nvptx/mkoffload.cc: Warn instead of error when reverse offload is not possible
Reverse offload requests at least -misa=sm_35; with this patch, a warning
instead of an error is shown, still permitting reverse offload for all
other configured device types. This is achieved by not calling
GOMP_offload_register_ver (and stopping generating pointless 'static const char'
variables, once known.)
The tool_name as progname changes adds "nvptx " and "gcn " to the
"mkoffload: warning/error:" diagnostic.
gcc/ChangeLog:
* config/nvptx/mkoffload.cc (process): Replace a fatal_error by
a warning + not enabling offloading if -misa=sm_30 prevents
reverse offload.
(main): Use tool_name as progname for diagnostic.
* config/gcn/mkoffload.cc (main): Likewise.
libgomp/ChangeLog:
* libgomp.texi (Offload-Target Specifics: nvptx): Document
that reverse offload requires >= -march=sm_35.
* testsuite/libgomp.c-c++-common/requires-4.c: Build for nvptx
with -misa=sm_35.
* testsuite/libgomp.c-c++-common/requires-5.c: Likewise.
* testsuite/libgomp.c-c++-common/requires-6.c: Likewise.
* testsuite/libgomp.c-c++-common/reverse-offload-1.c: Likewise.
* testsuite/libgomp.fortran/reverse-offload-1.f90: Likewise.
* testsuite/libgomp.c/reverse-offload-sm30.c: New test.
Jakub Jelinek [Mon, 12 Sep 2022 13:45:02 +0000 (15:45 +0200)]
libgomp: Fix up icv-6.c [PR106894]
The thing is,
make check
or
make check RUNTESTFLAGS="c.exp='icv-6.c' c++.exp='icv-6.c'"
in libgomp obj dir work fine, but
make -j32 -k check RUNTESTFLAGS="c.exp='icv-6.c' c++.exp='icv-6.c'"
fails.
The thing is that the testcase as written relies on OMP_NUM_THREADS not being
set in environment (as it takes priority over OMP_NUM_THREADS_ALL for the
host).
So, if either a user has OMP_NUM_THREADS=42 in the environment by himself, or
when doing make check with -jN, we trigger:
if test $$num_cpus -gt 8 && test -z "$$OMP_NUM_THREADS"; then \
OMP_NUM_THREADS=8; export OMP_NUM_THREADS; \
echo @@@ libgomp OMP_NUM_THREADS adjusted to 8 because of parallel
make check and too many CPUs; \
fi; \
in libgomp/testsuite/Makefile.am and so the test fails.
2022-09-12 Jakub Jelinek <jakub@redhat.com>
PR libgomp/106894
* testsuite/libgomp.c-c++-common/icv-6.c: Include string.h.
(main): Avoid tests for which corresponding non-_ALL suffixed variable
is in the environment, or for OMP_NUM_TEAMS on the device
OMP_NUM_TEAMS_DEV_?.
nvptx: libgomp+mkoffload.cc: Prepare for reverse offload fn lookup
Add support to nvptx for reverse lookup of function name to prepare for
'omp target device(ancestor:1)'.
gcc/ChangeLog:
* config/nvptx/mkoffload.cc (struct id_map): Add 'dim' member.
(record_id): Store func name without quotes, store dim separately.
(process): For GOMP_REQUIRES_REVERSE_OFFLOAD, check that -march is
at least sm_35, create '$offload_func_table' global array and init
with reverse-offload function addresses.
* config/nvptx/nvptx.cc (write_fn_proto_1, write_fn_proto): New
force_public attribute to force .visible.
(nvptx_declare_function_name): For "omp target
device_ancestor_nohost" attribut, force .visible/TREE_PUBLIC.
libgomp/ChangeLog:
* plugin/plugin-nvptx.c (GOMP_OFFLOAD_load_image): Read offload
function address table '$offload_func_table' if rev_fn_table
is not NULL.
Prepare for reverse-offloading function-pointer lookup by passing
a rev_fn_table argument to GOMP_OFFLOAD_load_image.
The argument will be NULL, unless GOMP_REQUIRES_REVERSE_OFFLOAD is
requested and devices not supported it, are filtered out.
(Up to and including this commit, no non-host device claims such
support and the caller currently always passes NULL.)
* These operations are single-precision float only and are only active
if unsafe_math_optimizations are enabled (due to potential numerical
precision issues).
Jakub Jelinek [Fri, 9 Sep 2022 13:35:18 +0000 (15:35 +0200)]
libgomp: Fix up OMP_PROC_BIND handling [PR106894]
While the first param is char (gomp_global_icv.bind_var), the second param
is char * (gomp_bind_var_list), so we shouldn't access it through *(char *).
2022-09-09 Jakub Jelinek <jakub@redhat.com>
PR libgomp/106894
* env.c (initialize_env) <case PARSE_BIND>: Use char ** instead of
char * for dest[1] initialization from params[1]. Formatting fixes.
Richard Biener [Wed, 7 Sep 2022 08:44:33 +0000 (10:44 +0200)]
tree-optimization/106860 - fix profile scaling in split_loop
The following fixes a mistake in loop splitting which assumes loop
latches have a single predecessor and that edge is from the exit
test. Instead work from the single exit edge we have to find the
edge towards the latch.
PR tree-optimization/106860
* tree-ssa-loop-split.cc (split_loop): Find the exit to
latch edge from the loop exit edge instead of from the
latch. Verify we're going to find it.
Richard Biener [Fri, 2 Sep 2022 11:36:13 +0000 (13:36 +0200)]
tree-optimization/106809 - compile time hog in VN
The dominated_by_p_w_unex function is prone to high compile time.
With GCC 12 we introduced a VN run for uninit diagnostics which now
runs into a degenerate case with bison generated code. Fortunately
this case is easy to fix with a simple extra check - a more
general fix needs more work.
PR tree-optimization/106809
* tree-ssa-sccvn.cc (dominaged_by_p_w_unex): Check we have
more than one successor before doing extra work.
omp/ompx sentinels are for vendor extensions; as they might be required for
the correctness of the program, a warning should be printable. This patch
documents in the OpenMP 5.2 table the existing warnings, including the new
warning for for fixed source form Fortran.
PR fortran/106670
gcc/fortran/ChangeLog:
* scanner.cc (skip_fixed_omp_sentinel): Add -Wsurprising warning
for 'omx' sentinels with -fopenmp.
* invoke.texi (-Wsurprising): Document additional warning case.
libgomp/ChangeLog:
* libgomp.texi (OpenMP 5.2): Add comment to ompx/omx entry.
gcc/testsuite/ChangeLog:
* c-c++-common/gomp/ompx-1.c: New test.
* c-c++-common/gomp/ompx-2.c: New test.
* g++.dg/gomp/ompx-attrs-1.C: New test.
* gfortran.dg/gomp/ompx-1.f90: New test.
* gfortran.dg/gomp/omx-1.f: New test.
* gfortran.dg/gomp/omx-2.f: New test.
This patch considers the environment variable syntax extension for
device-specific variants of environment variables from OpenMP 5.1 (see
OpenMP 5.1 specification, p. 75 and p. 639). An environment variable (e.g.
OMP_NUM_TEAMS) can have different suffixes:
_DEV (e.g. OMP_NUM_TEAMS_DEV): affects all devices but not the host.
_DEV_<device> (e.g. OMP_NUM_TEAMS_DEV_42): affects only device with
number <device>.
no suffix (e.g. OMP_NUM_TEAMS): affects only the host.
In future OpenMP versions also suffix _ALL will be introduced (see discussion
https://github.com/OpenMP/spec/issues/3179). This is also considered in this
patch:
_ALL (e.g. OMP_NUM_TEAMS_ALL): affects all devices and the host.
The precedence is as follows (descending). For the host:
1. no suffix
2. _ALL
For devices:
1. _DEV_<device>
2. _DEV
3. _ALL
That means, _DEV_<device> is used whenever available. Otherwise _DEV is used if
available, and at last _ALL. If there is no value for any of the variable
variants, default values are used as already implemented before.
This patch concerns parsing (a), storing (b), output (c) and transmission to the
device (d):
(a) The actual number of devices and the numbering are not known when parsing
the environment variables. Thus all environment variables are iterated and
searched for device-specific ones.
(b) Only configured device-specific variables are stored. Thus, a linked list
is used.
(c) The output is done in omp_display_env (see specification p. 468f). Global
ICVs are tagged with [all], see https://github.com/OpenMP/spec/issues/3179.
ICVs which are not global but aren't handled device-specific yet are tagged
with [host]. omp_display_env outputs the initial values of the ICVs. That is
why a dedicated data structure is introduced for the inital values only
(gomp_initial_icv_list).
(d) Device-specific ICVs are transmitted to the device via GOMP_ADDITIONAL_ICVS.
libgomp/ChangeLog:
* config/gcn/icv-device.c (omp_get_default_device): Return device-
specific ICV.
(omp_get_max_teams): Added for GCN devices.
(omp_set_num_teams): Likewise.
(ialias): Likewise.
* config/nvptx/icv-device.c (omp_get_default_device): Return device-
specific ICV.
(omp_get_max_teams): Added for NVPTX devices.
(omp_set_num_teams): Likewise.
(ialias): Likewise.
* env.c (struct gomp_icv_list): New struct to store entries of initial
ICV values.
(struct gomp_offload_icv_list): New struct to store entries of device-
specific ICV values that are copied to the device and back.
(struct gomp_default_icv_values): New struct to store default values of
ICVs according to the OpenMP standard.
(parse_schedule): Generalized for different variants of OMP_SCHEDULE.
(print_env_var_error): Function that prints an error for invalid values
for ICVs.
(parse_unsigned_long_1): Removed getenv. Generalized.
(parse_unsigned_long): Likewise.
(parse_int_1): Likewise.
(parse_int): Likewise.
(parse_int_secure): Likewise.
(parse_unsigned_long_list): Likewise.
(parse_target_offload): Likewise.
(parse_bind_var): Likewise.
(parse_stacksize): Likewise.
(parse_boolean): Likewise.
(parse_wait_policy): Likewise.
(parse_allocator): Likewise.
(omp_display_env): Extended to output different variants of environment
variables.
(print_schedule): New helper function for omp_display_env which prints
the values of run_sched_var.
(print_proc_bind): New helper function for omp_display_env which prints
the values of proc_bind_var.
(enum gomp_parse_type): Collection of types used for parsing environment
variables.
(ENTRY): Preprocess string lengths of environment variables.
(OMP_VAR_CNT): Preprocess table size.
(OMP_HOST_VAR_CNT): Likewise.
(INT_MAX_STR_LEN): Constant for the maximal number of digits of a device
number.
(gomp_get_icv_flag): Returns if a flag for a particular ICV is set.
(gomp_set_icv_flag): Sets a flag for a particular ICV.
(print_device_specific_icvs): New helper function for omp_display_env to
print device specific ICV values.
(get_device_num): New helper function for parse_device_specific.
Extracts the device number from an environment variable name.
(get_icv_member_addr): Gets the memory address for a particular member
of an ICV struct.
(gomp_get_initial_icv_item): Get a list item of gomp_initial_icv_list.
(initialize_icvs): New function to initialize a gomp_initial_icvs
struct.
(add_initial_icv_to_list): Adds an ICV struct to gomp_initial_icv_list.
(startswith): Checks if a string starts with a given prefix.
(initialize_env): Extended to parse the new syntax of environment
variables.
* icv-device.c (omp_get_max_teams): Added.
(ialias): Likewise.
(omp_set_num_teams): Likewise.
* icv.c (omp_set_num_teams): Moved to icv-device.c.
(omp_get_max_teams): Likewise.
(ialias): Likewise.
* libgomp-plugin.h (GOMP_DEVICE_NUM_VAR): Removed.
(GOMP_ADDITIONAL_ICVS): New target-side struct that
holds the designated ICVs of the target device.
* libgomp.h (enum gomp_icvs): Collection of ICVs.
(enum gomp_device_num): Definition of device numbers for _ALL, _DEV, and
no suffix.
(enum gomp_env_suffix): Collection of possible suffixes of environment
variables.
(struct gomp_initial_icvs): Contains all ICVs for which we need to store
initial values.
(struct gomp_default_icv):New struct to hold ICVs for which we need
to store initial values.
(struct gomp_icv_list): Definition of a linked list that is used for
storing ICVs for the devices and also for _DEV, _ALL, and without
suffix.
(struct gomp_offload_icvs): New struct to hold ICVs that are copied to
a device.
(struct gomp_offload_icv_list): Definition of a linked list that holds
device-specific ICVs that are copied to devices.
(gomp_get_initial_icv_item): Get a list item of gomp_initial_icv_list.
(gomp_get_icv_flag): Returns if a flag for a particular ICV is set.
* libgomp.texi: Updated.
* plugin/plugin-gcn.c (GOMP_OFFLOAD_load_image): Extended to read
further ICVs from the offload image.
* plugin/plugin-nvptx.c (GOMP_OFFLOAD_load_image): Likewise.
* target.c (gomp_get_offload_icv_item): Get a list item of
gomp_offload_icv_list.
(get_gomp_offload_icvs): New. Returns the ICV values
depending on the device num and the variable hierarchy.
(gomp_load_image_to_device): Extended to copy further ICVs to a device.
* testsuite/libgomp.c-c++-common/icv-5.c: New test.
* testsuite/libgomp.c-c++-common/icv-6.c: New test.
* testsuite/libgomp.c-c++-common/icv-7.c: New test.
* testsuite/libgomp.c-c++-common/icv-8.c: New test.
* testsuite/libgomp.c-c++-common/omp-display-env-1.c: New test.
* testsuite/libgomp.c-c++-common/omp-display-env-2.c: New test.
where one can see a missing COND_EXPR. It seems to be a latent
issue to me given the code can be exercised, it just maybe misses
a 'need_new' testcase combined with 'cond_stmt'. Appearantly
the if (cond_stmt) code is just to avoid creating a temporary
(and possibly to preserve the condition compute if used elsewhere
since the original stmt is going to be deleted). The following
makes the failure go away for me in my patched tree and it
also survives libgomp and gomp testing in an unpatched tree.
2022-05-13 Richard Biener <rguenther@suse.de>
* omp-expand.cc (expand_omp_atomic_cas): Do not short-cut
computation of the new value.
This patch implements doacross(sink: omp_cur_iteration - 1) that the
previous patchset emitted a sorry on during omp expansion.
It can be implemented with existing library functions.
To recap, depend(source)/doacross(source:)/doacross(source:omp_cur_iteration)
is implemented calling GOMP_doacross_post or GOMP_doacross_ull_post,
called with an array of long or unsigned long long elements, one for
all collapsed loops together and one for each further ordered loop if any.
We initialize that array in each thread when grabbing further set of iterations
and update it at the end of loops, so that it represents the current iteration
(as 0 based counters). When the worksharing loop is created, we tell the
library through another similar array the counts (the loop needs to be
rectangular) in each dimension, first element is count of all logical iterations
in the collapsed loops.
depend(sink:v1 op N1, v2 op N2, ...) is then implemented by conditionally calling
GOMP_doacross_wait/GOMP_doacross_ull_wait. For N? of 0 there is no check,
otherwise if it wants to wait in a particular dimension for a previous iteration,
we check that the corresponding iterator isn't the first one (or first few),
where the previous iterator in that dimension would be out of range, and similarly
for checking of next iteration in a dimension that it isn't the last one (or last few)
where it would be similarly out of bounds. Then the collapsed loop counters are
folded into a single 0 based counter (first argument) and then other 0 based
iterations counters on what iteration it should wait for.
Now, doacross(sink: omp_cur_iteration - 1) is supposed to wait for the previous
logical iteration in the combined iteration space of all ordered loops.
For the very first iteration in that combined iteration space it does nothing,
there is no previous iteration. And similarly it does nothing if there
are more ordered loops than collapsed loop and it isn't the first logical
iteration of the combined loops inside of the collapsed loops, because as implemented
we know the previous iteration in that case is always executed by the same thread
as the current one.
In the implementation, we use the same value as is stored in the first element
of the array for GOMP_doacross_post/GOMP_doacross_ull_post, if that value is 0,
we do nothing. The rest is different based on if ordered argument is equal to
collapse or not. If it is, then we otherwise call
GOMP_doacross_wait/GOMP_doacross_ull_wait with a single argument, one less than
that counter we compare against 0.
If ordered argument is bigger than collapse, we add a per-thread boolean variable
.first.N, which we set to true at the start of the outermost ordered loop inside
of the collapsed set of loops and set to false at the end of the innermost
ordered loop. If .first.N is false, we don't do anything (we know the previous
iteration was handled by the current thread and by my reading of the spec we don't
need to emit even a memory barrier in that case, because it is just synchronization
with the same thread), otherwise we call GOMP_doacross_wait/GOMP_doacross_ull_wait
with the first argument one less than the counter we compare against 0, and then
one less than 2nd and following counts if iterations we pass to the workshare
initialization. If say .counts.N passed to the workshare initialization is
{ 256, 13, 5, 2 } for collapse(3) ordered(6) loop, then
GOMP_doacross_post/GOMP_doacross_ull_post is called with arguments equal to
.ordereda.N[0] - 1, 12, 4, 1.
2022-09-08 Jakub Jelinek <jakub@redhat.com>
gcc/
* omp-expand.cc (expand_omp_ordered_sink): Add CONT_BB argument.
Add doacross(sink:omp_cur_iteration-1) support.
(expand_omp_ordered_source_sink): Clear counts[fd->ordered + 1].
Adjust expand_omp_ordered_sink caller.
(expand_omp_for_ordered_loops): If counts[fd->ordered + 1] is
non-NULL, set that variable to true at the start of outermost
non-collapsed loop and set it to false at the end of innermost
ordered loop.
(expand_omp_for_generic): If fd->ordered, allocate
1 + (fd->ordered - fd->collapse) further elements in counts array.
Copy to counts + 2 + fd->ordered the counts of fd->collapse ..
fd->ordered - 1 loop if any.
gcc/testsuite/
* c-c++-common/gomp/doacross-7.c: New test.
libgomp/
* libgomp.texi (OpenMP 5.2): Mention that omp_cur_iteration is now
fully supported.
* testsuite/libgomp.c/doacross-4.c: New test.
* testsuite/libgomp.c/doacross-5.c: New test.
* testsuite/libgomp.c/doacross-6.c: New test.
* testsuite/libgomp.c/doacross-7.c: New test.
Jakub Jelinek [Thu, 8 Sep 2022 12:01:14 +0000 (14:01 +0200)]
libgomp: Clarify that omp_display_env is fully implemented
OpenMP 5.2 added
"When called from within a target region the effect is unspecified."
restriction to omp_display_env, so it is ok not to support it in
target regions (worst case we could add an empty implementation
or one with __builtin_trap in there).
Jonathan Wakely [Mon, 22 Aug 2022 14:42:17 +0000 (15:42 +0100)]
libstdc++: Fix for explicit copy ctors in <thread> and <future> [PR106695]
When I changed std::thread and std::async to avoid unnecessary move
construction of temporaries, I introduced a regression where types with
an explicit copy constructor could not be passed to std::thread or
std::async. The fix is to add a constructor instead of using aggregate
initialization of an unnamed temporary.
libstdc++-v3/ChangeLog:
PR libstdc++/106695
* include/bits/std_thread.h (thread::_State_impl): Forward
individual arguments to _Invoker constructor.
(thread::_Invoker): Add constructor. Delete copies.
* include/std/future (__future_base::_Deferred_state): Forward
individual arguments to _Invoker constructor.
(__future_base::_Async_state_impl): Likewise.
* testsuite/30_threads/async/106695.cc: New test.
* testsuite/30_threads/thread/106695.cc: New test.
Jonathan Wakely [Mon, 22 Aug 2022 14:16:16 +0000 (15:16 +0100)]
libstdc++: Check for overflow in regex back-reference [PR106607]
Currently we fail to notice integer overflow when parsing a
back-reference expression, or when converting the parsed result from
long to int. This changes the result to be int, so no conversion is
needed, and uses the overflow-checking built-ins to detect an
out-of-range back-reference.
libstdc++-v3/ChangeLog:
PR libstdc++/106607
* include/bits/regex_compiler.tcc (_Compiler::_M_cur_int_value):
Use built-ins to check for integer overflow in back-reference
number.
* testsuite/28_regex/basic_regex/106607.cc: New test.
Jakub Jelinek [Wed, 7 Sep 2022 08:36:22 +0000 (10:36 +0200)]
openmp: Introduce gimple_omp_ordered_standalone_p
On Sat, Sep 03, 2022 at 10:07:27AM +0200, Jakub Jelinek via Gcc-patches wrote:
> Incrementally, I'd like to change the way we differentiate between
> stand-alone and block-associated ordered constructs, because the current
> way of looking for presence of doacross clause doesn't work well if those
> clauses are removed because they had been invalid (wrong syntax or
> unknown variables in it etc.)
The following, so far only lightly tested, patch implements that.
2022-09-06 Jakub Jelinek <jakub@redhat.com>
gcc/
* gimple.h (enum gf_mask): Add GF_OMP_ORDERED_STANDALONE enumerator.
(gimple_omp_subcode): Use GIMPLE_OMP_ORDERED instead of
GIMPLE_OMP_TEAMS as upper bound.
(gimple_omp_ordered_standalone_p, gimple_omp_ordered_standalone): New
inline functions.
* gimplify.cc (find_standalone_omp_ordered): Look for OMP_ORDERED with
NULL OMP_ORDERED_BODY rather than with OMP_DOACROSS clause.
(gimplify_expr): Call gimple_omp_ordered_standalone for OMP_ORDERED
with NULL OMP_ORDERED_BODY.
* omp-low.cc (check_omp_nesting_restrictions): Use
gimple_omp_ordered_standalone_p test instead of
omp_find_clause (..., OMP_CLAUSE_DOACROSS).
(lower_omp_ordered): Likewise.
* omp-expand.cc (expand_omp, build_omp_regions_1,
omp_make_gimple_edges): Likewise.
gcc/cp/
* pt.cc (tsubst_expr) <case OMP_ORDERED>: If OMP_BODY was NULL, keep
it NULL after instantiation too.
gcc/testsuite/
* c-c++-common/gomp/sink-3.c: Don't expect a superfluous error during
error recovery.
* c-c++-common/gomp/doacross-6.c (foo): Add further tests.
This removed a checking snippet which accidentally was left in in commit r13-2446-g938cda536019cd6a1bc0dd2346381185b420bbf8 ; this caused
fails in gfortran.dg/gomp/doacross-5.f90 (added in that very commit).
Note that a similar but refined check is now done in the middle end.
(The ME version additionally checks whether doacross is present.)
gcc/fortran/
* openmp.cc (resolve_omp_clauses): Remove ordered/linear
check as it is handled now in the middle end.
Jakub Jelinek [Tue, 6 Sep 2022 08:19:45 +0000 (10:19 +0200)]
openmp: Be consistent on parsing offsets between normal sink vector and omp_cur_iteration - 1
For normal sink vectors, we just check that the token is CPP_NUMBER and with
INTEGER_CST value, while for omp_cur_iteration I was additionally requiring
integer_type_node type (so only 1, 001, 0x0001 but not 1L or 1ULL etc.).
I think we need to clarify what we actually should allow in the standard, until
then it is better to be consistent.
2022-09-06 Jakub Jelinek <jakub@redhat.com>
gcc/c/
* c-parser.cc (c_parser_omp_clause_doacross_sink): Don't verify val
in omp_cur_iteration - 1 has integer_type_node type.
gcc/cp/
* parser.cc (cp_parser_omp_clause_doacross_sink): Don't verify val
in omp_cur_iteration - 1 has integer_type_node type.
gcc/testsuite/
* c-c++-common/gomp/doacross-6.c (corge): Don't expect an error here.
Add a few further tests.
Jakub Jelinek [Tue, 6 Sep 2022 08:16:35 +0000 (10:16 +0200)]
openmp: Fix ICE when splitting invalid depend(source)/depend(sink:vec)
As we now create OMP_CLAUSE_DOACROSS rather than OMP_CLAUSE_DEPEND when
depend is used with source/sink modifiers, c_omp_split_clauses can see
OMP_CLAUSE_DOACROSS clause too before we diagnose it as erroneous.
The following patch treats it like OMP_CLAUSE_DEPEND during
the splitting but adds an assertion.
nvptx: Silence unused variable warning in output_constant_pool_contents()
Similar to the rs6000 code, nvptx defines ASM_OUTPUT_DEF_FROM_DECLS as well as
ASM_OUTPUT_DEF. Make sure that the define's parameters are used by referencing
them as (void) to silence a warning in output_constant_pool_contents().
Jakub Jelinek [Mon, 5 Sep 2022 07:39:04 +0000 (09:39 +0200)]
openmp: Partial OpenMP 5.2 doacross and omp_cur_iteration support
The following patch implements part of the OpenMP 5.2 changes related
to ordered loops and with the assumed resolution of
https://github.com/OpenMP/spec/issues/3302 issues.
The changes are:
1) the depend clause on stand-alone ordered constructs has been renamed
to doacross (because depend clause has different syntax on other
constructs) with some syntax changes below, depend clause is deprecated
(we'll deprecate stuff on the GCC side only when we have everything else
from 5.2 implemented)
depend(source) -> doacross(source:) or doacross(source:omp_cur_iteration)
depend(sink:vec) -> doacross(sink:vec) (where vec has the same syntax
as before)
2) in 5.1 and before it has been significant whether ordered clause has or
doesn't have an argument, if it didn't, only block-associated ordered
could appear in the body, if it did, only stand-alone ordered could appear
in the body, all loops had to be perfectly nested, no associated
range-based for loops, no linear clause on work-sharing loop and ordered
clause with an argument wasn't allowed on composite for simd.
In 5.2, whether ordered clause has or doesn't have an argument is
insignificant (except for bugs in the standard, #3302 mentions those),
if the argument is missing, it is simply treated as equal to collapse
argument (if any, otherwise 1). The implementation better should be able
to differentiate between ordered and doacross loops at compile time
which previously was through the absence or presence of the argument,
now it is done through looking at the body of the construct lexically
and looking for stand-alone ordered constructs. If there are any,
it is to be handled as doacross loop, otherwise it is ordered loop
(but in that case ordered argument if present must be equal to collapse
argument - 5.2 says instead it must be one, but that is clearly wrong
and mentioned in #3302) - stand-alone ordered constructs must appear
lexically in the body (and had to before as well). For the restrictions
mentioned above, the for simd restriction is gone (stand-alone ordered
can't appear in simd construct, so that is enough), and the other rules
are expected to be changed into something related to presence of
stand-alone ordered constructs in the body
3) 5.2 allows a new syntax, doacross(sink:omp_cur_iteration-1), which
means wait for previous iteration in the iteration space of all the
associated loops
The following patch implements that, except that we sorry for now
on the doacross(sink:omp_cur_iteration-1) syntax during omp expansion
because library side isn't done yet for it. It doesn't implement it for
the Fortran FE either.
Incrementally, I'd like to change the way we differentiate between
stand-alone and block-associated ordered constructs, because the current
way of looking for presence of doacross clause doesn't work well if those
clauses are removed because they had been invalid (wrong syntax or
unknown variables in it etc.) and of course implement
doacross(sink:omp_cur_iteration-1).
2022-09-03 Jakub Jelinek <jakub@redhat.com>
gcc/
* tree-core.h (enum omp_clause_code): Add OMP_CLAUSE_DOACROSS.
(enum omp_clause_depend_kind): Remove OMP_CLAUSE_DEPEND_SOURCE
and OMP_CLAUSE_DEPEND_SINK, add OMP_CLAUSE_DEPEND_INVALID.
(enum omp_clause_doacross_kind): New type.
(struct tree_omp_clause): Add subcode.doacross_kind member.
* tree.h (OMP_CLAUSE_DEPEND_SINK_NEGATIVE): Remove.
(OMP_CLAUSE_DOACROSS_KIND): Define.
(OMP_CLAUSE_DOACROSS_SINK_NEGATIVE): Define.
(OMP_CLAUSE_DOACROSS_DEPEND): Define.
(OMP_CLAUSE_ORDERED_DOACROSS): Define.
* tree.cc (omp_clause_num_ops, omp_clause_code_name): Add
OMP_CLAUSE_DOACROSS entries.
* tree-nested.cc (convert_nonlocal_omp_clauses,
convert_local_omp_clauses): Handle OMP_CLAUSE_DOACROSS.
* tree-pretty-print.cc (dump_omp_clause): Don't handle
OMP_CLAUSE_DEPEND_SOURCE and OMP_CLAUSE_DEPEND_SINK. Handle
OMP_CLAUSE_DOACROSS.
* gimplify.cc (gimplify_omp_depend): Don't handle
OMP_CLAUSE_DEPEND_SOURCE and OMP_CLAUSE_DEPEND_SINK.
(gimplify_scan_omp_clauses): Likewise. Handle OMP_CLAUSE_DOACROSS.
(gimplify_adjust_omp_clauses): Handle OMP_CLAUSE_DOACROSS.
(find_standalone_omp_ordered): New function.
(gimplify_omp_for): When OMP_CLAUSE_ORDERED is present, search
body for OMP_ORDERED with OMP_CLAUSE_DOACROSS and if found,
set OMP_CLAUSE_ORDERED_DOACROSS.
(gimplify_omp_ordered): Don't handle OMP_CLAUSE_DEPEND_SINK or
OMP_CLAUSE_DEPEND_SOURCE, instead check OMP_CLAUSE_DOACROSS, adjust
diagnostics that presence or absence of ordered clause parameter
is irrelevant. Handle doacross(sink:omp_cur_iteration-1). Use
actual user name of the clause - doacross or depend - in diagnostics.
* omp-general.cc (omp_extract_for_data): Don't set fd->ordered
if !OMP_CLAUSE_ORDERED_DOACROSS (t). If
OMP_CLAUSE_ORDERED_DOACROSS (t) but !OMP_CLAUSE_ORDERED_EXPR (t),
set fd->ordered to -1 and set it after the loop in that case to
fd->collapse.
* omp-low.cc (check_omp_nesting_restrictions): Don't handle
OMP_CLAUSE_DEPEND_SOURCE nor OMP_CLAUSE_DEPEND_SINK, instead check
OMP_CLAUSE_DOACROSS. Use actual user name of the clause - doacross
or depend - in diagnostics. Diagnose mixing of stand-alone and
block associated ordered constructs binding to the same loop.
(lower_omp_ordered_clauses): Don't handle OMP_CLAUSE_DEPEND_SINK,
instead handle OMP_CLAUSE_DOACROSS.
(lower_omp_ordered): Look for OMP_CLAUSE_DOACROSS instead of
OMP_CLAUSE_DEPEND.
(lower_depend_clauses): Don't handle OMP_CLAUSE_DEPEND_SOURCE and
OMP_CLAUSE_DEPEND_SINK.
* omp-expand.cc (expand_omp_ordered_sink): Emit a sorry for
doacross(sink:omp_cur_iteration-1).
(expand_omp_ordered_source_sink): Use
OMP_CLAUSE_DOACROSS_SINK_NEGATIVE instead of
OMP_CLAUSE_DEPEND_SINK_NEGATIVE. Use actual user name of the clause
- doacross or depend - in diagnostics.
(expand_omp): Look for OMP_CLAUSE_DOACROSS clause instead of
OMP_CLAUSE_DEPEND.
(build_omp_regions_1): Likewise.
(omp_make_gimple_edges): Likewise.
* lto-streamer-out.cc (hash_tree): Handle OMP_CLAUSE_DOACROSS.
* tree-streamer-in.cc (unpack_ts_omp_clause_value_fields): Likewise.
* tree-streamer-out.cc (pack_ts_omp_clause_value_fields): Likewise.
gcc/c-family/
* c-pragma.h (enum pragma_omp_clause): Add PRAGMA_OMP_CLAUSE_DOACROSS.
* c-omp.cc (c_finish_omp_depobj): Check also for OMP_CLAUSE_DOACROSS
clause and diagnose it. Don't handle OMP_CLAUSE_DEPEND_SOURCE and
OMP_CLAUSE_DEPEND_SINK. Assert kind is not OMP_CLAUSE_DEPEND_INVALID.
gcc/c/
* c-parser.cc (c_parser_omp_clause_name): Handle doacross.
(c_parser_omp_clause_depend_sink): Renamed to ...
(c_parser_omp_clause_doacross_sink): ... this. Add depend_p argument.
Handle parsing of doacross(sink:omp_cur_iteration-1). Use
OMP_CLAUSE_DOACROSS_SINK_NEGATIVE instead of
OMP_CLAUSE_DEPEND_SINK_NEGATIVE, build OMP_CLAUSE_DOACROSS instead
of OMP_CLAUSE_DEPEND and set OMP_CLAUSE_DOACROSS_DEPEND flag on it.
(c_parser_omp_clause_depend): Use OMP_CLAUSE_DOACROSS_SINK and
OMP_CLAUSE_DOACROSS_SOURCE instead of OMP_CLAUSE_DEPEND_SINK and
OMP_CLAUSE_DEPEND_SOURCE, build OMP_CLAUSE_DOACROSS for depend(source)
and set OMP_CLAUSE_DOACROSS_DEPEND on it.
(c_parser_omp_clause_doacross): New function.
(c_parser_omp_all_clauses): Handle PRAGMA_OMP_CLAUSE_DOACROSS.
(c_parser_omp_depobj): Use OMP_CLAUSE_DEPEND_INVALID instead of
OMP_CLAUSE_DEPEND_SOURCE.
(c_parser_omp_for_loop): Don't diagnose here linear clause together
with ordered with argument.
(c_parser_omp_simd): Don't diagnose ordered clause with argument on
for simd.
(OMP_ORDERED_DEPEND_CLAUSE_MASK): Add PRAGMA_OMP_CLAUSE_DOACROSS.
(c_parser_omp_ordered): Handle also doacross and adjust for it
diagnostic wording.
* c-typeck.cc (c_finish_omp_clauses): Handle OMP_CLAUSE_DOACROSS.
Don't handle OMP_CLAUSE_DEPEND_SOURCE and OMP_CLAUSE_DEPEND_SINK.
gcc/cp/
* parser.cc (cp_parser_omp_clause_name): Handle doacross.
(cp_parser_omp_clause_depend_sink): Renamed to ...
(cp_parser_omp_clause_doacross_sink): ... this. Add depend_p
argument. Handle parsing of doacross(sink:omp_cur_iteration-1). Use
OMP_CLAUSE_DOACROSS_SINK_NEGATIVE instead of
OMP_CLAUSE_DEPEND_SINK_NEGATIVE, build OMP_CLAUSE_DOACROSS instead
of OMP_CLAUSE_DEPEND and set OMP_CLAUSE_DOACROSS_DEPEND flag on it.
(cp_parser_omp_clause_depend): Use OMP_CLAUSE_DOACROSS_SINK and
OMP_CLAUSE_DOACROSS_SOURCE instead of OMP_CLAUSE_DEPEND_SINK and
OMP_CLAUSE_DEPEND_SOURCE, build OMP_CLAUSE_DOACROSS for depend(source)
and set OMP_CLAUSE_DOACROSS_DEPEND on it.
(cp_parser_omp_clause_doacross): New function.
(cp_parser_omp_all_clauses): Handle PRAGMA_OMP_CLAUSE_DOACROSS.
(cp_parser_omp_depobj): Use OMP_CLAUSE_DEPEND_INVALID instead of
OMP_CLAUSE_DEPEND_SOURCE.
(cp_parser_omp_for_loop): Don't diagnose here linear clause together
with ordered with argument.
(cp_parser_omp_simd): Don't diagnose ordered clause with argument on
for simd.
(OMP_ORDERED_DEPEND_CLAUSE_MASK): Add PRAGMA_OMP_CLAUSE_DOACROSS.
(cp_parser_omp_ordered): Handle also doacross and adjust for it
diagnostic wording.
* pt.cc (tsubst_omp_clause_decl): Use
OMP_CLAUSE_DOACROSS_SINK_NEGATIVE instead of
OMP_CLAUSE_DEPEND_SINK_NEGATIVE.
(tsubst_omp_clauses): Handle OMP_CLAUSE_DOACROSS.
(tsubst_expr): Use OMP_CLAUSE_DEPEND_INVALID instead of
OMP_CLAUSE_DEPEND_SOURCE.
* semantics.cc (cp_finish_omp_clause_depend_sink): Rename to ...
(cp_finish_omp_clause_doacross_sink): ... this.
(finish_omp_clauses): Handle OMP_CLAUSE_DOACROSS. Don't handle
OMP_CLAUSE_DEPEND_SOURCE and OMP_CLAUSE_DEPEND_SINK.
gcc/fortran/
* trans-openmp.cc (gfc_trans_omp_clauses): Use
OMP_CLAUSE_DOACROSS_SINK_NEGATIVE instead of
OMP_CLAUSE_DEPEND_SINK_NEGATIVE, build OMP_CLAUSE_DOACROSS
clause instead of OMP_CLAUSE_DEPEND and set OMP_CLAUSE_DOACROSS_DEPEND
on it.
gcc/testsuite/
* c-c++-common/gomp/doacross-2.c: Adjust expected diagnostics.
* c-c++-common/gomp/doacross-5.c: New test.
* c-c++-common/gomp/doacross-6.c: New test.
* c-c++-common/gomp/nesting-2.c: Adjust expected diagnostics.
* c-c++-common/gomp/ordered-3.c: Likewise.
* c-c++-common/gomp/sink-3.c: Likewise.
* gfortran.dg/gomp/nesting-2.f90: Likewise.
Peter Bergner [Thu, 1 Sep 2022 02:14:36 +0000 (21:14 -0500)]
rs6000: Don't ICE when we disassemble an MMA variable [PR101322]
When we expand an MMA disassemble built-in with C++ using a pointer that
is cast to a valid MMA type, the type isn't passed down to the expand
machinery and we end up using the base type of the pointer which leads to
an ICE. This patch enforces we always use the correct MMA type regardless
of the pointer type being used.
2022-08-31 Peter Bergner <bergner@linux.ibm.com>
gcc/
PR target/101322
* config/rs6000/rs6000-builtin.cc (rs6000_gimple_fold_mma_builtin):
Enforce the use of a valid MMA pointer type.
gcc/testsuite/
PR target/101322
* g++.target/powerpc/pr101322.C: New test.
cselib: add function to check if SET is redundant [PR106187]
A SET operation that writes memory may have the same value as an
earlier store but if the alias sets of the new and earlier store do
not conflict then the set is not truly redundant. This can happen,
for example, if objects of different types share a stack slot.
To fix this we define a new function in cselib that first checks for
equality and if that is successful then finds the earlier store in the
value history and checks the alias sets.
The routine is used in two places elsewhere in the compiler:
cfgcleanup and postreload.
gcc/ChangeLog:
PR rtl-optimization/106187
* alias.h (mems_same_for_tbaa_p): Declare.
* alias.cc (mems_same_for_tbaa_p): New function.
* dse.cc (record_store): Use it instead of open-coding
alias check.
* cselib.h (cselib_redundant_set_p): Declare.
* cselib.cc: Include alias.h
(cselib_redundant_set_p): New function.
* cfgcleanup.cc: (mark_effect): Use cselib_redundant_set_p instead
of rtx_equal_for_cselib_p.
* postreload.cc (reload_cse_simplify): Use cselib_redundant_set_p.
(reload_cse_noop_set_p): Delete.
Richard Earnshaw [Wed, 11 May 2022 12:08:40 +0000 (13:08 +0100)]
arm: correctly handle misaligned MEMs on MVE [PR105463]
Vector operations in MVE must be aligned to the element size, so if we
are asked for a misaligned move in a wider mode we must recast it to a
form suitable for the known alignment (larger elements have better
address offset ranges, so there is some advantage to using wider
element sizes if possible). Whilst fixing this, also rework the
predicates used for validating operands - the Neon predicates are
not right for MVE.
gcc/ChangeLog:
PR target/105463
* config/arm/mve.md (*movmisalign<mode>_mve_store): Use
mve_memory_operand.
(*movmisalign<mode>_mve_load): Likewise.
* config/arm/vec-common.md (movmisalign<mode>): Convert to generator
form...
(@movmisalign<mode>): ... thus. Use generic predicates and then
rework operands if they are not valid. For MVE rework to a
narrower element size if the alignment is not high enough.
Andrew Stubbs [Fri, 5 Aug 2022 12:28:50 +0000 (13:28 +0100)]
omp-simd-clone: Allow fixed-lane vectors
The vecsize_int/vecsize_float has an assumption that all arguments will use
the same bitsize, and vary the number of lanes according to the element size,
but this is inappropriate on targets where the number of lanes is fixed and
the bitsize varies (i.e. amdgcn).
With this change the vecsize can be left zero and the vectorization factor will
be the same for all types.
gcc/ChangeLog:
* doc/tm.texi: Regenerate.
* omp-simd-clone.cc (simd_clone_adjust_return_type): Allow zero
vecsize.
(simd_clone_adjust_argument_types): Likewise.
* target.def (compute_vecsize_and_simdlen): Document the new
vecsize_int and vecsize_float semantics.
Andrew Stubbs [Fri, 15 Jul 2022 08:47:36 +0000 (09:47 +0100)]
amdgcn: Vector procedure call ABI
Adjust the (unofficial) procedure calling ABI such that vector arguments are
passed in vector registers, not on the stack. Scalar arguments continue to
be passed in scalar registers, making a total of 12 argument registers.
The return value is also moved to a vector register (even for scalars; it
would be possible to retain the scalar location, using untyped_call, but
there's no obvious advantage in doing so).
Jakub Jelinek [Thu, 1 Sep 2022 09:07:44 +0000 (11:07 +0200)]
Fix up dump_printf_loc format attribute and adjust uses [PR106782]
As discussed on IRC, the r13-2299-g68c61c2daa1f bug only got missed
because dump_printf_loc had incorrect format attribute and therefore
almost no -Wformat=* checking was performed on it.
3, 0 are suitable for function with (whatever, whatever, const char *, va_list)
arguments, not for (whatever, whatever, const char *, ...), that one should
use 3, 4.
There are 3 spots where the mismatch was worse though, two using %u or %d
for unsigned HOST_WIDE_INT argument and one %T for enum argument (promoted
to int) and this backport just fixes those spots.
2022-09-01 Jakub Jelinek <jakub@redhat.com>
PR other/106782
* tree-vect-slp.cc (vect_print_slp_tree): Use
HOST_WIDE_INT_PRINT_UNSIGNED instead of %u.
* tree-vect-loop.cc (vect_estimate_min_profitable_iters): Use
HOST_WIDE_INT_PRINT_UNSIGNED instead of %d.
* tree-vect-slp-patterns.cc (vect_pattern_validate_optab): Use %G
instead of %T and STMT_VINFO_STMT (SLP_TREE_REPRESENTATIVE (node))
instead of SLP_TREE_DEF_TYPE (node).
The OG11 version contained code that was not part of the upstream version
as the approach changed; however, when OG12 was created, those were
accidentally applied to OG12.
This commit removes (reverts) the additional code, i.e. the OG12 commits:
* dwarf2cfi.cc (get_cfa_from_loc_descr): Support register spans
with DW_OP_piece and DW_OP_LLVM_piece_end.
* dwarf2out.cc (build_cfa_loc): Support register spans.
include/ChangeLog:
* dwarf2.def (DW_OP_LLVM_piece_end): New extension operator.
Tobias Burnus [Tue, 30 Aug 2022 19:44:10 +0000 (21:44 +0200)]
OpenMP: Support reverse offload (middle end part)
gcc/ChangeLog:
* internal-fn.cc (expand_GOMP_TARGET_REV): New.
* internal-fn.def (GOMP_TARGET_REV): New.
* lto-cgraph.cc (lto_output_node, verify_node_partition): Mark
'omp target device_ancestor_host' as in_other_partition and don't
error if absent.
* omp-low.cc (create_omp_child_function): Mark as 'noclone'.
* omp-expand.cc (expand_omp_target): For reverse offload, remove
sorry, use device = GOMP_DEVICE_HOST_FALLBACK and create
empty-body nohost function.
* omp-offload.cc (execute_omp_device_lower): Handle
IFN_GOMP_TARGET_REV.
(pass_omp_target_link::execute): For ACCEL_COMPILER, don't
nullify fn argument for reverse offload
libgomp/ChangeLog:
* libgomp.texi (OpenMP 5.0): Mark 'ancestor' as implemented but
refer to 'requires'.
* testsuite/libgomp.c-c++-common/reverse-offload-1-aux.c: New test.
* testsuite/libgomp.c-c++-common/reverse-offload-1.c: New test.
* testsuite/libgomp.fortran/reverse-offload-1-aux.f90: New test.
* testsuite/libgomp.fortran/reverse-offload-1.f90: New test.
Tobias Burnus [Tue, 30 Aug 2022 17:48:37 +0000 (19:48 +0200)]
gfortran.dg/gomp/depend-6.f90: minor fix + dump update
Contains the minor fix of upstream commit r13-2152-gf05e3b2c63f3307ba405900f1a80c25b2e87b0a3
to avoid setting the same array element twice, which is:
Exactly the same as previous commit for depend-4.f90, r13-2151.
Additionally, it updates the expected tree dumps due to differences
between GCC 12/OG12 and mainline (GCC13); the latter seems to
use a restricted pointer of type '&a[1]' while OG12 has pointerplus
expressions like 'a + 4'.
The changes include -m32/-m64 handling for the depobj var and in
two cases, the expected count increases from 1 to 2 for code like
'D.\[0-9\]+ = daa'.
Marek Polacek [Mon, 29 Aug 2022 20:54:05 +0000 (16:54 -0400)]
c++: __has_builtin gives the wrong answer [PR106759]
We've supported __is_nothrow_constructible since r11-4386, but
names_builtin_p didn't know about it, so it gave the wrong answer for
#if __has_builtin(__is_nothrow_constructible)
...
#endif
I've tested all C++-only built-ins and only two were missing.
PR c++/106759
gcc/cp/ChangeLog:
* cp-objcp-common.cc (names_builtin_p): Handle RID_IS_NOTHROW_ASSIGNABLE
and RID_IS_NOTHROW_CONSTRUCTIBLE.
Peter Bergner [Sun, 28 Aug 2022 00:44:16 +0000 (19:44 -0500)]
rs6000: Allow conversions of MMA pointer types [PR106017]
GCC incorrectly disables conversions between MMA pointer types, which
are allowed with clang. The original intent was to disable conversions
between MMA types and other other types, but pointer conversions should
have been allowed. The fix is to just remove the MMA pointer conversion
handling code altogether.
H.J. Lu [Thu, 18 Aug 2022 21:17:33 +0000 (14:17 -0700)]
x86: Cast stride to __PTRDIFF_TYPE__ in AMX intrinsics
On 64-bit Windows, long is 32 bits and can't be used as stride in memory
operand when base is a pointer which is 64 bits. Cast stride to
__PTRDIFF_TYPE__, instead of long.
The following patch expands IEEE_CLASS inline in the FE but only for the
powerpc64le-linux IEEE quad real(kind=16), using the __builtin_fpclassify
builtin and explicit check of the MSB mantissa bit in place of missing
__builtin_signbit builtin.
Jakub Jelinek [Wed, 24 Aug 2022 07:57:09 +0000 (09:57 +0200)]
i386: Fix up mode iterators that weren't expanded [PR106721]
Currently, when md file reader sees <something> and something is valid mode
(or code) attribute but which doesn't include case for the current mode
(or code), it just keeps the <something> untouched.
I went through all cases matching <[a-zA-Z] in tmp-mddump.md after make mddump.
One of the cases was related to the V*HF mode additions and there was one typo.
From what I can see, this has been voted in as a DR and as it means
we warn less often than before in -std={gnu,c}++2{0,3} modes or with
-Wvolatile, I wonder if it shouldn't be backported to affected release
branches as well.
2022-08-16 Jakub Jelinek <jakub@redhat.com>
* typeck.cc (cp_build_modify_expr): Implement
P2327R1 - De-deprecating volatile compound operations. Don't warn
for |=, &= or ^= with volatile lhs.
* expr.cc (mark_use) <case MODIFY_EXPR>: Adjust warning wording,
leave out simple.
* g++.dg/cpp2a/volatile1.C: Adjust for de-deprecation of volatile
compound |=, &= and ^= operations.
* g++.dg/cpp2a/volatile3.C: Likewise.
* g++.dg/cpp2a/volatile5.C: Likewise.
Jakub Jelinek [Mon, 15 Aug 2022 11:56:57 +0000 (13:56 +0200)]
ifcvt: Fix up noce_convert_multiple_sets [PR106590]
The following testcase is miscompiled on x86_64-linux.
The problem is in the noce_convert_multiple_sets optimization.
We essentially have:
if (g == 1)
{
g = 1;
f = 23;
}
else
{
g = 2;
f = 20;
}
and for each insn try to create a conditional move sequence.
There is code to detect overlap with the regs used in the condition
and the destinations, so we actually try to construct:
tmp_g = g == 1 ? 1 : 2;
f = g == 1 ? 23 : 20;
g = tmp_g;
which is fine. But, we actually try to create two different
conditional move sequences in each case, seq1 with the whole
(eq (reg/v:HI 82 [ g ]) (const_int 1 [0x1]))
condition and seq2 with cc_cmp
(eq (reg:CCZ 17 flags) (const_int 0 [0]))
to rely on the earlier present comparison. In each case, we
compare the rtx costs and choose the cheaper sequence (seq1 if both
have the same cost).
The problem is that with the skylake tuning,
tmp_g = g == 1 ? 1 : 2;
is actually expanded as
tmp_g = (g == 1) + 1;
in seq1 (which clobbers (reg 17 flags)) and as a cmov in seq2
(which doesn't). The tuning says both have the same cost, so we
pick seq1. Next we check sequences for
f = g == 1 ? 23 : 20; and here the seq2 cmov is cheaper, but it
uses (reg 17 flags) which has been clobbered earlier.
The following patch fixes that by detecting if we in the chosen
sequence clobber some register mentioned in cc_cmp or rev_cc_cmp,
and if yes, arranges for only seq1 (i.e. sequences that emit the
comparison itself) to be used after that.
2022-08-15 Jakub Jelinek <jakub@redhat.com>
PR rtl-optimization/106590
* ifcvt.cc (check_for_cc_cmp_clobbers): New function.
(noce_convert_multiple_sets_1): If SEQ sets or clobbers any regs
mentioned in cc_cmp or rev_cc_cmp, don't consider seq2 for any
further conditional moves.
Harald Anlauf [Tue, 23 Aug 2022 20:16:14 +0000 (22:16 +0200)]
Fortran: improve error recovery while simplifying size of bad array [PR103694]
gcc/fortran/ChangeLog:
PR fortran/103694
* simplify.cc (simplify_size): The size expression of an array cannot
be simplified if an error occurs while resolving the array spec.
gcc/testsuite/ChangeLog:
PR fortran/103694
* gfortran.dg/pr103694.f90: New test.
Since 256-bit vector integer comparison is under TARGET_AVX2,
and gimple folding for vblendvpd/vblendvps/vpblendvb relies on that.
Restrict gimple fold condition to TARGET_AVX2.
Jonathan Wakely [Tue, 23 Aug 2022 14:46:16 +0000 (15:46 +0100)]
libstdc++: Fix visit<void>(v) for non-void visitors [PR106589]
The optimization for the common case of std::visit forgot to handle the
edge case of passing zero variants to a non-void visitor and converting
the result to void.
libstdc++-v3/ChangeLog:
PR libstdc++/106589
* include/std/variant (__do_visit): Handle is_void<R> for zero
argument case.
* testsuite/20_util/variant/visit_r.cc: Check std::visit<void>(v).
Kewen Lin [Tue, 16 Aug 2022 05:18:51 +0000 (00:18 -0500)]
vect: Don't allow vect_emulated_vector_p type in vectorizable_call [PR106322]
As PR106322 shows, in some cases for some vector type whose
TYPE_MODE is a scalar integral mode instead of a vector mode,
it's possible to obtain wrong target support information when
querying with the scalar integral mode. For example, for the
test case in PR106322, on ppc64 32bit vectorizer gets vector
type "vector(2) short unsigned int" for scalar type "short
unsigned int", its mode is SImode instead of V2HImode. The
target support querying checks umul_highpart optab with SImode
and considers it's supported, then vectorizer further generates
.MULH IFN call for that vector type. Unfortunately it's wrong
to use SImode support for that vector type multiply highpart
here.
This patch is to teach vectorizable_call analysis not to allow
vect_emulated_vector_p type for both vectype_in and vectype_out
as Richi suggested.
PR tree-optimization/106322
gcc/ChangeLog:
* tree-vect-stmts.cc (vectorizable_call): Don't allow
vect_emulated_vector_p type for both vectype_in and vectype_out.
gcc/testsuite/ChangeLog:
* gcc.target/i386/pr106322.c: New test.
* gcc.target/powerpc/pr106322.c: New test.
Kewen.Lin [Tue, 16 Aug 2022 05:24:07 +0000 (00:24 -0500)]
rs6000: Adjust mov optabs for opaque modes [PR103353]
As PR103353 shows, we may want to continue to expand built-in
function __builtin_vsx_lxvp, even if we have already emitted
error messages about some missing required conditions. As
shown in that PR, without one explicit mov optab on OOmode
provided, it would call emit_move_insn recursively.
So this patch is to allow the mov pattern to be generated during
expanding phase if compiler has already seen errors.
PR target/103353
gcc/ChangeLog:
* config/rs6000/mma.md (define_expand movoo): Move TARGET_MMA condition
check to preparation statements and add handlings for !TARGET_MMA.
(define_expand movxo): Likewise.
Jonathan Wakely [Mon, 22 Aug 2022 16:24:27 +0000 (17:24 +0100)]
libstdc++: Document linker option for C++23 <stacktrace> [PR105678]
libstdc++-v3/ChangeLog:
PR libstdc++/105678
* doc/xml/manual/using.xml: Document -lstdc++_libbacktrace
requirement for using std::stacktrace. Also adjust -frtti and
-fexceptions to document non-default (i.e. negative) forms.
* doc/html/*: Regenerate.
Additionally, it updates the expected tree dumps due to differences
between GCC 12/OG12 and mainline (GCC13); the latter seems to
use a restricted pointer of type '&a[1]' while OG12 has pointerplus
expressions like 'a + 4'.
The changes include -m32/-m64 handling for the depobj var and in
two cases, the expected count increases from 1 to 2 for code like
'D.\[0-9\]+ = daa'.
gcc/testsuite/
* gfortran.dg/gomp/depend-4.f90: Update expected tree dumps.
Tobias Burnus [Tue, 23 Aug 2022 11:00:38 +0000 (13:00 +0200)]
Fortran: OpenMP fix declare simd inside modules and absent linear step [PR106566]
gcc/fortran/ChangeLog:
PR fortran/106566
* openmp.cc (gfc_match_omp_clauses): Fix setting linear-step value
to 1 when not specified.
(gfc_match_omp_declare_simd): Accept module procedures.
gcc/testsuite/ChangeLog:
PR fortran/106566
* gfortran.dg/gomp/declare-simd-4.f90: New test.
* gfortran.dg/gomp/declare-simd-5.f90: New test.
* gfortran.dg/gomp/declare-simd-6.f90: New test.
Tobias Burnus [Tue, 23 Aug 2022 09:35:01 +0000 (11:35 +0200)]
gcn/mkoffload: Cleanup temporary dbgobj file
The file (suffix ".mkoffload.dbg.o") used to save the dbgobj data
data has to be passed to maybe_unlink for cleanup or -v -save-temps stderr
diagnostic. That was missed before.
This is a partial backport of commit r13-2125, "mkoffload: Cleanup
temporary omp_requires_file", only for GCN's mkoffload and its dbgobj
file as 'omp requires' is not supported on GCC 12 and, hence,
omp_requires_file does not exist on this branch.
gcc/ChangeLog:
* config/gcn/mkoffload.cc (main): Add dbgobj to files_to_cleanup.
Tobias Burnus [Mon, 22 Aug 2022 07:53:25 +0000 (09:53 +0200)]
lto-wrapper.cc: Delete offload_names temp files in case of error [PR106686]
Usually, the caller takes care of the .o files for the offload compilers
(suffix: ".target.o"). However, if an error occurs during processing
(e.g. fatal error by lto1), they were not deleted.
gcc/ChangeLog:
PR lto/106686
* lto-wrapper.cc (free_array_of_ptrs): Move before tool_cleanup.
(tool_cleanup): Unlink offload_names.
(compile_offload_image): Take filename argument to set it early.
(compile_images_for_offload_targets): Update call; set
offload_names to NULL after freeing the array.
Tobias Burnus [Fri, 19 Aug 2022 14:21:43 +0000 (16:21 +0200)]
mkoffload: Cleanup temporary omp_requires_file
The file (suffix ".mkoffload.omp_requires") used to save the 'omp requires'
data has to be passed to maybe_unlink for cleanup or -v -save-temps stderr
diagnostic. That was missed before. - For GCN, the same has to be done for
the files with suffix ".mkoffload.dbg.o".
gcc/ChangeLog:
* config/gcn/mkoffload.cc (main): Add omp_requires_file and dbgobj to
files_to_cleanup.
* config/i386/intelmic-mkoffload.cc (prepare_target_image): Add
omp_requires_file to temp_files.
* config/nvptx/mkoffload.cc (omp_requires_file): New global static var.
(main): Remove local omp_requires_file var.
(tool_cleanup): Handle omp_requires_file.
projects / thirdparty / gcc.git / log
