[llvm] [RISCV] Add combine for shadd family of instructions. (PR #130829)

[Min-Yih Hsu via llvm-commits]

https://github.com/mshockwave commented:

@stefanp-synopsys I left some comments w.r.t the profitability of this transformation. Also, can you clarify your motivation again? The example you gave in the patch description:
```
define i32 @foo(i32 %x, i32 %y) {
  %1 = shl i32 %x, 2
  %2 = add i32 %1, 24
  %3 = add i32 %y, %2
  ret i32 %3
}
```
Already yielded the code you desired:
```
        sh2add  a0, a0, a1
        addi    a0, a0, 24
```
even _without_ this patch -- I believed it has been done through DAGCombiner's existing reassociate optimizations.

Your patch hinted that you want to use shXadd even when the ADDI is used more than one time (which was not covered by the reassociate optimization I mentioned). But function `t8` in your test case doesn't seem to agree with it.

https://github.com/llvm/llvm-project/pull/130829