As for CUDA, installing older Developer Command Line Tools with clang 9.0, the
compilation advances much further, but then the linker fails with
```
ld: warning: directory not found for option '-L/usr/local/cuda/lib64'
ld: warning: directory not found for option '-L/usr/local/cuda/lib64/stubs'
Undefined symbols for architecture x86_64:
"mxnet::op::CuDNNAlgoReg<mxnet::op::ConvolutionParam>::Get()", referenced
from:
mxnet::op::CuDNNConvolutionOp<float>::SelectAlgo(mxnet::RunContext
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, cudnnDataType_t, cudnnDataType_t) in convolution_gpu.o
mxnet::op::CuDNNConvolutionOp<double>::SelectAlgo(mxnet::RunContext
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, cudnnDataType_t, cudnnDataType_t) in convolution_gpu.o
mxnet::op::CuDNNConvolutionOp<mshadow::half::half_t>::SelectAlgo(mxnet::RunContext
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, cudnnDataType_t, cudnnDataType_t) in convolution_gpu.o
"mxnet::op::CuDNNAlgoReg<mxnet::op::DeconvolutionParam>::Get()", referenced
from:
mxnet::op::CuDNNDeconvolutionOp<float>::SelectAlgo(mxnet::RunContext
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, cudnnDataType_t, cudnnDataType_t) in deconvolution_gpu.o
mxnet::op::CuDNNDeconvolutionOp<double>::SelectAlgo(mxnet::RunContext
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, cudnnDataType_t, cudnnDataType_t) in deconvolution_gpu.o
mxnet::op::CuDNNDeconvolutionOp<mshadow::half::half_t>::SelectAlgo(mxnet::RunContext
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, std::__1::vector<nnvm::TShape, std::__1::allocator<nnvm::TShape> >
const&, cudnnDataType_t, cudnnDataType_t) in deconvolution_gpu.o
ld: symbol(s) not found for architecture x86_64
```
`-L/usr/local/cuda/lib64` is the CUDA location on Linux, not on the Mac. This
must be hard-coded somewhere. Adding a soft-link doesn't help with the above
error message though.
[ Full content available at:
https://github.com/apache/incubator-mxnet/issues/12568 ]
This message was relayed via gitbox.apache.org for [email protected]
