Hello,
here is updated patch I intend to commit after bootstrap/regtest on
x86_64-linux and some additional testing on Chromium/libreoffice (it works
on Firefox).
I turned the cache into class. I tried to avoid global variable, but
ended up with the pointer to current cache because the use in ipa-
> On 2015.03.27 at 00:46 +0100, Jan Hubicka wrote:
> > > On 26 March 2015 at 22:12, Jan Hubicka wrote:
> > > After this fix, I can see build failures in glibc:
> > > key_call.c:574:1: internal compiler error: in inline_call, at
> > > ipa-inline-transform.c:386
> >
> > Can you please send me prepr
On 2015.03.27 at 00:46 +0100, Jan Hubicka wrote:
> > On 26 March 2015 at 22:12, Jan Hubicka wrote:
> > After this fix, I can see build failures in glibc:
> > key_call.c:574:1: internal compiler error: in inline_call, at
> > ipa-inline-transform.c:386
>
> Can you please send me preprocessed testca
Hi,
I reproduced the problem of Firefox LTO (though small testcase would be welcome)
The problem is in a way we walk edges that now can be turned from indirect to
direct.
I am working on a fix.
Honza
On 3/27/15 08:05, Chen Gang wrote:
> For misalignment memory access, c6x gcc will cause issue, so need remove
> "clobber (match_scratch ...)" which will be symmetric with "movmisalign
> _load", then pass compiling and generate correct assembly code.
>
> The related issue:
>
> [root@localhost
For misalignment memory access, c6x gcc will cause issue, so need remove
"clobber (match_scratch ...)" which will be symmetric with "movmisalign
_load", then pass compiling and generate correct assembly code.
The related issue:
[root@localhost c6x]# cat test.i
int oxu_driver_init(void)
{
The ubsan pass relies on rebuilding the cgraph edges in case it adds any
new statements. Without adding the new callgraph edges the IPA inline
analysis might be upset and respond with ICE. There are more ways how to
fix this: we could schedule the pass_rebuild_cgraph_edges right after the
ubsan p
> On 26 March 2015 at 22:12, Jan Hubicka wrote:
> > Hi,
> > this patch missed hunk adding CIF code that I commited now
> > * cif-code.def (CILK_SPAWN): New code
>
> Hi,
> After this fix, I can see build failures in glibc:
> key_call.c:574:1: internal compiler error: in inline_call, at
> i
Jan,
It appears that
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61250 is due to a problem
in linemap_macro_map_lookup. It is hard to debug as the pch test
harness doesn't produce a simple logged compile failure to re-execute
but I can look at it from a core file in gdb. Unfortunately your
pr
I am looking at ways of dynamically realigning the runtime stack in GCC.
Currently it looks like x86 is the only architecture that supports this.
The issue that I am trying to address is MSA registers on MIPS. The O32
MIPS ABI specifies an 8 byte aligned stack but MSA registers should be 16
byte
On 26 March 2015 at 22:12, Jan Hubicka wrote:
> Hi,
> this patch missed hunk adding CIF code that I commited now
> * cif-code.def (CILK_SPAWN): New code
Hi,
After this fix, I can see build failures in glibc:
key_call.c:574:1: internal compiler error: in inline_call, at
ipa-inline-transfor
On Thu, Mar 26, 2015 at 11:25 AM, Bill Schmidt
wrote:
>> This is a follow-up to
>> https://gcc.gnu.org/ml/gcc-patches/2015-03/msg01310.html, which
>> backported the POWER-specific little-endian swap optimization pass to
>> the 4.9 branch. We also need to backport this to the 4.8 branch. This
>>
> For now it will blow up with a GCC error in the back end, which does
> not know about this attribute yet.
I guess that we don't want to have a half-backed implementation so I have
installed the gigi part. Tested on x86_64-suse-linux.
2015-03-26 Eric Botcazou
* gcc-interface/trans
Hi,
this patch missed hunk adding CIF code that I commited now
* cif-code.def (CILK_SPAWN): New code.
--- trunk/gcc/cif-code.def 2015/03/26 20:37:53 221709
+++ trunk/gcc/cif-code.def 2015/03/26 21:10:28 221710
@@ -124,6 +124,10 @@
DEFCIFCODE(ATTRIBUTE_MISMATCH, CIF_FINAL
On 03/26/2015 04:38 PM, Sandra Loosemore wrote:
On 03/26/2015 01:25 PM, Vladimir Makarov wrote:
Hi, I neglected to write about RA changes for the previous releases
and people asked me to write about RA changes for GCC-5. So here is what
I'd like to add to gcc-5/changes.html. I'll do it tomor
Hi!
On Mon, Mar 09, 2015 at 01:19:38PM +0100, Dominik Vogt wrote:
> @@ -11368,6 +11349,7 @@ static void
> s390_reorg (void)
> {
>bool pool_overflow = false;
> + int hw_before, hw_after;
>
>/* Make sure all splits have been performed; splits after
> machine_dependent_reorg might
Hello!
Attached patch prevents non-matching input and output memory operands
for various vextract insn patterns. If an optimization pass propagates
non-matching memory operand to the pattern, then RA can't satisfy
matching "0" constraint.
2015-03-26 Uros Bizjak
PR target/65561
* confi
On Thu, Mar 26, 2015 at 13:09:19 +0100, Jakub Jelinek wrote:
> On Mon, Mar 23, 2015 at 10:44:39PM +0300, Ilya Verbin wrote:
> > If it is too late for such global changes (rework initialization in libgomp,
> > change mic and ptx plugins), then here is a small workaround patch to fix
> > offloading f
On 03/26/2015 01:25 PM, Vladimir Makarov wrote:
Hi, I neglected to write about RA changes for the previous releases
and people asked me to write about RA changes for GCC-5. So here is what
I'd like to add to gcc-5/changes.html. I'll do it tomorrow. So any
comments will be appreciated.
Index
Hi,
this patch makes the verifier to not check caller frequency. This makes the
check to be local for function being called on and allows to avoid fixup_cfg
in the patch for nothrow.
The patch also resolves old FIXME about unreachable edges that found two
latent bugs - one in inliner fixed by pre
We have a couple of bugs where durations with unsigned representations
result in negative or huge time_t values, which result in not sleeping
at all or sleeping for billions of years (or merely for decades if you
have a 32-bit time_t).
This change simply returns early for time points in the past
> + In GCC-4.9 and earlier releases PIC hard register was fixed
> + and was not used for other purposes when PIC code was
> + generated. Reuse of PIC hard register was implemented in RA
> + for GCC-5.0. It improves generated PIC code performance as
> + mor
Applied, thanks.
In future please CC me on C++ patches so that I see them faster. And
feel free to ping as often as once a week.
Jason
Hi, I neglected to write about RA changes for the previous releases
and people asked me to write about RA changes for GCC-5. So here is what
I'd like to add to gcc-5/changes.html. I'll do it tomorrow. So any
comments will be appreciated.
Thanks.
Index: changes.html
==
On March 26, 2015 6:01:21 PM GMT+01:00, Jan Hubicka wrote:
>Hi,
>this is updated patch I am testing (along with the fixes to verifiew
>and
>inliner I will commit independently).
OK.
Thanks,
Richard.
> * passes.def: Add pass_nothrow.
> * ipa-pure-const.c: (pass_data_nothrow): New.
>
On March 26, 2015 5:50:13 PM GMT+01:00, Jakub Jelinek wrote:
>Hi!
>
>This patch fixes
>FAIL: 26_numerics/complex/50880.cc execution test
>on powerpc64le-linux (and presumably arm*-linux-gnueabi) by basically
>reverting the r220390 change for the case when ref->type is not
>INTEGRAL_TYPE_P - otherw
Hi,
this patch fixes several issues in the inliner. First is the fact that
edge_set_predicate does not mark indirect and speculative edge as unreachable
correctly (found by my updated verification patch).
Other issue is that can_inline_edge_p was wrong in several ways. It did
use DECL_STRUCT_FUNC
It's simple enough to make potential_constant_expression_1 handle MEM_REF.
Tested x86_64-pc-linux-gnu, applying to trunk.
commit c25738941b9d32a9653bca90a896407667685d91
Author: Jason Merrill
Date: Thu Mar 26 13:11:13 2015 -0400
PR c++/65525
* constexpr.c (potential_constant_express
Due to differences between the master Go library and the Go 1.4 Go
library used by gccgo, I left out a key part of the patch to support
32-bit PPC relocations in the debug/elf package, which is used by cgo.
Matthias reported this as PR 65587. This patch fixes the problem.
Bootstrapped and ran Go t
On 03/26/2015 12:16 PM, Aldy Hernandez wrote:
Is there a reason we pass a collapsible option to add_subscript_info? It
seems like it's value is always !is_ada(). Can we just get rid of the
argument, and calculate its value within the function, or was it there
for some future flexibility?
On tr
Hi,
this is updated patch I am testing (along with the fixes to verifiew and
inliner I will commit independently).
* passes.def: Add pass_nothrow.
* ipa-pure-const.c: (pass_data_nothrow): New.
(pass_nothrow): New.
(pass_nothrow::execute): New.
(make_pass_not
Hi!
This patch fixes
FAIL: 26_numerics/complex/50880.cc execution test
on powerpc64le-linux (and presumably arm*-linux-gnueabi) by basically
reverting the r220390 change for the case when ref->type is not
INTEGRAL_TYPE_P - otherwise TYPE_PRECISION on say complex type or
vector type means very diff
> >>
> >> So please re-check with the above execute_fixup_cfg removed
> >> (your simple patch doesn't handle direct recursion optimistically).
> >
> >Ah, you are right. I will add it - that is easy enoug to check.
>
> But then you need to do the fixup or remember the direct recursion sites and
>
On March 26, 2015 3:41:12 PM GMT+01:00, Jan Hubicka wrote:
>>
>> Does that work on callgraph cycles?
>
>Yes. Because ...
>> > +
>> > + node->set_nothrow_flag (true);
>> > + if (dump_file)
>> > +fprintf (dump_file, "Function found to be nothrow: %s\n",
>> > + current_function_name ());
Hi Jason.
Is there a reason we pass a collapsible option to add_subscript_info?
It seems like it's value is always !is_ada(). Can we just get rid of
the argument, and calculate its value within the function, or was it
there for some future flexibility?
I'd like to queue this patch for GCC 6
The following GDB test is failing because typedef's are being marked as
completed after their type DIEs are generated in early debug:
FAIL: gdb.base/vla-ptr.exp: print td_vla
We need to revisit the partially created DIE in late debug and fill in
the bound information when available. W
On 03/25/2015 05:09 PM, Sebastian Pop wrote:
Specifically, it seems to me that copy_bbs should be refactored into
copy_bbs and copy_bbs_for_threading or somesuch. Where those
routines call into refactored common subroutines, but obviously
handle wiring up the outgoing edges from the copied block
On 03/25/2015 07:07 PM, Jason Merrill wrote:
On 03/25/2015 05:05 PM, Aldy Hernandez wrote:
Or we could cheat and just remove them as mainline does, but only when
reusing a declaration (as in the attached patch).
This seems right to me.
Jason
Ok thanks, committed.
Aldy
On 04/03/15 11:13, Alex Velenko wrote:
Hi,
This patch fixes arm pr45701 scan assembly tests. Those test register r3 being
used to maintain stack double word alignment. Recent optimizations reduced
number of local variables needed in those tests, removing necessity to push r3.
Testcase fixed by a
Oops. Fixed post title.
On Thu, 2015-03-26 at 10:23 -0500, Bill Schmidt wrote:
> Hi,
>
> This is a follow-up to
> https://gcc.gnu.org/ml/gcc-patches/2015-03/msg01310.html, which
> backported the POWER-specific little-endian swap optimization pass to
> the 4.9 branch. We also need to backport th
Hi,
This is a follow-up to
https://gcc.gnu.org/ml/gcc-patches/2015-03/msg01310.html, which
backported the POWER-specific little-endian swap optimization pass to
the 4.9 branch. We also need to backport this to the 4.8 branch. This
patch does that.
The patch is very similar to the 4.9 backport,
On 02 Feb 13:05, Jakub Jelinek wrote:
> On Tue, Jul 29, 2014 at 06:45:01PM +0400, Ilya Tocar wrote:
> > Hi,
> >
> > As discussed here in https://gcc.gnu.org/ml/gcc/2014-01/msg00189.html
> > Gcc should complain about pragma omp declare target without
> > corresponding pragma omp end declare target.
>
> Does that work on callgraph cycles?
Yes. Because ...
> > +
> > + node->set_nothrow_flag (true);
> > + if (dump_file)
> > +fprintf (dump_file, "Function found to be nothrow: %s\n",
> > +current_function_name ());
> > + return execute_fixup_cfg ();
>
> Err - this doesn't make se
> >
> > + lto_apply_location_cache ();
> > +
>
> Please add a comment here. It might be possible to write a
Will do.
> lto_location_cache_decl_source_location (TYPE_NAME (t1)) that
> looks if the location is in the cache and otherwise falls back to
> DECL_SOURCE_LOCATION. That would be kind
This fixes PR6. As I already thought we should guard the
noreturn-shouldn't-have-a-lhs checking properly so it doesn't
apply to not yet discovered noreturn calls.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2015-03-26 Richard Biener
PR middle-end/6555
On Thu, Mar 26, 2015 at 10:00 AM, Bill Schmidt
wrote:
> Hi,
>
> While working on a backport, I discovered a bug that only exists in the
> 4.8 branch (the buggy code was rewritten with the direct move additions
> in 4.9). A load-and-extract of the first doubleword in memory can
> always be done wi
Hi,
While working on a backport, I discovered a bug that only exists in the
4.8 branch (the buggy code was rewritten with the direct move additions
in 4.9). A load-and-extract of the first doubleword in memory can
always be done with the lxsd[u]x instruction. In 4.8 I wrongly added
code to use t
Hi!
Rather than adding 130 or so missing headers to PLUGIN_HEADERS,
this just adds all *.h and *.def files from gcc/ directory.
For files from different directories, one has to still add them
into PLUGIN_HEADERS.
Tested on x86_64-linux and i686-linux, ok for trunk?
2015-03-26 Jakub Jelinek
On 25/03/15 12:04 -0700, Richard Henderson wrote:
On 03/25/2015 11:49 AM, Jonathan Wakely wrote:
On 25/03/15 11:36 -0700, Richard Henderson wrote:
On 03/25/2015 09:22 AM, Jonathan Wakely wrote:
On 25/03/15 11:39 -0700, Richard Henderson wrote:
On 03/25/2015 09:22 AM, Jonathan Wakely wrote:
+
On 03/26/2015 02:50 PM, Jakub Jelinek wrote:
On Thu, Mar 26, 2015 at 09:53:03AM +0300, Marat Zakirov wrote:
gcc/ChangeLog:
2015-03-12 Marat Zakirov
* asan.c (asan_emit_stack_protection): Support for misalign accesses.
(asan_expand_check_ifn): Likewise.
* common.opt
On Mon, Mar 23, 2015 at 10:44:39PM +0300, Ilya Verbin wrote:
> If it is too late for such global changes (rework initialization in libgomp,
> change mic and ptx plugins), then here is a small workaround patch to fix
> offloading from libraries. Likely, it will not affect OpenACC programs with
> o
On 12/02/15 13:23 -0800, Richard Henderson wrote:
When we fixed PR54005, making sure that atomic_is_lock_free returns the same
value for all objects of a given type, we probably should have changed the
interface so that we would pass size and alignment rather than size and object
pointer.
Instea
On Thu, Mar 26, 2015 at 09:53:03AM +0300, Marat Zakirov wrote:
> gcc/ChangeLog:
>
> 2015-03-12 Marat Zakirov
>
> * asan.c (asan_emit_stack_protection): Support for misalign accesses.
> (asan_expand_check_ifn): Likewise.
> * common.opt: New flag -fasan-catch-misaligned.
>
On 15/03/15 23:27 -0700, Tim Shen wrote:
All it needs is something like:
enum syntax_option_type : unsigned { };
constexpr syntax_option_type icase = ...;
Acknowledged.
Here's the simple version of it.
OK for trunk, thanks.
So I've dug into this a bit further, as follows.
Firstly, changing the test (without -O) to use an 'i' constraint, fixes the test
(however, an "i" constraint is not correct for the instructions/asm where the
"S" constraint is used, e.g. in the Linux kernel). This is because
parse_input_constra
On Thu, 26 Mar 2015, Jakub Jelinek wrote:
> Hi!
>
> As discussed earlier, var-tracking assumes no pseudos are in the IL
> at the point it is run, which isn't the case for nvptx.
> While it would be also reasonable to kill debug stmts earlier or
> disable var-tracking, making it clear that var-tra
Hi!
As discussed earlier, var-tracking assumes no pseudos are in the IL
at the point it is run, which isn't the case for nvptx.
While it would be also reasonable to kill debug stmts earlier or
disable var-tracking, making it clear that var-tracking can't handle this
case is IMHO desirable too.
Ok
On Thu, 26 Mar 2015, Jakub Jelinek wrote:
> On Thu, Mar 26, 2015 at 10:32:18AM +0100, Richard Biener wrote:
> > I think I simply didn't want to change more testcases at that point,
> > but I can't see how followup passes at the very point wouldn't
> > clean things up very quickly (via fre or copyp
Hi!
On Mon, 23 Mar 2015 22:44:39 +0300, Ilya Verbin wrote:
> On Mon, Mar 09, 2015 at 14:45:55 +, Julian Brown wrote:
> > On Fri, 6 Mar 2015 17:01:13 +0300
> > Ilya Verbin wrote:
> >
> > > On Thu, Feb 26, 2015 at 20:25:11 +0300, Ilya Verbin wrote:
> > > > On Wed, Feb 25, 2015 at 10:36:08 +01
> [1. text/x-patch; sh_multilib_sh1_sh2a_ml_1.patch]
This works, thanks. I see the following produced:
vega>rpm -qlp gcc-sh-linux-gnu-5.0.0-0.1.fc23.x86_64.rpm | grep libgcc
/usr/lib/gcc/sh-linux-gnu/5.0.0/libgcc.a
/usr/lib/gcc/sh-linux-gnu/5.0.0/m2/libgcc.a
/usr/lib/gcc/sh-linux-gnu/
On Thu, Mar 26, 2015 at 10:32:18AM +0100, Richard Biener wrote:
> I think I simply didn't want to change more testcases at that point,
> but I can't see how followup passes at the very point wouldn't
> clean things up very quickly (via fre or copyprop). Eventually
> it was -Og and pass_fold_builti
On Thu, 26 Mar 2015, Jakub Jelinek wrote:
> On Thu, Mar 26, 2015 at 09:33:56AM +0100, Richard Biener wrote:
> > this hunk which I think is not really necessary given that
> > the late object-size pass now runs right before FRE which
>
> Not really immediately before that, but a few passes appart.
Dear Andre,
Everything works as expected with your new patch.
I have changed the test for pr57305 to
use iso_c_binding
implicit none
integer :: i
real, target :: e
class(*), allocatable, target :: a(:)
do i = 1, 3
e = i
call add_element_poly(a,e)
select type (a)
type is (real)
Tested on x86_64-unknown-linux-gnu, applied.
Richard.
2015-03-26 Richard Biener
PR middle-end/65519
* gnat.dg/specs/opt2.ads: Fix testcase.
Index: gcc/testsuite/gnat.dg/specs/opt2.ads
===
--- gcc/testsuite/gnat.
ping... https://gcc.gnu.org/ml/gcc-patches/2015-03/msg00262.html
On Thu, Mar 26, 2015 at 09:33:56AM +0100, Richard Biener wrote:
> this hunk which I think is not really necessary given that
> the late object-size pass now runs right before FRE which
Not really immediately before that, but a few passes appart.
And in the -Og case, while the immediately next pass
On Thu, 26 Mar 2015, Jan Hubicka wrote:
> Hi,
> this patch (as suggested by Richard) adds very simple discovery of
> DECL_NOTHORW to build_ssa passes.
>
> The reason is that in 4.9 we did build_ssa in parallel with early optimization
> that does nothrow discovery as part of local pure const. Bou
On Wed, 25 Mar 2015, Jan Hubicka wrote:
> Hi,
> linemap is optimized for situation where parser enters positions into it in
> source order.
> LTO does not work this way - it attach locations to trees and reads them more
> or less
> randomly. This results in large memory use of linemaps, slow loo
On Wed, 25 Mar 2015, Jakub Jelinek wrote:
> Hi!
>
> As discussed in the PR, fixing this issue for real (make sure we at least
> until the objsz pass don't lose information on which field's address if any
> has been taken) is probably too dangerous at this point, so this patch
> just adds a simple
On Thu, Mar 26, 2015 at 2:11 AM, Jan Hubicka wrote:
>> Hello!
>>
>> Attached patch introduces bootstrap-lto-noplugin bootstrap
>> configuration for hosts that do not support linker plugin (e.g. CentOS
>> 5.11 with binutils 2.17). Also, the patch adds some additional
>> documentation to bootstrap-l
On Thu, 2015-03-26 at 09:45 +0900, Kaz Kojima wrote:
> Oleg Endo wrote:
> > Sorry, I'm a bit confused. Do you mean something like the attached
> > patch? If so, it means that the original patch for this problem
> > https://gcc.gnu.org/ml/gcc-patches/2015-03/msg00424.html
> > should have been in
ping?
Thanks,
Kugan
On 17/03/15 12:19, Kugan wrote:
>
>
> On 17/03/15 03:48, Kyrill Tkachov wrote:
>>
>> On 16/03/15 13:15, Kugan wrote:
>>> On 16/03/15 23:32, Kugan wrote:
>> lower-subreg.c:compute_costs() only cares about the cost of a (set
>> (reg)
>> (const_int )) move but I thi
72 matches
Mail list logo
