[Mlir-commits] [llvm] [mlir] [mlir][llvmir][OpenMP] Translate affinity clause in task construct to llvmir (PR #182223)

[Tom Eccles]

================
@@ -2229,6 +2224,78 @@ class TaskContextStructManager {
   /// The type of the structure
   llvm::Type *structTy = nullptr;
 };
+
+/// IteratorInfo extracts and prepares loop bounds information from an
+/// mlir::omp::IteratorOp for lowering to LLVM IR.
+///
+/// It computes the per-dimension trip counts and the total linearized trip
+/// count, casted to i64. These are used to build a canonical loop and to
+/// reconstruct the physical induction variables inside the loop body.
+class IteratorInfo {
+private:
+  llvm::SmallVector<llvm::Value *> lowerBounds;
+  llvm::SmallVector<llvm::Value *> upperBounds;
+  llvm::SmallVector<llvm::Value *> steps;
+  llvm::SmallVector<llvm::Value *> trips;
+  unsigned dims;
+  llvm::Value *totalTrips;
+
+  llvm::Value *lookUpAsI64(mlir::Value val, const LLVM::ModuleTranslation &mt,
+                           llvm::IRBuilderBase &builder) {
+    llvm::Value *v = mt.lookupValue(val);
+    if (!v)
+      return nullptr;
+    if (v->getType()->isIntegerTy(64))
+      return v;
+    if (v->getType()->isIntegerTy())
+      return builder.CreateSExtOrTrunc(v, builder.getInt64Ty());
+    return nullptr;
+  }
+
+public:
+  IteratorInfo(mlir::omp::IteratorOp itersOp,
+               mlir::LLVM::ModuleTranslation &moduleTranslation,
+               llvm::IRBuilderBase &builder) {
+    dims = itersOp.getLoopLowerBounds().size();
+    this->lowerBounds.resize(dims);
----------------
tblah wrote:

ultra-nit: we don't usually use the `this` pointer to access members unless that is required to disambiguate something. However I can see some instances of this under mlir/ (although none in this file) and it doesn't seem to be prohibited in the LLVM Coding Standards so this isn't that important to me.

https://github.com/llvm/llvm-project/pull/182223