Yuechao Pan

Try to make sure the graph is allocated on CPU and has a CSR representation? Compiling Gunrock with -g will enable CPU side debugging, and will tell which line segfaults...

Thanks to @yzhwang for handling this. The working partitioners should be in the partitioner folder; the ones in the app folder are the ones for v0.5, and I don't think...

Louvain has a serial CPU reference, and the result is checked against the output of the CPU reference. https://github.com/gunrock/gunrock/blob/master/examples/louvain/test_louvain.cu#L76 So, if the result with race conditions still matches the CPU...

Yep. The iterative updates can mitigate the unwanted effects of data races. PR is another algorithm that has this property, and the Gunrock implementation uses rank_curr & rank_next to avoid...

There seems to be some mismatch between the parameter names in the original command line flags and what's currently supported in Gunrock's PR, I will use the current names. A...

@achalagarwal Do you know which kernel hit the illegal memory access issue? Form the printout, it seems to be mgpu, which I guess would be one of the 2 advances...

Thanks for uploading the code for large graph and the procedure of running it. It compiles and runs on our K40. I tried to give different sizes, and it seems...

Consider this PR a WIP to for finding the cause(s) of convergence difference between at scale and GA. The two potential places are 1) whether to load optimizer slots when...

The MLPerf reference BERT model is in fact slightly modified from the Google Research model. As far as I remembered, the math should be the same; input dataset is different....