onnxruntime: Static build fails to register ops
You provide an SO file in your releases. Could you provide a monolythic static library as well. When building from source, you get the following static libraries:
libonnxruntime_common.a
libonnxruntime_framework.a
libonnxruntime_graph.a
libonnxruntime_mlas.a
libonnxruntime_optimizer.a
libonnxruntime_providers.a
libonnxruntime_session.a
libonnxruntime_test_utils.a
libonnxruntime_util.a
libonnx_test_data_proto.a
libonnx_test_runner_common.a
lib/libgmock.a
lib/libgtest.a
onnx/libonnx.a
onnx/libonnx_proto.a
(there may be more)
These could be combined using a .mri script. For example, create the file libonnxruntime.mri which contains:
create libonnxruntime.a
addlib libonnxruntime_common.a
addlib libonnxruntime_framework.a
addlib libonnxruntime_graph.a
addlib libonnxruntime_mlas.a
addlib libonnxruntime_optimizer.a
addlib libonnxruntime_providers.a
addlib libonnxruntime_session.a
addlib libonnxruntime_test_utils.a
addlib libonnxruntime_util.a
addlib libonnx_test_data_proto.a
addlib libonnx_test_runner_common.a
addlib lib/libgmock.a
addlib lib/libgtest.a
addlib onnx/libonnx.a
addlib onnx/libonnx_proto.a
save
end
About this issue
- Original URL
- State: closed
- Created 4 years ago
- Comments: 15 (4 by maintainers)
This repository is insanely good! It’s so fast. And with an MRI script like the one above, it is a totally dependency free, single file static library (or shared library with no scripting). You guys should be really pushing for this to be the goto inference library. It’s much better than OpenCV’s DNN library or Tencent’s NCNN, or feathercnn, or paddlepaddle, or anything else I’ve tried.