I've been testing target mips on the tcg-search-2 branch -- I haven't
spotted any obvious regressions and performance gain was visible in some
of my tests. On qemu-system-mips64el the Linux kernel build time went
down from 465m to 305m. And in a virtual memory torture test I can see
slightly above
Version 2, updated based on comments. Notable changes:
(1) Move breakpoint recognition after insn_start. Now we really
do never have zero insns per TB. Assertions added.
(2) Comments added.
(3) Minor tweeks to the encoding, reflected in the comments.
I examined a handfull of test c
