In
computer science
Computer science is the study of computation, information, and automation. Computer science spans Theoretical computer science, theoretical disciplines (such as algorithms, theory of computation, and information theory) to Applied science, ...
, performance prediction means to estimate the execution time or other performance factors (such as
cache
Cache, caching, or caché may refer to:
Science and technology
* Cache (computing), a technique used in computer storage for easier data access
* Cache (biology) or hoarding, a food storing behavior of animals
* Cache (archaeology), artifacts p ...
misses) of a program on a given computer. It is being widely used for computer architects to evaluate new computer designs, for
compiler
In computing, a compiler is a computer program that Translator (computing), translates computer code written in one programming language (the ''source'' language) into another language (the ''target'' language). The name "compiler" is primaril ...
writers to explore new optimizations, and also for advanced developers to tune their programs.
There are many approaches to predict program 's performance on computers. They can be roughly divided into three major categories:
* simulation-based prediction
* profile-based prediction
* analytical modeling
Simulation-based prediction
Performance data can be directly obtained from computer
simulator
A simulation is an imitative representation of a process or system that could exist in the real world. In this broad sense, simulation can often be used interchangeably with model. Sometimes a clear distinction between the two terms is made, in ...
s, within which each instruction of the target program is actually dynamically executed given a particular input data set. Simulators can predict program's performance very accurately, but takes considerable time to handle large programs. Examples include the PACE and Wisconsin Wind Tunnel simulators as well as the more recent WARPP simulation toolkit,
[http://www2.warwick.ac.uk/fac/sci/dcs/research/hpsg/publications/pubs/simutools09.pdf] which attempts to significantly reduce the time required for parallel system simulation.
Another approach, based on
trace-based simulation does not run every instruction, but runs a trace file which store important program events only. This approach loses some flexibility and accuracy compared to cycle-accurate simulation mentioned above but can be much faster. The generation of traces often consumes considerable amounts of storage space and can severely impact the runtime of applications if large amount of data are recorded during execution.
Profile-based prediction
The classic approach of performance prediction treats a program as a set of
basic block
In compiler construction, a basic block is a straight-line code sequence with no branches in except to the entry and no branches out except at the exit. This restricted form makes a basic block highly amenable to analysis. Compilers usually decom ...
s connected by execution path. Thus the execution time of the whole program is the sum of execution time of each basic block multiplied by its execution frequency, as shown in the following formula:
The execution frequencies of basic blocks are generated from a
profiler, which is why this method is called profile-based prediction. The execution time of a basic block is usually obtained from a simple instruction scheduler.
Classic profile-based prediction worked well for early single-issue, in-order execution processors, but fails to accurately predict the performance of modern processors. The major reason is that modern processors can issue and execute several instructions at the same time, sometimes out of the original order and cross the boundary of basic blocks.
References
{{reflist
Software optimization