[Mlir-commits] [mlir] [MLIR][OpenMP] Normalize lowering of omp.loop_nest (PR #127217)

[Tom Eccles]

================
@@ -2295,6 +2253,61 @@ convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
           .failed())
     return failure();
 
+  llvm::ConstantInt *simdlen = nullptr;
+  if (std::optional<uint64_t> simdlenVar = simdOp.getSimdlen())
+    simdlen = builder.getInt64(simdlenVar.value());
+
+  llvm::ConstantInt *safelen = nullptr;
+  if (std::optional<uint64_t> safelenVar = simdOp.getSafelen())
+    safelen = builder.getInt64(safelenVar.value());
+
+  llvm::MapVector<llvm::Value *, llvm::Value *> alignedVars;
+  llvm::omp::OrderKind order = convertOrderKind(simdOp.getOrder());
+  llvm::BasicBlock *sourceBlock = builder.GetInsertBlock();
+  std::optional<ArrayAttr> alignmentValues = simdOp.getAlignments();
+  mlir::OperandRange operands = simdOp.getAlignedVars();
+  for (size_t i = 0; i < operands.size(); ++i) {
+    llvm::Value *alignment = nullptr;
+    llvm::Value *llvmVal = moduleTranslation.lookupValue(operands[i]);
+    llvm::Type *ty = llvmVal->getType();
+    if (auto intAttr = llvm::dyn_cast<IntegerAttr>((*alignmentValues)[i])) {
----------------
tblah wrote:

The clause is defined as taking an `I64ArrayAttr` so I don't think this cast can fail for valid IR. It might as well use `llvm::cast` so compilation hits a hard error if this assumption is ever broken (instead of merely skipping the clause, which could be hard to spot).

https://github.com/llvm/llvm-project/pull/127217