CSE for LEA-like operations

Currently binaryen can't do CSE for such cases:

i32.load(p + ((i + 0) << 2)
i32.load(p + ((i + 1) << 2)
i32.load(p + ((i + 2) << 2)

Which usually appear when we have array's access like:

p[i + 0]
p[i + 1]
p[i + 2]

I'm wondering how better to tech binaryen handle such cases and optimize it to this?

let pi = p + (i << 2)
i32.load(pi + (0 << 2))
i32.load(pi + (1 << 2))
i32.load(pi + (2 << 2))

Probably the best approach is always distribute (x op1 C1) op2 C2 to ((x op2 C2) op1 (C1 op2 C2)) in optimization instructions? WDYT?

No file, test, or entry point is named; begin by locating Binaryen's existing CSE and optimization-instruction handling and compare it with the LEA-like load examples. Done means agreeing on the transformation's correctness and implementing it with coverage for the shown indexed loads.