[llvm] [VectorCombine] Fold bitcast(bitreverse.v8i8(bitcast(IntTy))) into bswap+bitreverse (PR #209037)

[Simon Pilgrim via llvm-commits]

================
@@ -6112,12 +6112,56 @@ bool VectorCombine::foldBitcastOfVPLoad(Instruction &I) {
   return true;
 }
 
-/// Fold the following cases into a single byte-level bit-reverse operation
-/// and accepts bswap and bitreverse intrinsics:
-///   bswap(bitreverse(x)) --> bitcast(bitreverse(bitcast(x)))
-///   bitreverse(bswap(x)) --> bitcast(bitreverse(bitcast(x)))
+// Fold bitcast(bitreverse(<N x i8>)(bitcast(IntTy X))) --> bitreverse(bswap(X))
+// and the forward direction:
+//   bswap(bitreverse(x)) --> bitcast(bitreverse(bitcast(x)))
+//   bitreverse(bswap(x)) --> bitcast(bitreverse(bitcast(x)))
+// The direction chosen depends on which is cheaper for the target.
----------------
RKSimon wrote:

please rewrite this to better build on the existing comment - I was kind of expecting you to just replace `-->` with `<-->` and add something about costs

https://github.com/llvm/llvm-project/pull/209037