[llvm] improve debug messages in delinearization and dependence analysis (PR #156339)

[Sebastian Pop via llvm-commits]

================
@@ -354,18 +354,27 @@ void llvm::computeAccessFunctions(ScalarEvolution &SE, const SCEV *Expr,
     if (!AR->isAffine())
       return;
 
+  // Clear output vector.
+  Subscripts.clear();
----------------
sebpop wrote:

The interface has not changed, I can move this one in a separate patch if needed.
I tried to avoid any refactoring to minimize my changes.

One of the clean-ups is to return a struct with {Sizes, Subscripts, Delinearized_p} instead of passing those to the functions.

Also we should only have one delinearize() interface instead of 4 or 5 different ways to call delinearize.
The code looks pretty ugly and difficult to modify: every pass using delinearization LCA, DA, and polly has its own way and reason to call the internals of delinearization.

https://github.com/llvm/llvm-project/pull/156339