[llvm] [SLPVectorizer][AMDGPU] Look-through vectorization of AMDGPU intrinsics (PR #175821)

[Matt Arsenault via llvm-commits]

================
@@ -223,6 +223,13 @@ static cl::opt<unsigned> LoopAwareTripCount(
     cl::desc("Loop trip count, considered by the cost model during "
              "modeling (0=loops are ignored and considered flat code)"));
 
+/// Enables look-through vectorization for non-vectorizable operations.
+static cl::opt<bool>
+    LookThroughOperations("slp-look-through-intrinsics", cl::init(true),
+                          cl::Hidden,
+                          cl::desc("When true, look through non-vectorizable "
+                                   "operations to vectorize their operands"));
----------------
arsenm wrote:

No reason to add this option 

https://github.com/llvm/llvm-project/pull/175821