Skip to content

8345465: Fix performance regression on x64 after JDK-8345120 #22539

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 9 commits into from
Closed

8345465: Fix performance regression on x64 after JDK-8345120 #22539

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
2aaec5b
Use aliases for legacy code.
minborg Dec 3, 2024
1de3f13
Merge branch 'master' into strlen-bench-update
minborg Dec 3, 2024
f509b6c
Go back to use MemorySegment types
minborg Dec 3, 2024
779b7f9
Fix regression on x64
minborg Dec 4, 2024
be612ab
Merge branch 'master' into strlen-bench-update
minborg Dec 4, 2024
4d41488
Update comment
minborg Dec 4, 2024
96c3a55
Improve short string cases
minborg Dec 5, 2024
98285a6
Merge branch 'master' into strlen-bench-update
minborg Dec 10, 2024
e4b0b47
Simplify code
minborg Dec 10, 2024
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
29 changes: 17 additions & 12 deletions src/java.base/share/classes/jdk/internal/foreign/StringSupport.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,8 +130,8 @@ public static int strlenByte(final AbstractMemorySegmentImpl segment,
final long toOffset) {
final long length = toOffset - fromOffset;
segment.checkBounds(fromOffset, length);
if (length == 0) {
// The state has to be checked explicitly for zero-length segments
if (length < Byte.BYTES) {
// There can be no null terminator present
segment.scope.checkValidState();
throw nullNotFound(segment, fromOffset, toOffset);
}
Expand Down Expand Up @@ -164,7 +164,8 @@ public static int strlenShort(final AbstractMemorySegmentImpl segment,
final long toOffset) {
final long length = toOffset - fromOffset;
segment.checkBounds(fromOffset, length);
if (length == 0) {
if (length < Short.BYTES) {
// There can be no null terminator present
segment.scope.checkValidState();
throw nullNotFound(segment, fromOffset, toOffset);
}
Expand Down Expand Up @@ -199,19 +200,23 @@ public static int strlenInt(final AbstractMemorySegmentImpl segment,
final long toOffset) {
final long length = toOffset - fromOffset;
segment.checkBounds(fromOffset, length);
if (length == 0) {
if (length < Integer.BYTES) {
// There can be no null terminator present
segment.scope.checkValidState();
throw nullNotFound(segment, fromOffset, toOffset);
}
final long longBytes = length & LONG_MASK;
final long longLimit = fromOffset + longBytes;
long offset = fromOffset;
for (; offset < longLimit; offset += Long.BYTES) {
long val = SCOPED_MEMORY_ACCESS.getLongUnaligned(segment.sessionImpl(), segment.unsafeGetBase(), segment.unsafeGetOffset() + offset, !Architecture.isLittleEndian());
if (mightContainZeroInt(val)) {
for (int j = 0; j < Long.BYTES; j += Integer.BYTES) {
if (SCOPED_MEMORY_ACCESS.getIntUnaligned(segment.sessionImpl(), segment.unsafeGetBase(), segment.unsafeGetOffset() + offset + j, !Architecture.isLittleEndian()) == 0) {
return requireWithinStringSize(offset + j - fromOffset, segment, fromOffset, toOffset);
// For quad byte strings, it does not pay off to use long scanning on x64
if (!Architecture.isX64()) {
final long longBytes = length & LONG_MASK;
final long longLimit = fromOffset + longBytes;
for (; offset < longLimit; offset += Long.BYTES) {
long val = SCOPED_MEMORY_ACCESS.getLongUnaligned(segment.sessionImpl(), segment.unsafeGetBase(), segment.unsafeGetOffset() + offset, !Architecture.isLittleEndian());
if (mightContainZeroInt(val)) {
for (int j = 0; j < Long.BYTES; j += Integer.BYTES) {
if (SCOPED_MEMORY_ACCESS.getIntUnaligned(segment.sessionImpl(), segment.unsafeGetBase(), segment.unsafeGetOffset() + offset + j, !Architecture.isLittleEndian()) == 0) {
return requireWithinStringSize(offset + j - fromOffset, segment, fromOffset, toOffset);
}
}
}
}
Expand Down
24 changes: 12 additions & 12 deletions test/micro/org/openjdk/bench/java/lang/foreign/InternalStrLen.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,20 +55,20 @@
"--enable-native-access=ALL-UNNAMED"})
public class InternalStrLen {

private AbstractMemorySegmentImpl singleByteSegment;
private AbstractMemorySegmentImpl singleByteSegmentMisaligned;
private AbstractMemorySegmentImpl doubleByteSegment;
private AbstractMemorySegmentImpl quadByteSegment;
private MemorySegment singleByteSegment;
private MemorySegment singleByteSegmentMisaligned;
private MemorySegment doubleByteSegment;
private MemorySegment quadByteSegment;

@Param({"1", "4", "16", "251", "1024"})
int size;

@Setup
public void setup() {
var arena = Arena.ofAuto();
singleByteSegment = (AbstractMemorySegmentImpl) arena.allocate((size + 1L) * Byte.BYTES);
doubleByteSegment = (AbstractMemorySegmentImpl) arena.allocate((size + 1L) * Short.BYTES);
quadByteSegment = (AbstractMemorySegmentImpl) arena.allocate((size + 1L) * Integer.BYTES);
singleByteSegment = arena.allocate((size + 1L) * Byte.BYTES);
doubleByteSegment = arena.allocate((size + 1L) * Short.BYTES);
quadByteSegment = arena.allocate((size + 1L) * Integer.BYTES);
Stream.of(singleByteSegment, doubleByteSegment, quadByteSegment)
.forEach(s -> IntStream.range(0, (int) s.byteSize() - 1)
.forEach(i -> s.set(
Expand All @@ -79,7 +79,7 @@ public void setup() {
singleByteSegment.set(ValueLayout.JAVA_BYTE, singleByteSegment.byteSize() - Byte.BYTES, (byte) 0);
doubleByteSegment.set(ValueLayout.JAVA_SHORT, doubleByteSegment.byteSize() - Short.BYTES, (short) 0);
quadByteSegment.set(ValueLayout.JAVA_INT, quadByteSegment.byteSize() - Integer.BYTES, 0);
singleByteSegmentMisaligned = (AbstractMemorySegmentImpl) arena.allocate(singleByteSegment.byteSize() + 1).
singleByteSegmentMisaligned = arena.allocate(singleByteSegment.byteSize() + 1).
asSlice(1);
MemorySegment.copy(singleByteSegment, 0, singleByteSegmentMisaligned, 0, singleByteSegment.byteSize());
}
Expand All @@ -106,22 +106,22 @@ public int elementQuad() {

@Benchmark
public int chunkedSingle() {
return StringSupport.strlenByte(singleByteSegment, 0, singleByteSegment.byteSize());
return StringSupport.strlenByte((AbstractMemorySegmentImpl) singleByteSegment, 0, singleByteSegment.byteSize());
}

@Benchmark
public int chunkedSingleMisaligned() {
return StringSupport.strlenByte(singleByteSegmentMisaligned, 0, singleByteSegment.byteSize());
return StringSupport.strlenByte((AbstractMemorySegmentImpl) singleByteSegmentMisaligned, 0, singleByteSegment.byteSize());
}

@Benchmark
public int chunkedDouble() {
return StringSupport.strlenShort(doubleByteSegment, 0, doubleByteSegment.byteSize());
return StringSupport.strlenShort((AbstractMemorySegmentImpl) doubleByteSegment, 0, doubleByteSegment.byteSize());
}

@Benchmark
public int changedElementQuad() {
return StringSupport.strlenInt(quadByteSegment, 0, quadByteSegment.byteSize());
return StringSupport.strlenInt((AbstractMemorySegmentImpl) quadByteSegment, 0, quadByteSegment.byteSize());
}

// These are the legacy methods
Expand Down