> *mapSymTypes) const {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+
for (const omp::Object &object : objects) {
llvm::SmallVector<mlir::Value> bounds;
std::stringstream asFortran;
+ std::optional<omp::Object> parentObj;
lower::AddrAndBoundsInfo info =
lower::gatherDataOperandAddrAndBounds<mlir::omp::MapBoundsOp,
@@ -907,28 +908,46 @@ void ClauseProcessor::processMapObjects(
object.ref(), clauseLocation, asFortran, bounds,
treatIndexAsSection);
+ mlir::Value baseOp = info.rawInput;
+ if (object.sym()->owner().IsDerivedType()) {
+ omp::ObjectList objectList = gatherObjects(object, semaCtx);
+ assert(!objectList.empty() &&
+ "could not find parent objects of derived type member");
+ parentObj = objectList[0];
+ parentMemberIndices.emplace(parentObj.value(),
+ OmpMapParentAndMemberData{});
+
+ if (isMemberOrParentAllocatableOrPointer(object, semaCtx)) {
+ llvm::SmallVector<int64_t> indices;
+ generateMemberPlacementIndices(object, indices, semaCtx);
+ baseOp = createParentSymAndGenIntermediateMaps(
+ clauseLocation, converter, semaCtx, stmtCtx, objectList, indices,
+ parentMemberIndices[parentObj.value()], asFortran.str(),
+ mapTypeBits);
+ }
+ }
+
// Explicit map captures are captured ByRef by default,
// optimisation passes may alter this to ByCopy or other capture
// types to optimise
- mlir::Value baseOp = info.rawInput;
auto location = mlir::NameLoc::get(
mlir::StringAttr::get(firOpBuilder.getContext(), asFortran.str()),
baseOp.getLoc());
mlir::omp::MapInfoOp mapOp = createMapInfoOp(
firOpBuilder, location, baseOp,
/*varPtrPtr=*/mlir::Value{}, asFortran.str(), bounds,
- /*members=*/{}, /*membersIndex=*/mlir::DenseIntElementsAttr{},
+ /*members=*/{}, /*membersIndex=*/mlir::ArrayAttr{},
static_cast<
std::underlying_type_t<llvm::omp::OpenMPOffloadMappingFlags>>(
mapTypeBits),
mlir::omp::VariableCaptureKind::ByRef, baseOp.getType());
- if (object.sym()->owner().IsDerivedType()) {
- addChildIndexAndMapToParent(object, parentMemberIndices, mapOp, semaCtx);
+ if (parentObj.has_value()) {
+ addChildIndexAndMapToParent(
+ object, parentMemberIndices[parentObj.value()], mapOp, semaCtx);
} else {
mapVars.push_back(mapOp);
- if (mapSyms)
- mapSyms->push_back(object.sym());
+ mapSyms->push_back(object.sym());
if (mapSymTypes)
mapSymTypes->push_back(baseOp.getType());
if (mapSymLocs)
@@ -949,9 +968,7 @@ bool ClauseProcessor::processMap(
llvm::SmallVector<const semantics::Symbol *> localMapSyms;
llvm::SmallVectorImpl<const semantics::Symbol *> *ptrMapSyms =
mapSyms ? mapSyms : &localMapSyms;
- std::map<const semantics::Symbol *,
- llvm::SmallVector<OmpMapMemberIndicesData>>
- parentMemberIndices;
+ std::map<Object, OmpMapParentAndMemberData> parentMemberIndices;
bool clauseFound = findRepeatableClause<omp::clause::Map>(
[&](const omp::clause::Map &clause, const parser::CharBlock &source) {
@@ -1003,17 +1020,15 @@ bool ClauseProcessor::processMap(
mapSymLocs, mapSymTypes);
});
- insertChildMapInfoIntoParent(converter, parentMemberIndices, result.mapVars,
- *ptrMapSyms, mapSymTypes, mapSymLocs);
-
+ insertChildMapInfoIntoParent(converter, semaCtx, stmtCtx, parentMemberIndices,
+ result.mapVars, mapSymTypes, mapSymLocs,
+ ptrMapSyms);
return clauseFound;
}
bool ClauseProcessor::processMotionClauses(lower::StatementContext &stmtCtx,
mlir::omp::MapClauseOps &result) {
- std::map<const semantics::Symbol *,
- llvm::SmallVector<OmpMapMemberIndicesData>>
- parentMemberIndices;
+ std::map<Object, OmpMapParentAndMemberData> parentMemberIndices;
llvm::SmallVector<const semantics::Symbol *> mapSymbols;
auto callbackFn = [&](const auto &clause, const parser::CharBlock &source) {
@@ -1034,9 +1049,9 @@ bool ClauseProcessor::processMotionClauses(lower::StatementContext &stmtCtx,
clauseFound =
findRepeatableClause<omp::clause::From>(callbackFn) || clauseFound;
- insertChildMapInfoIntoParent(converter, parentMemberIndices, result.mapVars,
- mapSymbols,
- /*mapSymTypes=*/nullptr, /*mapSymLocs=*/nullptr);
+ insertChildMapInfoIntoParent(
+ converter, semaCtx, stmtCtx, parentMemberIndices, result.mapVars,
+ /*mapSymTypes=*/nullptr, /*mapSymLocs=*/nullptr, &mapSymbols);
return clauseFound;
}
@@ -1110,9 +1125,7 @@ bool ClauseProcessor::processUseDeviceAddr(
llvm::SmallVectorImpl<mlir::Type> &useDeviceTypes,
llvm::SmallVectorImpl<mlir::Location> &useDeviceLocs,
llvm::SmallVectorImpl<const semantics::Symbol *> &useDeviceSyms) const {
- std::map<const semantics::Symbol *,
- llvm::SmallVector<OmpMapMemberIndicesData>>
- parentMemberIndices;
+ std::map<Object, OmpMapParentAndMemberData> parentMemberIndices;
bool clauseFound = findRepeatableClause<omp::clause::UseDeviceAddr>(
[&](const omp::clause::UseDeviceAddr &clause,
const parser::CharBlock &source) {
@@ -1125,9 +1138,9 @@ bool ClauseProcessor::processUseDeviceAddr(
&useDeviceSyms, &useDeviceLocs, &useDeviceTypes);
});
- insertChildMapInfoIntoParent(converter, parentMemberIndices,
- result.useDeviceAddrVars, useDeviceSyms,
- &useDeviceTypes, &useDeviceLocs);
+ insertChildMapInfoIntoParent(converter, semaCtx, stmtCtx, parentMemberIndices,
+ result.useDeviceAddrVars, &useDeviceTypes,
+ &useDeviceLocs, &useDeviceSyms);
return clauseFound;
}
@@ -1136,9 +1149,8 @@ bool ClauseProcessor::processUseDevicePtr(
llvm::SmallVectorImpl<mlir::Type> &useDeviceTypes,
llvm::SmallVectorImpl<mlir::Location> &useDeviceLocs,
llvm::SmallVectorImpl<const semantics::Symbol *> &useDeviceSyms) const {
- std::map<const semantics::Symbol *,
- llvm::SmallVector<OmpMapMemberIndicesData>>
- parentMemberIndices;
+ std::map<Object, OmpMapParentAndMemberData> parentMemberIndices;
+
bool clauseFound = findRepeatableClause<omp::clause::UseDevicePtr>(
[&](const omp::clause::UseDevicePtr &clause,
const parser::CharBlock &source) {
@@ -1151,9 +1163,9 @@ bool ClauseProcessor::processUseDevicePtr(
&useDeviceSyms, &useDeviceLocs, &useDeviceTypes);
});
- insertChildMapInfoIntoParent(converter, parentMemberIndices,
- result.useDevicePtrVars, useDeviceSyms,
- &useDeviceTypes, &useDeviceLocs);
+ insertChildMapInfoIntoParent(converter, semaCtx, stmtCtx, parentMemberIndices,
+ result.useDevicePtrVars, &useDeviceTypes,
+ &useDeviceLocs, &useDeviceSyms);
return clauseFound;
}
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index 0c8e7bd47ab5a6..0f6f0c9863582d 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -178,8 +178,7 @@ class ClauseProcessor {
lower::StatementContext &stmtCtx, mlir::Location clauseLocation,
const omp::ObjectList &objects,
llvm::omp::OpenMPOffloadMappingFlags mapTypeBits,
- std::map<const semantics::Symbol *,
- llvm::SmallVector<OmpMapMemberIndicesData>> &parentMemberIndices,
+ std::map<Object, OmpMapParentAndMemberData> &parentMemberIndices,
llvm::SmallVectorImpl<mlir::Value> &mapVars,
llvm::SmallVectorImpl<const semantics::Symbol *> *mapSyms,
llvm::SmallVectorImpl<mlir::Location> *mapSymLocs = nullptr,
diff --git a/flang/lib/Lower/OpenMP/Clauses.h b/flang/lib/Lower/OpenMP/Clauses.h
index 51bf0eab0f8d07..34639673a1bd5b 100644
--- a/flang/lib/Lower/OpenMP/Clauses.h
+++ b/flang/lib/Lower/OpenMP/Clauses.h
@@ -55,6 +55,13 @@ struct IdTyTemplate {
return designator == other.designator;
}
+ // Defining an "ordering" which allows types derived from this to be
+ // utilised in maps and other containers that require comparison
+ // operators for ordering
+ bool operator<(const IdTyTemplate &other) const {
+ return symbol < other.symbol;
+ }
+
operator bool() const { return symbol != nullptr; }
};
@@ -76,6 +83,10 @@ struct ObjectT<Fortran::lower::omp::IdTyTemplate<Fortran::lower::omp::ExprTy>,
Fortran::semantics::Symbol *sym() const { return identity.symbol; }
const std::optional<ExprTy> &ref() const { return identity.designator; }
+ bool operator<(const ObjectT<IdTy, ExprTy> &other) const {
+ return identity < other.identity;
+ }
+
IdTy identity;
};
} // namespace tomp::type
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 8195f4a897a90b..a1a6d8816e4bc1 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -935,7 +935,7 @@ static void genBodyOfTargetOp(
firOpBuilder, copyVal.getLoc(), copyVal,
/*varPtrPtr=*/mlir::Value{}, name.str(), bounds,
/*members=*/llvm::SmallVector<mlir::Value>{},
- /*membersIndex=*/mlir::DenseIntElementsAttr{},
+ /*membersIndex=*/mlir::ArrayAttr{},
static_cast<
std::underlying_type_t<llvm::omp::OpenMPOffloadMappingFlags>>(
llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_IMPLICIT),
@@ -1792,7 +1792,7 @@ genTargetOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
mlir::Value mapOp = createMapInfoOp(
firOpBuilder, location, baseOp, /*varPtrPtr=*/mlir::Value{},
name.str(), bounds, /*members=*/{},
- /*membersIndex=*/mlir::DenseIntElementsAttr{},
+ /*membersIndex=*/mlir::ArrayAttr{},
static_cast<
std::underlying_type_t<llvm::omp::OpenMPOffloadMappingFlags>>(
mapFlag),
diff --git a/flang/lib/Lower/OpenMP/Utils.cpp b/flang/lib/Lower/OpenMP/Utils.cpp
index 47bc12e1b8a030..2e25f374934b94 100644
--- a/flang/lib/Lower/OpenMP/Utils.cpp
+++ b/flang/lib/Lower/OpenMP/Utils.cpp
@@ -13,9 +13,15 @@
#include "Utils.h"
#include "Clauses.h"
+#include <DirectivesCommon.h>
+
+#include <flang/Evaluate/fold.h>
#include <flang/Lower/AbstractConverter.h>
+#include <flang/Lower/ConvertExprToHLFIR.h>
#include <flang/Lower/ConvertType.h>
#include <flang/Lower/PFTBuilder.h>
+#include <flang/Lower/StatementContext.h>
+#include <flang/Lower/SymbolMap.h>
#include <flang/Optimizer/Builder/FIRBuilder.h>
#include <flang/Optimizer/Builder/Todo.h>
#include <flang/Parser/parse-tree.h>
@@ -23,9 +29,6 @@
#include <flang/Semantics/tools.h>
#include <llvm/Support/CommandLine.h>
-#include <algorithm>
-#include <numeric>
-
llvm::cl::opt<bool> treatIndexAsSection(
"openmp-treat-index-as-section",
llvm::cl::desc("In the OpenMP data clauses treat `a(N)` as `a(N:N)`."),
@@ -117,14 +120,12 @@ void gatherFuncAndVarSyms(
symbolAndClause.emplace_back(clause, *object.sym());
}
-mlir::omp::MapInfoOp
-createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location loc,
- mlir::Value baseAddr, mlir::Value varPtrPtr, std::string name,
- llvm::ArrayRef<mlir::Value> bounds,
- llvm::ArrayRef<mlir::Value> members,
- mlir::DenseIntElementsAttr membersIndex, uint64_t mapType,
- mlir::omp::VariableCaptureKind mapCaptureType, mlir::Type retTy,
- bool partialMap) {
+mlir::omp::MapInfoOp createMapInfoOp(
+ fir::FirOpBuilder &builder, mlir::Location loc, mlir::Value baseAddr,
+ mlir::Value varPtrPtr, std::string name, mlir::ArrayRef<mlir::Value> bounds,
+ mlir::ArrayRef<mlir::Value> members, mlir::ArrayAttr membersIndex,
+ uint64_t mapType, mlir::omp::VariableCaptureKind mapCaptureType,
+ mlir::Type retTy, bool partialMap) {
if (auto boxTy = llvm::dyn_cast<fir::BaseBoxType>(baseAddr.getType())) {
baseAddr = builder.create<fir::BoxAddrOp>(loc, baseAddr);
retTy = baseAddr.getType();
@@ -145,11 +146,174 @@ createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location loc,
builder.getIntegerAttr(builder.getIntegerType(64, false), mapType),
builder.getAttr<mlir::omp::VariableCaptureKindAttr>(mapCaptureType),
builder.getStringAttr(name), builder.getBoolAttr(partialMap));
-
return op;
}
-static int
+omp::ObjectList gatherObjects(omp::Object obj,
+ semantics::SemanticsContext &semaCtx) {
+ omp::ObjectList objList;
+ std::optional<omp::Object> baseObj = obj;
+ while (baseObj.has_value()) {
+ objList.push_back(baseObj.value());
+ baseObj = getBaseObject(baseObj.value(), semaCtx);
+ }
+ return omp::ObjectList{llvm::reverse(objList)};
+}
+
+bool isDuplicateMemberMapInfo(OmpMapParentAndMemberData &parentMembers,
+ llvm::SmallVectorImpl<int64_t> &memberIndices) {
+ for (auto memberData : parentMembers.memberPlacementIndices)
+ if (std::equal(memberIndices.begin(), memberIndices.end(),
+ memberData.begin()))
+ return true;
+ return false;
+}
+
+static void generateArrayIndices(lower::AbstractConverter &converter,
+ fir::FirOpBuilder &firOpBuilder,
+ lower::StatementContext &stmtCtx,
+ mlir::Location clauseLocation,
+ llvm::SmallVectorImpl<mlir::Value> &indices,
+ omp::Object object) {
+ if (auto maybeRef = evaluate::ExtractDataRef(*object.ref())) {
+ evaluate::DataRef ref = *maybeRef;
+ if (auto *arr = std::get_if<evaluate::ArrayRef>(&ref.u)) {
+ for (auto v : arr->subscript()) {
+ if (std::holds_alternative<Triplet>(v.u)) {
+ llvm_unreachable("Triplet indexing in map clause is unsupported");
+ } else {
+ auto expr =
+ std::get<Fortran::evaluate::IndirectSubscriptIntegerExpr>(v.u);
+ mlir::Value subscript = fir::getBase(
+ converter.genExprValue(toEvExpr(expr.value()), stmtCtx));
+ mlir::Value one = firOpBuilder.createIntegerConstant(
+ clauseLocation, firOpBuilder.getIndexType(), 1);
+ subscript = firOpBuilder.createConvert(
+ clauseLocation, firOpBuilder.getIndexType(), subscript);
+ indices.push_back(firOpBuilder.create<mlir::arith::SubIOp>(
+ clauseLocation, subscript, one));
+ }
+ }
+ }
+ }
+}
+
+// When mapping members of derived types, there is a chance that one of the
+// members along the way to a mapped member is an descriptor. In which case
+// we have to make sure we generate a map for those along the way otherwise
+// we will be missing a chunk of data required to actually map the member
+// type to device. This function effectively generates these maps and the
+// appropriate data accesses required to generate these maps. It will avoid
+// creating duplicate maps, as duplicates are just as bad as unmapped
+// descriptor data in a lot of cases for the runtime (and unnecessary
+// data movement should be avoided where possible)
+mlir::Value createParentSymAndGenIntermediateMaps(
+ mlir::Location clauseLocation, lower::AbstractConverter &converter,
+ semantics::SemanticsContext &semaCtx, lower::StatementContext &stmtCtx,
+ omp::ObjectList &objectList, llvm::SmallVector<int64_t> &indices,
+ OmpMapParentAndMemberData &parentMemberIndices, std::string asFortran,
+ llvm::omp::OpenMPOffloadMappingFlags mapTypeBits) {
+
+ auto arrayExprWithSubscript = [](omp::Object obj) {
+ if (auto maybeRef = evaluate::ExtractDataRef(*obj.ref())) {
+ evaluate::DataRef ref = *maybeRef;
+ if (auto *arr = std::get_if<evaluate::ArrayRef>(&ref.u))
+ return !arr->subscript().empty();
+ }
+ return false;
+ };
+
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ lower::AddrAndBoundsInfo parentBaseAddr = lower::getDataOperandBaseAddr(
+ converter, firOpBuilder, *objectList[0].sym(), clauseLocation);
+ mlir::Value curValue = parentBaseAddr.addr;
+
+ // Iterate over all objects in the objectList, this should consist of all
+ // record types between the parent and the member being mapped (including
+ // the parent). The object list may also contain array objects as well,
+ // this can occur when specifying bounds or a specific element access
+ // within a member map, we skip these.
+ size_t currentIndex = 0;
+ for (size_t i = 0; i < objectList.size(); ++i) {
+ if (fir::SequenceType arrType = mlir::dyn_cast<fir::SequenceType>(
+ fir::unwrapPassByRefType(curValue.getType()))) {
+ if (arrayExprWithSubscript(objectList[i])) {
+ llvm::SmallVector<mlir::Value> indices;
+ generateArrayIndices(converter, firOpBuilder, stmtCtx, clauseLocation,
+ indices, objectList[i]);
+ assert(!indices.empty() && "missing expected indices for map clause");
+ curValue = firOpBuilder.create<fir::CoordinateOp>(
+ clauseLocation, firOpBuilder.getRefType(arrType.getEleTy()),
+ curValue, indices);
+ }
+ }
+
+ if (fir::RecordType recordType = mlir::dyn_cast<fir::RecordType>(
+ fir::unwrapPassByRefType(curValue.getType()))) {
+ mlir::Value idxConst = firOpBuilder.createIntegerConstant(
+ clauseLocation, firOpBuilder.getIndexType(), indices[currentIndex]);
+ mlir::Type memberTy =
+ recordType.getTypeList().at(indices[currentIndex]).second;
+ curValue = firOpBuilder.create<fir::CoordinateOp>(
+ clauseLocation, firOpBuilder.getRefType(memberTy), curValue,
+ idxConst);
+
+ if ((currentIndex == indices.size() - 1) ||
+ !fir::isTypeWithDescriptor(memberTy)) {
+ currentIndex++;
+ continue;
+ }
+
+ llvm::SmallVector<int64_t> interimIndices(
+ indices.begin(), std::next(indices.begin(), currentIndex + 1));
+ if (!isDuplicateMemberMapInfo(parentMemberIndices, interimIndices)) {
+ // Generate initial bounds operations using the standard lowering
+ ...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/111192
More information about the llvm-branch-commits
mailing list