[auto-merge] bot-auto-merge-release/26.04 to main [skip ci] [bot]

[nvauto]

auto-merge triggered by github actions on bot-auto-merge-release/26.04 to create a PR keeping main up-to-date. If this PR is unable to be merged due to conflicts, it will remain open until manually fix.

[nvauto]

FAILURE - Unable to auto-merge. Manual operation is required.

{'message': 'Pull Request is not mergeable', 'documentation_url': 'https://docs.github.com/rest/pulls/pulls#merge-a-pull-request', 'status': '405'}

Please use the following steps to fix the merge conflicts manually:

# Assume upstream is NVIDIA/spark-rapids-jni remote
git fetch upstream bot-auto-merge-release/26.04 main
git checkout -b fix-auto-merge-conflict-4411 upstream/main
git merge upstream/bot-auto-merge-release/26.04
# Fix any merge conflicts caused by this merge
git commit -am "Merge bot-auto-merge-release/26.04 into main"
git push <personal fork> fix-auto-merge-conflict-4411
# Open a PR targets NVIDIA/spark-rapids-jni main

IMPORTANT: Before merging this PR, be sure to change the merging strategy to Create a merge commit (repo admin only).

Once this PR is merged, the auto-merge PR should automatically be closed since it contains the same commit hashes

[greptile-apps]

Greptile Summary

This PR merges the release/26.04 branch into main, introducing Spark Bloom Filter V2 format support across the entire bloom filter stack (C++ CUDA kernels, JNI bridge, and Java API).

Key changes:

• New V2 format: 16-byte header (version, num_hashes, seed, num_longs) alongside the existing 12-byte V1 header. V2 uses 64-bit hash indexing (int64_t bloom_filter_bits) and a configurable seed for MurmurHash3_32, matching Spark's BloomFilterImplV2.
• Kernel versioning: gpu_bloom_filter_put and bloom_probe_functor are now templated on int Version, using if constexpr to select between V1 and V2 algorithms.
• Atomics upgrade: atomicOr replaced with cuda::atomic_ref::fetch_or(mask, relaxed) for modern CUDA style.
• Const-correctness: unpack_bloom_filter now takes and returns const spans.
• JNI/Java API: New four-argument BloomFilter.create(version, numHashes, bloomFilterBits, seed) overload; old two-argument overload deprecated. VERSION_1, VERSION_2, DEFAULT_SEED constants exposed.
• Test coverage: All V1 tests kept and renamed; equivalent V2 tests added; Java tests parameterised over both versions.

Confidence Score: 5/5

Safe to merge; only P2 style/edge-case findings remain, no correctness or data-integrity issues.

The V2 algorithm correctly matches the referenced Spark Java source (BloomFilterImplV2). Header serialisation, endian-swapping, const-correctness, atomic writes, and merge validation are all handled correctly. Both V1 and V2 paths have symmetric test coverage. The two findings are a minor signed/unsigned comparison inconsistency in bloom_filter_probe and an overly generous JNI upper bound that defers rejection to a less descriptive C++ error — neither affects correctness.

src/main/cpp/src/BloomFilterJni.cpp (JNI size bound) and src/main/cpp/src/bloom_filter.cu (bloom_filter_probe signed/size_t comparison)

Important Files Changed

Filename	Overview
src/main/cpp/src/bloom_filter.cu	Core implementation extended with V2 format support (64-bit hash indexing, configurable seed, 16-byte header). Template parameters and const-correctness improved; minor signed/unsigned comparison inconsistency in bloom_filter_probe.
src/main/cpp/src/bloom_filter.hpp	New V1/V2 header structs with static_assert size checks, version constants, and helper function bloom_filter_header_size_for_version added cleanly.
src/main/cpp/src/BloomFilterJni.cpp	JNI entry point updated with version and seed parameters; adds upper-bound check for bloomFilterBits, but the advertised maximum can exceed what the C++ layer can service, yielding a cryptic downstream error.
src/main/cpp/benchmarks/bloom_filter.cu	Benchmark split into V1 and V2 variants; unused lambda parameter fixed.
src/main/cpp/tests/bloom_filter.cu	Comprehensive V2 test suite added mirroring all V1 tests; seed variation test included.
src/main/java/com/nvidia/spark/rapids/jni/BloomFilter.java	New create(version, numHashes, bits, seed) API added; legacy two-arg overload deprecated cleanly; VERSION_1/VERSION_2/DEFAULT_SEED constants exposed.
src/test/java/com/nvidia/spark/rapids/jni/BloomFilterTest.java	Tests parameterised over VERSION_1 and VERSION_2; V2-specific seed test and mixed-version merge failure test added.

Sequence Diagram

sequenceDiagram
    participant Java as BloomFilter.java
    participant JNI as BloomFilterJni.cpp
    participant CPP as bloom_filter.cu
    participant GPU as CUDA kernel

    Java->>JNI: creategpu(version, numHashes, bloomFilterBits, seed)
    JNI->>JNI: validate bloomFilterBits range
    JNI->>CPP: bloom_filter_create(version, numHashes, longs, seed)
    CPP->>CPP: get_bloom_filter_stride(version, longs)
    CPP->>CPP: pack_bloom_filter_header (V1=12B / V2=16B, big-endian)
    CPP-->>JNI: list_scalar (header + zeroed bits)
    JNI-->>Java: Scalar handle

    Java->>JNI: put(bloomFilter, cv)
    JNI->>CPP: bloom_filter_put(scalar, column)
    CPP->>CPP: unpack_bloom_filter to header, buffer, bits, seed
    CPP->>GPU: gpu_bloom_filter_put<Version, nullable>(buffer, bits, input, num_hashes, seed)
    Note over GPU: V1: loop 1..N, 32-bit combined hash
    Note over GPU: V2: loop 0..N-1, 64-bit combined hash seeded h1*INT32_MAX

    Java->>JNI: probe(bloomFilter, cv)
    JNI->>CPP: bloom_filter_probe(scalar, column)
    CPP->>CPP: unpack_bloom_filter to header, buffer, bits, seed
    CPP->>GPU: thrust::transform(bloom_probe_functor<Version>{})
    GPU-->>CPP: bool column
    CPP-->>JNI: column pointer
    JNI-->>Java: ColumnVector

Loading

_{Reviews (6): Last reviewed commit: "Merge branch 'main' into bot-auto-merge-..." | Re-trigger Greptile}

[nvauto]

FAILURE - Unable to auto-merge. Manual operation is required.

{'message': 'Pull Request is not mergeable', 'documentation_url': 'https://docs.github.com/rest/pulls/pulls#merge-a-pull-request', 'status': '405'}

Please use the following steps to fix the merge conflicts manually:

# Assume upstream is NVIDIA/spark-rapids-jni remote
git fetch upstream bot-auto-merge-release/26.04 main
git checkout -b fix-auto-merge-conflict-4411 upstream/main
git merge upstream/bot-auto-merge-release/26.04
# Fix any merge conflicts caused by this merge
git commit -am "Merge bot-auto-merge-release/26.04 into main"
git push <personal fork> fix-auto-merge-conflict-4411
# Open a PR targets NVIDIA/spark-rapids-jni main

IMPORTANT: Before merging this PR, be sure to change the merging strategy to Create a merge commit (repo admin only).

Once this PR is merged, the auto-merge PR should automatically be closed since it contains the same commit hashes

[nvauto]

FAILURE - Unable to auto-merge. Manual operation is required.

{'message': 'Base branch was modified. Review and try the merge again.', 'documentation_url': 'https://docs.github.com/rest/pulls/pulls#merge-a-pull-request', 'status': '405'}

Please use the following steps to fix the merge conflicts manually:

# Assume upstream is NVIDIA/spark-rapids-jni remote
git fetch upstream bot-auto-merge-release/26.04 main
git checkout -b fix-auto-merge-conflict-4411 upstream/main
git merge upstream/bot-auto-merge-release/26.04
# Fix any merge conflicts caused by this merge
git commit -am "Merge bot-auto-merge-release/26.04 into main"
git push <personal fork> fix-auto-merge-conflict-4411
# Open a PR targets NVIDIA/spark-rapids-jni main

IMPORTANT: Before merging this PR, be sure to change the merging strategy to Create a merge commit (repo admin only).

Once this PR is merged, the auto-merge PR should automatically be closed since it contains the same commit hashes

[pxLi]

cc @mythrocks for the automerge conflict resolve, thanks

<<<<<<< bot-auto-merge-release/26.04
inline int32_t byte_swap_int32(int32_t val)
=======
__device__ inline cuda::std::pair<cudf::size_type, cudf::bitmask_type> gpu_get_hash_mask(
  bloom_hash_type h, cudf::size_type bloom_filter_bits)
>>>>>>> main

[nvauto]

FAILURE - Unable to auto-merge. Manual operation is required.

{'message': 'Pull Request is not mergeable', 'documentation_url': 'https://docs.github.com/rest/pulls/pulls#merge-a-pull-request', 'status': '405'}

Please use the following steps to fix the merge conflicts manually:

# Assume upstream is NVIDIA/spark-rapids-jni remote
git fetch upstream bot-auto-merge-release/26.04 main
git checkout -b fix-auto-merge-conflict-4411 upstream/main
git merge upstream/bot-auto-merge-release/26.04
# Fix any merge conflicts caused by this merge
git commit -am "Merge bot-auto-merge-release/26.04 into main"
git push <personal fork> fix-auto-merge-conflict-4411
# Open a PR targets NVIDIA/spark-rapids-jni main

IMPORTANT: Before merging this PR, be sure to change the merging strategy to Create a merge commit (repo admin only).

Once this PR is merged, the auto-merge PR should automatically be closed since it contains the same commit hashes

[nvauto]

FAILURE - Unable to auto-merge. Manual operation is required.

{'message': 'Pull Request is not mergeable', 'documentation_url': 'https://docs.github.com/rest/pulls/pulls#merge-a-pull-request', 'status': '405'}

Please use the following steps to fix the merge conflicts manually:

# Assume upstream is NVIDIA/spark-rapids-jni remote
git fetch upstream bot-auto-merge-release/26.04 main
git checkout -b fix-auto-merge-conflict-4411 upstream/main
git merge upstream/bot-auto-merge-release/26.04
# Fix any merge conflicts caused by this merge
git commit -am "Merge bot-auto-merge-release/26.04 into main"
git push <personal fork> fix-auto-merge-conflict-4411
# Open a PR targets NVIDIA/spark-rapids-jni main

IMPORTANT: Before merging this PR, be sure to change the merging strategy to Create a merge commit (repo admin only).

Once this PR is merged, the auto-merge PR should automatically be closed since it contains the same commit hashes

[mythrocks]

A strange place for the auto-merge to fail. I've committed the fix.

[mythrocks]

Build