Hi,
In my testing with cost tweaking for vector with length, I found
two cases below didn't get the expected output. Since the expected
instructions reply on the vectorization occurrence, we don't expect
vectorization gets disabled by cost model.
To make it not fragile, the fix is to force it wi
The convert_like* macros were introduced in
2000-03-05 Nathan Sidwell
* call.c (convert_like): Macrofy.
(convert_like_with_context): New macro.
but now we can use overloading so we can do away with the macros.
I've also taken this chance to rename _real to _internal to make it
Hi!
On Tue, Jul 14, 2020 at 12:15:01PM -0500, will schmidt wrote:
> We've got a scenario with a combination of old hardware, gcc-8 and
> binutils where gcc will ICE during it's selftest. This ICE was exposed
> when the builtin processing for better #pragma support was added, where
> we no longe
This patch tunes the default GPU thread count for OpenMP and OpenACC on
AMD GCN devices. It chooses a sensible default if no attributes are
given at all, increases the number of OpenACC gangs if only one worker
per gang is specified, and increases the number of workers otherwise.
The tuning is
Hi!
On Wed, Jul 08, 2020 at 12:59:29PM -0700, Carl Love wrote:
> [PATCH 5/6] rs6000, Add vector splat builtin support
> +(define_insn "xxspltiw_v4si"
> + [(set (match_operand:V4SI 0 "register_operand" "=wa")
> + (unspec:V4SI [(match_operand:SI 1 "s32bit_cint_operand" "n")]
> +
CMPXCHG instruction sets ZF flag if the values in the destination operand
and EAX register are equal; otherwise the ZF flag is cleared and value
from destination operand is loaded to EAX. Following assembly:
movl%esi, %eax
lock cmpxchgl %edx, (%rdi)
cmpl%esi, %eax
On 01/07/2020 3:28 pm, Tom de Vries wrote:
I looked at the implementation, and it looks ok to me, though I think we
need to make explicit in a comment what the assumptions are:
- that we have read and write access to the entire word, and
Is there a situation where an 8/16-bit portion of memory
On Tue, 14 Jul 2020, Patrick Palka wrote:
> This implements the floating-point std::to_chars overloads for float,
> double and long double. We use the Ryu library to compute the shortest
> round-trippable fixed and scientific forms of a number for float, double
> and long double. We also use Ryu
On Tue, 14 Jul 2020, Patrick Palka wrote:
> This performs the following modifications to our local copy of Ryu in
> order to make it more easily usable for our std::to_chars implementation:
>
> * Remove all #includes
> * Remove copy_special_str routines
> * Adjust the exponent formatting to
Expand strncmp to "repz cmpsb" only with -minline-all-stringops since
"repz cmpsb" can be much slower than strncmp function implemented with
vector instructions, see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=43052
gcc/
PR target/95458
* config/i386/i386-expand.c (ix86_expand_c
On Wed, 15 Jul 2020, Richard Biener wrote:
> But note one of the issues is that when not cross-compiling we're
> using a single libiberty for target and host objects (likewise
There shouldn't be a target libiberty, since commit
8499116aa30a46993deff5acf73985df6b16fb8b (re PR regression/47836 (So
gcc/ChangeLog:
* config.in: Regenerate.
* config/s390/s390.c (print_operand): Emit vector alignment hints
for target z13, if AS accepts them. For other targets the logic
stays the same.
* config/s390/s390.h (TARGET_VECTOR_LOADSTORE_ALIGNMENT_HINTS): Define
From: Andreas Krebbel
The IBM z14 POP adds an optional alignment operand to the vl, vst,
vlm, and vstm instruction (vector loads and stores). Vectors residing
on 8 or 16 byte boundaries might get loaded or stored faster on some
models given the instruction uses the proper hint operand. A wrong
h
Backport for supporting vector load/store alignment hints for GCC8.
Bootstrapped and regtested on s390x with and without patched gas. Ok for
releases/gcc-8 branch?
Andreas Krebbel (1):
S/390: Support vector load/store alignment hints
Stefan Schulze Frielinghaus (1):
S/390: Emit vector alignm
Hi,
here is an enhancement to gcc, which allows load/store groups with size being
non-power-of-2 to be vectorized.
Current implementation is using interleaving permutations to transform
load/store groups. That is where power-of-2 requirements comes from.
For N-element vectors simplest approch wo
It turned out that using -fdefault-integer-8 on 32bit systems didn't
work with omp_alloc – as that gave a 64bit / kind=8 integer for the
(c_)size_t argument. [As reported by Jakub on #gcc.]
Fixed (and committed to mainline + OG10) as attached; the most trivial
version of the patch didn't not work
On Wed, Jul 15, 2020 at 8:20 AM Richard Biener wrote:
>
> On Wed, 15 Jul 2020, H.J. Lu wrote:
>
> > With --enable-cet, require CET support only for the final GCC build.
> > Don't enable CET without CET support in stage1 nor for build support.
>
> I wonder if we want to do sth less fragile than tes
On Wed, 15 Jul 2020, H.J. Lu wrote:
> With --enable-cet, require CET support only for the final GCC build.
> Don't enable CET without CET support in stage1 nor for build support.
I wonder if we want to do sth less fragile than testing
for ../stage_current - for example the toplevel make could
exp
Hi!
I've missed
+FAIL: libgomp.c/loop-21.c execution test
during testing of the recent patch. The problem is that while
for the number of iterations computation it doesn't matter if we compute
min_inner_iterations as (m2 * first + n2 + (adjusted step) + m1 * first + n1) /
step
or (m2 * last + n2
With --enable-cet, require CET support only for the final GCC build.
Don't enable CET without CET support in stage1 nor for build support.
config/
PR bootstrap/96202
* cet.m4 (GCC_CET_HOST_FLAGS): Don't enable CET without CET
support in stage1 nor for build support.
gcc/
cp_parser_declaration copies tokens to local variables, before
inspecting (some of) their fields. There's no need. Just point at them
in the token buffer -- they don't move. Also, we never look at the
second token if the first is EOF, so no need for some kind of dummy
value in that case.
This adjusts the description of Flatten, removes an obsolete comment
and uses Compile_Time_Known_Value as now done in Is_Static_Element.
No functional changes.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* exp_aggr.adb (Flatten): Adjust description.
(Convert_To_Po
GNAT would in some cases not resolve a Predicate_Failure aspect properly
and generate spurious errors of the form:
cannot find unique type for raise expression
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* sem_ch13.adb (Check_Aspect_At_End_Of_Declarations): Add proper
This patch modifies the parser to recognize
iterated_element_associations, which may include a key_exprewsion to be
used in a named aggregate such as a map. The new syntactic node
N_Iterated_Element_Association is recognized throughout the compiler.
The patch also extends the analysis and expansion
Renamed_Entity is only valid for a few entities, using it on any entity
passed to Is_Renaming can result in crashes. Fixing this requires
making sure that Is_Renaming only uses Renamed_Entity on entities where
this is allowed and uses Is_Renaming_Of_Object everywhere else.
Tested on x86_64-pc-lin
If an exception is raised early in Allocate_Any_Controlled, no lock is
taken yet and Unlock is called on a lock which isn't taken.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* libgnat/s-stposu.adb (Allocate_Any_Controlled): Fix logic in
lock/unlock.diff --git a/gc
This prevents the compiler from generating elaboration code for a record
declared with an initial value and an alignment aspect. The expression
of an alignment aspect must be static so, in practice, there is no need
to defer the elaboration of the object just because of it.
Tested on x86_64-pc-li
Cleanup frontend code before routine Find_Overlaid_Entity will be reused
in GNATprove backend.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* sem_util.adb (Find_Overlaid_Entity): Fix style in comment.
(Note_Possible_Modification): Simplify repeated calls to Ekind.di
Target name (i.e. "@"), which was introduced to Ada 202X in AI12-0125,
denotes a constant object (RM 3.3(21.2/5)), even though target_name
itself is not an object (RM 3.3(2)).
This patch allows @ to be appear as a prefix for Address attribute
(which requires an object). Also, it enables constructs
In some cases where a procedure call is expected but a function is
provided such as "Interfaces.C."=" (x, y);" GNAT would not generate any
error message.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* sem_ch6.adb (Analyze_Procedure_Call): Detect use of operators
in
In expression like "(Arr with delta Low .. High => New_Component_Value)"
bounds Low .. High might denote a null range. In this case both Low and
High can be any values from the base type of the array's index type;
they don't need to belong to the array's index type itself.
This patch removes unnec
Use aspect SPARK_Mode with value Off in the spec and body of standard
containers, bounded and unbounded versions, so that it is clearer that
they cannot be used in SPARK code. Formal containers should be used
instead.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* libgnat/a
Detecting effectively volatile objects in restricted contexts happens in
two routines: Is_Effectively_Volatile_Object and Is_OK_Volatile_Context.
Their handling of type conversions and slices were different; also none
of them has been dealing with qualified expressions, which has been just
added to
This patch extends static functions and allow them on Intrinsic imported
subprograms in addition to expression functions, under -gnatX. We also
implement compile time evaluation of Shift_Left/Right operators as a
first set of useful static-compatible intrinsics.
Tested on x86_64-pc-linux-gnu, com
In a case of a complex precondition expression with quantifiers, we can
get a crash in Resolve_Type_Conversion when trying to emit a -gnatwr
warning.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* sem_res.adb (Resolve_Type_Conversion): Protect against null
entity.
This ACATS test shows GNAT was not enforcing legality rules related to
subpools.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* exp_ch3.adb (Freeze_Type): Remove warning in expander,
replaced by a corresponding error in sem_ch13.adb. Replace
RTE_Available by
The RM C.6(19) clause says that atomic or volatile objects of types
that are not by reference must be passed by copy in a call if the
type of the formal is not atomic or volatile respectively. But this
requirement does not apply to initialization procedures, which are
generated by the compiler, an
The predicate returns true only if an aspect requiring delaying like
Alignment or Address is the first aspect in the list. The change
also contains a small consistency fix for Freeze_Object_Declaration.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* sem_ch3.adb (Delayed_As
Apart from the usual editorial tweaks, this documents the discrepancy
between the aspect and the non-aspect cases for alignment settings in
object declarations.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* einfo.ads (Delayed Freezing and Elaboration): Minor tweaks.
This patch fixes an error in the compiler whereby an allocator for a
limited type may cause spurious accessibility errors due to a
miscalculation of access levels on interally generated temporaries.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* exp_ch6.adb (Make_Build_In_P
This changes the Sloc used to expand a timed entry call from that
of the Select to that of the Delay statement for coverage purposes.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* exp_ch9.adb (Expand_N_Timed_Entry_Call): Use the Sloc of
the delay statement in the e
When we detect effectively volatile array type we only need to examine
the type of array component if the array itself has no
Has_Volatile_Components property.
Tested on x86_64-pc-linux-gnu, committed on trunk
gcc/ada/
* sem_prag.adb (Atomic_Components): Simplify with Ekind_In.
(
This implements additional functionality for the Ada 202x container
aggregates, in particular the use of iterated_component_association in
both Unnamed (positional) and Named (keyed) aggregates for types for
which the Aspect Aggregate is defined.
Tested on x86_64-pc-linux-gnu, committed on trunk
On Wed, Jul 15, 2020 at 03:06:18PM +0200, Rainer Orth wrote:
> I must admit I missed that in that terrible nested maze of #ifdef's
> compiler-rt uses.
>
> > I mean, while the ifndef/define change is guarded by #if SANITIZER_SOLARIS,
> > the last changed line is not. I'm afraid I don't know if
> >
Parser error recovery can get confused by the tokens within a deferred
pragma, as treats those as regular tokens. This adjusts the recovery so
that the pragma is treated as a unit. Also, the preprocessor now
ensures that we never have an EOF token inside a pragma -- the pragma is
always close
Hi Jakub,
> On Tue, Jul 14, 2020 at 02:32:57PM +0200, Rainer Orth wrote:
>> The latest Solaris 11.4/x86 update uncovered a libsanitizer bug that
>> caused one test to FAIL for 32-bit:
>>
>> +FAIL: c-c++-common/asan/null-deref-1.c -O0 output pattern test
>> +FAIL: c-c++-common/asan/null-deref-1
> This should be:
>
> PR target/96190
> * config/sparc/linux.h (ENDFILE_SPEC): Use GNU_USER_TARGET_ENDFILE_SPEC
> to get crtendS.o for !no-pie mode.
> * config/sparc/linux64.h(ENDFILE_SPEC): Ditto.
* config/sparc/linux64.h (ENDFILE_SPEC): Ditto.
> OK for mainline
Status
==
The GCC 10 branch is now frozen for the GCC 10.2 release, all changes
to he branch require a RM approval.
Quality Data
Priority # Change from last report
--- ---
P1
P2 218 + 2
P3
> the stack clash protection mechanism in the x86 back-end was implemented by
> largely duplicating the existing stack checking implementation. Now the only
> significant difference between them is the probing window, which is shifted by
> 1 probing interval (not 2 as documented in explow.c), but
On 6/24/20 7:32 PM, Tobias Burnus wrote:
While testing, I encountered two bugs, one relating to kind=4
character string (patch pending review; PR95837)
not part of testcase)
As that PR has been committed, I updated the testcase
to check character(kind=4) as well. (I also removed
the unused var
Hi Julian, Tobias!
On 2020-07-14T13:43:37+0200, I wrote:
> On 2020-06-16T15:39:44-0700, Julian Brown wrote:
>> As mentioned in the blurb for the previous patch, an "attach" operation
>> for a Fortran pointer with an array descriptor must copy that array
>> descriptor to the target.
>
> Heh, I see
On Tue, Jul 14, 2020 at 1:21 AM Richard Sandiford
wrote:
>
> Jakub Jelinek writes:
> > On Wed, Jul 08, 2020 at 03:10:14PM +0100, Richard Sandiford wrote:
> >> gcc/
> >> PR target/95726
> >> * config/aarch64/aarch64.c (aarch64_attribute_table): Add
> >> "Advanced SIMD type".
> >>
Hi Jojo,
> diff --git a/gcc/Makefile.in b/gcc/Makefile.in
> index 2ba76656dbf..3306510a9a8 100644
> --- a/gcc/Makefile.in
> +++ b/gcc/Makefile.in
> @@ -1253,6 +1253,13 @@ ANALYZER_OBJS = \
> # We put the *-match.o and insn-*.o files first so that a parallel make
> # will build them sooner, becau
Hi,
The size of generated file like insn-emit.c is very huge
if we add lots of define_expand patten or define_insn patten.
It’s present easily when there are much more intrinsic interfaces
it costs much more time to create toolchain with compiling these files
:(
Hi,
the stack clash protection mechanism in the x86 back-end was implemented by
largely duplicating the existing stack checking implementation. Now the only
significant difference between them is the probing window, which is shifted by
1 probing interval (not 2 as documented in explow.c), but
> -Original Message-
> From: Richard Sandiford [mailto:richard.sandif...@arm.com]
> Sent: Tuesday, July 7, 2020 10:08 PM
> To: xiezhiheng
> Cc: Richard Biener ; gcc-patches@gcc.gnu.org
> Subject: Re: [PATCH PR94442] [AArch64] Redundant ldp/stp instructions
> emitted at -O3
>
> xiezhiheng
> In --enable-default-pie mode compiler should switch from
> using crtend.o to crtendS.o. On sparc it is especially important
> because crtend.o contains PIC-unfriendly code.
>
> We use GNU_USER_TARGET_ENDFILE_SPEC as a baseline spec to get
> crtendS.o instead of crtend.o in !no-pie mode.
>
> gcc
Hi,
This is a simple fix for PR96195.
For the test case, GCC generates the following gimple statement in
pass_vect:
vect__21.16_58 = zp.simdclone.2 (vect_vec_iv_.15_56);
The mode of vect__21.16_58 is VNx2SI while the mode of zp.simdclone.2
(vect_vec_iv_.15_5
This patch provides standard vec_extract and vec_set patterns to the
nvptx backend, to extract an element from a PTX vector and set an
element of a PTX vector respectively. PTX vectors (I hesitate to
call them SIMD vectors) may contain up to four elements, so vector
modes up to size four are supp
59 matches
Mail list logo
