Trace Tree (TT) is a technique to collect program execution traces, which is commonly used in JIT environments. Its main features are the ability to perform loop unrolling and function inlining at no cost, while detecting application loop kernels. In this paper we evaluate a TT implementation in a DBT environment. We show that, under DBT, trace trees suffer from severe code duplication, considerably degrading its performance. In order to take advantage of the TTs interesting features in DBTs, we propose a variation called Compact Trace Trees (CTTs), which we show to be faster and to reduce code duplication.
Mendeley saves you time finding and organizing research
There are no full text links
Choose a citation style from the tabs below