Jikes RVM

look in VM.boot().  there is a special call to setup the write buffer for the
concurrent collector before VM_Allocator.boot() is called.

david

Hezi Azatchi wrote:

> Hi,
>
> Which writebarrier is invoked before the VM_Allocator.boot() is executed?
>
> In particular, the concurrent writebarrier implementation uses
> a buffer which is initialized at VM_Allocator.setupProcessor(), thus this
> writebarrier can't be used before the buffer is initialized, so which
> method performs the updates before that stage?
>
> Thanks,
> --Hezi.
>
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jikesrvm-researchers@...
> http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jikesrvm-researchers

I was wondering whether there is any reason for not allocating the 
INSTRUCTION[] (BTW: why do you use this "alias" for byte[]?) in a 
separate memory area which is not garbage collected.

Isn't it the case that classes don't get unloaded (no ClassLoader 
objects that could get collected), and thus compiled code basically is 
immortal. If that's true, wouldn't it be a good idea to allocate the 
byte[] in VM_Assembler.getMachineCode() in a special area of memory into 
which the GCs don't look?

Are there any big hurdles to overcome in implementing this idea? I am 
thinking about doing it to reduce the influence of my heavy 
instrumentations (blowing up code size by a factor of 3 to 10) on the 
memory system.

-Matthias