Jason Merrill [Mon, 2 Mar 2020 19:42:47 +0000 (14:42 -0500)]
c++: Fix attributes with lambda and trailing return type.
My fix for 60503 fixed handling of C++11 attributes following the
lambda-declarator. My patch for 89640 re-added support for GNU attributes,
but attributes after the trailing return type were parsed as applying to the
return type rather than to the function. This patch adjusts parsing of a
trailing-return-type to ignore GNU attributes at the end of the declaration
so that they will be applied to the declaration as a whole.
I also considered parsing the attributes between the closing paren and the
trailing-return-type, and tried a variety of approaches to implementing
that, but I think it's better to stick with the documented rule that "An
attribute specifier list may appear immediately before the comma, '=' or
semicolon terminating the declaration of an identifier...." Anyone
disagree?
Meanwhile, C++ committee discussion about the lack of any way to apply
attributes to a lambda op() seems to have concluded that they should go
between the introducer and declarator, so I've implemented that as well.
gcc/cp/ChangeLog
2020-03-02 Jason Merrill <jason@redhat.com>
PR c++/90333
PR c++/89640
PR c++/60503
* parser.c (cp_parser_type_specifier_seq): Don't parse attributes in
a trailing return type.
(cp_parser_lambda_declarator_opt): Parse C++11 attributes before
parens.
Jakub Jelinek [Thu, 27 Feb 2020 09:45:30 +0000 (10:45 +0100)]
gimplify: Don't optimize register const vars to static [PR93949]
The following testcase is rejected, while it was accepted in 3.4 and earlier
(before tree-ssa merge).
The problem is that we decide to promote the const variable to TREE_STATIC,
but TREE_STATIC DECL_REGISTER VAR_DECLs may only be the global register vars
and so assemble_variable/make_decl_rtl diagnoses it.
Either we do what the following patch does, where we could consider
register as a hint the user doesn't want such optimization, because if
something is forced static, it is not "register" anymore and register static
is not valid in C either, or we could clear DECL_REGISTER instead, but would
still need to punt at least on DECL_HARD_REGISTER cases.
Jakub Jelinek [Thu, 27 Feb 2020 10:21:52 +0000 (11:21 +0100)]
sccvn: Punt on ref->size not multiple of 8 for memset (, 123, ) in 9.x [PR93945]
And here is the corresponding 9.x change where we the patch just punts if
ref->size is not whole bytes, like we already punt if offseti is not byte
aligned.
2020-02-27 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/93945
* tree-ssa-sccvn.c (vn_reference_lookup_3): For memset with non-zero
second operand, require ref->size to be a multiple of BITS_PER_UNIT.
Carl Love [Thu, 27 Feb 2020 00:22:46 +0000 (18:22 -0600)]
PPC64, fix documentation for __builtin_crypto_vpmsum* builtin functions.
PR target/91276 - Doc typos in __builtin_crypto_vpmsum*
gcc/ChangeLog:
2020-02-26 Carl Love <cel@us.ibm.com>
PR target/91276
* doc/extend.texi (PowerPC AltiVec Built-in Functions available on
ISA 3.0): The builtin-function name __builtin_crypto_vpmsumb is only
for the vector unsigned short arguments. It is also listed as the
name of the built-in for arguments vector unsigned short,
vector unsigned int and vector unsigned long long built-ins. The
name of the builtins for these arguments should be:
__builtin_crypto_vpmsumh, __builtin_crypto_vpmsumw and
__builtin_crypto_vpmsumd respectively.
Marek Polacek [Fri, 20 Dec 2019 23:30:04 +0000 (23:30 +0000)]
PR c++/92745 - bogus error when initializing array of vectors.
In r268428 I changed reshape_init_r in such a way that when it sees
a nested { } in a CONSTRUCTOR with missing braces, it just returns
the initializer:
+ else if (COMPOUND_LITERAL_P (stripped_init)
...
+ ++d->cur;
+ gcc_assert (!BRACE_ENCLOSED_INITIALIZER_P (stripped_init));
+ return init;
But as this test shows, that's incorrect: if TYPE is an array, we need
to proceed to reshape_init_array_1 which will iterate over the array
initializers:
6006 /* Loop until there are no more initializers. */
6007 for (index = 0;
6008 d->cur != d->end && (!sized_array_p || index <= max_index_cst);
6009 ++index)
6010 {
and update d.cur accordingly. In other words, when reshape_init gets
we recurse on the first element:
{col[0][0], col[1][0], col[2][0], col[3][0]}
and we can't just move d.cur to point to
{col[0][1], col[1][1], col[2][1], col[3][1]}
and return; we need to iterate, so that d.cur ends up being properly
updated, and after all initializers have been seen, points to d.end.
Currently we skip the loop, wherefore we hit this:
6502 /* Make sure all the element of the constructor were used. Otherwise,
6503 issue an error about exceeding initializers. */
6504 if (d.cur != d.end)
6505 {
6506 if (complain & tf_error)
6507 error ("too many initializers for %qT", type);
6508 return error_mark_node;
6509 }
gcc/cp/ChangeLog
2019-12-19 Marek Polacek <polacek@redhat.com>
PR c++/92745 - bogus error when initializing array of vectors.
* decl.c (reshape_init_r): For a nested compound literal, do
call reshape_init_{class,array,vector}.
gcc/testsuite/ChangeLog
2019-12-19 Marek Polacek <polacek@redhat.com>
Jakub Jelinek <jakub@redhat.com>
PR c++/92745 - bogus error when initializing array of vectors.
* g++.dg/cpp0x/initlist118.C: New test.
* g++.dg/cpp0x/initlist118.C: Add -Wno-psabi -w to dg-options.
Jason Merrill [Wed, 26 Feb 2020 18:03:23 +0000 (13:03 -0500)]
cgraph: A COMDAT decl always has non-zero address.
We should be able to assume that a template instantiation or other COMDAT
has non-zero address even if MAKE_DECL_ONE_ONLY for the target sets
DECL_WEAK and we haven't yet decided to emit a definition in this
translation unit.
gcc/ChangeLog
2020-02-26 Jason Merrill <jason@redhat.com>
PR c++/92003
* symtab.c (symtab_node::nonzero_address): A DECL_COMDAT decl has
non-zero address even if weak and not yet defined.
Jason Merrill [Wed, 26 Feb 2020 18:03:23 +0000 (13:03 -0500)]
c++: Fix constexpr vs. omitted aggregate init.
Value-initialization is importantly different from {}-initialization for
this testcase, where the former calls the deleted S constructor and the
latter initializes S happily.
gcc/cp/ChangeLog
2020-02-26 Jason Merrill <jason@redhat.com>
PR c++/90951
* constexpr.c (cxx_eval_array_reference): {}-initialize missing
elements instead of value-initializing them.
Jason Merrill [Wed, 26 Feb 2020 18:03:23 +0000 (13:03 -0500)]
c++: Fix decltype of empty pack expansion of parm.
In unevaluated context, we only substitute a single PARM_DECL, not the
entire chain, but the handling of an empty pack expansion was missing that
check.
gcc/cp/ChangeLog
2020-02-26 Jason Merrill <jason@redhat.com>
PR c++/93140
* pt.c (tsubst_decl) [PARM_DECL]: Check cp_unevaluated_operand in
handling of TREE_CHAIN for empty pack.
Jonathan Wakely [Wed, 26 Feb 2020 16:31:19 +0000 (16:31 +0000)]
libstdc++: Fix undefined behaviour in random dist serialization (PR93205)
The deserialization functions for random number distributions fail to
check the stream state before using the extracted values. In some cases
this leads to using indeterminate values to resize a vector, and then
filling that vector with indeterminate values.
No values that affect control flow should be used without checking that a
good value was read from the stream.
Additionally, where reasonable to do so, defer modifying any state in
the distribution until all values have been successfully read, to avoid
modifying some of the distribution's parameters and leaving others
unchanged.
Backport from mainline
2020-01-09 Jonathan Wakely <jwakely@redhat.com>
PR libstdc++/93205
* include/bits/random.h (operator>>): Check stream operation succeeds.
* include/bits/random.tcc: (operator>>): Likewise.
(__extract_params): New function to fill a vector from a stream.
* testsuite/26_numerics/random/pr60037-neg.cc: Adjust dg-error line.
Matheus Castanho [Thu, 13 Feb 2020 23:43:39 +0000 (23:43 +0000)]
rs6000: fixinc: Skip machine_name fix for powerpc*-*-linux*
Some system headers can be broken by the machine_name fix performed
by GCC during the fixincludes step. According to the comment in
fixincludes/fixinc.h:130 :
On some platforms, machine_name doesn't work properly and
breaks some of the header files. Since everything works
properly without it, just wipe the macro list to
disable the fix.
So we can just skip it to avoid trouble.
Backport from trunk
2020-02-13 Matheus Castanho <msc@linux.ibm.com>
fixincludes/
* fixinc.in: Skip machine_name fix on powerpc*-*-linux*.
Jonathan Wakely [Wed, 26 Feb 2020 15:32:34 +0000 (15:32 +0000)]
libstdc++: Replace glibc-specific check for clock_gettime (PR 93325)
It's wrong to assume that clock_gettime is unavailable on any *-*-linux*
target that doesn't have glibc 2.17 or later. Use a generic test instead
of using __GLIBC_PREREQ. Only do that test when is_hosted=yes so that we
don't get an error for cross targets without a working linker.
This ensures that C library's clock_gettime will be used on non-glibc
targets, instead of an incorrect syscall to SYS_clock_gettime.
Backport from mainline
2020-01-28 Jonathan Wakely <jwakely@redhat.com>
PR libstdc++/93325
* acinclude.m4 (GLIBCXX_ENABLE_LIBSTDCXX_TIME): Use AC_SEARCH_LIBS for
clock_gettime instead of explicit glibc version check.
* configure: Regenerate.
Jonathan Wakely [Wed, 26 Feb 2020 15:04:53 +0000 (15:04 +0000)]
libstdc++: Fix regressions in unique_ptr::swap (PR 93562)
The requirements for this function are only that the deleter is
swappable, but we incorrectly require that the element type is complete
and that the deleter can be swapped using std::swap (which requires it
to be move cosntructible and move assignable).
The fix is to add __uniq_ptr_impl::swap which swaps the pointer and
deleter individually, instead of using the generic std::swap on the
tuple containing them.
PR libstdc++/93562
* include/bits/unique_ptr.h (__uniq_ptr_impl::swap): Define.
(unique_ptr::swap, unique_ptr<T[], D>::swap): Call it.
* testsuite/20_util/unique_ptr/modifiers/93562.cc: New test.
Jonathan Wakely [Wed, 26 Feb 2020 14:20:55 +0000 (14:20 +0000)]
libstdc++: Fix freestanding build (PR 92376)
In a freestanding library we don't install the <pstl/pstl_config.h>
header, so don't try to include it unless it exists.
Explicitly declare aligned alloc functions for freestanding, because
<cstdlib> doesn't declare them.
Backport from mainline
2020-01-17 Jonathan Wakely <jwakely@redhat.com>
PR libstdc++/92376
* include/bits/c++config: Only do PSTL config when the header is
present, to fix freestanding.
* libsupc++/new_opa.cc [!_GLIBCXX_HOSTED]: Declare allocation
functions if they were detected by configure.
Jonathan Wakely [Wed, 26 Feb 2020 14:00:07 +0000 (14:00 +0000)]
PR libstdc++/78552 only construct std::locale for C locale once
Backport from mainline
2019-10-09 Jonathan Wakely <jwakely@redhat.com>
PR libstdc++/78552
* src/c++98/locale_init.cc (locale::classic()): Do not construct a new
locale object for every call.
(locale::_S_initialize_once()): Construct C locale here.
Jiufu Guo [Mon, 17 Feb 2020 02:48:39 +0000 (10:48 +0800)]
rs6000: mark clobber for registers changed by untpyed_call
As PR93047 said, __builtin_apply/__builtin_return does not work well with
-frename-registers. This is caused by return register(e.g. r3) is used to
rename another register, before return register is stored to stack.
This patch fix this issue by emitting clobber for those egisters which
maybe changed by untyped call.
The following testcase is miscompiled in 8+.
The problem is that check_no_overlap has a special case for INTEGER_CST
marked stores (i.e. stores of constants), if both all currenly merged stores
and the one under consideration for merging with them are marked that way,
it anticipates that other INTEGER_CST marked stores that overlap with those
and precede those (have smaller info->order) could be merged with those and
doesn't punt for them.
In PR86844 and PR87859 fixes I've then added quite large code that is
performed after check_no_overlap and tries to find out if we need and can
merge further INTEGER_CST marked stores, or need to punt.
Unfortunately, that code is there only in the overlapping case code and
the testcase below shows that we really need it even in the adjacent store
case. After sort_by_bitpos we have:
bitpos width order rhs_code
96 32 3 INTEGER_CST
128 32 1 INTEGER_CST
128 128 2 INTEGER_CST
192 32 0 MEM_REF
Because of the missing PR86844/PR87859-ish code in the adjacent store
case, we merge the adjacent (memory wise) stores 96/32/3 and 128/32/1,
and then we consider the 128-bit store which is in program-order in between
them, but in this case we punt, because the merging would extend the
merged store region from bitpos 96 and 64-bits to bitpos 96 and 160-bits
and that has an overlap with an incompatible store (the MEM_REF one).
The problem is that we can't really punt this way, because the 128-bit
store is in between those two we've merged already, so either we manage
to merge even that one together with the others, or would need to avoid
already merging the 96/32/3 and 128/32/1 stores together.
Now, rather than copying around the PR86844/PR87859 code to the other spot,
we can actually just use the overlapping code, merge_overlapping is really
a superset of merge_into, so that is what the patch does. If doing
adjacent store merge for rhs_code other than INTEGER_CST, I believe the
current code is already fine, check_no_overlap in that case doesn't make
the exception and will punt if there is some earlier (smaller order)
non-mergeable overlapping store. There is just one case that could be
problematic, if the merged_store has BIT_INSERT_EXPRs in them and the
new store is a constant store (INTEGER_CST rhs_code), then check_no_overlap
would do the exception and still would allow the special case. But we
really shouldn't have the special case in that case, so this patch also
changes check_no_overlap to just have a bool whether we should have the
special case or not.
Note, as I said in the PR, for GCC11 we could consider performing some kind
of cheap DSE during the store merging (perhaps guarded with flag_tree_dse).
And another thing to consider is only consider as problematic non-mergeable
stores that not only have order smaller than last_order as currently, but
also have order larger than first_order, as in this testcase if we actually
ignored (not merged with anything at all) the 192/32/0 store, because it is
not in between the other stores we'd merge, it would be fine to merge the
other 3 stores, though of course the testcase can be easily adjusted by
putting the 192/32 store after the 128/32 store and then this patch would be
still needed. Though, I think I'd need more time thinking this over.
2020-02-26 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/93820
* gimple-ssa-store-merging.c (check_no_overlap): Change RHS_CODE
argument to ALL_INTEGER_CST_P boolean.
(imm_store_chain_info::try_coalesce_bswap): Adjust caller.
(imm_store_chain_info::coalesce_immediate_stores): Likewise. Handle
adjacent INTEGER_CST store into merged_store->only_constants like
overlapping one.
Christophe Lyon [Wed, 19 Feb 2020 20:55:23 +0000 (20:55 +0000)]
ARM: Fix -mpure-code for v6m
When running the testsuite with -fdisable-rtl-fwprop2 and -mpure-code
for cortex-m0, I noticed that some testcases were failing because we
still generate "ldr rX, .LCY", which is what we want to avoid with
-mpure-code. This is latent since a recent improvement in fwprop
(PR88833).
In this patch I change the thumb1_movsi_insn pattern so that it emits
the desired instruction sequence when arm_disable_literal_pool is set.
To achieve that, I introduce a new required_for_purecode attribute to
enable the corresponding alternative in thumb1_movsi_insn and take the
actual instruction sequence length into account.
Backport from mainline
2020-02-25 Christophe Lyon <christophe.lyon@linaro.org>
* config/arm/arm.md (required_for_purecode): New attribute.
(enabled): Handle required_for_purecode.
* config/arm/thumb1.md (thumb1_movsi_insn): Add alternative to
work with -mpure-code.
Christophe Lyon [Tue, 17 Dec 2019 15:43:07 +0000 (15:43 +0000)]
ARM: Add support for -mpure-code in thumb-1 (v6m)
This patch extends support for -mpure-code to all thumb-1 processors,
by removing the need for MOVT.
Symbol addresses are built using upper8_15, upper0_7, lower8_15 and
lower0_7 relocations, and constants are built using sequences of
movs/adds and lsls instructions.
The extension of the *thumb1_movhf pattern uses always the same size
(6) although it can emit a shorter sequence when possible. This is
similar to what *arm32_movhf already does.
CASE_VECTOR_PC_RELATIVE is now false with -mpure-code, to avoid
generating invalid assembly code with differences from symbols from
two different sections (the difference cannot be computed by the
assembler).
Tests pr45701-[12].c needed a small adjustment to avoid matching
upper8_15 when looking for the r8 register.
Test no-literal-pool.c is augmented with __fp16, so it now uses
-mfp16-format=ieee.
Test thumb1-Os-mult.c generates an inline code sequence with
-mpure-code and computes the multiplication by using a sequence of
add/shift rather than using the multiply instruction, so we skip it in
presence of -mpure-code.
With -mcpu=cortex-m0, the pure-code/no-literal-pool.c fails because
code like:
static char *p = "Hello World";
char *
testchar ()
{
return p + 4;
}
By contrast, when using -mcpu=cortex-m4, the code looks like:
.section .rodata
.LC0:
.ascii "Hello World\000"
.data
p:
.word .LC0
testchar:
push {r7}
add r7, sp, #0
movw r3, #:lower16:p
movt r3, #:upper16:p
ldr r3, [r3]
adds r3, r3, #4
mov r0, r3
mov sp, r7
pop {r7}
bx lr
I haven't found yet how to make code for cortex-m0 apply upper/lower
relocations to "p" instead of .LC2. The current code looks functional,
but could be improved.
Jakub Jelinek [Tue, 25 Feb 2020 12:56:47 +0000 (13:56 +0100)]
combine: Fix find_split_point handling of constant store into ZERO_EXTRACT [PR93908]
git is miscompiled on s390x-linux with -O2 -march=zEC12 -mtune=z13.
I've managed to reduce it into the following testcase. The problem is that
during combine we see the s->k = -1; bitfield store and change the SET_SRC
from a pseudo into a constant:
(set (zero_extract:DI (mem/j:HI (plus:DI (reg/v/f:DI 60 [ s ])
(const_int 10 [0xa])) [0 +0 S2 A16])
(const_int 2 [0x2])
(const_int 7 [0x7]))
(const_int -1 [0xffffffffffffffff]))
This on s390x with the above option isn't recognized as valid instruction,
so find_split_point decides to handle it as IOR or IOR/AND.
src is -1, mask is 3 and pos is 7.
src != mask (this is also incorrect, we want to set all (both) bits in the
bitfield), so we go for IOR/AND, but instead of trying
mem = (mem & ~0x180) | ((-1 << 7) & 0x180)
we actually try
mem = (mem & ~0x180) | (-1 << 7)
and that is further simplified into:
mem = mem | (-1 << 7)
aka
mem = mem | 0xff80
which doesn't set just the 2-bit bitfield, but also many other bitfields
that shouldn't be touched.
We really should do:
mem = mem | 0x180
instead.
The problem is that we assume that no bits but those low len (2 here) will
be set in the SET_SRC, but there is nothing that can prevent that, we just
should ignore the other bits.
The following patch fixes it by masking src with mask, this way already
the src == mask test will DTRT, and as the code for or_mask uses
gen_int_mode, if the most significant bit is set after shifting it left by
pos, it will be properly sign-extended.
2020-02-25 Jakub Jelinek <jakub@redhat.com>
PR rtl-optimization/93908
* combine.c (find_split_point): For store into ZERO_EXTRACT, and src
with mask.
Eric Botcazou [Tue, 25 Feb 2020 11:34:00 +0000 (12:34 +0100)]
Fix link failure with debug info in LTO mode
This fixes a regression whereby the program fails to link with debug
info in LTO mode because of an undefined reference to a symbol coming
from the object files containing the early debug info.
* dwarf2out.c (dwarf2out_size_function): Run in early-DWARF mode.
Roman Zhuykov [Tue, 25 Feb 2020 11:32:42 +0000 (14:32 +0300)]
doc: backport proper description of --enable-checking behavior
This patch rewords the whole description to fix minor issues:
- documents 'gimple' and 'types' checks,
- clarifies what happens when option is used without '=list',
- fixes inaccurate wrong wording about release snapshots,
- describes that release checks can only be disabled explicitly.
Backport from master
2020-02-24 Roman Zhuykov <zhroma@ispras.ru>
* doc/install.texi (--enable-checking): Properly document current
behavior.
(--enable-stage1-checking): Minor clarification about bootstrap.
vect: Fix offset calculation for -ve strides [PR93767]
This PR is a regression caused by r256644, which added support for alias
checks involving variable strides. One of the changes in that commit
was to split the access size out of the segment length. The PR shows
that I hadn't done that correctly for the handling of negative strides
in vect_compile_time_alias. The old code was:
where vect_get_scalar_dr_size (a) was cancelling out the subtraction
of the access size inherent in "- const_length_a". Taking the access
size out of the segment length meant that the addition was no longer
needed/correct.
2020-02-24 Richard Sandiford <richard.sandiford@arm.com>
gcc/
Backport from mainline
2020-02-19 Richard Sandiford <richard.sandiford@arm.com>
PR tree-optimization/93767
* tree-vect-data-refs.c (vect_compile_time_alias): Remove the
access-size bias from the offset calculations for negative strides.
gcc/testsuite/
Backport from mainline
2020-02-19 Richard Sandiford <richard.sandiford@arm.com>
PR tree-optimization/93767
* gcc.dg/vect/pr93767.c: New test.
Mark Eggleston [Mon, 24 Feb 2020 15:53:24 +0000 (15:53 +0000)]
fortran: ICE using SHAPE with FINDLOC PR93835
Backported from mainline
2020-02-24 Mark Eggleston <markeggleston@gcc.gnu.org>
PR fortran/93835
* simplify.c (simplify_findloc_nodim) : Fix whitespace issues.
(gfc_simplify_shape) : Create and initialise one shape value
for the result expression. Set shape value with the rank of
the source array.
PR fortran/93835
* gfortran.dg/pr77351.f90 : Check for one error instead of two.
* gfortran.dg/pr93835.f08 : New test.
* collect2.c (tool_cleanup): Avoid calling not signal-safe
functions.
(maybe_run_lto_and_relink): Avoid possible signal handler
access to unintialzed memory (lto_o_files).
Peter Bergner [Mon, 24 Feb 2020 00:22:57 +0000 (18:22 -0600)]
rs6000: Fix infinite loop building ghostscript and icu [PR93658]
Fix rs6000_legitimate_address_p(), which erroneously marks a valid Altivec
address as being invalid, which causes LRA's process_address() to go into
an infinite loop spilling the same address over and over again.
Include Mike's earlier commits that fix bugs this patch exposes.
Backport from master
2020-02-20 Peter Bergner <bergner@linux.ibm.com>
Michael Meissner [Mon, 24 Feb 2020 00:17:12 +0000 (18:17 -0600)]
Adjust how variable vector extraction is done.
Backport from master
2020-02-03 Michael Meissner <meissner@linux.ibm.com>
* config/rs6000/rs6000.c (get_vector_offset): New helper function
to calculate the offset in memory from the start of a vector of a
particular element. Add code to keep the element number in
bounds if the element number is variable.
(rs6000_adjust_vec_address): Move calculation of offset of the
vector element to get_vector_offset.
(rs6000_split_vec_extract_var): Do not do the initial AND of
element here, move the code to get_vector_offset.
Fix PR 93568 (thinko)
Backport from master
2020-02-05 Michael Meissner <meissner@linux.ibm.com>
PR target/93568
* config/rs6000/rs6000.c (get_vector_offset): Fix Q constraint assert
to use MEM.
Michael Meissner [Mon, 24 Feb 2020 00:13:00 +0000 (18:13 -0600)]
Fix bad code of vector extract of PC-relative address with variable element #.
Backport from master
2020-01-06 Michael Meissner <meissner@linux.ibm.com>
* config/rs6000/vsx.md (vsx_extract_<mode>_var, VSX_D iterator):
Use 'Q' for doing vector extract from memory.
(vsx_extract_v4sf_var): Use 'Q' for doing vector extract from
memory.
(vsx_extract_<mode>_var, VSX_EXTRACT_I iterator): Use 'Q' for
doing vector extract from memory.
(vsx_extract_<mode>_<VS_scalar>mode_var): Use 'Q' for doing vector
extract from memory.
Thomas König [Thu, 13 Feb 2020 21:22:04 +0000 (22:22 +0100)]
Use au->lock exclusively for locking in async I/O.
2020-02-18 Thomas Koenig <tkoenig@gcc.gnu.org>
Backport from trunk
PR fortran/93599
* io/async.c (destroy_adv_cond): Do not destroy lock.
(async_io): Make sure au->lock is locked for finishing of thread.
Do not lock/unlock around signalling emptysignal. Unlock au->lock
before return.
(init_adv_cond): Do not initialize lock.
(enqueue_transfer): Unlock after signal.
(enqueue_done_id): Likewise.
(enqueue_done): Likewise.
(enqueue_close): Likewise.
(enqueue_data_transfer): Likewise.
(async_wait_id): Do not lock/unlock around signalling au->work.
(async_wait): Unlock after signal.
* io/async.h (SIGNAL): Add comment about needed au->lock.
Remove locking/unlocking of advcond->lock.
(WAIT_SIGNAL_MUTEX): Add comment. Remove locking/unlocking of
advcond->lock. Unlock mutex only at the end. Loop on
__ghread_cond_wait returning zero.
(REVOKE_SIGNAL): Add comment. Remove locking/unlocking of
advcond->lock.
(struct adv_cond): Remove mutex from struct.
Fix handling of floating-point homogeneous aggregates.
2020-02-21 John David Anglin <danglin@gcc.gnu.org>
* gcc/config/pa/pa.c (pa_function_value): Fix check for word and
double-word size when handling aggregate return values.
* gcc/config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Fix to indicate
that homogeneous SFmode and DFmode aggregates are passed and returned
in general registers.
Uros Bizjak [Thu, 20 Feb 2020 20:58:57 +0000 (21:58 +0100)]
i386: Fix *vec_extractv2sf_1 and *vec_extractv2sf_1 shufps alternative [PR93828]
shufps moves two of the four packed single-precision floating-point values
from *destination* operand (first operand) into the low quadword of the
destination operand. Match source operand to the destination.
PR target/93828
* config/i386/mmx.md (*vec_extractv2sf_1): Match source operand
to destination operand for shufps alternative.
(*vec_extractv2si_1): Ditto.
Mark Eggleston [Wed, 19 Feb 2020 09:36:42 +0000 (09:36 +0000)]
[Fortran] ICE assign character pointer to non target PR93714
An ICE occurred if an attempt was made to assign a pointer to a
character variable that has an length incorrectly specified using
a real constant and does not have the target attribute.
Backported from mainline
2020-02-18 Mark Eggleston <markeggleston@gcc.gnu.org>
PR fortran/93714
* expr.c (gfc_check_pointer_assign): Move check for
matching character length to after checking the lvalue
attributes for target or pointer.
PR fortran/93714
* gfortran.dg/char_pointer_assign_6.f90: Look for no target
message instead of length mismatch.
* gfortran.dg/pr93714_1.f90
* gfortran.dg/pr93714_2.f90
Check for bitwise identity when encoding VECTOR_CSTs [PR92768]
This PR shows that we weren't checking for bitwise-identical values
when trying to encode a VECTOR_CST, so -0.0 was treated the same as
0.0 for -fno-signed-zeros. The patch adds a new OEP flag to select
that behaviour.
2020-02-18 Richard Sandiford <richard.sandiford@arm.com>
gcc/
Backport from mainline
2019-12-05 Richard Sandiford <richard.sandiford@arm.com>
PR middle-end/92768
* tree-core.h (OEP_BITWISE): New flag.
* fold-const.c (operand_compare::operand_equal_p): Handle it.
* tree-vector-builder.h (tree_vector_builder::equal_p): Pass it.
gcc/testsuite/
PR middle-end/92768
* gcc.dg/pr92768.c: New test.
Mark Eggleston [Tue, 18 Feb 2020 10:56:38 +0000 (10:56 +0000)]
[fortran] ICE in gfc_validate_kind(): Got bad kind [PR93580]
Caused by using invalid part_refs in kind specifications,
e.g. %re or %im on non-complex expressions and %len on
non character expressions.
Check whether %re, %im and %len are valid when checking
kind specification.
The original patch from Steven G. Kargl <kargl@gcc.gnu.org> only
checked for %re and %im.
Backported from mainline
2020-02-18 Mark Eggleston <markeggleston@gcc.gnu.org>
PR fortran/93580
* primary.c (gfc_match_varspec): If the symbol following %
is re or im and the primary expression type is not BT_COMPLEX
issue an error. If the symbol is len and the primary
expression type is not BT_CHARACTER is an error.
PR fortran/93580
* gfortran.dg/dg/pr93580.f90: New test.
predcom has the following code to stop one rogue load from
interfering with other store-load opportunities:
/* If A is read and B write or vice versa and there is unsuitable
dependence, instead of merging both components into a component
that will certainly not pass suitable_component_p, just put the
read into bad component, perhaps at least the write together with
all the other data refs in it's component will be optimizable. */
But when store-store commoning was added later, this had the effect
of ignoring loads that occur between two candidate stores.
There is code further up to handle loads and stores with unknown
dependences:
/* Don't do store elimination if there is any unknown dependence for
any store data reference. */
if ((DR_IS_WRITE (dra) || DR_IS_WRITE (drb))
&& (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know
|| DDR_NUM_DIST_VECTS (ddr) == 0))
eliminate_store_p = false;
But the store-load code above skips loads for *known* dependences
if (a) the load has already been marked "bad" or (b) the data-ref
machinery knows the dependence distance, but determine_offsets
can't handle the combination.
(a) happens to be the problem in the testcase, but a different
sequence could have given (b) instead. We have writes to individual
fields of a structure and reads from the whole structure. Since
determine_offsets requires the types to be the same, it returns false
for each such read/write combination.
This patch records which components have had loads removed and
prevents store-store commoning for them. It's a bit too pessimistic,
since there shouldn't be a problem if a "bad" load dominates all stores
in a component. But (a) we can't AFAIK use pcom_stmt_dominates_stmt_p
here and (b) the handling for that case would probably need to be
removed again if we handled more exotic cases in future.
2020-02-18 Richard Sandiford <richard.sandiford@arm.com>
gcc/
Backport from mainline
2020-01-28 Richard Sandiford <richard.sandiford@arm.com>
PR tree-optimization/93434
* tree-predcom.c (split_data_refs_to_components): Record which
components have had aliasing loads removed. Prevent store-store
commoning for all such components.
gcc/testsuite/
PR tree-optimization/93434
* gcc.c-torture/execute/pr93434.c: New test.
Don't pass booleans as mask types to simd clones [PR92710]
In this PR we assigned a vector mask type to the result of a comparison
and then tried to pass that mask type to a simd clone, which expected
a normal (non-mask) type instead.
This patch simply punts on call arguments that have a mask type.
A better fix would be to pattern-match the comparison to a COND_EXPR,
like we would if the comparison was stored to memory, but doing that
isn't gcc 9 or 10 material.
Note that this doesn't affect x86_64-linux-gnu because the ABI promotes
bool arguments to ints.
2020-02-18 Richard Sandiford <richard.sandiford@arm.com>
gcc/
Backport from mainline
2019-11-29 Richard Sandiford <richard.sandiford@arm.com>
Fix SLP downward group access classification [PR92420]
This PR was caused by the SLP handling in get_group_load_store_type
returning VMAT_CONTIGUOUS rather than VMAT_CONTIGUOUS_REVERSE for
downward groups.
A more elaborate fix would be to try to combine the reverse permutation
into SLP_TREE_LOAD_PERMUTATION for loads, but that's really a follow-on
optimisation and not backport material. It might also not necessarily
be a win, if the target supports (say) reversing and odd/even swaps
as independent permutes but doesn't recognise the combined form.
2020-02-18 Richard Sandiford <richard.sandiford@arm.com>
gcc/
Backport from mainline
2019-11-11 Richard Sandiford <richard.sandiford@arm.com>
PR tree-optimization/92420
* tree-vect-stmts.c (get_negative_load_store_type): Move further
up file.
(get_group_load_store_type): Use it for reversed SLP accesses.
gcc/testsuite/
PR tree-optimization/92420
* gcc.dg/vect/pr92420.c: New test.
Jakub Jelinek [Sat, 15 Feb 2020 11:53:44 +0000 (12:53 +0100)]
match.pd: Disallow side-effects in GENERIC for non-COND_EXPR to COND_EXPR simplifications [PR93744]
As the following testcases show (the first one reported, last two
found by code inspection), we need to disallow side-effects
in simplifications that turn some unconditional expression into conditional
one. From my little understanding of genmatch.c, it is able to
automatically disallow side effects if the same operand is used multiple
times in the match pattern, maybe if it is used multiple times in the
replacement pattern, and if it is used in conditional contexts in the match
pattern, could it be taught to handle this case too? If yes, perhaps
just the first hunk could be usable for 8/9 backports (+ the testcases).
2020-02-15 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/93744
* match.pd (((m1 >/</>=/<= m2) * d -> (m1 >/</>=/<= m2) ? d : 0,
A - ((A - B) & -(C cmp D)) -> (C cmp D) ? B : A,
A + ((B - A) & -(C cmp D)) -> (C cmp D) ? B : A): For GENERIC, make
sure @2 in the first and @1 in the other patterns has no side-effects.
* gcc.c-torture/execute/pr93744-1.c: New test.
* gcc.c-torture/execute/pr93744-2.c: New test.
* gcc.c-torture/execute/pr93744-3.c: New test.
Eric Botcazou [Fri, 14 Feb 2020 18:21:02 +0000 (19:21 +0100)]
Fix problematic TLS sequences for the Solaris linker
This is an old thinko pertaining to the interaction between TLS
sequences and delay slot filling: the compiler knows that it cannot
put instructions with TLS relocations into delay slots with the
original Sun TLS model, but it tests TARGET_SUN_TLS in this context,
which depends only on the assembler. So if the compiler is configured
with the GNU assembler and the Solaris linker, then TARGET_GNU_TLS is
set instead and the limitation is not enforced.
PR target/93704
* config/sparc/sparc.c (eligible_for_call_delay): Test HAVE_GNU_LD
in conjunction with TARGET_GNU_TLS in early return.
Jakub Jelinek [Fri, 14 Feb 2020 16:36:00 +0000 (17:36 +0100)]
c++: Fix thinko in enum_min_precision [PR61414]
When backporting the PR61414 fix to 8.4, I've noticed that the caching
of prec is actually broken, as it would fail to actually store the computed
precision into the hash_map's value and so next time we'd think the enum needs
0 bits.
2020-02-14 Jakub Jelinek <jakub@redhat.com>
PR c++/61414
* class.c (enum_min_precision): Change prec type from int to int &.
Richard Biener [Wed, 5 Feb 2020 13:04:29 +0000 (14:04 +0100)]
middle-end/90648 fend off builtin calls with not enough arguments from match
This adds guards to genmatch generated code before accessing call
expression or stmt arguments that might be out of bounds when
the user provided bogus prototypes for what we consider builtins.
2020-02-05 Richard Biener <rguenther@suse.de>
PR middle-end/90648
* genmatch.c (dt_node::gen_kids_1): Emit number of argument
checks before matching calls.
Richard Biener [Wed, 22 Jan 2020 11:38:12 +0000 (12:38 +0100)]
tree-optimization/93381 fix integer offsetting in points-to analysis
We were incorrectly assuming a merge operation is conservative enough
for not explicitely handled operations but we also need to consider
offsetting within fields when field-sensitive analysis applies.
2020-01-22 Richard Biener <rguenther@suse.de>
PR tree-optimization/93381
* tree-ssa-structalias.c (find_func_aliases): Assume offsetting
throughout, handle all conversions the same.
Richard Biener [Fri, 14 Feb 2020 08:23:06 +0000 (09:23 +0100)]
tree-optimization/93439 move clique bookkeeping to OMP expansion
Autopar was doing clique bookkeeping too early when creating destination
functions but then later introducing new cliques via versioning loops.
The following moves the bookkeeping to the actual outlining process.
2020-02-14 Richard Biener <rguenther@suse.de>
Backport from mainline
2020-01-28 Richard Biener <rguenther@suse.de>
PR tree-optimization/93439
* tree-parloops.c (create_loop_fn): Move clique bookkeeping...
* tree-cfg.c (move_sese_region_to_fn): ... here.
(verify_types_in_gimple_reference): Verify used cliques are
tracked.
Richard Biener [Fri, 14 Feb 2020 08:17:57 +0000 (09:17 +0100)]
debug/92763 keep DIEs that might be used in DW_TAG_inlined_subroutine
We were pruning type-local subroutine DIEs if their context is unused
despite us later needing those DIEs as abstract origins for inlines.
The patch makes code already present for -fvar-tracking-assignments
unconditional.
2020-02-14 Richard Biener <rguenther@suse.de>
Backport from mainline
2020-01-20 Richard Biener <rguenther@suse.de>
PR debug/92763
* dwarf2out.c (prune_unused_types): Unconditionally mark
called function DIEs.
Richard Biener [Fri, 14 Feb 2020 08:10:48 +0000 (09:10 +0100)]
middle-end/92674 delay purging EH edges when folding during inlining
2020-02-14 Richard Biener <rguenther@suse.de>
Backport from mainline
2019-11-27 Richard Biener <rguenther@suse.de>
PR middle-end/92674
* tree-inline.c (expand_call_inline): Delay purging EH/abnormal
edges and instead record blocks in bitmap.
(gimple_expand_calls_inline): Adjust.
(fold_marked_statements): Delay EH cleanup until all folding is
done.
(optimize_inline_calls): Do EH/abnormal cleanup for calls after
inlining finished.
Jakub Jelinek [Thu, 13 Feb 2020 20:00:09 +0000 (21:00 +0100)]
c: Fix ICE with cast to VLA [93576]
The following testcase ICEs, because the PR84305 changes try to evaluate
the size earlier. If size has side-effects, that is desirable, and the
side-effects will actually be wrapped in a SAVE_EXPR. The problem on this
testcase is that there are no side-effects, and c_fully_fold doesn't fold
those COMPOUND_EXPRs to constant, and while before gimplification we unshare
trees found in the expressions, the unsharing doesn't involve TYPE_SIZE etc.
of used types. Gimplification is destructive though, so when we gimplify
the two nested COMPOUND_EXPRs and then try to gimplify it the second time
for the TYPE_SIZEs, we ICE.
Now, we could use unshare_expr in what we push to *expr, SAVE_EXPRs and
their operands in there aren't unshared, but I really don't see a point of
evaluating expressions that don't have side-effects before, so instead
this just pushes there expressions that do have side-effects.
2020-02-13 Jakub Jelinek <jakub@redhat.com>
PR c/93576
* c-decl.c (grokdeclarator): If this_size_varies, only push size into
*expr if it has side effects.
Jakub Jelinek [Thu, 13 Feb 2020 09:43:27 +0000 (10:43 +0100)]
i386: Fix up _mm*_mask_popcnt_epi* [PR93696]
As mentioned in the PR and as
https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mask_popcnt_epi
also documents, _mm*_popcnt_epi* intrinsics are consistent with all other
unary AVX512* intrinsics regarding arguments, i.e. the
_mm*_whatever has just single argument (called a in the docs, and __A in the
GCC headers),
_mm*_mask_whatever has 3 arguments (called src, k, a in the docs and
_W, __U, __A in GCC headers) and
_mm*_maskz_whatever 2 arguments (called k, a in the docs and __U, __A in GCC
headers). Unfortunately, whomever implemented the _mm*_popcnt_epi*
intrinsics got it wrong for the _mm*_mask_popcnt_epi* ones, calling the
args __A, __U, __B and not passing them in the canonical order to the
builtins, making it API incompatible with ICC as well as clang (tested on
godbolts clang 7/8/9/trunk and ICC 19.0.{0,1}, older clang/ICC don't
understand those, so it isn't that it used to be broken even in other
compilers and got changed afterwards).
2020-02-13 Jakub Jelinek <jakub@redhat.com>
PR target/93696
* config/i386/avx512bitalgintrin.h (_mm512_mask_popcnt_epi8,
_mm512_mask_popcnt_epi16, _mm256_mask_popcnt_epi8,
_mm256_mask_popcnt_epi16, _mm_mask_popcnt_epi8,
_mm_mask_popcnt_epi16): Rename __B argument to __A and __A to __W,
pass __A to the builtin followed by __W instead of __A followed by
__B.
* config/i386/avx512vpopcntdqintrin.h (_mm512_mask_popcnt_epi32,
_mm512_mask_popcnt_epi64): Likewise.
* config/i386/avx512vpopcntdqvlintrin.h (_mm_mask_popcnt_epi32,
_mm256_mask_popcnt_epi32, _mm_mask_popcnt_epi64,
_mm256_mask_popcnt_epi64): Likewise.
Jakub Jelinek [Thu, 13 Feb 2020 07:17:07 +0000 (08:17 +0100)]
i386: Fix k*shift* intrinsics [PR93673]
As mentioned in the PR, the intrinsics allow counts from 0 to 255, but
we actually reject values from 128 to 255. That is because QImode
CONST_INTs can be only -128 to 127. Fixed by using const_0_to_255_operand
and dropping the modes for the operands with those predicates
(the IL actually contains the CONST_INT which has VOIDmode).
2020-02-13 Jakub Jelinek <jakub@redhat.com>
PR target/93673
* config/i386/sse.md (k<code><mode>): Drop mode from last operand and
use const_0_to_255_operand predicate instead of immediate_operand.
(avx512dq_fpclass<mode><mask_scalar_merge_name>,
avx512dq_vmfpclass<mode><mask_scalar_merge_name>,
vgf2p8affineinvqb_<mode><mask_name>,
vgf2p8affineqb_<mode><mask_name>): Drop mode from
const_0_to_255_operand predicated operands.
* gcc.target/i386/avx512f-pr93673.c: New test.
* gcc.target/i386/avx512dq-pr93673.c: New test.
* gcc.target/i386/avx512bw-pr93673.c: New test.
Jakub Jelinek [Wed, 12 Feb 2020 10:58:35 +0000 (11:58 +0100)]
i386: Fix up vec_extract_lo* patterns [PR93670]
The VEXTRACT* insns have way too many different CPUID feature flags (ATT
syntax)
vextractf128 $imm, %ymm, %xmm/mem AVX
vextracti128 $imm, %ymm, %xmm/mem AVX2
vextract{f,i}32x4 $imm, %ymm, %xmm/mem {k}{z} AVX512VL+AVX512F
vextract{f,i}32x4 $imm, %zmm, %xmm/mem {k}{z} AVX512F
vextract{f,i}64x2 $imm, %ymm, %xmm/mem {k}{z} AVX512VL+AVX512DQ
vextract{f,i}64x2 $imm, %zmm, %xmm/mem {k}{z} AVX512DQ
vextract{f,i}32x8 $imm, %zmm, %ymm/mem {k}{z} AVX512DQ
vextract{f,i}64x4 $imm, %zmm, %ymm/mem {k}{z} AVX512F
As the testcase shows and the patch too, we didn't get it right in all
cases.
The first hunk is about avx512vl_vextractf128v8s[if] incorrectly
requiring TARGET_AVX512DQ. The corresponding insn is the first
vextract{f,i}32x4 above, so it requires VL+F, and the builtins have it
correct (TARGET_AVX512VL implies TARGET_AVX512F):
BDESC (OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_avx512vl_vextractf128v8sf, "__builtin_ia32_extractf32x4_256_mask", IX86_BUILTIN_EXTRACTF32X4_256, UNKNOWN, (int) V4SF_FTYPE_V8SF_INT_V4SF_UQI)
BDESC (OPTION_MASK_ISA_AVX512VL, 0, CODE_FOR_avx512vl_vextractf128v8si, "__builtin_ia32_extracti32x4_256_mask", IX86_BUILTIN_EXTRACTI32X4_256, UNKNOWN, (int) V4SI_FTYPE_V8SI_INT_V4SI_UQI)
We only need TARGET_AVX512DQ for avx512vl_vextractf128v4d[if].
The second hunk is about vec_extract_lo_v16s[if]{,_mask}. These are using
the vextract{f,i}32x8 insns (AVX512DQ above), but we weren't requiring that,
but instead incorrectly && 1 for non-masked and && (64 == 64 && TARGET_AVX512VL)
for masked insns. This is extraction from ZMM, so it doesn't need VL for
anything. The hunk actually only requires TARGET_AVX512DQ when the insn
is masked, if it is not masked, when TARGET_AVX512DQ isn't available we can
use vextract{f,i}64x4 instead which is available already in TARGET_AVX512F
and does the same thing, extracts the low 256 bits from 512 bits vector
(often we split it into just nothing, but there are some special cases like
when using xmm16+ when we can't without AVX512VL).
The last hunk is about vec_extract_lo_v8s[if]{,_mask}. The non-_mask
suffixed ones are ok already and just split into nothing (lowpart subreg).
The masked ones were incorrectly requiring TARGET_AVX512VL and
TARGET_AVX512DQ, when we only need TARGET_AVX512VL.
2020-02-12 Jakub Jelinek <jakub@redhat.com>
PR target/93670
* config/i386/sse.md (VI48F_256_DQ): New mode iterator.
(avx512vl_vextractf128<mode>): Use it instead of VI48F_256. Remove
TARGET_AVX512DQ from condition.
(vec_extract_lo_<mode><mask_name>): Use <mask_avx512dq_condition>
instead of <mask_mode512bit_condition> in condition. If
TARGET_AVX512DQ is false, emit vextract*64x4 instead of
vextract*32x8.
(vec_extract_lo_<mode><mask_name>): Drop <mask_avx512dq_condition>
from condition.
Jakub Jelinek [Mon, 10 Feb 2020 21:44:40 +0000 (22:44 +0100)]
i386: Fix -mavx -mno-mavx2 ICE with VEC_COND_EXPR [PR93637]
As mentioned in the PR, for -mavx -mno-avx2 the backend does support
vcondv4div4df and vcondv8siv8sf optabs (while generally 32-byte vectors
aren't much supported in that case, it is performed using
vandps/vandnps/vorps). The problem is that after the last generic vector
lowering (where the VEC_COND_EXPR still compares two V4DF vectors and
has two V4DI last operands and V4DI result and so is considered ok) fre4
folds the condition into constant, at which point the middle-end during
expansion will try vcond_mask_optab and fall back to trying to expand it
as the constant vector < 0 vcondv4div4di, but neither of them is supported
for -mavx -mno-avx2 and thus we ICE.
So, the options I see is either what the following patch does, also support
vcond_mask_v4div4di and vcond_mask_v4siv4si already for TARGET_AVX, or
require for vcondv4div4df and vcondv8siv8sf TARGET_AVX2 rather than current
TARGET_AVX.
2020-02-10 Jakub Jelinek <jakub@redhat.com>
PR target/93637
* config/i386/sse.md (VI_256_AVX2): New mode iterator.
(vcond_mask_<mode><sseintvecmodelower>): Use it instead of VI_256.
Change condition from TARGET_AVX2 to TARGET_AVX.
Jakub Jelinek [Sat, 8 Feb 2020 09:59:40 +0000 (10:59 +0100)]
i386: Make xmm16-xmm31 call used even in ms ABI [PR65782]
On Tue, Feb 04, 2020 at 11:16:06AM +0100, Uros Bizjak wrote:
> I guess that Comment #9 patch form the PR should be trivially correct,
> but althouhg it looks obvious, I don't want to propose the patch since
> I have no means of testing it.
I don't have means of testing it either.
https://docs.microsoft.com/en-us/cpp/build/x64-calling-convention?view=vs-2019
is quite explicit that [xyz]mm16-31 are call clobbered and only xmm6-15 (low
128-bits only) are call preserved.
We are talking e.g. about
/* { dg-options "-O2 -mabi=ms -mavx512vl" } */
typedef double V __attribute__((vector_size (16)));
void foo (void);
V bar (void);
void baz (V);
void
qux (void)
{
V c;
{
register V a __asm ("xmm18");
V b = bar ();
asm ("" : "=x" (a) : "0" (b));
c = a;
}
foo ();
{
register V d __asm ("xmm18");
V e;
d = c;
asm ("" : "=x" (e) : "0" (d));
baz (e);
}
}
where according to the MSDN doc gcc incorrectly holds the c value
in xmm18 register across the foo call; if foo is compiled by some Microsoft
compiler (or LLVM), then it could clobber %xmm18.
If all xmm18 occurrences are changed to say xmm15, then it is valid to hold
the 128-bit value across the foo call (though, surprisingly, LLVM saves it
into stack anyway).
The other parts are I guess mainly about SEH. Consider e.g.
void
foo (void)
{
register double x __asm ("xmm14");
register double y __asm ("xmm18");
asm ("" : "=x" (x));
asm ("" : "=v" (y));
x += y;
y += x;
asm ("" : : "x" (x));
asm ("" : : "v" (y));
}
looking at cross-compiler output, with -O2 -mavx512f this emits
.file "abcdeq.c"
.text
.align 16
.globl foo
.def foo; .scl 2; .type 32; .endef
.seh_proc foo
foo:
subq $40, %rsp
.seh_stackalloc 40
vmovaps %xmm14, (%rsp)
.seh_savexmm %xmm14, 0
vmovaps %xmm18, 16(%rsp)
.seh_savexmm %xmm18, 16
.seh_endprologue
vaddsd %xmm18, %xmm14, %xmm14
vaddsd %xmm18, %xmm14, %xmm18
vmovaps (%rsp), %xmm14
vmovaps 16(%rsp), %xmm18
addq $40, %rsp
ret
.seh_endproc
.ident "GCC: (GNU) 10.0.1 20200207 (experimental)"
Does whatever assembler mingw64 uses even assemble this (I mean the
.seh_savexmm %xmm16, 16 could be problematic)?
I can find e.g.
https://stackoverflow.com/questions/43152633/invalid-register-for-seh-savexmm-in-cygwin/43210527
which then links to
https://gcc.gnu.org/PR65782
2020-02-08 Uroš Bizjak <ubizjak@gmail.com>
Jakub Jelinek <jakub@redhat.com>
PR target/65782
* config/i386/i386.h (CALL_USED_REGISTERS): Make
xmm16-xmm31 call-used even in 64-bit ms-abi.
Jakub Jelinek [Thu, 6 Feb 2020 08:19:08 +0000 (09:19 +0100)]
openmp: Fix handling of non-addressable shared scalars in parallel nested inside of target [PR93515]
As the following testcase shows, we need to consider even target to be a construct
that forces not to use copy in/out for shared on parallel inside of the target.
E.g. for parallel nested inside another parallel or host teams, we already avoid
copy in/out and we need to treat target the same.
2020-02-06 Jakub Jelinek <jakub@redhat.com>
PR libgomp/93515
* omp-low.c (use_pointer_for_field): For nested constructs, also
look for map clauses on target construct.
(scan_omp_1_stmt) <case GIMPLE_OMP_TARGET>: Bump temporarily
taskreg_nesting_level.
* testsuite/libgomp.c-c++-common/pr93515.c: New test.
Jakub Jelinek [Thu, 6 Feb 2020 08:15:13 +0000 (09:15 +0100)]
openmp: Notice reduction decl in outer contexts after adding it to shared [PR93515]
If we call omp_add_variable, following omp_notice_variable will already find it
on that construct and not go through outer constructs, the following patch fixes that.
Note, this still doesn't follow OpenMP 5.0 semantics on target combined with other
constructs with reduction/lastprivate/linear clauses, will handle that for GCC11.
2020-02-06 Jakub Jelinek <jakub@redhat.com>
PR libgomp/93515
* gimplify.c (gimplify_scan_omp_clauses) <do_notice>: If adding
shared clause, call omp_notice_variable on outer context if any.
Jakub Jelinek [Wed, 5 Feb 2020 22:35:08 +0000 (23:35 +0100)]
c++: Mark __builtin_convertvector operand as read [PR93557]
In C++ we weren't calling mark_exp_read on the __builtin_convertvector first
argument. I guess it could misbehave even with lambda implicit captures.
Fixed by calling decay_conversion on the argument, we use the argument as
rvalue so we want the standard lvalue to rvalue conversions, but as the
argument must be a vector type, e.g. integral promotions aren't really
needed.
2020-02-05 Jakub Jelinek <jakub@redhat.com>
PR c++/93557
* semantics.c (cp_build_vec_convert): Call decay_conversion on arg
prior to passing it to c_build_vec_convert.
projects / thirdparty / gcc.git / log
