L. C. Hsieh created SPARK-57024:
-----------------------------------
Summary: Use bulk fill APIs to materialize RLE runs in Parquet
vectorized reader
Key: SPARK-57024
URL: https://issues.apache.org/jira/browse/SPARK-57024
Project: Spark
Issue Type: Improvement
Components: SQL
Affects Versions: 4.3.0
Reporter: L. C. Hsieh
`VectorizedRleValuesReader` materializes RLE runs of nulls and definition
levels with degenerate per-element loops:
// VectorizedRleValuesReader.java
for (int k = 0; k < runLen; k++) {
nulls.putNull(valueOff + k);
}
for (int k = 0; k < runLen; k++) {
defLevels.putInt(levelIdx + k, runValue);
}
`WritableColumnVector` already exposes the bulk equivalents
`putNulls(rowId, count)` and `putInts(rowId, count, value)`. Switching the
callers to these bulk APIs lets the implementations use intrinsics — but
the bulk implementations themselves were also degenerate byte-by-byte
loops, so the win only materializes once both sides are fixed:
- `OnHeapColumnVector.putNulls` -> `Arrays.fill(byte[], ..., (byte) 1)`
- `OnHeapColumnVector.putInts(rowId, count, value)`
-> `Arrays.fill(int[], ..., value)`
- `OffHeapColumnVector.putNulls` -> `Platform.setMemory(addr, (byte) 1,
count)`
`Arrays.fill` is a JIT intrinsic backed by `_jbyte_fill` / `_jint_fill`
stubs, and `Unsafe.setMemory` lowers to a native memset; both are
significantly faster than the unrolled-by-JIT byte/int loops they replace
for run lengths above a handful of elements.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
