archive/jdk
1
0
Fork 0
mirror of https://github.com/openjdk/jdk.git synced 2025-08-27 06:45:07 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

8338967: Improve performance for MemorySegment::fill

Browse source 
Reviewed-by: mcimadamore, psandoz
This commit is contained in:
Per Minborg 2024-09-03 10:25:27 +00:00
parent 633fad8e53
commit 7a418fc074
3 changed files with 277 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

49
src/java.base/share/classes/jdk/internal/foreign/AbstractMemorySegmentImpl.java
View file

@ -51,6 +51,7 @@ import jdk.internal.access.foreign.UnmapperProxy;
import jdk.internal.misc.ScopedMemoryAccess;
import jdk.internal.reflect.CallerSensitive;
import jdk.internal.reflect.Reflection;
import jdk.internal.util.Architecture;
import jdk.internal.util.ArraysSupport;
import jdk.internal.util.Preconditions;
import jdk.internal.vm.annotation.ForceInline;
@ -188,10 +189,52 @@ public abstract sealed class AbstractMemorySegmentImpl
return StreamSupport.stream(spliterator(elementLayout), false);
}
// FILL_NATIVE_THRESHOLD must be a power of two and should be greater than 2^3
// Update the value for Aarch64 once 8338975 is fixed.
private static final long FILL_NATIVE_THRESHOLD = 1L << (Architecture.isAARCH64() ? 10 : 5);
@Override
public final MemorySegment fill(byte value){
checkAccess(0, length, false);
SCOPED_MEMORY_ACCESS.setMemory(sessionImpl(), unsafeGetBase(), unsafeGetOffset(), length, value);
@ForceInline
public final MemorySegment fill(byte value) {
checkReadOnly(false);
if (length == 0) {
// Implicit state check
checkValidState();
} else if (length < FILL_NATIVE_THRESHOLD) {
// 0 <= length < FILL_NATIVE_LIMIT : 0...0X...XXXX
// Handle smaller segments directly without transitioning to native code
final long u = Byte.toUnsignedLong(value);
final long longValue = u << 56 | u << 48 | u << 40 | u << 32 | u << 24 | u << 16 | u << 8 | u;
int offset = 0;
// 0...0X...X000
final int limit = (int) (length & (FILL_NATIVE_THRESHOLD - 8));
for (; offset < limit; offset += 8) {
SCOPED_MEMORY_ACCESS.putLong(sessionImpl(), unsafeGetBase(), unsafeGetOffset() + offset, longValue);
}
int remaining = (int) length - limit;
// 0...0X00
if (remaining >= 4) {
SCOPED_MEMORY_ACCESS.putInt(sessionImpl(), unsafeGetBase(), unsafeGetOffset() + offset, (int) longValue);
offset += 4;
remaining -= 4;
}
// 0...00X0
if (remaining >= 2) {
SCOPED_MEMORY_ACCESS.putShort(sessionImpl(), unsafeGetBase(), unsafeGetOffset() + offset, (short) longValue);
offset += 2;
remaining -= 2;
}
// 0...000X
if (remaining == 1) {
SCOPED_MEMORY_ACCESS.putByte(sessionImpl(), unsafeGetBase(), unsafeGetOffset() + offset, value);
}
// We have now fully handled 0...0X...XXXX
} else {
// Handle larger segments via native calls
SCOPED_MEMORY_ACCESS.setMemory(sessionImpl(), unsafeGetBase(), unsafeGetOffset(), length, value);
}
return this;
}