Skip to content

Latest commit

 

History

History
165 lines (132 loc) · 4.84 KB

INSTALL.md

File metadata and controls

165 lines (132 loc) · 4.84 KB
Raw

Installation of the CLAW Compiler

Requirements

The CLAW Compiler has the followings dependencies:

  • Java 1.8 or greater
  • Ant 1.9 or greater
  • cmake and make

Additional dependencies, if built with XCodeML tools from OMNI Compiler:

  • yacc, lex
  • C/C++ compiler (supports C99)
  • Fortran compiler (supports Fortran 90)
  • libxml2

Build & install

CLAW Compiler (clawfc) uses XCodeML tools (Fortran frontend and backend) from OMNI Compiler. They can be packaged and built together with the CLAW Compiler.

To build and install the CLAW Compiler, use the following commands.

Out-of-source build:

git clone git@github.com:claw-project/claw-compiler.git
cd claw-compiler
git submodule init
git submodule update --remote
export SRC_DIR="$(pwd)/claw-compiler"
mkdir <build_path>
cd <build_path>
cmake -DCMAKE_INSTALL_PREFIX=<install_path> -S ${SRC_DIR}
make
make install
Specific Java version

If you have several Java compiler version installed on your machine, you must set the cmake variable JAVA_HOME. E.g. cmake -DJAVA_HOME=<install_path> ...

External XCodeML tools at build-time

It is possible to build CLAW Compiler with already installed version of XCodeML tools. To do this, set the cmake option BUILD_OMNI_XCODEML_TOOLS to OFF and specify the path to the XCodeML tools installation directory in cmake variable OMNI_HOME. cmake -DBUILD_OMNI_XCODEML_TOOLS=OFF -DOMNI_HOME=<xcodeml_tools_install_dir> ...

Note however that each version of the CLAW Compiler is tightly bound to the specific version of the XCodeML tools, therefore only the version from CLAW repository with the same GIT commit as the one specified in cmake variable 'OMNI_GIT_COMMIT_HASH' will be compatible.

External XCodeML tools at runtime

By default CLAW Compiler will add XCodeML tools to its install directory. This can be disabled by setting cmake option ADD_OMNI_XCODEML_TOOLS_TO_INSTALL to OFF. Additionally, runtime path to the XCodeML tools can be overridden by setting the environment variable OMNI_HOME before running the CLAW compiler.

Offline build steps

If your system has no network connection to the Internet, you need to get the submodule and the ANT dependencies for the repository. In order to gather all the dependencies for offline build, you can use the following command:

cd claw-compiler
./scripts/offline.sh

Once you have all the dependencies for the repository, you can copy all the data to your target system and add OFFLINE as a CMake option as follows:

cmake -DOFFLINE=ON .
Switch git checkout from https to ssh

If you are building without external OMNI XCodeML tools and your network only allow ssh connection, set the cmake variable OMNI_GIT_REPOSITORY to git@github.com:claw-project/xcodeml-tools.git

Specific steps for Piz Daint

On Piz Daint, specific steps as to be performed in order to have a successful compilation.

First of all, Ant is not available on Piz Daint. To install it, follow the instruction here.

On Piz Daint, the Cray MPI wrapper must be used regardless of the selected programming environment. So if you are compiling with PGI or GNU, use the following commands:

git clone git@github.com:claw-project/claw-compiler.git
cd claw-compiler
mkdir build
cd build
FC=ftn CC=cc CXX=CC cmake -DCMAKE_INSTALL_PREFIX=<install_path> -DOMNI_MPI_FC="MPI_FC=ftn" -DOMNI_MPI_CC="MPI_CC=cc" ..
make
make install

It will use the PGI compiler or the GNU compiler going through the Cray MPI wrapper.

Test your installation with an example

Source code

File: simple_sample.f90

PROGRAM simple_sample
  CALL my_simple_subroutine
END PROGRAM simple_sample

SUBROUTINE my_simple_subroutine
  INTEGER :: i
  !$claw loop-fusion
  DO i=1,2
    PRINT *, 'First loop body:',i
  END DO

  !$claw loop-fusion
  DO i=1,2
    PRINT *, 'Second loop body:',i
  END DO

  !$claw loop-fusion
  DO i=1,2
    PRINT *, 'Third loop body:',i
  END DO
END
Compilation

Compile the original source code to compare the output

gfortran -o simple_sample1 simple_sample1.f90

Apply code transformation and compile the transformed source file

clawfc -o transformed_code.f90 simple_sample.f90  # Generate transformed_code
gfortran -o simple_sample2 transformed_code.f90   # Compile with std compiler
simple_sample1's output:
$ ./simple_sample1
First loop body:           1
First loop body:           2
Second loop body:          1
Second loop body:          2
Third loop body:           1
Third loop body:           2
simple_sample2's output:
$ ./simple_sample2
First loop body:           1
Second loop body:          1
Third loop body:           1
First loop body:           2
Second loop body:          2
Third loop body:           2