[pypy-svn] r63831 - pypy/extradoc/talk/icooolps2009
cfbolz at codespeak.net
cfbolz at codespeak.net
Wed Apr 8 15:11:56 CEST 2009
Author: cfbolz
Date: Wed Apr 8 15:11:56 2009
New Revision: 63831
Modified:
pypy/extradoc/talk/icooolps2009/paper.tex
Log:
Benchmark Psyco
Modified: pypy/extradoc/talk/icooolps2009/paper.tex
==============================================================================
--- pypy/extradoc/talk/icooolps2009/paper.tex (original)
+++ pypy/extradoc/talk/icooolps2009/paper.tex Wed Apr 8 15:11:56 2009
@@ -746,8 +746,8 @@
\hline
1 &Compiled to C, no JIT &23.44 $\pm$ 0.07 &1.00\\
2 &Compiled to C, with JIT &3.58 $\pm$ 0.05 &6.54\\
-3 &CPython 2.5.4 &4.96 $\pm$ 0.05 &4.73\\
-\hline
+3 &CPython 2.5.2 &4.96 $\pm$ 0.05 &4.73\\
+4 &CPython 2.5.2 + Psyco 1.6 &1.51 $\pm$ 0.05 &15.57\\\hline
\end{tabular}
\caption{Benchmarked function and results for the Python interpreter running
@@ -762,13 +762,15 @@
nearly all of the involved operations, like indexing into the tuple, addition
and comparison of \texttt{i}. We benchmarked PyPy's Python interpreter with the
JIT disabled, with the JIT enabled and
-CPython\footnote{\texttt{http://python.org}} 2.5.4 (the reference implementation of
-Python).
+CPython\footnote{\texttt{http://python.org}} 2.5.2 (the reference implementation of
+Python). In addition we benchmarked CPython using Psyco 1.6
+\cite{rigo_representation-based_2004}, a specializing JIT compiler for Python.
The results show that the tracing JIT speeds up the execution of this Python
function significantly, even outperforming CPython. To achieve this, the tracer
traces through the whole Python dispatching machinery, automatically inlining
-the relevant fast paths.
+the relevant fast paths. However, the manually tuned Psyco still performs a lot
+better than our prototype.
\section{Related Work}
More information about the Pypy-commit
mailing list