From: Sergey Grechanik
This patch avoids changing speculative bits of a register use when it is moved
up above a speculation check where that register is used as the address
register. We should only call ds_full_merge when the producer (speculation
check) writes to a register.
(A similar check
From: Sergey Grechanik
Sometimes we need to be able to call sched_get_condition_with_rev from
selective scheduler for an instruction with zero luid (i.e. before h_d_i_d
had been extended). On such occasion, we need to bypass the caching and use
"old", uncached lookup.
The patch factors out cach
I'm finally getting back to this.
This patch adds test suite support for the libstdc++ pretty-printers.
These tests require a new gdb, so following Mike Stump's advice and the
consensus of the libstdc++ list, we now check for an appropriate gdb
feature before deciding whether to run the tests.
I
Bernd Schmidt writes:
> +@findex simple_return
> +@item (simple_return)
> +Like @code{(return)}, but truly represents only a function return, while
> +@code{(return)} may represent an insn that also performs other functions
> +of the function epilogue. Like @code{(return)}, this may also occur in
On Wed, Aug 3, 2011 at 03:17, Guozhi Wei wrote:
> Hi
>
> I want to port r175009 from trunk to google/gcc-4_6 since it fixed 3 dejagnu
> failures for arm backend.
>
> Tested on x86-64 with
> make check-gcc RUNTESTFLAGS="vect.exp"
>
> Tested on arm qemu with
> make check-gcc RUNTESTFLAGS="--target_b
On 08/03/2011 03:50 AM, Torvald Riegel wrote:
> Fix rollback and queries of transaction ID.
>
> * beginend.cc (GTM::gtm_transaction::rollback): Roll back tx id as well.
> * query.cc (_ITM_getTransactionId): There is no active transaction if
> the current nesting level is
Richard Henderson wrote:
> On 08/03/2011 07:07 AM, Georg-Johann Lay wrote:
>> #include
>>
>> void foo ()
>> {
>> printf ("%d %d %d", 1, 2, 3);
>> printf ("%d %d %d", 3, 4, 5);
>> printf ("%d %d %d", 1, 4, 5);
>> }
>>
>> Attached the output: The compiler happily pushes onto the stack
>>
On 08/03/2011 08:47 AM, Georg-Johann Lay wrote:
> With ACCUMULATE_OUTGOING_ARGS it looks much better: there is just
> one such block in the prologue/epilogue.
>
> I think ACCUMULATE_OUTGOING_ARGS would be a win definitely.
That's what I thought too, but with the test case in PR49881
I couldn't ma
Mikael,
first, thanks for carefully reading the patch!
On 08/03/2011 01:43 PM, Mikael Morin wrote:
> > PS: It somehow took me quite some time to understand "subcomponent" even
> > though the standard is rather clear about it.
> Is it? It seems I haven't understood the constraint as you did.
It'
On Tuesday 02 August 2011 18:01:35 Mikael Morin wrote:
> Thanks. I'll see if I have additional fixes after updating.
Here they are.
Tobias, I would like your input on lto/pr47839_{0,1}.f90 for which you put the
cleanup directive for module pec_mod from pr47839_1.f90 in the pr47839_0.f90
file. As
Hi Rainer,
How should we proceed with this patch, especially given the quite
moderate comments from most affected target maintainers?
I have no objections to the patch from an ARM or FRV point of view.
Cheers
Nick
On 08/03/2011 05:55 PM, Mikael Morin wrote:
Tobias, I would like your input on lto/pr47839_{0,1}.f90 for which you put the
cleanup directive for module pec_mod from pr47839_1.f90 in the pr47839_0.f90
file. As it seems to be an overlook, this patches changes it. Is that correct?
I think either
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
On 08/03/11 08:07, Georg-Johann Lay wrote:
> Georg-Johann Lay wrote:
>> Richard Henderson wrote:
>>> On 08/01/2011 11:42 AM, Georg-Johann Lay wrote:
Is there a specific reason not to define
ACCUMULATE_OUTGOING_ARGS on AVR?
>>> Yes. So that
Richard Henderson wrote on 08/03/2011 04:15:26 PM:
> On 08/03/2011 07:07 AM, Ulrich Weigand wrote:
> > Was this assert intended to impose requirements on target's
> > restore_stack_block implemtations? What exactly are those
> > supposed to be?
>
> The assert was not intended to impose requiremen
On Wednesday 03 August 2011 18:22:55 Tobias Burnus wrote:
> > --- pr32921.f (révision 177274)
> > +++ pr32921.f (copie de travail)
> > @@ -46,5 +46,5 @@
> >
> > RETURN
> > END
> >
> > ! { dg-final { scan-tree-dump-times "stride" 4 "lim1" } }
> >
> > -! { dg-final
Hello.
This patch removes the initialization class_sizes array from reduce_class
function and instead it uses reg_class_size array. Also reduce_class function
use reg_class_contents array instead of class_contents, now. The local
class_contents array duplicate global array reg_class_content
Ping.
David
On Sun, Jul 31, 2011 at 11:58 PM, Xinliang David Li wrote:
> The attached patch fixed a minor bug in cfg fixup -- the outgoing edge
> profile count is not scaled after inlining leading to warnings printed
> in IR dump -- 'Invalid sum of ...'.
>
> Bootstrap and tested on x86-64/linux,
Ping.
David
On Fri, Jul 29, 2011 at 10:10 AM, Xinliang David Li wrote:
> The attached patch fixed the problem. The root cause of the problem is
> due to the ordering change of profile_estimation and tree_profile
> pass. In trunk, the function/node frequency is not computed after
> profile annota
Ok.
On Wed, Aug 3, 2011 at 7:31 AM, H.J. Lu wrote:
> On Tue, Aug 2, 2011 at 3:32 PM, Richard Henderson wrote:
>> I got Jeff Law to review the reload change on IRC
>> and committed the composite patch.
>>
>> Tested on x86_64, i586, avr, and h8300. Most other
>> tier1 targets ought not be affected, as
On 08/03/2011 08:46 AM, Richard Guenther wrote:
If that's reasonable then adding the malloc attribute should be, too.
Finally. Please. Doesn't C++0x maybe "fix" the issue we were
discussing to death?
Nope, as far as I can tell nobody raised it with the committee. I have now.
I think we ough
My linemap implementation just caught a very important failure due to
this patch (because of it's massive use of pph_in_string for
filenames).
We CANNOT free stream->encoder.r.file_data this early as it contains
all of the strings streamed in (which are refered to directly from all
over the place
On Wed, Aug 03, 2011 at 02:31:17PM +0200, Jakub Jelinek wrote:
> 2011-08-03 Jakub Jelinek
>
> PR middle-end/49905
> * decl.c (cxx_init_decl_processing): Add alloc_size (1) attribute
> for operator new and operator new [].
Unfortunately the patch caused a bunch of regressions,
On Mon, Aug 01, 2011 at 02:27:49PM +0200, Romain Geissler wrote:
> ping
I went ahead and bootstrapped/regtested/and make install tested your patch
(note your mailer wrapped it up so that it didn't apply cleanly), but now
that I think about it, it is wrong to put the gengtype binary into
$(libsubdi
Hi!
The following testcase fails because SLP creates
vect_something = {&c, &c};
stmt but update addresses taken after parloop doesn't discover the
ADDR_EXPRs in the ctor, thus TREE_ADDRESSABLE is dropped from c.
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux.
Committing to tr
On Fri, Jul 29, 2011 at 10:10 AM, Xinliang David Li
wrote:
The attached patch fixed the problem. The root cause of the problem is
due to the ordering change of profile_estimation and tree_profile
pass. In trunk, the function/node frequency is not computed after
profile annotation is done lea
The following patch fixes a new crash on SPEC2000 sixtrack on x86-64.
LRA failed to find a mode for a constant. When LRA can not find a mode
of operand it uses mode from machine description of the insn. It is not
enough, in very rare cases the mode can be found only from context. The
patch
On 08/03/2011 02:26 PM, Jakub Jelinek wrote:
As for the properties which the middle-end would like to assume or not
from the operator new/operator new[]:
1) for alloc_size it is whether the returned pointer has exactly the
requested bytes defined, i.e. can't return a buffer where only fewer bytes
> "Peter" == Peter Collingbourne writes:
Peter> 2011-08-02 Peter Collingbourne
Peter> * expr.c (expand_invoke) Use the type of the method rewrite
target.
Ok.
Tom
Hi,
this patch fixes problem with handling aliases in emultls. With new
representation of aliases in cgraph/varpool it is actually neccesary to walk
the aliases and create corresponding veirables for them.
The patch employ strategy of creating all variables for all aliases at a time
one of them is
> On 07/14/2011 04:07 PM, Richard Henderson wrote:
> > This finally brings us to something that can support shrink-wrapping.
> > As mentioned in the description of the last patch, this is 95% of
> > what Bernd had in his last 007-dw2cfg patch, except for the remember/
> > restore_state stuff. And
On 08/03/2011 01:10 PM, Jan Hubicka wrote:
> * tree-emultls.c (new_emutls_decl):Add alias_of parameter;
> handle aliases.
> (create_emultls_var):New function.
> (ipa_lower_emutls): Handle aliases correctly.
Looks ok.
r~
On 08/03/2011 01:24 PM, Jan Hubicka wrote:
> Cool, will this also help to handle alternate entry points, so we can move
> ahead
> on this area we got stuck years ago? (i.e. add support for alternate entry
> points on
> i386 skipping IP pointer load in PIC mode/using register passing conventions)
I've committed the attached patch, please add any other v3 changes
worth mentioning
Index: htdocs/gcc-4.7/changes.html
===
RCS file: /cvs/gcc/wwwdocs/htdocs/gcc-4.7/changes.html,v
retrieving revision 1.25
diff -u -r1.25 changes.html
--
> On 08/03/2011 01:24 PM, Jan Hubicka wrote:
> > Cool, will this also help to handle alternate entry points, so we can move
> > ahead
> > on this area we got stuck years ago? (i.e. add support for alternate entry
> > points on
> > i386 skipping IP pointer load in PIC mode/using register passing
Mikael Morin wrote:
Patch updated.
Looks good! Thanks for the further clean up!
Tobias
Richard Guenther wrote:
On Wed, 3 Aug 2011, Richard Guenther wrote:
Bootstrap and regtest is currently running on x86_64-unknown-linux-gnu,
the reversed loop one was ok (well, apart from those 2-3 fails).
Re-bootstrapping currently. The previous patch tested ok.
The patch looks OK. I don't k
Hey this fixes up the array::at issue as pointed out on this list.
In addition, it adds some more tuple markup for constexpr. I'm going to
stage in the tuple work, as some of it is more controversial.
tested x86/linux
-benjamin2011-08-03 Benjamin Kosnik
* include/std/tuple: Mark more cons
> +++ b/libstdc++-v3/acinclude.m4
> @@ -685,9 +685,9 @@ AC_DEFUN([GLIBCXX_EXPORT_INCLUDES], [
>fi
>
># Stuff in the actual top level. Currently only used by libsupc++
> to
> - # get unwind* headers from the gcc dir.
> - #TOPLEVEL_INCLUDES='-I$(toplevel_srcdir)/gcc
> -I$(toplevel_srcdi
Hello,
On Wednesday 03 August 2011 15:47:37 Richard Guenther wrote:
> Comments? Any idea why reversing the loop would break?
Yes, the list of scalarized expressions has to be created in the same order it
is consumed. Here the scalarized expressions are array indexes to be
precomputed out of th
Hi,
I am testing this patch to avoid nested PARALLEL call patterns. OK for
trunk if there are no regressions on Linux/AVX?
Thanks.
H.J.
---
2011-08-03 H.J. Lu
PR bootstrap/49964
* config/i386/i386.c (ix86_expand_call): Avoid nested PARALLEL
call pattern.
(ix
On 08/02/2011 06:27 PM, H.J. Lu wrote:
Here is the updated patch. I updated REG_VALUE_IN_UNWIND_CONTEXT
document and added ASSUME_EXTENDED_UNWIND_CONTEXT. OK for
trunk?
Let's say OK on Monday if nobody objects before then.
Jason
Tobias Burnus wrote:
Mikael, first, thanks for carefully reading the patch!
Updated patch attached. Changes:
- parse.c: Cleaned up a bit, use suggested wording, add missing
diagnostic (cf. my previous mail)
- resolve.c: use suggested wording
- coarray_lock_5.f90: Remove.
- coarray_lock_6.f90:
Joseph S. Myers wrote:
> The SPU format doesn't support NaNs, so the value of this field isn't
> particularly meaningful for SPU. I'm not clear why SPU is actually using
> fp-bit; what functions from fp-bit will actually end up getting used by
> code built for SPU, and how close does fp-bit ge
When a frame pointer is in use, we can optimize popping all
queued parameters via a simple move from the frame pointer
instead of an addition to the stack pointer.
The new sequence is 4 insns, the old sequence was 9 insns.
Committed.
r~
PR target/34888
* config/avr/avr.md: New s
Committed as obvious.
r~
* config/spu/spu.md: Use define_c_enum instead of define_constants.
(UNSPECV_BLOCKAGE, UNSPECV_LNOP, UNSPECV_SYNC): Rename from UNSPEC_*.
(UNSPECV_NOP): New.
diff --git a/gcc/config/spu/spu.md b/gcc/config/spu/spu.md
index 426437a..5742e0d 100644
Jan Hubicka writes:
>
> Cool, will this also help to handle alternate entry points, so we can move
> ahead
> on this area we got stuck years ago? (i.e. add support for alternate entry
> points on
> i386 skipping IP pointer load in PIC mode/using register passing conventions)
Also alternative e
These tests recently started failing at compile time for me due to no
declarations for unlink due to no include of unistd.h. So, I did the
obvious thing and included it.
tested x86/linux
-benjamin2011-08-03 Benjamin Kosnik
* testsuite/performance/27_io/filebuf_sputn_unbuf.cc: Include
unist
This patch came about because I noticed a comment about "the second
condition" in rs6000_emit_epilogue no longer matched the code. I was
going to fix the comment, then noticed the code could be simplified.
In rs6000_emit_epilogue, lr_save_offset is a small positive value (4,8
or 16), or zero when
This patch from Roberto Lublinerman fixes PR 49889, which is a case
where the compiler failed to give an error for a program which made a
tuple assignment from a symbol which was not a function at all.
Bootstrapped and ran Go testsuite on x86_64-unknown-linux-gnu.
Committed to mainline.
Ian
diff
2011/8/4 Richard Henderson :
> When a frame pointer is in use, we can optimize popping all
> queued parameters via a simple move from the frame pointer
> instead of an addition to the stack pointer.
>
> The new sequence is 4 insns, the old sequence was 9 insns.
>
> Committed.
It seems strange for
Hi,
Bootstrap in C++ mode broke after a recent commit for
arm-linux-gnueabi - Committed as obvious .
cheers
Ramana
2011-08-04 Ramana Radhakrishnan
* config/arm/arm.c (arm_set_fixed_conv_libfunc): Constify maybe_suffix.
Index: gcc/config/arm/arm.c
==
101 - 152 of 152 matches
Mail list logo
