[llvm] [RISCV][CostModel] Correct the cost of some reductions (PR #118072)

[Shih-Po Hung via llvm-commits]

================
@@ -1534,6 +1534,13 @@ RISCVTTIImpl::getArithmeticReductionCost(unsigned Opcode, VectorType *Ty,
              getRISCVInstructionCost(Opcodes, LT.second, CostKind) +
              getCmpSelInstrCost(Instruction::ICmp, ElementTy, ElementTy,
                                 CmpInst::ICMP_EQ, CostKind);
+    } else if (ISD == ISD::XOR) {
+      // Example sequences:
+      //   vsetvli a0, zero, e8, mf8, ta, ma
+      //   vcpop.m a0, v0
+      //   andi a0, a0, 1
+      Opcodes = {RISCV::VCPOP_M};
+      return LT.first + getRISCVInstructionCost(Opcodes, LT.second, CostKind);
----------------
arcbbb wrote:

`(LT.first - 1)` first appeared in this [commit](https://github.com/llvm/llvm-project/commit/536095a27c17e2f30765111f57d36c0ee9e211c7).
It seems to be used to calculate the additional cost when the LMUL exceeds 8.
For example,
```
define zeroext i1 @vreduce_and_nxv1i1(<vscale x 128 x i1> %v) {
;    vmand.mm    v8, v0, v8 // addtional cost when LMUL exceeds 8.
;    vmnot.m v8, v8
;    vcpop.m a0, v8
;    seqz    a0, a0
  %red = call i1 @llvm.vector.reduce.and.nxv128i1(<vscale x 128 x i1> %v)
  ret i1 %red
}
```

https://github.com/llvm/llvm-project/pull/118072