1. Carl Friedrich Bolz
  2. tracebench-pypy

Commits

Show all
Author Commit Message Date Builds
Jeff Terrace
Added pi, sum, min, and max
Armin Rigo
Tests and fix for ``for c in string'' when the string turns out to be always a single character.
Hakan Ardo
Merge jit-targets. It introduces an label ResOperation which can be used as a target for jump operations. Currently the optimizer only supports constructs with a single intermediate label (ie a preamble followed by a peeled loop). The LoopToken have been replaced by a JitCellToken that represents a JitCell and a TargetToken that represents a specific label. The frontend works with jumps between JitCellTokens. Each J…
Hakan Ardo
closing to be merged branch
Branches
jit-targets
Hakan Ardo
hg merge default
Branches
jit-targets
Hakan Ardo
Retracing does not fail in the same set of situations as before. In test_getattr_promote we previously got an retrace that was never used. Now it is not even passed to the backend
Branches
jit-targets
Armin Rigo
merge heads
Armin Rigo
fix comment
Alex Gaynor
Recommit the tolist() for numpy stuff, now that it translates.
Armin Rigo
Added a FAQ entry.
Hakan Ardo
hg merge default
Branches
jit-targets
Hakan Ardo
hg merge default
Branches
jit-targets
Hakan Ardo
a bit more cloning to get a short_preamble for the fallback that does not contain ops from the failed optimization attempt
Branches
jit-targets
Armin Rigo
fix
Armin Rigo
Wrap(r_singlefloat) is not supported by the std obj space either.
Armin Rigo
Hopefully fix test_pypy_c:test_call:test_stararg_virtual, by forcing getitems_fixedsize() to unroll small constant-length lists.
Armin Rigo
hg backout 3eba2ed546ad Does not translate. The issue is that to_builtin_type() is not RPython: the result of self.unbox() is going to be promoted to the widest possible integer, meaning that tests like "assert y.tolist() is True" would fail after translation (if it did translate).
Armin Rigo
Merge the SpecialisedTuples branch, started by Mark W. P. Add a number of interp-level classes for 'tuple', specialized to some small number of items, e.g. (int, int), (int, object), etc. It should be a win both memory-wise and speed-wise, as it reduces a lot the number of reads and checks done to access tuple elements.
Armin Rigo
hg merge default
Armin Rigo
Update the demo.
Armin Rigo
Fix: _hashlib was just broken for any non-standard hash function. Small extra clean-ups.
Armin Rigo
Fix. The testing framework was happily accepting app-level methods that are generators, and then running the yielded functions at interp-level :-( Add a hack to conftest.py to prevent that, and rewrite the test in a non-generator style. Now it fails, good.
Hakan Ardo
hg merge default
Branches
jit-targets
Hakan Ardo
clone the short_boxes to allow it to be reused in some fallabck if the first athempt to optimize fails
Branches
jit-targets
Armin Rigo
Typo.
Armin Rigo
Fix: make sure that PySequence_Fast() returns a W_ListObject or a W_TupleObject, not just some object of type list or tuple. The reason is that PySequence_Fast_GET_xxx() expects these exact interp-level types.
Armin Rigo
Dead import
Maciej Fijalkowski
Merge nedbat-sandbox branch, with one tweak, don't realy on identity of one, just use object() for identitiy checks.
Armin Rigo
Add a test for some of the nonstandard hashes (issue957)
Hakan Ardo
hg merge default
Branches
jit-targets
  1. Prev
  2. Next