Skip to content

[flang][cuda] Lower attribute for module variables #81226

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 3 commits into from
Feb 9, 2024
Merged

[flang][cuda] Lower attribute for module variables #81226

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c3e2d9c
[flang][cuda] Lower attribute for module variables
clementval Feb 8, 2024
41cf949
do not use hardcoded string attr name
clementval Feb 9, 2024
0f77cf0
remove duplicate cuda_attr string
clementval Feb 9, 2024
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
6 changes: 4 additions & 2 deletions flang/include/flang/Optimizer/Builder/FIRBuilder.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -230,12 +230,14 @@ class FirOpBuilder : public mlir::OpBuilder, public mlir::OpBuilder::Listener {
llvm::StringRef name,
mlir::StringAttr linkage = {},
mlir::Attribute value = {}, bool isConst = false,
bool isTarget = false);
bool isTarget = false,
fir::CUDAAttributeAttr cudaAttr = {});

fir::GlobalOp createGlobal(mlir::Location loc, mlir::Type type,
llvm::StringRef name, bool isConst, bool isTarget,
std::function<void(FirOpBuilder &)> bodyBuilder,
mlir::StringAttr linkage = {});
mlir::StringAttr linkage = {},
fir::CUDAAttributeAttr cudaAttr = {});

/// Create a global constant (read-only) value.
fir::GlobalOp createGlobalConstant(mlir::Location loc, mlir::Type type,
Expand Down
3 changes: 2 additions & 1 deletion flang/include/flang/Optimizer/Dialect/FIROps.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2737,7 +2737,8 @@ def fir_GlobalOp : fir_Op<"global", [IsolatedFromAbove, Symbol]> {
OptionalAttr<AnyAttr>:$initVal,
OptionalAttr<UnitAttr>:$constant,
OptionalAttr<UnitAttr>:$target,
OptionalAttr<StrAttr>:$linkName
OptionalAttr<StrAttr>:$linkName,
OptionalAttr<fir_CUDAAttributeAttr>:$cuda_attr
);

let regions = (region AtMostRegion<1>:$region);
Expand Down
16 changes: 11 additions & 5 deletions flang/lib/Lower/ConvertVariable.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -138,7 +138,8 @@ static bool isConstant(const Fortran::semantics::Symbol &sym) {
static fir::GlobalOp defineGlobal(Fortran::lower::AbstractConverter &converter,
const Fortran::lower::pft::Variable &var,
llvm::StringRef globalName,
mlir::StringAttr linkage);
mlir::StringAttr linkage,
fir::CUDAAttributeAttr cudaAttr = {});

static mlir::Location genLocation(Fortran::lower::AbstractConverter &converter,
const Fortran::semantics::Symbol &sym) {
Expand Down Expand Up @@ -462,7 +463,8 @@ void Fortran::lower::createGlobalInitialization(
static fir::GlobalOp defineGlobal(Fortran::lower::AbstractConverter &converter,
const Fortran::lower::pft::Variable &var,
llvm::StringRef globalName,
mlir::StringAttr linkage) {
mlir::StringAttr linkage,
fir::CUDAAttributeAttr cudaAttr) {
fir::FirOpBuilder &builder = converter.getFirOpBuilder();
const Fortran::semantics::Symbol &sym = var.getSymbol();
mlir::Location loc = genLocation(converter, sym);
Expand Down Expand Up @@ -500,8 +502,9 @@ static fir::GlobalOp defineGlobal(Fortran::lower::AbstractConverter &converter,
}
}
if (!global)
global = builder.createGlobal(loc, symTy, globalName, linkage,
mlir::Attribute{}, isConst, var.isTarget());
global =
builder.createGlobal(loc, symTy, globalName, linkage, mlir::Attribute{},
isConst, var.isTarget(), cudaAttr);
if (Fortran::semantics::IsAllocatableOrPointer(sym) &&
!Fortran::semantics::IsProcedure(sym)) {
const auto *details =
Expand Down Expand Up @@ -2219,7 +2222,10 @@ void Fortran::lower::defineModuleVariable(
// Do nothing. Mapping will be done on user side.
} else {
std::string globalName = converter.mangleName(sym);
defineGlobal(converter, var, globalName, linkage);
fir::CUDAAttributeAttr cudaAttr =
Fortran::lower::translateSymbolCUDAAttribute(
converter.getFirOpBuilder().getContext(), sym);
defineGlobal(converter, var, globalName, linkage, cudaAttr);
}
}

Expand Down
23 changes: 14 additions & 9 deletions flang/lib/Optimizer/Builder/FIRBuilder.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -271,27 +271,32 @@ mlir::Value fir::FirOpBuilder::createHeapTemporary(

/// Create a global variable in the (read-only) data section. A global variable
/// must have a unique name to identify and reference it.
fir::GlobalOp fir::FirOpBuilder::createGlobal(mlir::Location loc,
mlir::Type type,
llvm::StringRef name,
mlir::StringAttr linkage,
mlir::Attribute value,
bool isConst, bool isTarget) {
fir::GlobalOp fir::FirOpBuilder::createGlobal(
mlir::Location loc, mlir::Type type, llvm::StringRef name,
mlir::StringAttr linkage, mlir::Attribute value, bool isConst,
bool isTarget, fir::CUDAAttributeAttr cudaAttr) {
auto module = getModule();
auto insertPt = saveInsertionPoint();
if (auto glob = module.lookupSymbol<fir::GlobalOp>(name))
return glob;
setInsertionPoint(module.getBody(), module.getBody()->end());
auto glob =
create<fir::GlobalOp>(loc, name, isConst, isTarget, type, value, linkage);
llvm::SmallVector<mlir::NamedAttribute> attrs;
if (cudaAttr) {
auto globalOpName = mlir::OperationName(fir::GlobalOp::getOperationName(),
module.getContext());
attrs.push_back(mlir::NamedAttribute(
fir::GlobalOp::getCudaAttrAttrName(globalOpName), cudaAttr));
}
auto glob = create<fir::GlobalOp>(loc, name, isConst, isTarget, type, value,
linkage, attrs);
restoreInsertionPoint(insertPt);
return glob;
}

fir::GlobalOp fir::FirOpBuilder::createGlobal(
mlir::Location loc, mlir::Type type, llvm::StringRef name, bool isConst,
bool isTarget, std::function<void(FirOpBuilder &)> bodyBuilder,
mlir::StringAttr linkage) {
mlir::StringAttr linkage, fir::CUDAAttributeAttr cudaAttr) {
auto module = getModule();
auto insertPt = saveInsertionPoint();
if (auto glob = module.lookupSymbol<fir::GlobalOp>(name))
Expand Down
51 changes: 30 additions & 21 deletions flang/test/Lower/CUDA/cuda-data-attribute.cuf
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,53 +3,62 @@

! Test lowering of CUDA attribute on variables.

module cuda_var
real, constant :: mod_a_rc
! CHECK: fir.global @_QMcuda_varEmod_a_rc {cuda_attr = #fir.cuda<constant>} : f32
real, device :: mod_b_ra
! CHECK: fir.global @_QMcuda_varEmod_b_ra {cuda_attr = #fir.cuda<device>} : f32
real, allocatable, managed :: mod_c_rm
! CHECK: fir.global @_QMcuda_varEmod_c_rm {cuda_attr = #fir.cuda<managed>} : !fir.box<!fir.heap<f32>>
real, allocatable, pinned :: mod_d_rp
! CHECK: fir.global @_QMcuda_varEmod_d_rp {cuda_attr = #fir.cuda<pinned>} : !fir.box<!fir.heap<f32>>

contains

subroutine local_var_attrs
real, constant :: rc
real, device :: rd
real, allocatable, managed :: rm
real, allocatable, pinned :: rp
end subroutine

! CHECK-LABEL: func.func @_QPlocal_var_attrs()
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<constant>, uniq_name = "_QFlocal_var_attrsErc"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<device>, uniq_name = "_QFlocal_var_attrsErd"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFlocal_var_attrsErm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFlocal_var_attrsErp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)
! CHECK-LABEL: func.func @_QMcuda_varPlocal_var_attrs()
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<constant>, uniq_name = "_QMcuda_varFlocal_var_attrsErc"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<device>, uniq_name = "_QMcuda_varFlocal_var_attrsErd"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFlocal_var_attrsErm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)
! CHECK: %{{.*}}:2 = hlfir.declare %{{.*}} {cuda_attr = #fir.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFlocal_var_attrsErp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)

! FIR: %{{.*}} = fir.declare %{{.*}} {cuda_attr = #fir.cuda<constant>, uniq_name = "_QFlocal_var_attrsErc"} : (!fir.ref<f32>) -> !fir.ref<f32>
! FIR: %{{.*}} = fir.declare %{{.*}} {cuda_attr = #fir.cuda<device>, uniq_name = "_QFlocal_var_attrsErd"} : (!fir.ref<f32>) -> !fir.ref<f32>
! FIR: %{{.*}} = fir.declare %{{.*}} {cuda_attr = #fir.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFlocal_var_attrsErm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> !fir.ref<!fir.box<!fir.heap<f32>>>
! FIR: %{{.*}} = fir.declare %{{.*}} {cuda_attr = #fir.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFlocal_var_attrsErp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> !fir.ref<!fir.box<!fir.heap<f32>>>
! FIR: %{{.*}} = fir.declare %{{.*}} {cuda_attr = #fir.cuda<constant>, uniq_name = "_QMcuda_varFlocal_var_attrsErc"} : (!fir.ref<f32>) -> !fir.ref<f32>
! FIR: %{{.*}} = fir.declare %{{.*}} {cuda_attr = #fir.cuda<device>, uniq_name = "_QMcuda_varFlocal_var_attrsErd"} : (!fir.ref<f32>) -> !fir.ref<f32>
! FIR: %{{.*}} = fir.declare %{{.*}} {cuda_attr = #fir.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFlocal_var_attrsErm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> !fir.ref<!fir.box<!fir.heap<f32>>>
! FIR: %{{.*}} = fir.declare %{{.*}} {cuda_attr = #fir.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFlocal_var_attrsErp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> !fir.ref<!fir.box<!fir.heap<f32>>>

subroutine dummy_arg_constant(dc)
real, constant :: dc
end subroutine
! CHECK-LABEL: func.func @_QPdummy_arg_constant(
! CHECK-LABEL: func.func @_QMcuda_varPdummy_arg_constant(
! CHECK-SAME: %[[ARG0:.*]]: !fir.ref<f32> {fir.bindc_name = "dc", fir.cuda_attr = #fir.cuda<constant>}
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] {cuda_attr = #fir.cuda<constant>, uniq_name = "_QFdummy_arg_constantEdc"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] {cuda_attr = #fir.cuda<constant>, uniq_name = "_QMcuda_varFdummy_arg_constantEdc"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)

subroutine dummy_arg_device(dd)
real, device :: dd
end subroutine
! CHECK-LABEL: func.func @_QPdummy_arg_device(
! CHECK-LABEL: func.func @_QMcuda_varPdummy_arg_device(
! CHECK-SAME: %[[ARG0:.*]]: !fir.ref<f32> {fir.bindc_name = "dd", fir.cuda_attr = #fir.cuda<device>}) {
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] {cuda_attr = #fir.cuda<device>, uniq_name = "_QFdummy_arg_deviceEdd"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] {cuda_attr = #fir.cuda<device>, uniq_name = "_QMcuda_varFdummy_arg_deviceEdd"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)

subroutine dummy_arg_managed(dm)
real, allocatable, managed :: dm
end subroutine
! CHECK-LABEL: func.func @_QPdummy_arg_managed(
! CHECK-LABEL: func.func @_QMcuda_varPdummy_arg_managed(
! CHECK-SAME: %[[ARG0:.*]]: !fir.ref<!fir.box<!fir.heap<f32>>> {fir.bindc_name = "dm", fir.cuda_attr = #fir.cuda<managed>}) {
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] {cuda_attr = #fir.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFdummy_arg_managedEdm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] {cuda_attr = #fir.cuda<managed>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFdummy_arg_managedEdm"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)

subroutine dummy_arg_pinned(dp)
real, allocatable, pinned :: dp
end subroutine
! CHECK-LABEL: func.func @_QPdummy_arg_pinned(
! CHECK-LABEL: func.func @_QMcuda_varPdummy_arg_pinned(
! CHECK-SAME: %[[ARG0:.*]]: !fir.ref<!fir.box<!fir.heap<f32>>> {fir.bindc_name = "dp", fir.cuda_attr = #fir.cuda<pinned>}) {
! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] {cuda_attr = #fir.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFdummy_arg_pinnedEdp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)




! CHECK: %{{.*}}:2 = hlfir.declare %[[ARG0]] {cuda_attr = #fir.cuda<pinned>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMcuda_varFdummy_arg_pinnedEdp"} : (!fir.ref<!fir.box<!fir.heap<f32>>>) -> (!fir.ref<!fir.box<!fir.heap<f32>>>, !fir.ref<!fir.box<!fir.heap<f32>>>)

end module