[Builtin][ARM] Implement addsf3/__aeabi_fadd for Thumb1

Summary:
This patch implements addsf3/__aeabi_fadd in asm for Thumb1.
Compared with generic C version (lib/fp_add_impl.inc), it
1. all constants are materialized instead of loading from constant pool
2. no stack spills (C version uses 136 bytes stack space)
3. clz() is called only when necessary. (C version always calls it)

Reviewers: compnerd, rengolin, asl

Reviewed By: asl

Subscribers: efriedma, aemerson, mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D29485

llvm-svn: 294172