Using unsigned library work in the JDK
Xueming Shen
xueming.shen at oracle.com
Tue Feb 7 01:02:01 UTC 2012
Joe,
I did some non-scientific measure of the performance. test1() is to
measure the change
in a real "listing" operation on a ZipInputStream and test2() is a pure
access on the byte[]
directly. I don't see any significant difference on performance in both
-client and -server
vm. I would assume the change has no performance impact to the real zip
access.
In -client vm
test1: 108 vs 109
test2: 280 : 283
Simple test case attached (the input is a big enough jar file, such as
the rt.jar)
-Sherman
--------------------------------------------
public static void main(String[] args) throws IOException {
byte[] data = java.nio.file.Files.readAllBytes(new
File(args[0]).toPath());
ByteArrayInputStream bais = new ByteArrayInputStream(data);
//warm up
test1(bais, 100);
System.out.printf("t=%d%n", test1(bais, 500));
test2(data, 100);
System.out.printf("t=%d%n", test2(data, 100));
}
static long test1(ByteArrayInputStream bais, int n) throws
IOException {
long t = 0;
for (int i = 0; i < n; i++) {
bais.reset();
ZipInputStream zis = new ZipInputStream(bais);
ZipEntry e;
long t0 = System.currentTimeMillis();
while ((e = zis.getNextEntry()) != null) {
zis.closeEntry();
//System.out.println(new String(e.getName()));
}
long t1 = System.currentTimeMillis();
t += t1 - t0;
}
return t /= n;
}
static long test2(byte data[], int n) throws IOException {
int j = 10000;
long t = 0;
if (j >= data.length -1)
j = data.length -2;
for (int m = 0; m < n; m++) {
long t0 = System.nanoTime();
int sum = 0;
for (int i = 0; i < j; i++) {
sum += get16(data, i);
}
long t1 = System.nanoTime();
t += t1 - t0;
}
return t / n / 100;
}
private static final int get16(byte b[], int off) {
return (b[off] & 0xff) | ((b[off+1] & 0xff) << 8);
//return Byte.toUnsignedInt(b[off]) |
(Byte.toUnsignedInt(b[off+1]) << 8);
}
On 1/26/2012 5:33 PM, Joe Darcy wrote:
> Sherman, thanks for offering to investigate the performance impact, if
> any, of this change.
>
> -Joe
>
> PS All the jar/zip regression tests pass with this change.
>
> On 01/26/2012 09:52 AM, Xueming Shen wrote:
>> Joe,
>>
>> The changes look fine. However I have the same performance concern in
>> cases that the vm fails to inline the toUnsignedInt(), especially
>> for those
>> "simple" one line utility methods, such as the get16(), CH(), SH(), my
>> guess is that it might have a performance impact. It might not be a big
>> deal for operation like creating or extracting a jar/zip file but
>> probably will
>> slow down the loc/cen table reading only operation such as list a
>> jar/zip
>> file. I will try to get some performance data to see if the impact is
>> "significant" enough to be a real concern.
>>
>> -Sherman
>>
>> On 01/25/2012 08:37 PM, Joe Darcy wrote:
>>> Hello,
>>>
>>> As a follow-up to the recent push of unsigned library support in the
>>> JDK [1], I grepped -i for "0xff" in the JDK code base to look for
>>> candidate locations to use the new methods. I choose to first
>>> tackle some jar/zip code.
>>>
>>> Sherman, can you review the changes below?
>>>
>>> diff -r 303b67074666
>>> src/share/classes/java/util/jar/JarOutputStream.java
>>> --- a/src/share/classes/java/util/jar/JarOutputStream.java Tue
>>> Jan 24 15:13:27 2012 -0500
>>> +++ b/src/share/classes/java/util/jar/JarOutputStream.java Wed
>>> Jan 25 20:31:05 2012 -0800
>>> @@ -1,5 +1,5 @@
>>> /*
>>> - * Copyright (c) 1997, 2006, Oracle and/or its affiliates. All
>>> rights reserved.
>>> + * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All
>>> rights reserved.
>>> * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
>>> *
>>> * This code is free software; you can redistribute it and/or
>>> modify it
>>> @@ -135,7 +135,7 @@
>>> * The bytes are assumed to be in Intel (little-endian) byte
>>> order.
>>> */
>>> private static int get16(byte[] b, int off) {
>>> - return (b[off] & 0xff) | ((b[off+1] & 0xff) << 8);
>>> + return Byte.toUnsignedInt(b[off]) | (
>>> Byte.toUnsignedInt(b[off+1]) << 8);
>>> }
>>>
>>> /*
>>> diff -r 303b67074666 src/share/classes/java/util/jar/Manifest.java
>>> --- a/src/share/classes/java/util/jar/Manifest.java Tue Jan 24
>>> 15:13:27 2012 -0500
>>> +++ b/src/share/classes/java/util/jar/Manifest.java Wed Jan 25
>>> 20:31:05 2012 -0800
>>> @@ -1,5 +1,5 @@
>>> /*
>>> - * Copyright (c) 1997, 2006, Oracle and/or its affiliates. All
>>> rights reserved.
>>> + * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All
>>> rights reserved.
>>> * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
>>> *
>>> * This code is free software; you can redistribute it and/or
>>> modify it
>>> @@ -339,7 +339,7 @@
>>> return -1;
>>> }
>>> }
>>> - return buf[pos++] & 0xff;
>>> + return Byte.toUnsignedInt(buf[pos++]);
>>> }
>>>
>>> public int read(byte[] b, int off, int len) throws
>>> IOException {
>>> diff -r 303b67074666
>>> src/share/classes/java/util/zip/InflaterInputStream.java
>>> --- a/src/share/classes/java/util/zip/InflaterInputStream.java
>>> Tue Jan 24 15:13:27 2012 -0500
>>> +++ b/src/share/classes/java/util/zip/InflaterInputStream.java
>>> Wed Jan 25 20:31:05 2012 -0800
>>> @@ -1,5 +1,5 @@
>>> /*
>>> - * Copyright (c) 1996, 2006, Oracle and/or its affiliates. All
>>> rights reserved.
>>> + * Copyright (c) 1996, 2012, Oracle and/or its affiliates. All
>>> rights reserved.
>>> * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
>>> *
>>> * This code is free software; you can redistribute it and/or
>>> modify it
>>> @@ -119,7 +119,7 @@
>>> */
>>> public int read() throws IOException {
>>> ensureOpen();
>>> - return read(singleByteBuf, 0, 1) == -1 ? -1 :
>>> singleByteBuf[0] & 0xff;
>>> + return read(singleByteBuf, 0, 1) == -1 ? -1 :
>>> Byte.toUnsignedInt(singleByteBuf[0]);
>>> }
>>>
>>> /**
>>> diff -r 303b67074666
>>> src/share/classes/java/util/zip/ZipInputStream.java
>>> --- a/src/share/classes/java/util/zip/ZipInputStream.java Tue Jan
>>> 24 15:13:27 2012 -0500
>>> +++ b/src/share/classes/java/util/zip/ZipInputStream.java Wed Jan
>>> 25 20:31:05 2012 -0800
>>> @@ -1,5 +1,5 @@
>>> /*
>>> - * Copyright (c) 1996, 2009, Oracle and/or its affiliates. All
>>> rights reserved.
>>> + * Copyright (c) 1996, 2012, Oracle and/or its affiliates. All
>>> rights reserved.
>>> * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
>>> *
>>> * This code is free software; you can redistribute it and/or
>>> modify it
>>> @@ -435,7 +435,7 @@
>>> * The bytes are assumed to be in Intel (little-endian) byte
>>> order.
>>> */
>>> private static final int get16(byte b[], int off) {
>>> - return (b[off] & 0xff) | ((b[off+1] & 0xff) << 8);
>>> + return Byte.toUnsignedInt(b[off]) |
>>> (Byte.toUnsignedInt(b[off+1]) << 8);
>>> }
>>>
>>> /*
>>> diff -r 303b67074666
>>> src/share/demo/nio/zipfs/src/com/sun/nio/zipfs/ZipConstants.java
>>> ---
>>> a/src/share/demo/nio/zipfs/src/com/sun/nio/zipfs/ZipConstants.java Tue
>>> Jan 24 15:13:27 2012 -0500
>>> +++
>>> b/src/share/demo/nio/zipfs/src/com/sun/nio/zipfs/ZipConstants.java Wed
>>> Jan 25 20:31:05 2012 -0800
>>> @@ -185,11 +185,11 @@
>>> */
>>> ///////////////////////////////////////////////////////
>>> static final int CH(byte[] b, int n) {
>>> - return b[n] & 0xff;
>>> + return Byte.toUnsignedInt(b[n]);
>>> }
>>>
>>> static final int SH(byte[] b, int n) {
>>> - return (b[n] & 0xff) | ((b[n + 1] & 0xff) << 8);
>>> + return Byte.toUnsignedInt(b[n]) | (Byte.toUnsignedInt(b[n +
>>> 1]) << 8);
>>> }
>>>
>>> static final long LG(byte[] b, int n) {
>>>
>>> If the changes look good, I'll file a bug for them, etc.
>>>
>>> In case other people want to look over candidates sites in different
>>> areas, I've included the list of JDK files using "0xff" below.
>>>
>>> Thanks,
>>>
>>> -Joe
>>>
>>> [1] 4504839: Java libraries should provide support for unsigned
>>> integer arithmetic
>>> http://mail.openjdk.java.net/pipermail/core-libs-dev/2012-January/008926.html
>>>
>>>
>>> .
>
More information about the core-libs-dev
mailing list