alioth/nixpkgs
3
0
Fork 0
forked from mirrors/nixpkgs
Code Releases Activity

python3Packages.pytorch: remove oneDNN dependency

Browse source 
oneDNN was added as a dependency, but it is not actually used by
PyTorch. PyTorch uses oneDNN from the vendored iDeep dependency.

Using a system-provided oneDNN is currently not a supported build
option.
This commit is contained in:
Daniël de Kok 2020-08-26 10:46:22 +02:00
parent ff926743b6
commit a4ed797948
1 changed files with 5 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
pkgs/development/python-modules/pytorch/default.nix
View file

@ -4,7 +4,7 @@
openMPISupport ? false, openmpi ? null,
buildDocs ? false,
cudaArchList ? null,
numpy, pyyaml, cffi, click, typing, cmake, oneDNN, hypothesis, numactl, psutil,
numpy, pyyaml, cffi, click, typing, cmake, hypothesis, numactl, psutil,
linkFarm, symlinkJoin,
# virtual pkg that consistently instantiates blas across nixpkgs
@ -159,9 +159,9 @@ in buildPythonPackage rec {
USE_MKL = blas.implementation == "mkl";
# Unlike MKL, MKLDNN is FOSS, so we enable support for it by default. Note
# that this was renamed to dnnl and then renamed again to oneDNN upstream, but
# pytorch still calls it by the old name mkldnn.
# Unlike MKL, oneDNN (née MKLDNN) is FOSS, so we enable support for
# it by default. PyTorch currently uses its own vendored version
# of oneDNN through Intel iDeep.
USE_MKLDNN = mklDnnSupport;
USE_MKLDNN_CBLAS = mklDnnSupport;
@ -210,7 +210,7 @@ in buildPythonPackage rec {
ninja
] ++ lib.optionals cudaSupport [ cudatoolkit_joined ];
buildInputs = [ blas blas.provider oneDNN ]
buildInputs = [ blas blas.provider ]
++ lib.optionals cudaSupport [ cudnn magma nccl ]
++ lib.optionals stdenv.isLinux [ numactl ];