Loading...
Please wait, while we are loading the content...
Similar Documents
Learning Cost Semantics for Modeling Running Time of OCaml Programs
| Content Provider | Semantic Scholar |
|---|---|
| Author | Das, Ankush Hoffmann, Jan |
| Copyright Year | 2016 |
| Abstract | Programmers would greatly benefit from a source-language model of the running time of compiled code that goes beyond asymptotics. However, it is a difficult problem to accurately model the running time of compiled code at the source level. In this work, we present an operational cost semantics to model the running time of OCaml programs on a specific hardware architecture. Our hypothesis is that the running time of a program can be modeled as a linear function of the number of executions of some high level constructs that we have carefully selected. We learn the coefficients of this linear function by applying traditional machine learning algorithms to automatically adapt the cost semantics to match the measured runtime of training examples on a modern hardware platform. With this learned cost semantics, we are able to model the running time of simple OCaml programs within a reasonable margin of error. Our experiments show that effects of low-level features like memory caches seem to amortize while modeling garbage collection remains a major challenge. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://lola.cse.buffalo.edu/abstracts/LOLA_2016_3.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
