datapulverizer/kernelmatrixbenchmark

Benchmarking Kernel Matrix Calculations in D, Julia, and Chapel

kernelmatrixbenchmark's Issues

slow ndslice is weird

With #2 ndslice has the same speed on my machine or it is a bit faster than custom matrix implementation.

Please provide the build command you are using to build a common D benchmark and Mir D benchmark. Looks like something has been missing.

Test --no-ieee-float for Chapel

@dataPulverizer -- thanks very much for a very detailed and thorough discussion of these benchmarks.

Would you be able to try compiling the Chapel code with --no-ieee-float? This should be similar to the -ffast-math flag for ldc (and in fact, throws the same flag for the backend C compiler). I find that it makes a significant difference (~x5) for a number of the benchmarks on my laptop (although not in all cases, notably Log and Power).

Thanks in advance!