Skip to content

laurenarnett/TensorFlock

master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
lrm
 
 
 
 
src
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

TensorFlock

Joseph Morag - Language Guru (jm4157@columbia.edu) Lauren Arnett - System Architect (lba2138@columbia.edu) Elena Ariza - Manager (esa2150@columbia.edu) Nick Buonincontri - Tester (nb2413@columbia.edu)

Compilation of TensorFlock

External Dependencies

TensorFlock requires the following to be installed on your local system:

  • opam, version 1.2.2 or greater
  • OCaml, version 4.06 or greater, installed with opam -- Though it has not been tested we've tried to avoid language features added after 4.03; earlier versions of OCaml will likely work, but we make no guarantees.
  • LLVM, version 3.8
  • OCaml's LLVM bindings, version 3.8, installed with opam -- LLVM and OCaml's LLVM bindings may use later versions but the versions must match

Build

First, you need to declare where LLVM's binaries are located on your system. If LLVM's bin is already in your path then you can skip this step.

To declare LLVM's location cp the file named local.example to local.mkInlocal.mkadd this assignment:LLVM_PATH = /path/to/llvm/bin`.

The Makefile contains a target to first verify lli can be found on the system path, then if the LLVM_PATH variable has been set - always in that order, the path takes precedence over the LLVM_PATH variable. If neither condition is met it will error with a message when attempting to run make targets that require LLVM.

If using the Docker container described below, you do not need to setup local.mk. The container has LLVM's bin in the system path.

Run make to compile TensorFlock's compiler. TensorFlock is packaged with an opam file that will install any needed dependencies as a local pin.

After running make the compiler will be named toplevel.native and can be found in the root directory of the project.

Compiling and Executing TensorFlock Programs

The TensorFlock compiler can take one of four flags:

  • -a: prints the AST
  • -s: runs the semantic checker
  • -l: prints generated LLVM IR
  • -c: generates LLVM bytecode, saved to output.ll

For example the compiler is executed by running: ./toplevel.native -c tests/some_file.tf

"Hello World"

For ease of demonstration a make demo target has been added to the Makefile. This compiles the TensorFlock compiler, which in turn compiles our demo program to LLVM bytecode, which is then passed to the LLVM interpreter.

Testing

Run make test to execute the test runner.

Testing Scheme

Our test runner checks three levels of compilation: parsing, semantic checking, and codegen.

Parsing and semantic checking are verified on the basis of the exit code of the compilation process. Passing tests are expected to exit on 0, failing tests are expected to exit on some n > 0.

In addition to the manner described above, passing codegen tests are also tested by comparing the output of the executed program with a canonical value. (The value is saved in a corresponding file with the extension .pass eg: some_test.tf => some_test.pass

Docker

To ensure a consistent development environment we've created a Docker image with all external dependencies installed, as well as provide build targets to make, run, and test TensorFlock in a corresponding Docker containers.

The Dockerfile, which declares how the image is built, can be found in the docker directory.

To use Docker, first install Docker on your host system and verify that the Docker demon is running. Then:

  • make docker-make: runs make inside a container and exits
  • make docker-test: runs make test inside a container and exits
  • make docker-shell: drops you into a shell of a running container, inside /root/TensorFlock

Some notes about our Docker setup:

  • The initial run of the container on the host will pull down the image, which is ~1GB. (We used Ubuntu 16.04 as a base system; in retrospect a smaller distro would have been better.)
  • The TensorFlock directory on the host machine is effectively mounted to /root/TensorFlock inside the container. Disk writes on the host in this directory will appear in the client, and vice-versa.
  • The container is discarded upon exit. Any changes to the client outside of /root/TensorFlock are lost when the container stops. This is done to maintain the consistency of the build environment as well as prevent the buildup of old containers on the host.

About

A small functional tensor language with Einstein summation notation convention and shape-checking at compile-time.

Resources

Stars

Watchers

Forks

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •