gcc.target/riscv/attribute-10.c will failed on rv64 target because it
the abi is incompatible with rv64, added a valid arch before the
incorrect -march option could resolve it.
From 727069fa18d0c0351391ae4a5e9661023f0ca12e Mon Sep 17 00:00:00 2001
From: kito
Date: Thu, 1 Aug 2019 06:33:48 +
Su
Hi!
I've backported following 5 patches from trunk to 9.2,
bootstrapped/regtested on x86_64-linux and i686-linux, committed to
gcc-9-branch.
Jakub
2019-08-01 Jakub Jelinek
Backported from mainline
2019-07-17 Jakub Jelinek
PR tree-optimization/91157
More extensive testing of the last week's strlen patch for
PR91183 on various non-mainstream targets and with better tests
has exposed a few gaps and a couple of bugs. The attached patch
addresses all in one change. I considered splitting it up but
in the end decided the changes were small and s
On 7/31/19 4:36 AM, Richard Biener wrote:
On Tue, Jul 30, 2019 at 4:54 PM Andrew MacLeod wrote:
Everything in this compiler is historical. We did everything we could to
save memory back in the old days, I know, I was there when we did this.
Combining the lattice and the range made sense in
It was pointed out recently in another forum that GCC doesn't
document attribute alias for variables. It was also noted in
the same discussion that the semantics of accessing aliases
and their targets can have "surprising" effects because GCC
(as do other compilers) assumes that distinct declarat
On Wed, Jul 31, 2019 at 08:43:50PM +0200, Marc Glisse wrote:
> On Wed, 31 Jul 2019, Jonathan Wakely wrote:
>
> >So something like the attached patch. The glibc says the
> >values I used have enough digits for IEEE quad-precision:
> >
> >/* The above constants are not adequate for computation usi
Hi any feedback on this?
Thanks.
On Tue, Jul 23, 2019 at 02:49:13PM -0400, Arvind Sankar wrote:
> There seems to be a typo in gcc/jit/Make-lang.in where it references an
> undefined LIBGCCJIT_SYMLINK instead of LIBGCCJIT_SONAME_SYMLINK.
> ---
> gcc/jit/Make-lang.in | 5 ++---
> 1 file changed, 2
On Wed, Jul 31, 2019 at 9:41 AM Maxim Blinov wrote:
> gcc/ChangeLog:
> 2019-07-31 Maxim Blinov
>
> * common/config/riscv/riscv-common.c: Check -march string ends
> with null.
>
> gcc/testsuite/ChangeLog:
> 2019-07-31 Maxim Blinov
>
> * gcc.target/riscv/attribute-10.c:
On 31/07/19 21:16 +0200, Jakub Jelinek wrote:
On Wed, Jul 31, 2019 at 07:58:55PM +0100, Jonathan Wakely wrote:
> Perhaps add __extension__ before the literals too?
Does that do anything for Q literals? I thought I'd tried it
previously and decided it didn't. I'm happy to add it though.
You're
On Wed, 2019-07-31 at 15:14 -0400, Jason Merrill wrote:
> On 7/24/19 5:24 PM, Paolo Carlini wrote:
> > this takes care of the four locations, two warnings and two errors.
> > I'm
> > also adding the missing complain & tf_warning or tf_error guards,
> > I
> > don't have a SFINAE testcase failing b
One of the features of constexpr is that it doesn't allow UB; and such UB must
be detected at compile-time. So running your code in a context that requires
a constant expression should ensure that the code in question is free of UB.
In effect, constexpr can serve as a sanitizer. E.g. this article
On Wed, Jul 31, 2019 at 07:58:55PM +0100, Jonathan Wakely wrote:
> > Perhaps add __extension__ before the literals too?
>
> Does that do anything for Q literals? I thought I'd tried it
> previously and decided it didn't. I'm happy to add it though.
You're right, it works well that way only in C,
On 29/07/19 15:27 +0100, Jonathan Wakely wrote:
The recursive_init_error class is defined in a header, with an inline
constructor, but the definition of the vtable and destructor are not
exported from the shared library. With -fkeep-inline-functions the
constructor gets emitted in user code, and
On 7/24/19 5:24 PM, Paolo Carlini wrote:
this takes care of the four locations, two warnings and two errors. I'm
also adding the missing complain & tf_warning or tf_error guards, I
don't have a SFINAE testcase failing but since the function takes a
tsubst_flags_t argument I think it's the right
* include/std/memory (make_obj_using_allocator): Qualify call to
uses_allocator_construction_args.
Tested x86_64-linux, committed to trunk
Backport to gcc-9-branch to follow.
commit 0c3543005cee13158806fe2909de9920522d2f9e
Author: redi
Date: Wed Jul 31 19:08:56 2019 +
On 31/07/19 20:50 +0200, Jakub Jelinek wrote:
On Wed, Jul 31, 2019 at 07:37:03PM +0100, Jonathan Wakely wrote:
On 31/07/19 19:07 +0200, Marc Glisse wrote:
> On Wed, 31 Jul 2019, Jonathan Wakely wrote:
>
> > The values of the constants are taken from Glibc where the equivalent
> > constant exists
On 7/15/19 9:47 AM, Matthew Beliveau wrote:
Okay I kept the TYPE_MAIN_VARIANT and dropped the accidental new line!
Hopefully this should be fine!
OK, thanks.
Jason
On 7/12/19 4:16 PM, Marek Polacek wrote:
In order to fix 89906 in GCC 8, we need to backport 86098.
I think the patch is safe to be backported.
Tested x86_64-linux, ok for 8?
OK.
Jason
On 7/17/19 1:51 PM, Marek Polacek wrote:
This was a wrong code issue where we printed
2 3 1
1 2 3
instead of
1 2 3
1 2 3
but it was fixed by r271705. I don't know of a good way to check
the auto... expansion here so I used dg-output.
You could call another function with the expansion and check
On Wed, Jul 31, 2019 at 07:37:03PM +0100, Jonathan Wakely wrote:
> On 31/07/19 19:07 +0200, Marc Glisse wrote:
> > On Wed, 31 Jul 2019, Jonathan Wakely wrote:
> >
> > > The values of the constants are taken from Glibc where the equivalent
> > > constant exists, or by rounding the actual constant t
Previously, with init-capture the type of the closure field was a
DECLTYPE_TYPE of the initializer. But since each time we tsubst a lambda we
get a different lambda, that meant that if the initializer is a lambda, we'd
end up with different closure types in the field and initializer after
substitu
build_clone uses copy_node to duplicate the TEMPLATE_INFO for a clone, but
this clears TREE_CHAIN, which was TI_ARGS in a TEMPLATE_INFO.
Tested x86_64-pc-linux-gnu, applying to trunk.
* cp-tree.h (struct tree_template_info): Use tree_base instead of
tree_common. Add tmpl and args
On Wed, 31 Jul 2019, Jonathan Wakely wrote:
So something like the attached patch. The glibc says the
values I used have enough digits for IEEE quad-precision:
/* The above constants are not adequate for computation using `long double's.
Therefore we provide as an extension constants with sim
On 31/07/19 19:07 +0200, Marc Glisse wrote:
On Wed, 31 Jul 2019, Jonathan Wakely wrote:
The values of the constants are taken from Glibc where the equivalent
constant exists, or by rounding the actual constant to the same number
of digits as the Glibc constants have.
How does it behave with _
Hi,
noticed this when checking a qsort patch, applying as obvious.
In the C days, -Wmissing-prototypes would have caught this, but C++ needs a
different warning (-Wmissing-declarations), so we get inevitable bitrot :/
Alexander
* ipa-devirt.c (type_warning_cmp): Make static.
(de
On 7/26/19 9:28 AM, Iain Sandoe wrote:
> The patch needs one small amendment to succeed in bootstrap.
> I applied the amended to i686, x86_64 and powerpc Darwin with no apparent
> new problems. From the Darwin perspective, the patch is OK with the
> amendment.
Ok, thanks. For completeness, I h
On Wed, 31 Jul 2019, Jonathan Wakely wrote:
The values of the constants are taken from Glibc where the equivalent
constant exists, or by rounding the actual constant to the same number
of digits as the Glibc constants have.
How does it behave with __float128? I think that with -std=gnu++2a, it
ping
PR81800 is about the lrint inline giving spurious FE_INEXACT exceptions.
The previous change for PR81800 didn't fix this: when lrint is disabled
in the backend, the midend will simply use llrint. This actually makes
things worse since llrint now also ignores FE_INVALID exceptions!
The
ping
In aarch64_classify_symbol symbols are allowed full-range offsets on
relocations.
This means the offset can use all of the +/-4GB offset, leaving no offset
available
for the symbol itself. This results in relocation overflow and link-time
errors
for simple expressions like &globa
ping
With -mcpu=generic the function alignment is currently 8, however almost all
supported cores prefer 16 or higher, so increase the default to 16:12.
This gives ~0.2% performance increase on SPECINT2017, while codesize is 0.12%
larger.
ChangeLog:
2019-05-31 Wilco Dijkstra
ping
Remove the remaining Neon adddi3, subdi3 and negdi2 patterns. As a result
adddi3, subdi3 and negdi2 can now always be expanded early irrespectively of
whether Neon is available. Also expand the extenddi patterns at the same
time. Several Neon arch attributes are no longer used and
Hi Martin, thanks for reviewing.
> I would expect the missing quotes around the option to trigger
> a -Wformat-diag warning. The %<%s%s> should also be flagged by
> the same warning. Changing the format string as follows should
> avoid the warnings:
>
> error_at (loc, "%<-march=%s%>: unexpect
The values of the constants are taken from Glibc where the equivalent
constant exists, or by rounding the actual constant to the same number
of digits as the Glibc constants have.
P0631R4 Math Constants
* include/Makefile.am: Add new header.
* include/Makefile.in: Regenera
ping
Like the logical operations, expand all shifts early rather than only
sometimes. The Neon shift expansions are never emitted (not even with
-fneon-for-64bits), so they are not useful. So all the late expansions
and Neon shift patterns can be removed, and shifts are more optimized
a
ping
Cleanup the logical DImode operations since the current implementation is way
too complicated. Thumb-1, Thumb-2, VFP/Neon and iwMMXt all work differently,
resulting in a bewildering number of expansions, patterns and splits across
several md files. All this complexity is counterpr
On 7/31/19 5:19 AM, Maxim Blinov wrote:
This patch adds the same check that is present in binutils/bfd/elfxx-riscv.c.
Checks that we have reached the end of the string after all the
parsing routines have been run. Without this check, GCC silently
succeeds on erroneous input, and produces an asse
On 7/31/19 12:01 PM, Martin Liška wrote:
> Anyway, the auto-detection of jobserver seems very ugly to me :/
> I tend to remove the support and start working on a proper implementation
> that can be used not only by make build system.
Can you Jakub test the attached patch please?
Do it reach a ulim
Here is the patch for
* Implement C++20 p0202 - Add constexpr Modifiers to Functions in
and Headers.
* Implement C++20 p1023 - constexpr comparison operators for std::array.
Relative to the last effort it is rebased on more recent trunk and I
added to .
There's some chance that I'll have
On 7/31/19 4:00 AM, Richard Biener wrote:
On Tue, 30 Jul 2019, Martin Sebor wrote:
On 7/30/19 9:19 AM, Jakub Jelinek wrote:
On Tue, Jul 30, 2019 at 09:16:45AM -0600, Martin Sebor wrote:
Say something like this POC:
We have those already, see ctz_hwi, clz_hwi etc.
The thing is that BITMAP_WO
On Tue, 16 Jul 2019 at 14:42, Kyrill Tkachov
wrote:
>
>
> On 7/16/19 12:18 PM, Kyrill Tkachov wrote:
> > Hi Christophe
> >
> > On 5/15/19 1:39 PM, Christophe Lyon wrote:
> > > In FDPIC mode, the trampoline generated to support pointers to nested
> > > functions looks like:
> > >
> > >.
* include/std/bit: Add Doxygen comments.
Committed to trunk.
At some point I need to resume working on Doxygen improvements and
switch the Doxygen generation to C++17 mode, so we document all the
new C++17 stuff.
commit c76147753665e66ac8258f9b471c5cb65e28067e
Author: redi
Date: Wed
On Tue, Jul 30, 2019 at 09:22:00AM +0200, Jakub Jelinek wrote:
> Hi!
>
> Neither c_expr_sizeof_expr nor c_expr_sizeof_type bother with filling up
> the locations in c_expr struct they return. Normally, this isn't a problem,
> as the sole caller of those calls set_c_expr_source_range. It doesn't
Committed as revision 273933.
Dave
On 2019-06-14 11:44 a.m., co...@sdf.org wrote:
> This adds netbsd/hppa support. I tested it on the shiny new QEMU-git
> which can now boot NetBSD too :-)
>
> Files are very similar to the linux code.
>
> Please let me know if any changes need to be made.
>
> Mat
Martin Liška schrieb:
On 7/29/19 3:46 PM, Georg-Johann Lay wrote:
Hi Martin.
In SVN r273662 you changed a comment in avr.c from __gnu_lto_v1 to
__gnu_lto_slim without changing the relevant code:
Yes.
/* __gnu_lto_slim is just a marker for the linker injected by toplev.c.
There is no
On Wed, Jul 31, 2019 at 11:30 AM Jakub Jelinek wrote:
>
> Hi!
>
> On Wed, Jul 31, 2019 at 10:51:22AM +0200, Uros Bizjak wrote:
> > OK.
>
> Thanks. This follow-up implements the same for mmx with sse for V8QImode,
> the testcase shows that it is useful too. The difference is quite large:
>
> -
There's some longer latent issue with PTAs handling of
offsetted MEM_REFs in get_constraint_for_component_ref when there
are subfields involved. The following solves this by handling
offsetted MEM_REFs manually.
Bootstrap & regtest running on x86_64-unknown-linux-gnu.
Richard.
2019-07-31 Ric
On 30/07/2019 21:51, Bernd Edlinger wrote:
+/* { dg-options "-marm -march=armv6 -mno-unaligned-access -mfloat-abi=soft
-mabi=aapcs -O3" } */
This isn't going to work as-is, we test many combinations of the
compiler, either with explicit dejagnu settings or with the compiler
defaults and t
The following fixes wrong-code when SLP ends up swapping operands
of a reduction stmt, confusing code-generation later. Fixed by
not doing such swapping.
Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Richard.
2019-07-31 Richard Biener
PR tree-optimization/91293
On Wed, 2019-07-31 at 10:42 +0200, Martin Liška wrote:
> Hi.
>
> As seen here:
> https://github.com/RPGillespie6/fastcov/pull/18/files/f184dd8b6fc14e0
> 75dfc0ea93de7be5c96298ddb#r308735088
>
> GCOV uses json::number for branch count, line count and similar.
> However, the json library
> uses a d
On Mon, Jul 29, 2019 at 10:00:53PM +0100, Kwok Cheung Yeung wrote:
> --- a/gcc/omp-low.c
> +++ b/gcc/omp-low.c
> @@ -11636,9 +11636,40 @@ lower_omp_target (gimple_stmt_iterator *gsi_p,
> omp_context *ctx)
> tkind = GOMP_MAP_FIRSTPRIVATE_INT;
> type = TREE_TYPE (ovar);
>
On Mon, Jul 29, 2019 at 09:55:52PM +0100, Kwok Cheung Yeung wrote:
> +/* True if OpenMP should treat this DECL as an optional argument. */
> +
> +bool
> +gfc_omp_is_optional_argument (const_tree decl)
> +{
> + return DECL_LANG_SPECIFIC (decl)
> + && GFC_DECL_OPTIONAL_ARGUMENT (decl);
I thin
I've committed this patch to correct a misunderstanding of the ISA.
The ISA documentation implies that "s_waitcnt expcnt(0)" should be used
after memory writes, but apparently it really only means this for GDS
writes (and pixel exports, I think).
The patch simply removes most of these instruc
> diff --git a/gcc/gcc.c b/gcc/gcc.c
> index a4323eb146e..e43f46246ad 100644
> --- a/gcc/gcc.c
> +++ b/gcc/gcc.c
> @@ -8268,6 +8268,44 @@ driver::maybe_run_linker (const char *argv0) const
> {
>int tmp = execution_count;
>
> + /* Detect jobserver and drop it if it's not working.
On Sat, 27 Jul 2019, Uros Bizjak wrote:
> On Sat, Jul 27, 2019 at 12:07 PM Uros Bizjak wrote:
>
> > > How would one write smaxsi3 as a splitter to be split after
> > > reload in the case LRA assigned the GPR alternative? Is it
> > > even worth doing? Even the SSE reg alternative can be split
>
This patch adds the same check that is present in binutils/bfd/elfxx-riscv.c.
Checks that we have reached the end of the string after all the
parsing routines have been run. Without this check, GCC silently
succeeds on erroneous input, and produces an assembly with a truncated
arch attribute.
tes
This is the last bit I had in my tree - as seen elsewhere you
eventually run into testcases triggering issues, so this fixes
EVRP not releasing any of its temporary ranges it pushes to
the lattice. It also cleans up the lattice entry swapping
"const correctness" issue by adding a new swap_vr_val
On Fri, Jul 26, 2019 at 12:56 PM Eric Botcazou wrote:
>
> Hi,
>
> one of the effects of -fnon-call-exceptions is that the memory accesses are
> considered trapping by default, i.e. unless you can prove otherwise. If, in
> addition to this, the code is covered by an exception handler, such memory
On Wed, Jul 31, 2019 at 10:37 AM Martin Liška wrote:
>
> On 7/30/19 4:00 PM, Marc Glisse wrote:
> > On Tue, 30 Jul 2019, Martin Liška wrote:
> >
> >> Ah, that's bad, both of them need a care:
> >
> > Yes, that makes more sense to me, thanks.
> >
> >>> I tried to experiment to understand, but it is
On 7/31/19 11:12 AM, Jan Hubicka wrote:
>> and see that in some cases jobserver is 0 and in other cases jobserver is 1.
>> Examples of both:
>> 0 kw -j2 --jobserver-auth=3,6 --
>> total 0
>> lrwx--. 1 jakub jakub 64 Jul 31 10:41 0 -> /dev/pts/4
>> l-wx--. 1 jakub jakub 64 Jul 31 10:41 1 ->
On Tue, 30 Jul 2019, Martin Sebor wrote:
> On 7/30/19 9:19 AM, Jakub Jelinek wrote:
> > On Tue, Jul 30, 2019 at 09:16:45AM -0600, Martin Sebor wrote:
> > > Say something like this POC:
> >
> > We have those already, see ctz_hwi, clz_hwi etc.
> > The thing is that BITMAP_WORD isn't always the same
On Tue, Jul 30, 2019 at 3:39 PM Martin Liška wrote:
>
> On 7/30/19 3:09 PM, Marc Glisse wrote:
> > On Tue, 30 Jul 2019, Martin Liška wrote:
> >
> >> On 7/30/19 1:35 PM, Marc Glisse wrote:
> >>> + /* Some delete operators have size as 2nd argument. */
> >>>
> >>> Some delete operators hav
Hi!
On Wed, Jul 31, 2019 at 10:51:22AM +0200, Uros Bizjak wrote:
> OK.
Thanks. This follow-up implements the same for mmx with sse for V8QImode,
the testcase shows that it is useful too. The difference is quite large:
- movq$0, -72(%rsp)
- movl$bytes, %eax
movqb
gcc/testsuite/ChangeLog:
* g++.dg/lto/pr87906_0.C: Add dg-require-effective-target shared check.
---
gcc/testsuite/g++.dg/lto/pr87906_0.C | 1 +
1 file changed, 1 insertion(+)
diff --git a/gcc/testsuite/g++.dg/lto/pr87906_0.C
b/gcc/testsuite/g++.dg/lto/pr87906_0.C
index 434d9fbe142..6a0
> On Wed, Jul 31, 2019 at 11:12:15AM +0200, Jan Hubicka wrote:
> > > so, if one is lucky enough and at least one of the two file descriptors in
> > > MAKEFLAGS --jobserver-auth= is closed, it will work fine, but if they are
> > > open, even when they have nothing to do with make, it will fail miser
On Wed, Jul 31, 2019 at 11:12:15AM +0200, Jan Hubicka wrote:
> > so, if one is lucky enough and at least one of the two file descriptors in
> > MAKEFLAGS --jobserver-auth= is closed, it will work fine, but if they are
> > open, even when they have nothing to do with make, it will fail miserably.
>
> and see that in some cases jobserver is 0 and in other cases jobserver is 1.
> Examples of both:
> 0 kw -j2 --jobserver-auth=3,6 --
> total 0
> lrwx--. 1 jakub jakub 64 Jul 31 10:41 0 -> /dev/pts/4
> l-wx--. 1 jakub jakub 64 Jul 31 10:41 1 -> pipe:[75026106]
> lr-x--. 1 jakub jakub 6
Hi Joel,
On 7/17/19 12:19 PM, Joel Hutton wrote:
At present it is possible to call the CMSE functions for checking
addresses (such as cmse_check_address_range) and forget to check/use
the return value. This patch makes the interfaces more robust against
programmer error by marking these functio
On Wed, Jul 31, 2019 at 10:21:16AM +0200, Martin Liška wrote:
> On 7/31/19 10:08 AM, Jan Hubicka wrote:
> > Hi,
> >
> >> We do not detect jobserver because of Dejagnu is not using it.
> >> And yes, we default to -flto= in LTO tests now. That's
> >> what is causing issues right now.
> >
> > Why th
On Wed, Jul 31, 2019 at 9:10 AM Jakub Jelinek wrote:
>
> Hi!
>
> As mentioned in the PR, we can use psadbw to shorten the final reductions to
> scalar for 8-bit elements. E.g. for -mavx2 the difference is:
> - vmovdqa %xmm1, %xmm0
> - vextracti128$0x1, %ymm1, %xmm1
> - vpadd
On Wed, Jul 31, 2019 at 7:51 AM Tejas Joshi wrote:
>
> Hi.
>
> > > * gcc.target/i386/avx-vround-roundeven-1.c: New test.
> > > * gcc.target/i386/avx-vround-roundeven-2.c: New test.
> >
> > roundss and roundsd are sse4_1 instructions, also please change tests
> > to use -O2:
>
> I have made
Hi.
As seen here:
https://github.com/RPGillespie6/fastcov/pull/18/files/f184dd8b6fc14e075dfc0ea93de7be5c96298ddb#r308735088
GCOV uses json::number for branch count, line count and similar. However, the
json library
uses a double as an internal representation for numbers. That's no desirable in
On 7/30/19 4:00 PM, Marc Glisse wrote:
> On Tue, 30 Jul 2019, Martin Liška wrote:
>
>> Ah, that's bad, both of them need a care:
>
> Yes, that makes more sense to me, thanks.
>
>>> I tried to experiment to understand, but it is complicated because
>>> including disables the optimization:
>>>
>
On Tue, Jul 30, 2019 at 4:54 PM Andrew MacLeod wrote:
>
> On 7/30/19 4:55 AM, Richard Biener wrote:
> > On Fri, Jul 26, 2019 at 4:32 PM Andrew MacLeod wrote:
> >> On 7/25/19 11:37 PM, Jeff Law wrote:
> >>> On 7/24/19 12:33 PM, Richard Biener wrote:
> On July 24, 2019 8:18:57 PM GMT+02:00, Je
On 7/31/19 10:08 AM, Jan Hubicka wrote:
> Hi,
>
>> We do not detect jobserver because of Dejagnu is not using it.
>> And yes, we default to -flto= in LTO tests now. That's
>> what is causing issues right now.
>
> Why the error messages are
> make[4]: *** write jobserver: Bad file descriptor. St
Hi,
> We do not detect jobserver because of Dejagnu is not using it.
> And yes, we default to -flto= in LTO tests now. That's
> what is causing issues right now.
Why the error messages are
make[4]: *** write jobserver: Bad file descriptor. Stop.
make[4]: *** Waiting for unfinished jobs
make
On Wed, Jul 31, 2019 at 09:50:47AM +0200, Martin Liška wrote:
> > Why doesn't the jobserver work in the tests? Is that because of missing +
> > somewhere in the Makefiles or is something unsetting MFLAGS or MAKEFLAGS
> > env vars?
>
> Is Dejagnu using make internally to invoke individual tests?
Otherwise a sequence of _2 = _1 + 4; _3 = _2 + 4; ... can blow the
stack.
Bootstrap and regtest running on x86_64-unknown-linux-gnu.
Richard.
2019-07-31 Richard Biener
PR tree-optimization/91178
* tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address):
Use tail-recu
Hi!
This undoes the splitting of private clauses to parallel instead of
innermost. I should one day change the clause splitting to match exactly
what was agreed on in OpenMP 5.0, without the exceptions because we've done
something differently in the past, but I'm afraid it will be a lot of work.
On 7/31/19 9:34 AM, Jakub Jelinek wrote:
> On Wed, Jul 31, 2019 at 09:20:40AM +0200, Martin Liška wrote:
>> One possible solution will be to adjust lto.exp:
>>
>>set LTO_OPTIONS [list \
>>{-O0 -flto -flto-partition=none -fuse-linker-plugin} \
>>{-O2 -flto -flto-parti
On 7/31/19 9:40 AM, Jan Hubicka wrote:
>> Neither will work very well, we have thousands of -flto tests outside of
>> lto.exp, e.g. dg-torture.exp (or libgomp and others) cycle through various
>> options including -flto etc.
>>
>> Some env var would be useful I guess, though shouldn't it have GCC i
> Neither will work very well, we have thousands of -flto tests outside of
> lto.exp, e.g. dg-torture.exp (or libgomp and others) cycle through various
> options including -flto etc.
>
> Some env var would be useful I guess, though shouldn't it have GCC in the
> name? I mean, if we run into these
On Wed, Jul 31, 2019 at 09:20:40AM +0200, Martin Liška wrote:
> One possible solution will be to adjust lto.exp:
>
> set LTO_OPTIONS [list \
> {-O0 -flto -flto-partition=none -fuse-linker-plugin} \
> {-O2 -flto -flto-partition=none -fuse-linker-plugin
> -fno-fat-lt
Steve Ellcey writes:
> On Tue, 2019-07-30 at 14:31 +0100, Richard Sandiford wrote:
>>
>> > -
>> > - tree new_type = build_distinct_type_copy (TREE_TYPE (node-
>> > >decl));
>> > - TYPE_ARG_TYPES (new_type) = new_reversed;
>>
>> I think you still need this line, just applied to the exis
Thanks for these comments.
Feng
From: Michael Matz
Sent: Tuesday, July 30, 2019 1:59:04 AM
To: Feng Xue OS
Cc: Richard Biener; gcc-patches@gcc.gnu.org
Subject: Re: Ping agian: [PATCH V2] Loop split upon semi-invariant condition
(PR tree-optimization/8913
On 7/31/19 1:32 AM, Jakub Jelinek wrote:
> This broke a lot of tests.
Whoops.
> The logs show
> spawn -ignore SIGHUP /home/jakub/src/gcc/obj31/gcc/xgcc
> -B/home/jakub/src/gcc/obj31/gcc/ c_lto_pr83954_0.o c_lto_pr83954_1.o
> -fno-diagnostics-show-
> caret -fno-diagnostics-show-line-numbers -fdi
Hi!
As mentioned in the PR, we can use psadbw to shorten the final reductions to
scalar for 8-bit elements. E.g. for -mavx2 the difference is:
- vmovdqa %xmm1, %xmm0
- vextracti128$0x1, %ymm1, %xmm1
- vpaddb %xmm1, %xmm0, %xmm0
- vpsrldq $8, %xmm0, %xmm1
- vpadd
86 matches
Mail list logo
