Speed Impact experiment on GCJ

Tue Feb 14 16:14:00 GMT 2006

Hello:
It looks like you didn't specify optimization in your native compile. 
This can often make a big difference. You might want to try adding "-O2 
-g0" to your compliation flags and try again.
We are big-time users of gcj. On a real-world blend of Java use cases, 
native-complied GCJ modules are almost always about 30-40% faster than 
the same code interpreted on Sun (tm) JVM (tm). Sometimes the disparity 
is even more skewed in native-gcj's favor. If you don't get results 
like this, something you're doing is wrong.
hth,
craig vanderborgh
voxware incorporated
Rui Wang wrote:
> Hi,
> I did a simple test to analyse the speed impact of using GCJ between
> java bytecode and native machine code. But the result looks not very 
> promising, the data below shows that there
> is rather large decrease of speed instead of increase. I couldn't 
> figure out why, any suggestions are appreciated.
> The results of my experiment are the followings:
>> For native version:
>> Number of calls Execution time (ms)
> 200.0 1221
> 300.0 1831
> 400.0 2446
> 500.0 3060
> For java version:
> Number of calls Execution time(ms)
> 200.0 663
> 300.0 1007
> 400.0 1310
> 500.0 1643
> This is how I approached the experiment:
> The experiment is done on Linux with Fedora core. I have three java 
> classes: Client.java IFace.java and Impl.java.
>> For Client.java
>> public class Client {
> public static void main(String[] args) throws Exception {
> BufferedReader tasks = new BufferedReader(new
> FileReader(args[0])),
> sDetails = new BufferedReader(new
> FileReader(args[1]));
> BufferedWriter output = new BufferedWriter(new
> FileWriter(args[2]));
> String line = null;
> List servers = new LinkedList();
> while((line = sDetails.readLine()) != null)
> {
> Impl obj = new Impl();
> servers.add(obj);
> }
> double currTask = 0;
> int r = 0, index = 0;
> long start = 0, end = 0, time = 0;
> double throughput = 0, responseTime=0;
>> while((line = tasks.readLine()) != null)
> {
> currTask = Double.parseDouble(line);
> start = System.currentTimeMillis();
> for (int i = 0; i < currTask; i++)
> {
> r = ((Impl)servers.get(index)).ping(1000,
> 1000);
> if((index +1) < servers.size())
> index++;
> else
> index = 0;
> }
> end = System.currentTimeMillis();
>> index = 0;
> time = end - start;
>> output.write(currTask+"\t"+time);
> output.newLine();
> }
> tasks.close();
> sDetails.close();
> output.close();
> }
> } 
> For Impl.java
> public class Impl implements IFace {
> public Impl(){
> }
>> public int ping(int i, int j) {
> int count= 0;
> for(int n = 0 ; n < i*j; n++)
> count++;
> return count;
> }
> }
>> For IFace.java
> public interface IFace {
> public int ping(int i, int j);
> }
>> I also have a script to compile and run the app.
> PREFIX=${HOME}/gcc/bin/
> GCJ=${PREFIX}gcj
> GRMIC=${PREFIX}grmic
> GRMIREGISTRY=${PREFIX}grmiregistry
>> export PATH=$PATH:/home/pg/n1704655/gcc/bin
> export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/pg/n1704655/gcc/lib
>> $GCJ -C -fCLASSPATH=./ *java
> sleep 2
> echo finish compile
>> $GCJ --classpath=./ -c Client.class -o Client.o
> $GCJ --classpath=./ -c IFace.class -o IFace.o
> $GCJ --classpath=./ -c Impl.class -o Impl.o
> sleep 2
> echo finish creat object files
>> $GCJ *.o --main=Client -o client
>> ./client task.test servers.d result.d
> echo client started
>> Any ideas why? In the code, I don't think I did anything that will slow
> things down.
> Thank you very much
>> Rui
>>>>
Confidentiality Note: This message may contain information which is privileged or confidential and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient, or the employee responsible for delivering the message to the intended recipient, you are hereby NOTIFIED that any dissemination, distribution, retention, archiving, or copying of this communication is strictly prohibited. If you received this email in error, please notify Voxware immediately by return email to the sender.