Pass number of registers involved into leaf vector functions. This increases compilation speed of interpreter.cc by about 25% and also makes it easier to see that code doesn't rely on vlmul directly but only or number of registers involved. Test: m berberis_all Change-Id: Ie9d1e57d05bb05ac88f98fae8f2cbf14f5389459
Berberis: dynamic binary translator to run Android apps with riscv64 native code on x86_64 devices or emulators.