TAPAS: Generating Parallel Accelerators fromParallel Programs

Citation

	@inproceedings{margem2018tapas,
		title={TAPAS: Generating Parallel Accelerators from Parallel Programs},
		author={Margerm, Steven and Amirali, Sharifian and Guha, Apala and Shriraman, Arrvindh},
		booktitle={Microarchitecture (MICRO), 2018 51th Annual IEEE/ACM International Symposium on},
		pages={1--14},
		year={2018},
		organization={IEEE}
	

Getting Started

Official supported environment for building and running TAPAS is ubuntu 16.04. You have to run these commands to install required packages from ubuntu package repositories :

# apt-get install build-essential cmake sbt verilator

Build

To build TAPAS you should run following commands :

$ git clone [email protected]/sfu-arch/TAPAS.git
$ cd TAPAS
$ git clone --recursive https://github.com/sfu-arch/Tapir-Meta.git
$ cd Tapir-Meta/
$ ./build.sh
$ source ./setup-env.sh
$ cd ..; mkdir build; cd build;
$ cmake -DLLVM_DIR=<your repository>/Tapir-Meta/tapir/build/lib/cmake/llvm/ -DTAPIR=ON ..
$ make

Running tests

Inside test directory there are set of test example which show the generality of TAPAS. To get the generated accelerator files for test cases you can run:

$ # in your code repository
$ cd test/c/
$ make all

For each test case there is going to be one sacala file which has the detailed implementation of TAPAS.

How to run generator on your code (detailed way)?

For generating .scala for your code the following steps need to be taken :

1. Emit the llvm ir (.ll) for your code with supported Tapir/TAPAS compiler(<your repository code location>/Tapir-Meta/tapir/build/bin/clang).
   • You can out put llvm ir with this command ./Tapir-Meta/tapir/build/bin/clang -emit-llvm [Your source code]
2. Run opt with -mem2reg -loop-simplify -loop-simplifycfg -disable-loop-vectorization -dce arguments on your llvm ir (.ll) code.
3. Run TAPAS generator on your .ll file like this :
   • <your repository>/build/bin/dandelion -fn-name=[output file of last step] -config=../../scripts/config.json -o output.scala

How to run generator on your code (simple way)?

You can simply put your .c file in the <your repository>/test/c directory and simply run :

$ make all

After that you should have the .scala file beside your .c file with a same name.

Help

./bin/dandelion -help

dandelion options:

  -aa-trace                - Alias analysis trace
  -config=<config_file>    - Target function name
  -fn-name=<Function name> - Target function name
  -l-ex                    - Extracting loops
  -o=<filename>            - tapas output file
  -test-file               - Printing Test file

Generic Options:

  -help                    - Display available options (-help-hidden for more)
  -help-list               - Display list of available options (-help-list-hidden for more)
  -version                 - Display the version of this program

Extra Documentation

For more information you can look at the wiki. https://github.com/sfu-arch/dandelion-lib/wiki

Contributors

• Amirali Sharifian (Simon Fraser University)
• Steven Margem (Simon Fraser University)
• Navid Rahimi (Simon Fraser University)
• Apala Guha (Simon Fraser University)
• Arrvindh Shriraman (Simon Fraser University)