Performance tuning for the XO

[Mitch Bradley]

C. Scott Ananian wrote:
> We make currently make heavy use of hashing in our upgrade &
> verification infrastructure.  I'd like to find the fastest possible
> implementation of the SHA-256 and/or RIPEMD-160 algorithms.  Can
> anyone offer me advice on the proper compilation options and
> strategies for maximum performance on the processor in the XO?  I
> don't see the -mxo option to gcc yet.
>  --scott
>   

Some timing for libtomcrypt with various compilation options.  All times 
are for hashing 1MiB of data from memory, timed under Open Firmware with 
interrupts disabled.

Hash
	-DLTC_SMALL_CODE, -Os
	large code, -Os
	large code, -O2
sha256
	249 mS
	202 mS
	189 mS
sha512
	373 mS
	347 mS
	358 mS
rmd160
	111 mS
	112 mS
	112 mS


-DLTC_SMALL_CODE causes rolled (as opposed to unrolled) loops in 
sha256.c and sha512.c, but is not used in rmd160.c

The code size for the bios_hash binary is about 20K with rolled loops, 
about 44K with unrolled loops.

So, with this code, the slow version does sha256 at about 4 MiB/sec, the 
fast version does sha256 at about 5.3 MiB/sec, and rm160 goes at 9 
MiB/sec in all cases.