simonb@netbsd.org said:
> The binary was compiled with -mtune=strongarm to use the 32x32->64 bit
> multiply.


That won't cause it to use 32x32->64 multiply.  For that you need to 
change the architecture that you are compiling for.  Hence:

	-march=armv4	-- Use full armv4 instruction set
			   (Not compatible with RISC PC)

	-march=armv3m	-- Use architecture 3 plus enhanced multiply
			   instructions (Compatible with RISC PC)

Having done that, you can then tune the instruction ordering for optimal 
StrongARM performance by giving -mtune=strongarm.

So, -march tells the compiler which instructions it is legal to chose 
from, -mtune=foo tells the compiler that it needs to select from that set 
for best performance on a cpu of type foo.  By doing it this way we get 
much finer control over the code generated.

Richard.