Skip to content

[SPARK-21967][CORE] org.apache.spark.unsafe.types.UTF8String#compareTo Should Compare 8 Bytes at a Time for Better Performance #19180

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

Closed
wants to merge 1 commit into from
Closed

[SPARK-21967][CORE] org.apache.spark.unsafe.types.UTF8String#compareTo Should Compare 8 Bytes at a Time for Better Performance #19180

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
24 changes: 19 additions & 5 deletions common/unsafe/src/main/java/org/apache/spark/unsafe/types/UTF8String.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,7 +64,8 @@ public final class UTF8String implements Comparable<UTF8String>, Externalizable,
5, 5, 5, 5,
6, 6};

private static boolean isLittleEndian = ByteOrder.nativeOrder() == ByteOrder.LITTLE_ENDIAN;
private static final boolean IS_LITTLE_ENDIAN =
ByteOrder.nativeOrder() == ByteOrder.LITTLE_ENDIAN;

private static final UTF8String COMMA_UTF8 = UTF8String.fromString(",");
public static final UTF8String EMPTY_UTF8 = UTF8String.fromString("");
Expand Down Expand Up @@ -220,7 +221,7 @@ public long getPrefix() {
// After getting the data, we use a mask to mask out data that is not part of the string.
long p;
long mask = 0;
if (isLittleEndian) {
if (IS_LITTLE_ENDIAN) {
if (numBytes >= 8) {
p = Platform.getLong(base, offset);
} else if (numBytes > 4) {
Expand Down Expand Up @@ -1097,10 +1098,23 @@ public UTF8String copy() {
@Override
public int compareTo(@Nonnull final UTF8String other) {
int len = Math.min(numBytes, other.numBytes);
// TODO: compare 8 bytes as unsigned long
for (int i = 0; i < len; i ++) {
int wordMax = (len / 8) * 8;
long roffset = other.offset;
Object rbase = other.base;
for (int i = 0; i < wordMax; i += 8) {
long left = getLong(base, offset + i);
long right = getLong(rbase, roffset + i);
if (left != right) {
if (IS_LITTLE_ENDIAN) {
return Long.compareUnsigned(Long.reverseBytes(left), Long.reverseBytes(right));
} else {
return Long.compareUnsigned(left, right);
}
}
}
for (int i = wordMax; i < len; i++) {
// In UTF-8, the byte should be unsigned, so we should compare them as unsigned int.
int res = (getByte(i) & 0xFF) - (other.getByte(i) & 0xFF);
int res = (getByte(i) & 0xFF) - (Platform.getByte(rbase, roffset + i) & 0xFF);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this really make a difference? It's the same call but the other object has also already calculated the base and offset. Hey maybe so, just wondering if it's a benchmark issue or a real improvement.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@srowen yea this was visible in the benchmark. Likely because base and offset aren't final, so you actually do the field lookup on every invocation I think.

if (res != 0) {
return res;
}
Expand Down