Hi!

As discussed in the PR, the assumption that in !optimize functions
all SSA_NAMEs for a particular PARM_DECL or RESULT_DECL can be coalesced
together is wrong for -flto, if you compile with optimizations the object
file and then link with -O0 -flto, because the SSA_NAMEs can already
overlap.

As we don't have info right now if a particular function has always been
compiled with !optimize during the LTO pipeline, this patch fixes the
wrong-code by always assuming it might be optimized at some point if
in_lto_p.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2014-03-04  Jakub Jelinek  <ja...@redhat.com>

        PR lto/60404
        * cfgexpand.c (expand_used_vars): Do not assume all SSA_NAMEs
        of PARM/RESULT_DECLs must be coalesced with optimize && in_lto_p.
        * tree-ssa-coalesce.c (coalesce_ssa_name): Use MUST_COALESCE_COST - 1
        cost for in_lto_p.

        * gcc.dg/lto/pr60404_0.c: New test.
        * gcc.dg/lto/pr60404_1.c: New file.
        * gcc.dg/lto/pr60404_2.c: New file.

--- gcc/cfgexpand.c.jj  2014-03-04 10:57:18.000000000 +0100
+++ gcc/cfgexpand.c     2014-03-04 11:12:12.684524383 +0100
@@ -1652,10 +1652,12 @@ expand_used_vars (void)
         debug info, there is no need to do so if optimization is disabled
         because all the SSA_NAMEs based on these DECLs have been coalesced
         into a single partition, which is thus assigned the canonical RTL
-        location of the DECLs.  */
+        location of the DECLs.  If in_lto_p, we can't rely on optimize,
+        a function could be compiled with -O1 -flto first and only the
+        link performed at -O0.  */
       if (TREE_CODE (SSA_NAME_VAR (var)) == VAR_DECL)
        expand_one_var (var, true, true);
-      else if (DECL_IGNORED_P (SSA_NAME_VAR (var)) || optimize)
+      else if (DECL_IGNORED_P (SSA_NAME_VAR (var)) || optimize || in_lto_p)
        {
          /* This is a PARM_DECL or RESULT_DECL.  For those partitions that
             contain the default def (representing the parm or result itself)
--- gcc/tree-ssa-coalesce.c.jj  2014-01-03 11:40:57.000000000 +0100
+++ gcc/tree-ssa-coalesce.c     2014-03-04 11:15:46.955296026 +0100
@@ -1289,9 +1289,12 @@ coalesce_ssa_name (void)
                     _require_ that all the names originating from it be
                     coalesced, because there must be a single partition
                     containing all the names so that it can be assigned
-                    the canonical RTL location of the DECL safely.  */
+                    the canonical RTL location of the DECL safely.
+                    If in_lto_p, a function could have been compiled
+                    originally with optimizations and only the link
+                    performed at -O0, so we can't actually require it.  */
                  const int cost
-                   = TREE_CODE (SSA_NAME_VAR (a)) == VAR_DECL
+                   = (TREE_CODE (SSA_NAME_VAR (a)) == VAR_DECL || in_lto_p)
                      ? MUST_COALESCE_COST - 1 : MUST_COALESCE_COST;
                  add_coalesce (cl, SSA_NAME_VERSION (a),
                                SSA_NAME_VERSION (*slot), cost);
--- gcc/testsuite/gcc.dg/lto/pr60404_0.c.jj     2014-03-04 11:19:48.503909557 
+0100
+++ gcc/testsuite/gcc.dg/lto/pr60404_0.c        2014-03-04 11:09:51.590333222 
+0100
@@ -0,0 +1,15 @@
+/* { dg-lto-do run } */
+/* { dg-lto-options { { -O1 -flto } } } */
+/* { dg-extra-ld-options { -O0 } } */
+
+extern void fn2 (int);
+int a[1], b;
+
+int
+main ()
+{
+  fn2 (0);
+  if (b != 0 || a[b] != 0)
+    __builtin_abort ();
+  return 0;
+}
--- gcc/testsuite/gcc.dg/lto/pr60404_1.c.jj     2014-03-04 11:19:51.057894910 
+0100
+++ gcc/testsuite/gcc.dg/lto/pr60404_1.c        2014-03-04 09:32:35.000000000 
+0100
@@ -0,0 +1,4 @@
+void
+fn1 (int p)
+{
+}
--- gcc/testsuite/gcc.dg/lto/pr60404_2.c.jj     2014-03-04 11:19:54.578874717 
+0100
+++ gcc/testsuite/gcc.dg/lto/pr60404_2.c        2014-03-04 09:33:18.000000000 
+0100
@@ -0,0 +1,9 @@
+extern int b;
+extern void fn1 (int);
+
+void
+fn2 (int p)
+{
+  b = p++;
+  fn1 (p);
+}

        Jakub

Reply via email to