[llvm] [SLP] Avoid seeding related affine loop address computations (PR #226220)

[Alexey Bataev via llvm-commits]

================
@@ -36032,6 +36035,13 @@ bool SLPVectorizerPass::vectorizeOnceUsedSeeds(BasicBlock *BB, BoUpSLP &R) {
         !isOnceUsedSeed(&I) || isNonVectorizableInst(&I, TLI) ||
         R.hasResolvedUser(&I))
       continue;
+    // Index chains of collected GEPs are handled by vectorizeGEPIndices.
+    if (isGEPCandidateIndexBundle({&I}, *SE, *LI, SLPReVec, [&](auto *GEP) {
----------------
alexey-bataev wrote:

```suggestion
    if (!GEPs.empty() && isGEPCandidateIndexBundle({&I}, *SE, *LI, SLPReVec, [&](auto *GEP) {
```

https://github.com/llvm/llvm-project/pull/226220