Re: [Jikesrvm-researchers] use of available bits in VM_JavaHeader.java

[Nihit <ni...@st...>]

Hi David

I am using Watson Mark & Sweep GC.

I am doing precisely what has been suggested by you. Since 
ADDRESS_BASED_HASHING is set to false by default, There are two bits 
available in GC Header. I did VM_MiscHeader.REQUESTED_BITS = 2. Now I am 
calling VM_ObjectModel.setAvailableBit(ref,1,true) from initializeHeader() 
of VM_MiscHeader.java. I am checking this bit in VM_Allocator.java of 
watson/markSweep in the method allocateScalar(). The result is that the 
test method returns false (bit not set) always. One of the reasons 
suggested before was that the memory region might not be set when the test 
function is called i.e. it may be premature to call this testAvailableBit 
method in allocateScalar(). If this is the case, when can I call 
testAvailableBit().

Also I noticed when I tested the available bit in the same method where I 
set the bit, testAvailableBit() return true (bit is set). This does show 
that the use of available bit in MiscHeader does work.

Suggest me what would be the appropritate place for setting & testing 
these bits.

Nihit Kasabwala
M.Tech CSE, IIIT-Hyderabad
India


On Thu, 1 May 2003, David P Grove wrote:

> The object model definitions are more than a little confusing.  We're 
> trying to break things into nice modular pieces, but still have all of the 
> important values be compile time constants.  This efficiency requirement 
> leads to some amount of non-obviousness in the definitions 
> 
> In theory, all you have to do is change VM_MiscHeader.REQUESTED_BITS to 
> the number of bits that you want..  The convention is that the GC system 
> takes available bits 0 through VM_AllocatorHeader.REQUESTED_BITS-1.   You 
> access them by using the VM_ObjectModel functions to test/set available 
> bits.  If you request more bits than are available you should get an 
> assertion failure during bootimage writing. 
> 
> The use of bits by the MiscHeader isn't something we do very often, but it 
> is supposed to work.
> 
> Hope this helps,
> 
> --dave
> 
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jik...@ww...
> http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jikesrvm-researchers
>