call Alveo (accelerator card) from MATLAB example

An example of how to execute an operation in Xilinx Alveo accelerator card from MATLAB.

This README.md is under CC BY 4.0 license , although whole this repository is not free to use because the license of the referd repository is unknown.

How to execute

There are some limitations in the MEX execution of MATLAB, so it is not likely to be possible to run acceleration in Alveo as it is.

• The MEX executable does not accept standard input.
  • We need to write the path to the xclbin file directly to the source code or to pass path via mex API.
• MATLAB overwrites the LD_PRELOAD environment variable.
  • In the dynamic linking of the library, an error occurs when executing it because different versions of the library are linked depending on the environment.

Please see

• https://github.com/nixir/callAlveoFromMATLAB/blob/main/run.fish
• https://github.com/nixir/callAlveoFromMATLAB/blob/main/codegen/mex/fir_filter/interface/_coder_fir_filter_api.c

Build

1. Move ./codegen/mex/fir_filter
2. Set your card name at design.cfg
3. Build with ninja (Maybe need some uncomment. Please see build.ninja)
4. Copy mex file and xclbin to top directory

Outputs

kodama@XXXXXXX ~/R/callAlveoFromMATLAB> fish run.fish
MATLAB is selecting SOFTWARE OPENGL rendering.

                 < M A T L A B (R) >
       Copyright 1984-2019 The MathWorks, Inc.
   R2019b Update 1 (9.7.0.1216025) 64-bit (glnxa64)
                  September 26, 2019


To get started, type doc.
For product information, visit www.mathworks.com.

>> FIRfilter
    0.0065    0.4935    0.4935    0.0065

Elapsed time is 0.004253 seconds.
/usr/local/MATLAB/R2019b/sys/opengl/lib/glnxa64:/usr/local/MATLAB/R2019b/sys/os/glnxa64:/usr/local/MATLAB/R2019b/bin/glnxa64:/usr/local/MATLAB/R2019b/extern/lib/glnxa64:/usr/local/MATLAB/R2019b/sys/java/jre/glnxa64/jre/lib/amd64/native_threads:/usr/local/MATLAB/R2019b/sys/java/jre/glnxa64/jre/lib/amd64/server:/opt/xilinx/xrt/lib:/tools/Xilinx/Vitis/2019.2/lib/lnx64.o/Ubuntu/18:/tools/Xilinx/Vitis/2019.2/lib/lnx64.o/Ubuntu:/tools/Xilinx/Vitis/2019.2/lib/lnx64.o:/tools/Xilinx/Vitis/2019.2/runtime/lib/x86_64:/opt/xilinx/xrt/lib:/tools/Xilinx/Vitis/2019.2/lib/lnx64.o/Ubuntu/18:/tools/Xilinx/Vitis/2019.2/lib/lnx64.o/Ubuntu:/tools/Xilinx/Vitis/2019.2/lib/lnx64.o:/tools/Xilinx/Vitis/2019.2/runtime/lib/x86_64:/tools/Xilinx/Vivado/2019.2/lnx64/tools/opencv/opencv_gcc/:/tools/Xilinx/Vivado/2019.2/lnx64/tools/opencv/opencv_gcc/

INFO: Found 2 platforms
INFO: Selected platform 1 from Xilinx
INFO: Found 1 devices
CL_DEVICE_NAME xilinx_u250_xdma_201830_2
Selected xilinx_u250_xdma_201830_2 as the target device
INFO: Make context
INFO: create a computer context
INFO!create a command commands!
INFO: loading xclbin ./fir_filter.xilinx_u250_xdma_201830_2.xclbin
INFO:load kernel from xclbin!
INFO:create compute program from binary
INFO: build program executable!
INFO: create compute kernel_fir_filter!
FPGA execution time is 38671 nano sec
Elapsed time is 0.617696 seconds.

ans =

  logical

   1

>>

Reference

• Vitis-Tutorial https://github.com/Xilinx/Vitis-Tutorials/blob/master/docs/my-first-program/reference-files/src/host.c

  • (Above file added this file https://github.com/nixir/callAlveoFromMATLAB/blob/main/codegen/mex/fir_filter/interface/_coder_fir_filter_api.c)

• And These codes generated by MATLAB Coder (Academic License)