Adds tunes needed for zen4 microarchitecture. I added two new knobs.
TARGET_AVX512_SPLIT_REGS which is used to specify that internally 512 vectors
are split to 256 vectors. This affects vectorization costs and reassociation
width. It probably should also affect RTX costs however I doubt it is very useful
since RTL optimizers are usually not judging between 256 and 512 vectors.
I also added X86_TUNE_AVOID_256FMA_CHAINS. Since fma has improved in zen4 this
flag may not be a win except for very specific benchmarks. I am still doing some
more detailed testing here.
Oherwise I disabled gathers on zen4 for 2 parts nad 4 parts. We can open code them
and since the latencies has only increased since zen3 opencoding is better than
actual instrucction. This shows at 4 tsvc benchmarks.
I ended up setting AVX256_OPTIMAL. This is a compromise. There are some tsvc
benchmarks that increase noticeably (up to 250%) however there are also few
regressions. Most of these can be solved by incrasing vec_perm cost in the
vectorizer. However this does not cure about 14% regression on x264 that is
quite important. Here we produce vectorized loops for avx512 that probably
would be faster if the loops in question had high enough iteration count.
We hit this problem with avx256 too: since the loop iterates few times, only
prologues/epilogues are used. Adding another round of prologue/epilogue
code does not make it better.
Finally I enabled avx stores for constnat sized memcpy and memset. I am not
sure why this is an opt-in feature. I think for most hardware this is a win.
gcc/ChangeLog:
2022-12-22 Jan Hubicka <hubicka@ucw.cz>
* config/i386/i386-expand.cc (ix86_expand_set_or_cpymem): Add
TARGET_AVX512_SPLIT_REGS
* config/i386/i386-options.cc (ix86_option_override_internal):
Honor x86_TONE_AVOID_256FMA_CHAINS.
* config/i386/i386.cc (ix86_vec_cost): Honor TARGET_AVX512_SPLIT_REGS.
(ix86_reassociation_width): Likewise.
* config/i386/i386.h (TARGET_AVX512_SPLIT_REGS): New tune.
* config/i386/x86-tune.def (X86_TUNE_USE_GATHER_2PARTS): Disable
for znver4.
(X86_TUNE_USE_GATHER_4PARTS): Likewise.
(X86_TUNE_AVOID_256FMA_CHAINS): Set for znver4.
(X86_TUNE_AVOID_512FMA_CHAINS): New utne; set for znver4.
(X86_TUNE_AVX256_OPTIMAL): Add znver4.
(X86_TUNE_AVX512_SPLIT_REGS): New tune.
(X86_TUNE_AVX256_MOVE_BY_PIECES): Add znver1-3.
(X86_TUNE_AVX256_STORE_BY_PIECES): Add znver1-3.
(X86_TUNE_AVX512_MOVE_BY_PIECES): Add znver4.
(X86_TUNE_AVX512_STORE_BY_PIECES): Add znver4.
Update cost of znver4 mostly based on data measued by Agner Fog.
Compared to previous generations x87 became bit slower which is probably not
big deal (and we have minimal benchmarking coverage for it). One interesting
improvement is reducation of FMA cost. I also updated costs of AVX256
loads/stores based on latencies (not throughput which is twice of avx256).
Overall AVX512 vectorization seems to improve noticeably some of TSVC
benchmarks but since internally 512 vectors are split to 256 vectors it is
somewhat risky and does not win in SPEC scores (mostly by regressing benchmarks
with loop that have small trip count like x264 and exchange), so for now I am
going to set AVX256_OPTIMAL tune but I am still playing with it. We improved
since ZNVER1 on choosing vectorization size and also have vectorized
prologues/epilogues so it may be possible to make avx512 small win overall.
2022-12-22 Jan Hubicka <hubicka@ucw.cz>
* config/i386/x86-tune-costs.h (znver4_cost): Upate costs of FP and SSE
moves, division multiplication, gathers, L2 cache size, and more
complex FP instrutions.
This fixes the following on LLP64 mingw-w64 target:
Excess errors:
gcc/testsuite/gcc.c-torture/compile/pr55569.c:13:12: warning: overflow in conversion from 'long long unsigned int' to 'long int' changes value from '4611686018427387903' to '-1' [-Woverflow]
gcc/testsuite/gcc.c-torture/compile/pr55569.c:13:34: warning: iteration 2147483647 invokes undefined behavior [-Waggressive-loop-optimizations]
gcc/testsuite/ChangeLog:
* gcc.c-torture/compile/pr55569.c: fix excess errors.
Signed-off-by: Jonathan Yong <10walls@gmail.com>
Even though this PR was reported with an ubsan issue, the problem is
tree_nonzero_bits is being called with an expression which is a vector type.
This fixes three patterns I noticed which does that.
And adds a testcase for one of the patterns.
OK? Bootstrapped and tested on x86_64-linux-gnu with no regressions
gcc/ChangeLog:
PR tree-optimization/105532
* match.pd (~(X >> Y) -> ~X >> Y): Check if it is an integral
type before calling tree_nonzero_bits.
(popcount(X) + popcount(Y)): Likewise.
(popcount(X&C1)): Likewise.
gcc/testsuite/ChangeLog:
* gcc.c-torture/compile/vector-shift-1.c: New test.
[Sync'ed from the binutils-gdb repo]
This patch uses the toplevel configure parts for GMP/MPFR for
gdb. The only thing is that gdb now requires MPFR for building.
Before it was a recommended but not required library.
Also this allows building of GMP and MPFR with the toplevel
directory just like how it is done for GCC.
We now error out in the toplevel configure of the version
of GMP and MPFR that is wrong.
OK after GDB 13 branches? Build gdb 3 ways:
with GMP and MPFR in the toplevel (static library used at that point for both)
With only MPFR in the toplevel (GMP distro library used and MPFR built from source)
With neither GMP and MPFR in the toplevel (distro libraries used)
Changes from v1:
* Updated gdb/README and gdb/doc/gdb.texinfo.
* Regenerated using unmodified autoconf-2.69
Thanks,
Andrew Pinski
ChangeLog:
* Makefile.def: Add configure-gdb dependencies
on all-gmp and all-mpfr.
* configure.ac: Split out MPC checking from MPFR.
Require GMP and MPFR if the gdb directory exist.
* Makefile.in: Regenerate.
* configure: Regenerate.
Instead of trying to have the GPU do CPU-with-OS-like things, this new barriers
implementation for NVPTX uses simplistic bar.* synchronization instructions.
Tasks are processed after threads have joined, and only if team->task_count != 0
It is noted that: there might be a little bit of performance forfeited for
cases where earlier arriving threads could've been used to process tasks ahead
of other threads, but that has the requirement of implementing complex
futex-wait/wake like behavior, which is what we're try to avoid with this patch.
It is deemed that task processing is not what GPU target offloading is usually
used for.
Implementation highlight notes:
1. gomp_team_barrier_wake() is now an empty function (threads never "wake" in
the usual manner)
2. gomp_team_barrier_cancel() now uses the "exit" PTX instruction.
3. gomp_barrier_wait_last() now is implemented using "bar.arrive"
4. gomp_team_barrier_wait_end()/gomp_team_barrier_wait_cancel_end():
The main synchronization is done using a 'bar.red' instruction. This reduces
across all threads the condition (team->task_count != 0), to enable the task
processing down below if any thread created a task.
(this bar.red usage means that this patch is dependent on the prior NVPTX
bar.red GCC patch)
PR target/99555
libgomp/ChangeLog:
* config/nvptx/bar.c (generation_to_barrier): Remove.
(futex_wait,futex_wake,do_spin,do_wait): Remove.
(GOMP_WAIT_H): Remove.
(#include "../linux/bar.c"): Remove.
(gomp_barrier_wait_end): New function.
(gomp_barrier_wait): Likewise.
(gomp_barrier_wait_last): Likewise.
(gomp_team_barrier_wait_end): Likewise.
(gomp_team_barrier_wait): Likewise.
(gomp_team_barrier_wait_final): Likewise.
(gomp_team_barrier_wait_cancel_end): Likewise.
(gomp_team_barrier_wait_cancel): Likewise.
(gomp_team_barrier_cancel): Likewise.
* config/nvptx/bar.h (gomp_barrier_t): Remove waiters, lock fields.
(gomp_barrier_init): Remove init of waiters, lock fields.
(gomp_team_barrier_wake): Remove prototype, add new static inline
function.
This patch adds support for the PTX 'bar.red' (i.e. "barrier reduction")
instruction, in the form of nvptx-specific __builtin_nvptx_bar_red_[and/or/popc]
built-in functions.
gcc/ChangeLog:
* config/nvptx/nvptx.cc (nvptx_print_operand): Add 'p' case, adjust
comments.
(enum nvptx_builtins): Add NVPTX_BUILTIN_BAR_RED_AND,
NVPTX_BUILTIN_BAR_RED_OR, and NVPTX_BUILTIN_BAR_RED_POPC.
(nvptx_expand_bar_red): New function.
(nvptx_init_builtins):
Add DEFs of __builtin_nvptx_bar_red_[and/or/popc].
(nvptx_expand_builtin): Use nvptx_expand_bar_red to expand
NVPTX_BUILTIN_BAR_RED_[AND/OR/POPC] cases.
* config/nvptx/nvptx.md (define_c_enum "unspecv"): Add
UNSPECV_BARRED_AND, UNSPECV_BARRED_OR, and UNSPECV_BARRED_POPC.
(BARRED): New int iterator.
(barred_op,barred_mode,barred_ptxtype): New int attrs.
(nvptx_barred_<barred_op>): New define_insn.
Add the patch that fixes i686 Darwin build.
Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
libffi/ChangeLog:
* LOCAL_PATCHES: Add patch to fix i686 darwin build.
This addresses a number of issues in the X86 Darwin 32b port for libffi.
1. The pic symbol stubs are weak definitions; the correct section placement
for these depends on the linker version in use. We do not have access
to that information, but we can use the target OS version (assumes that
the user has installed the latest version of xcode available).
When a coalesced section is in use (OS versions earlier than Darwin12 /
OSX 10.8), its name must differ from __TEXT,__text since otherwise that
would correspond to altering the attributes of the .text section (which
produces a diagnostic from the assembler).
Here we use __TEXT, __textcoal_nt for this which is what GCC emits for
these stubs.
For later versions than Darwin 12 (OS X 10.8) we can place the stubs in
the .text section (if we do not we get a diagnostic from clang -cc1as
saying that the use of coalesced sections for this is deprecated).
2. The EH frame is specified manually, since there is no support for .cfi_
directives in 'cctools' assemblers. The implementation needs to provide
offsets for CFA advance, code size and to the CIE as signed values
rather than relocations. However the cctools assembler will produce a
relocation for expressions like ' .long Lxx-Lyy' which then leads to a
link-time error. We correct this by forming the offset values using
' .set' directives and then assigning the results of them.
3. The register numbering used by m32 X86 Darwin EH frames is not the same
as the DWARF debug numbering (the Frame and Stack pointer numbers are
swapped).
4. The FDE address encoding used by the system tools is '0x10' (PCrel + abs)
where the value provided was PCrel + sdata4.
5. GCC does not use compact unwind at present, and it was not implemented
until Darwin10 / OSX 10.6. There were some issues with function location
in 10.6 so that the solution here suppresses emitting the compact unwind
section until Darwin11 / OSX 10.7.
Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
libffi/ChangeLog:
* src/x86/sysv.S (COMDAT): Amend section use for Darwin, accounting
cases where coalesced is needed. (eh_frame): Rework to avoid relocs
that cause builf fails on earlier Darwin. Adjust register numbers
to account for X86 m32 Darwin differences between EH and debug.
The following avoids passing down error_mark_node to fold_convert.
PR middle-end/107994
* gimplify.cc (gimplify_expr): Catch errorneous comparison
operand.
Update documentation of -fwhole-program which was wrongly
claiming that it is useless with LTO whole it is useful for LTO without plugin
and extends -fwhole-program to also work with incremental linking.
This is useful when building kernel where the incremental link is de-facto fina
binary and only some explicitly marked symbols needs to remain.
Bootstrapped/regtested x86_64-linux, comitted.
gcc/ChangeLog:
2022-12-21 Jan Hubicka <hubicka@ucw.cz>
* doc/invoke.texi: Fix documentation of -fwhole-program with LTO
and document behaviour for incremental linking.
gcc/lto/ChangeLog:
2022-12-21 Jan Hubicka <hubicka@ucw.cz>
* lto-common.cc (lto_resolution_read): With incremental linking
and whole program ignore turn LDPR_PREVAILING_DEF_IRONLY to
LDPR_PREVAILING_DEF_IRONLY_EXP
* lto-lang.cc (lto_post_options): Do not clear flag_whole_program
for incremental link
Lto profiledbootstrap was failing for me on {powerpc64le,s390x}-linux with
modula 2 enabled, with:
cc1gm2: internal compiler error: the location value is corrupt
0x11a3d2d m2assert_AssertLocation(unsigned int)
../../gcc/m2/gm2-gcc/m2assert.cc:40
0x11a3d2d m2statement_BuildAssignmentTree
../../gcc/m2/gm2-gcc/m2statement.cc:177
ICE. The problem was that caller (m2assert_AssertLocation used
location_t M2Options_OverrideLocation (location_t);
prototype with the libcpp/line-map.h
typedef unsigned int location_t;
typedef, but the callee defined in Modula 2 was using:
TYPE
location_t = INTEGER ;
and
PROCEDURE OverrideLocation (location: location_t) : location_t ;
Now, on powerpc64le-linux unsigned int is returned and passed zero extended
into 64-bits, while signed int is returned and passed sign-extended into 64-bits
and Modula 2 INTEGER is signed 32-bit type, so when the caller then compared
M2Options_OverrideLocation (location) != location
and powerpc64le-linux performed the comparison as 64-bit compare, there
was a mismatch for location_t of 0x8000007 or others with the MSB set.
Fixed by making Modula 2 location_t a CARDINAL, which is 32-bit unsigned type.
2022-12-21 Jakub Jelinek <jakub@redhat.com>
PR modula2/108153
* gm2-gcc/m2linemap.def (location_t): Use CARDINAL instead of INTEGER.
DECL_OMP_PRIVATIZED_MEMBER vars are artificial vars with DECL_VALUE_EXPR
of this->field used just during gimplification and omp lowering/expansion
to privatize individual fields in methods when needed.
As the following testcase shows, when not in templates, they were handled
right, but in templates we actually called cp_finish_decl on them and
that can result in their destruction, which is obviously undesirable,
we should only destruct the privatized copies of them created in omp
lowering.
Fixed thusly.
2022-12-21 Jakub Jelinek <jakub@redhat.com>
PR c++/108180
* pt.cc (tsubst_expr): Don't call cp_finish_decl on
DECL_OMP_PRIVATIZED_MEMBER vars.
* testsuite/libgomp.c++/pr108180.C: New test.
The make silent the following 2 warnings:
jit/jit-playback.h:785:16: warning: private field 'm_source_file' is not used [-Wunused-private-field]
jit/jit-playback.h:804:16: warning: private field 'm_line' is not used [-Wunused-private-field]
gcc/jit/ChangeLog:
* jit-playback.h: Use unused attribute.
In Fedora build libstdc++.so is built with assertions enabled and
FAIL: 20_util/to_chars/float128_c++23.cc execution test
was failing on all arches. The problem is that it called 5 argument version
of to_chars with chars_format{}, which C++ says is invalid:
http://eel.is/c++draft/charconv.to.chars#12
Preconditions: fmt has the value of one of the enumerators of chars_format.
The following patch fixes it by skipping the second part of the test
which needs the 5 argument to_chars for chars_format{}, but because
it is strictly speaking invalid also for 4 argument to_chars, it uses
3 argument to_chars instead of 4 argument to_chars with last argument
chars_format{}.
2022-12-21 Jakub Jelinek <jakub@redhat.com>
* testsuite/20_util/to_chars/float16_c++23.cc (test): Use 3 argument
std::to_chars if fmt is std::chars_format{}, rather than 4 argument.
* testsuite/20_util/to_chars/float128_c++23.cc (test): Likewise, and
skip second part of testing that requires 5 argument std::to_chars.
On non-Cygwin Windows, use '.' and expect the documented fail when opening
a directory (EACCESS). As gfortran does not set __WIN32__ this check is
done on the C side. (On __CYGWIN__, __WIN32__ is not set - but to make it
clear, !__CYGWIN__ is used in #if.)
On non-Windows, replace the 'call system' shell call by the POSIX functions
stat/mkdir/rmdir for better compatibility, especially on embedded systems;
additionally add some more checks. In particular, confirm that 'close' with
status='delete' indeed deleted the directory.
gcc/testsuite/ChangeLog:
* gfortran.dg/read_dir-aux.c: New; provides my_mkdir, my_rmdir,
my_verify_not_exists and expect_open_to_fail.
* gfortran.dg/read_dir.f90: Call those; expect that opening a
directory fails on Windows.
The hunk for setting flag OPTION_MASK_P10_FUSION locates wrongly
between the if and else if block for OPTION_MASK_MMA. This is
to fix this oversight accordingly.
gcc/ChangeLog:
* config/rs6000/rs6000.cc (rs6000_option_override_internal): Fix the
location for OPTION_MASK_P10_FUSION flag setting.
In function fold_convert_const_real_from_real, when the modes of
two types involved in fp conversion are the same, we can simply
take it as copy, rebuild with the exactly same TREE_REAL_CST and
the target type. It is more efficient and helps to avoid possible
unexpected signalling bit clearing in [1].
[1] https://gcc.gnu.org/pipermail/gcc-patches/2022-December/608533.html
gcc/ChangeLog:
* fold-const.cc (fold_convert_const_real_from_real): Treat floating
point conversion to a type with same mode as copy instead of normal
convertFormat.
As PR106736 shows, it's unexpected to use __vector_quad and
__vector_pair types without MMA support, it would cause ICE
when expanding the corresponding assignment. We can't guard
these built-in types registering under MMA support as Peter
pointed out in that PR, because the registering is global,
it doesn't work for target pragma/attribute support with MMA
enabled. The existing verify_type_context mentioned in [2]
can help to make the diagnostics invalid built-in type uses
better, but as Richard pointed out in [4], it can't deal with
all cases. As the discussions in [1][3], this patch is to
check the invalid use of built-in types __vector_quad and
__vector_pair in mov pattern of OOmode and XOmode, on the
currently being expanded gimple assignment statement. It
still puts an assertion in else arm rather than just makes
it go through, it's to ensure we can catch any other possible
unexpected cases in time if there are.
[1] https://gcc.gnu.org/pipermail/gcc/2022-December/240218.html
[2] https://gcc.gnu.org/pipermail/gcc/2022-December/240220.html
[3] https://gcc.gnu.org/pipermail/gcc/2022-December/240223.html
[4] https://gcc.gnu.org/pipermail/gcc-patches/2022-December/608083.html
PR target/106736
gcc/ChangeLog:
* config/rs6000/mma.md (define_expand movoo): Call function
rs6000_opaque_type_invalid_use_p to check and emit error message for
the invalid use of opaque type.
(define_expand movxo): Likewise.
* config/rs6000/rs6000-protos.h
(rs6000_opaque_type_invalid_use_p): New function declaration.
(currently_expanding_gimple_stmt): New extern declaration.
* config/rs6000/rs6000.cc (rs6000_opaque_type_invalid_use_p): New
function.
gcc/testsuite/ChangeLog:
* gcc.target/powerpc/pr106736-1.c: New test.
* gcc.target/powerpc/pr106736-2.c: Likewise.
* gcc.target/powerpc/pr106736-3.c: Likewise.
* gcc.target/powerpc/pr106736-4.c: Likewise.
* gcc.target/powerpc/pr106736-5.c: Likewise.
Unlike protected_set_expr_location, this variant can return
a different tree.
gcc/ChangeLog:
* fold-const.cc (fold_convert_loc): Check return value of
protected_set_expr_location_unshare.
If the DECL_VALUE_EXPR of a VAR_DECL has EXPR_LOCATION set, then any use of
that variable looks like it has that location, which leads to the debugger
jumping back and forth for both lambdas and structured bindings.
Rather than fix all the uses, it seems simplest to remove any EXPR_LOCATION
when setting DECL_VALUE_EXPR. So the cp/ hunks aren't necessary, but they
avoid the need to unshare to remove the location.
PR c++/84471
PR c++/107504
gcc/cp/ChangeLog:
* coroutines.cc (transform_local_var_uses): Don't
specify a location for DECL_VALUE_EXPR.
* decl.cc (cp_finish_decomp): Likewise.
gcc/ChangeLog:
* fold-const.cc (protected_set_expr_location_unshare): Not static.
* tree.h: Declare it.
* tree.cc (decl_value_expr_insert): Use it.
include/ChangeLog:
* ansidecl.h (ATTRIBUTE_WARN_UNUSED_RESULT): Add __.
gcc/testsuite/ChangeLog:
* g++.dg/tree-ssa/value-expr1.C: New test.
* g++.dg/tree-ssa/value-expr2.C: New test.
* g++.dg/analyzer/pr93212.C: Move warning.
Unfortunately the extract_autos_r fix in r13-4799-ga7c8036b26082d is
derailed by the fact that walk_tree_1 currently walks the elements of a
TREE_VEC in reverse, which means for A<auto, auto> in the below testcase
extract_autos_r ends up adjusting the TEMPLATE_TYPE_IDX of the first
auto instead of the second one, and the first auto is the canonical auto
of level 2 (and index 0), so we rightfully trigger the sanity check
added in that commit.
It seems TREE_VEC and VECTOR_CST are the only trees we walk in reverse,
and this has been the case since r21884 whence the original version of
walk_tree_1 was introduced. But that's arguably unnatural and not
consistent with how we walk all other compound trees such as CONSTRUCTORs
and EXPR_P trees in forward order.
So this patch makes walk_tree_1 walk TREE_VEC (and VECTOR_CST) in
forward order as well, which fixes the testcase. Doing so revealed that
keep_template_parm grows the list of found template parameters from the
front, which previously compensated for the TREE_VEC behavior, so now we
should grow it from the back.
PR c++/101886
gcc/cp/ChangeLog:
* pt.cc (find_template_parameter_info::parm_list_tail):
New data member.
(keep_template_parm): Use parm_list_tail to append rather
than prepend to parm_list.
gcc/ChangeLog:
* tree.cc (walk_tree_1) <case TREE_VEC>: Walk the elements
in forward instead of reverse order.
<case VECTOR_CST>: Likewise.
gcc/testsuite/ChangeLog:
* g++.dg/concepts/diagnostic12.C: Adjust expected order of
template parameters within pretty printed parameter mapping.
* g++.dg/concepts/auto6.C: New test.
'extern "C"' coroutines are permitted by the standard and expected to work
(although constructing useful cases could be challenging). In order to
permit this we need to arrange for the outlined helper functions to be
named properly, even when no mangling is required. To do this, we append
the actor and destroy suffixes in all cases.
Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
gcc/cp/ChangeLog:
* mangle.cc (write_mangled_name): Append the helper function
suffixes here...
(write_encoding): ... rather than here.
gcc/testsuite/ChangeLog:
* g++.dg/coroutines/torture/extern-c-coroutine.C: New test.
This patch fixes some issues with substitution into a C++20 template
parameter object wrapper:
* The first testcase demonstrates a situation where the same_type_p
assert in relevant case of tsubst_copy doesn't hold, because (partial)
substitution of {int,} into the wrapper's TREE_TYPE yields A<int> but
substitution into the underlying TEMPLATE_PARM_INDEX is a nop due to
tsubst's level==1/tf_partial early exit tests, hence TREE_TYPE in the
latter case remains A<T>. So this patch just gets rid of the assert;
the type mismatch doesn't seem to be a problem in practice.
* In the second testcase, dependent substitution into the underlying
TEMPLATE_PARM_INDEX yields a CALL_EXPR with empty TREE_TYPE, which
tsubst_copy doesn't expect. This patch fixes this by handling empty
TREE_TYPE the same way as a non-const TREE_TYPE. Moreover, after
this substitution we're left with a VIEW_CONVERT_EXPR wrapping a
CALL_EXPR instead of a TEMPLATE_PARM_INDEX, which during the subsequent
non-dependent substitution tsubst_copy doesn't expect either. So
this patch also relaxes tsubst_copy to accept such VIEW_CONVERT_EXPR
too.
* In the third testcase, we end up never resolving the call to
f.modify() because tsubst_copy doesn't do overload resolution.
This patch fixes this by moving the handling of these
VIEW_CONVERT_EXPR wrappers from tsubst_copy to tsubst_copy_and_build.
For good measure tsubst_copy_and_build should also handle
REF_PARENTHESIZED_P wrappers instead of delegating to tsubst_copy.
PR c++/103346
PR c++/104278
PR c++/102553
gcc/cp/ChangeLog:
* pt.cc (tsubst_copy) <case VIEW_CONVERT_EXPR>: Move the
handling of C++20 template parameter object wrappers to ...
(tsubst_copy_and_build) <case VIEW_CONVERT_EXPR>: ... here.
Accept non-TEMPLATE_PARM_INDEX inner operand. Handle empty
TREE_TYPE on substituted inner operand. Remove same_type_p
assert. Also handle REF_PARENTHESIZED_P VIEW_CONVERT_EXPRs.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/nontype-class52a.C: New test.
* g++.dg/cpp2a/nontype-class53.C: New test.
* g++.dg/cpp2a/nontype-class54.C: New test.
* g++.dg/cpp2a/nontype-class55.C: New test.
If there is only one argument to a PHI which is defined, an equivalency is
created between the def and the argument. It is safe to consider the def
equal to the argument, but it is dangerous to assume the argument is also
equivalent to the def as there may be branches which change the range on the
path to the PHI on that argument
This patch avoid using that relation in range-on-entry calculations.
PR tree-optimization/108139
gcc/
* gimple-range-cache.cc (ranger_cache::fill_block_cache): Do not
use equivalences originating from PHIS.
gcc/testsuite/
* gcc.dg/pr108139.c: New.
This documents that GDC 9.4 or later is required to build the D
language rather than GDC 9.1 which suffers from PR94240.
PR d/104749
* doc/install.texi (GDC): Document GDC 9.4 or later is required
to build the D language frontend.
The Make-lang.in was missing the link serialization support.
PR rust/108113
gcc/rust
* Make-lang.in (rust.serial): New variable.
(rust1$(exeext)): Depend on $(rust.prev). Call LINK_PROGRESS.
Signed-off-by: Marc Poulhiès <dkm@kataplop.net>
Hi,
When checking eq/ne with a constant which has only 16bits, it can be
optimized to check the rotated data. By this, the constant building
is optimized.
As the example in PR103743:
For "in == 0x8000000000000000LL", this patch generates:
rotldi 3,3,1 ; cmpldi 0,3,1
instead of:
li 9,-1 ; rldicr 9,9,0,0 ; cmpd 0,3,9
Compare with previous version:
https://gcc.gnu.org/pipermail/gcc-patches/2022-August/600475.html.
This patch refactor the code according to review comments.
e.g. updating function names/comments/code.
This patch pass bootstrap and regtest on ppc64 and ppc64le.
Is it ok for trunk? Thanks for comments!
BR,
Jeff(Jiufu)
PR target/103743
gcc/ChangeLog:
* config/rs6000/rs6000-protos.h (can_be_rotated_to_lowbits): New.
(can_be_rotated_to_positive_16bits): New.
(can_be_rotated_to_negative_15bits): New.
* config/rs6000/rs6000.cc (can_be_rotated_to_lowbits): New definition.
(can_be_rotated_to_positive_16bits): New definition.
(can_be_rotated_to_negative_15bits): New definition.
* config/rs6000/rs6000.md (*rotate_on_cmpdi): New define_insn_and_split.
(eqne): Move earlier.
gcc/testsuite/ChangeLog:
* gcc.target/powerpc/pr103743.c: New test.
* gcc.target/powerpc/pr103743_1.c: New test.
It's OK to rely on conditionally-supported features in #if CHECKING_P, since
that isn't defined in stage 1.
gcc/ChangeLog:
* sort.cc: Disable -Wconditionally-supported in
CHECKING_P code.
We currently declare __builtin_source_location with a const void* return
type instead of the actual type const std::source_location::__impl*, and
later when folding this builtin we obtain the actual type via name lookup.
But the below testcase demonstrates this approach seems to interact
poorly with modules, since we may import an entity that uses
std::source_location::current() in its default argument (or DMI) without
necessarily importing <source_location>, and thus the name lookup for
std::source_location will fail at the call site (when using the default
argument) unless we also import <source_location>.
This patch fixes this by instead initially declaring the builtin with an
auto return type and updating it appropriately upon its first use (in
standard code the first/only use would be in the definition of
std::source_location). Thus when folding calls to this builtin we can
get at its return type through the type of the CALL_EXPR and avoid
needing to do a name lookup.
PR c++/100881
gcc/cp/ChangeLog:
* constexpr.cc (cxx_eval_builtin_function_call): Adjust calls
to fold_builtin_source_location.
* cp-gimplify.cc (cp_gimplify_expr): Likewise.
(cp_fold): Likewise.
(get_source_location_impl_type): Remove location_t parameter and
adjust accordingly. No longer static.
(fold_builtin_source_location): Take a CALL_EXPR tree instead of a
location and obtain the impl type from its return type.
* cp-tree.h (enum cp_tree_index): Remove CPTI_SOURCE_LOCATION_IMPL
enumerator.
(source_location_impl): Remove.
(fold_builtin_source_location): Adjust parameter type.
(get_source_location_impl_type): Declare.
* decl.cc (cxx_init_decl_processing): Declare
__builtin_source_location with auto return type instead of
const void*.
(require_deduced_type): Update the return type of
__builtin_source_location.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/srcloc3.C: Adjust expected note s/evaluating/using.
* g++.dg/cpp2a/srcloc4.C: Likewise.
* g++.dg/cpp2a/srcloc5.C: Likewise.
* g++.dg/cpp2a/srcloc6.C: Likewise.
* g++.dg/cpp2a/srcloc7.C: Likewise.
* g++.dg/cpp2a/srcloc8.C: Likewise.
* g++.dg/cpp2a/srcloc9.C: Likewise.
* g++.dg/cpp2a/srcloc10.C: Likewise.
* g++.dg/cpp2a/srcloc11.C: Likewise.
* g++.dg/cpp2a/srcloc12.C: Likewise.
* g++.dg/cpp2a/srcloc13.C: Likewise.
* g++.dg/modules/pr100881_a.C: New test.
* g++.dg/modules/pr100881_b.C: New test.
In extract_autos_r, we need to recompute TYPE_CANONICAL for the template
type parameter after adjusting its index, otherwise we end up with a
comptypes ICE for the below testcase. Note that such in-place type
adjustment isn't generally safe to do since the type could be the
TYPE_CANONICAL of another (unadjusted) type, but in this case the
canonical auto (of some level and 0 index) is the first auto (of that
level) that's created, and so any auto that we do end up adjusting can't
be the canonical one.
PR c++/101886
gcc/cp/ChangeLog:
* pt.cc (extract_autos_r): Recompute TYPE_CANONICAL after
adjusting the template type parameter's index. Simplify
by using TEMPLATE_TYPE_IDX. Add some sanity checks.
gcc/testsuite/ChangeLog:
* g++.dg/concepts/auto5.C: New test.
The lowercase constants are more consistent with the standard, and it is
unlikely that the uppercase versions would've been accepted.
gcc/cp/ChangeLog:
* contracts.cc: Rename references to
contract_violation_continuation_mode constants to be lowercase.
libstdc++-v3/ChangeLog:
* include/experimental/contract: Lowercase the constants in
contract_violation_continuation_mode.
Both C99 and latest C2X say that compound literal shall have an object type
(complete object type in the latter case) or array of unknown bound,
so complit with function type is invalid. When the initializer had to be
non-empty for such case, we used to diagnose it as incorrect initializer,
but with (fntype){} now allowed we just ICE on it.
The following patch diagnoses that.
2022-12-19 Jakub Jelinek <jakub@redhat.com>
PR c/108043
* c-parser.cc (c_parser_postfix_expression_after_paren_type): Diagnose
compound literals with function type.
* gcc.dg/pr108043.c: New test.
* gcc.dg/c99-complit-2.c (foo): Adjust expected diagnostics for
complit with function type.
Here we crash because check_function_format was using TREE_PURPOSE
directly rather than using get_attribute_name.
PR c/98487
gcc/c-family/ChangeLog:
* c-format.cc (check_function_format): Use get_attribute_name.
gcc/testsuite/ChangeLog:
* c-c++-common/Wsuggest-attribute-1.c: New test.
The PR (which isn't resolved by this commit) pointed out to me that GCC
should build with -Wconditionally-supported to support bootstrapping with a
C++11 compiler that makes different choices.
PR c++/64867
gcc/ChangeLog:
* configure.ac (strict_warn): Add -Wconditionally-supported.
* configure: Regenerate.
vect_update_ivs_after_vectorizer can end up emitting a signed
IV update when the loop body performed an unsigned computation.
The following makes sure to perform that update in the type
of the loop update type to avoid undefined behavior on overflow.
PR tree-optimization/108164
* tree-vect-loop-manip.cc (vect_update_ivs_after_vectorizer):
Perform vect_step_op_add update in the appropriate type.
* gcc.dg/pr108164.c: New testcase.
The ACLE requires that __ARM_FEATURE_CLZ be defined if the hardware
supports it; it's also clear that this doesn't mean the current ISA,
so we must define this even when compiling for Thumb1 if the target
supports CLZ in A32.
This brings GCC into alignment with Clang.
gcc/ChangeLog:
* config/arm/arm-c.cc (__ARM_FEATURE_CLZ): Fix definition of
preprocessor macro when target has CLZ in another ISA.
Since store instructions doesn't care about tail policy, we remove
vste from "ta" attribute. Hence, we could have more fusion chances
and better optimization.
gcc/ChangeLog:
* config/riscv/vector.md: Remove vste.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-1.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-10.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-11.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-12.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-13.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-14.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-15.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-16.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-17.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-18.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-19.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-2.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-20.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-21.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-22.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-23.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-24.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-25.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-26.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-27.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-28.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-29.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-3.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-30.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-31.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-32.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-33.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-34.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-35.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-36.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-37.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-38.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-39.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-4.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-40.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-41.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-42.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-43.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-44.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-45.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-46.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-5.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-6.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-7.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-8.c: New test.
* gcc.target/riscv/rvv/vsetvl/vlmax_back_prop-9.c: New test.
