Skip to content

Latest commit

 

History

History
37 lines (21 loc) · 949 Bytes

notes.md

File metadata and controls

37 lines (21 loc) · 949 Bytes
Raw

Profiling GPU computing example

CPU mode

...

Most time will be spent in numpy.core.multiarray.dot

GPU mode

$ python -m cProfile -o gpu.log <>
[]: import pstats
[]: p = pstats.Stats('gpu.log')
[]: p.sort_stats('tottime').print_stats(10)

Most time will be spent in Theano's compiled function.

The matrix dot will be called around 1.2M times.

Tips

Numpy object alignment issue

If Theano complains the numpy object is not aligned, you can use <obj>.flags to check the alignment (discussion).

I personally meet this issue when using scipy.io.loadmat(), the numpy array objects are not aligned.

If Nvidia GPU only supports float32

If you are using Theano to manipulate float64 data, but the Nvidia GPU only supports float32, you can indicate the parameter allow_input_downcast=True in theano.function() when building the expression.