hannospreeuw / kernel-tuning-for-sagecal Goto Github PK

The current performance benchmarks for kernel_coherencies assume there are only point sources in the sky map. We need a realistic dataset to test with in order to measure the performance of kernel_coherencies on real data. In particular a data set with with a representative distribution of different models of objects in the sky, including points, gaussians, disks, and rings. Only then, we can see if more work needs to be done to avoid the possible divergence that could lead to performance issues with real data.

I just experimented with using the math function sincosf versus the math intrinsic __sincosf, the overal performance of kernel_array_beam improves by a factor of 3 when using the intrinsic __sincosf over the function sincosf (tested only with the use_kernel=0 use_shared_mem=1 kernel configuration). The trade-off is of course precision, it depends on the application whether or not this is a problem.

The CUDA programming guide states the following about the precision of the __sincosf intrinsic:
For x in [-π,π], the maximum absolute error is 2^(-21.19), and larger otherwise,
see here.

I guess we need to test with real data and then judge depending on the results whether or not the error is problematic for the application.

@HannoSpreeuw

./tune_kernel_coherencies.py
geeft vaak
........
pycuda._driver.MemoryError: cuModuleLoadData failed: out of memory -
PyCUDA WARNING: a clean-up operation failed (dead context maybe?)

We need to find what exactly the problem is here. What version of Kernel Tuner are you using at the moment?