nixpkgs/pkgs/applications/science/math/cntk/default.nix
Markus Kowalewski 6dba41fbcb
mpi: use mpi attribute consistently as the default MPI implementations
Use the attribute mpi to provide a system wide default MPI
implementation. The default is openmpi (as before).
This now allows for overriding the MPI implentation by using
the overlay mechanism. Build all packages with mpich instead
of the default openmpi can now be achived like this:
self: super:
 {
   mpi = super.mpich;
 }

All derivations that have been using "mpi ? null" to provide optional
building with MPI have been change in the following way to allow for
optional builds with MPI:
{ ...
, mpi
, useMpi ? false
}
2021-01-23 12:15:13 +01:00

104 lines
3 KiB
Nix

{ lib, stdenv, fetchgit, fetchFromGitHub, cmake
, openblas, blas, lapack, opencv3, libzip, boost, protobuf, mpi
, onebitSGDSupport ? false
, cudaSupport ? false, addOpenGLRunpath, cudatoolkit, nvidia_x11
, cudnnSupport ? cudaSupport, cudnn
}:
assert cudnnSupport -> cudaSupport;
assert blas.implementation == "openblas" && lapack.implementation == "openblas";
let
# Old specific version required for CNTK.
cub = fetchFromGitHub {
owner = "NVlabs";
repo = "cub";
rev = "1.7.4";
sha256 = "0ksd5n1lxqhm5l5cd2lps4cszhjkf6gmzahaycs7nxb06qci8c66";
};
in stdenv.mkDerivation rec {
pname = "CNTK";
version = "2.7";
# Submodules
src = fetchgit {
url = "https://github.com/Microsoft/CNTK";
rev = "v${version}";
sha256 = "18l9k7s966a26ywcf7flqyhm61788pcb9fj3wk61jrmgkhy2pcns";
};
nativeBuildInputs = [ cmake ] ++ lib.optional cudaSupport addOpenGLRunpath;
# Force OpenMPI to use g++ in PATH.
OMPI_CXX = "g++";
buildInputs = [ openblas opencv3 libzip boost protobuf mpi ]
++ lib.optional cudaSupport cudatoolkit
++ lib.optional cudnnSupport cudnn;
configureFlags = [
"--with-opencv=${opencv3}"
"--with-libzip=${libzip.dev}"
"--with-openblas=${openblas}"
"--with-boost=${boost.dev}"
"--with-protobuf=${protobuf}"
"--with-mpi=${mpi}"
"--cuda=${if cudaSupport then "yes" else "no"}"
# FIXME
"--asgd=no"
] ++ lib.optionals cudaSupport [
"--with-cuda=${cudatoolkit}"
"--with-gdk-include=${cudatoolkit}/include"
"--with-gdk-nvml-lib=${nvidia_x11}/lib"
"--with-cub=${cub}"
] ++ lib.optional onebitSGDSupport "--1bitsgd=yes";
configurePhase = ''
sed -i \
-e 's,^GIT_STATUS=.*,GIT_STATUS=,' \
-e 's,^GIT_COMMIT=.*,GIT_COMMIT=v${version},' \
-e 's,^GIT_BRANCH=.*,GIT_BRANCH=v${version},' \
-e 's,^BUILDER=.*,BUILDER=nixbld,' \
-e 's,^BUILDMACHINE=.*,BUILDMACHINE=machine,' \
-e 's,^BUILDPATH=.*,BUILDPATH=/homeless-shelter,' \
-e '/git does not exist/d' \
Tools/generate_build_info
patchShebangs .
mkdir build
cd build
${lib.optionalString cudnnSupport ''
mkdir cuda
ln -s ${cudnn}/include cuda
export configureFlags="$configureFlags --with-cudnn=$PWD"
''}
../configure $configureFlags
'';
installPhase = ''
mkdir -p $out/bin
# Moving to make patchelf remove references later.
mv lib $out
cp bin/cntk $out/bin
'';
postFixup = lib.optionalString cudaSupport ''
for lib in $out/lib/*; do
addOpenGLRunpath "$lib"
done
'';
meta = with lib; {
# Newer cub is included with cudatoolkit now and it breaks the build.
# https://github.com/Microsoft/CNTK/issues/3191
broken = cudaSupport;
homepage = "https://github.com/Microsoft/CNTK";
description = "An open source deep-learning toolkit";
license = if onebitSGDSupport then licenses.unfreeRedistributable else licenses.mit;
platforms = [ "x86_64-linux" ];
maintainers = with maintainers; [ abbradar ];
};
}