On 2011.12.03 at 15:35 +, Jonathan Wakely wrote:
> On 12 November 2011 15:14, Jonathan Wakely wrote:
> > On 12 November 2011 15:04, Marc Glisse wrote:
> >>
> >> Debug-mode seems to check that first,last is a valid range, is a heap, but
> >> not that it is not empty. Maybe it could?
> >
> > Good
Back to this...
Bernd Schmidt writes:
>> gcc/
>> * reload1.c (reload_regs_reach_end_p): Replace with...
>> (reload_reg_rtx_reaches_end_p): ...this function.
>> (new_spill_reg_store): Update commentary.
>> (emit_input_reload_insns): Don't clear new_spill_reg_store here.
>>
Ping for:
http://gcc.gnu.org/ml/gcc-patches/2011-11/msg02449.html
which fixes a wrong-code bug in reorg.c for MIPS.
Richard
I think I must have messed up my previous testing of MIPS16 __sync stuff.
This patch fixes one bit of fallout, as seen in ia64-sync-1.c. If the
result of sync_lock_test_and_set is unused, the optab helper routine
maybe_emit_sync_lock_test_and_set gets called with a "target" of const0_rtx.
If the o
[Sorry Kaz, there might be some overlapping work here. I wrote the
patch before realising there was a PR about it.]
This patch fixes link failures in various atomic tests on MIPS16.
The internal optab names use "ior" to be consistent with the GCC
rtl convention, but the external names use "or" i
Several profiling tests fail for MIPS16. The problem is that MIPS has
native TLS support, but the ABI has not "yet" been extended to MIPS16.
MIPS16 is supposed to be link-compatible with non-MIPS16, so we can't
use emultls, and must simply say sorry().
This patch adds dg-require-profiling to the
On Sat, Dec 3, 2011 at 5:54 PM, Michael Zolotukhin
wrote:
>> I mean, that, when 256-bit vectorization is enabled we still use 128bit
>> vectorization if the arrays are too short for 256bit vectorization. You'll
>> lose this test coverage when you change the array sizes.
> That's true, but do we n
Hello!
> Attached is a patch which fixes bug target/51393:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51393
>
> Also attached, avx_bug.c is a minimal example to reproduce the bug
> (requires an AVX-capable CPU):
>
> $ gcc -O3 -mavx avx_bug.c
> $ ./a.out 0x8000
> in = 0x80
A while back I added the target_globals structure, to allow a backend
to switch between two very different ISA modes without paying the full
target_reinit penalty each time. This made a huge difference to compile
time, but had a drawback: the target_globals structure contained both
GGC and non-GGC
Richard Sandiford writes:
> Richard Sandiford writes:
>> Bernd Schmidt writes:
The reason I'm suddenly "reviewing" the code now is that it
doesn't prevent shrink-wrapping, because nothing adds register 2
to the liveness info of the affected blocks. The temporary prologue
val
Tom de Vries writes:
> OK, factored out delete_label now.
>
> Bootstrapped and reg-tested on x86_64.
>
> Ok for next stage1?
Looks good codewise. I'm just a bit worried about the name "delete_label".
"delete_insn (label)" should always do the right thing for a pure deletion;
the point of the new
On 4 December 2011 09:37, Markus Trippelsdorf wrote:
>
> You forgot to change the second one with the comparison functor.
Doh! Thanks, fixed now.
* include/bits/stl_heap.h (pop_heap): Check for non-empty range in
overload taking a predicate.
* testsuite/25_algorithms/pop_h
Hi,
Attached is a patch to fix miscompilation in
arm.md:*minmax_arithsi.
The following testcase, reduced from efgcvt_r.c:fcvt_r in glibc, gets
miscompiled:
extern void abort (void);
int __attribute__((noinline))
foo (int a, int b)
{
int max = (b > 0) ? b : 0;
return max - a;
}
int
main (vo
Hi,
This patch adds a missing exit phi node for outer loop in
vectorization of double reduction.
Bootstrapped and tested on powerpc64-suse-linux.
Committed.
Ira
ChangeLog:
PR middle-end/51285
* tree-vect-loop.c (vect_create_epilog_for_reduction): Create exit
phi nodes f
Thomas Koenig wrote:
Regression-tested. OK for trunk?
2011-11-29 Thomas Koenig
PR fortran/51338
* dependency.c (are_identical_variables): Handle case where
end fields of substring references are NULL.
2011-11-29 Thomas Koenig
PR fortran/51338
* gfo
> Looks good codewise.
Seconded, modulo the file: the function should be in cfgrtl.c instead.
> I'm just a bit worried about the name "delete_label".
> "delete_insn (label)" should always do the right thing for a pure deletion;
> the point of the new routine is that it also moves instructions.
> This introduces host-dependent code generation differences, right?
> You can simply use int64_t for code that is run on the host only.
Well, no, there is an entire file dedicated to this business (hwint.h).
--
Eric Botcazou
> I don't think that is related to C++ switch, because C++03 doesn't have
> long long, only C++11 and C99 has it. We apparently are using int64_t or
> uint64_t in a couple of places already though:
IMHO they should be audited and fixed if need be.
> ada/tb-gcc.c: uwx_get_reg ((struct uwx_env *)
On Sun, Dec 4, 2011 at 4:31 PM, Eric Botcazou wrote:
>> This introduces host-dependent code generation differences, right?
>> You can simply use int64_t for code that is run on the host only.
>
> Well, no, there is an entire file dedicated to this business (hwint.h).
No to what? To the fact that
On Sun, Dec 4, 2011 at 4:37 PM, Richard Guenther
wrote:
> On Sun, Dec 4, 2011 at 4:31 PM, Eric Botcazou wrote:
>>> This introduces host-dependent code generation differences, right?
>>> You can simply use int64_t for code that is run on the host only.
>>
>> Well, no, there is an entire file dedic
Georg-Johann Lay schrieb:
Denis Chertykov wrote:
Georg-Johann Lay:
I attached a patch but I fail to find the right configure options for
gcc/binutils as the testsuite complains
./avr/bin/ld: bad -plugin option
Maybe the patch can be pre-approved so that the others can proceed with their
wo
> No to what? To the fact that HOST_WIDEST_INT is host-dependent
> and thus should not be used to drive code generation? Or no to the
> fact that we can (and do) use int64_t as host integer type?
No to the fact that int64_t should be used (and the occurrences in the LTO code
are OK). hwint.h i
Georg-Johann Lay wrote:
http://gcc.gnu.org/ml/gcc-patches/2011-11/msg02574.html
As this is a blocker and I am blocked myself by a collect2 issue:
Eric, Denis, could one of you test the patch and apply it if it is okay?
It is PR51409.
In addition, please add PR49868 to the ChangeLog. Thanks.
> > No to what? To the fact that HOST_WIDEST_INT is host-dependent
> > and thus should not be used to drive code generation? Or no to the
> > fact that we can (and do) use int64_t as host integer type?
>
> No to the fact that int64_t should be used (and the occurrences in the LTO
> code
> are
2011/12/4 Georg-Johann Lay :
> Georg-Johann Lay wrote:
>>
>>
>> http://gcc.gnu.org/ml/gcc-patches/2011-11/msg02574.html
>>
>> As this is a blocker and I am blocked myself by a collect2 issue:
>>
>> Eric, Denis, could one of you test the patch and apply it if it is okay?
>> It is PR51409.
I'm sorr
On Saturday 03 December 2011 20:12:50 Tobias Burnus wrote:
> Another OOP-related patch: If one uses type extension, the first
> REF_COMPONENT does not necessarily refer directly to a component in the
> linked list starting at sym->ts.u.derived->components.
>
> Using simply ref->u.c.component direc
> * resource.c (init_resource_info): Only consider EXIT_IGNORE_STACK
> if there is in epilogue.
OK, but keep the original order of the tests, i.e. EXIT_IGNORE_STACK is tested
only after we know there is a frame pointer:
if (!(frame_pointer_needed
&& EXIT_IGNORE_STACK
On Dec 4, 2011, at 3:29 AM, Richard Sandiford
wrote:
> The problem is that MIPS has
> native TLS support, but the ABI has not "yet" been extended to MIPS16.
> MIPS16 is supposed to be link-compatible with non-MIPS16, so we can't
> use emultls, and must simply say sorry().
>
> This patch adds dg-
On Dec 4, 2011, at 4:02 AM, Richard Sandiford
wrote:
> A while back I added the target_globals structure, to allow a backend
> to switch between two very different ISA modes without paying the full
> target_reinit penalty each time. This made a huge difference to compile
> time, but had a drawba
This patch declares the Type Traits doxygen group in not
in *and* . It also makes sure
doxygen comments are attached to the actual trait such as
is_member_pointer, not a helper such as __is_member_pointer_helper. I
added brief descriptions for some of the less self-explanatory traits
(enable_if,
On Sat, Dec 03, 2011 at 10:45:18PM -0800, Mike Stump wrote:
> On Dec 3, 2011, at 7:25 AM, Jack Howarth wrote:
> > FSF gcc currently doesn't handle -fno-pie and friends properly under Lion.
> > The darwin11 linker now defaults to -pie
>
> > Okay for gcc trunk and backports to gcc-4_5-branch/gcc-4_6
On Dec 4, 2011, at 9:09 AM, Jack Howarth wrote:
> On Sat, Dec 03, 2011 at 10:45:18PM -0800, Mike Stump wrote:
>> On Dec 3, 2011, at 7:25 AM, Jack Howarth wrote:
>>> FSF gcc currently doesn't handle -fno-pie and friends properly under Lion.
>>> The darwin11 linker now defaults to -pie
>>
>>> Okay
> Actually, the real problem is that the Cygwin-targeted version of gnat
> shouldn't need those definitions in the first place. Cygwin provides a
> fairly complete Linux/Posix feature-set, and doing an end-run around it by
> using the underlying winsock API isn't usually a good idea, so I think
Hi all,
as Dominique has found, Fortran 2008 allows the BOZ edit descriptors now
also with REAL and COMPLEX arguments. (See PR for quotes from the standard.)
Build and regtested on x86-64-linux.
OK for the trunk?
Tobias
PS: Thank you, Mikael, for reviewing my ASSOCIATE patch!
2011-12-04 Tob
On 12/04/2011 03:07 AM, Richard Sandiford wrote:
> * optabs.c (maybe_emit_sync_lock_test_and_set): Pass a null target
> to emit_library_call_value.
> (expand_atomic_compare_and_swap): Likewise.
Ok.
r~
Hi,
for this ice on invalid, 4.7 Regression, the idea is just early
returning error_mark_node from build_functional_cast, after the error,
like in all the other error conditions explicitly dealt with there,
instead of setting type = error_mark_node.
The catch is, for testcases like auto25.C:
I've committed this patch to break apart the gcov finalization routines, I
believe this will make it easier to fix the problem shown up by bug 51113 --
although this patch does not. Notable changes:
* rename coverage_begin_output to coverage_begin_function for consistency with
coverage_end_fu
On Sun, Dec 04, 2011 at 09:28:55AM -0800, Mike Stump wrote:
> On Dec 4, 2011, at 9:09 AM, Jack Howarth wrote:
> > On Sat, Dec 03, 2011 at 10:45:18PM -0800, Mike Stump wrote:
> >> On Dec 3, 2011, at 7:25 AM, Jack Howarth wrote:
> >>> FSF gcc currently doesn't handle -fno-pie and friends properly un
On Sun, Dec 04, 2011 at 02:00:20PM -0500, Jack Howarth wrote:
> > > at -m32/-m64 on x86_64-apple-darwin11 due to the -pie linker default.
> > > Iain had wanted
> > > to leave these in place to encourage boehm-gc to be fixed but I doubt
> > > that is a realistic
> > > goal in the near/middle term.
On Sun, Dec 04, 2011 at 08:18:32PM +0100, Jakub Jelinek wrote:
> On Sun, Dec 04, 2011 at 02:00:20PM -0500, Jack Howarth wrote:
> > > > at -m32/-m64 on x86_64-apple-darwin11 due to the -pie linker default.
> > > > Iain had wanted
> > > > to leave these in place to encourage boehm-gc to be fixed but
On 4 Dec 2011, at 06:35, Mike Stump wrote:
On Nov 30, 2011, at 6:28 AM, Iain Sandoe wrote:
While trying to track down the vector unwind problems on ppc-
darwin, I made some tidy-ups for "save_world()".
In the end, that was not where the main problem, lay - but I did
find a few things wrong t
On 4 Dec 2011, at 20:19, Jack Howarth wrote:
On Sun, Dec 04, 2011 at 08:18:32PM +0100, Jakub Jelinek wrote:
On Sun, Dec 04, 2011 at 02:00:20PM -0500, Jack Howarth wrote:
at -m32/-m64 on x86_64-apple-darwin11 due to the -pie linker
default. Iain had wanted
to leave these in place to encourage
On 12/03/2011 09:20 AM, Iain Sandoe wrote:
> version 2 is a modification of your original:
>
> a) -FRAME+BASE(r1) cannot be guaranteed to be vec-aligned in general (it
> isn't on m32 darwin)
>
> ... so I've taken the liberty of rounding the gtm_buffer object and then
> pointing r4 at original
On Sun, Dec 04, 2011 at 08:35:07PM +, Iain Sandoe wrote:
>
> On 4 Dec 2011, at 20:19, Jack Howarth wrote:
>
>> On Sun, Dec 04, 2011 at 08:18:32PM +0100, Jakub Jelinek wrote:
>>> On Sun, Dec 04, 2011 at 02:00:20PM -0500, Jack Howarth wrote:
>> at -m32/-m64 on x86_64-apple-darwin11 due to the
> What about this way? I've groupped the two variables into a structure
> to make it clear it is internal internal_arg_pointer_based_exp* state,
> scanning is done in a separate function and the SCAN argument is gone,
> instead the internal_arg_pointer_based_exp_scan function disables scanning
> d
Hi Jonathan,
On Sat, 3 Dec 2011, Jonathan Wakely wrote:
> How's this? I think I got all the versions and dates correct, but I
> must say I find keeping some of this info in the manual to be tedious
> and unnecessary.
I agree, there is (too) much detailed and extra contents there
which does not a
On Sun, Dec 04, 2011 at 09:53:42PM +0100, Eric Botcazou wrote:
> > What about this way? I've groupped the two variables into a structure
> > to make it clear it is internal internal_arg_pointer_based_exp* state,
> > scanning is done in a separate function and the SCAN argument is gone,
> > instead
Hi Richard,
On 4 Dec 2011, at 20:45, Richard Henderson wrote:
On 12/03/2011 09:20 AM, Iain Sandoe wrote:
version 2 is a modification of your original:
a) -FRAME+BASE(r1) cannot be guaranteed to be vec-aligned in
general (it isn't on m32 darwin)
... so I've taken the liberty of rounding t
> I think it is. Those called during internal_arg_pointer_based_exp_scan
> will see scan_start equal to pc_rtx and won't scan, and for the calls after
> it, while scan_start won't be pc_rtx, as it is after scan, it is either
> NULL_RTX with no insns in the sequence, or some insn whose NEXT_INSN is
Hi,
The problem here seems to be that we don't perform the enumeration
constant resolving in finish_id_expression when the DECL is a
USING_DECL. Consequently, I think we shall strip the USING_DECL before
checking for a CONST_DECL.
Tested x86_64-unknown-linux-gnu without regressions. OK to commit
Hi,
The attached patch adds atomic patterns with software atomic
sequences. They are enabled when a new option -msoft-atomic
is specified and the option is default for sh-linux.
Regtested on sh4-unknown-linux-gnu with no new failures and
the doc patch is tested with "make info dvi pdf".
Applied o
On 4 December 2011 21:08, Gerald Pfeifer wrote:
> Hi Jonathan,
>
> On Sat, 3 Dec 2011, Jonathan Wakely wrote:
>> How's this? I think I got all the versions and dates correct, but I
>> must say I find keeping some of this info in the manual to be tedious
>> and unnecessary.
>
> I agree, there is (t
Is there a reason not to just do
decl = strip_using_decl (decl);
early in finish_id_expression?
Jason
On 2011/12/5 12:39 AM, Mike Stump wrote:
> On Dec 4, 2011, at 3:29 AM, Richard Sandiford
> wrote:
>> The problem is that MIPS has
>> native TLS support, but the ABI has not "yet" been extended to MIPS16.
>> MIPS16 is supposed to be link-compatible with non-MIPS16, so we can't
>> use emultls, and
Ok, will several tests with short arrays be enough for that or should
we keep all the original tests plus new ones with longer arrays?
Michael
On 4 December 2011 15:44, Richard Guenther wrote:
> On Sat, Dec 3, 2011 at 5:54 PM, Michael Zolotukhin
> wrote:
>>> I mean, that, when 256-bit vectoriza
On Fri, Dec 2, 2011 at 11:59 AM, Xinliang David Li wrote:
> ;
>>
>> +/* Determine whether LOOP contains floating-point computation. */
>> +bool
>> +loop_has_FP_comp(struct loop *loop)
>> +{
>> + rtx set, dest;
>
> This probably should be extended to detect other long latency
> operations in the f
Latest patch which improves the efficiency as described below is
included here. Boostrapped and checked again with
x86_64-unknown-linux-gnu. Could someone review?
Thanks,
Teresa
2011-12-04 Teresa Johnson
* loop-unroll.c (decide_unroll_constant_iterations): Call loop
unroll tar
2011/12/5 Jason Merrill :
> Is there a reason not to just do
>
> decl = strip_using_decl (decl);
>
> early in finish_id_expression?
Not really, I've already tried it and it works. I wasn't sure it was
correct not to return a USING_DECL in aIl cases -- they are numerous
in this huge function. If y
58 matches
Mail list logo
