Author: Carl Friedrich Bolz <[email protected]>
Branch: extradoc
Changeset: r3795:73a66fe07d24
Date: 2011-06-27 15:49 +0200
http://bitbucket.org/pypy/extradoc/changeset/73a66fe07d24/
Log: XXXs
diff --git a/talk/icooolps2011/jit-hints.pdf b/talk/icooolps2011/jit-hints.pdf
index
c78b3b84550a3db53382fb1fb1a9a97c0596a4ef..afbc33e62d31d10aa178450e5a83b3e086fee9b8
GIT binary patch
[cut]
diff --git a/talk/iwtc11/paper.bib b/talk/iwtc11/paper.bib
--- a/talk/iwtc11/paper.bib
+++ b/talk/iwtc11/paper.bib
@@ -109,6 +109,16 @@
year = {2009}
},
+@inproceedings{bolz_runtime_2011,
+ address = {Lancaster, {UK}},
+ title = {Runtime Feedback in a {Meta-Tracing} {JIT} for Efficient
Dynamic Languages},
+ abstract = {Meta-tracing {JIT} compilers can be applied to a variety of
different languages without explicitly encoding language semantics into the
compiler. So far, they lacked a way to give the language implementor control
over runtime feedback. This restricted their performance. In this paper we
describe the mechanisms in {PyPy’s} meta-tracing {JIT} that can be used
to control runtime feedback in language-specific ways. These mechanisms are
flexible enough to express classical {VM} techniques such as maps and runtime
type feedback.},
+ booktitle = {{ICOOOLPS}},
+ publisher = {{ACM}},
+ author = {Bolz, Carl Friedrich and Cuni, Antonio and Fijałkowski,
Maciej and Leuschel, Michael and Rigo, Armin and Pedroni, Samuele},
+ year = {2011}
+},
+
@inproceedings{chang_tracing_2009,
address = {Washington, {DC}},
title = {Tracing for Web 3.0: Trace Compilation for the Next Generation
Web Applications},
diff --git a/talk/iwtc11/paper.tex b/talk/iwtc11/paper.tex
--- a/talk/iwtc11/paper.tex
+++ b/talk/iwtc11/paper.tex
@@ -45,7 +45,6 @@
\usepackage{listings}
\usepackage{beramono}
-
\definecolor{gray}{rgb}{0.3,0.3,0.3}
\lstset{
@@ -444,9 +443,6 @@
\section{Making Trace Optimizations Loop Aware}
-XXX make clear that the preamble is not necessarily the \emph{first} iteration
-of a loop
-
Before the trace is passed to a backend compiling it into machine code
it needs to be optimized to achieve better performance.
The focus of this paper
@@ -486,6 +482,8 @@
However, the peeled loop can then be optimized using the assumption that a
previous iteration has happened.
+XXX (samuele): the point about the first iteration is hard to understand
+
When applying optimizations to this two-iteration trace
some care has to taken as to how the arguments of the two
\lstinline{jump} operations and the input arguments of the peeled loop are
@@ -752,8 +750,8 @@
In the general case, each allocation-removed object in the jump arguments is
exploded into a
vector of variables containing the values of all registered
-fields\footnote{This is sometimes called \emph{scalar replacement}. XXX check
-whether that's true}. If some of the fields are themselves references to
+fields\footnote{This is sometimes called \emph{scalar replacement}.}.
+If some of the fields are themselves references to
allocation-removed objects they are recursively exploded
to make the vector contain only concrete variables. Some care has
to be taken to always place the fields in the same order when
@@ -945,7 +943,8 @@
We can observe that PyPy (even without loop peeling) is orders of magnitude
faster than either CPython or Psyco. This is due to the JIT compilation
-advantages and optimizations we discussed in XXX [ref to other paper]. Loop
+advantages and optimizations we discussed in previous work
+\cite{bolz_allocation_2011, bolz_runtime_2011}. Loop
peeling gives an additional XXX on average, which makes benchmark times
comparable with native-compiled C code. Missing performance we attribute to
the relative immaturity of PyPy's JIT assembler backend as well as missing
@@ -960,8 +959,10 @@
\section{Related Work}
\label{sec:related}
-All the optimizations presented here are completely standard
-\cite{muchnick_advanced_1997}. XXX
+The effect of combining a one ass optimization with loop peeling gives
+completely standard loop invariant code motion optimizations
+\cite{muchnick_advanced_1997}. We do not claim any novelty in the effect, but
+think that our implementation scheme is a very simple one.
Mike Pall, the author of LuaJIT\footnote{\texttt{http://luajit.org/}} seems to
have developped the described technique independently. There are no papers
about
_______________________________________________
pypy-commit mailing list
[email protected]
http://mail.python.org/mailman/listinfo/pypy-commit
