24 lines
1.1 KiB
TeX
24 lines
1.1 KiB
TeX
\section{Related Work}
|
|
%% I want to talk in this section about:
|
|
%% - work of Ertl
|
|
%% - work of Wang et al.
|
|
%% - how iso c became unusable for os dev
|
|
%% - ubs that are used for optimizations
|
|
|
|
Little work has been done in the area of detecting the performance
|
|
speedup of UB optimizations in big software projects. Wang et
|
|
al.~\cite{wang2012undefined} and Ertl~\cite{ertl2015every} provide
|
|
metrics for this class of optimizations based on SPECint benchmark.
|
|
|
|
Wang et al. state that they observed a decrease in performance of $7.2\%$
|
|
with GCC and $9.0\%$ with Clang for 456.hammer and $6.3\%$ with GCC and
|
|
$11.8\%$ with Clang for 462.libquantum. The experiments were conducted
|
|
with UB optimizations turned off.
|
|
|
|
Ertl states that with Clang-3.1 and UB optimizatoins turned on the
|
|
speedup factor is $1.017$ for SPECint 2006. Furthermore, for a specific
|
|
class of programs, i.e. Jon Bentley's traveling salesman problem, the
|
|
speedup factor can reach values greater than $2.7$ if the programmer
|
|
issues source-level optimizations by hand, surpasing the UB
|
|
optimizatoins issued by the compiler.
|