Wolfgang Bangerth | 1 Jun 01:11 2003

[Patch, committed] sourcebuild.texi: remove two references to GNATS

I just committed the following. The reference to gnats.html still needs to 
be purged, but let's be incremental.

Tested on x86-linux-gnu.


Index: sourcebuild.texi
RCS file: /cvs/gcc/gcc/gcc/doc/sourcebuild.texi,v
retrieving revision 1.19
diff -c -r1.19 sourcebuild.texi
*** sourcebuild.texi	5 May 2003 21:59:35 -0000	1.19
--- sourcebuild.texi	31 May 2003 23:08:09 -0000
*** 466,475 ****

   <at> itemize  <at> bullet
   <at> item
! At least one GNATS category for bugs in that front end and runtime
  libraries.  This category needs to be mentioned in
   <at> file{gcc/gccbug.in}, and in  <at> file{gnats.html} on the GCC web site, as
! well as being added to the GNATS database.
   <at> item
  Normally, one or more maintainers of that front end listed in
   <at> file{MAINTAINERS}.
--- 466,475 ----

   <at> itemize  <at> bullet
(Continue reading)

Joseph S. Myers | 1 Jun 01:12 2003

Re: Patch to check asm_fprintf format specifiers [take 2]

On Sat, 31 May 2003, Kaveh R. Ghazi wrote:

> built.  So I lean towards having more implemented rather than less.

Whereas I lean to not having formats that are unlikely to be use use in
asm_fprintf - that is, floating point and %p - and starting the checking
off restrictive; it can be expanded later if necessary and it's an
inevitable part of this checking approach that it will only work well when
building with the same compiler version as being built.


Joseph S. Myers
jsm28 <at> cam.ac.uk

Aldy Hernandez | 1 Jun 01:44 2003

Re: [janitor] toplev.c cleanups

On Thursday, May 29, 2003, at 06:47  PM, Geoff Keating wrote:

> This is kind of hard to review properly, but I like the idea, I don't
> think the patch is making anything worse, and I'm going to trust that
> you didn't accidentally delete any important optimisation passes :-).
> So the patch is OK.

Yes, I know this is very hard to review properly-- probably more time 
consuming than hard, that's why I didn't touch any logic.  If I need to 
do that, I'll do in subsequent patches that can be more straightforward 
to review.

Further testing revealed a small thinko in the last patch.  Below is 
the patch I'm committing.

Tested on x86 linux.


2003-05-31  Aldy Hernandez  <aldyh <at> redhat.com>

	* toplev.c (botch): Remove.
	(do_abort): Remove.
	(set_Wunused): Comment.
	(set_Wextra): Comment.
	Remove ^L's.
	(rest_of_compilation): Factor out common code into functions.
	(rest_of_handle_inlining): New.
(Continue reading)

law | 1 Jun 04:04 2003

Re: [tree-ssa] New tests

In message <1054356456.21692.408.camel <at> frodo.toronto.redhat.com>, Diego Novillo
 >On Fri, 2003-05-30 at 19:18, law <at> redhat.com wrote:
 >> In message <1054335774.21692.385.camel <at> frodo.toronto.redhat.com>, Diego Nov
 >>  writes:
 >>  >On Fri, 2003-05-30 at 18:09, law <at> redhat.com wrote:
 >>  >
 >>  >> The first shows a case where Diego's hack to avoid overlapping
 >>  >> lifetimes fails, thus leading to an abort during the out of SSA
 >>  >> translation.
 >>  >> 
 >>  >Really?  How come var_is_live is never called on this test case?  The
 >>  >var_is_live kludge is only called when we're doing redundant assignments
 >>  >and copy-prop while renaming.  In fact, the bug was that we are *not*
 >>  >calling the hack.
 >> No, we're calling it just fine.
 >You must have something different in your tree.  I'm working with a
 >pristine copy and we are NOT calling var_is_live for this testcase!
We're looking at different testcases -- I was referring back to the 
java one, which I consider more important.

There's a variety of ways to fix the one derived from rs6000.c, the
right one is to fix the hashing routines, but after a day poking at
them, I've come to the conclusion that's more difficult than it may
at first appear.

Just for fun, insert this code into avail_expr_eq just before it's
about to return true (both cases).  Then do some builds and watch
(Continue reading)

law | 1 Jun 04:05 2003

Re: [tree-ssa] Removing redundant loads

In message <wvlbrxk0vve.fsf <at> prospero.boston.redhat.com>, Jason Merrill writes:
 >On Fri, 30 May 2003 18:34:45 -0600, law <at> redhat.com wrote:
 >> In message <wvln0h40yjy.fsf <at> prospero.boston.redhat.com>, Jason Merrill writ
 >>  >> For example, if you pass it this tree for both operands, you'll get a
 >>  >> false back from operand_b because this expression has tree_side_effects
 >>  >> set.
 >>  >
 >>  >Why does that expression have TREE_SIDE_EFFECTS set?
 >> Because we unconditionally set TREE_SIDE_EFFECTS on STMT_EXPRs.  That
 >> gets propagated up to the COMPONENT_REFs.
 >The gimplifier should have cleared that flag while processing the
 >COMPONENT_REF.  I'll check it out.
Thanks.   FWIW, I'm offline all next week, so I won't be able to poke
at anything you may come up with until June 10/11.


Daniel Berlin | 1 Jun 07:55 2003

[tree-ssa]: Fix PTA for memory tag changes

With this patch, C, C++, Java, Fortran, and Objective C bootstrap with PTA
turned on by default.
It does increase the runtimes of CCP and DCE slightly in large cases,
since they do more work now.

There are a few C++ files where it is a little slower than i'd like (it
takes 8% of the compile-time, which in these cases is 1 second out of 12),
but this is actually due to the overhead of calls to decl_function_context.

2003-06-01  Daniel Berlin  <dberlin <at> dberlin.org

	* tree-alias-common.c: Remove setting of DECL_CONTEXT in temp vars,
	it's done in create_tmp_alias_var for us.
	(ptr_may_alias_var): Check if the variables are memory tags, and get
	the associated pointers if they are.

Index: tree-alias-common.c
RCS file: /cvs/gcc/gcc/gcc/Attic/tree-alias-common.c,v
retrieving revision
diff -u -3 -p -r1.1.2.26 tree-alias-common.c
--- tree-alias-common.c	19 May 2003 21:37:58 -0000
+++ tree-alias-common.c	1 Jun 2003 05:28:23 -0000
 <at>  <at>  -288,7 +288,6  <at>  <at>  get_alias_var (expr)
 	   expression. */
 	tree temp = create_tmp_alias_var (void_type_node, "aliastmp");
 	alias_typevar tempvar;
-	DECL_CONTEXT (temp) = currptadecl;
 	tempvar = current_alias_ops->add_var (current_alias_ops,
(Continue reading)

Joel Brobecker | 1 Jun 09:01 2003

[RFA] Fix libiberty compilation failure on systems running AiX 4.x


The following change in libiberty works for Aix 5.x, but breaks the
build on AiX 4.x:

        2003-02-22  Kaveh R. Ghazi  <ghazi <at> caip.rutgers.edu>

                * configure.in: Check for sys/systemcfg.h and
                * physmem.c: Add support for AIX.  Tweek formatting as per
                upstream coreutils beta.

This is because the _system_configuration structure does not contain
a field named "physmem", which is used in physmem.c:

          /* This works on AIX.  */
          return _system_configuration.physmem;

For the record, I have pasted at then end of this message the definition
of this varible on a machine running 4.3.2 .

Given that the only field in _system_configuration used is "physmem",
the attached patch seemed to be the simplest change to get back the
build on AiX 4.x machines. I essentially deactivated the test for
_system_configuration on machines running AiX 4.x.

2003-06-01  J. Brobecker  <brobecker <at> gnat.com>

(Continue reading)

Andreas Tobler | 1 Jun 09:26 2003

Re: [RFA/C] Break out C options into a separate file

Neil Booth wrote:

> Index: opts.sh
> ===================================================================
> RCS file: opts.sh
> diff -N opts.sh
> --- /dev/null	1 Jan 1970 00:00:00 -0000
> +++ opts.sh	22 May 2003 19:51:37 -0000

> +AWK=/usr/bin/awk
> +SORT=/usr/bin/sort

Just for the record, my ppc linux box has sort under /bin. So a 
bootstrap fails here unless I symlink around.

Would it be possible to check for awk/sort and take the corresponding one?


Andreas Jaeger | 1 Jun 10:14 2003

[Janitor] Unused macros in dwarfout.c

This patch removes unused macros in dwarfout.c.

Ok to commit if bootstrap passes on i686-linux-gnu?


2003-06-01  Andreas Jaeger  <aj <at> suse.de>

	* dwarfout.c: Remove USE_ORDERING_ATTRIBUTE,
	(ordering_attribute): Removed.
	(output_array_type_die): Remove USE_ORDERING_ATTRIBUTE dependend
	(src_coords_attribute): Removed.
	(name_and_src_coords_attributes): Remove DWARF_DECL_COORDINATES usage.
	(dwarfout_init): Remove DWARF_TIMESTAMPS usage.


Index: gcc/dwarfout.c
--- gcc/dwarfout.c	13 May 2003 18:06:45 -0000	1.123
+++ gcc/dwarfout.c	1 Jun 2003 08:02:10 -0000
 <at>  <at>  -338,13 +338,9  <at>  <at>  Software Foundation, 59 Temple Place - S
 		 might be configured to generate a default value of 0xffffffff
 		 in this field (in cases where it is desired to have object
 		 files created at different times from identical source files
-		 be byte-for-byte identical).  By default, these timestamps
(Continue reading)

Andreas Jaeger | 1 Jun 12:01 2003

[Janitor] Remove DONT_REDUCE_ADDR from loop.c

The macro is not defined anywhere, I'm bootstrapping on i686-linux-gnu

Ok to commit?


2003-06-01  Andreas Jaeger  <aj <at> suse.de>

	* loop.c (check_insn_for_givs): Remove DONT_REDUCE_ADDR macro.

Index: gcc/loop.c
--- gcc/loop.c	31 May 2003 13:23:28 -0000	1.457
+++ gcc/loop.c	1 Jun 2003 09:49:58 -0000
 <at>  <at>  -5578,14 +5578,12  <at>  <at>  check_insn_for_givs (loop, p, not_every_

   /* Look for givs which are memory addresses.  */
   /* This resulted in worse code on a VAX 8600.  I wonder if it
      still does.  */
   if (GET_CODE (p) == INSN)
     find_mem_givs (loop, PATTERN (p), p, not_every_iteration,

   /* Update the status of whether giv can derive other givs.  This can
(Continue reading)