search

GPUOpen-ProfessionalCompute-Libraries / amdovx-modules

AMD OpenVX modules: such as, neural network inference, 360 video stitching, etc.
110 stars 50 forks source link

Segmentation Fault running "./anntest weights.bin" #187

Closed arunnura123 closed 6 years ago

arunnura123 commented 6 years ago

I see the following segmentation Fault when I try running the anntest.

  1. Onnx model ( Googlenet ) is converted to NNIR
  2. NNIR to Vx format using the python utilities

./anntest weights.bin OK: loaded 21 kernels from libvx_nn.so OK: OpenVX using GPU device#0 (gfx900) [OpenCL 1.2 ] [SvmCaps 0 1] MIOpen Error: std::bad_alloc ERROR: fatal error occured at /home/amd/Miopen-opencl/amdovx-modules/vx_nn/src/convolution_layer.cpp#200 Segmentation fault (core dumped)

Could you help me on what I am missing

Thanks, Arun

rgiduthuri commented 6 years ago

Interesting...

Looks like memory allocation inside MIOpen convolution initialization failed. How much memory is on the GPU? Is some other app running on GPU at the same time?

rgiduthuri commented 6 years ago

Can you share the onnx model that you’re using?

arunnura123 commented 6 years ago

I am using https://github.com/onnx/models/tree/master/bvlc_googlenet 16GB Memory . I will try uninstalling rocm and rebuilding again.

Thanks

arunnura123 commented 6 years ago

This issue seems to be resolved after I installed "apt-get install miopen-opencl" instead of building from source.