Skip to content

[CIR] Upstream ArraySubscriptExpr from function parameter with pointer base #135493

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
Apr 15, 2025
Merged

[CIR] Upstream ArraySubscriptExpr from function parameter with pointer base #135493

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
650a240
[CIR] Upstream ArraySubscriptExpr for base type as pointer
AmrDeveloper Apr 11, 2025
2bb03de
Address code review comments
AmrDeveloper Apr 14, 2025
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
43 changes: 38 additions & 5 deletions clang/lib/CIR/CodeGen/CIRGenExpr.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -552,7 +552,19 @@ CIRGenFunction::emitArraySubscriptExpr(const clang::ArraySubscriptExpr *e) {
// in lexical order (this complexity is, sadly, required by C++17).
assert((e->getIdx() == e->getLHS() || e->getIdx() == e->getRHS()) &&
"index was neither LHS nor RHS");
const mlir::Value idx = emitScalarExpr(e->getIdx());

auto emitIdxAfterBase = [&]() -> mlir::Value {
const mlir::Value idx = emitScalarExpr(e->getIdx());

// Extend or truncate the index type to 32 or 64-bits.
auto ptrTy = mlir::dyn_cast<cir::PointerType>(idx.getType());
if (ptrTy && mlir::isa<cir::IntType>(ptrTy.getPointee()))
cgm.errorNYI(e->getSourceRange(),
"emitArraySubscriptExpr: index type cast");
return idx;
};

const mlir::Value idx = emitIdxAfterBase();
if (const Expr *array = getSimpleArrayDecayOperand(e->getBase())) {
LValue arrayLV;
if (const auto *ase = dyn_cast<ArraySubscriptExpr>(array))
Expand All @@ -566,13 +578,34 @@ CIRGenFunction::emitArraySubscriptExpr(const clang::ArraySubscriptExpr *e) {
arrayLV.getAddress(), e->getType(), idx, cgm.getLoc(e->getExprLoc()),
/*shouldDecay=*/true);

return LValue::makeAddr(addr, e->getType(), LValueBaseInfo());
const LValue lv = LValue::makeAddr(addr, e->getType(), LValueBaseInfo());

if (getLangOpts().ObjC && getLangOpts().getGC() != LangOptions::NonGC) {
cgm.errorNYI(e->getSourceRange(), "emitArraySubscriptExpr: ObjC with GC");
}

return lv;
}

// The base must be a pointer; emit it with an estimate of its alignment.
cgm.errorNYI(e->getSourceRange(),
"emitArraySubscriptExpr: The base must be a pointer");
return {};
assert(e->getBase()->getType()->isPointerType() &&
"The base must be a pointer");

LValueBaseInfo eltBaseInfo;
const Address ptrAddr = emitPointerWithAlignment(e->getBase(), &eltBaseInfo);
// Propagate the alignment from the array itself to the result.
Copy link
Contributor

Choose a reason for hiding this comment

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

What happened to the call to EmitIdxAfterBase()? It looks like you've simplified the code here (and in the earlier implementation), but I'm not sure we don't need some of what has been eliminated. At least when we add support for sanitizers we'll want the bounds check in emitIdxAfterBase.

Copy link
Member Author

Choose a reason for hiding this comment

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

I think yes, I should keep it as lambda for the bounds checks in the future, I was thinking also of keeping it as a function but for now i will keep emitIdxAfterBase

const Address addxr = emitArraySubscriptPtr(
*this, cgm.getLoc(e->getBeginLoc()), cgm.getLoc(e->getEndLoc()), ptrAddr,
e->getType(), idx, cgm.getLoc(e->getExprLoc()),
/*shouldDecay=*/false);

const LValue lv = LValue::makeAddr(addxr, e->getType(), eltBaseInfo);

if (getLangOpts().ObjC && getLangOpts().getGC() != LangOptions::NonGC) {
cgm.errorNYI(e->getSourceRange(), "emitArraySubscriptExpr: ObjC with GC");
}

return lv;
}

LValue CIRGenFunction::emitBinaryOperatorLValue(const BinaryOperator *e) {
Expand Down
122 changes: 110 additions & 12 deletions clang/test/CIR/CodeGen/array.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -350,20 +350,118 @@ void func7() {
// OGCG: %[[ARR:.*]] = alloca [1 x ptr], align 8
// OGCG: call void @llvm.memset.p0.i64(ptr align 8 %[[ARR]], i8 0, i64 8, i1 false)

void func8(int p[10]) {}
// CIR: cir.func @func8(%arg0: !cir.ptr<!s32i>
// CIR: cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, ["p", init]
void func8(int arr[10]) {
Copy link
Contributor

Choose a reason for hiding this comment

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

Can you add a test case where the parameter is a raw pointer. Something like this?

void func10(int *a) {
  int e = a[5];
}

Copy link
Member Author

Choose a reason for hiding this comment

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

Done

int e = arr[0];
int e2 = arr[1];
}

// LLVM: define void @func8(ptr {{%.*}})
// LLVM-NEXT: alloca ptr, i64 1, align 8
// CIR: cir.func @func8(%[[ARG:.*]]: !cir.ptr<!s32i>
// CIR: %[[ARR:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, ["arr", init]
// CIR: %[[INIT:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, ["e", init]
// CIR: %[[INIT_2:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, ["e2", init]
// CIR: cir.store %[[ARG]], %[[ARR]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
// CIR: %[[IDX:.*]] = cir.const #cir.int<0> : !s32i
// CIR: %[[TMP_1:.*]] = cir.load %[[ARR]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!s32i>
// CIR: %[[ELE_0:.*]] = cir.ptr_stride(%[[TMP_1]] : !cir.ptr<!s32i>, %[[IDX]] : !s32i), !cir.ptr<!s32i>
// CIR: %[[TMP_2:.*]] = cir.load %[[ELE_0]] : !cir.ptr<!s32i>, !s32i
// CIR: cir.store %[[TMP_2]], %[[INIT]] : !s32i, !cir.ptr<!s32i>
// CIR: %[[IDX_1:.*]] = cir.const #cir.int<1> : !s32i
// CIR: %[[TMP_3:.*]] = cir.load %[[ARR]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!s32i>
// CIR: %[[ELE_1:.*]] = cir.ptr_stride(%[[TMP_3]] : !cir.ptr<!s32i>, %[[IDX_1]] : !s32i), !cir.ptr<!s32i>
// CIR: %[[TMP_4:.*]] = cir.load %[[ELE_1]] : !cir.ptr<!s32i>, !s32i
// CIR: cir.store %[[TMP_4]], %[[INIT_2]] : !s32i, !cir.ptr<!s32i>

// LLVM: define void @func8(ptr %[[ARG:.*]])
// LLVM: %[[ARR:.*]] = alloca ptr, i64 1, align 8
// LLVM: %[[INIT:.*]] = alloca i32, i64 1, align 4
// LLVM: %[[INIT_2:.*]] = alloca i32, i64 1, align 4
// LLVM: store ptr %[[ARG]], ptr %[[ARR]], align 8
// LLVM: %[[TMP_1:.*]] = load ptr, ptr %[[ARR]], align 8
// LLVM: %[[ELE_0:.*]] = getelementptr i32, ptr %[[TMP_1]], i64 0
// LLVM: %[[TMP_2:.*]] = load i32, ptr %[[ELE_0]], align 4
// LLVM: store i32 %[[TMP_2]], ptr %[[INIT]], align 4
// LLVM: %[[TMP_3:.*]] = load ptr, ptr %[[ARR]], align 8
// LLVM: %[[ELE_1:.*]] = getelementptr i32, ptr %[[TMP_3]], i64 1
// LLVM: %[[TMP_4:.*]] = load i32, ptr %[[ELE_1]], align 4
// LLVM: store i32 %[[TMP_4]], ptr %[[INIT_2]], align 4

// OGCG: %[[ARR:.*]] = alloca ptr, align 8
// OGCG: %[[INIT:.*]] = alloca i32, align 4
// OGCG: %[[INIT_2:.*]] = alloca i32, align 4
// OGCG: store ptr {{%.*}}, ptr %[[ARR]], align 8
// OGCG: %[[TMP_1:.*]] = load ptr, ptr %[[ARR]], align 8
// OGCG: %[[ELE_0:.*]] = getelementptr inbounds i32, ptr %[[TMP_1]], i64 0
// OGCG: %[[TMP_2:.*]] = load i32, ptr %[[ELE_0]], align 4
// OGCG: store i32 %[[TMP_2]], ptr %[[INIT]], align 4
// OGCG: %[[TMP_3:.*]] = load ptr, ptr %[[ARR]], align 8
// OGCG: %[[ELE_1:.*]] = getelementptr inbounds i32, ptr %[[TMP_3]], i64 1
// OGCG: %[[TMP_2:.*]] = load i32, ptr %[[ELE_1]], align 4
// OGCG: store i32 %[[TMP_2]], ptr %[[INIT_2]], align 4

// OGCG: alloca ptr, align 8
void func9(int arr[10][5]) {
int e = arr[1][2];
}

void func9(int pp[10][5]) {}
// CIR: cir.func @func9(%arg0: !cir.ptr<!cir.array<!s32i x 5>>
// CIR: cir.alloca !cir.ptr<!cir.array<!s32i x 5>>, !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>
// CIR: cir.func @func9(%[[ARG:.*]]: !cir.ptr<!cir.array<!s32i x 5>>
// CIR: %[[ARR:.*]] = cir.alloca !cir.ptr<!cir.array<!s32i x 5>>, !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, ["arr", init]
// CIR: %[[INIT:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, ["e", init]
// CIR: cir.store %[[ARG]], %[[ARR]] : !cir.ptr<!cir.array<!s32i x 5>>, !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>
// CIR: %[[IDX:.*]] = cir.const #cir.int<2> : !s32i
// CIR: %[[IDX_1:.*]] = cir.const #cir.int<1> : !s32i
// CIR: %[[TMP_1:.*]] = cir.load %[[ARR]] : !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, !cir.ptr<!cir.array<!s32i x 5>>
// CIR: %[[ARR_1:.*]] = cir.ptr_stride(%[[TMP_1]] : !cir.ptr<!cir.array<!s32i x 5>>, %[[IDX_1]] : !s32i), !cir.ptr<!cir.array<!s32i x 5>>
// CIR: %[[ARR_1_PTR:.*]] = cir.cast(array_to_ptrdecay, %[[ARR_1]] : !cir.ptr<!cir.array<!s32i x 5>>), !cir.ptr<!s32i>
// CIR: %[[ARR_1_2:.*]] = cir.ptr_stride(%[[ARR_1_PTR]] : !cir.ptr<!s32i>, %[[IDX]] : !s32i), !cir.ptr<!s32i>
// CIR: %[[TMP_2:.*]] = cir.load %[[ARR_1_2]] : !cir.ptr<!s32i>, !s32i
// CIR: cir.store %[[TMP_2]], %[[INIT]] : !s32i, !cir.ptr<!s32i>

// LLVM: define void @func9(ptr %[[ARG:.*]])
// LLVM: %[[ARR:.*]] = alloca ptr, i64 1, align 8
// LLVM: %[[INIT:.*]] = alloca i32, i64 1, align 4
// LLVM: store ptr %[[ARG]], ptr %[[ARR]], align 8
// LLVM: %[[TMP_1:.*]] = load ptr, ptr %[[ARR]], align 8
// LLVM: %[[ARR_1:.*]] = getelementptr [5 x i32], ptr %[[TMP_1]], i64 1
// LLVM: %[[ARR_1_PTR:.*]] = getelementptr i32, ptr %[[ARR_1]], i32 0
Copy link
Contributor

Choose a reason for hiding this comment

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

This is not related to the current PR, but it's unfortunate that we're generating a getelementptr for the array_to_ptrdecay cast. That seems entirely unnecessary.

Copy link
Member

Choose a reason for hiding this comment

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

Yea, seems like lowering could be slightly smarter and fold this right away

Copy link
Member Author

Choose a reason for hiding this comment

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

Yes, I can take a look and handle it in other PR

Copy link
Member

Choose a reason for hiding this comment

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

You might need to change similar in the incubator

// LLVM: %[[ARR_1_2:.*]] = getelementptr i32, ptr %[[ARR_1_PTR]], i64 2
// LLVM: %[[TMP_2:.*]] = load i32, ptr %[[ARR_1_2]], align 4
// LLVM: store i32 %[[TMP_2]], ptr %[[INIT]], align 4

// OGCG: %[[ARR:.*]] = alloca ptr, align 8
// OGCG: %[[INIT:.*]] = alloca i32, align 4
// OGCG: store ptr {{%.*}}, ptr %[[ARR]], align 8
// OGCG: %[[TMP_1:.*]] = load ptr, ptr %[[ARR]], align 8
// OGCG: %[[ARR_1:.*]] = getelementptr inbounds [5 x i32], ptr %[[TMP_1]], i64 1
// OGCG: %[[ARR_1_2:.*]] = getelementptr inbounds [5 x i32], ptr %[[ARR_1]], i64 0, i64 2
// OGCG: %[[TMP_2:.*]] = load i32, ptr %[[ARR_1_2]], align 4
// OGCG: store i32 %[[TMP_2]], ptr %[[INIT]], align 4

void func10(int *a) {
int e = a[5];
}

// LLVM: define void @func9(ptr {{%.*}})
// LLVM-NEXT: alloca ptr, i64 1, align 8
// CIR: cir.func @func10(%[[ARG:.*]]: !cir.ptr<!s32i>
// CIR: %[[ARR:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, ["a", init]
// CIR: %[[INIT:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, ["e", init]
// CIR: cir.store %[[ARG]], %[[ARR]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
// CIR: %[[IDX:.*]] = cir.const #cir.int<5> : !s32i
// CIR: %[[TMP_1:.*]] = cir.load %[[ARR]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!s32i>
// CIR: %[[ELE:.*]] = cir.ptr_stride(%[[TMP_1]] : !cir.ptr<!s32i>, %[[IDX]] : !s32i), !cir.ptr<!s32i>
// CIR: %[[TMP_2:.*]] = cir.load %[[ELE]] : !cir.ptr<!s32i>, !s32i
// CIR: cir.store %[[TMP_2]], %[[INIT]] : !s32i, !cir.ptr<!s32i>

// LLVM: define void @func10(ptr %[[ARG:.*]]) {
// LLVM: %[[ARR:.*]] = alloca ptr, i64 1, align 8
// LLVM: %[[INIT:.*]] = alloca i32, i64 1, align 4
// LLVM: store ptr %[[ARG]], ptr %[[ARR]], align 8
// LLVM: %[[TMP_1:.*]] = load ptr, ptr %[[ARR]], align 8
// LLVM: %[[ELE:.*]] = getelementptr i32, ptr %[[TMP_1]], i64 5
// LLVM: %[[TMP_2:.*]] = load i32, ptr %[[ELE]], align 4
// LLVM: store i32 %[[TMP_2]], ptr %[[INIT]], align 4

// OGCG: alloca ptr, align 8
// OGCG: %[[ARR:.*]] = alloca ptr, align 8
// OGCG: %[[INIT:.*]] = alloca i32, align 4
// OGCG: store ptr {{%.*}}, ptr %[[ARR]], align 8
// OGCG: %[[TMP_1:.*]] = load ptr, ptr %[[ARR]], align 8
// OGCG: %[[ELE:.*]] = getelementptr inbounds i32, ptr %[[TMP_1]], i64 5
// OGCG: %[[TMP_2:.*]] = load i32, ptr %[[ELE]], align 4
// OGCG: store i32 %[[TMP_2]], ptr %[[INIT]], align 4
Loading