[llvm] [LAA] Add stencil group merging to reduce runtime pointer checks (PR #187252)

[Igor Kirillov via llvm-commits]

================
@@ -1310,6 +1367,11 @@ void RuntimePointerChecking::mergeStencilGroups(PredicatedScalarEvolution &PSE,
     const SCEV *BaseLow = Pointers[Member0].Start;
     const SCEV *BaseHigh = Pointers[Member0].End;
 
+    // Keep stencil decomposition and stride-limit arithmetic within 64 bits.
+    // All offsets relative to BaseLow have the same index width.
+    if (SE->getTypeSizeInBits(BaseLow->getType()) > 64)
----------------
igogo-x86 wrote:

Now there is - `stencil_wide_index_small_coeff`

https://github.com/llvm/llvm-project/pull/187252