Re: [PATCH] move many gc hashtab to hash_table

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Mon, Oct 13, 2014 at 8:33 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Fri, Oct 10, 2014 at 1:44 PM, Trevor Saunders <tsaunders@mozilla.com> wrote:
>> On Fri, Oct 10, 2014 at 02:26:00PM -0600, Jeff Law wrote:
>>> On 10/06/14 21:24, tsaunders@mozilla.com wrote:
>>> >From: Trevor Saunders <tsaunders@mozilla.com>
>>> >
>>> >Hi,
>>> >
>>> >This changes almost all of the ggc htab that don't use the if_marked option to
>>> >be hash_tables.  I added a for_user gty attribute so that types could be used
>>> >from user marking routines without either using the mangled names or being
>>> >referenced in a template argument which was previously the only way to get
>>> >gengtype to emit user callable marking routines.
>>> >
>>> >bootstrapped + regtested on x86_64-unknown-linux-gnu, ran make all-gcc for
>>> >ppc64-linux, ppc64-apple-darwin, mips-elf, m32c-elf, mep-elf, and
>>> >x86_64-apple-darwin.  I haven't heard back yet on my request for a cfarm
>>> >account so I haven't tried bootstrapping on other archs, but more testing is
>>> >most welcome. Ok?
>>> >
>>> >Trev
>>> >
>>> >
>>> >gcc/
>>> >
>>> >* asan.c, cfgloop.c, cfgloop.h, cgraph.c, cgraph.h,
>>> >     config/darwin.c, config/m32c/m32c.c, config/mep/mep.c,
>>> >     config/mips/mips.c, config/rs6000/rs6000.c, dwarf2out.c,
>>> >     function.c, function.h, gimple-ssa.h, libfuncs.h, optabs.c,
>>> >     output.h, rtl.h, sese.c, symtab.c, tree-cfg.c, tree-dfa.c,
>>> >     tree-ssa.c, varasm.c: Use hash-table instead of hashtab.
>>> >     * doc/gty.texi (for_user): Document new option.
>>> >     * gengtype.c (create_user_defined_type): Don't try to get a struct for
>>> >     char.
>>> >     (walk_type): Don't error out on for_user option.
>>> >     (write_func_for_structure): Emit user marking routines if requested by
>>> >     for_user option.
>>> >     (write_local_func_for_structure): Likewise.
>>> >     (main): Mark types with for_user option as used.
>>> >     * ggc.h (gt_pch_nx): Add overload for unsigned int.
>>> >     * hash-map.h (hash_map::hash_entry::pch_nx_helper): AddOverloads.
>>> >     * hash-table.h (ggc_hasher): New struct.
>>> >     (hash_table::create_ggc): New function.
>>> >     (gt_pch_nx): New overload for hash_table.
>>> >
>>> >java/
>>> >
>>> >     * class.c, decl.c, except.c, expr.c, java-tree.h, lang.c: Use
>>> >     hash_table instead of hashtab.
>>> >
>>> >objc/
>>> >
>>> >     * objc-act.c: use hash_table instead of hashtab.
>>> >
>>> >cp/
>>> >
>>> >     * cp-gimplify.c, cp-tree.h, decl.c, mangle.c, name-lookup.c,
>>> >     pt.c, semantics.c, tree.c, typeck2.c: Use hash_table instead of
>>> >     hashtab.
>>> >
>>> >fortran/
>>> >
>>> >     * trans-decl.c, trans.c, trans.h: Use hash_table instead of hashtab.
>>> >
>>> >c-family/
>>> >
>>> >     * c-common.c: Use hash_table instead of hashtab.
>>> Whee, meaning we no longer have to pass through void * pointers and the
>>> visually ugly casting that requires in the callbacks?!?  Yea!
>>
>> yeah, you can even use an iterator now if you like
>>
>>> >-    ++*((unsigned HOST_WIDE_INT *) data);
>>> >+    ++* data;
>>> I think coding standards require no space here.  Obviously a nit.
>>
>> oops, definitely should be fixed :)
>>
>>> This is obviously a fairly mechanical change.  I did some spot checking and
>>> it looks good.  I don't expect any fallout.
>>
>> Well, its large enough I expect *something* but I doubt it'll be a big
>> deal.
>>
>>> Good for the trunk.
>>
>> thanks for the review, I know it was rather big.
>
> This breaks bootstrap on Linux/x86:
>
> https://gcc.gnu.org/ml/gcc-regression/2014-10/msg00237.html
>
> This patch fixes it.

This is what I checked in.


-- 
H.J.
---
Index: ChangeLog
===================================================================
--- ChangeLog (revision 216150)
+++ ChangeLog (working copy)
@@ -1,3 +1,9 @@
+2014-10-13  H.J. Lu  <hongjiu.lu@intel.com>
+
+ * mangle.c (mangle_conv_op_name_for_type): Cast elements to
+ unsigned long.
+ (print_template_statistics): Cast size and elements to long.
+
 2014-10-12  Trevor Saunders  <tsaunders@mozilla.com>

  * cp-gimplify.c, cp-tree.h, decl.c, mangle.c, name-lookup.c,
Index: mangle.c
===================================================================
--- mangle.c (revision 216150)
+++ mangle.c (working copy)
@@ -3744,7 +3744,8 @@ mangle_conv_op_name_for_type (const tree
       char buffer[64];

        /* Create a unique name corresponding to TYPE.  */
-      sprintf (buffer, "operator %lu", conv_type_names->elements ());
+      sprintf (buffer, "operator %lu",
+       (unsigned long) conv_type_names->elements ());
       identifier = get_identifier (buffer);
       *slot = identifier;

Index: pt.c
===================================================================
--- pt.c (revision 216150)
+++ pt.c (working copy)
@@ -22324,12 +22324,12 @@ void
 print_template_statistics (void)
 {
   fprintf (stderr, "decl_specializations: size %ld, %ld elements, "
-   "%f collisions\n", decl_specializations->size (),
-   decl_specializations->elements (),
+   "%f collisions\n", (long) decl_specializations->size (),
+   (long) decl_specializations->elements (),
    decl_specializations->collisions ());
   fprintf (stderr, "type_specializations: size %ld, %ld elements, "
-   "%f collisions\n", type_specializations->size (),
-   type_specializations->elements (),
+   "%f collisions\n", (long) type_specializations->size (),
+   (long) type_specializations->elements (),
    type_specializations->collisions ());
 }