ARM32: Combine LSR into ADD for Div by 2^n

Combining LSR into ADD reduces a number of used registers from three to
two and a needed number of instructions.

This CL implements this optimization.

Test: 411-checker-hdiv-hrem-pow2
Test: test.py --host --optimizing --jit --gtest --interpreter
Test: test.py --32 --target --optimizing --jit --interpreter
Test: run-gtests.sh
Change-Id: I230d2623e745884fe4278a860740829ee63750c6
3 files changed