[llvm] [Codegen][LegalizeIntegerTypes] Improve shift through stack (PR #96151)

[via llvm-commits]

================
@@ -723,98 +723,117 @@ define void @lshr_16bytes(ptr %src.ptr, ptr %byteOff.ptr, ptr %dst) nounwind {
 ;
 ; RV32I-LABEL: lshr_16bytes:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    addi sp, sp, -48
-; RV32I-NEXT:    sw s0, 44(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s1, 40(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s2, 36(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    lbu a3, 0(a0)
-; RV32I-NEXT:    lbu a4, 1(a0)
+; RV32I-NEXT:    addi sp, sp, -32
+; RV32I-NEXT:    lbu a3, 1(a0)
----------------
futog wrote:

the same as for aarch64, plus it seems that the load/store alignment used in the test, which is 1, does not mix well with the legalized code.

https://github.com/llvm/llvm-project/pull/96151