Skip to content

[llvm][OpenMP] Handle complex types in atomic read #111377

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 2 commits into from
Oct 23, 2024
Merged

[llvm][OpenMP] Handle complex types in atomic read #111377

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
82df319
[llvm][OpenMP] Handle complex types in atomic read
NimishMishra Oct 21, 2024
db1731b
Add FIXME
NimishMishra Oct 21, 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
16 changes: 15 additions & 1 deletion llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7836,7 +7836,7 @@ OpenMPIRBuilder::createAtomicRead(const LocationDescription &Loc,
"OMP Atomic expects a pointer to target memory");
Type *XElemTy = X.ElemTy;
assert((XElemTy->isFloatingPointTy() || XElemTy->isIntegerTy() ||
XElemTy->isPointerTy()) &&
XElemTy->isPointerTy() || XElemTy->isStructTy()) &&
"OMP atomic read expected a scalar type");

Value *XRead = nullptr;
Expand All @@ -7846,6 +7846,20 @@ OpenMPIRBuilder::createAtomicRead(const LocationDescription &Loc,
Builder.CreateLoad(XElemTy, X.Var, X.IsVolatile, "omp.atomic.read");
XLD->setAtomic(AO);
XRead = cast<Value>(XLD);
} else if (XElemTy->isStructTy()) {
// FIXME: Add checks to ensure __atomic_load is emitted iff the
// target does not support `atomicrmw` of the size of the struct
LoadInst *OldVal = Builder.CreateLoad(XElemTy, X.Var, "omp.atomic.read");
OldVal->setAtomic(AO);
const DataLayout &LoadDL = OldVal->getModule()->getDataLayout();
unsigned LoadSize =
LoadDL.getTypeStoreSize(OldVal->getPointerOperand()->getType());
OpenMPIRBuilder::AtomicInfo atomicInfo(
&Builder, XElemTy, LoadSize * 8, LoadSize * 8, OldVal->getAlign(),
OldVal->getAlign(), true /* UseLibcall */, X.Var);
auto AtomicLoadRes = atomicInfo.EmitAtomicLoadLibcall(AO);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Libcalls should only be used if the device does not support atomicrmw of that size. E.g. when the struct is only 4 bytes long, emit atomicrmw instead of call __atomic_load.

I think this is part of the larger issue of handling atomics properly. For the immediate problem, this should work, but I suggest to add a TODO/FIXME comment here.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@Meinersbur I have added a FIXME for this point. Thanks. If everything else looks okay, can we merge this?

XRead = AtomicLoadRes.first;
OldVal->eraseFromParent();
} else {
// We need to perform atomic op as integer
IntegerType *IntCastTy =
Expand Down
22 changes: 22 additions & 0 deletions mlir/test/Target/LLVMIR/openmp-llvm.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1525,6 +1525,28 @@ llvm.func @_QPomp_atomic_capture_complex() {

// -----

// CHECK-LABEL: define void @omp_atomic_read_complex() {
llvm.func @omp_atomic_read_complex(){

// CHECK: %[[a:.*]] = alloca { float, float }, i64 1, align 8
// CHECK: %[[b:.*]] = alloca { float, float }, i64 1, align 8
// CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
// CHECK: call void @__atomic_load(i64 8, ptr %[[b]], ptr %[[ATOMIC_TEMP_LOAD]], i32 0)
// CHECK: %[[LOADED_VAL:.*]] = load { float, float }, ptr %[[ATOMIC_TEMP_LOAD]], align 8
// CHECK: store { float, float } %[[LOADED_VAL]], ptr %[[a]], align 4
// CHECK: ret void
// CHECK: }

%0 = llvm.mlir.constant(1 : i64) : i64
%1 = llvm.alloca %0 x !llvm.struct<(f32, f32)> {bindc_name = "ib"} : (i64) -> !llvm.ptr
%2 = llvm.mlir.constant(1 : i64) : i64
%3 = llvm.alloca %2 x !llvm.struct<(f32, f32)> {bindc_name = "ia"} : (i64) -> !llvm.ptr
omp.atomic.read %1 = %3 : !llvm.ptr, !llvm.struct<(f32, f32)>
llvm.return
}

// -----

// Checking an order-dependent operation when the order is `expr binop x`
// CHECK-LABEL: @omp_atomic_update_ordering
// CHECK-SAME: (ptr %[[x:.*]], i32 %[[expr:.*]])
Expand Down
Loading