1. vinodm
  2. pypy

Source

pypy / pypy / jit / metainterp / executor.py

Author Commit Message Date Builds
Antonio Cuni
hg merge default
Branches
jitypes2
Armin Rigo
Backout 4d4b01d56455 and investigate the failure, probably related, of benchmarks.
Armin Rigo
Giant hack. See comments in optimizeopt/virtualize.py
Armin Rigo
hg merge default
Branches
jitypes2
Alex Gaynor
Merged default into numpy-exp.
Branches
numpy-exp
Maciej Fijalkowski
merge default
Branches
numpy-exp
Antonio Cuni
hg merge default
Branches
jitypes2
Armin Rigo
hg merge default
Armin Rigo
hg merge default
Branches
jit-lsprofile
Maciej Fijalkowski
Backed out changeset 124ebb7828dd, merge of out-of-line-guards-2. It seems to have broken benchmarks, investigation needed
Armin Rigo
First test passing on the front-end.
Antonio Cuni
partial merge of default; _ctypes/function.py had very hard conflicts, I'll try to manually merge the changes in the next commits
Branches
jitypes2
Maciej Fijalkowski
in-progress. Start implementing vector operations. It's a bit annoying because we don't (can't?) have a type representing sse vector while not jitted so a bit of dance is required.
Branches
numpy-exp
Armin Rigo
Replace 'call_may_force' with 'call_release_gil' in order to call external C functions. The idea is that 'call_release_gil' should release the GIL and, for the GC, close the stack in the real backend.
Branches
jitypes2
Alex Gaynor
Merged default into numpy-exp.
Branches
numpy-exp
Armin Rigo
Start supporting both 32-bit and 64-bit. Type differences: BoxFloats store floats on 64-bit but longlongs on 32-bit, and read_timestamp() returns an int on 64-bit but a r_longlong on 32-bit.
Branches
jit-lsprofile
Alex Gaynor
(alex, fijal): make jit aware of read_timestamp
Branches
jit-lsprofile
Maciej Fijalkowski
Make a basic SSE operation on array work. It doesn't guarantee as of now that arrays are aligned. Have to think for a second about how to do it, probably by operating on raw arrays with no length and calling special functions for POSIX. *VERY* experimental
Branches
numpy-exp
Armin Rigo
More fixes. metainterp/test passes again.
Branches
jit-longlong-2
Armin Rigo
In-progress.
Branches
jit-longlong-2
Armin Rigo
In-progress: turns the BoxFloat and ConstFloat attribute '.value' to be of type SignedLongLong on 32-bit machines. This is needed because it seems not reliable to store longlongs in a float field; see the test.
Branches
jit-longlong-2
Amaury Forgeot d'Arc
hg merge default
Branches
fast-forward
Dan Villiom Podlaski Christiansen
No-op merge to register the merge done in 0e3099e3cc33 with Mercurial.
Branches
fast-forward
Armin Rigo
Merge branch/rsre-jit: put the JIT hints in the rsre module. Requires small extensions in the JIT: - "green fields": the ability to write 'foo.bar' in the list of greens in the JitDriver, where 'foo' is written in the list of reds. - the ability to only write jitdriver.jit_merge_point(), without any jitdriver.can_enter_jit(). In that case the jit_merge_point() plays both roles. The difference with putting explicitly a can_enter_jit() just before is that such a can_enter_jit() i…
Antonio Cuni
merge the jitffi branch: it provides a new jit-friendly rlib.libffi module, and an app-level _ffi module which exposes it.
Maciej Fijalkowski
(arigo, fijal merging) Merge jit-str branch. This branch allows JIT to do something saner about string operations (for example virtualize a slice)
Carl Friedrich Bolz
(cfbolz, arigo checking): constant-fold ovf operations
Maciej Fijalkowski
(hopefully) fix test_basic in x86 backend
Benjamin Peterson
want a pointer not an integer
Armin Rigo
issue545 testing Change CALL_PURE into a temporary operation. The idea is to fix issue545 by preventing multiple calls to a pure function. Now a call to a pure function is traced as a normal CALL, then immediately turned into a CALL_PURE by whacking at the history, and the value that it returned is inserted as first argument. If optimizeopt can constant-fold it away, it thus has the returned value available. I…
  1. Prev
  2. 1
  3. 2
  4. 3
  5. 4
  6. Next