Re: [lto] add direct-to-ELF serialization to lto1

[Diego Novillo <dnovillo at google dot com>]

On Wed, Jun 25, 2008 at 14:56, Ollie Wild <aaw@google.com> wrote:

> +  if (obs->first_block == NULL)
> +    {
> +      /* This is the first time the stream has been written
> +	 into.  */
> +      obs->block_size = 1024;

Does it make sense to #define or --param this number?

> +	 pointer to the new block.  */
> +      tptr = obs->current_block;
> +      tptr->ptr = (char *)new_block;

Space after typecast.

> +    }
> +
> +  /* Set the place for the next char at the first position after the
> +     chain to the next block.  */
> +  obs->current_pointer
> +    = ((char *)new_block) + sizeof (struct lto_char_ptr_base);

Likewise.

> @@ -36,6 +36,7 @@ Boston, MA 02110-1301, USA.  */
>  #include "tm.h"
>  #include "libiberty.h"
>  #include "ggc.h"
> +#include "lto-section-out.h"

Needs corresponding changes in lto/Make-lang.in

> +  /* Grab the current output file and do some basic assertion checking.  */
> +  file = (lto_elf_file *)lto_get_current_out_file (),

Space after type cast.

> +  gcc_assert (file);
> +  gcc_assert (file->elf);
> +  gcc_assert (!file->scn);
> +
> +  /* Create a new section.  */
> +  scn = elf_newscn(file->elf);

Space after '('.

> +
> +  /* Free any ELF data buffers.  */
> +  cur = elf_file->data;
> +  while (cur)
> +    {
> +      tmp = cur;
> +      cur = (struct lto_char_ptr_base *)cur->ptr;

Space after typecast.

> +  /* This is some bogus wrapper code for development testing.  It will be
> +     replaced once some basic WPA partitioning logic is implemented.  To use
> +     this pass "-flto -fsyntax-only" to the lto1 invocation.  */
> +  if (flag_generate_lto)
> +    {
> +      lto_file *file;
> +
> +      file = lto_elf_file_open ("bogus.lto.o", /*writable=*/true);
> +      if (!file)
> +	fatal_error ("lto_elf_file_open() failed");
> +      lto_set_current_out_file (file);
> +    }
> +

I'm confused by this hunk.  What are you testing here?  It also
seems that it will trigger with just -flto.  Where does
-fsyntax-only come to play?

> -/* Open the ELF input file indicated by FILENAME.  Return */
> -extern lto_file *lto_elf_file_open (const char *filename);
> +/* Open the ELF file indicated by FILENAME.  Return */

Return what?


With the exception of the -fsyntax-only test and the langhooks,
the rest looks fine to me.


Diego.