* [PATCHx2,java] Fix java tree checking assert and
@ 2010-12-11 13:38 Dave Korn
2010-12-11 13:41 ` ... and tree ssa loop prefetch pass bad gimple creation, I meant to say... [was [PATCHx2,java] Fix java tree checking assert and] Dave Korn
2010-12-13 12:43 ` [PATCHx2,java] Fix java tree checking assert and Andrew Haley
0 siblings, 2 replies; 4+ messages in thread
From: Dave Korn @ 2010-12-11 13:38 UTC (permalink / raw)
To: Java Patches, GCC Patches
[-- Attachment #1: Type: text/plain, Size: 1678 bytes --]
Hello lists,
The attached patches fix a couple of problems(*) exposed by the tree ssa
loop prefetch pass in recent trunk builds of jc1:
- loop prefetch generates an invalid gimple call to __builtin_prefetch(),
because it uses the integer_three_node global tree; because java doesn't take
advantage of the core build_common_tree_nodes_2() routine but sets up all the
nodes itself, this was missing any initialisation - integer_three_node is a
recent addition to the core, and the prefetch pass was adjusted to use it, but
java was inadvertently omitted.
- turning on tree dump can cause a crash when it tries to generate a name for
a nested function, because put_decl_node doesn't understand that the context
of a function_decl can be another function_decl.
The first I fixed by adding the missing initialisation - long-term I guess
it would be better to see if java couldn't use build_common_tree_nodes_2() in
some way, but that's not stage3 material. The second I fixed by making it
treat the nested function as just another level of scope, adding another
period-separated segment to the generated name.
java/ChangeLog:
* decl.c (java_init_decl_processing): Initialise integer_three_node.
* lang.c (put_decl_node): Handle nested function decls.
I verified that this fixed the failing compile that I had seen in the
testsuite, now I'm taking the combined patchset for a full bootstrap and test
cycle on i686-pc-cygwin. OK if that completes without regressions?
cheers,
DaveK
--
(*) - http://gcc.gnu.org/ml/gcc/2010-12/msg00308.html
http://gcc.gnu.org/ml/gcc/2010-12/msg00315.html
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29710
[-- Attachment #2: java-tree-dump-checking-fix.diff --]
[-- Type: text/x-c, Size: 609 bytes --]
Index: gcc/java/lang.c
===================================================================
--- gcc/java/lang.c (revision 167688)
+++ gcc/java/lang.c (working copy)
@@ -402,7 +402,9 @@ put_decl_node (tree node, int verbosity)
if verbosity is higher than 1. */
&& verbosity >= 1)
{
- put_decl_node (TYPE_NAME (DECL_CONTEXT (node)),
+ put_decl_node (TREE_CODE (DECL_CONTEXT (node)) == FUNCTION_DECL
+ ? DECL_CONTEXT (node)
+ : TYPE_NAME (DECL_CONTEXT (node)),
verbosity);
put_decl_string (".", 1);
}
[-- Attachment #3: java-loop-prefetch-crash-fix.diff --]
[-- Type: text/x-c, Size: 542 bytes --]
Index: gcc/java/decl.c
===================================================================
--- gcc/java/decl.c (revision 167688)
+++ gcc/java/decl.c (working copy)
@@ -614,6 +614,7 @@ java_init_decl_processing (void)
integer_zero_node = build_int_cst (NULL_TREE, 0);
integer_one_node = build_int_cst (NULL_TREE, 1);
integer_two_node = build_int_cst (NULL_TREE, 2);
+ integer_three_node = build_int_cst (NULL_TREE, 3);
integer_four_node = build_int_cst (NULL_TREE, 4);
integer_minus_one_node = build_int_cst (NULL_TREE, -1);
^ permalink raw reply [flat|nested] 4+ messages in thread
* ... and tree ssa loop prefetch pass bad gimple creation, I meant to say... [was [PATCHx2,java] Fix java tree checking assert and]
2010-12-11 13:38 [PATCHx2,java] Fix java tree checking assert and Dave Korn
@ 2010-12-11 13:41 ` Dave Korn
2010-12-13 12:43 ` [PATCHx2,java] Fix java tree checking assert and Andrew Haley
1 sibling, 0 replies; 4+ messages in thread
From: Dave Korn @ 2010-12-11 13:41 UTC (permalink / raw)
To: Java Patches, GCC Patches
ibid.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCHx2,java] Fix java tree checking assert and
2010-12-11 13:38 [PATCHx2,java] Fix java tree checking assert and Dave Korn
2010-12-11 13:41 ` ... and tree ssa loop prefetch pass bad gimple creation, I meant to say... [was [PATCHx2,java] Fix java tree checking assert and] Dave Korn
@ 2010-12-13 12:43 ` Andrew Haley
2010-12-15 11:05 ` Dave Korn
1 sibling, 1 reply; 4+ messages in thread
From: Andrew Haley @ 2010-12-13 12:43 UTC (permalink / raw)
To: gcc-patches
On 12/11/2010 12:10 PM, Dave Korn wrote:
>
> Hello lists,
>
>
> The attached patches fix a couple of problems(*) exposed by the tree ssa
> loop prefetch pass in recent trunk builds of jc1:
>
> - loop prefetch generates an invalid gimple call to __builtin_prefetch(),
> because it uses the integer_three_node global tree; because java doesn't take
> advantage of the core build_common_tree_nodes_2() routine but sets up all the
> nodes itself, this was missing any initialisation - integer_three_node is a
> recent addition to the core, and the prefetch pass was adjusted to use it, but
> java was inadvertently omitted.
>
> - turning on tree dump can cause a crash when it tries to generate a name for
> a nested function, because put_decl_node doesn't understand that the context
> of a function_decl can be another function_decl.
>
> The first I fixed by adding the missing initialisation - long-term I guess
> it would be better to see if java couldn't use build_common_tree_nodes_2() in
> some way, but that's not stage3 material. The second I fixed by making it
> treat the nested function as just another level of scope, adding another
> period-separated segment to the generated name.
>
> java/ChangeLog:
>
> * decl.c (java_init_decl_processing): Initialise integer_three_node.
> * lang.c (put_decl_node): Handle nested function decls.
>
> I verified that this fixed the failing compile that I had seen in the
> testsuite, now I'm taking the combined patchset for a full bootstrap and test
> cycle on i686-pc-cygwin. OK if that completes without regressions?
Yes, that looks OK.
Andrew.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCHx2,java] Fix java tree checking assert and
2010-12-13 12:43 ` [PATCHx2,java] Fix java tree checking assert and Andrew Haley
@ 2010-12-15 11:05 ` Dave Korn
0 siblings, 0 replies; 4+ messages in thread
From: Dave Korn @ 2010-12-15 11:05 UTC (permalink / raw)
To: Andrew Haley; +Cc: gcc-patches
On 13/12/2010 10:25, Andrew Haley wrote:
> On 12/11/2010 12:10 PM, Dave Korn wrote:
>> java/ChangeLog:
>>
>> * decl.c (java_init_decl_processing): Initialise integer_three_node.
>> * lang.c (put_decl_node): Handle nested function decls.
>>
>> I verified that this fixed the failing compile that I had seen in the
>> testsuite, now I'm taking the combined patchset for a full bootstrap and test
>> cycle on i686-pc-cygwin. OK if that completes without regressions?
>
> Yes, that looks OK.
>
> Andrew.
Committed revision 167849.
cheers,
DaveK
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2010-12-15 10:00 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-12-11 13:38 [PATCHx2,java] Fix java tree checking assert and Dave Korn
2010-12-11 13:41 ` ... and tree ssa loop prefetch pass bad gimple creation, I meant to say... [was [PATCHx2,java] Fix java tree checking assert and] Dave Korn
2010-12-13 12:43 ` [PATCHx2,java] Fix java tree checking assert and Andrew Haley
2010-12-15 11:05 ` Dave Korn
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).