rok commented on code in PR #14191:
URL: https://github.com/apache/arrow/pull/14191#discussion_r1025273509
##########
cpp/src/parquet/encoding.cc:
##########
@@ -2060,6 +2062,220 @@ class DictByteArrayDecoderImpl : public
DictDecoderImpl<ByteArrayType>,
}
};
+// ----------------------------------------------------------------------
+// DeltaBitPackEncoder
+
+constexpr uint32_t kValuesPerBlock = 128;
+constexpr uint32_t kMiniBlocksPerBlock = 4;
+
+template <typename DType>
+class DeltaBitPackEncoder : public EncoderImpl, virtual public
TypedEncoder<DType> {
+ public:
+ using T = typename DType::c_type;
+ using TypedEncoder<DType>::Put;
+
+ explicit DeltaBitPackEncoder(const ColumnDescriptor* descr, MemoryPool* pool,
+ const uint32_t values_per_block =
kValuesPerBlock,
+ const uint32_t mini_blocks_per_block =
kMiniBlocksPerBlock)
+ : EncoderImpl(descr, Encoding::DELTA_BINARY_PACKED, pool),
+ values_per_block_(values_per_block),
+ mini_blocks_per_block_(mini_blocks_per_block),
+ values_per_mini_block_(values_per_block / mini_blocks_per_block),
+ deltas_(values_per_block, ::arrow::stl::allocator<T>(pool)),
+ bits_buffer_(AllocateBuffer(pool, (values_per_block + 3) * sizeof(T))),
+ sink_(pool),
+ bit_writer_(bits_buffer_->mutable_data(),
+ static_cast<int>(bits_buffer_->size())) {
+ if (values_per_mini_block_ % 32 != 0) {
+ throw ParquetException(
+ "the number of values in a miniblock must be multiple of 32, but
it's " +
+ std::to_string(values_per_mini_block_));
+ }
+ }
+
+ std::shared_ptr<Buffer> FlushValues() override;
+
+ int64_t EstimatedDataEncodedSize() override { return sink_.length(); }
+
+ void Put(const ::arrow::Array& values) override;
+
+ void Put(const T* buffer, int num_values) override;
+
+ void PutSpaced(const T* src, int num_values, const uint8_t* valid_bits,
+ int64_t valid_bits_offset) override;
+
+ void FlushBlock();
+
+ private:
+ const uint32_t values_per_block_;
+ const uint32_t mini_blocks_per_block_;
+ const uint32_t values_per_mini_block_;
+ uint32_t values_current_block_{0};
Review Comment:
`{0}` here initialises the variable are at the correct value (starting with
zero values in the buffer) and more also initialises them to prevent undefined
behaviour. I think I've introduced a bug like that in one of the timestamp
kernels.
As for the `length` vs `values` - I'm taking the names from the decoder
implementation. It seems the whole `encoder.cc` uses `length` more in way of
`type_length` than `vector_length` and talks about array length more in the way
of value count. I don't have a strong opinion either way but I'd like to be
consistent here. So we can keep it as is or switch at least the decoder as well.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscr...@arrow.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
