[flang-commits] [flang] [flang] Add struct passing target rewrite hooks and partial X86-64 impl (PR #74829)

[via flang-commits]

================
@@ -58,6 +59,63 @@ static void typeTodo(const llvm::fltSemantics *sem, mlir::Location loc,
   }
 }
 
+/// Return the size of alignment of FIR types.
+/// TODO: consider moving this to a DataLayoutTypeInterface implementation
+/// for FIR types. It should first be ensured that it is OK to open the gate of
+/// target dependent type size inquiries in lowering. It would also not be
+/// straightforward given the need for a kind map that would need to be
+/// converted in terms of mlir::DataLayoutEntryKey.
+static std::pair<std::uint64_t, unsigned short>
+getSizeAndAlignment(mlir::Location loc, mlir::Type ty,
+                    const mlir::DataLayout &dl,
+                    const fir::KindMapping &kindMap) {
+  if (mlir::isa<mlir::IntegerType, mlir::FloatType, mlir::ComplexType>(ty)) {
+    llvm::TypeSize size = dl.getTypeSize(ty);
+    unsigned short alignment = dl.getTypeABIAlignment(ty);
+    return {size, alignment};
+  }
+  if (auto firCmplx = mlir::dyn_cast<fir::ComplexType>(ty)) {
+    auto [floatSize, floatAlign] =
+        getSizeAndAlignment(loc, firCmplx.getEleType(kindMap), dl, kindMap);
+    return {llvm::alignTo(floatSize, floatAlign) + floatSize, floatAlign};
+  }
+  if (auto real = mlir::dyn_cast<fir::RealType>(ty))
+    return getSizeAndAlignment(loc, real.getFloatType(kindMap), dl, kindMap);
+
+  if (auto seqTy = mlir::dyn_cast<fir::SequenceType>(ty)) {
+    auto [eleSize, eleAlign] =
+        getSizeAndAlignment(loc, seqTy.getEleTy(), dl, kindMap);
+
+    std::uint64_t size =
+        llvm::alignTo(eleSize, eleAlign) * seqTy.getConstantArraySize();
+    return {size, eleAlign};
+  }
+  if (auto recTy = mlir::dyn_cast<fir::RecordType>(ty)) {
+    std::uint64_t size = 0;
+    unsigned short align = 8;
----------------
jeanPerier wrote:

I changed it to zero. Setting it to a minimum of 8 is specific to X86_64 stack passing. So I moved the setting of byval to a minimum of 8 there.

I started at eight because of the note that the note 14 page 23 in the [System V ABI](https://gitlab.com/x86-psABIs/x86-64-ABI/-/jobs/artifacts/master/raw/x86-64-ABI/abi.pdf?job=build). "14 Therefore the stack will always be eightbyte aligned."

This is also done in clang here:
https://github.com/llvm/llvm-project/blob/18959c46e3ced1f7ad12a82e9f30bafe9d1f1733/clang/lib/CodeGen/Targets/X86.cpp#L2200C5-L2200C5

https://github.com/llvm/llvm-project/pull/74829