Commits

Show all
Author Commit Message Labels Comments Date
Antonio Cuni
(antocuni, arigo): mark more modules as is_core, because they really contain extensions to pyopcode.py
Branches
core-only-tracing
Antonio Cuni
bah, actually fix rpython
Branches
core-only-tracing
Antonio Cuni
tentative rpython fix
Branches
core-only-tracing
Antonio Cuni
(antocuni, arigo, romain around): try to implement a reasonable is_core_function policy for the pypy interpreter
Branches
core-only-tracing
Antonio Cuni
(antocuni, arigo): improve this test
Branches
core-only-tracing
Antonio Cuni
(antocuni, arigo): use a boolean instead of a string to store the value of core-only-mode, and don't disable inlining automatically (for now at least)
Branches
core-only-tracing
Antonio Cuni
(antocuni, arigo): if we are in core mode, we know for sure that the jitcode we are tracing is_core
Branches
core-only-tracing
Antonio Cuni
always disable inlining when setting the jitmode to core-only
Branches
core-only-tracing
Antonio Cuni
rename 'fast' mode to 'core-only'
Branches
core-only-tracing
Antonio Cuni
make sure to inline core-to-core calls
Branches
core-only-tracing
Antonio Cuni
make sure not to inline indirect calls when in core-only mode
Branches
core-only-tracing
Antonio Cuni
fix tests
Branches
core-only-tracing
Antonio Cuni
implement the 'fast' jit mode, in which we trace only inside the 'core' graphs, and do residual calls to everything else; still in-progress, at least one case is missing, see next checkin
Branches
core-only-tracing
Antonio Cuni
(antocuni, arigo, romain) complain if a graph is marked as access_direct but not core, because in this case we would have a wrong behaviour when tracing in core-only mode
Branches
core-only-tracing
Antonio Cuni
(antocuni, arigo, romain): introduce the is_core flag on JitCodes, which will be used to select which graphs to inline when tracing in core-only mode
Branches
core-only-tracing
Antonio Cuni
(antocuni, arigo, bivab, romain): a branch where to experiment with 'lightweight tracing', in which we inline only the opcode implementations and nothing else
Branches
core-only-tracing
Antonio Cuni
(benol) merge the jvm-improvements branch, which fixes (again :-)) the translation for the JVM backend
Antonio Cuni
close about-to-be-merged branch
Branches
jvm-improvements
Michał Bendowski
Fix the implementation of compute_unique_id for _builtin_type.
Branches
jvm-improvements
Michał Bendowski
Simpler implementations of float2longlong and longlong2float.
Branches
jvm-improvements
Michał Bendowski
Implemented float2longlong and longlong2float for the JVM. Also removed the oo_primitive for pypy__rotateLeft - it's not needed on 32 bit architecture (and JVM backend doesn't support 64 bit anyway).
Branches
jvm-improvements
Michał Bendowski
Handle the 'jit_is_virtual' opcode by always returning False
Branches
jvm-improvements
Michał Bendowski
Add a missing cast from Unsigned to UnsignedLongLong in the JVM backend.
Branches
jvm-improvements
Michał Bendowski
Declare oo_primitives that should implement some rffi operations. For now the actual implementations are missing, but once we get the JVM backend to work in some way, this will have to be revisited.
Branches
jvm-improvements
Michał Bendowski
Fix compute_unique_id to support built-ins in ootype. Otherwise the translation fails because it doesn't know how to apply compute_unique_id to a String. In the jvm backend this is implemented by System.identityHashCode() which can be applied to our representations of built-ins equally well as for instances.
Branches
jvm-improvements
Michał Bendowski
Fix userspace builders in ootype Implement the getlength() method of StringBuilders in ootype.
Branches
jvm-improvements
Michał Bendowski
Add files generated by PyCharm to .hgignore
Branches
jvm-improvements
Maciej Fijalkowski
mere
Maciej Fijalkowski
A bit experimental - try to preallocate the size of unicode join and remove a pointless performance hack (the general optimization should work already)
mattip
add failing test, add fix
  1. Prev
  2. Next