[CST-2] Comp.arch 1997 p7 q3

[Martin Harper]

Andrei Legostaev wrote:

> http://www.cl.cam.ac.uk/tripos/y1997p7q3.pdf
> 
> Last paragraph: "What special instruction set features or compiler
> optimisations  might be used  to reduce the overhead of sub-word accesses".
> 
> There is only a reference to these "features" in the slides, no details.
> Presumably there is more in the Additional Material handout.  Could somebody
> please point me to the right page or just tell me the answer?
> 
> :-)
> 
> Thanks in advance!

Compiler optimisations: padding to remove them. This might be 
cheating... ;-)

The following: loop i, doing { a[i] = b[i] XOR c[i] }, where a, b, c are 
(possibly unaligned) arrays of unpadded bytes, and i is large; can be 
optimised by rewriting it to do word loads, stores, XORs, and dealing 
with the stuff at the start and the end of the loop specially {but 
probably still by doing word-sized reads and writes, depending on the ISA}

Obviously there are similar rewrites that can be used on other pieces of 
code.

Special instruction features I'm not sure of. Does this perhaps refer to 
Load Word Right(LWR) and the like?

Martin