My plan was to have (like NumPy) a subset that is optimized for for matrix (two-dimensional) operations.<br><br><div class="gmail_quote">On Fri, Aug 14, 2009 at 4:51 AM, Jakub Piotr Cłapa <span dir="ltr"><<a href="mailto:jpc-ml@zenburn.net">jpc-ml@zenburn.net</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">Regarding efficient numeric computation the CorePy project [1] is quite interesting. For a start Noel Welsh has some nice sample code for generating x86 instruction streams dynamicaly [2]. Maybe such a specialised thing would be better than stuffing to much into the mzscheme JIT.<br>
<br>
As far as array indexing is concerned the NumPy matrix manipulation is quite nice but I find the multi-dimensional arrays to be difficult to grasp. OTOH it is just about next-column increment and next-row stride so you cannot use it to for example vectorize image algorithms operating on multiple image tiles (like 8x8 tiles of JPEG or image segmentation in DjVu). Specialized JITing of iterations would allow for doing more fancy algorithms without multiple-dimensions.<br>
<br>
[1]: <a href="http://www.corepy.org/" target="_blank">http://www.corepy.org/</a><br>
[2]: <a href="http://github.com/noelwelsh/assembler/tree/master" target="_blank">http://github.com/noelwelsh/assembler/tree/master</a><br>
<br>
-- <br>
regards,<br><font color="#888888">
Jakub Piotr Cłapa</font><div><div></div><div class="h5"><br>
_________________________________________________<br>
For list-related administrative tasks:<br>
<a href="http://list.cs.brown.edu/mailman/listinfo/plt-scheme" target="_blank">http://list.cs.brown.edu/mailman/listinfo/plt-scheme</a><br>
</div></div></blockquote></div><br>