- 03 Nov, 2005 3 commits
-
-
Samuele Pedroni authored
-
Samuele Pedroni authored
-
Samuele Pedroni authored
-
- 02 Nov, 2005 4 commits
-
-
Michael Hudson authored
-
Samuele Pedroni authored
-
Michael Hudson authored
namespace -- before a (recursive!) dictionary called 'globals' would appear.
-
Samuele Pedroni authored
-
- 01 Nov, 2005 7 commits
-
-
Michael Hudson authored
-
Samuele Pedroni authored
make the ll_strcpy signature less dangerous for lifetime issues, with the new one sensible keepalive can be generated on inlining...
-
Samuele Pedroni authored
-
Samuele Pedroni authored
-
Samuele Pedroni authored
disentangle the fortunes of bpnn.py (it was broken) from pypy/lib/random.py which doesn't need to be RPython and could also go away at some point.
-
Samuele Pedroni authored
fixes the output of the example in getting-started
-
Michael Hudson authored
-
- 31 Oct, 2005 18 commits
-
-
Samuele Pedroni authored
-
Samuele Pedroni authored
-
Armin Rigo authored
-
Christian Tismer-Sperling authored
changed interface to stackless. optimized stackless to do a full analysis which call-site really needs stackless support. This almost halves the speed difference between pypy-c-stackless and pypy-c. This is most of what can be done. There is a small opportunity to special-case tail recursive functions, but I doubt this is much fruit.
-
Christian Tismer-Sperling authored
-
Samuele Pedroni authored
-
Christian Tismer-Sperling authored
This has the advantage that the picture gets much more complete. Sometimes it now gets over-complete, because transitions which already existed before some transformations, may be found twice, now. A cleanup may be needed at some point, but for now it is good enough. See for instance targetrpystonedalone.py, where I have put a recursion just for testing it with graphics (and actually stackless). The recursion in pysone_main is shown twice, now.
-
Christian Tismer-Sperling authored
The problem was that we do some initial callgraph, but in the later transformations, we don't update that, and it would be not so trivial to enforce this. Therefore, I created a property complete_callgraph, which is recomputed whenever new graphs are added.
-
Armin Rigo authored
(cbuild.py is quite messy, so this involved some refactoring...)
-
Christian Tismer-Sperling authored
-
Christian Tismer-Sperling authored
-
Armin Rigo authored
-
Armin Rigo authored
-
Armin Rigo authored
-
Armin Rigo authored
-
Michael Hudson authored
did anything noteworthy get discussed over the weekend?
-
Christian Tismer-Sperling authored
out that the problem is not about the physical distance of code, but we need to use coloring of cache lines, instead.
-
Christian Tismer-Sperling authored
-
- 30 Oct, 2005 8 commits
-
-
Samuele Pedroni authored
-
Samuele Pedroni authored
-
Armin Rigo authored
* add a link to draft-dynamic-language-translation, where there is at least a few paragraphs about what GenC does
-
Armin Rigo authored
Do we really want this at all? Or should it be longer?
-
Armin Rigo authored
* Concrete mode execution * Constant propagation * Termination with non-static aspects
-
Armin Rigo authored
-
Armin Rigo authored
-
Armin Rigo authored
-