Skip to content

[flang][cuda] Allocate derived-type with CUDA componement in managed memory #146797

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 2, 2025
Merged

[flang][cuda] Allocate derived-type with CUDA componement in managed memory #146797

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 25 additions & 2 deletions flang/lib/Lower/ConvertVariable.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -702,6 +702,29 @@ static void instantiateGlobal(Fortran::lower::AbstractConverter &converter,
mapSymbolAttributes(converter, var, symMap, stmtCtx, cast);
}

bool needCUDAAlloc(const Fortran::semantics::Symbol &sym) {
if (Fortran::semantics::IsDummy(sym))
return false;
if (const auto *details{
sym.GetUltimate()
.detailsIf<Fortran::semantics::ObjectEntityDetails>()}) {
if (details->cudaDataAttr() &&
(*details->cudaDataAttr() == Fortran::common::CUDADataAttr::Device ||
*details->cudaDataAttr() == Fortran::common::CUDADataAttr::Managed ||
*details->cudaDataAttr() == Fortran::common::CUDADataAttr::Unified ||
*details->cudaDataAttr() == Fortran::common::CUDADataAttr::Shared ||
*details->cudaDataAttr() == Fortran::common::CUDADataAttr::Pinned))
return true;
const Fortran::semantics::DeclTypeSpec *type{details->type()};
const Fortran::semantics::DerivedTypeSpec *derived{type ? type->AsDerived()
: nullptr};
if (derived)
if (FindCUDADeviceAllocatableUltimateComponent(*derived))
return true;
}
return false;
}

//===----------------------------------------------------------------===//
// Local variables instantiation (not for alias)
//===----------------------------------------------------------------===//
Expand Down Expand Up @@ -732,7 +755,7 @@ static mlir::Value createNewLocal(Fortran::lower::AbstractConverter &converter,
if (ultimateSymbol.test(Fortran::semantics::Symbol::Flag::CrayPointee))
return builder.create<fir::ZeroOp>(loc, fir::ReferenceType::get(ty));

if (Fortran::semantics::NeedCUDAAlloc(ultimateSymbol)) {
if (needCUDAAlloc(ultimateSymbol)) {
cuf::DataAttributeAttr dataAttr =
Fortran::lower::translateSymbolCUFDataAttribute(builder.getContext(),
ultimateSymbol);
Expand Down Expand Up @@ -1087,7 +1110,7 @@ static void instantiateLocal(Fortran::lower::AbstractConverter &converter,
Fortran::lower::defaultInitializeAtRuntime(converter, var.getSymbol(),
symMap);
auto *builder = &converter.getFirOpBuilder();
if (Fortran::semantics::NeedCUDAAlloc(var.getSymbol()) &&
if (needCUDAAlloc(var.getSymbol()) &&
!cuf::isCUDADeviceContext(builder->getRegion())) {
cuf::DataAttributeAttr dataAttr =
Fortran::lower::translateSymbolCUFDataAttribute(builder->getContext(),
Expand Down
15 changes: 13 additions & 2 deletions flang/lib/Semantics/tools.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1095,9 +1095,20 @@ bool IsDeviceAllocatable(const Symbol &symbol) {
}

std::optional<common::CUDADataAttr> GetCUDADataAttr(const Symbol *symbol) {
const auto *object{
const auto *details{
symbol ? symbol->detailsIf<ObjectEntityDetails>() : nullptr};
return object ? object->cudaDataAttr() : std::nullopt;
if (details) {
const Fortran::semantics::DeclTypeSpec *type{details->type()};
const Fortran::semantics::DerivedTypeSpec *derived{
type ? type->AsDerived() : nullptr};
if (derived) {
if (FindCUDADeviceAllocatableUltimateComponent(*derived)) {
return common::CUDADataAttr::Managed;
}
}
return details->cudaDataAttr();
}
return std::nullopt;
}

bool IsAccessible(const Symbol &original, const Scope &scope) {
Expand Down
14 changes: 12 additions & 2 deletions flang/test/Lower/CUDA/cuda-derived.cuf
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,16 @@ module m1

type t1; real, device, allocatable :: a(:); end type
type t2; type(t1) :: b; end type
contains
subroutine sub1()
type(ty_device) :: a
end subroutine

! CHECK-LABEL: func.func @_QMm1Psub1()
! CHECK: %[[ALLOC:.*]] = cuf.alloc !fir.type<_QMm1Tty_device{x:!fir.box<!fir.heap<!fir.array<?xi32>>>}> {bindc_name = "a", data_attr = #cuf.cuda<managed>, uniq_name = "_QMm1Fsub1Ea"} -> !fir.ref<!fir.type<_QMm1Tty_device{x:!fir.box<!fir.heap<!fir.array<?xi32>>>}>>
! CHECK: %[[DECL:.*]]:2 = hlfir.declare %[[ALLOC]] {data_attr = #cuf.cuda<managed>, uniq_name = "_QMm1Fsub1Ea"} : (!fir.ref<!fir.type<_QMm1Tty_device{x:!fir.box<!fir.heap<!fir.array<?xi32>>>}>>) -> (!fir.ref<!fir.type<_QMm1Tty_device{x:!fir.box<!fir.heap<!fir.array<?xi32>>>}>>, !fir.ref<!fir.type<_QMm1Tty_device{x:!fir.box<!fir.heap<!fir.array<?xi32>>>}>>)
! CHECK: cuf.free %[[DECL]]#0 : !fir.ref<!fir.type<_QMm1Tty_device{x:!fir.box<!fir.heap<!fir.array<?xi32>>>}>> {data_attr = #cuf.cuda<managed>}

end module

program main
Expand All @@ -16,5 +26,5 @@ program main
end

! CHECK-LABEL: func.func @_QQmain() attributes {fir.bindc_name = "main"}
! CHECK: %{{.*}} = fir.alloca !fir.type<_QMm1Tty_device{x:!fir.box<!fir.heap<!fir.array<?xi32>>>}> {bindc_name = "a", uniq_name = "_QFEa"}
! CHECK: %{{.*}} = fir.alloca !fir.type<_QMm1Tt2{b:!fir.type<_QMm1Tt1{a:!fir.box<!fir.heap<!fir.array<?xf32>>>}>}> {bindc_name = "b", uniq_name = "_QFEb"}
! CHECK: %{{.*}} = cuf.alloc !fir.type<_QMm1Tty_device{x:!fir.box<!fir.heap<!fir.array<?xi32>>>}> {bindc_name = "a", data_attr = #cuf.cuda<managed>, uniq_name = "_QFEa"}
! CHECK: %{{.*}} = cuf.alloc !fir.type<_QMm1Tt2{b:!fir.type<_QMm1Tt1{a:!fir.box<!fir.heap<!fir.array<?xf32>>>}>}> {bindc_name = "b", data_attr = #cuf.cuda<managed>, uniq_name = "_QFEb"}
Loading