Skip to content

Files

Latest commit

Taras Seredamli
Taras Sereda
and
mli
debugging mxnet c++ operator (apache#4926)
Mar 15, 2017
99cd4ad · Mar 15, 2017

History

History
This branch is 6755 commits behind apache/mxnet:master.

python-howto

README.md

Python Howto Examples

  • Configuring Net to get Multiple Ouputs

  • Configuring Image Record Iterator

  • Set break point in C++ code of the symbol using gdb under Linux:

    • Build mxnet with following values:

      	DEBUG=1 
	CUDA=0 #to make sure convolution-inl.h will be used 
	CUDNN=0 #to make sure convolution-inl.h will be used

    • run python under gdb: gdb --args python debug_conv.py

    • in gdb set break point on particular line of the code and run execution:

(gdb) break src/operator/convolution-inl.h:120 (gdb) run Breakpoint 1, mxnet::op::ConvolutionOp<mshadow::cpu, float>::Forward (this=0x12219d0, ctx=..., in_data=std::vector of length 3, capacity 4 = {...}, req=std::vector of length 1, capacity 1 = {...}, out_data=std::vector of length 1, capacity 1 = {...}, aux_args=std::vector of length 0, capacity 0) at src/operator/./convolution-inl.h:121 121 data.shape_[1] / param_.num_group * param_.kernel[0] * param_.kernel[1]); (gdb) list 116 } 117 Tensor<xpu, 4, DType> data = in_data[conv::kData].get<xpu, 4, DType>(s); 118 Shape<3> wmat_shape = 119 Shape3(param_.num_group, 120 param_.num_filter / param_.num_group, 121 data.shape_[1] / param_.num_group * param_.kernel[0] * param_.kernel[1]); 122 Tensor<xpu, 3, DType> wmat = 123 in_data[conv::kWeight].get_with_shape<xpu, 3, DType>(wmat_shape, s); 124 Tensor<xpu, 4, DType> out = out_data[conv::kOut].get<xpu, 4, DType>(s); 125 #if defined(CUDACC) ```