You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@ws.apache.org by ve...@apache.org on 2017/05/21 13:47:21 UTC
svn commit: r1795701 -
/webservices/axiom/trunk/components/core-streams/src/main/java/org/apache/axiom/core/stream/serializer/Serializer.java
Author: veithen
Date: Sun May 21 13:47:21 2017
New Revision: 1795701
URL: http://svn.apache.org/viewvc?rev=1795701&view=rev
Log:
Optimize Serializer for better CPU cache utilization.
Modified:
webservices/axiom/trunk/components/core-streams/src/main/java/org/apache/axiom/core/stream/serializer/Serializer.java
Modified: webservices/axiom/trunk/components/core-streams/src/main/java/org/apache/axiom/core/stream/serializer/Serializer.java
URL: http://svn.apache.org/viewvc/webservices/axiom/trunk/components/core-streams/src/main/java/org/apache/axiom/core/stream/serializer/Serializer.java?rev=1795701&r1=1795700&r2=1795701&view=diff
==============================================================================
--- webservices/axiom/trunk/components/core-streams/src/main/java/org/apache/axiom/core/stream/serializer/Serializer.java (original)
+++ webservices/axiom/trunk/components/core-streams/src/main/java/org/apache/axiom/core/stream/serializer/Serializer.java Sun May 21 13:47:21 2017
@@ -38,6 +38,12 @@ import org.apache.axiom.core.stream.seri
* @xsl.usage internal
*/
public final class Serializer implements XmlHandler {
+ /**
+ * The number of characters to process at once. Chosen small enough to leverage processor caches
+ * and large enough to reduce method invocation overhead.
+ */
+ private static final int CHUNK_SIZE = 4096;
+
private static final int MIXED_CONTENT = 0;
private static final int TAG = 1;
private static final int ATTRIBUTE_VALUE = 2;
@@ -87,7 +93,7 @@ public final class Serializer implements
* Reusing this buffer means not creating a new character array
* everytime and it runs faster.
*/
- private char[] charsBuff = new char[60];
+ private final char[] charsBuff = new char[CHUNK_SIZE];
public Serializer(Writer out) {
writer = new WriterXmlWriter(out);
@@ -332,138 +338,129 @@ public final class Serializer implements
final XmlWriter writer = this.writer;
final int context = this.context;
+ final String illegalCharacterSequence = illegalCharacterSequences[context];
- String illegalCharacterSequence = illegalCharacterSequences[context];
- if (illegalCharacterSequence != null) {
- int matchedIllegalCharacters = this.matchedIllegalCharacters;
- for (int i = 0; i < length; i++) {
- while (true) {
- if (chars[start+i] == illegalCharacterSequence.charAt(matchedIllegalCharacters)) {
- if (++matchedIllegalCharacters == illegalCharacterSequence.length()) {
- throw new IllegalCharacterSequenceException("Illegal character sequence \"" + illegalCharacterSequence + "\"");
- }
- break;
- } else if (matchedIllegalCharacters > 0) {
- int offset = 1;
- loop: while (offset < matchedIllegalCharacters) {
- for (int j = 0; j < matchedIllegalCharacters - offset; j++) {
- if (illegalCharacterSequence.charAt(j) != illegalCharacterSequence.charAt(j+offset)) {
- offset++;
- continue loop;
- }
- }
- break;
- }
- matchedIllegalCharacters -= offset;
- } else {
- break;
- }
- }
- }
- this.matchedIllegalCharacters = matchedIllegalCharacters;
- }
-
- if (context == CDATA_SECTION || context == COMMENT || context == PROCESSING_INSTRUCTION) {
- // TODO: this doesn't take care of illegal characters
- try {
- writer.write(chars, start, length);
- } catch (IOException ex) {
- throw new StreamException(ex);
- }
- return;
- }
-
try {
int i;
final int end = start + length;
int lastDirtyCharProcessed = start - 1; // last non-clean character that was processed
// that was processed
+ int matchedIllegalCharacters = this.matchedIllegalCharacters;
int squareBrackets = this.squareBrackets;
for (i = start; i < end; i++) {
char ch = chars[i];
- String replacement = null;
- boolean generateCharacterReference = false;
- if (ch <= 0x1F) {
- // Range 0x00 through 0x1F inclusive
- //
- // This covers the non-whitespace control characters
- // in the range 0x1 to 0x1F inclusive.
- // It also covers the whitespace control characters in the same way:
- // 0x9 TAB
- // 0xA NEW LINE
- // 0xD CARRIAGE RETURN
- //
- // We also cover 0x0 ... It isn't valid
- // but we will output "�"
-
- // The default will handle this just fine, but this
- // is a little performance boost to handle the more
- // common TAB, NEW-LINE, CARRIAGE-RETURN
- switch (ch) {
- case 0x09:
- if (context == ATTRIBUTE_VALUE) {
- replacement = "	";
+ if (illegalCharacterSequence != null) {
+ while (true) {
+ if (ch == illegalCharacterSequence.charAt(matchedIllegalCharacters)) {
+ if (++matchedIllegalCharacters == illegalCharacterSequence.length()) {
+ throw new IllegalCharacterSequenceException("Illegal character sequence \"" + illegalCharacterSequence + "\"");
}
break;
- case 0x0A:
- if (context == ATTRIBUTE_VALUE) {
- replacement = " ";
+ } else if (matchedIllegalCharacters > 0) {
+ int offset = 1;
+ loop: while (offset < matchedIllegalCharacters) {
+ for (int j = 0; j < matchedIllegalCharacters - offset; j++) {
+ if (illegalCharacterSequence.charAt(j) != illegalCharacterSequence.charAt(j+offset)) {
+ offset++;
+ continue loop;
+ }
+ }
+ break;
}
+ matchedIllegalCharacters -= offset;
+ } else {
break;
- case 0x0D:
- replacement = " ";
- // Leave whitespace carriage return as a real character
- break;
- default:
- generateCharacterReference = true;
- break;
+ }
}
}
- else if (ch < 0x7F) {
- switch (ch) {
- case '<':
- replacement = "<";
- break;
- case '>':
- if (context == MIXED_CONTENT && squareBrackets >= 2) {
- replacement = ">";
- }
- break;
- case '&':
- replacement = "&";
- break;
- case '"':
- if (context == ATTRIBUTE_VALUE) {
- replacement = """;
- }
+
+ String replacement = null;
+ boolean generateCharacterReference = false;
+
+ if (context == MIXED_CONTENT || context == ATTRIBUTE_VALUE) {
+ if (ch <= 0x1F) {
+ // Range 0x00 through 0x1F inclusive
+ //
+ // This covers the non-whitespace control characters
+ // in the range 0x1 to 0x1F inclusive.
+ // It also covers the whitespace control characters in the same way:
+ // 0x9 TAB
+ // 0xA NEW LINE
+ // 0xD CARRIAGE RETURN
+ //
+ // We also cover 0x0 ... It isn't valid
+ // but we will output "�"
+
+ // The default will handle this just fine, but this
+ // is a little performance boost to handle the more
+ // common TAB, NEW-LINE, CARRIAGE-RETURN
+ switch (ch) {
+ case 0x09:
+ if (context == ATTRIBUTE_VALUE) {
+ replacement = "	";
+ }
+ break;
+ case 0x0A:
+ if (context == ATTRIBUTE_VALUE) {
+ replacement = " ";
+ }
+ break;
+ case 0x0D:
+ replacement = " ";
+ // Leave whitespace carriage return as a real character
+ break;
+ default:
+ generateCharacterReference = true;
+ break;
+ }
+ } else if (ch < 0x7F) {
+ switch (ch) {
+ case '<':
+ replacement = "<";
+ break;
+ case '>':
+ if (context == MIXED_CONTENT && squareBrackets >= 2) {
+ replacement = ">";
+ }
+ break;
+ case '&':
+ replacement = "&";
+ break;
+ case '"':
+ if (context == ATTRIBUTE_VALUE) {
+ replacement = """;
+ }
+ }
+ } else if (ch <= 0x9F) {
+ // Range 0x7F through 0x9F inclusive
+ // More control characters, including NEL (0x85)
+ generateCharacterReference = true;
+ } else if (ch == 0x2028) {
+ // LINE SEPARATOR
+ replacement = "
";
+ }
+
+ if (ch == ']') {
+ squareBrackets++;
+ } else {
+ squareBrackets = 0;
}
- } else if (ch <= 0x9F) {
- // Range 0x7F through 0x9F inclusive
- // More control characters, including NEL (0x85)
- generateCharacterReference = true;
- } else if (ch == 0x2028) {
- // LINE SEPARATOR
- replacement = "
";
}
- if (replacement != null || generateCharacterReference) {
- int startClean;
- startClean = lastDirtyCharProcessed + 1;
+
+ int startClean = lastDirtyCharProcessed + 1;
+ int lengthClean = i - startClean;
+ if (replacement != null || generateCharacterReference || lengthClean == CHUNK_SIZE) {
if (startClean < i) {
- int lengthClean = i - startClean;
writer.write(chars, startClean, lengthClean);
}
if (replacement != null) {
writer.write(replacement);
- } else {
+ } else if (generateCharacterReference) {
writer.writeCharacterReference(ch);
}
lastDirtyCharProcessed = i;
- } else if (ch == ']') {
- squareBrackets++;
- } else {
- squareBrackets = 0;
}
}
@@ -474,6 +471,7 @@ public final class Serializer implements
int lengthClean = i - startClean;
writer.write(chars, startClean, lengthClean);
}
+ this.matchedIllegalCharacters = matchedIllegalCharacters;
this.squareBrackets = squareBrackets;
} catch (IOException ex) {
throw new StreamException(ex);
@@ -485,11 +483,13 @@ public final class Serializer implements
}
void characters(String s, int start, int length) throws StreamException {
- if (length > charsBuff.length) {
- charsBuff = new char[length * 2 + 1];
+ while (length > 0) {
+ int count = Math.min(length, CHUNK_SIZE);
+ s.getChars(start, start+count, charsBuff, 0);
+ characters(charsBuff, 0, count);
+ start += count;
+ length -= count;
}
- s.getChars(start, length, charsBuff, 0);
- characters(charsBuff, 0, length);
}
@Override