Export: other / legacy¶

n2d2 "mnist24_16c4s2_24c5s2_150_10.ini" -export CPP_OpenCL

Export types:

C C export using OpenMP;
C_HLS C export tailored for HLS with Vivado HLS;
CPP_OpenCL C++ export using OpenCL;
CPP_Cuda C++ export using Cuda;
CPP_cuDNN C++ export using cuDNN;
SC_Spike SystemC spike export.

Other program options related to the exports:

Option [default value]	Description
`-nbbits` [8]	Number of bits for the weights and signals. Must be 8, 16, 32 or 64 for integer export, or -32, -64 for floating point export. The number of bits can be arbitrary for the `C_HLS` export (for example, 6 bits). It must be -32 for the `CPP_TensorRT` export, the precision is directly set at runtime
`-calib` [0]	Number of stimuli used for the calibration. 0 = no calibration (default), -1 = use the full test dataset for calibration
`-calib-passes` [2]	Number of KL passes for determining the layer output values distribution truncation threshold (0 = use the max. value, no truncation)
`-no-unsigned`	If present, disable the use of unsigned data type in integer exports
`-db-export` [-1]	Max. number of stimuli to export (0 = no dataset export, -1 = unlimited)

C export¶

Test the exported network:

cd export_C_int8
make
./bin/n2d2_test

The result should look like:

...
1652.00/1762    (avg = 93.757094%)
1653.00/1763    (avg = 93.760635%)
1654.00/1764    (avg = 93.764172%)
Tested 1764 stimuli
Success rate = 93.764172%
Process time per stimulus = 187.548186 us (12 threads)

Confusion matrix:
-------------------------------------------------
| T \ E |       0 |       1 |       2 |       3 |
-------------------------------------------------
|     0 |     329 |       1 |       5 |       2 |
|       |  97.63% |   0.30% |   1.48% |   0.59% |
|     1 |       0 |     692 |       2 |       6 |
|       |   0.00% |  98.86% |   0.29% |   0.86% |
|     2 |      11 |      27 |     609 |      55 |
|       |   1.57% |   3.85% |  86.75% |   7.83% |
|     3 |       0 |       0 |       1 |      24 |
|       |   0.00% |   0.00% |   4.00% |  96.00% |
-------------------------------------------------
T: Target    E: Estimated

CPP_OpenCL export¶

The OpenCL export can run the generated program in GPU or CPU architectures. Compilation features:

Preprocessor command [default value]	Description
`PROFILING` [0]	Compile the binary with a synchronization between each layers and return the mean execution time of each layer. This preprocessor option can decrease performances.
`GENERATE_KBIN` [0]	Generate the binary output of the OpenCL kernel .cl file use. The binary is store in the /bin folder.
`LOAD_KBIN` [0]	Indicate to the program to load an OpenCL kernel as a binary from the /bin folder instead of a .cl file.
`CUDA` [0]	Use the CUDA OpenCL SDK locate at \({/usr/local/cuda}\)
`MALI` [0]	Use the MALI OpenCL SDK locate at \({/usr/Mali_OpenCL_SDK_vXXX}\)
`INTEL` [0]	Use the INTEL OpenCL SDK locate at \({/opt/intel/opencl}\)
`AMD` [1]	Use the AMD OpenCL SDK locate at \({/opt/AMDAPPSDK-XXX}\)

Program options related to the OpenCL export:

Option [default value]	Description
`-cpu`	If present, force to use a CPU architecture to run the program
`-gpu`	If present, force to use a GPU architecture to run the program
`-batch` [1]	Size of the batch to use
`-stimulus` [NULL]	Path to a specific input stimulus to test. For example: -stimulus \({/stimulus/env0000.pgm}\) command will test the file env0000.pgm of the stimulus folder.

Test the exported network:

cd export_CPP_OpenCL_float32
make
./bin/n2d2_opencl_test -gpu

CPP_cuDNN export¶

The cuDNN export can run the generated program in NVIDIA GPU architecture. It use CUDA and cuDNN library. Compilation features:

Preprocessor command [default value]	Description
`PROFILING` [0]	Compile the binary with a synchronization between each layers and return the mean execution time of each layer. This preprocessor option can decrease performances.
`ARCH32` [0]	Compile the binary with the 32-bits architecture compatibility.

Program options related to the cuDNN export:

Option [default value]	Description
`-batch` [1]	Size of the batch to use
`-dev` [0]	CUDA Device ID selection
`-stimulus` [NULL]	Path to a specific input stimulus to test. For example: -stimulus \({/stimulus/env0000.pgm}\) command will test the file env0000.pgm of the stimulus folder.

Test the exported network:

cd export_CPP_cuDNN_float32
make
./bin/n2d2_cudnn_test

C_HLS export¶

Test the exported network:

cd export_C_HLS_int8
make
./bin/n2d2_test

Run the High-Level Synthesis (HLS) with Xilinx Vivado HLS:

vivado_hls -f run_hls.tcl

Layer compatibility table¶

Layer compatibility table in function of the export type:

Layer compatibility table	Export type
Layer compatibility table	C	C_HLS	CPP_OpenCL	CPP_TensorRT
Conv	✓	✓	✓	✓
Pool	✓	✓	✓	✓
Fc	✓	✓	✓	✓
Softmax	✓	✗	✓	✓
FMP	✓	✗	✓	✗
Deconv	✗	✗	✗	✓
ElemWise	✗	✗	✗	✓
Resize	✓	✗	✗	✓
Padding	✗	✗	✗	✓
LRN	✗	✗	✗	✓
Anchor	✗	✗	✗	✓
ObjectDet	✗	✗	✗	✓
ROIPooling	✗	✗	✗	✓
RP	✗	✗	✗	✓

BatchNorm is not mentionned because batch normalization parameters are automatically fused with convolutions parameters with the command “-fuse”.