Source

extradoc / talk / stm2012 / stmimpl.rst

Author Commit Message Labels Comments Date
Armin Rigo
This causes a deadlock. Remove it.
Armin Rigo
Replace ValidateDuringCommit with ValidateDuringTransaction(1). Clean up list_of_read_objects.
Armin Rigo
Fix a subtle bug
Maciej Fijalkowski
backout unnecessary changes
Maciej Fijalkowski
one abstract
Armin Rigo
Cannot compress the whole chain! Discovered this by "targetjit1" in pypy.
Maciej Fijalkowski
add a missing file
Maciej Fijalkowski
add some text to start
Armin Rigo
Speed up allocation by leaving the value of h_revision undefined. Also allows the GC the abuse this h_revision as it wants on newly allocated objects (e.g. on all nursery objects).
Armin Rigo
Mention pointer equality.
Armin Rigo
Tweaks
Armin Rigo
Move down and expand the section about "Barrier placement in the source code".
Armin Rigo
Add a table of barriers.
Armin Rigo
Bug and fix. Found by arigo/hack/stm/python/, which turns out to be a useful hack.
Armin Rigo
Move this comment.
Armin Rigo
A hard bug
Armin Rigo
Kill ValidateForInevitable(). Using ValidateDuringCommit(), which has the correct behavior of detecting locks.
Armin Rigo
Fixes
Armin Rigo
Finish the first version of inevitable transactions
Armin Rigo
in-progress
Armin Rigo
Updates from trying it out in arigo/hack/stm/c2.
Armin Rigo
Fixes
Armin Rigo
Tweaks
Armin Rigo
Needs to describe Local GC, at least up to the effect it needs to have on the flags.
Armin Rigo
Fix again...
Armin Rigo
Bug fix. This is all nice but is missing some form of runnable code and an actual test suite or formal proof :-/
Armin Rigo
Four steps again...
Armin Rigo
Another deadlock to work around here.
Armin Rigo
Update the description (and fix the name of) smp_wmb.
Armin Rigo
Another simplification, which more clearly shows that we were missing a ``smp_wb``.
  1. Prev
  2. 1
  3. 2
  4. Next