Difference between CuSparse csrsv* and csrsv2*

I just wonder what’s the difference between the functions csrsv_analysis, csrsv_solve, csrilu0 and csrsv2_analysis, csrsv2_solve, csrilu02.

Unfortunately, I did not find any information in the documentation or on the internet.

I am not sure what you mean by “not … any information”. For example, a cursory look at the CUSPARSE documentation (http://docs.nvidia.com/cuda/pdf/CUSPARSE_Library.pdf) shows that both csrsv_analysis and csrsv_solve are documented. I know nothing about sparse matrix methods, but would assume that one has to use the analysis function before one can use the solve function.

What kind of additional information would you expect to find in the CUSPARSE library documentation?

Have you checked the examples that come with the CUDA installation to see whether there is one that demonstrates the use of some or all of the functions mentioned?

The question is not about the difference between csrsv_analysis/csrsv_solv/csrilu0. That’s of course well documented and explained with examples. It was about the difference between the mentioned group of functions and the other group of functions with a “2” in them: csrsv2_analysis/csrsv2_solv/csrilu02. Reading the documentation and examples does not point out any difference, they seem to do absolutely the same.

This presentation may be of interest:

http://on-demand.gputechconf.com/gtc/2014/presentations/S4188-avoid-global-synchronization-domino-scheme.pdf

for understanding the difference between csrsv and csrsv2

@_Michael: blush. Sorry, I completely misread your question. I promise to work on my reading comprehension …