Synopsis

This software package contains the Auto Tuning Optimization System (ATOS) tools.

Travis CI build status

Overview

To easily install the ATOS software packages and all its dependencies, please go to the atos-build project and follow the instructions described here.

After installation the following tools are available:

• Top level tools:
  • atos-explore: automatic optimization system for a build and run system that generate a build configuration to be used by atos-play (see atos-explore -h)
  • atos-graph: graphs the optimization space and best trade-offs frontier (see atos-graph -h)
  • atos-play: build system optimizer for a build system (see atos-play -h)
• Companion tools:
  • atos-audit: auditing tool for a build system that generates a build template to be used by atos-build (see atos-audit -h)
  • atos-raudit: auditing tool for a run system that generates a run template to be used by atos-run (see atos-raudit -h)
  • atos-deps: specify executables targets and complete the auditing process (see atos-deps -h)
  • atos-build: replay build audited by atos-audit with custome options injection (see atos-build -h)
  • atos-run: rerun benchmark audited by atos-raudit (see atos-run -h)
  • atos-profile: specific rebuild!rerun for generating application profiles (see atos-profile -h)

Examples

Execute 'make examples' for an example exploration on two version of sha1sum (C and C++ versions). It will do the exploration and open the optimization space graph. See the INSTALL file if any trouble.

Usage

A simple usage of the atos-explore tool is to provide a script for building the application (compile.sh for instance) and a script for running the benchmarks (exec.sh for instance), then run the atos-explore tool: For instance:

atos-explore -b ./compile.sh -r ./exec.sh
....
Best Size configuration: ...
Best Performance configuration: ...

Then in order to choose your best trade-off configuration run the atos-graph tool:

atos-graph
... display trade-off graph

By clicking on the best trade-off point that you choose on the drawn optimization space frontier, get the local_id and pass it to atos-play.

Last, in order to replay the build for your chosen configuration local_id:

atos-play -l <local_id>
... compile your application for your chosen best trade-off

Or simply for best size or best perf:

atos-play -f time
... compile your application for best performance
atos-play -f size
... compile your application for best code size