[llvm] [GlobalISel] Migrate various generic wip_match_opcode combines to MIR-pattern. (PR #220213)

[Vikash Gupta via llvm-commits]

================
@@ -448,11 +448,25 @@ def bitreverse_lshr : GICombineRule<
 
 // Combine (shl (add x, c1), c2) -> (add (shl x, c2), c1 << c2)
 // Combine (shl (or x, c1), c2) -> (or (shl x, c2), c1 << c2)
-def commute_shift : GICombineRule<
-  (defs root:$d, build_fn_matchinfo:$matchinfo),
-  (match (wip_match_opcode G_SHL):$d,
-         [{ return Helper.matchCommuteShift(*${d}, ${matchinfo}); }]),
-  (apply [{ Helper.applyBuildFn(*${d}, ${matchinfo}); }])>;
+// Apply rebuilds the matched opcode, so it stays inline C++. The builder is
+// used rather than an apply pattern so that (shl c1, c2) is constant folded.
----------------
vg0204 wrote:

As I going through Combine.td, I found it invokes matchConstantFoldBinOp that would do constant folding in next iteration, but it will need pre-requisite patch to support vector outputs, so I start working on that in parallel

https://github.com/llvm/llvm-project/pull/220213