Am 2015-08-12 05:27, schrieb Henry Gomersall:
I'm struggling to realise your improvements. Simply installing from master with `pip install -e .` (or similar) is actually substantially slower than the pypi installation.
Are you talking about the installation process itself or the execution time? Next, what precisely are you measuring? What OpenCL implementation are you using? What array size are you using?
For what it's worth, the version that is now in git is an entirely different animal than the version that is currently on the package index: one is based on boost Python (the one on the package index, the old one) and the other is based on cffi. In my measurements, I have not noticed much of a speed differential between the two versions, but I am interested in learning about different findings.
I noticed that the pypi installation (`pip install pyopencl`) uses boost, the config for which seems to have been stripped out of the master version. Am I missing something regarding the use of libffi in building from source that yields the slowdown I've noticed?
While cffi has the option of using ctypes and thereby libffi, PyOpenCL uses cffi in its ahead-of-time mode and therefore never incurs the penalty of touching libffi.
(it's something like half the speed now, so not insignificant!).
It would be valuable if you could share a profile, such as the file gathered using
python -m cProfile -o out clmath_test.py Thanks for investigating! Andreas _______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
