Various algorithmic improvements that reduce compile times (thanks Alexandre Pilkiewicz):

- Lattice: preserve sharing in "combine" operation
- Kildall: use splay heaps (lib/Heaps.v) for node sets
- RTLgen: add a "nop" before loops so that natural enumeration of nodes
  coincides with (reverse) postorder
- Maps: add PTree.map1 operation, use it in RTL and LTL.
- Driver: increase minor heap size


git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@1543 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e

1 files changed, 1 insertions, 1 deletions

diff --git a/backend/RTLgen.v b/backend/RTLgen.v
index e918449..a9c8f97 100644
--- a/backend/RTLgen.v
+++ b/backend/RTLgen.v
@@ -591,7 +591,7 @@ Fixpoint transl_stmt (map: mapping) (s: stmt) (nd: node)
       do n1 <- reserve_instr;
       do n2 <- transl_stmt map sbody n1 nexits ngoto nret rret;
       do xx <- update_instr n1 (Inop n2);
-      ret n1
+      add_instr (Inop n2)
   | Sblock sbody =>
       transl_stmt map sbody nd (nd :: nexits) ngoto nret rret
   | Sexit n =>