[flang-commits] [flang] [mlir] [flang][mlir] Add support for implicit linearization in omp.simd (PR #150386)

[via flang-commits]

================
@@ -3025,9 +3025,36 @@ genStandaloneSimd(lower::AbstractConverter &converter, lower::SymMap &symTable,
   simdArgs.priv.vars = simdClauseOps.privateVars;
   simdArgs.reduction.syms = simdReductionSyms;
   simdArgs.reduction.vars = simdClauseOps.reductionVars;
+
+  std::vector<mlir::Attribute> typeAttrs;
+  // If attributes from explicit `linear(...)` clause are present,
+  // carry them forward.
+  if (simdClauseOps.linearVarTypes && !simdClauseOps.linearVarTypes.empty())
+    typeAttrs.assign(simdClauseOps.linearVarTypes.begin(),
+                     simdClauseOps.linearVarTypes.end());
+
+  for (auto [loopVar, loopStep] : llvm::zip(iv, loopNestClauseOps.loopSteps)) {
+    // TODO: Implicit linearization is skipped if iv is a pointer
+    // or an allocatable, due to potential mismatch between the linear
+    // variable type (example !fir.ref<!fir.box<!fir.heap<i32>>>)
+    // and the linear step size (example: i64). Handle this type mismatch
+    // gracefully.
+    if (loopVar->test(Fortran::semantics::Symbol::Flag::OmpLinear) &&
+        !(Fortran::semantics::IsAllocatableOrPointer(*loopVar) ||
+          Fortran::semantics::IsAllocatableOrPointer(loopVar->GetUltimate()))) {
+      const mlir::Value variable = converter.getSymbolAddress(*loopVar);
+      mlir::Type ty = converter.genType(*loopVar);
+      typeAttrs.push_back(mlir::TypeAttr::get(ty));
----------------
NimishMishra wrote:

Okay, that's a good point. I will test on this a bit more.

https://github.com/llvm/llvm-project/pull/150386