SIL SideEffects: handle program termination point functions

We can ignore any memory writes in a program termination point, because it's not relevant for the caller.
But we need to consider memory reads, otherwise preceeding memory writes would be eliminated by dead-store-elimination in the caller.
E.g. String initialization for error strings which are printed by the program termination point.
Regarding ownership: a program termination point must not touch any reference counted objects.

Author: eeckstein

SwiftCompilerSources/Sources/SIL/Effects.swift

@@ -128,6 +128,14 @@ extension Function {
     default:
       break
     }
+    if isProgramTerminationPoint {
+      // We can ignore any memory writes in a program termination point, because it's not relevant
+      // for the caller. But we need to consider memory reads, otherwise preceeding memory writes
+      // would be eliminated by dead-store-elimination in the caller. E.g. String initialization
+      // for error strings which are printed by the program termination point.
+      // Regarding ownership: a program termination point must not touch any reference counted objects.
+      return SideEffects.GlobalEffects(memory: SideEffects.Memory(read: true))
+    }
     var result = SideEffects.GlobalEffects.worstEffects
     switch effectAttribute {
     case .none:

SwiftCompilerSources/Sources/SIL/Function.swift

@@ -121,6 +121,10 @@ final public class Function : CustomStringConvertible, HasShortDescription, Hash
     }
   }
 
+  /// True if the callee function is annotated with @_semantics("programtermination_point").
+  /// This means that the function terminates the program.
+  public var isProgramTerminationPoint: Bool { hasSemanticsAttribute("programtermination_point") }
+
   /// Kinds of effect attributes which can be defined for a Swift function.
   public enum EffectAttribute {
     /// No effect attribute is specified.