Hi,
this patch fixes wrong type propagation across global variable.
Bootstrapped/regtested x86_64-linux, will commit it shortly.
PR ipa/63166
* ipa-prop.c (compute_known_type_jump_func): Fix conditional.
* g++.dg/lto/pr63166_0.ii: New testcase.
* g++.dg/lto/pr63166
Ping.
On Thu, Sep 4, 2014 at 11:41 AM, Andrey Turetskiy
wrote:
> Ping.
>
> On Tue, Aug 26, 2014 at 10:41 AM, Andrey Turetskiy
> wrote:
>> Ping.
>>
>> On Tue, Aug 19, 2014 at 12:45 PM, Kirill Yukhin
>> wrote:
>>> Hello,
>>> On 12 Aug 10:58, Andrey Turetskiy wrote:
All remarks from
htt
Hi,
Currently most thumb1_size_rtx_costs are "guessed" from performance view,
not size.
The patch adjusts some of them according the instruction patterns defined in
thumb1.md. It also replaces several hard coded "4" to COSTS_N_INSNS (1).
No make check regression for Cortex-M0 on qemu.
For Cortex
On Tue, Sep 09, 2014 at 10:51:23PM +, Joseph S. Myers wrote:
> On Thu, 28 Aug 2014, Maxim Ostapenko wrote:
>
> > diff --git a/gcc/diagnostic.c b/gcc/diagnostic.c
> > index 0cc7593..67b8c5b 100644
> > --- a/gcc/diagnostic.c
> > +++ b/gcc/diagnostic.c
> > @@ -492,7 +492,7 @@ diagnostic_action_af
On Tue, Sep 09, 2014 at 04:42:04PM +0200, Richard Biener wrote:
> On Tue, Sep 9, 2014 at 4:30 PM, Jakub Jelinek wrote:
> > On Tue, Sep 09, 2014 at 04:25:23PM +0200, Richard Biener wrote:
> >> why wasn't 'result' built using simplify_gen_* in the first place? I also
> >
> > It is built using cseli
On Tue, Sep 09, 2014 at 03:37:26PM +0100, Alan Lawrence wrote:
> Following this, we're seeing ICEs in tests in gcc.dg/pch.exp and
> g++.dg/pch.exp,
> with cross-builds (hosted on x86_64) targetting bare metal AArch64 and ARM
> (aarch64-none-elf, aarch64_be-none-elf and arm-none-eabi; I haven't tes
2014-09-09 Robert Suchanek
gcc/
* lra-lives.c (process_bb_lives): Replace assignment with bitwise OR
assignment.
gcc/testsuite/
* gcc.target/mips/20140909.c: New test.
---
gcc/lra-lives.c |6 +++---
gcc/testsuite/gcc.target/mips/2014090
On Tue, Sep 9, 2014 at 3:29 PM, Segher Boessenkool
wrote:
> These patterns use subfic now. subfic clobbers the carry. Other code
> already preferably uses xor. Let's do the same here.
>
> Bootstrapped and tested as usual, no regressions. Is this okay to apply?
>
>
> Segher
>
>
> 2014-09-09 Se
On Thu, 28 Aug 2014, Maxim Ostapenko wrote:
> diff --git a/gcc/diagnostic.c b/gcc/diagnostic.c
> index 0cc7593..67b8c5b 100644
> --- a/gcc/diagnostic.c
> +++ b/gcc/diagnostic.c
> @@ -492,7 +492,7 @@ diagnostic_action_after_output (diagnostic_context
> *context,
> real_abort ();
>dia
On Mon, 1 Sep 2014, Maciej W. Rozycki wrote:
> This fixes an issue with the mode used for register save slots on the
> stack where e500 processor support is enabled along non-e500 multilibs.
> In that case the HARD_REGNO_CALLER_SAVE_MODE macro definition from
> gcc/config/rs6000/e500.h overri
On 09/09/14 23:03 +0200, François Dumont wrote:
On 09/09/2014 19:29, Jonathan Wakely wrote:
On 14/08/14 21:22 +0200, François Dumont wrote:
I am preparing a patch for profile mode so I will submit
modification for this mode with this big patch.
btw, François, for profile mode I think we shoul
On Fri, Sep 5, 2014 at 6:52 PM, Thomas Schwinge wrote:
> Hi!
>
> On Tue, 06 Jan 2009 00:10:02 +0200, Janne Blomqvist
> wrote:
>> attached is a substantially reworked low level I/O library for gfortran.
>> [...]
>
> Due to a recent improvement in GCC, it is now pointing out an issue in
> the foll
On 09/09/2014 19:29, Jonathan Wakely wrote:
On 14/08/14 21:22 +0200, François Dumont wrote:
I am preparing a patch for profile mode so I will submit modification
for this mode with this big patch.
btw, François, for profile mode I think we should just do something
like this patch.
I feel quit
On Fri, 5 Sep 2014, Manuel L?pez-Ib??ez wrote:
> This adds a new option property CppReason which maps to a warning
> reason code in cpplib.h. This allows us to auto-generate
> cpp_reason_option_codes[], which maps from CPP warning codes to GCC
> ones, thus making a bit harder to forget to update t
On 09/05/2014 10:51 AM, David Sherwood wrote:
> Hi,
>
> I have a potential fix for a gcc testsuite failure for aarch64 in big
> endian, i.e.
>
> FAIL: tmpdir-gcc.dg-struct-layout-1/t028 c_compat_x_tst.o compile, (internal
> compiler error)
> FAIL: tmpdir-gcc.dg-struct-layout-1/t028 c_compat_y_tst.o
These patterns use subfic now. subfic clobbers the carry. Other code
already preferably uses xor. Let's do the same here.
Bootstrapped and tested as usual, no regressions. Is this okay to apply?
Segher
2014-09-09 Segher Boessenkool
* config/rs6000/htm.md (tabort, tabortdc, tabo
On 09/09/14 11:34, David Malcolm wrote:
[dropping DJ from CC as this doesn't relate to rl78]
[adding Denis to CC due to avr-related discussion]
The reason for the as_a here is that avr_out_plus and
avr_out_bitop are mostly called with insns [1], but are also called with
SET patterns [2], which
Richard, thanks for the review. The revised patch is attached. Is this
one OK (after testing is done)?
David
2014-09-08 Xinliang David Li
PR target/63209
* config/arm/arm.md (movcond_addsi): Handle case where source
and target operands are the same
2014-09-08 Xinlia
Rainer Orth writes:
> Hi Richard,
>>> It seems the new get_some_local_dynamic_name implementation in
>>> function.c lost the non-NULL check the sparc.c version had. I'm
>>> currently testing the following patch:
>>
>> Could you do a:
>>
>> call debug_rtx (...)
>>
>> on the insn that contains a
Hi Richard,
>> It seems the new get_some_local_dynamic_name implementation in
>> function.c lost the non-NULL check the sparc.c version had. I'm
>> currently testing the following patch:
>
> Could you do a:
>
> call debug_rtx (...)
>
> on the insn that contains a null pointer? Normally insn pa
Committed to branch dmalcolm/jit:
gcc/testsuite/ChangeLog.jit:
* jit.dg/test-nested-loops.c: New test case.
* jit.dg/all-non-failing-tests.h: Add test-nested-loops.c.
* jit.dg/test-combination.c (create_code): Likewise.
(verify_code): Likewise.
* jit.dg/test
> > I fixed this (in patch 5) by introducing a new local rtx "set" for the
> > result of single_set, and hence not overwriting "insn" within the loop.
> > That said I've only tested that it compiles for rl78, I've not yet
> > forced line 3605 to execute, and not simulated the resulting code.
> See
On Fri, 5 Sep 2014, Manuel L?pez-Ib??ez wrote:
> gcc/ChangeLog:
>
> 2014-09-05 Manuel L?pez-Ib??ez
>
> * doc/invoke.texi (Wnormalized=): Update.
>
> libcpp/ChangeLog:
>
> 2014-09-05 Manuel L?pez-Ib??ez
>
> * include/cpplib.h (struct cpp_options): Declare warn_normalize as
>
On Fri, Jun 27, 2014 at 09:13:07AM +0200, Jakub Jelinek wrote:
> The patch adds two new (trivial handlers) to libubsan, as it is maintained
> in llvm's compiler-rt, will talk to them if they are interested in those
> and what exact wording and form (AFAIK clang also added the gcc
> {,returns_}nonnu
Hi!
On Fri, Jun 27, 2014 at 09:13:07AM +0200, Jakub Jelinek wrote:
> The patch adds two new (trivial handlers) to libubsan, as it is maintained
> in llvm's compiler-rt, will talk to them if they are interested in those
> and what exact wording and form (AFAIK clang also added the gcc
> {,returns_}
On Tue, 2014-09-09 at 09:53 -0600, Jeff Law wrote:
> On 09/08/14 15:29, David Malcolm wrote:
> > gcc/ChangeLog:
[...]
> > * config/avr/avr.c (avr_out_plus): Add checked cast to rtx_insn *
> > when invoking single_set in region guarded by INSN_P.
> > (avr_out_bitop): Likewise.
> > (_
On 09/09/14 16:58, Xinliang David Li wrote:
> The attached patch fixed the problem reported in PR63209. The problem
> exists in both gcc4.9 and trunk.
>
> Regression test on arm-unknown-linux-gnueabi passes.
>
> Ok for gcc-4.9 and trunk?
>
No, this isn't right. I don't think you need a new pat
This was a tricky one, because std::deque always has a map of nodes
allocated, so needs to re-allocate in its move constructor. That makes
move assignment complicated, as the rvalue object must be left in a
valid state owning memory that can be freed by its (possibly
moved-from) allocator.
The _M
On 14/08/14 21:22 +0200, François Dumont wrote:
I am preparing a patch for profile mode so I will submit modification
for this mode with this big patch.
btw, François, for profile mode I think we should just do something
like this patch.
I feel quite strongly that if using Debug Mode or Profil
On Tue, Sep 9, 2014 at 12:21 PM, Bill Schmidt
wrote:
> Hi,
>
> When collecting my list of invalid UNSPECs for the analyze_swaps pass, I
> missed UNSPEC_VSX_CVDPSPN (which looked too much like
> UNSPEC_VSX_CVSPDPN). This patch adds it to the list. I have an idea
> for allowing it to be used, but
On 09/04/2014 10:30 AM, Zamyatin, Igor wrote:
>
>> -Original Message-
>> From: gcc-patches-ow...@gcc.gnu.org [mailto:gcc-patches-
>> ow...@gcc.gnu.org] On Behalf Of Vladimir Makarov
>> Sent: Thursday, September 04, 2014 12:19 AM
>> To: Ilya Enkovich
>> Cc: g...@gnu.org; gcc-patches; Evgeny
Hi,
When collecting my list of invalid UNSPECs for the analyze_swaps pass, I
missed UNSPEC_VSX_CVDPSPN (which looked too much like
UNSPEC_VSX_CVSPDPN). This patch adds it to the list. I have an idea
for allowing it to be used, but won't have time to work on it now, so I
added some commentary to
The attached patch fixed the problem reported in PR63209. The problem
exists in both gcc4.9 and trunk.
Regression test on arm-unknown-linux-gnueabi passes.
Ok for gcc-4.9 and trunk?
(I sent the patch last night, but it got lost somehow)
David
Index: ChangeLog
==
On 09/08/14 15:29, David Malcolm wrote:
gcc/ChangeLog:
* rtl.h (single_set_2): Strengthen first param from const_rtx to
const rtx_insn *, and move prototype to above...
(single_set): ...this. Convert this from a macro to an inline
function, enforcing the requireme
$ arm-none-eabi-gcc ./foo.c -S -o -
# before this patch
...
.bss
.align 2
bufbuf.4078:
.space 8
.ident "GCC: (GNU) 5.0.0 20140908 (experimental)"
# after this patch (add size directive )
...
.bss
.align 2
bufbuf.4078:
.space 8
.size bufbuf.4078, 8
.ident "GCC: (GNU) 5.0.0 201
On 09/08/14 15:29, David Malcolm wrote:
gcc/ChangeLog:
* caller-save.c (rtx saveinsn): Strengthen this variable from rtx
to rtx_insn *.
(restinsn): Likewise.
* config/aarch64/aarch64-protos.h (aarch64_simd_attr_length_move):
Likewise for param.
* co
Trevor Saunders writes:
> I also do think the advantages of using members outways the cost.
>
> For one thing functions with all caps names are just weird. I think the
> more important reason though is that it will help make rtx_insn be a
> separate class sometime in the far future, since at some
Hi all,
As Christophe mentioned at
https://gcc.gnu.org/ml/gcc-patches/2014-09/msg00202.html
These tests fail on big-endian. The reason is that the input is not
aligned to 128 bit forcing the use of a movmisalign which we don't
support on big-endian.
A solution is to force the alignment of th
Attached is a further revision of the patch, now dealing with check-c++.
Roughly 50% speedup here at '-j32' (18m vs 12m). For my setup
(--enable-languages=c,c++,fortran) I have now improved all targets called in
'make -j32 -k check'. The latter is now 30% faster (15m vs 20m). Note that
there ar
Now with gzipped figure.. why do these bounce ?
> But if there are jobs that just take 1s to complete, then clearly it doesn't
> make sense to split them off as separate job. I think we don't need 100%
> even split, but at least roughly is highly desirable.
Let me add some data, attached is a gr
On 09/08/14 15:29, David Malcolm wrote:
gcc/
* rtl.h (INSN_LOCATION): Strengthen param from const_rtx to
const rtx_insn *, and from rtx to rtx_insn * for the other
overloaded variant.
(RTL_LOCATION): Add a checked cast to rtx_insn * when invoking
INSN_LOCAT
On 09/08/14 15:29, David Malcolm wrote:
gcc/ChangeLog:
* combine.c (try_combine): Eliminate checked cast on result of
gen_rtx_INSN.
* emit-rtl.c (gen_rtx_INSN): New function, improving over the prior
autogenerated one by strengthening the return type and params 2 a
On 09/09/2014 06:33 PM, Jakub Jelinek wrote:
On Tue, Sep 09, 2014 at 06:27:10PM +0400, Yury Gribov wrote:
On 09/09/2014 06:14 PM, VandeVondele Joost wrote:
I certainly don't want to claim that the patch I have now is perfect,
it is rather an incremental improvement on the current setup.
I'd s
On 09/08/14 15:29, David Malcolm wrote:
gcc/
* ira.c (ira_update_equiv_info_by_shuffle_insn): Use NULL rather
than NULL_RTX.
(no_equiv): Likewise.
(update_equiv_regs): Likewise.
(setup_reg_equiv): Likewise. Strengthen locals "elem",
"prev_elem", "n
On 09/08/14 15:29, David Malcolm wrote:
Here's another 5 patches that make use of rtx_def subclasses, which
are non-trivial enough *not* to fall under the pre-approval Jeff
granted.
In particular, patches 4 and 5 update the signature of recog_memoized
and single_set respectively to require an rt
On Tue, Sep 9, 2014 at 4:30 PM, Jakub Jelinek wrote:
> On Tue, Sep 09, 2014 at 04:25:23PM +0200, Richard Biener wrote:
>> why wasn't 'result' built using simplify_gen_* in the first place? I also
>
> It is built using cselib_expand_value_rtx_cb, which calls the various
> simplify_*_operation and
Following this, we're seeing ICEs in tests in gcc.dg/pch.exp and g++.dg/pch.exp,
with cross-builds (hosted on x86_64) targetting bare metal AArch64 and ARM
(aarch64-none-elf, aarch64_be-none-elf and arm-none-eabi; I haven't tested
armeb-none-eabi; builds targeting linux are OK), for *release build
On Tue, Sep 09, 2014 at 06:27:10PM +0400, Yury Gribov wrote:
> On 09/09/2014 06:14 PM, VandeVondele Joost wrote:
> >I certainly don't want to claim that the patch I have now is perfect,
> >it is rather an incremental improvement on the current setup.
>
> I'd second this. Writing patterns manually
On Tue, Sep 09, 2014 at 04:25:23PM +0200, Richard Biener wrote:
> why wasn't 'result' built using simplify_gen_* in the first place? I also
It is built using cselib_expand_value_rtx_cb, which calls the various
simplify_*_operation and simplify_rtx too.
> note that debug_insns can have all sorts
On 09/09/2014 06:14 PM, VandeVondele Joost wrote:
I certainly don't want to claim that the patch I have now is perfect,
it is rather an incremental improvement on the current setup.
I'd second this. Writing patterns manually seems rather inefficient and
error-prone
(not undoable of course but
On Tue, Sep 9, 2014 at 1:50 PM, Alan Modra wrote:
> On Fri, Sep 05, 2014 at 11:00:04AM +0930, Alan Modra wrote:
>> Of course it would be better to repair the damage done to debug info
>> rather than rejecting it outright..
>
> This cures PR60655 on PowerPC by passing the horrible debug_loc
> expre
> If you get whitespace right, one can provide multiple different wildcards to
> a single *.exp file, e.g.
> make check-gcc RUNTESTFLAGS="dg.exp='p[0-9A-Za-qs-z]* pr[9A-Za-z]*'" should
> cover all tests starting with p other than pr[0-8]*.c (where you could split
> say pr[0-2]* into another job, pr
Kito Cheng writes:
> .align 2
> bufbuf.4078:
> .space 8
> ...
> .ident "GCC: (GNU) 5.0.0 20140828 (experimental)"
>
>
> And then the size info will missing:
>
> $ arm-none-eabi-objdump ./foo.o -t
>
> ./foo.o: file format elf32-littlearm
>
> SYMBOL TABLE:
> ldf *ABS* zoo
Richard Earnshaw writes:
> .space puts space in the current section, I believe. .comm and .lcomm
> put the space in the common section; that's very different behaviour.
The local common section is the .bss section.
Andreas.
--
Andreas Schwab, SUSE Labs, sch...@suse.de
GPG Key fingerprint = 0
On Mon, Sep 8, 2014 at 6:41 PM, Segher Boessenkool
wrote:
> When reload decides it needs to reload something that was assigned a
> register before it will only look at the contraint, not the predicate.
> This means that for the *bool3 I added with predicate logical_operand
> (that is registers, an
On 09/09/14 14:32, Ramana Radhakrishnan wrote:
> On Mon, Sep 1, 2014 at 4:30 AM, Kito Cheng wrote:
>> Hi all:
>>
>> In arm-*-elf target some variable will missing size directive,
>>
>> for example:
>>
>> foo.c:
>>
>> void foo (void) {
>> static char bufbuf[8];
>> }
>>
>> $ arm-none-eabi-gcc ./fo
On Mon, Sep 1, 2014 at 4:30 AM, Kito Cheng wrote:
> Hi all:
>
> In arm-*-elf target some variable will missing size directive,
>
> for example:
>
> foo.c:
>
> void foo (void) {
> static char bufbuf[8];
> }
>
> $ arm-none-eabi-gcc ./foo.c -S -o -
>
> ...
> .align 2
> bufbuf.4078:
> .space 8
Th
On Tue, Sep 09, 2014 at 10:57:09AM +, VandeVondele Joost wrote:
> > No. As I wrote earlier, splitting on filenames and test counts only is only
> > very rough split, all the splits really need to be backed out by real timing
> > data from popular targets.
>
> Furthermore, for parallel perf
Quote shell variables that may contain whitespace in the configure
file. I cannot remember exactly whether I actually ran into a
problem, but the patch won't hurt anyway.
ChangeLog
2014-09-05 Dominik Vogt
* libgo/configure.ac (GO_SYSCALL_OS_ARCH_FILE): Use double quotes around
This is an optional extension of the -fgo-dump-spec option to
handle copmplex data types. It is not necessary for s390[x] but
was easy to write on top of the previous patch.
gcc/ChangeLog
2014-09-05 Dominik Vogt
* godump.c (go_format_type): Represent "float _Complex" and
"doub
This patch extends the -fdump-go-spec option to handle bitfields
and unions and fixes handlinx of zero length arrays. All of this
is necessary for s390[x] since the system headers use these
features. Please check the commit comment for a detailed
description of the patch.
gcc/ChangeLog
-
Eases the rediculously tight minimum stack size for goprocesses on
64 bit systems.
ChangeLog
2014-09-05 Dominik Vogt
* libgo/runtime/proc.c (runtime_newosproc): Set the thread stack size
explicitly only on 32 bit systems.
Ciao
Dominik ^_^ ^_^
--
Dominik Vogt
IBM Germany
>
This optional cleanup patch fixes some sloppy programming in the
x86 libgo/go/debug/elf library that had given me a very hard time
to debug and fix when porting the code to s390[x]. See commit
comment for details.
ChangeLog
2014-09-05 Dominik Vogt
* libgo/go/debug/elf/file.go (applyRe
This patch fixes the compiler flags in libgo/mksysinfo.sh. In one
place, some compiler flags were missing that are consistently used
elswhere, resulting in an error message.
ChangeLog
2014-09-05 Dominik Vogt
* libgo/mksysinfo.sh (OUT):
Add the same compile flags that configure
The current Gccgo cannot handle 64 bit symbol tables on s390[x].
The attached patch fixes that.
gcc/go/ChangeLog
2014-09-05 Dominik Vogt
* gofrontend/import-archive.cc (interpret_header): Recognize 64-bit
symbol tables ("/SYM64/ ").
Ciao
Dominik ^_^ ^_^
--
Dominik
A test case added to golang for the previous patch.
gcc/testsuite/ChangeLog
2014-09-05 Dominik Vogt
* go.test/test/recover.go (test1): Test recover() from deferred
recursive function.
Ciao
Dominik ^_^ ^_^
--
Dominik Vogt
IBM Germany
>From a3cc3d043d580d083f5dd573cb66d299c
This patch fixes bug 60406 in an architecture independent way.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60406
The key change in this patch is to use
_Unwind_FindEnclosingFunction() to identify the defering function.
However, this does not work on platforms that still use SJLJ
exceptions.
Cha
The following series' of patches introduce s390[x] support for
Gccgo. For practical reasons they will be submitted in separate
parts. This is the first part of the submissions with
architecture independent bug fixes and enhancements that affect
the following directories:
gcc
On Tue, Sep 9, 2014 at 1:41 PM, Kirill Yukhin wrote:
> On 30 Aug 10:21, Uros Bizjak wrote:
> Hello,
>> It looks to me that _SCALAR unspecs are redundant, and should be
>> possible to use UNSPEC_REDUCE for all patterns without unwanted
>> matching.
> Updated patch in the bottom
> gcc/
> * c
On Tue, Sep 9, 2014 at 1:28 PM, Kirill Yukhin wrote:
> Hello Uroš,
> On 28 Aug 20:44, Uros Bizjak wrote:
>> On Thu, Aug 28, 2014 at 3:48 PM, Kirill Yukhin
>> wrote:
>> Please document these changes as:
>>
>> [new_pattern_name]: Rename from [old_pattern_name] and use VF_AVXyy
>> mode iterator.
>>
On 9 September 2014 12:03, wrote:
>
>
>> On Sep 9, 2014, at 2:50 AM, Marcus Shawcroft
>> wrote:
>>
>> +static unsigned HOST_WIDE_INT
>> +aarch64_asan_shadow_offset (void)
>> +{
>> + return (HOST_WIDE_INT_1 << 36);
>> +}
>> +
>>
>> Looking around various other ports I see magic numbers includin
-Winvalid-offsetof was introduced before __builtin_offsetof, and the
comments both in the code and the testsuite note that even though
offsetof on a non-standard-layout type is undefined, the usual expansion
seems to be well-defined. So -Winvalid-offsetof should move to
offsetof-specific code,
On Fri, Sep 05, 2014 at 11:00:04AM +0930, Alan Modra wrote:
> Of course it would be better to repair the damage done to debug info
> rather than rejecting it outright..
This cures PR60655 on PowerPC by passing the horrible debug_loc
expressions we have through simplify_rtx. Not only do we get
On 30 Aug 10:21, Uros Bizjak wrote:
Hello,
> It looks to me that _SCALAR unspecs are redundant, and should be
> possible to use UNSPEC_REDUCE for all patterns without unwanted
> matching.
Updated patch in the bottom
gcc/
* config/i386/i386.c
(ix86_expand_args_builtin): Handle avx512
Hello Uroš,
On 30 Aug 10:37, Uros Bizjak wrote:
> On Fri, Aug 29, 2014 at 4:00 PM, Kirill Yukhin
> wrote:
> Please note that you can use UNSPEC_COMPRESS everywhere, there is no
> need for UNSPEC_COMPRESS_STORE and can be deleted. On a related note,
> it looks to me that UNSPEC_COMPRESS patterns c
This adds the capability to specify a type to be used when building
an expression.
Committed.
2014-09-09 Richard Biener
* genmatch.c (struct expr): Add expr_type member.
(expr::gen_transform): Honor explicitely specified type.
(parse_expr): Parse ':' on the operator a
Hello Uroš,
On 28 Aug 20:44, Uros Bizjak wrote:
> On Thu, Aug 28, 2014 at 3:48 PM, Kirill Yukhin
> wrote:
> Please document these changes as:
>
> [new_pattern_name]: Rename from [old_pattern_name] and use VF_AVXyy
> mode iterator.
>
> > -(define_insn "fma_fmsub_"
> > +(define_insn
> > "fma_fms
Hello,
The branch was rebased and updated again:
https://gcc.gnu.org/git/?p=gcc.git;a=shortlog;h=refs/heads/kyukhin/gomp4-offload
Now offloading works without -flto option. Also there is no need to set
LIBGOMP_PLUGIN_PATH.
https://gcc.gnu.org/wiki/Offloading was updated accordingly.
All tests
On Tue, Sep 09, 2014 at 10:55:40AM +0100, Ramana Radhakrishnan wrote:
> On Sun, Aug 31, 2014 at 4:45 PM, Gerald Pfeifer wrote:
> > On Wed, 5 Feb 2014, James Greenhalgh wrote:
> >> As far as I know the behaviour of this flag has always been this way.
> >> So is this also OK to backport to release b
On Tue, Aug 19, 2014 at 4:04 PM, Kyrill Tkachov wrote:
> Hi all,
>
> This patch updates some FP load/store multiple patterns and in the
> vfp_output_vstmd case when the instruction is used as a push to sp it now
> emits a vpush mnemonic instead of vstmdb.
>
> Ok for trunk?
Ok - thanks,
Ramana
>
On Tue, Aug 19, 2014 at 4:04 PM, Kyrill Tkachov wrote:
> Hi all,
>
> The ARM ARM explicitly says that fmstat is translated into 'vmrs APSR_nzcvm,
> FPSCR' in UAL syntax so this patch does that.
>
> Ok for trunk?
Ok
Ramana
>
> Thanks,
> Kyrill
>
> 2014-08-19 Kyrylo Tkachov
>
> * config/arm
On Tue, Aug 19, 2014 at 4:04 PM, Kyrill Tkachov wrote:
> Hi all,
>
> These are the fp<->fp and int<->fp convert patterns. IMHO these are much
> more readable than the pre-UAL ones because they use two suffixes to
> indicate which data types are being converted to and from.
>
> Ok for trunk?
>
> Th
On Tue, Aug 19, 2014 at 4:04 PM, Kyrill Tkachov wrote:
> Hi all,
>
> The sqrt and floating point compare patterns.
> For the case when we compare with floating-point 0 the ARM ARM uses the
> syntax 'vcmp {s,d}, #0.0'
> but current gas has a bug and doesn't accept that form, only 'vcmp {s,d},
> #0'
On Tue, Aug 19, 2014 at 4:04 PM, Kyrill Tkachov wrote:
> Hi all,
>
> This patch switches over the FP mul+add, mul+neg+add etc patterns. Mixing
> some of the mnemonics up is a danger but the ARM ARM has a handy section
> that maps each mnemonic to its UAL equivalent.
>
> Ok for trunk?
>
Ok this is
On Tue, Aug 19, 2014 at 4:22 PM, Kyrill Tkachov wrote:
> Hi all,
>
> Nothing too controversial here, convert the concerned patterns to UAL.
> The size of the data types in the operation is expressed in the .f32 or
> .f64 suffix.
>
> Ok for trunk?
Ok - in an ideal world I'd have just done these wi
On Tue, Aug 19, 2014 at 4:22 PM, Kyrill Tkachov wrote:
> Hi all,
>
> In this patch the move patterns are updated.
> For the fconst case where the constant is encoded in a decimal
> representation before going into the immediate field of the assembly
> instruction UAL syntax allows for the real ope
On Tue, Aug 19, 2014 at 4:04 PM, Kyrill Tkachov wrote:
> Hi all,
>
> This patch series converts the arm backend to output unified assembly syntax
> for the VFP instructions.
> This makes it more readable since most UAL mnemonics also include various
> type suffixes such as .f32 and .f64 that quick
On Mon, Aug 18, 2014 at 11:31 AM, Jiong Wang wrote:
> this patch enable auto-vectorization for copysignf by using vector
> bit selection instruction on arm32 when neon available.
>
> for a simple testcase:
>
> for (i = 0; i < N; i++)
> r[i] = __builtin_copysignf (a[i], b[i]);
>
>
> assuming
> No. As I wrote earlier, splitting on filenames and test counts only is only
> very rough split, all the splits really need to be backed out by real timing
> data from popular targets.
I'm actually doing quite some testing trying to get a reasonable balance,
checking 'completed in' in all *.l
On 8 September 2014 17:39, Alan Lawrence wrote:
> gcc/ChangeLog:
>
> * config/aarch64/arm_neon.h (aarch64_vset_lane_any): New (*2).
> (vset_lane_f32, vset_lane_f64, vset_lane_p8, vset_lane_p16,
> vset_lane_s8, vset_lane_s16, vset_lane_s32, vset_lane_s64,
> vset_lan
On 8 September 2014 17:31, Alan Lawrence wrote:
> This adds a test thath checks the result of a vset_lane intrinsic is
> identical to the input apart from one value being changed.
>
> Test checks only one index per vset_lane_xxx in a somewhat adhoc fashion as
> the index has to be a compile-time i
On 8 September 2014 17:45, Alan Lawrence wrote:
> Now that float64x1 is a vector type, we can convert between it and other
> same-sized vector types with a simple cast, as is already done for all other
> reinterprets not involving float64x1.
>
> This allows to remove the reinterpretv1df and reinte
On 9 September 2014 11:20, Alan Lawrence wrote:
> The 'scan-assembler-not not' test in
> gcc.target/aarch64/simd/int_comparisons_1.c fails on aarch64-linux-gnu
> because the compiler adds a ".note" at the end of the .s file. This patch
> tightens the regex to only match a not with surrounding whit
On 04/09/14 07:08, Bin Cheng wrote:
> @@ -1872,7 +1892,9 @@ const struct tune_params arm_cortex_a53_tune =
>{true, true}, /* Prefer non short
> circuit. */
>&arm_default_vec_cost, /* Vectorizer costs. */
>false,
On 09/09/14 17:47 +0800, Tony Wang wrote:
Hi there,
This is a updated patch to fix pr56846. Bootstrapped on
x86_64-unknown-linux-gnu and no regression for
regression test on Linux and a cross regression test on arm-none-eabi.
OK for the trunk?
gcc/libstdc++-v3/ChangeLog:
2014-09-09 Tony Wan
> +# ls -1 | ../../../contrib/generate_tcl_patterns.sh 300
> "dg.exp=gfortran.dg/"
>
> How does this work with subdirectories? Can we replace ls with find?
The input to the script is general, you can use this to your advantage. For
example, I've been using:
ls -1 g++.*/* | cut -c5- | ../../.
On Tue, Sep 9, 2014 at 12:06 PM, Kugan
wrote:
>
>
> On 08/09/14 19:48, Richard Biener wrote:
>> On Sun, Sep 7, 2014 at 11:50 AM, Kugan
>> wrote:
>>> On 05/09/14 19:50, Richard Biener wrote:
>>>
Well - the best way would be to expose the target specifics to GIMPLE
at some point in the op
On 8 September 2014 12:03, Alan Lawrence wrote:
> This adds a test of all the variants of vst2, vst2q, vst3, vst3q, vst4, and
> vst4q. These all use typexNxM structs and the OI/CI/XImode mechanism, so the
> test cross-checks this against plain ol' vld1(q?).
>
> Cross-tested on aarch64-none-elf (pa
On 8 September 2014 11:49, Alan Lawrence wrote:
> At present there is no test coverage of the vld2_lane, vld2q_lane,
> vld3_lane, vld3q_lane, vld4_lane, vld4q_lane intrinsics. So this adds a test
> using the vld1 and vst1 intrinsics.
>
> Passing on aarch64-none-elf.
>
> Failing on aarch64_be-none-
+
+main (int argc, char **argv)
+{
+
int main ?
Otherwise OK /Marcus
On 8 September 2014 11:43, Alan Lawrence wrote:
> This adds a test of the vld2_dup, vld2q_dup, vld3_dup, vld3q_dup, vld4_dup
> and vld4q_dup instrinsics.
>
> Passing on aarch64-none-elf and aarch64_be-none-elf.
>
> gcc/testsu
1 - 100 of 122 matches
Mail list logo
