[BUG] Invalid memory read in cudf::io::inflate_kernel #13567
Description
The nightly memcheck tests found an invalid global device memory read in cudf::io::inflate_kernel
========= Invalid __global__ read of size 4 bytes
========= at 0x11fa0 in void cudf::io::inflate_kernel<(int)128>(cudf::device_span<const cudf::device_span<const unsigned char, (unsigned long)18446744073709551615>, (unsigned long)18446744073709551615>, cudf::device_span<const cudf::device_span<unsigned char, (unsigned long)18446744073709551615>, (unsigned long)18446744073709551615>, cudf::device_span<cudf::io::compression_result, (unsigned long)18446744073709551615>, cudf::io::gzip_header_included)
========= by thread (0,0,0) in block (2605,0,0)
========= Address 0x7fba9f00654c is out of bounds
========= and is inside the nearest allocation at 0x7fba9a000000 of size 83912015 bytes
========= Saved host backtrace up to driver entry point at kernel launch time
========= Host Frame: [0x304e32]
========= in /usr/lib/x86_64-linux-gnu/libcuda.so.1
========= Host Frame: [0x1488c]
========= in /opt/conda/envs/test/bin/gtests/libcudf/../../../lib/libcudart.so.11.0
========= Host Frame:cudaLaunchKernel [0x6c318]
========= in /opt/conda/envs/test/bin/gtests/libcudf/../../../lib/libcudart.so.11.0
========= Host Frame:cudf::io::gpuinflate(cudf::device_span<cudf::device_span<unsigned char const, 18446744073709551615ul> const, 18446744073709551615ul>, cudf::device_span<cudf::device_span<unsigned char, 18446744073709551615ul> const, 18446744073709551615ul>, cudf::device_span<cudf::io::compression_result, 18446744073709551615ul>, cudf::io::gzip_header_included, rmm::cuda_stream_view) [0x1429446]
========= in /opt/conda/envs/test/bin/gtests/libcudf/../../../lib/libcudf.so
========= Host Frame:cudf::io::text::(anonymous namespace)::bgzip_data_chunk_reader::decompression_blocks::decompress(rmm::cuda_stream_view) [0x15b3ad1]
========= in /opt/conda/envs/test/bin/gtests/libcudf/../../../lib/libcudf.so
========= Host Frame:cudf::io::text::(anonymous namespace)::bgzip_data_chunk_reader::get_next_chunk(unsigned long, rmm::cuda_stream_view) [0x15b44d8]
========= in /opt/conda/envs/test/bin/gtests/libcudf/../../../lib/libcudf.so
...
The error can be recreated using the DATA_CHUNK_SOURCE_TEST as follows:
compute-sanitizer --tool memcheck gtests/DATA_CHUNK_SOURCE_TEST --gtest_filter=DataChunkSourceTest.BgzipSource --rmm_mode=cuda