Turn off copy optimization when returning a composite by reference.

The copy optimization is not correct in case of overlap between destination
and source.  We would need to use an hypothetical __builtin_memmove_aligned
that can cope with overlap to implement the copy at return of callee.

4 files changed, 50 insertions, 5 deletions

diff --git a/cparser/StructReturn.ml b/cparser/StructReturn.ml
index 660f1d9b..5e5602f3 100644
--- a/cparser/StructReturn.ml
+++ b/cparser/StructReturn.ml
@@ -299,10 +299,15 @@ let rec transf_expr env ctx e =
       transf_call env ctx None fn args e.etyp
 
 (* Function calls returning a composite by reference: add first argument.
-    ctx = Effects:   lv = f(...)     -> f(&lv, ...)      [copy optimization]
+    ctx = Effects:   lv = f(...)     -> f(&newtemp, ...), lv = newtemp
                      f(...)          -> f(&newtemp, ...)
     ctx = Val:       lv = f(...)     -> f(&newtemp, ...), lv = newtemp
                      f(...)          -> f(&newtemp, ...), newtemp
+
+   We used to do a copy optimization:
+     ctx = Effects:   lv = f(...)     -> f(&lv, ...)
+   but it is not correct in case of overlap (see test/regression/struct12.c) 
+
    Function calls returning a composite by value:
     ctx = Effects:   lv = f(...)     -> newtemp = f(...), lv = newtemp
                      f(...)          -> f(...)
@@ -332,13 +337,14 @@ and transf_call env ctx opt_lhs fn args ty =
           | Effects, None ->
               let tmp = new_temp ~name:"_res" ty in
               {edesc = ECall(fn', eaddrof tmp :: args'); etyp = TVoid []}
-          | Effects, Some lhs ->
-              {edesc = ECall(fn', eaddrof lhs :: args'); etyp = TVoid []}
+          (* Copy optimization, turned off as explained above *)
+       (* | Effects, Some lhs ->
+              {edesc = ECall(fn', eaddrof lhs :: args'); etyp = TVoid []} *)
           | Val, None ->
               let tmp = new_temp ~name:"_res" ty in
               ecomma {edesc = ECall(fn', eaddrof tmp :: args'); etyp = TVoid []}
                      tmp
-          | Val, Some lhs ->
+          | _, Some lhs ->
               let tmp = new_temp ~name:"_res" ty in
               ecomma {edesc = ECall(fn', eaddrof tmp :: args'); etyp = TVoid []}
                      (eassign lhs tmp)
diff --git a/test/regression/Makefile b/test/regression/Makefile
index da7d5755..e2d94aa9 100644
--- a/test/regression/Makefile
+++ b/test/regression/Makefile
@@ -15,7 +15,7 @@ LIBS=$(LIBMATH)
 TESTS=int32 int64 floats floats-basics \
   expr1 expr6 funptr2 initializers initializers2 initializers3 \
   volatile1 volatile2 volatile3 \
-  funct3 expr5 struct7 struct8 struct11 casts1 casts2 char1 \
+  funct3 expr5 struct7 struct8 struct11 struct12 casts1 casts2 char1 \
   sizeof1 sizeof2 binops bool for1 switch switch2 compound \
   decl1 interop1 bitfields9 ptrs3
 
diff --git a/test/regression/Results/struct12 b/test/regression/Results/struct12
new file mode 100644
index 00000000..e69de29b
--- /dev/null
+++ b/test/regression/Results/struct12
diff --git a/test/regression/struct12.c b/test/regression/struct12.c
new file mode 100644
index 00000000..39e62b28
--- /dev/null
+++ b/test/regression/struct12.c
@@ -0,0 +1,39 @@
+/*  This is was originally a regression test for bug 43784 of gcc.
+    See ISO/IEC 9899:TC3 §6.8.6.4p4 and footnote 139. */
+
+#include <stdio.h>
+
+struct s {
+  unsigned char a[256];
+};
+union u {
+  struct { struct s b; int c; } d;
+  struct { int c; struct s b; } e;
+};
+
+static union u v;
+static struct s *p = &v.d.b;
+static struct s *q = &v.e.b;
+
+static struct s __attribute__((noinline)) rp(void)
+{
+  return *p;
+}
+
+static void qp(void)
+{
+  *q = rp();
+}
+
+int main()
+{
+  int i;
+  for (i = 0; i < 256; i++)
+    p->a[i] = i;
+  qp();
+  for (i = 0; i < 256; i++)
+    if (q->a[i] != i)
+      printf("ERROR at %d: %d\n", i, q->a[i]);
+  return 0;
+}
+