assemble_integer & align

assemble_integer & align

[Denis Chertykov]

Today (after long delay) I have tried to build my port (avr) and have
founded a changes in varasm.c:assemble_integer.
ChangeLog entry was:
2001-08-17  Richard Henderson  <rth@redhat.com>
	* defaults.h (UNALIGNED_SHORT_ASM_OP, UNALIGNED_INT_ASM_OP,
	UNALIGNED_DOUBLE_INT_ASM_OP, ASM_BYTE_OP): Move from ...
	* dwarf2asm.c: ... here.
	* dwarfout.c: Remove them.

	* varasm.c (assemble_integer): Add align parameter.
...
	(assemble_integer): Handle unaligned data.

Diff was:
...
...
 int
-assemble_integer (x, size, force)
+assemble_integer (x, size, align, force)
      rtx x;
-     int size;
+     unsigned int size;
+     unsigned int align;
      int force;
 {
   /* First try to use the standard 1, 2, 4, 8, and 16 byte
      ASM_OUTPUT... macros.  */

-  switch (size)
-    {
+  if (align >= size * BITS_PER_UNIT)
+    switch (size)
+      {
--------------------------------------------------------------------

My main problem is 'if (align >= size * BITS_PER_UNIT)' because my
ALIGN always 8 and if size > 1 then `assemble_integer' will output
all integers as:
...
...
      if (asm_op)
	{
	  fputs (asm_op, asm_out_file);
	  output_addr_const (asm_out_file, x);
	  fputc ('\n', asm_out_file);
	  return 1;
	}
--------------------------------------------------------
My version of ASM_OUTPUT_SHORT is a port specific.
void
asm_output_short (file, value)
     FILE *file;
     rtx value;
{
  if (SYMBOL_REF_FLAG (value) || GET_CODE (value) == LABEL_REF)
    {
      fprintf (file, "\t.word pm(");
      output_addr_const (file, (value));
      fprintf (file, ")\n");
    }
  else
    {
      fprintf (file, "\t.word ");
      output_addr_const (file, (value));
      fprintf (file, "\n");
    }
}
----------------------------------------
How can I resolv this problem ?

Denis.

Re: assemble_integer & align

[Richard Henderson]

On Sat, Sep 22, 2001 at 11:38:44AM +0400, Denis Chertykov wrote:
> My main problem is 'if (align >= size * BITS_PER_UNIT)' because my
> ALIGN always 8 and if size > 1 then `assemble_integer' will output
> all integers as:

Why is your align always 8?  Is that BIGGEST_ALIGNMENT?

If so, we can perhaps modify the tests there to take that
into account.


r~

Re: assemble_integer & align

[Denis Chertykov]

Richard Henderson <rth@redhat.com> writes:

> On Sat, Sep 22, 2001 at 11:38:44AM +0400, Denis Chertykov wrote:
> > My main problem is 'if (align >= size * BITS_PER_UNIT)' because my
> > ALIGN always 8 and if size > 1 then `assemble_integer' will output
> > all integers as:
> 
> Why is your align always 8?  Is that BIGGEST_ALIGNMENT?

Yes. (You can see it in avr/avr.h). AVR is an 8 bit target which isn't
needed in alignments at all.

Denis.

PS: Is you your timezone GMT-7 ?

Re: assemble_integer & align

[Richard Henderson]

On Sat, Sep 22, 2001 at 01:19:22PM +0400, Denis Chertykov wrote:
> Yes. (You can see it in avr/avr.h). AVR is an 8 bit target which isn't
> needed in alignments at all.

Oh, duh.  This should work for you then.  I'll wait for an
success ack before applying though.

> PS: Is you your timezone GMT-7 ?

Yes, until daylight savings time ends in another month or so.
Then we'll be back at GMT-8.


r~


	* varasm.c (assemble_integer): Bound aligned data check by
	BIGGEST_ALIGNMENT.

Index: varasm.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/varasm.c,v
retrieving revision 1.209
diff -c -p -d -u -r1.209 varasm.c
--- varasm.c	2001/09/22 13:23:14	1.209
+++ varasm.c	2001/09/24 20:11:36
@@ -1913,7 +1913,7 @@ assemble_integer (x, size, align, force)
   /* First try to use the standard 1, 2, 4, 8, and 16 byte
      ASM_OUTPUT... macros.  */
 
-  if (align >= size * BITS_PER_UNIT)
+  if (align >= MIN (size * BITS_PER_UNIT, BIGGEST_ALIGNMENT))
     switch (size)
       {
 #ifdef ASM_OUTPUT_CHAR