[llvm] [LSR] Add unequal cost eval for dropping solutions (PR #178039)

[John Brawn via llvm-commits]

john-brawn-arm wrote:

Looking at the tests added in lsr-baseline-solution.ll it looks like this change sometimes makes things worse. e.g. in _Z6add1_iPim we currently generate
```
_Z6add1_iPim:                           # @_Z6add1_iPim
        .cfi_startproc
# %bb.0:                                # %entry
        beqz    a1, .LBB0_3
# %bb.1:                                # %for.body.preheader
        slli    a1, a1, 2
        add     a1, a0, a1
.LBB0_2:                                # %for.body
                                        # =>This Inner Loop Header: Depth=1
        lw      a2, 0(a0)
        addi    a2, a2, 1
        sw      a2, 0(a0)
        addi    a0, a0, 4
        bne     a0, a1, .LBB0_2
.LBB0_3:                                # %for.cond.cleanup
        ret
```
but with this change we generate
```
_Z6add1_iPim:                           # @_Z6add1_iPim
        .cfi_startproc
# %bb.0:                                # %entry
        beqz    a1, .LBB0_3
# %bb.1:                                # %for.body.preheader
        li      a2, 0
.LBB0_2:                                # %for.body
                                        # =>This Inner Loop Header: Depth=1
        slli    a3, a2, 2
        add     a3, a0, a3
        lw      a4, 0(a3)
        addi    a4, a4, 1
        addi    a2, a2, 1
        sw      a4, 0(a3)
        bne     a2, a1, .LBB0_2
.LBB0_3:                                # %for.cond.cleanup
        ret
```


https://github.com/llvm/llvm-project/pull/178039