Re: [PATCH] Fix bss_initializer_p (PR target/83100)

[Richard Biener <rguenther@suse.de>]

On Sat, 25 Nov 2017, Jakub Jelinek wrote:

> Hi!
> 
> As the testcases show, it is essential that bss_initializer_p returns
> true for DECL_COMMON vars without initializer, even when they are
> TREE_READONLY, otherwise they aren't really common, e.g.
>   if (DECL_COMMON (decl))
>     {
>       /* If the decl has been given an explicit section name, or it resides
>          in a non-generic address space, then it isn't common, and shouldn't
>          be handled as such.  */
>       gcc_assert (DECL_SECTION_NAME (decl) == NULL
>                   && ADDR_SPACE_GENERIC_P (as));
>       if (DECL_THREAD_LOCAL_P (decl))
>         return tls_comm_section;
>       else if (TREE_PUBLIC (decl) && bss_initializer_p (decl))
>         return comm_section;
>     }
> falls through into non-common code, and on section anchors target sometimes
> ICE.
> 
> Fixed thusly, bootstrapped/regtested on powerpc64le-linux, bootstrapped
> also on {powerpc64,x86_64,i686}-linux, ok for trunk if the pending
> regtests on the above 3 targets succeed?

Ok.

Richard.

> 2017-11-24  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR target/83100
> 	* varasm.c (bss_initializer_p): Return true for DECL_COMMON
> 	TREE_READONLY decls.
> 
> 	* gcc.dg/pr83100-1.c: New test.
> 	* gcc.dg/pr83100-2.c: New test.
> 	* gcc.dg/pr83100-3.c: New test.
> 	* gcc.dg/pr83100-4.c: New test.
> 
> --- gcc/varasm.c.jj	2017-11-21 20:23:02.000000000 +0100
> +++ gcc/varasm.c	2017-11-24 21:43:55.616951823 +0100
> @@ -986,9 +986,9 @@ decode_reg_name (const char *name)
>  bool
>  bss_initializer_p (const_tree decl)
>  {
> -  /* Do not put constants into the .bss section, they belong in a readonly
> -     section.  */
> -  return (!TREE_READONLY (decl)
> +  /* Do not put non-common constants into the .bss section, they belong in
> +     a readonly section.  */
> +  return ((!TREE_READONLY (decl) || DECL_COMMON (decl))
>  	  && (DECL_INITIAL (decl) == NULL
>  	      /* In LTO we have no errors in program; error_mark_node is used
>  	         to mark offlined constructors.  */
> --- gcc/testsuite/gcc.dg/pr83100-1.c.jj	2017-11-24 21:56:52.957497438 +0100
> +++ gcc/testsuite/gcc.dg/pr83100-1.c	2017-11-24 22:01:03.433437167 +0100
> @@ -0,0 +1,7 @@
> +/* PR target/83100 */
> +/* { dg-do compile { target *-*-linux* *-*-gnu* } } */
> +/* { dg-options "-O2 -fcommon -fdata-sections" } */
> +
> +const int a;
> +
> +/* { dg-final { scan-assembler "comm" } } */
> --- gcc/testsuite/gcc.dg/pr83100-2.c.jj	2017-11-24 21:58:54.475012758 +0100
> +++ gcc/testsuite/gcc.dg/pr83100-2.c	2017-11-24 22:01:09.108367832 +0100
> @@ -0,0 +1,15 @@
> +/* PR target/83100 */
> +/* { dg-do run } */
> +/* { dg-options "-O2 -fcommon -fdata-sections" } */
> +/* { dg-additional-sources pr83100-3.c } */
> +/* { dg-skip-if "-fdata-sections not supported" { hppa*-*-hpux* nvptx-*-* } } */
> +
> +const int a;
> +
> +int
> +main ()
> +{
> +  if (a != 7)
> +    __builtin_abort ();
> +  return 0;
> +}
> --- gcc/testsuite/gcc.dg/pr83100-3.c.jj	2017-11-24 21:59:57.072247956 +0100
> +++ gcc/testsuite/gcc.dg/pr83100-3.c	2017-11-24 22:01:14.344303860 +0100
> @@ -0,0 +1,6 @@
> +/* PR target/83100 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -fcommon -fdata-sections" } */
> +/* { dg-skip-if "-fdata-sections not supported" { hppa*-*-hpux* nvptx-*-* } } */
> +
> +const int a = 7;
> --- gcc/testsuite/gcc.dg/pr83100-4.c.jj	2017-11-24 22:00:13.405048405 +0100
> +++ gcc/testsuite/gcc.dg/pr83100-4.c	2017-11-24 22:01:20.281231323 +0100
> @@ -0,0 +1,7 @@
> +/* PR target/83100 */
> +/* { dg-do compile { target *-*-linux* *-*-gnu* } } */
> +/* { dg-options "-O2 -fno-common -fdata-sections" } */
> +
> +const int a;
> +
> +/* { dg-final { scan-assembler "rodata.a" } } */
> 
> 	Jakub
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nuernberg)