[llvm] [RISCV] Add vector hasAndNot to enable optimizations (PR #132438)

[Craig Topper via llvm-commits]

================
@@ -2580,9 +2580,9 @@ define <vscale x 1 x i8> @not_signbit_mask_nxv1i8(<vscale x 1 x i8> %a, <vscale
 ; CHECK-ZVKB-LABEL: not_signbit_mask_nxv1i8:
 ; CHECK-ZVKB:       # %bb.0:
 ; CHECK-ZVKB-NEXT:    vsetvli a0, zero, e8, mf8, ta, ma
-; CHECK-ZVKB-NEXT:    vmsgt.vi v0, v8, -1
-; CHECK-ZVKB-NEXT:    vmv.v.i v8, 0
-; CHECK-ZVKB-NEXT:    vmerge.vvm v8, v8, v9, v0
+; CHECK-ZVKB-NEXT:    vsra.vi v8, v8, 7
+; CHECK-ZVKB-NEXT:    vnot.v v8, v8
+; CHECK-ZVKB-NEXT:    vand.vv v8, v8, v9
----------------
topperc wrote:

The not immediate ended up being 255 instead of -1 like the pattern expected. The upper bits of splat_vector are ignored so either is correct.

This patch fixes it, but I'm still investigating other way to fix it that don't double the isel patterns.

```
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td b/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td
index fcbb2dbc76a3..775c161b56e8 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td
@@ -603,13 +603,15 @@ multiclass VPatUnarySDNode_V<SDPatternOperator op, string instruction_name,
 // This should match the logic in RISCVDAGToDAGISel::selectVSplat
 def riscv_splat_vector : PatFrag<(ops node:$rs1),
                                  (riscv_vmv_v_x_vl undef, node:$rs1, srcvalue)>;
-def riscv_vnot : PatFrag<(ops node:$rs1), (xor node:$rs1,
-                                               (riscv_splat_vector -1))>;
+class riscv_vnot<int sew>
+    : PatFrags<(ops node:$rs1),
+               [(xor node:$rs1, (riscv_splat_vector -1)),
+                (xor node:$rs1, (riscv_splat_vector !srl(-1, !sub(64, sew))))]>;

 foreach vti = AllIntegerVectors in {
   let Predicates = !listconcat([HasStdExtZvkb],
                                GetVTypePredicates<vti>.Predicates) in {
-    def : Pat<(vti.Vector (and (riscv_vnot vti.RegClass:$rs1),
+    def : Pat<(vti.Vector (and (riscv_vnot<vti.SEW> vti.RegClass:$rs1),
                                vti.RegClass:$rs2)),
               (!cast<Instruction>("PseudoVANDN_VV_"#vti.LMul.MX)
                  (vti.Vector (IMPLICIT_DEF)),
```

https://github.com/llvm/llvm-project/pull/132438