Abstract
This paper presents the structure of program execution traces. This study extends previous works from many researchers aiming to quantify ILP. Our goal is to understand the general structure of a run and the instruction parallelism it offers. This structure is mainly made of two opposing parts. One has a high ILP which can infinitely increase when the trace lengthens and the other has a very low ILP (close to 1) which can extend infinitely with the trace. The first part comes from loops controls (loops with a counter set from a constant). The second part comes from functions parameters transmissions. Our results show that, without any modification of the ILP distribution, a processor must conciliate a high resource demand at the start of the run and a nearly sequential execution at the end.