About compiling process in GCJ and garbage collection

Hi there,

I have asked some questions about the plugin loading problem ( refer
to Andrew's answer: http://gcc.gnu.org/ml/java/2011-09/msg00001.html
and Bryce's answer: http://gcc.gnu.org/ml/java/2011-09/msg00003.html )
And now I know ecj is the workhorse when compiling source code to
bytecode.

I still have a question about the compiling process:

if I use GCJ to compile the java source code to machine code directly,
 it will first use ecj to compile java source to bytecode, and then
compile the bytecode(or other information) to machine code.( right? )
BUT, I write a plugin, which removes a certain statement in gimple
tree, and the gcj dump the gimple tree into file. The plugin works.
That means that  GCJ does do something in the process  of compiling
the source code, otherwise the ecj MUST use the same tree
structure(GENERIC, GIMPLE) as gcc does to compile source code in order
to provide information to the compiler of GCJ to do the rest process(
in my case, to provide information for GCJ to remove a certain
statement), which is kind of impossible since ecj is not part of gcc
but GIMPLE is.
Can you help correct me, and explain this?

I have tried the usage of "gcj -v" to print the compiling information,
and I see there are a few files being produced in /tmp, I think that
it is the bytecode of java code, or the STRING type stuff( because I
didn't find any string in the dumped file. )


Thanks,
J. Li