[CIR][ABI][AArch64][Lowering] Support calls for struct types > 128 bits (#1074)

As the title says, this PR adds support for calls with struct types >
128 bits, building upon this
[PR](#1068).

The idea is gotten from the original Codegen, and I have added a couple
of tests.

Author: bruteforceboy

clang/include/clang/CIR/MissingFeatures.h

@@ -404,6 +404,8 @@ struct MissingFeatures {
   static bool itaniumRecordLayoutBuilderFinishLayout() { return false; }
 
   static bool mustProgress() { return false; }
+
+  static bool skipTempCopy() { return false; }
 };
 
 } // namespace cir

clang/lib/CIR/Dialect/Transforms/TargetLowering/LowerFunction.cpp

@@ -704,6 +704,14 @@ LowerFunction::buildFunctionEpilog(const LowerFunctionInfo &FI) {
         if (auto al = findAlloca(ret)) {
           rewriter.replaceAllUsesWith(al.getResult(), RVAddr);
           rewriter.eraseOp(al);
+          rewriter.setInsertionPoint(ret);
+
+          auto retInputs = ret.getInput();
+          assert(retInputs.size() == 1 && "return should only have one input");
+          if (auto load = mlir::dyn_cast<LoadOp>(retInputs[0].getDefiningOp()))
+            if (load.getResult().use_empty())
+              rewriter.eraseOp(load);
+
           rewriter.replaceOpWithNewOp<ReturnOp>(ret);
         }
       });
@@ -952,6 +960,15 @@ mlir::Value LowerFunction::rewriteCallOp(FuncType calleeTy, FuncOp origCallee,
   return CallResult;
 }
 
+mlir::Value createAlloca(mlir::Location loc, mlir::Type type,
+                         LowerFunction &CGF) {
+  auto align = CGF.LM.getDataLayout().getABITypeAlign(type);
+  auto alignAttr = CGF.getRewriter().getI64IntegerAttr(align.value());
+  return CGF.getRewriter().create<AllocaOp>(
+      loc, CGF.getRewriter().getType<PointerType>(type), type,
+      /*name=*/llvm::StringRef(""), alignAttr);
+}
+
 // NOTE(cir): This method has partial parity to CodeGenFunction's EmitCall
 // method in CGCall.cpp. When incrementing it, use the original codegen as a
 // reference: add ABI-specific stuff and skip codegen stuff.
@@ -984,10 +1001,12 @@ mlir::Value LowerFunction::rewriteCallOp(const LowerFunctionInfo &CallInfo,
   CIRToCIRArgMapping IRFunctionArgs(LM.getContext(), CallInfo);
   llvm::SmallVector<mlir::Value, 16> IRCallArgs(IRFunctionArgs.totalIRArgs());
 
+  mlir::Value sRetPtr;
   // If the call returns a temporary with struct return, create a temporary
   // alloca to hold the result, unless one is given to us.
   if (RetAI.isIndirect() || RetAI.isCoerceAndExpand() || RetAI.isInAlloca()) {
-    cir_cconv_unreachable("NYI");
+    sRetPtr = createAlloca(loc, RetTy, *this);
+    IRCallArgs[IRFunctionArgs.getSRetArgNo()] = sRetPtr;
   }
 
   cir_cconv_assert(!cir::MissingFeatures::swift());
@@ -1082,6 +1101,32 @@ mlir::Value LowerFunction::rewriteCallOp(const LowerFunctionInfo &CallInfo,
 
       break;
     }
+    case ABIArgInfo::Indirect:
+    case ABIArgInfo::IndirectAliased: {
+      assert(NumIRArgs == 1);
+      // TODO(cir): For aggregate types
+      // We want to avoid creating an unnecessary temporary+copy here;
+      // however, we need one in three cases:
+      // 1. If the argument is not byval, and we are required to copy the
+      // 2. If the argument is byval, RV is not sufficiently aligned, and
+      //    source.  (This case doesn't occur on any common architecture.)
+      //    we cannot force it to be sufficiently aligned.
+      // 3. If the argument is byval, but RV is not located in default
+      //    or alloca address space.
+      cir_cconv_assert(!::cir::MissingFeatures::skipTempCopy());
+
+      mlir::Value alloca = findAlloca(I->getDefiningOp());
+
+      // since they are a ARM-specific feature.
+      if (::cir::MissingFeatures::undef())
+        cir_cconv_unreachable("NYI");
+
+      IRCallArgs[FirstIRArg] = alloca;
+
+      // NOTE(cir): Skipping Emissions, lifetime markers.
+
+      break;
+    }
     default:
       llvm::outs() << "Missing ABIArgInfo::Kind: " << ArgInfo.getKind() << "\n";
       cir_cconv_unreachable("NYI");
@@ -1217,6 +1262,10 @@ mlir::Value LowerFunction::rewriteCallOp(const LowerFunctionInfo &CallInfo,
       // done in CIRGen
       return RetVal;
     }
+    case ABIArgInfo::Indirect: {
+      auto load = rewriter.create<LoadOp>(loc, sRetPtr);
+      return load.getResult();
+    }
     default:
       llvm::errs() << "Unhandled ABIArgInfo kind: " << RetAI.getKind() << "\n";
       cir_cconv_unreachable("NYI");

clang/test/CIR/CallConvLowering/AArch64/aarch64-cc-structs.c

@@ -153,6 +153,43 @@ void pass_eq_128(EQ_128 s) {}
 // LLVM:   %[[#V2:]] = load ptr, ptr %[[#V1]], align 8
 void pass_gt_128(GT_128 s) {}
 
+// CHECK: cir.func @get_gt_128(%arg0: !cir.ptr<!ty_GT_128_> {{.*}}, %arg1: !cir.ptr<!ty_GT_128_>
+// CHECK: %[[#V0:]] = cir.alloca !cir.ptr<!ty_GT_128_>, !cir.ptr<!cir.ptr<!ty_GT_128_>>, [""] {alignment = 8 : i64}
+// CHECK: cir.store %arg1, %[[#V0]] : !cir.ptr<!ty_GT_128_>, !cir.ptr<!cir.ptr<!ty_GT_128_>>
+// CHECK: %[[#V1:]] = cir.load %[[#V0]] : !cir.ptr<!cir.ptr<!ty_GT_128_>>, !cir.ptr<!ty_GT_128_>
+// CHECK: cir.copy %[[#V1]] to %arg0 : !cir.ptr<!ty_GT_128_>
+// CHECK: cir.return
+
+// LLVM: void @get_gt_128(ptr %[[#V0:]], ptr %[[#V1:]])
+// LLVM: %[[#V3:]] = alloca ptr, i64 1, align 8
+// LLVM: store ptr %[[#V1]], ptr %[[#V3]], align 8
+// LLVM: %[[#V4:]] = load ptr, ptr %[[#V3]], align 8
+// LLVM: call void @llvm.memcpy.p0.p0.i32(ptr %[[#V0]], ptr %[[#V4]], i32 24, i1 false)
+// LLVM: ret void
+GT_128 get_gt_128(GT_128 s) {
+  return s;
+}
+
+// CHECK: cir.func no_proto @call_and_get_gt_128(%arg0: !cir.ptr<!ty_GT_128_>
+// CHECK: %[[#V0:]] = cir.alloca !ty_GT_128_, !cir.ptr<!ty_GT_128_>, {{.*}} {alignment = 8 : i64}
+// CHECK: %[[#V1:]] = cir.alloca !ty_GT_128_, !cir.ptr<!ty_GT_128_>, {{.*}} {alignment = 8 : i64}
+// CHECK: cir.call @get_gt_128(%[[#V1]], %arg0) : (!cir.ptr<!ty_GT_128_>, !cir.ptr<!ty_GT_128_>) -> ()
+// CHECK: %[[#V2:]] = cir.load %[[#V1]] : !cir.ptr<!ty_GT_128_>, !ty_GT_128_
+// CHECK: cir.store %[[#V2]], %[[#V0]] : !ty_GT_128_, !cir.ptr<!ty_GT_128_>
+// CHECK: cir.return
+
+// LLVM: void @call_and_get_gt_128(ptr %[[#V0:]])
+// LLVM: %[[#V2:]] = alloca %struct.GT_128, i64 1, align 8
+// LLVM: %[[#V3:]] = alloca %struct.GT_128, i64 1, align 8
+// LLVM: call void @get_gt_128(ptr %[[#V3]], ptr %[[#V0]])
+// LLVM: %[[#V4:]] = load %struct.GT_128, ptr %[[#V3]], align 8
+// LLVM: store %struct.GT_128 %[[#V4]], ptr %[[#V2]], align 8
+// LLVM: ret void
+GT_128 call_and_get_gt_128() {
+  GT_128 s;
+  s = get_gt_128(s);
+  return s;
+}
 // CHECK: cir.func @passS(%arg0: !cir.array<!u64i x 2> 
 // CHECK:   %[[#V0:]] = cir.alloca !ty_S, !cir.ptr<!ty_S>, [""] {alignment = 4 : i64}
 // CHECK:   %[[#V1:]] = cir.alloca !cir.array<!u64i x 2>, !cir.ptr<!cir.array<!u64i x 2>>, ["tmp"] {alignment = 8 : i64}