Learn C, the execution-speed:writing-speed ratio is very good, only very occasionally will you need to resort to raw ASM. I'd skip Python completely if you want speed (plus C is a useful thing to know).

ASM is a pita to use basically, and I only resort to it if I really have to (not very often thankfully ). To learn it, I scoured the web, mainly looking at existing ARM ASM code (the ARM Linux site has some useful info too: http://www.arm.linux.org.uk/developer/) and also eventually bought a book (ARM Assembly Language - an introduction by J.R. Gibson), however I've not done much ASM since then, and therefore haven't really read the book, so can't actually recommend it in any way other than saying that it's the one that looked best to me at the time