Jakub Jelinek wrote on 17/04/2011 05:26:14 PM:
>
> On Sun, Apr 17, 2011 at 11:30:31AM +0300, Ira Rosen wrote:
> > We already have this check in vect_build_slp_tree(). It didn't work for
the
> > testcase because it doesn't take into account the type of definition. I
> > agree that it's better to
On 04/17/2011 11:39 AM, Steve Kargl wrote:
On Sun, Apr 17, 2011 at 10:14:37AM -0700, Jerry DeLisle wrote:
Hi Folks,
This patch implements the adjustments required for the various rounding
modes as stated in the Fortran Standard F2008, paragraph 10.7.5.2.2. In
the process of testing this patch
And a minor change I noticed while looking at this area.
commit bd069335f569a8c148aec0daf81b3e3a57f31ad4
Author: Jason Merrill
Date: Sat Apr 16 17:31:36 2011 -0700
* tree.c (get_target_expr): Handle VEC_INIT_EXPR.
diff --git a/gcc/cp/tree.c b/gcc/cp/tree.c
index ad004bb..25f2c32 10064
The switch to using build_value_init for value-initialization of
non-class types caused us to incorrectly allow functional casts to array
type, leading to an ICE.
Tested x86_64-pc-linux-gnu, applied to trunk.
commit 4c4d50b6852b51cd6a227a09f330a086eb3b1d63
Author: Jason Merrill
Date: Sat Apr
... looks like I missed std::partitition. Tested x86_64-linux
parallel-mode, committed to mainline.
Paolo.
//
2011-04-17 Paolo Carlini
* testsuite/25_algorithms/partition/moveable.cc: Actually run
it in parallel-mode for check-parallel.
Index: testsuite/25_al
> > notice that OBJ_TYPE_REF is useless since foo got devitualized and it holds
> > alive
> > the useless s._vptr.S = &_ZTV1S[2]. We want to drop OBJ_TYPE_REF at time
> > we fold
> > first argument to constant.
>
> At some point we dropped all OBJ_TYPE_REF with direct fn but that lead to
> issu
Hi,
this (partially) fixes the LTO bootstrap failure with Ada enabled on the
mainline. The problem is an ICE whose origin is as follows:
1. during LTO stage, copy-prop eliminates a DECL_IN_CONSTANT_POOL variable
from the GIMPLE IR, except for a reference in a GIMPLE_DEBUG statement
(ADDR_EXPR
> * config/sparc/sparc.h (GENERAL_OR_I64, REGISTER_MOVE_COST):
> Remove. * config/sparc/sparc.c (TARGET_REGISTER_MOVE_COST): Define.
> (general_or_i64_p, sparc_register_move_cost): New function.
OK modulo:
> @@ -9124,6 +9128,36 @@
> }
> }
>
> +static inline bool
> +general_or_i64_p
This removes the bunch of -Wxxx flags that are currently passed to link
gnatbind, by replacing $(ALL_CFLAGS) with $(CFLAGS) as is done for gnat1.
This (marginally) helps during LTO bootstrap.
Tested on i586-suse-linux, applied on the mainline and 4.6 branch.
2011-04-17 Eric Botcazou
On Saturday 16 April 2011 22:25:19 Tobias Burnus wrote:
> In simplify.c, currently an error is returned if the cobounds are
> exceeded. One should probably downgrade those to warnings.
OK with that change.
As far as I know, it is (weird but) valid to use the image_index result for
something comple
Actually, after reinvestigating this, "AFAIK, the vectorizer is smart
enough to automagically increase alignment on global arrays" so is not
of concern to Freescale. Also, Changing the alignment of arrays within
structs would change the field layout and would violate the ABI and be
incompatible wit
... applied mainline and branch this follow-up, see audit-trail for details.
Tested x86_64-linux.
Paolo.
PS: Jon, looks like we are missing some constraining in the templated
constructor and and assignment operator of the unique_ptr specialization?
2011-04-17 Daniel Krugle
> Adding assertion sounds good to me -- the only problem is that problem
> like this hard to be triggered during testing, thus making assertion
> less useful as it can be. Is it better to add the assertion with
> checking is enabled while still keeping the correction code?
Hi,
since we speak about
On Sun, Apr 17, 2011 at 1:39 PM, Steven Bosscher wrote:
> On Sun, Apr 17, 2011 at 9:39 PM, Easwaran Raman wrote:
>> @@ -372,8 +366,9 @@
>> to elements will conflict. In case of unions we have
>> to be careful as type based aliasing rules may say
>>
On Sun, Apr 17, 2011 at 9:39 PM, Easwaran Raman wrote:
> @@ -372,8 +366,9 @@
> to elements will conflict. In case of unions we have
> to be careful as type based aliasing rules may say
> access to the same memory does not conflict. So play
> -
Richard Sandiford writes:
> PR 45074 showed up a rather embarrassing oversight in the MIPS backend:
> global registers were still being treated as call-saved.
>
> Fixed with the attached patch. Tested on mips64-linux-gnu and applied.
>
> Richard
>
>
> gcc/
> * config/mips/mips.c
On Sun, 17 Apr 2011, Jan Hubicka wrote:
> Hi,
> this patch drops vtable_method filed. I never understood what it is about but
> reading
> PR20991 I am convinced that we hit the same problem with work on new
> devirutalization code.
> I implemented what Mark describe as "ideal solution" there - i
Hello.
This patch removes obsolete REGISTER_MOVE_COST macro from SPARC back end in
the GCC and introduces equivalent TARGET_REGISTER_MOVE_COST target hooks.
Bootstrapped and regression tested on sparc64-unknown-linux-gnu.
OK to install?
* config/sparc/sparc.h (GENERAL_OR_I64, R
Hi,
This patch impoves the heuristic used in assigning stack location to
stack variables. Currently, if there are 3 variables A, B and C with
their sizes in increasing order and A and C have a conflict, it will
put A and B in a partition and C in a separate partition with a total
frame size of si
Adding assertion sounds good to me -- the only problem is that problem
like this hard to be triggered during testing, thus making assertion
less useful as it can be. Is it better to add the assertion with
checking is enabled while still keeping the correction code?
Thanks,
David
On Sun, Apr 17,
2011/4/17 Denis Chertykov :
> 2011/4/15 Georg-Johann Lay :
>> Finally, I exposed alternative #3 of the insns to the register
>> allocator, because it is not possible to distinguish between
>> overlapping or non-overlapping regs, and #3 does not need a scratch.
>>
>> Ran C-testsuite with no regressi
Hi,
hmm, looks we both run into same issue and developed different fixes.
>
> Good point. The change above was added based on 4.4.3 where the sum
> computation has no capping like above. Yes, with the current capping,
> the negative value won't result. However, it is still good to guard
> the scal
On Sun, Apr 17, 2011 at 8:36 AM, Jan Hubicka wrote:
>> Hi, in current FDO implementation, the source file version used in
>> profile-generate needs to strictly match the version used in
>> profile-use -- simple formating changes will invalidate the profile
>> data (use of it will lead to compiler
On Sun, Apr 17, 2011 at 10:14:37AM -0700, Jerry DeLisle wrote:
> Hi Folks,
>
> This patch implements the adjustments required for the various rounding
> modes as stated in the Fortran Standard F2008, paragraph 10.7.5.2.2. In
> the process of testing this patch we have found another bug related
On Sun, Apr 17, 2011 at 8:29 AM, Jan Hubicka wrote:
>> Hi please review the attached patch.
>>
>> Ok when bootstrap and test finish?
>>
>> Thanks,
>>
>> David
>>
>>
>>
>> 2011-04-07 Xinliang David Li
>>
>> * ipa-cp.c (ipcp_update_profiling): Correct
>> negative scale factor due to i
PR 45074 showed up a rather embarrassing oversight in the MIPS backend:
global registers were still being treated as call-saved.
Fixed with the attached patch. Tested on mips64-linux-gnu and applied.
Richard
gcc/
* config/mips/mips.c (mips_cfun_call_saved_reg_p): Handle global
PR45074 is a case where expand_absneg_bit needs to decompose a
multi-word negation operation into separate word_mode operations.
The target of the operation is a global FP register variable.
For various reasons, CANNOT_CHANGE_MODE_CLASS forbids changes
from DFmode to SImode for FP registers, so a
Hi Folks,
This patch implements the adjustments required for the various rounding modes as
stated in the Fortran Standard F2008, paragraph 10.7.5.2.2. In the process of
testing this patch we have found another bug related to D and E editing which is
now PR48651. In the meantime, this regress
Comitted as obvious.
Index: ChangeLog
===
--- ChangeLog (revision 172613)
+++ ChangeLog (working copy)
@@ -1,5 +1,12 @@
2011-04-17 Jan Hubicka
+ * cgrpah.h (struct cgraph_node): Remove finalized_by_frontend.
+ *
Hi,
this patch drops vtable_method filed. I never understood what it is about but
reading
PR20991 I am convinced that we hit the same problem with work on new
devirutalization code.
I implemented what Mark describe as "ideal solution" there - i.e. teaching
cgraph that
virtual functions might bec
> Hi, in current FDO implementation, the source file version used in
> profile-generate needs to strictly match the version used in
> profile-use -- simple formating changes will invalidate the profile
> data (use of it will lead to compiler error or ICE). There are two
> main problems that lead t
> Hi please review the attached patch.
>
> Ok when bootstrap and test finish?
>
> Thanks,
>
> David
>
>
>
> 2011-04-07 Xinliang David Li
>
> * ipa-cp.c (ipcp_update_profiling): Correct
> negative scale factor due to insane profile data.
> Index: ipa-cp.c
> ===
> On Fri, Apr 15, 2011 at 15:20, Xinliang David Li wrote:
> > This is a trivial patch to deal with bad profile data (from
> > multi-threaded programs) that leads to divide by zero error.
> >
> > Ok for trunk?
> >
> > Thanks,
> >
> > David
> >
> >
> >
> > 2011-04-15 Xinliang David Li
> >
> >
This fixes profiled LTO bootstrap with Ada enabled on the 4.6 branch. The bug
is a segfault in merge_profile_summaries:
Program received signal SIGSEGV, Segmentation fault.
merge_profile_summaries (file_data_vec=0x718c9000)
at /home/eric/svn/gcc-4_6-branch/gcc/lto-cgraph.c:1504
1504
> > For me, I get everything inlined except for:
>
> Oops! looking at my mail, I realized that I did not timed fatigue, but ac.
> With the patch the only difference for fatigue is that the threshold for
> full inlining decreased from 125 to 113.
See, we make progress. 10 more patches like this an
On Sun, Apr 17, 2011 at 11:30:31AM +0300, Ira Rosen wrote:
> We already have this check in vect_build_slp_tree(). It didn't work for the
> testcase because it doesn't take into account the type of definition. I
> agree that it's better to move it here and base the vector/scalar decision
> on it, b
> For me, I get everything inlined except for:
Oops! looking at my mail, I realized that I did not timed fatigue, but ac.
With the patch the only difference for fatigue is that the threshold for
full inlining decreased from 125 to 113.
Sorry for the noise.
Dominique
gcc-patches-ow...@gcc.gnu.org wrote on 12/04/2011 04:59:16 PM:
>
> This patch adds vec_load_lanes and vec_store_lanes optabs for
instructions
> like NEON's vldN and vstN. The optabs are defined this way because the
> vectors must be allocated to a block of consecutive registers.
>
> Tested on x86
Hello, gentle maintainer.
This is a message from the Translation Project robot.
A revised PO file for textual domain 'gcc' has been submitted
by the German team of translators. The file is available at:
http://translationproject.org/latest/gcc/de.po
(This file, 'gcc-4.6.0.de.po', has just
This is the last of three regressions caused by my introduction of
reallocation on assignment. The comments in the patch adequately
explain what is being done.
Bootstrapped and regtested on FC9/x86_64 - OK for trunk and 4.6?
Paul
2011-04-17 Paul Thomas
PR fortran/48462
* tra
> Honza,
>
> Your patch seems to make --param max-inline-insns-auto= ineffective:
>
> gfc -Ofast -funroll-loops -ftree-loop-linear -fomit-frame-pointer --param
> max-inline-insns-auto=4000 -fwhole-program -fstack-arrays ac.f90
> 8.105u 0.005s 0:08.11 99.8% 0+0k 0+5io 0pf+0w
>
> while the timi
> >
> > I never really tuned the stack frame growth heuristics since it did not
> > cause any problems
> > in the benchmarks. On fortran this is quite different because of the large
> > i/o blocks
> > hitting it very commonly, so I will look into making it more permissive.
> > We definitely
> >
Hi,
Thanks Diego for committing this patch.
Yes, I would like to submit the patch to trunk. The reason is for this
patch is that setlocale() in bionicC always returns NULL (bionicC does
not support setlocale()), however libstdc++ does not handle the NULL
return value of setlocale(xxx,NULL) and t
Hello,
The attached patch extends the current implementation to analyze
instructions with REG_INC_NOTE.
Tested on ppc64-redhat-linux (bootstrap and regtest) SPU (only regtest)
and arm-linux-gnueabi (bootstrap c and regtest) configured with
--with-arch=armv7-a --with-mode=thumb.
OK for mainline?
Hi,
I propose to add a phrase to the "rpath"-section (Dynamic linker is
unable to find GCC libraries), which explains how to use "-R" as
compiler-option (by using "-Wl,")
Axel
--- old/faq.html2011-04-17 13:55:36.0 +0200
+++ new/faq.html2011-04-17 14:00:28.0 +0200
@@ -244,
Hello,
>
> New params need documentation in doc/invoke.texi. Please also change the
> maximum value to 2, not 1. Ok with that changes.
When changing the max value to 2, any value that is greater than
2 is rejected due to the following comment in params.def:
- The maximum acceptable value
On Sun, Apr 17, 2011 at 10:35 AM, Jan Hubicka wrote:
>> AFAICT revision 172430 fixed the original problem in pr45810:
>>
>> gfc -Ofast -fwhole-program fatigue.f90 : 6.301u 0.003s 0:06.30
>> gfc -Ofast -fwhole-program -flto fatigue.f90 : 6.263u 0.003s 0:06.26
>>
>> However if I play with --pa
Honza,
Your patch seems to make --param max-inline-insns-auto= ineffective:
gfc -Ofast -funroll-loops -ftree-loop-linear -fomit-frame-pointer --param
max-inline-insns-auto=4000 -fwhole-program -fstack-arrays ac.f90
8.105u 0.005s 0:08.11 99.8% 0+0k 0+5io 0pf+0w
while the timing was ~2.7s wit
On Sun, Apr 17, 2011 at 8:33 AM, Revital Eres wrote:
> Hello,
>
> Following Richard's suggestion here is the patch again.
> Tested on ppc64-redhat-linux.
>
> OK for mainline?
New params need documentation in doc/invoke.texi. Please also change the
maximum value to 2, not 1. Ok with that changes
> Hi,
>
> since r172430 lto bootstrap is broken, as well as the attached testcase
> (pr48622) and cpu2006 compilation (pr48645). The inline summary writer
> used a different order for size and time than the reader expected.
>
> I've committed the below patch as obvious (r172603) after verifyin
The compiler ICEs on the call to a function returning a discriminated record
type with variant part. The problem is again an incorrect sharing of a tree
node between two types.
Tested on i586-suse-linux, applied on the mainline.
2011-04-17 Eric Botcazou
* gcc-interface/decl.c (gna
Honza,
> I will also look into the estimate_size ICE reported today.
This has been fixed by revision 172603 and now the thresholds are
the same with/without -flto. For the fine tuning I have posted
some results on pr48636.
Cheers,
Dominique
gcc-patches-ow...@gcc.gnu.org wrote on 12/04/2011 04:38:54 PM:
> vectorizable_store contains the code:
>
> for (j = 0; j < ncopies; j++)
> {
> for (i = 0; i < vec_num; i++)
>{
> ...
> if (j == 0)
>STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
>
2011/4/15 Georg-Johann Lay :
> Denis Chertykov schrieb:
>> 2011/4/14 Georg-Johann Lay :
>>> Denis Chertykov schrieb:
2011/4/14 Georg-Johann Lay :
> The "rotl3" expanders (mode \in {HI,SI,DI}) violates synopsis by
> using 4 operands instead of 3. This runs in ICE in top of
> optabs.
> AFAICT revision 172430 fixed the original problem in pr45810:
>
> gfc -Ofast -fwhole-program fatigue.f90 : 6.301u 0.003s 0:06.30
> gfc -Ofast -fwhole-program -flto fatigue.f90 : 6.263u 0.003s 0:06.26
>
> However if I play with --param max-inline-insns-auto=*, I get
>
> gfc -Ofast -fwhole
Jakub Jelinek wrote on 16/04/2011 09:16:23 AM:
>
> Hi!
>
> As the attached testcase shows, while the current detection of what
> shifts are by scalar and what shifts are by vector shift count
> may work well for loop vectorizer (vect_internal_def being vector
> shift, vect_external_def or vect_c
On Sun, 17 Apr 2011 02:55:33 +0200 (CEST)
"Nicola Pero" wrote:
> While reading GCC code, I noticed that in gengtype-state.c
> the equality function in a string hashtable is set to strcmp.
>
> But that returns 0 (ie, false for hashtable.c) when the strings
> are equal! I can't see how that hasht
Hello,
The attached patch avoids considering debug_insn when calculating SCCs.
With this change the existence of debug_insn does not influence
the scheduling order and rec_MII.
Bootstrap and regtest on ppc64-redhat-linux and regtest on arm-linux-gnueabi.
OK for mainline?
Thanks,
Revital
Change
Hello,
Following Richard's suggestion here is the patch again.
Tested on ppc64-redhat-linux.
OK for mainline?
Thanks,
Revital
Changelog
* params.def (sms-min-sc): New param flag.
* modulo-sched.c (sms_schedule): Use it.
=== modified file 'gcc/modulo-sched.c'
--- gcc/modulo-sch
Hello,
* Diego Novillo wrote on Sat, Apr 16, 2011 at 10:14:32PM CEST:
> 2011-04-15 Rong Xu
>
> * Makefile.in (install-leaf): Install gcov-io.h, gcov-iov.h
> gcov-io.c and libgcov.c to $DESTDIR/$inst_libdir/gcov-src.
> --- a/libgcc/Makefile.in
> +++ b/libgcc/Makefile.in
> @@ -9
60 matches
Mail list logo
