Closed fhs closed 9 years ago
Maintaining this amount of benchmark code by hand will be a lot work. I still think everything should be autogenerated, including the stubs if you end up factoring out some of the code in a package like testblas.
You have less than 300 lines of code that generates lots of benchmarks. You can clean it up a little and maybe make it shorter by using the reflect package to get the method parameter names and types.
Sorry, I didn't mean to suggest it shouldn't be automated. I was suggesting the actual benchmark code live in testblas, and then just have stubs in goblas and cblas like is there for the other benchmarks.
This doesn't merge anymore. Closing.
The original isn't great code. I think the better fix is to have the benchmark code itself live in testblas, and then have the BenchmarkXxx stubs live in goblas and cblas.