The End of Numeric Error: An interview with John L. Gustafson

Re:The End of Numeric Error: An interview with John L. Gusta... by azag0 on Tuesday April 26, 2016 @01:31PM (#11573873)

I started a DFT code in Julia two years ago but didn't get too far. I always felt that I have to think how to make the code fast in Julia (types/vectorization), whereas I just know it will be fast in Fortran. So I always end up writing modern Fortran C-binded to Python. Was it just me not paying enough time to get used to Julia or was/is it a real problem?

[ Parent | on HN ]

Re:The End of Numeric Error: An interview with John L. Gusta... by acidflask on Tuesday April 26, 2016 @04:21PM (#11575293)

Have you actually profiled the Julia code you wrote and ensured that you have followed the performance tips in the manual? If you did it correctly all the time should be spent in linear algebra operations, and so it shouldn't matter what language you implemented the code in since the performance is primarily determined by your underlying BLAS library.

[ Parent | on HN ]

Re:The End of Numeric Error: An interview with John L. Gusta... by azag0 on Wednesday April 27, 2016 @07:40AM (#11579467)

It's not that I couldn't make it fast. I followed just the steps you mentioned and got to a really fast code. But the process didn't feel intuitive.

[ Parent | on HN ]

Re:The End of Numeric Error: An interview with John L. Gusta... by dnautics on Tuesday April 26, 2016 @01:48PM (#11574034)

probably the best way to do it would have been to bind the fortran functions to julia.

[ Parent | on HN ]