Fast constant multiplies on an x86

I've recently done some work in the area of constant multiplies for
the Pentium (the mul and imul instructions are 9/11 clocks, so there
is quite some room for improvment.) I have a method for producing
optimal code sequences for constant multiplies (similar in
functionality to gnu's super optimizer I suppose.)


Anyhow, a friend of mine who used to work at Weitek says that Intel
used to have a document that they distributed which listed scores of
simulated constant multiplies. I am curious as to its quality, for
comparison purposes. Basically I want to track down this
documentation if it exists. If its only available to commercial
compiler vendors, then I might be interested in opening up a
discussion with said vendor. I'd be interested in pointers either
way.


For a precursor to my work, optimized only for number of instructions,
you can see:


        http://www.geocities.com/SiliconValley/9498/amult.html


--
Paul Hsieh
http://www.geocities.com/SiliconValley/9498/mailme.html
--

Return to the comp.compilers page.
Search the comp.compilers archives again.