yuhc / gpu-rodinia Goto Github PK

Thank you for your kind help in advance.

I'm trying to move on to Rodinia 3.1 from 3.0.
Is there any building dependencies on version 3.1?

1. Does is support cuda 4.2?
2. Is there any specific gcc version that I need to use?

There are some info written regarding pervious version(2.3) but I can't find any updates for 3.1 version.

Virginia university link provided in the README for data is not working.
Can the link be updated? I cant seem to find the data anywhere.

Hello, please let me know which version of Rodinia is supported by CUDA 4.0 & GPGPUSim 3.2.2.
Better can you give a version to the version compatibility list between CUDA/GPGPUSim & Rodinia?
Thanks!

I want to run rodinia on sm_60 and cuda-10.0. The default input size for backprop, 64k, is small for real hardware. When I run 1M elements, I get this error:

$ ./backprop 1048576
Random number generator seed: 7
Input layer size : 1048576
Starting training kernel
Performing GPU computation
bpnn kernel error: invalid configuration argument

Is there any way to fix that or the program is unable to handle that size?

With gcc-4.8.5 I get this error while building graphgen

graphgen.cpp:31:36: error: ‘linear_congruential’ was not declared in this scope
 #define LINEAR_CONGRUENTIAL_ENGINE linear_congruential

In fact no where linear_congruential is defined based on my searches.

Some of the applications such as BACKPROP, PATHFINDER, and BFS does not have a verification option to verify the computation results's correctness done on GPU.

I would like to know whether you will add the verification options as well or not.

Thanks,
Burak Topçu

How can I compile and run kmeans with visual studio 2015?
I have linking errors!
please help me

HI dear author,
It's an honor to write a letter here, I wonder what's wrong with the cuda benchmarks when I type $ make

thank you
best regards
William

What is the .domn file ?

How to read it with meshlab, meshio ?

I can't load the results and the mesh in paraview ?

Thanks you.

I am a tester and am currently using this tool to test the performance of the CPU and GPU. May I ask if the server does not have a separate graphics card, how does this tool run? If it's convenient, could you give me your WeChat account

I need to enlarge the inputs. Not all applications are staright forward. For example, in bfs, I can create my own graph input. However, for b+ I don't know how to pick up the necessary parameters to enlarge. Any guide line?

Hey, is there a certain reason why e.g. the kmeans_openmp benchmark is compiled and linked with -g and -O2, when best performance should be expected without -g and with -O3?
Makefile

Dear Auther,
I want to use the benchmark of CUDA.But after I enter the website--http://lava.cs.virginia.edu/Rodinia/download_links.htm--and click the rodinia_3.1.tar.bz2(the other part will appear the same problem).The fig that can be seen will display.

Thus,where can I go to find the dateset?
Best wishes to you!
Thank you!

It seems there is a bug in rodinia

https://github.com/yuhc/gpu-rodinia/blob/master/opencl/lavaMD/kernel/kernel_gpu_opencl.cl#L120

It uses variables in local address space in the non-outermost scope of a kernel function, which violates OpenCL spec v1.2 s6.5.2 Variables declared in the __local address space inside a kernel function must occur at kernel function scope. https://www.khronos.org/registry/OpenCL/specs/opencl-1.2.pdf

Can App Developer check and resolve this issue. Due to this, test is failing on OpenCL devices.