Skip to content

GH-55277: [C++][Gandiva] Fix stack overflow in decimal evaluation with JIT optimization disabled #55278

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

Open
dmitry-chirkov-dremio wants to merge 2 commits into
base: main
Choose a base branch
from
+46 −6
Open

GH-55277: [C++][Gandiva] Fix stack overflow in decimal evaluation with JIT optimization disabled #55278

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
ad4b916
GH-55277: [C++][Gandiva] Add regression for decimal cast stack overflow
dmitry-chirkov-dremio Oct 27, 2026
4757838
GH-55277: [C++][Gandiva] Allocate decimal return slots in the entry b…
dmitry-chirkov-dremio Oct 27, 2026
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: 10 additions & 6 deletions cpp/src/gandiva/decimal_ir.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -246,9 +246,10 @@ DecimalIR::ValueWithOverflow DecimalIR::AddWithOverflowCheck(const ValueFull& x,
// This is pretty complex, so use CPP fns.
llvm::Value* DecimalIR::AddLarge(const ValueFull& x, const ValueFull& y,
const ValueFull& out) {
auto block = ir_builder()->GetInsertBlock();
auto out_high_ptr = new llvm::AllocaInst(types()->i64_type(), 0, "out_hi", block);
auto out_low_ptr = new llvm::AllocaInst(types()->i64_type(), 0, "out_low", block);
auto& entry = ir_builder()->GetInsertBlock()->getParent()->getEntryBlock();

@balaatur balaatur Oct 28, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

Could this be merged with the other change ? It looks like the code is duplicated. Maybe via a helper function ?

llvm::IRBuilder<> entry_builder(&entry, entry.begin());
auto out_high_ptr = entry_builder.CreateAlloca(types()->i64_type(), nullptr, "out_hi");
auto out_low_ptr = entry_builder.CreateAlloca(types()->i64_type(), nullptr, "out_low");
auto x_split = ValueSplit::MakeFromInt128(this, x.value());
auto y_split = ValueSplit::MakeFromInt128(this, y.value());

Expand Down Expand Up @@ -454,9 +455,12 @@ llvm::Value* DecimalIR::CallDecimalFunction(const std::string& function_name,
llvm::Value* result = nullptr;
if (return_type == i128) {
// for i128 ret, replace with two int64* args, and join them.
auto block = ir_builder()->GetInsertBlock();
auto out_high_ptr = new llvm::AllocaInst(i64, 0, "out_hi", block);
auto out_low_ptr = new llvm::AllocaInst(i64, 0, "out_low", block);
// Decimal calls can be emitted inside the record loop or a CASE branch.
// Allocate return slots once per invocation, rather than once per row.
auto& entry = ir_builder()->GetInsertBlock()->getParent()->getEntryBlock();
llvm::IRBuilder<> entry_builder(&entry, entry.begin());
auto out_high_ptr = entry_builder.CreateAlloca(i64, nullptr, "out_hi");
auto out_low_ptr = entry_builder.CreateAlloca(i64, nullptr, "out_low");
dis_assembled_args.push_back(out_high_ptr);
dis_assembled_args.push_back(out_low_ptr);

Expand Down
36 changes: 36 additions & 0 deletions cpp/src/gandiva/tests/decimal_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,42 @@ std::vector<Decimal128> TestDecimal::MakeDecimalVector(std::vector<std::string>
return ret;
}

TEST_F(TestDecimal, CastLargeBatch) {
// Allocating decimal return slots in the record loop can exhaust the stack
// when optimization is disabled. A single int32-to-decimal cast is sufficient.
constexpr int kNumRecords = 1 << 19;
auto input_field = arrow::field("large_batch_decimal", arrow::int32());
auto schema = arrow::schema({input_field});
auto decimal_type = arrow::decimal128(10, 0);
auto decimal = TreeExprBuilder::MakeFunction(
"castDECIMAL", {TreeExprBuilder::MakeField(input_field)}, decimal_type);
auto expression =
TreeExprBuilder::MakeExpression(decimal, arrow::field("result", decimal_type));

std::vector<int32_t> values(kNumRecords);
std::vector<bool> validity(kNumRecords);
std::vector<Decimal128> expected_values(kNumRecords);
for (int i = 0; i < kNumRecords; ++i) {
values[i] = i % 17 - 8;
validity[i] = i % 11 != 0;
expected_values[i] = Decimal128(values[i]);
}
auto batch = arrow::RecordBatch::Make(schema, kNumRecords,
{MakeArrowArrayInt32(values, validity)});
auto expected = MakeArrowArrayDecimal(decimal_type, expected_values, validity);

for (bool optimize : {false, true}) {
SCOPED_TRACE(optimize);
auto config = std::make_shared<Configuration>(optimize);
std::shared_ptr<Projector> projector;
ASSERT_OK(Projector::Make(schema, {expression}, config, &projector));
arrow::ArrayVector output;
ASSERT_OK(projector->Evaluate(*batch, pool_, &output));
ASSERT_EQ(output.size(), 1);
EXPECT_ARROW_ARRAY_EQUALS(expected, output[0]);
}
}

TEST_F(TestDecimal, TestSimple) {
// schema for input fields
constexpr int32_t precision = 36;
Expand Down
Loading