--- Comment #56 from Leandro Lucarella <> 2010-07-26 06:40:21 
PDT ---
(In reply to comment #55)
> If/when this makes it into D2 I'm going to store the info in a parallel array
> kind of like the bits are.

Will you store a pointer to the pointer bitmask? In that case you'll have an
overhead of 4096/16 = 256, 256*4 = 1KiB (32 bits; 2KiB for 64) for each
allocated page, even when a page might have all its data marked as NO_SCAN.
That's 25% overhead (50% for 64 bits). Even if that overhead is non-GC'ed
memory, so it won't get scanned or anything, it looks like too much. If you
plan to store it only when necessary and being dependent on the bin size, it
would be very hard to address, I think.

I'll try storing the bits directly for small bins (16..128 for 32 bits, ..256
for 64) first, it seems easier and it looks like it could help.

Configure issuemail:
------- You are receiving this mail because: -------

Reply via email to