[PATCH] D14614: [WinEH] Find root frame correctly in CLR funclets

[Joseph Tremoulet via llvm-commits]

JosephTremoulet added inline comments.

================
Comment at: lib/Target/X86/X86FrameLowering.cpp:1206-1213
@@ +1205,10 @@
+      unsigned PSPSlotOffset = getPSPSlotOffsetFromSP(MF);
+      addRegOffset(
+          BuildMI(MBB, MBBI, DL, TII.get(X86::MOV64rm)).addReg(Establisher),
+          Establisher, false, PSPSlotOffset);
+      // Save the root establisher back into the current funclet's (mostly
+      // empty) frame, in case a sub-funclet or the GC needs it.
+      addRegOffset(BuildMI(MBB, MBBI, DL, TII.get(X86::MOV64mr)), StackPtr,
+                   false, PSPSlotOffset)
+          .addReg(Establisher);
+    }
----------------
Do I need to do anything additional here to capture appropriate dataflow constraints?  The MachineInstrs this generates dump as:


```
	MOV64rm %RCX, %RCX, 1, %noreg, 40, %noreg
	MOV64mr %RSP, 1, %noreg, 40, %noreg, %RCX
```

and I don't know if I need to be marking rsp as a kill, doing something to get it to show up as a LHS, etc.


================
Comment at: lib/Target/X86/X86FrameLowering.cpp:1273-1275
@@ +1272,5 @@
+    unsigned PSPSlotOffset = getPSPSlotOffsetFromSP(MF);
+    addRegOffset(BuildMI(MBB, MBBI, DL, TII.get(X86::MOV64mr)), StackPtr, false,
+                 PSPSlotOffset)
+        .addReg(StackPtr);
+  }
----------------
Similarly, this begets


```
	MOV64mr %RSP, 1, %noreg, 40, %noreg, %RSP
```

and I'm wondering if the store needs to be marked volatile like C++'s UnwindHelp store is, etc.


http://reviews.llvm.org/D14614