1 \documentclass[pdftex,
a4paper,
11pt
]{report}
3 \usepackage[utf8
]{inputenc}
4 \usepackage[T1]{fontenc}
5 \usepackage[english
]{babel
}
11 \title{PIPS - List of code transformations
}
18 \chapter{List of Pips transformations
}
20 \section{Memory allocation alteration
}
22 \section{Loop transformations
}
24 \section{Inter-procedural transformations
}
26 \section{Base blocs transformations
}
28 \section{Dead code removal
}
34 \item[loop unrolling
]{
35 is a loop transformation. Unrolling a loop by a factor of $n$ consists in the substitution of a loop
36 body by itself, replicated $n$ times. A prelude and/or postlude are
37 added to preserve the number of iteration.
}
40 is a function transformation. Inlining a function
41 \texttt{foo
} in its caller
\texttt{bar
} consists in the substitution
42 of the calls to
\texttt{foo
} in
\texttt{bar
} by the function body
43 after replacement of the formal parameters by their effective
47 \item[forward substitution
]{
48 is the process of replacing a reference read in an
49 expression by the latest expression affected to it.
}
52 is a loop transformation that replaces two loops by a single loops whose
53 body is the concatenation of the bodies of the two initial loops.
}
56 is a loop nest transformation that changes the loop execution order
57 through a partitions of the iteration space into
58 chunks, so that the iteration is performed over each chunk and in the
61 \item[reduction detection
]{
62 is an analysis that identifies statements that perform a reduction over a
65 \item[parallelism detection
]{
66 is a common name for analysis that detect if a loop can be run in parallel.
}
68 \item[parallelism extraction
]{
69 is a common name for code transformations that modifies loop nest to make it legal to
70 run them in parallel.
}
72 \item[directive generation
]{
73 is a common name for code transformations that annotate the code with directives.
}
75 \item[constant propagation
]{
76 is a pass that replaces a variable by its value when this value is
77 known at compile time.
}
79 \item[instruction selection
]{
80 is the process of mapping parts of the IR to machine instructions.
}
82 \item[goto elimination
]{
83 is the process of replacing
\texttt{goto
} instructions by a hierarchical control flow
86 is the process of extracting part of a function body into a new function
87 and replacing it in the initial function by a function call.
}
89 \item[common subexpression elimination
]{
90 is the process of replacing similar expressions by a variable that holds
91 the result of their evaluation.
}
93 \item[loop interchange
]{is a loop transformation that permutes two
94 loops from a loop nest.
}
96 \item[loop unswitching
]{is a loop transformation that replaces a
97 loop containing a test independent from the loop execution by a test
98 containing the loop without the test in both true and false branch.
}
100 \item[statement isolation
]{is the process of replacing all
101 variables referenced in a statement by newly declared variables. A
103 and an epilogue are added to copy old variable values to new variable, back
106 \item[dead code elimination
]{is the process of pruning from a
107 function all the statements whose results are
110 \item[array linearization
]{is the process of converting
111 multidimensional array into unidimensional arrays, possibly with a
112 conversion from array to pointer.
}
114 \item[privatization
]{is the process of detecting variables that
115 are private to a loop body, i.e.\ written first, then read.
}
117 \item[loop normalization
]{is a loop transformation that changes
118 the loop initial increment value or the loop range to enforce certain values, generally
121 \item[iteration clamping
]{is a loop transformation that extends
122 the loop range but guards the loop body with the former range.
}
124 \item[flatten code
]{is the process of pruning a function body from
125 declaration blocks so that all declarations are made at the top level.
}
127 \item[strength reduction
]{is the process of replacing an operation
128 by an operation of lower cost.
}
130 \item[split update operator
]{is the process of replacing an update
131 operator by its expanded form.
}
133 \item[n address code generation
]{is the process of splitting
134 complex expression in simpler ones that take at most $n$ operands.
}
136 \item[memory footprint reduction
]{is the process of tiling a loop
137 to make sure the iteration over the tile has a memory footprint bounded by
140 \item[redundant load-store elimination
]{is an inter procedural transformation
141 that optimizes data transfers by delaying and merging them.
}
143 \item[invariant code motion
]{is a loop transformation that moves outside
145 the code from its body that is independent from the iteration.
}
149 \item[loop rerolling
]{finds manually unrolled loop and replace
150 them by their non-unrolled version.
}