From: Pierre Guillou <guillou@cri.ensmp.fr>
Date: Tue, 11 Feb 2014 10:42:57 +0000 (+0100)
Subject: added content of Serge Guelton's thesis transformations glossary
X-Git-Url: https://svn.cri.ensmp.fr/git/pipstransfo.git/commitdiff_plain/61e96cd04cf67182f81330cfbb76b9b2c5c7b892?ds=sidebyside;hp=--cc

added content of Serge Guelton's thesis transformations glossary
---

61e96cd04cf67182f81330cfbb76b9b2c5c7b892
diff --git a/pipstransfo.tex b/pipstransfo.tex
index e69de29..4cef713 100644
--- a/pipstransfo.tex
+++ b/pipstransfo.tex
@@ -0,0 +1,153 @@
+\documentclass[pdftex, a4paper, 11pt]{report}
+
+\usepackage[utf8]{inputenc}
+\usepackage[T1]{fontenc}
+\usepackage[english]{babel}
+
+\usepackage{listings}
+\usepackage{hyperref}
+
+
+\title{PIPS - List of code transformations}
+
+
+
+\begin{document}
+
+
+\chapter{List of Pips transformations}
+
+\section{Memory allocation alteration}
+
+\section{Loop transformations}
+
+\section{Inter-procedural transformations}
+
+\section{Base blocs transformations}
+
+\section{Dead code removal}
+
+
+
+
+\begin{description}
+\item[loop unrolling]{
+  is a loop transformation. Unrolling a loop by a factor of $n$ consists in the substitution of a loop
+  body by itself, replicated $n$ times. A prelude and/or postlude are
+  added to preserve the number of iteration.}
+
+\item[inlining]{
+  is a function transformation. Inlining a function
+  \texttt{foo} in its caller \texttt{bar} consists in the substitution
+  of the calls to \texttt{foo} in \texttt{bar} by the function body
+  after replacement of the formal parameters by their effective
+  parameters.
+}
+
+\item[forward substitution]{
+  is the process of replacing a reference read in an
+  expression by the latest expression affected to it.}
+
+\item[loop fusion]{
+  is a loop transformation that replaces two loops by a single loops whose
+  body is the concatenation of the bodies of the two initial loops.}
+
+\item[loop tiling]{
+  is a loop nest transformation that changes the loop execution order
+  through a partitions of the iteration space into
+  chunks, so that the iteration is performed over each chunk and in the
+  chunks.}
+
+\item[reduction detection]{
+  is an analysis that identifies statements that perform a reduction over a
+  variable.}
+
+\item[parallelism detection]{
+  is a common name for analysis that detect if a loop can be run in parallel.}
+
+\item[parallelism extraction]{
+  is a common name for code transformations that modifies loop nest to make it legal to
+  run them in parallel.}
+
+\item[directive generation]{
+  is a common name for code transformations that annotate the code with directives.}
+
+\item[constant propagation]{
+  is a pass that replaces a variable by its value when this value is
+  known at compile time.}
+
+\item[instruction selection]{
+  is the process of mapping parts of the IR to machine instructions.}
+
+\item[goto elimination]{
+  is the process of replacing \texttt{goto} instructions by a hierarchical control flow
+  graph.}
+\item[outlining]{
+  is the process of extracting part of a function body into a new function
+  and replacing it in the initial function by a function call.}
+
+\item[common subexpression elimination]{
+  is the process of replacing similar expressions by a variable that holds
+  the result of their evaluation.}
+
+\item[loop interchange]{is a loop transformation that permutes two
+  loops from a loop nest.}
+
+\item[loop unswitching]{is a loop transformation that replaces a
+  loop containing a test independent from the loop execution by a test
+  containing the loop without the test in both true and false branch.}
+
+\item[statement isolation]{is the process of replacing all
+  variables referenced in a statement by newly declared variables. A
+  prologue
+  and an epilogue are added to copy old variable values to new variable, back
+  and forth.}
+
+\item[dead code elimination]{is the process of pruning from a
+  function all the statements whose results are
+  never used.}
+
+\item[array linearization]{is the process of converting
+  multidimensional array into unidimensional arrays, possibly with a
+  conversion from array to pointer.}
+
+\item[privatization]{is the process of  detecting variables that
+  are private to a loop body, i.e.\ written first, then read.}
+
+\item[loop normalization]{is a loop transformation that changes
+  the loop initial increment value or the loop range to enforce certain values, generally
+  1.}
+
+\item[iteration clamping]{is a loop transformation that extends
+  the loop range but guards the loop body with the former range.}
+
+\item[flatten code]{is the process of pruning a function body from
+  declaration blocks so that all declarations are made at the top level.}
+
+\item[strength reduction]{is the process of replacing an operation
+  by an operation of lower cost.}
+
+\item[split update operator]{is the process of replacing an update
+  operator by its expanded form.}
+
+\item[n address code generation]{is the process of splitting
+  complex expression in simpler ones that take at most $n$ operands.}
+
+\item[memory footprint reduction]{is the process of tiling a loop
+  to make sure the iteration over the tile has a memory footprint bounded by
+  a given value.}
+
+\item[redundant load-store elimination]{is an inter procedural transformation
+  that optimizes data transfers by delaying and merging them.}
+
+\item[invariant code motion]{is a loop transformation that moves outside
+  of the loop
+  the code from its body that is independent from the iteration.}
+
+
+
+\item[loop rerolling]{finds manually unrolled loop and replace
+  them by their non-unrolled version.}
+\end{description}
+
+\end{document}