diff options
author | Chih-Hsuan Yen | 2022-10-29 00:19:50 +0800 |
---|---|---|
committer | Chih-Hsuan Yen | 2022-10-29 00:19:50 +0800 |
commit | 9c392fb542979981fe0026e0fe3cc361a5f00a36 (patch) | |
tree | 7568dee7109670c8408ab190a23e5cedb92b13d2 | |
parent | 58086ca4650eddd8024f3ad58e7b996160c71bda (diff) | |
download | aur-9c392fb542979981fe0026e0fe3cc361a5f00a36.tar.gz |
python-onnxruntime: re-enable LTO
As mentioned in [1], LTO works again.
Also switches to the new way to configure parallelism of nvcc. Upstream
introduces a similar mechanism in onnxruntime 1.10 [2], and my original
approach (added in [3]) is always overridden.
[1] https://aur.archlinux.org/packages/python-onnxruntime#comment-886838
[2] https://github.com/microsoft/onnxruntime/pull/8974
[3] https://github.com/archlinuxcn/repo/commit/99c193b303a416811c9cdac6b12c30804ad5acb6
-rw-r--r-- | .SRCINFO | 1 | ||||
-rw-r--r-- | PKGBUILD | 4 |
2 files changed, 1 insertions, 4 deletions
@@ -29,7 +29,6 @@ pkgbase = python-onnxruntime depends = re2 depends = openmpi depends = libprotobuf-lite.so - options = !lto source = git+https://github.com/microsoft/onnxruntime#tag=v1.13.1 source = git+https://github.com/onnx/onnx.git source = git+https://github.com/dcleblanc/SafeInt.git @@ -37,8 +37,6 @@ sha512sums=('SKIP' '8f0bd7ae59f86f002c88368a8c2852b9613363771aae61f91a90bfc13dcd3173e43d7988a59ccef86657cf6abfcc53837bbf445c216a7994a765a7e0770d0f5f' '7d55b0d4232183a81c20a5049f259872150536eed799d81a15e7f10b5c8b5279b443ba96d7b97c0e4338e95fc18c9d6f088e348fc7002256ee7170d25b27d80d' 'ab48d27be98a88d3c361e1d0aac3b1e078096c0902ba7a543261a1c24faed0f1f44947a1b7ea1f264434cd2199b9d563d2447c14b6afbdf9900e68a65f7d2619') -# CUDA seems not working with LTO -options+=('!lto') if [[ $_ENABLE_CUDA = 1 ]]; then pkgname+=(onnxruntime-cuda) @@ -103,7 +101,6 @@ build() { # total processes may be much larger than the number of cores - let # the scheduler handle it. cmake_args+=( - -DCMAKE_CUDA_FLAGS="-t0" -DCMAKE_CUDA_ARCHITECTURES="$_CUDA_ARCHITECTURES" -DCMAKE_CUDA_STANDARD_REQUIRED=ON -DCMAKE_CXX_STANDARD_REQUIRED=ON @@ -112,6 +109,7 @@ build() { -DCMAKE_CUDA_COMPILER:PATH=/opt/cuda/bin/nvcc -Donnxruntime_CUDNN_HOME=/usr -Donnxruntime_USE_NCCL=ON + -Donnxruntime_NVCC_THREADS=0 ) fi |