Regarding ANUGA GPU ​Version​

[Girishchandra-Yendargaye]

Hello Sir,

This is regarding Easily availability to a GPU capable version ( Beta version is currently working).

Can you please tell us where can we get this version.

Can you please tell us Which CUDA version and GPU CARD does it supports.

Thank you,

Girishchandra Y.

[stoiver]

This is an old project (a student project from back in 2014). Here is my fork https://github.com/stoiver/anuga-cuda which implemented a GPU version of the anuga evolve loop together with one or two of the simple Boundary conditions. Here is a short paper describing the work https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6969525 I would love to get back to this project and roll the methods into the production version of the code. Any GPU coders out there interested in helping? An idea would be to completely isolate the GPU code to work only with triangles in the interior of the mesh (and so avoid needing to have GPU code for BCs etc). Cheers Steve ============================== Stephen Roberts Undergraduate Convenor Mathematical Sciences Institute Room 4.74 Hanna Neumann Building #145 The Australian National University Canberra, ACT 2600 AUSTRALIA Ph: +61 2 61254445 CRICOS: 00120C

…

________________________________ From: Girishchandra-Yendargaye <[email protected]> Sent: Tuesday, 30 October 2018 5:01 PM To: budiaji/anuga-cuda Cc: Subscribed Subject: [budiaji/anuga-cuda] Regarding ANUGA GPU Version (#3) Hello Sir, This is regarding Easily availability to a GPU capable version ( Beta version is currently working). Can you please tell us where can we get this version. Can you please tell us Which CUDA version and GPU CARD does it supports. Thank you, Girishchandra Y. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<#3>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AFTlJGYEsu4r_3QsQpjVzTczb6KLy3cSks5up-sigaJpZM4YBPG9>.

[schoeller]

@stoiver : I don't know whether this is of any help, but I tried using Offloading through OpenMP on Ubuntu 19.04. I have not managed to profile yet, but it does not thrown any errors at least.

I did

1# Install packages
sudo apt-get install gcc-offload-nvptx

2# Enable compiler args
extra_args = ['-fopenmp', '-foffload=nvptx-none', '-fno-stack-protector', '-fopt-info-optimized-omp']

3# Edit pragma parts for example
#pragma omp parallel for private(i) to #pragma omp target teams distribute parallel for
I presume that there is a lot of improvement here. I am not competent in this matter.

4# Build and install using setup.py

5# Enabling profiling in Python
import cProfile

6# Log profile
A) Log-file: nvprof --log-file gpu.log python -m cProfile -s time runDamBreak.py > cpu.log
B) Graphical profiler: nvprof -o gpu.prof python -m cProfile runDamBreak.py

I presume one would have to study how effective this truly is.

[samcom12]

Following

[HarshaUgaveI]

Hello sir,
This is regarding installation of https://github.com/stoiver/anuga-cuda .
Facing boost issue while installing anuga_cuda. All the depenedencies and ANUGA is installed along with boost
Please provide installation guild and solution for the same.

Thank You.