Matrix Exponential function - benchmark about matrix

[c2akula]

I am skeptical of the concurrent implementation because the performance gain over the serial implementation is very small and that too in only a few cases. That means, the concurrent implementation is either incorrect, or I am looking at the results the wrong way. I'll need some guidance here.

[c2akula]

exponential matrix consuming too much memory, ran out of 2 GB of memory within 30seconds. Input matrix size 1000x1000

[btracey]

Is this an issue or just a comment? If just a comment, in the future it would have been better to include it in the original issue asking to add Exp, or to open a thread on gonum-dev

Did you test with GOMAXPROCS set?

[c2akula]

I haven't yet, I will do it. It's still a work in progress, I am working on optimizing the algorithm. It uses too much memory due to too many allocations. I have tried an implementation where the client allocates the inputs and the function just uses the inputs. I was able to bring down memory allocation fro 69MB to 6.5MB. for this specific function. I'll post the new benchmark once I have completed the new optimized version of the algorithm.