[4.5:r99336] Cortex-A8 optimisation: jump/loop alignment.

Assigning upstream submission to Julian.

I've benchmarked this patch on A8 (imx51) using "-O3 -static -mcpu=cortex-a8" in GCC 4.6 and found *no* measurable speed improvement.

In fact, I've benchmarked every permutation of -falign-functions={4,8,16,32,64} and -falign-jumps={4,8,16,32,64}, and found that none of them give a measurable improvement.

Therefore, I suggest dropping this patch on A8, and possibly on A9 also, if the benchmark results turn out the same.

Benchmarking on A9 shows little better results. There are a few configurations that give a tiny improvement, but they're far from consistent across all the benchmark instances.

I'm going to drop this patch.