[LSV] Merge contiguous chains across scalar types (llvm#154069)

This change enables the LoadStoreVectorizer to merge and vectorize
contiguous chains even when their scalar element types differ, as long
as the total bitwidth matches. To do so, we rebase offsets between
chains, normalize value types to a common integer type, and insert the
necessary casts around loads and stores. This uncovers more
vectorization opportunities and explains the expected codegen updates
across AMDGPU tests.

Key changes:
- Chain merging
  - Build contiguous subchains and then merge adjacent ones when:
- They refer to the same underlying pointer object and address space.
    - They are either all loads or all stores.
    - A constant leader-to-leader delta exists.
- Rebasing one chain into the other's coordinate space does not overlap.
    - All elements have equal total bit width.
- Rebase the second chain by the computed delta and append it to the
first.

- Type normalization and casting
- Normalize merged chains to a common integer type sized to the total
bits.
- For loads: create a new load of the normalized type, copy metadata,
and cast back to the original type for uses if needed.
  - For stores: bitcast the value to the normalized type and store that.
- Insert zext/trunc for integer size changes; use bit-or-pointer casts
when sizes match.

- Cleanups
  - Erase replaced instructions and DCE pointer operands when safe.
- New helpers: computeLeaderDelta, chainsOverlapAfterRebase,
rebaseChain, normalizeChainToType, and allElemsMatchTotalBits.

Impact:
- Increases vectorization opportunities across mixed-typed but
size-compatible access chains.
- Large set of expected AMDGPU codegen diffs due to more/changed
vectorization.

This PR resolves llvm#97715.

Author: gandhi56

llvm/include/llvm/Transforms/Utils/Local.h

@@ -431,7 +431,7 @@ LLVM_ABI void combineAAMetadata(Instruction *K, const Instruction *J);
 
 /// Copy the metadata from the source instruction to the destination (the
 /// replacement for the source instruction).
-LLVM_ABI void copyMetadataForLoad(LoadInst &Dest, const LoadInst &Source);
+LLVM_ABI void copyMetadataForAccess(Instruction &Dest, Instruction &Source);
 
 /// Patch the replacement so that it is not more restrictive than the value
 /// being replaced. It assumes that the replacement does not get moved from

llvm/lib/Target/AMDGPU/AMDGPULowerBufferFatPointers.cpp

@@ -1035,7 +1035,7 @@ bool LegalizeBufferContentTypesVisitor::visitLoadImpl(
       LoadInst *NewLI = IRB.CreateAlignedLoad(
           LoadableType, NewPtr, commonAlignment(OrigLI.getAlign(), ByteOffset),
           Name + ".off." + Twine(ByteOffset));
-      copyMetadataForLoad(*NewLI, OrigLI);
+      copyMetadataForAccess(*NewLI, OrigLI);
       NewLI->setAAMetadata(
           AANodes.adjustForAccess(ByteOffset, LoadableType, DL));
       NewLI->setAtomic(OrigLI.getOrdering(), OrigLI.getSyncScopeID());

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

@@ -415,7 +415,7 @@ void PointerReplacer::replace(Instruction *I) {
                               LT->getAlign(), LT->getOrdering(),
                               LT->getSyncScopeID());
     NewI->takeName(LT);
-    copyMetadataForLoad(*NewI, *LT);
+    copyMetadataForAccess(*NewI, *LT);
 
     IC.InsertNewInstWith(NewI, LT->getIterator());
     IC.replaceInstUsesWith(*LT, NewI);
@@ -606,7 +606,7 @@ LoadInst *InstCombinerImpl::combineLoadToNewType(LoadInst &LI, Type *NewTy,
       Builder.CreateAlignedLoad(NewTy, LI.getPointerOperand(), LI.getAlign(),
                                 LI.isVolatile(), LI.getName() + Suffix);
   NewLoad->setAtomic(LI.getOrdering(), LI.getSyncScopeID());
-  copyMetadataForLoad(*NewLoad, LI);
+  copyMetadataForAccess(*NewLoad, LI);
   return NewLoad;
 }
 

llvm/lib/Transforms/Scalar/SROA.cpp

@@ -3280,7 +3280,7 @@ class AllocaSliceRewriter : public InstVisitor<AllocaSliceRewriter, bool> {
       // Copy any metadata that is valid for the new load. This may require
       // conversion to a different kind of metadata, e.g. !nonnull might change
       // to !range or vice versa.
-      copyMetadataForLoad(*NewLI, LI);
+      copyMetadataForAccess(*NewLI, LI);
 
       // Do this after copyMetadataForLoad() to preserve the TBAA shift.
       if (AATags)

llvm/lib/Transforms/Utils/Local.cpp

@@ -3100,54 +3100,70 @@ void llvm::combineAAMetadata(Instruction *K, const Instruction *J) {
   combineMetadata(K, J, /*DoesKMove=*/true, /*AAOnly=*/true);
 }
 
-void llvm::copyMetadataForLoad(LoadInst &Dest, const LoadInst &Source) {
+void llvm::copyMetadataForAccess(Instruction &DestI, Instruction &SourceI) {
   SmallVector<std::pair<unsigned, MDNode *>, 8> MD;
-  Source.getAllMetadata(MD);
-  MDBuilder MDB(Dest.getContext());
-  Type *NewType = Dest.getType();
-  const DataLayout &DL = Source.getDataLayout();
+  SourceI.getAllMetadata(MD);
+  MDBuilder MDB(DestI.getContext());
+  Type *NewType = DestI.getType();
+
+  // Only needed for range metadata on loads.
+  const DataLayout *DL = nullptr;
+  const LoadInst *LSource = dyn_cast<LoadInst>(&SourceI);
+  if (LSource)
+    DL = &LSource->getDataLayout();
+
   for (const auto &MDPair : MD) {
     unsigned ID = MDPair.first;
     MDNode *N = MDPair.second;
-    // Note, essentially every kind of metadata should be preserved here! This
-    // routine is supposed to clone a load instruction changing *only its type*.
-    // The only metadata it makes sense to drop is metadata which is invalidated
-    // when the pointer type changes. This should essentially never be the case
-    // in LLVM, but we explicitly switch over only known metadata to be
-    // conservatively correct. If you are adding metadata to LLVM which pertains
-    // to loads, you almost certainly want to add it here.
+
     switch (ID) {
+    // Applies to both loads and stores as-is.
     case LLVMContext::MD_dbg:
-    case LLVMContext::MD_tbaa:
     case LLVMContext::MD_prof:
-    case LLVMContext::MD_fpmath:
     case LLVMContext::MD_tbaa_struct:
-    case LLVMContext::MD_invariant_load:
     case LLVMContext::MD_alias_scope:
     case LLVMContext::MD_noalias:
     case LLVMContext::MD_nontemporal:
-    case LLVMContext::MD_mem_parallel_loop_access:
     case LLVMContext::MD_access_group:
     case LLVMContext::MD_noundef:
     case LLVMContext::MD_noalias_addrspace:
-      // All of these directly apply.
-      Dest.setMetadata(ID, N);
+    case LLVMContext::MD_mem_parallel_loop_access:
+      DestI.setMetadata(ID, N);
+      break;
+
+    // Load-only metadata.
+    case LLVMContext::MD_fpmath:
+    case LLVMContext::MD_invariant_load:
+      if (isa<LoadInst>(DestI))
+        DestI.setMetadata(ID, N);
       break;
 
     case LLVMContext::MD_nonnull:
-      copyNonnullMetadata(Source, N, Dest);
+      if (auto *LDest = dyn_cast<LoadInst>(&DestI)) {
+        if (LSource)
+          copyNonnullMetadata(*LSource, N, *LDest);
+      }
       break;
 
     case LLVMContext::MD_align:
     case LLVMContext::MD_dereferenceable:
     case LLVMContext::MD_dereferenceable_or_null:
-      // These only directly apply if the new type is also a pointer.
+      // Applies to both loads and stores only if the new type is also a
+      // pointer.
       if (NewType->isPointerTy())
-        Dest.setMetadata(ID, N);
+        DestI.setMetadata(ID, N);
       break;
 
     case LLVMContext::MD_range:
-      copyRangeMetadata(DL, Source, N, Dest);
+      if (auto *LDest = dyn_cast<LoadInst>(&DestI)) {
+        if (LSource && DL)
+          copyRangeMetadata(*DL, *LSource, N, *LDest);
+      }
+      break;
+
+    case LLVMContext::MD_tbaa:
+      if (isa<LoadInst>(DestI))
+        DestI.setMetadata(ID, N);
       break;
     }
   }