[Numpy-tickets] [NumPy] #425: Faster clip for scalar clipping values
NumPy
numpy-tickets at scipy.net
Sun Jan 21 00:22:09 CST 2007
#425: Faster clip for scalar clipping values
------------------------------+---------------------------------------------
Reporter: david cournapeau | Owner: somebody
Type: enhancement | Status: new
Priority: normal | Milestone:
Component: Other | Version: devel
Severity: normal | Resolution:
Keywords: |
------------------------------+---------------------------------------------
Comment (by david cournapeau):
I don't know much about how f-structure functions are supposed to work:
- my understanding is that they work on C datatype: for NPY_DOUBLE, the
fastclip would be double double fasctlip(double in, double min, double
max) ? Can the function call be inlined ?
- how does it work when you have mixed type ?
Most of the hard work of my patch is done to do the dataype casting, and
being sure that the data fed to function which work at the C datatype
level are the one expected. Other than that, I think once we agree on the
fastclip signature, modifying the patch would be trivial.
Another functionality which seems like it can be speed up that way is
PyArray_PutMask. Those two functions are often used in matplotlib, and for
some plot I am regularly using (specgram and such), those are the two main
culprints (on numpy side), as they are something like between 20 times
slower than a direct C implementation.
What do you mean by "magic function lists" ? I don't know anything about
numeric.
--
Ticket URL: <http://projects.scipy.org/scipy/numpy/ticket/425#comment:2>
NumPy <http://projects.scipy.org/scipy/numpy>
The fundamental package needed for scientific computing with Python.
More information about the Numpy-tickets
mailing list