1. Pypy
  2. Untitled project
  3. pypy

Commits

Show all
Author Commit Message Date Builds
Armin Rigo
Rewrite the 'jit_conditional_call' operation to 'conditional_call' and always allow it to occur, even if not jitted. Avoids the mess of '*args' and playing around with 'if we_are_jitted()'.
Armin Rigo
Start adapting the codewriter for the new rlist.py format
Armin Rigo
Apply the diff of incminimark.py to minimark.py.
Armin Rigo
Test and fix
Armin Rigo
Kill llmemory.gcarrayofptr_xxx. Found out that we can write the special case in gc/base.py without using them, but using only the size of a GCREF as a special constant.
Armin Rigo
Armin Rigo
Test and fix
Armin Rigo
Be more eager in overallocating lists (but not extra eager in this checkin; need to measure...)
Armin Rigo
Fix for refcounting. Deleted items are kept alive with refcounting, but too bad.
Armin Rigo
One point of this refactoring is to allow list deletions to work without replacing the items with NULL.
Armin Rigo
Use GcArray(overallocated=True) in rlist.py.
Armin Rigo
hg merge default
Armin Rigo
Add this file from the py3k branch, so that it appears as http://doc.pypy.org/en/latest/release-pypy3-2.1.0-beta1.html
Armin Rigo
Add a link to download.html, and mention the issue with -jN
Armin Rigo
Pffff, I see no way to pass down the original MAKEFLAGS to the subprocess. This makes the "-jN" option unusable. As a result, increase the time estimate...
Armin Rigo
Clarify
Armin Rigo
Improve the documentation of -jN
Armin Rigo
Mention the MAKEFLAGS=-j.. option
Armin Rigo
merge heads
Armin Rigo
Tweak the Makefile to run "rpython" using python rather than pypy, if there is no pypy installed. Gives a proper warning in this case.
Armin Rigo
Document the branch.
Armin Rigo
hg merge jit-counter Tweak the jit counters: decay them at minor collection (actually only every 32 minor collection is enough). Should avoid the "memory leaks" observed in long-running processes, actually created by the jit compiling more and more rarely executed paths. Done with refactorings to move them to a single table; should reduce total memory usage. Clean-up, particularly for guard_value.
Armin Rigo
close branch
Branches
jit-counter
Armin Rigo
Retry to tweak these numbers
Branches
jit-counter
Armin Rigo
Translation fix
Branches
jit-counter
Armin Rigo
Adjust "decay" for measured usage (translate.py --annotate): massively reduce the per-minor-collection count, but not the per-new-loop count. Done by only calling decay_all_counters() every 64 minor collections rather than every time.
Branches
jit-counter
Armin Rigo
Oups.
Branches
jit-counter
Armin Rigo
Use systematically unsigned numbers for the timetable index. Generates a bit simpler code.
Branches
jit-counter
Armin Rigo
Increase the size of the timetable cache. It seems to suffer from a bit too many collisions when running translate.py (not too surprizing but still)
Branches
jit-counter
Armin Rigo
Turned out to be a bad idea to re-enable "don't-inline" functions after a short while. Maybe after a long time would make sense, but it seems really counter-productive for now. Fixed, and use a single word instead of three boolean flags.
Branches
jit-counter
  1. Prev
  2. Next