rfc: DW_AT_vector

rfc: DW_AT_vector

[Aldy Hernandez]

hullo.

it was brought to my attention that the dwarf2 information emitted
by gcc for vector types is an array of ints/floats within a structure.

this causes problems for gdb because vectors and structures may have
different calling conventions.  for example, in altivec, vectors
go in registers and get returned in registers.  this may not necessarily
be true for structures on the same platform so gdb is unable to get
the correct information.

richard suggested:

  I suppose we could invent either (1) a DW_AT_vector flag to go along
  with the current struct-of-array-of representation or (2) a new
  DW_TAG_vector_type, which looks cleaner but wouldn't work at all with
  older gdb or other debuggers.

i am completely clueless in all this, having never heard of DW_AT_
or DW_TAG_ until about 30 minutes ago.  but of course, that doesn't
keep me from coming up with a patch which i obviously think is correct. :)

i suggest coming up with a DW_AT_vector flag to go along with the
structure business.  this will work with any debugger, and will not
break any type of compatibility.  then gdb can be tweaked to look for
the right values if a structure has a vector tag as well.

of course, i'm willing to volunteer to code the DW_TAG_vector_type
stuff, but only as a last and painful resort :).

here is a proof of concept patch for gcc, gdb, binutils, etc.

comments?

2002-02-22  Aldy Hernandez  <aldyh@redhat.com>

	* gdb/dwarf2read.c (dwarf_attr_name): Add case for DW_AT_vector.

	* include/elf/dwarf2.h (dwarf_attribute): Add DW_AT_vector.

	* binutils/readelf.c (get_AT_name): Add case for DW_AT_vector.

	* dwarf2.h (dwarf_attribute): Add DW_AT_vector.

	* dwarf2out.c (gen_struct_or_union_type_die): Add vector tag.
	(dwarf_attr_name): Add case for DW_AT_vector.

Index: gcc/dwarf2out.c
===================================================================
RCS file: /cvs/uberbaum/gcc/dwarf2out.c,v
retrieving revision 1.353
diff -c -p -r1.353 dwarf2out.c
*** dwarf2out.c	2002/01/30 00:19:23	1.353
--- dwarf2out.c	2002/02/22 03:39:41
*************** dwarf_attr_name (attr)
*** 4164,4169 ****
--- 4164,4171 ----
        return "DW_AT_body_begin";
      case DW_AT_body_end:
        return "DW_AT_body_end";
+     case DW_AT_vector:
+       return "DW_AT_vector";
      case DW_AT_VMS_rtnbeg_pd_address:
        return "DW_AT_VMS_rtnbeg_pd_address";
  
*************** gen_struct_or_union_type_die (type, cont
*** 10888,10893 ****
--- 10890,10897 ----
  	  && ! decl_function_context (TYPE_STUB_DECL (type)))
  	VARRAY_PUSH_TREE (incomplete_types, type);
      }
+   if (TREE_CODE (type) == VECTOR_TYPE)
+     add_AT_flag (type_die, DW_AT_vector, 1);
  }
  
  /* Generate a DIE for a subroutine _type_.  */
Index: gcc/dwarf2.h
===================================================================
RCS file: /cvs/uberbaum/gcc/dwarf2.h,v
retrieving revision 1.22
diff -c -p -r1.22 dwarf2.h
*** dwarf2.h	2002/01/28 23:23:26	1.22
--- dwarf2.h	2002/02/22 03:39:41
*************** enum dwarf_attribute
*** 239,244 ****
--- 239,245 ----
      DW_AT_src_coords = 0x2104,
      DW_AT_body_begin = 0x2105,
      DW_AT_body_end   = 0x2106,
+     DW_AT_vector     = 0x2107,
      /* VMS Extensions.  */
      DW_AT_VMS_rtnbeg_pd_address = 0x2201
    };
Index: gdb/dwarf2read.c
===================================================================
RCS file: /cvs/uberbaum/gdb/dwarf2read.c,v
retrieving revision 1.48
diff -c -p -r1.48 dwarf2read.c
*** dwarf2read.c	2002/02/15 22:42:33	1.48
--- dwarf2read.c	2002/02/22 03:39:43
*************** dwarf_attr_name (register unsigned attr)
*** 5161,5166 ****
--- 5161,5168 ----
        return "DW_AT_body_begin";
      case DW_AT_body_end:
        return "DW_AT_body_end";
+     case DW_AT_vector:
+       return "DW_AT_vector";
      default:
        return "DW_AT_<unknown>";
      }
Index: binutils/readelf.c
===================================================================
RCS file: /cvs/uberbaum/binutils/readelf.c,v
retrieving revision 1.151
diff -c -p -r1.151 readelf.c
*** readelf.c	2002/02/13 18:14:42	1.151
--- readelf.c	2002/02/22 03:39:46
*************** get_AT_name (attribute)
*** 6455,6460 ****
--- 6455,6461 ----
      case DW_AT_src_coords:           return "DW_AT_src_coords";
      case DW_AT_body_begin:           return "DW_AT_body_begin";
      case DW_AT_body_end:             return "DW_AT_body_end";
+     case DW_AT_vector:               return "DW_AT_vector";
      default:
        {
  	static char buffer [100];
Index: gcc/dwarf2out.c
===================================================================
RCS file: /cvs/uberbaum/gcc/dwarf2out.c,v
retrieving revision 1.353
diff -c -p -r1.353 dwarf2out.c
*** dwarf2out.c	2002/01/30 00:19:23	1.353
--- dwarf2out.c	2002/02/22 03:40:26
*************** dwarf_attr_name (attr)
*** 4164,4169 ****
--- 4164,4171 ----
        return "DW_AT_body_begin";
      case DW_AT_body_end:
        return "DW_AT_body_end";
+     case DW_AT_vector:
+       return "DW_AT_vector";
      case DW_AT_VMS_rtnbeg_pd_address:
        return "DW_AT_VMS_rtnbeg_pd_address";
  
*************** gen_struct_or_union_type_die (type, cont
*** 10888,10893 ****
--- 10890,10897 ----
  	  && ! decl_function_context (TYPE_STUB_DECL (type)))
  	VARRAY_PUSH_TREE (incomplete_types, type);
      }
+   if (TREE_CODE (type) == VECTOR_TYPE)
+     add_AT_flag (type_die, DW_AT_vector, 1);
  }
  
  /* Generate a DIE for a subroutine _type_.  */
Index: gcc/dwarf2.h
===================================================================
RCS file: /cvs/uberbaum/gcc/dwarf2.h,v
retrieving revision 1.22
diff -c -p -r1.22 dwarf2.h
*** dwarf2.h	2002/01/28 23:23:26	1.22
--- dwarf2.h	2002/02/22 03:40:27
*************** enum dwarf_attribute
*** 239,244 ****
--- 239,245 ----
      DW_AT_src_coords = 0x2104,
      DW_AT_body_begin = 0x2105,
      DW_AT_body_end   = 0x2106,
+     DW_AT_vector     = 0x2107,
      /* VMS Extensions.  */
      DW_AT_VMS_rtnbeg_pd_address = 0x2201
    };
Index: gdb/dwarf2read.c
===================================================================
RCS file: /cvs/uberbaum/gdb/dwarf2read.c,v
retrieving revision 1.48
diff -c -p -r1.48 dwarf2read.c
*** dwarf2read.c	2002/02/15 22:42:33	1.48
--- dwarf2read.c	2002/02/22 03:40:29
*************** dwarf_attr_name (register unsigned attr)
*** 5161,5166 ****
--- 5161,5168 ----
        return "DW_AT_body_begin";
      case DW_AT_body_end:
        return "DW_AT_body_end";
+     case DW_AT_vector:
+       return "DW_AT_vector";
      default:
        return "DW_AT_<unknown>";
      }
Index: include/elf/dwarf2.h
===================================================================
RCS file: /cvs/uberbaum/include/elf/dwarf2.h,v
retrieving revision 1.8
diff -c -p -r1.8 dwarf2.h
*** dwarf2.h	2002/01/28 23:26:53	1.8
--- dwarf2.h	2002/02/22 03:40:29
*************** enum dwarf_attribute
*** 328,333 ****
--- 328,334 ----
      DW_AT_src_coords = 0x2104,
      DW_AT_body_begin = 0x2105,
      DW_AT_body_end   = 0x2106,
+     DW_AT_vector     = 0x2107,
      /* VMS Extensions.  */
      DW_AT_VMS_rtnbeg_pd_address = 0x2201
    };
Index: binutils/readelf.c
===================================================================
RCS file: /cvs/uberbaum/binutils/readelf.c,v
retrieving revision 1.151
diff -c -p -r1.151 readelf.c
*** readelf.c	2002/02/13 18:14:42	1.151
--- readelf.c	2002/02/22 03:40:35
*************** get_AT_name (attribute)
*** 6455,6460 ****
--- 6455,6461 ----
      case DW_AT_src_coords:           return "DW_AT_src_coords";
      case DW_AT_body_begin:           return "DW_AT_body_begin";
      case DW_AT_body_end:             return "DW_AT_body_end";
+     case DW_AT_vector:               return "DW_AT_vector";
      default:
        {
  	static char buffer [100];

Re: rfc: DW_AT_vector

[Andrew Cagney]

Er, just an aside, any chance of GCC, GDB and BINUTILS agreeing to have 
only one version of this file? :-)

Andrew


> diff -c -p -r1.22 dwarf2.h
> *** dwarf2.h	2002/01/28 23:23:26	1.22
> --- dwarf2.h	2002/02/22 03:39:41
> *************** enum dwarf_attribute
> *** 239,244 ****
> --- 239,245 ----
>       DW_AT_src_coords = 0x2104,
>       DW_AT_body_begin = 0x2105,
>       DW_AT_body_end   = 0x2106,
> +     DW_AT_vector     = 0x2107,
>       /* VMS Extensions.  */
>       DW_AT_VMS_rtnbeg_pd_address = 0x2201
>     };
> 

> diff -c -p -r1.8 dwarf2.h
> *** dwarf2.h	2002/01/28 23:26:53	1.8
> --- dwarf2.h	2002/02/22 03:40:29
> *************** enum dwarf_attribute
> *** 328,333 ****
> --- 328,334 ----
>       DW_AT_src_coords = 0x2104,
>       DW_AT_body_begin = 0x2105,
>       DW_AT_body_end   = 0x2106,
> +     DW_AT_vector     = 0x2107,
>       /* VMS Extensions.  */
>       DW_AT_VMS_rtnbeg_pd_address = 0x2201
>     };
> 

Re: rfc: DW_AT_vector

[Elena Zannoni]

Aldy Hernandez writes:
 > hullo.
 > 
 > it was brought to my attention that the dwarf2 information emitted
 > by gcc for vector types is an array of ints/floats within a structure.
 > 
 > this causes problems for gdb because vectors and structures may have
 > different calling conventions.  for example, in altivec, vectors
 > go in registers and get returned in registers.  this may not necessarily
 > be true for structures on the same platform so gdb is unable to get
 > the correct information.
 > 
 > richard suggested:
 > 
 >   I suppose we could invent either (1) a DW_AT_vector flag to go along
 >   with the current struct-of-array-of representation or (2) a new
 >   DW_TAG_vector_type, which looks cleaner but wouldn't work at all with
 >   older gdb or other debuggers.
 > 
 > i am completely clueless in all this, having never heard of DW_AT_
 > or DW_TAG_ until about 30 minutes ago.  but of course, that doesn't
 > keep me from coming up with a patch which i obviously think is correct. :)
 > 
 > i suggest coming up with a DW_AT_vector flag to go along with the
 > structure business.  this will work with any debugger, and will not
 > break any type of compatibility.  then gdb can be tweaked to look for
 > the right values if a structure has a vector tag as well.
 > 
 > of course, i'm willing to volunteer to code the DW_TAG_vector_type
 > stuff, but only as a last and painful resort :).
 > 

This would work, yes, even though I would personally prefer to have a
new vector type. It represents more closely what these vectors really
are.

Anyway, you'll need to talk to the dwarf2/3 committee as well.


Elena


 > here is a proof of concept patch for gcc, gdb, binutils, etc.
 > 
 > comments?
 > 
 > 2002-02-22  Aldy Hernandez  <aldyh@redhat.com>
 > 
 > 	* gdb/dwarf2read.c (dwarf_attr_name): Add case for DW_AT_vector.
 > 
 > 	* include/elf/dwarf2.h (dwarf_attribute): Add DW_AT_vector.
 > 
 > 	* binutils/readelf.c (get_AT_name): Add case for DW_AT_vector.
 > 
 > 	* dwarf2.h (dwarf_attribute): Add DW_AT_vector.
 > 
 > 	* dwarf2out.c (gen_struct_or_union_type_die): Add vector tag.
 > 	(dwarf_attr_name): Add case for DW_AT_vector.
 > 
 > Index: gcc/dwarf2out.c
 > ===================================================================
 > RCS file: /cvs/uberbaum/gcc/dwarf2out.c,v
 > retrieving revision 1.353
 > diff -c -p -r1.353 dwarf2out.c
 > *** dwarf2out.c	2002/01/30 00:19:23	1.353
 > --- dwarf2out.c	2002/02/22 03:39:41
 > *************** dwarf_attr_name (attr)
 > *** 4164,4169 ****
 > --- 4164,4171 ----
 >         return "DW_AT_body_begin";
 >       case DW_AT_body_end:
 >         return "DW_AT_body_end";
 > +     case DW_AT_vector:
 > +       return "DW_AT_vector";
 >       case DW_AT_VMS_rtnbeg_pd_address:
 >         return "DW_AT_VMS_rtnbeg_pd_address";
 >   
 > *************** gen_struct_or_union_type_die (type, cont
 > *** 10888,10893 ****
 > --- 10890,10897 ----
 >   	  && ! decl_function_context (TYPE_STUB_DECL (type)))
 >   	VARRAY_PUSH_TREE (incomplete_types, type);
 >       }
 > +   if (TREE_CODE (type) == VECTOR_TYPE)
 > +     add_AT_flag (type_die, DW_AT_vector, 1);
 >   }
 >   
 >   /* Generate a DIE for a subroutine _type_.  */
 > Index: gcc/dwarf2.h
 > ===================================================================
 > RCS file: /cvs/uberbaum/gcc/dwarf2.h,v
 > retrieving revision 1.22
 > diff -c -p -r1.22 dwarf2.h
 > *** dwarf2.h	2002/01/28 23:23:26	1.22
 > --- dwarf2.h	2002/02/22 03:39:41
 > *************** enum dwarf_attribute
 > *** 239,244 ****
 > --- 239,245 ----
 >       DW_AT_src_coords = 0x2104,
 >       DW_AT_body_begin = 0x2105,
 >       DW_AT_body_end   = 0x2106,
 > +     DW_AT_vector     = 0x2107,
 >       /* VMS Extensions.  */
 >       DW_AT_VMS_rtnbeg_pd_address = 0x2201
 >     };
 > Index: gdb/dwarf2read.c
 > ===================================================================
 > RCS file: /cvs/uberbaum/gdb/dwarf2read.c,v
 > retrieving revision 1.48
 > diff -c -p -r1.48 dwarf2read.c
 > *** dwarf2read.c	2002/02/15 22:42:33	1.48
 > --- dwarf2read.c	2002/02/22 03:39:43
 > *************** dwarf_attr_name (register unsigned attr)
 > *** 5161,5166 ****
 > --- 5161,5168 ----
 >         return "DW_AT_body_begin";
 >       case DW_AT_body_end:
 >         return "DW_AT_body_end";
 > +     case DW_AT_vector:
 > +       return "DW_AT_vector";
 >       default:
 >         return "DW_AT_<unknown>";
 >       }
 > Index: binutils/readelf.c
 > ===================================================================
 > RCS file: /cvs/uberbaum/binutils/readelf.c,v
 > retrieving revision 1.151
 > diff -c -p -r1.151 readelf.c
 > *** readelf.c	2002/02/13 18:14:42	1.151
 > --- readelf.c	2002/02/22 03:39:46
 > *************** get_AT_name (attribute)
 > *** 6455,6460 ****
 > --- 6455,6461 ----
 >       case DW_AT_src_coords:           return "DW_AT_src_coords";
 >       case DW_AT_body_begin:           return "DW_AT_body_begin";
 >       case DW_AT_body_end:             return "DW_AT_body_end";
 > +     case DW_AT_vector:               return "DW_AT_vector";
 >       default:
 >         {
 >   	static char buffer [100];
 > Index: gcc/dwarf2out.c
 > ===================================================================
 > RCS file: /cvs/uberbaum/gcc/dwarf2out.c,v
 > retrieving revision 1.353
 > diff -c -p -r1.353 dwarf2out.c
 > *** dwarf2out.c	2002/01/30 00:19:23	1.353
 > --- dwarf2out.c	2002/02/22 03:40:26
 > *************** dwarf_attr_name (attr)
 > *** 4164,4169 ****
 > --- 4164,4171 ----
 >         return "DW_AT_body_begin";
 >       case DW_AT_body_end:
 >         return "DW_AT_body_end";
 > +     case DW_AT_vector:
 > +       return "DW_AT_vector";
 >       case DW_AT_VMS_rtnbeg_pd_address:
 >         return "DW_AT_VMS_rtnbeg_pd_address";
 >   
 > *************** gen_struct_or_union_type_die (type, cont
 > *** 10888,10893 ****
 > --- 10890,10897 ----
 >   	  && ! decl_function_context (TYPE_STUB_DECL (type)))
 >   	VARRAY_PUSH_TREE (incomplete_types, type);
 >       }
 > +   if (TREE_CODE (type) == VECTOR_TYPE)
 > +     add_AT_flag (type_die, DW_AT_vector, 1);
 >   }
 >   
 >   /* Generate a DIE for a subroutine _type_.  */
 > Index: gcc/dwarf2.h
 > ===================================================================
 > RCS file: /cvs/uberbaum/gcc/dwarf2.h,v
 > retrieving revision 1.22
 > diff -c -p -r1.22 dwarf2.h
 > *** dwarf2.h	2002/01/28 23:23:26	1.22
 > --- dwarf2.h	2002/02/22 03:40:27
 > *************** enum dwarf_attribute
 > *** 239,244 ****
 > --- 239,245 ----
 >       DW_AT_src_coords = 0x2104,
 >       DW_AT_body_begin = 0x2105,
 >       DW_AT_body_end   = 0x2106,
 > +     DW_AT_vector     = 0x2107,
 >       /* VMS Extensions.  */
 >       DW_AT_VMS_rtnbeg_pd_address = 0x2201
 >     };
 > Index: gdb/dwarf2read.c
 > ===================================================================
 > RCS file: /cvs/uberbaum/gdb/dwarf2read.c,v
 > retrieving revision 1.48
 > diff -c -p -r1.48 dwarf2read.c
 > *** dwarf2read.c	2002/02/15 22:42:33	1.48
 > --- dwarf2read.c	2002/02/22 03:40:29
 > *************** dwarf_attr_name (register unsigned attr)
 > *** 5161,5166 ****
 > --- 5161,5168 ----
 >         return "DW_AT_body_begin";
 >       case DW_AT_body_end:
 >         return "DW_AT_body_end";
 > +     case DW_AT_vector:
 > +       return "DW_AT_vector";
 >       default:
 >         return "DW_AT_<unknown>";
 >       }
 > Index: include/elf/dwarf2.h
 > ===================================================================
 > RCS file: /cvs/uberbaum/include/elf/dwarf2.h,v
 > retrieving revision 1.8
 > diff -c -p -r1.8 dwarf2.h
 > *** dwarf2.h	2002/01/28 23:26:53	1.8
 > --- dwarf2.h	2002/02/22 03:40:29
 > *************** enum dwarf_attribute
 > *** 328,333 ****
 > --- 328,334 ----
 >       DW_AT_src_coords = 0x2104,
 >       DW_AT_body_begin = 0x2105,
 >       DW_AT_body_end   = 0x2106,
 > +     DW_AT_vector     = 0x2107,
 >       /* VMS Extensions.  */
 >       DW_AT_VMS_rtnbeg_pd_address = 0x2201
 >     };
 > Index: binutils/readelf.c
 > ===================================================================
 > RCS file: /cvs/uberbaum/binutils/readelf.c,v
 > retrieving revision 1.151
 > diff -c -p -r1.151 readelf.c
 > *** readelf.c	2002/02/13 18:14:42	1.151
 > --- readelf.c	2002/02/22 03:40:35
 > *************** get_AT_name (attribute)
 > *** 6455,6460 ****
 > --- 6455,6461 ----
 >       case DW_AT_src_coords:           return "DW_AT_src_coords";
 >       case DW_AT_body_begin:           return "DW_AT_body_begin";
 >       case DW_AT_body_end:             return "DW_AT_body_end";
 > +     case DW_AT_vector:               return "DW_AT_vector";
 >       default:
 >         {
 >   	static char buffer [100];

Re: rfc: DW_AT_vector

[Aldy Hernandez]

>  > of course, i'm willing to volunteer to code the DW_TAG_vector_type
>  > stuff, but only as a last and painful resort :).
>  > 
> 
> This would work, yes, even though I would personally prefer to have a
> new vector type. It represents more closely what these vectors really
> are.
> 

why can't you achieve the functionality with DW_AT_vector?  it has the
added benefit that other debuggers can read vectors quite intuitively.

aldy

Re: rfc: DW_AT_vector

[Richard Henderson]

On Thu, Feb 21, 2002 at 11:32:56PM -0500, Andrew Cagney wrote:
> Er, just an aside, any chance of GCC, GDB and BINUTILS agreeing to have 
> only one version of this file? :-)

That's way too progressive.

But if you insist on this sort of "outside the box"
thinking, I suppose you could move the file to the
top-level include directory.


r~

Re: rfc: DW_AT_vector

[Andrew Cagney]

> On Thu, Feb 21, 2002 at 11:32:56PM -0500, Andrew Cagney wrote:
> 
>> Er, just an aside, any chance of GCC, GDB and BINUTILS agreeing to have 
>> only one version of this file? :-)
> 
> 
> That's way too progressive.
> 
> But if you insist on this sort of "outside the box"
> thinking, I suppose you could move the file to the
> top-level include directory.

Can instead, GCC simply shadow bfd's existing include/elf directory 
structure?

enjoy,
Andrew

Re: rfc: DW_AT_vector

[Richard Henderson]

On Fri, Feb 22, 2002 at 09:26:09AM -0500, Andrew Cagney wrote:
> Can instead, GCC simply shadow bfd's existing include/elf directory 
> structure?

Sure.


r~