[PATCH][LTO] Fix PR48437

[Richard Guenther]

On Wed, 7 Dec 2011, Diego Novillo wrote:

> On 12/07/11 10:54, Richard Guenther wrote:
> > On Wed, 7 Dec 2011, Diego Novillo wrote:
> > 
> > > On 12/07/11 10:46, Richard Guenther wrote:
> > > > On Wed, 7 Dec 2011, Diego Novillo wrote:
> > > > 
> > > > > On 12/07/11 09:52, Richard Guenther wrote:
> > > > > 
> > > > > > Index: gcc/lto-streamer-out.c
> > > > > > ===================================================================
> > > > > > *** gcc/lto-streamer-out.c	(revision 182081)
> > > > > > --- gcc/lto-streamer-out.c	(working copy)
> > > > > > *************** tree_is_indexable (tree t)
> > > > > > *** 129,134 ****
> > > > > > --- 129,144 ----
> > > > > >        else if (TREE_CODE (t) == VAR_DECL&&    decl_function_context
> > > > > > (t)
> > > > > >      	&&    !TREE_STATIC (t))
> > > > > >          return false;
> > > > > > +   /* If this is a decl generated for block local externs for
> > > > > > +      debug info generation, stream it unshared alongside
> > > > > > BLOCK_VARS.
> > > > > > */
> > > > > > +   else if (VAR_OR_FUNCTION_DECL_P (t)
> > > > > > + 	   /* ???  The following tests are a literal match on what
> > > > > > + 	      c-decl.c:pop_scope does.  */
> > > > > 
> > > > > Factor it into a common routine then?
> > > > 
> > > > pop_scope of course _sets_ the values that way, it doesn't test them.
> > > 
> > > Yes.  I meant factoring, so future users have something to call, instead
> > > of
> > > three seemingly random flag checks.  pop_scope could also be calling some
> > > complementary setter instead of doing the seemingly random flag setting.
> > 
> > I don't see a good way to factor out the flags setting.  Factoring out
> > the copying maybe.  But well... we can do that when a 2nd user
> > comes along?
> 
> The problem is that the 2nd user will cut-n-paste from this one. However, if
> you find adding a little function too strenuous, I guess it's not too big a
> deal.

Testing this kind of patches turns out to be quite time-consuming
(I do a LTO bootstrap and two SPEC 2k6 builds (-g and -g0)), so
yeah ...

The following is what I ended up LTO bootstrapping (finished ok),
testing is still in progress, as is SPEC 2k6 build.

It fixes PR49945 in a similar way, by streaming VLA types locally as well.

I'm going to apply it tomorrow, when full testing hopefully finished

Thanks,
Richard.

2011-12-08  Richard Guenther  <rguenther@suse.de>

	PR lto/48437
	PR lto/49945
	* lto-streamer-out.c (tree_is_indexable): Exclude block-local
	extern declarations.

	PR lto/48437
	* gcc.dg/lto/20111207-2_0.c: New testcase.
	* gcc.dg/guality/pr48437.c: Likewise.

Index: gcc/lto-streamer-out.c
===================================================================
*** gcc/lto-streamer-out.c	(revision 182081)
--- gcc/lto-streamer-out.c	(working copy)
*************** tree_is_indexable (tree t)
*** 129,134 ****
--- 129,147 ----
    else if (TREE_CODE (t) == VAR_DECL && decl_function_context (t)
  	   && !TREE_STATIC (t))
      return false;
+   /* If this is a decl generated for block local externs for
+      debug info generation, stream it unshared alongside BLOCK_VARS.  */
+   else if (VAR_OR_FUNCTION_DECL_P (t)
+ 	   /* ???  The following tests are a literal match on what
+ 	      c-decl.c:pop_scope does.  */
+ 	   && TREE_PUBLIC (t)
+ 	   && DECL_EXTERNAL (t)
+ 	   && DECL_CONTEXT (t)
+ 	   && TREE_CODE (DECL_CONTEXT (t)) == FUNCTION_DECL)
+     return false;
+   else if (TYPE_P (t)
+ 	   && variably_modified_type_p (t, NULL_TREE))
+     return false;
    else
      return (TYPE_P (t) || DECL_P (t) || TREE_CODE (t) == SSA_NAME);
  }
Index: gcc/testsuite/gcc.dg/lto/20111207-2_0.c
===================================================================
*** gcc/testsuite/gcc.dg/lto/20111207-2_0.c	(revision 0)
--- gcc/testsuite/gcc.dg/lto/20111207-2_0.c	(revision 0)
***************
*** 0 ****
--- 1,17 ----
+ /* { dg-lto-do run } */
+ 
+ int
+ test (void)
+ {
+   int f (void);
+   return 0;
+ }
+ 
+ int
+ main (void)
+ {
+   int f (void);
+   int test (void);
+ 
+   return test ();
+ }
Index: gcc/testsuite/gcc.dg/guality/pr48437.c
===================================================================
*** gcc/testsuite/gcc.dg/guality/pr48437.c	(revision 0)
--- gcc/testsuite/gcc.dg/guality/pr48437.c	(revision 0)
***************
*** 0 ****
--- 1,17 ----
+ /* PR lto/48437 */
+ /* { dg-do run } */
+ /* { dg-options "-g" } */
+ 
+ #include "../nop.h"
+ 
+ int i __attribute__((used));
+ int main()
+ {
+   volatile int i;
+   for (i = 3; i < 7; ++i)
+     {
+       extern int i;
+       asm volatile (NOP : : : "memory"); /* { dg-final { gdb-test 14 "i" "0" } } */
+     }
+   return 0;
+ }