[flang-commits] [flang] [flang] Inline scalar-to-scalar TRANSFER for same-size trivial types (PR #191589)

[via flang-commits]

================
@@ -8680,6 +8680,36 @@ IntrinsicLibrary::genTransfer(mlir::Type resultType,
 
   assert(args.size() >= 2); // args.size() == 2 when size argument is omitted.
 
+  bool absentSize = (args.size() == 2);
+
+  // Inline scalar-to-scalar transfers when the result is a trivial type
+  // (integer, real, etc.) and both source and result have the same storage
+  // size.
+  if (absentSize && fir::isa_trivial(resultType)) {
+    mlir::Value sourceBase = fir::getBase(args[0]);
+    mlir::Type sourceType = fir::unwrapRefType(sourceBase.getType());
+    if (fir::isa_ref_type(sourceBase.getType()) &&
+        !mlir::isa<fir::SequenceType>(sourceType)) {
+      auto sourceSizeAndAlign = fir::getTypeSizeAndAlignment(
+          loc, sourceType, builder.getDataLayout(), builder.getKindMap());
+      auto resultSizeAndAlign = fir::getTypeSizeAndAlignment(
+          loc, resultType, builder.getDataLayout(), builder.getKindMap());
+      if (sourceSizeAndAlign && resultSizeAndAlign &&
+          sourceSizeAndAlign->first == resultSizeAndAlign->first) {
+        if (mlir::isa<mlir::IntegerType, mlir::FloatType>(sourceType) &&
+            mlir::isa<mlir::IntegerType, mlir::FloatType>(resultType)) {
+          mlir::Value val = fir::LoadOp::create(builder, loc, sourceBase);
+          if (sourceType != resultType)
+            val = mlir::arith::BitcastOp::create(builder, loc, resultType, val);
+          return val;
+        }
+        mlir::Type refTy = builder.getRefType(resultType);
+        mlir::Value cast = builder.createConvert(loc, refTy, sourceBase);
----------------
jeanPerier wrote:

TBAA in Fortran is not related to the storage type, the "types" are based on the variable names and category (dummies, pointers...).
TBAA will be added on the loads much later before codegeneration and it "goes through" pointer type conversion to identify the Fortran variable being loaded.
So nothing is needed here.

https://github.com/llvm/llvm-project/pull/191589