Kyle Jones <kyle_jones(a)wonderworks.com&gt; writes: Because the current code is basically an unrolled memory compare loop. I think the compiler can do better on a real memcmp[1] and in addition the function becomes simple enough to inline. I checked that with my changes, compare_runes is indeed inlined. [When gcc uses its builtin memcmp it uses a very tight loop] I think I was hoping since that the fetches are basically linear through memory (the inner loop in redisplay output is comparing two arrays) the prefetching would help. Maybe we can force the compiler to inline compare_runes some other way. (on a machine where memcmp is a function call the memcmp trick doesn't work. It just replaces one function call with another). Jan Footnotes: [1] i.e. I am betting the compiler use a builtin memcmp. However looking at the sparc output it seems that gcc doesn't always do that.