summarylogtreecommitdiffstats
AgeCommit message (Collapse)Author
2019-03-05Update git submodules. Fix build issue with cudnn path.Daniel Bermond
2018-11-22conflicts: add python-pytorchDaniel Bermond
2018-11-22Fix build (disable fbgemm). Add missing libs. Fix insecure rpath.Daniel Bermond
2018-11-10Do not rename source clones. Enable tests.Daniel Bermond
2018-11-08build() and package() improvements. Update submodules and options.Daniel Bermond
2018-10-25Update libibverbs dependencyDaniel Bermond
libibverbs was merged into rdma-core. References ---------- https://lists.archlinux.org/pipermail/aur-requests/2018-October/027305.html
2018-10-25FixesDaniel Bermond
2018-10-03Update conflictsDaniel Bermond
2018-09-30Update method for choosing the cuda architecture to match upstreamDaniel Bermond
The current method for choosing the cuda architecture in the build is by using the TORCH_CUDA_ARCH_LIST environment variable. According to comments in the upstream source code, in the future this method will change to a cmake option of the same name. References ---------- https://github.com/pytorch/pytorch/blob/93ecf4d72ada7fd6cbcf9ec582ecb0e37344bc7c/cmake/public/utils.cmake#L165-L190
2018-09-29Fix build. Remove nanopb git submodule. Add system eigen option.Daniel Bermond
gcc54 is no longer on the official repositories. Fortunatelly, upstream caffe2 git master now builds fine with gcc7, which is the current default cuda compiler (cuda 10.0). nanopb submodule was removed by upstream. The eigen patch to fix build with system eigen is no longer needed because upstream have implemented the cmake option USE_SYSTEM_EIGEN_INSTALL, which we can set to OFF here. References ---------- https://github.com/pytorch/pytorch/commit/964e30de1d1e35b8f5b092ad5126a15cf573724b https://github.com/pytorch/pytorch/commit/c172ffb6321db128fb42f77554869f66406c06c4 https://github.com/pytorch/pytorch/commit/684b55d76257141aa9ea6eb55923896ca4f7d029
2018-08-22Initial commitDaniel Bermond