TF/keras/pytorch native build on Summit

To activate env:

source activate pyTFenv


To deactivate:

source datctivate

Create python virutal environment for ML

unloaded existing python module - as I had already python 3.7 by now)
look for all python modules

load python/3.7.0-anaconda3-5.3.0

$ conda -V

Update conda

conda update conda

conda create -n MLenv python=3.7 anaconda

Now you are asked for update step.

Then activate the environment:

source activate MLenv

Now we have our python 3.7 environment
We can install all the frameworks we need here.

To deactivate the environment,

source deactivate

.
.
.
.
.
.
.
.

6. Delete a no longer needed virtual environment

1. To delete a conda environment, enter the following, where yourenvname is the name of the environment you wish to delete.

conda remove -n yourenvname -all




I figured out that TensorFlow 1.12.0 only works with Python version 3.5.2. I had Python 3.7 but that didn't work. So, I had to downgrade Python and then I could install TensorFlow to make it work.

To downgrade your python version from 3.7 to 3.6

conda install python=3.6.8

Environment location:

/ccs/home/uma/.conda/envs/pyTFptEnv

Reviving Aria 10 work on Oswald03

xxk@oswald03:~/gamesskernel/openarcOS/test/examples/openarc/matmul/bin$ env CL_CONTEXT_E
MULATOR_DEVICE_INTELFPGA=1 ./matmul_ACC
CPU Elapsed time = 0.392515 sec
Accelerator Elapsed time = 22.417378 sec
Verification Successful err = 0.000000e+00



On Oswald >
matmul example
change the makefile

 16:52:09 udk@oswald03:~/gamesskernel/openarcOS/test/examples/openarc/matmul/bin$ ls
__all_sources.cl   matmul_ACC         openarc_kernel_EmulatorDeviceEmulatedDevice.aocx
binBuilder_opencl  openarc_kernel.cl  Timer
16:53:03 udk@oswald03:~/gamesskernel/openarcOS/test/examples/openarc/matmul/bin$ cd ..
16:53:06 udk@oswald03:~/gamesskernel/openarcOS/test/examples/openarc/matmul$ cd cetus_output/
16:53:31 udk@oswald03:~/gamesskernel/openarcOS/test/examples/openarc/matmul/cetus_output$ ls
matmul.cpp  openarc_kernel  openarc_kernel.aoco  openarc_kernel.aocx  openarc_kernel.cl
16:53:32 udk@oswald03:~/gamesskernel/openarcOS/test/examples/openarc/matmul/cetus_output$ aoc -v openarc_kernel.cl -o ../bin/openarc_kernel_p510t_sch_ax115PCIe510n.aocx --board p510t_sch_ax115
Warning: Please use -board= instead of --board
aoc: Environment checks are completed successfully.
aoc: If necessary for the compile, your BAK files will be cached here: /var/tmp/aocl/
You are now compiling the full flow!!
aoc: Selected target board p510t_sch_ax115
aoc: Running OpenCL parser....
aoc: OpenCL parser completed successfully.
aoc: Optimizing and doing static analysis of code...
aoc: Linking with IP library ...
Checking if memory usage is larger than 100%
aoc: First stage compilation completed successfully.
Compiling for FPGA. This process may take a long time, please be patient.

unzip bz2

tar xvfj filename

Creating OpenACC C++ libraries

OpenACC code targeted at an Nvidia GPU must be compiled with the PGI compiler using at least the following options:

pgcc source_code.c -ta=nvidia -acc

scc1% pgcc -o mycode -acc -Minfo mycode.c

In the above example, -acc turns on the OpenACC feature while -Minfo returns additional information on the compilation.

pgi/15.10.0 supports pgCC for c++
Newer pgi/18 doesn't recognize pgCC on Titan

A good read

http://www.drdobbs.com/parallel/creating-and-using-libraries-with-openac/240012502?pgno=2

https://kristerw.blogspot.com/2017/04/building-gcc-with-support-for-nvidia.html

acc benchmark code
https://github.com/EPCCed/epcc-openacc-benchmarks/blob/master/