Skip to content

Validate CUDA NMS mask size #32014

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
Akshay Sonawane (apsonawane) merged 1 commit into from
Aug 13, 2026
+39 −6
Merged

Validate CUDA NMS mask size #32014

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
18 changes: 12 additions & 6 deletions onnxruntime/core/providers/cuda/object_detection/non_max_suppression_impl.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,11 +16,14 @@

#include "core/providers/cuda/cu_inc/common.cuh"

#include <limits>

Check warning on line 19 in onnxruntime/core/providers/cuda/object_detection/non_max_suppression_impl.cu

View workflow job for this annotation

GitHub Actions / Optional Lint C++ 

[cpplint] reported by reviewdog 🐶
 Found C++ system header after other header. Should be: non_max_suppression_impl.h, c system, c++ system, other.  [build/include_order] [4]

Raw Output:
onnxruntime/core/providers/cuda/object_detection/non_max_suppression_impl.cu:19:  Found C++ system header after other header. Should be: non_max_suppression_impl.h, c system, c++ system, other.  [build/include_order] [4]

#include <thrust/count.h>
#include <thrust/device_vector.h>
#include <thrust/execution_policy.h>

#include "non_max_suppression_impl.h"
#include "core/common/safeint.h"
#include "core/providers/cpu/object_detection/non_max_suppression_helper.h"
#include "core/providers/cuda/cuda_common.h"

Expand Down Expand Up @@ -241,15 +244,18 @@
auto iptr = reinterpret_cast<std::uintptr_t>(d_sorted_boxes_float_ptr);
ORT_ENFORCE((iptr & 15) == 0);

const int bit_mask_len =
(num_boxes + kNmsBoxesPerThread - 1) / kNmsBoxesPerThread;
int max_nms_mask_size = num_boxes * bit_mask_len;
const int bit_mask_len = num_boxes == 0 ? 0 : 1 + (num_boxes - 1) / kNmsBoxesPerThread;
const size_t max_nms_mask_size = SafeInt<size_t>(num_boxes) * bit_mask_len;
ORT_RETURN_IF_NOT(max_nms_mask_size <= static_cast<size_t>(std::numeric_limits<int>::max()),
"CUDA NonMaxSuppression mask size exceeds the int index range.");
const int max_nms_mask_size_int = static_cast<int>(max_nms_mask_size);

IAllocatorUniquePtr<void> d_nms_mask_ptr{allocator(max_nms_mask_size * sizeof(int))};
IAllocatorUniquePtr<void> d_nms_mask_ptr{allocator(SafeInt<size_t>(max_nms_mask_size) * sizeof(int))};
auto* d_nms_mask = static_cast<int*>(d_nms_mask_ptr.get());

int blocksPerGrid = (int)(ceil(static_cast<float>(max_nms_mask_size) / GridDim::maxThreadsPerBlock));
SetZero<int><<<blocksPerGrid, GridDim::maxThreadsPerBlock, 0, stream>>>(max_nms_mask_size, d_nms_mask);
int blocksPerGrid = static_cast<int>((max_nms_mask_size + GridDim::maxThreadsPerBlock - 1) /
GridDim::maxThreadsPerBlock);
SetZero<int><<<blocksPerGrid, GridDim::maxThreadsPerBlock, 0, stream>>>(max_nms_mask_size_int, d_nms_mask);

int* d_delete_mask = d_nms_mask;
int* h_selected_count = h_nkeep;
Expand Down
27 changes: 27 additions & 0 deletions onnxruntime/test/providers/cpu/object_detection/non_max_suppression_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,10 @@
#include "gtest/gtest.h"
#include "test/providers/provider_test_utils.h"

#ifdef USE_CUDA
#include "test/util/include/default_providers.h"
#endif

namespace onnxruntime {
namespace test {

Expand Down Expand Up @@ -404,5 +408,28 @@ TEST(NonMaxSuppressionOpTest, WithIOUThresholdOpset11) {
test.Run();
}

#ifdef USE_CUDA
TEST(NonMaxSuppressionOpTest, CudaRejectsMaskSizeOutsideIntRange) {
auto cuda_provider = DefaultCudaExecutionProvider();
if (cuda_provider == nullptr) {
GTEST_SKIP() << "CUDA execution provider is not available.";
}

constexpr int64_t num_boxes = 262144;
OpTester test("NonMaxSuppression", 11, kOnnxDomain);
test.AddInput<float>("boxes", {1, num_boxes, 4}, std::vector<float>(num_boxes * 4));
test.AddInput<float>("scores", {1, 1, num_boxes}, std::vector<float>(num_boxes));
test.AddInput<int64_t>("max_output_boxes_per_class", {}, {1});
test.AddInput<float>("iou_threshold", {}, {0.5f});
test.AddOptionalInputEdge<float>();
test.AddOutput<int64_t>("selected_indices", {0, 3}, {});

std::vector<std::unique_ptr<IExecutionProvider>> execution_providers;
execution_providers.push_back(std::move(cuda_provider));
test.Run(OpTester::ExpectResult::kExpectFailure, "mask size exceeds the int index range",
{}, nullptr, &execution_providers);
}
#endif

} // namespace test
} // namespace onnxruntime
Loading