[X86] Fix callee side of receiving byval args on the stack

[hans]

Summary

See the discussion in https://discourse.llvm.org/t/generic-llvm-ir-windows-x64-argument-passing-issue-in-llvm-11-0-0-and-later/71350

D51842 implemented byval lowering for Win64. D83175 made the call lowering honor the "from now on treat this as a regular pointer" comment also when the argument gets passed on the stack. However, it didn't update the callee side.

[rnk]

lgtm

I checked the code and as I understand it, byval has no effect on the first four pointer arguments, those are passed in the 4 register parameters indirectly. After that, byval works normally as it does for i686 and other platforms. The argument is placed on the stack and it can be materialized.

llvm/lib/Target/X86/X86ISelLowering.cpp

if (VA.isRegLoc()) {
       }
     } else {
       assert(VA.isMemLoc());
       ArgValue =
           LowerMemArgument(Chain, CallConv, Ins, dl, DAG, VA, MFI, InsIndex);
     }
 
     // If value is passed via pointer - do a load.
-    if (VA.getLocInfo() == CCValAssign::Indirect && !Ins[I].Flags.isByVal())
+    if (VA.getLocInfo() == CCValAssign::Indirect &&
+        !(Ins[I].Flags.isByVal() && VA.isRegLoc())) {
       ArgValue =
           DAG.getLoad(VA.getValVT(), dl, Chain, ArgValue, MachinePointerInfo());
+    }
 
     InVals.push_back(ArgValue);
   }
 
   for (unsigned I = 0, E = Ins.size(); I != E; ++I) {
     if (Ins[I].Flags.isSwiftAsync()) {
       auto X86FI = MF.getInfo<X86MachineFunctionInfo>();
       if (Subtarget.is64Bit())

llvm/test/CodeGen/X86/win64-byval.ll

 ; CHECK-NEXT:    callq foo2
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addq $136, %rsp
 ; CHECK-NEXT:    retq
 ; CHECK-NEXT:    .seh_endproc
   call void @foo2(ptr byval({ float, double }) @G, ptr byval({ float, double }) @G, ptr byval({ float, double }) @G, ptr byval({ float, double }) @G, ptr byval({ float, double }) @G, i64 10)
   ret void
 }
+
+define i64 @receive_byval_arg_via_stack_arg(i64* byval(i64), i64* byval(i64), i64* byval(i64), i64* byval(i64), i64* byval(i64) %x) {
+; CHECK-LABEL: receive_byval_arg_via_stack_arg:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; CHECK-NEXT:    movq (%rax), %rax
+; CHECK-NEXT:    retq
+  %r = load i64, i64* %x
+  ret i64 %r
+}