[llvm] [RISCV] Add a pass to remove ADDI by reassociating to fold into load/store address. (PR #127151)

[Craig Topper via llvm-commits]

topperc wrote:

> Is there an overlap between `RISCVMergeBaseOffset` and `RISCVFoldMemOffset`? Based on my rough understanding, they are doing similar things, right?

There's currently no overlap. RISCVMergeBaseOffset does 2 different optimizations both involving LUI/AUIPIC+ADDI representing global variables.

The first optimization is to look for constant additions that can be folded into the offset of LUI/AUIPC+ADDI. This changes the LUI/AUIPC+ADDI not the load/store.

The second optimization tries to fold the ADDI from the LUI/AUIPIC+ADDI into the load/store address. This only handles the case of the load/store being the direct user. This is similar to this new pass but it only handles the simple direct user case and it works on ADDIs containing symbol references not immediates. More complex cases like the new pass handles can't be done for symbol references that can be GP relaxed. The GP relaxation optimization requires the LUI/AUIPC to be directly used by the instruction that has the `lo` or `pcrel_lo` relocation. There can't be any arithmetic between them.

https://github.com/llvm/llvm-project/pull/127151