To test the speed of performing different base16 to binary and binary to base16 conversions on various platforms. Converting base16 to binary and back is very important to certain applications, doing this as fast as possible for the platform in question (software / hardware combo) may not be a trivial problem.
make
./test FILE
Where FILE is the input to operate on. Up to 32 MiB of FILE will be read in and
operated on by each different test. For a quick test, use /dev/urandom for
FILE.
Using a lookup table for converting binary to base16 results in a 10x speedup
measured on my Intel Core-i7 system and a 14x speedup measured on my BeagleBone
when compared to using sprintf(). This is very significant!
Using a lookup table for converting base16 to binary results in a 3x speedup
measured on my Intel Core-i7 system but only a slightly less than 2x speedup
measured on my Beaglebone when compared to doing a memcpy() followed by
strtol(). This isn't very significant but is still a worthwhile improvement.
It's interesting to note that the speedups between my Intel Core-i7 and BeagleBone rely on system configuration, not just processing power difference. If they were simply linear with processing power difference, I'd expect to see similar gains for any algorithm changes which is not what's observed (14x vs 10x followed by 1.8x vs 3x). I wonder if a Cortex-A8 with a 32 bit wide memory bus would fair better than the BeagleBone's 16 bit wide bus.
After more investigation, it seems isxdigit() is quite expensive. By removing
calls to isxdigit() the speedup becomes more dramatic but at the possible
expense of encountering error conditions. Without isxdigit() the base16 to
binary LUT measures a 6x improvement on BeagleBone and a 5x improvement on my
Intel Core-i7.