[BUG] Memcheck error in ParquetDictionaryTest/ParquetSizedTest.DictionaryTest/22

[davidwendt]

Found an out-of-bounds memory write in the PARQUET_TEST at ParquetDictionaryTest/ParquetSizedTest.DictionaryTest/22
The error does not occur for any other sizes 1,24 -- only for 22.

Partial output from compute-sanitizer:

========= COMPUTE-SANITIZER
Note: Google Test filter = ParquetDictionaryTest/ParquetSizedTest.DictionaryTest/22
[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from ParquetDictionaryTest/ParquetSizedTest
[ RUN      ] ParquetDictionaryTest/ParquetSizedTest.DictionaryTest/22
========= Invalid __global__ write of size 1 bytes
=========     at 0xdfb0 in void cudf::io::parquet::gpu::gpuEncodePages<(int)128>(cudf::device_span<cudf::io::parquet::gpu::EncPage, (unsigned long)18446744073709551615>, cudf::device_span<cudf::device_span<const unsigned char, (unsigned long)18446744073709551615>, (unsigned long)18446744073709551615>, cudf::device_span<cudf::device_span<unsigned char, (unsigned long)18446744073709551615>, (unsigned long)18446744073709551615>, cudf::device_span<cudf::io::compression_result, (unsigned long)18446744073709551615>)
=========     by thread (92,0,0) in block (158,0,0)
=========     Address 0x7fd632539b50 is out of bounds
=========     and is 1 bytes after the nearest allocation at 0x7fd62c000000 of size 106142544 bytes
=========     Saved host backtrace up to driver entry point at kernel launch time
=========     Host Frame: [0x305c18]
=========                in /usr/lib/x86_64-linux-gnu/libcuda.so.1
=========     Host Frame: [0x1402c]
=========                in /conda/envs/rapids/lib/libcudart.so.11.0
=========     Host Frame:cudaLaunchKernel [0x68808]
=========                in /conda/envs/rapids/lib/libcudart.so.11.0
=========     Host Frame:cudf::io::parquet::gpu::EncodePages(cudf::device_span<cudf::io::parquet::gpu::EncPage, 18446744073709551615ul>, cudf::device_span<cudf::device_span<unsigned char const, 18446744073709551615ul>, 18446744073709551615ul>, cudf::device_span<cudf::device_span<unsigned char, 18446744073709551615ul>, 18446744073709551615ul>, cudf::device_span<cudf::io::compression_result, 18446744073709551615ul>, rmm::cuda_stream_view) [0x17117d3]
=========                in /conda/envs/rapids/lib/libcudf.so
=========     Host Frame:cudf::io::detail::parquet::writer::impl::encode_pages(cudf::detail::hostdevice_2dvector<cudf::io::parquet::gpu::EncColumnChunk>&, cudf::device_span<cudf::io::parquet::gpu::EncPage, 18446744073709551615ul>, unsigned long, unsigned int, unsigned int, unsigned int, unsigned int, cudf::io::statistics_chunk const*, cudf::io::statistics_chunk const*, cudf::io::statistics_chunk const*) [0x172e4e5]
=========                in /conda/envs/rapids/lib/libcudf.so
=========     Host Frame:cudf::io::detail::parquet::writer::impl::write(cudf::table_view const&, std::vector<cudf::io::partition_info, std::allocator<cudf::io::partition_info> > const&) [0x173275b]
=========                in /conda/envs/rapids/lib/libcudf.so
=========     Host Frame:cudf::io::detail::parquet::writer::write(cudf::table_view const&, std::vector<cudf::io::partition_info, std::allocator<cudf::io::partition_info> > const&) [0x1734388]
=========                in /conda/envs/rapids/lib/libcudf.so
=========     Host Frame:cudf::io::write_parquet(cudf::io::parquet_writer_options const&, rmm::mr::device_memory_resource*) [0x164bc5c]
=========                in /conda/envs/rapids/lib/libcudf.so
=========     Host Frame:ParquetSizedTest_DictionaryTest_Test::TestBody() [0xad84f]
=========                in /cudf/cpp/build/gtests/PARQUET_TEST

Use the following command to reproduce the bug:

compute-sanitizer --tool memcheck gtests/PARQUET_TEST --gtest_filter=ParquetDictionaryTest/ParquetSizedTest.DictionaryTest/22 --rmm_mode=cuda

This is occurring on the current 22.12. I've not verified it occurs on a previous branch.

[davidwendt]

Nightlys are failing for this.
https://gpuci.gpuopenanalytics.com/job/rapidsai/job/gpuci/job/cudf/job/branches/job/cudf-gpu-build-branch-22.12/1/CUDA=11.5/testReport/junit/cudamemcheck/PARQUET_TEST/ParquetDictionaryTest_ParquetSizedTest_DictionaryTest_22/

[GregoryKimball]

@vuule would you please take a look at this memcheck failure?