Experience with performance decrease PENTIUM 4 to ARM9 ?

[Peter Meier <peter dot meier at ar-solutions dot de>]

Hello, 

I am researching, if a realtime computer vision application, currently
running on PCs is possible to port to an ARM9 platform, while
fullfilling the realtime constrains. I downloaded the instruction cycle
summary for ARM and have the cycles and bus activity for the basic
assembler operations, like LDR, LDM MUL.

But since I am a c++ programmer, I am having problems to convert a c++
a*b into assembler code, especially how many load instructions I need to
do on average for a MULT, etc. . Are there any tables or figures, that
convert a 32bit integer MULT into cycle times or are there rough
estimations how much slower an ARM-Processor is compared to an Intel
Pentium 4? Any exprerienced cross-compiler-users?

I have allready changed the algorithm to work with a minimal amount of
floating point operations.

 

Any answers, links, papers are very welcome.

Thanks a lot, Peter