Skip to content

perf(parquet): hybrid 2-byte unroll + tail loop for readUleb128Int #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
mashraf-222 wants to merge 1 commit into
base: master
Choose a base branch
from
Open

perf(parquet): hybrid 2-byte unroll + tail loop for readUleb128Int #2

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 23 additions & 23 deletions lib/trino-parquet/src/main/java/io/trino/parquet/ParquetReaderUtils.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,35 +52,35 @@ public static int readUleb128Int(SimpleSliceInputStream input)
{
byte[] inputBytes = input.getByteArray();
int offset = input.getByteArrayOffset();
// Manual loop unrolling shows improvements in BenchmarkReadUleb128Int
int inputByte = inputBytes[offset];
int value = inputByte & 0x7F;
if ((inputByte & 0x80) == 0) {
// Fast paths for 1-byte and 2-byte values (most common for small dictionary indices
// and page/group counts). For 3+ byte values, fall through to a tight loop. This beats
// the fully unrolled 5-branch version on long input sequences (where the unrolled
// version's branch chain depth dominates), while preserving straight-line code for
// the common short values (where the unrolled version wins).
byte b0 = inputBytes[offset];
if ((b0 & 0x80) == 0) {
input.skip(1);
return value;
return b0;
}
inputByte = inputBytes[offset + 1];
value |= (inputByte & 0x7F) << 7;
if ((inputByte & 0x80) == 0) {
byte b1 = inputBytes[offset + 1];
int value = (b0 & 0x7F) | ((b1 & 0x7F) << 7);
if ((b1 & 0x80) == 0) {
input.skip(2);
return value;
}
inputByte = inputBytes[offset + 2];
value |= (inputByte & 0x7F) << 14;
if ((inputByte & 0x80) == 0) {
input.skip(3);
return value;
}
inputByte = inputBytes[offset + 3];
value |= (inputByte & 0x7F) << 21;
if ((inputByte & 0x80) == 0) {
input.skip(4);
return value;
int consumed = 2;
int shift = 14;
byte b;
do {
b = inputBytes[offset + consumed];
value |= (b & 0x7F) << shift;
shift += 7;
consumed++;
}
inputByte = inputBytes[offset + 4];
verify((inputByte & 0x80) == 0, "ULEB128 variable-width integer should not be longer than 5 bytes");
input.skip(5);
return value | inputByte << 28;
while ((b & 0x80) != 0 && shift < 35);
verify((b & 0x80) == 0, "ULEB128 variable-width integer should not be longer than 5 bytes");
input.skip(consumed);
return value;
}

public static long readUleb128Long(SimpleSliceInputStream input)
Expand Down