| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
When a stack is occupying less than 1/4 of the memory it owns, and is
larger than a megablock, we release half of it. Shrinking is O(1), it
doesn't need to copy the stack.
|
| |
|
|
|
|
|
|
|
| |
eg. use +RTS -g2 -RTS for 2 threads. Only major GCs are parallelised,
minor GCs are still sequential. Don't use more threads than you
have CPUs.
It works most of the time, although you won't see much speedup yet.
Tuning and more work on stability still required.
|
| |
|
|
|
|
|
|
| |
The main goal here is to reduce fragmentation, which turns out to be
the case of #743. While I was here I found some opportunities to
improve performance too. The code is rather more complex, but it also
contains a long comment describing the strategy, so please take a look
at that for the details.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
In the free list, we don't strictly speaking need to have every block
in a coalesced group point to the head block, although this is an
invariant for non-free blocks. Dropping this invariant for the free
list means that coalesce() is O(1) rather than O(N), and freeGroup()
is therefore O(N) not O(N^2).
The bad case probably didn't happen most of the time, indeed it has
never shown up in a profile that I've seen. I had a report from a
while back that this was a problem with really large heaps, though.
Fortunately the fix is easy.
|
|
|
In preparation for parallel GC, split up the monolithic GC.c file into
smaller parts. Also in this patch (and difficult to separate,
unfortunatley):
- Don't include Stable.h in Rts.h, instead just include it where
necessary.
- consistently use STATIC_INLINE in source files, and INLINE_HEADER
in header files. STATIC_INLINE is now turned off when DEBUG is on,
to make debugging easier.
- The GC no longer takes the get_roots function as an argument.
We weren't making use of this generalisation.
|
