This patch gets CSE to re-use constants already inside a vector rather than
re-materializing the constant again.
Basically consider the following case:
#include <stdint.h>
#include <arm_neon.h>
uint64_t
test (uint64_t a, uint64x2_t b, uint64x2_t* rt)
{
uint64_t arr[2] = { 0x0942430810234076UL, 0x0942430810234076UL};
uint64_t res = a | arr[0];
uint64x2_t val = vld1q_u64 (arr);
*rt = vaddq_u64 (val, b);
return res;
}
The actual behavior is inconsequential however notice that the same constants
are used in the vector (arr and later val) and in the calculation of res.
The code we generate for this however is quite sub-optimal:
test:
adrp x2, .LC0
sub sp, sp, #16
ldr q1, [x2, #:lo12:.LC0]
mov x2, 16502
movk x2, 0x1023, lsl 16
movk x2, 0x4308, lsl 32
add v1.2d, v1.2d, v0.2d
movk x2, 0x942, lsl 48
orr x0, x0, x2
str q1, [x1]
add sp, sp, 16
ret
.LC0:
.xword 667169396713799798
.xword 667169396713799798
Essentially we materialize the same constant twice. The reason for this is
because the front-end lowers the constant extracted from arr[0] quite early on.
If you look into the result of fre you'll find
<bb 2> :
arr[0] = 667169396713799798;
arr[1] = 667169396713799798;
res_7 = a_6(D) | 667169396713799798;
_16 = __builtin_aarch64_ld1v2di (&arr);
_17 = VIEW_CONVERT_EXPR<uint64x2_t>(_16);
_11 = b_10(D) + _17;
*rt_12(D) = _11;
arr ={v} {CLOBBER};
return res_7;
Which makes sense for further optimization. However come expand time if the
constant isn't representable in the target arch it will be assigned to a
register again.
(insn 8 5 9 2 (set (reg:V2DI 99)
(const_vector:V2DI [
(const_int 667169396713799798 [0x942430810234076]) repeated x2
])) "cse.c":7:12 -1
(nil))
...
(insn 14 13 15 2 (set (reg:DI 103)
(const_int 667169396713799798 [0x942430810234076])) "cse.c":8:12 -1
(nil))
(insn 15 14 16 2 (set (reg:DI 102 [ res ])
(ior:DI (reg/v:DI 96 [ a ])
(reg:DI 103))) "cse.c":8:12 -1
(nil))
And since it's out of the immediate range of the scalar instruction used
combine won't be able to do anything here.
This will then trigger the re-materialization of the constant twice.
To fix this this patch extends CSE to be able to generate an extract for a
constant from another vector, or to make a vector for a constant by duplicating
another constant.
Whether this transformation is done or not depends entirely on the costing for
the target for the different constants and operations.
I Initially also investigated doing this in PRE, but PRE requires at least 2 BB
to work and does not currently have any way to remove redundancies within a
single BB and it did not look easy to support.
gcc/ChangeLog:
* cse.c (add_to_set): New.
(find_sets_in_insn): Register constants in sets.
(canonicalize_insn): Use auto_vec instead.
(cse_insn): Try materializing using vec_dup.
* rtl.h (simplify_context::simplify_gen_vec_select,
simplify_gen_vec_select): New.
* simplify-rtx.c (simplify_context::simplify_gen_vec_select): New.
Some targets have overriden the default unroll factor and so do not have enough
data to succeed for SLP vectorization if loop vect is turned off.
To fix this just always unroll in these testcases.
gcc/testsuite/ChangeLog:
PR testsuite/103000
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-double.c:
Force unroll.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-float.c: likewise
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-pattern-float.c:
Likewise
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-pattern-half-float.c:
Likewise.
This patch adds support to GCC's diagnostic subsystem for escaping certain
bytes and Unicode characters when quoting source code.
Specifically, this patch adds a new flag rich_location::m_escape_on_output
which is a hint from a diagnostic that non-ASCII bytes in the pertinent
lines of the user's source code should be escaped when printed.
The patch sets this for the following diagnostics:
- when complaining about stray bytes in the program (when these
are non-printable)
- when complaining about "null character(s) ignored");
- for -Wnormalized= (and generate source ranges for such warnings)
The escaping is controlled by a new option:
-fdiagnostics-escape-format=[unicode|bytes]
For example, consider a diagnostic involing a source line containing the
string "before" followed by the Unicode character U+03C0 ("GREEK SMALL
LETTER PI", with UTF-8 encoding 0xCF 0x80) followed by the byte 0xBF
(a stray UTF-8 trailing byte), followed by the string "after", where the
diagnostic highlights the U+03C0 character.
By default, this line will be printed verbatim to the user when
reporting a diagnostic at it, as:
beforeπXafter
^
(using X for the stray byte to avoid putting invalid UTF-8 in this
commit message)
If the diagnostic sets the "escape" flag, it will be printed as:
before<U+03C0><BF>after
^~~~~~~~
with -fdiagnostics-escape-format=unicode (the default), or as:
before<CF><80><BF>after
^~~~~~~~
if the user supplies -fdiagnostics-escape-format=bytes.
This only affects how the source is printed; it does not affect
how column numbers that are printed (as per -fdiagnostics-column-unit=
and -fdiagnostics-column-origin=).
gcc/c-family/ChangeLog:
* c-lex.c (c_lex_with_flags): When complaining about non-printable
CPP_OTHER tokens, set the "escape on output" flag.
gcc/ChangeLog:
* common.opt (fdiagnostics-escape-format=): New.
(diagnostics_escape_format): New enum.
(DIAGNOSTICS_ESCAPE_FORMAT_UNICODE): New enum value.
(DIAGNOSTICS_ESCAPE_FORMAT_BYTES): Likewise.
* diagnostic-format-json.cc (json_end_diagnostic): Add
"escape-source" attribute.
* diagnostic-show-locus.c
(exploc_with_display_col::exploc_with_display_col): Replace
"tabstop" param with a cpp_char_column_policy and add an "aspect"
param. Use these to compute m_display_col accordingly.
(struct char_display_policy): New struct.
(layout::m_policy): New field.
(layout::m_escape_on_output): New field.
(def_policy): New function.
(make_range): Update for changes to exploc_with_display_col ctor.
(default_print_decoded_ch): New.
(width_per_escaped_byte): New.
(escape_as_bytes_width): New.
(escape_as_bytes_print): New.
(escape_as_unicode_width): New.
(escape_as_unicode_print): New.
(make_policy): New.
(layout::layout): Initialize new fields. Update m_exploc ctor
call for above change to ctor.
(layout::maybe_add_location_range): Update for changes to
exploc_with_display_col ctor.
(layout::calculate_x_offset_display): Update for change to
cpp_display_width.
(layout::print_source_line): Pass policy
to cpp_display_width_computation. Capture cpp_decoded_char when
calling process_next_codepoint. Move printing of source code to
m_policy.m_print_cb.
(line_label::line_label): Pass in policy rather than context.
(layout::print_any_labels): Update for change to line_label ctor.
(get_affected_range): Pass in policy rather than context, updating
calls to location_compute_display_column accordingly.
(get_printed_columns): Likewise, also for cpp_display_width.
(correction::correction): Pass in policy rather than tabstop.
(correction::compute_display_cols): Pass m_policy rather than
m_tabstop to cpp_display_width.
(correction::m_tabstop): Replace with...
(correction::m_policy): ...this.
(line_corrections::line_corrections): Pass in policy rather than
context.
(line_corrections::m_context): Replace with...
(line_corrections::m_policy): ...this.
(line_corrections::add_hint): Update to use m_policy rather than
m_context.
(line_corrections::add_hint): Likewise.
(layout::print_trailing_fixits): Likewise.
(selftest::test_display_widths): New.
(selftest::test_layout_x_offset_display_utf8): Update to use
policy rather than tabstop.
(selftest::test_one_liner_labels_utf8): Add test of escaping
source lines.
(selftest::test_diagnostic_show_locus_one_liner_utf8): Update to
use policy rather than tabstop.
(selftest::test_overlapped_fixit_printing): Likewise.
(selftest::test_overlapped_fixit_printing_utf8): Likewise.
(selftest::test_overlapped_fixit_printing_2): Likewise.
(selftest::test_tab_expansion): Likewise.
(selftest::test_escaping_bytes_1): New.
(selftest::test_escaping_bytes_2): New.
(selftest::diagnostic_show_locus_c_tests): Call the new tests.
* diagnostic.c (diagnostic_initialize): Initialize
context->escape_format.
(convert_column_unit): Update to use default character width policy.
(selftest::test_diagnostic_get_location_text): Likewise.
* diagnostic.h (enum diagnostics_escape_format): New enum.
(diagnostic_context::escape_format): New field.
* doc/invoke.texi (-fdiagnostics-escape-format=): New option.
(-fdiagnostics-format=): Add "escape-source" attribute to examples
of JSON output, and document it.
* input.c (location_compute_display_column): Pass in "policy"
rather than "tabstop", passing to
cpp_byte_column_to_display_column.
(selftest::test_cpp_utf8): Update to use cpp_char_column_policy.
* input.h (class cpp_char_column_policy): New forward decl.
(location_compute_display_column): Pass in "policy" rather than
"tabstop".
* opts.c (common_handle_option): Handle
OPT_fdiagnostics_escape_format_.
* selftest.c (temp_source_file::temp_source_file): New ctor
overload taking a size_t.
* selftest.h (temp_source_file::temp_source_file): Likewise.
gcc/testsuite/ChangeLog:
* c-c++-common/diagnostic-format-json-1.c: Add regexp to consume
"escape-source" attribute.
* c-c++-common/diagnostic-format-json-2.c: Likewise.
* c-c++-common/diagnostic-format-json-3.c: Likewise.
* c-c++-common/diagnostic-format-json-4.c: Likewise, twice.
* c-c++-common/diagnostic-format-json-5.c: Likewise.
* gcc.dg/cpp/warn-normalized-4-bytes.c: New test.
* gcc.dg/cpp/warn-normalized-4-unicode.c: New test.
* gcc.dg/encoding-issues-bytes.c: New test.
* gcc.dg/encoding-issues-unicode.c: New test.
* gfortran.dg/diagnostic-format-json-1.F90: Add regexp to consume
"escape-source" attribute.
* gfortran.dg/diagnostic-format-json-2.F90: Likewise.
* gfortran.dg/diagnostic-format-json-3.F90: Likewise.
libcpp/ChangeLog:
* charset.c (convert_escape): Use encoding_rich_location when
complaining about nonprintable unknown escape sequences.
(cpp_display_width_computation::::cpp_display_width_computation):
Pass in policy rather than tabstop.
(cpp_display_width_computation::process_next_codepoint): Add "out"
param and populate *out if non-NULL.
(cpp_display_width_computation::advance_display_cols): Pass NULL
to process_next_codepoint.
(cpp_byte_column_to_display_column): Pass in policy rather than
tabstop. Pass NULL to process_next_codepoint.
(cpp_display_column_to_byte_column): Pass in policy rather than
tabstop.
* errors.c (cpp_diagnostic_get_current_location): New function,
splitting out the logic from...
(cpp_diagnostic): ...here.
(cpp_warning_at): New function.
(cpp_pedwarning_at): New function.
* include/cpplib.h (cpp_warning_at): New decl for rich_location.
(cpp_pedwarning_at): Likewise.
(struct cpp_decoded_char): New.
(struct cpp_char_column_policy): New.
(cpp_display_width_computation::cpp_display_width_computation):
Replace "tabstop" param with "policy".
(cpp_display_width_computation::process_next_codepoint): Add "out"
param.
(cpp_display_width_computation::m_tabstop): Replace with...
(cpp_display_width_computation::m_policy): ...this.
(cpp_byte_column_to_display_column): Replace "tabstop" param with
"policy".
(cpp_display_width): Likewise.
(cpp_display_column_to_byte_column): Likewise.
* include/line-map.h (rich_location::escape_on_output_p): New.
(rich_location::set_escape_on_output): New.
(rich_location::m_escape_on_output): New.
* internal.h (cpp_diagnostic_get_current_location): New decl.
(class encoding_rich_location): New.
* lex.c (skip_whitespace): Use encoding_rich_location when
complaining about null characters.
(warn_about_normalization): Generate a source range when
complaining about improperly normalized tokens, rather than just a
point, and use encoding_rich_location so that the source code
is escaped on printing.
* line-map.c (rich_location::rich_location): Initialize
m_escape_on_output.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
The std::begin and std::end overloads for std::valarray are defined in
terms of std::addressof(v[0]) which is undefined for an empty valarray.
libstdc++-v3/ChangeLog:
PR libstdc++/103022
* include/std/valarray (begin, end): Do not dereference an empty
valarray. Add noexcept and [[nodiscard]].
* testsuite/26_numerics/valarray/range_access.cc: Check empty
valarray. Check iterator properties. Run as well as compiling.
* testsuite/26_numerics/valarray/range_access2.cc: Likewise.
* testsuite/26_numerics/valarray/103022.cc: New test.
Chasing down stage3 miscomparisons is never fun, and having no way to
distinguish between jump threads registered by a particular
pass, is even harder. This patch adds debug counters for the individual
back threading passes. I've left the ethread pass alone, as that one is
usually benign, but we could easily add it if needed.
The fact that we can only pass one boolean argument to the passes
infrastructure has us do all sorts of gymnastics to differentiate
between the various back threading passes.
Tested on x86-64 Linux.
gcc/ChangeLog:
* dbgcnt.def: Add debug counter for back_thread[12] and
back_threadfull[12].
* passes.def: Pass "first" argument to each back threading pass.
* tree-ssa-threadbackward.c (back_threader::back_threader): Add
first argument.
(back_threader::debug_counter): New.
(back_threader::maybe_register_path): Call debug_counter.
This patch moves all the static functions into the pass class, and
cleans up things a little. The goal is to shuffle things around such
that we can add debug counters that depend on different threading
passes, but it's a clean-up on its own right.
Tested on x86-64 Linux.
gcc/ChangeLog:
* tree-ssa-threadbackward.c (BT_NONE): New.
(BT_SPEED): New.
(BT_RESOLVE): New.
(back_threader::back_threader): Add flags.
Move loop initialization here.
(back_threader::~back_threader): New.
(back_threader::find_taken_edge_switch): Change solver and ranger
to pointers.
(back_threader::find_taken_edge_cond): Same.
(back_threader::find_paths_to_names): Same.
(back_threader::find_paths): Same.
(back_threader::dump): Same.
(try_thread_blocks): Merge into thread_blocks.
(back_threader::thread_blocks): New.
(do_early_thread_jumps): Merge into thread_blocks.
(do_thread_jumps): Merge into thread_blocks.
(back_threader::thread_through_all_blocks): Remove.
gcc/
PR tree-optimization/103003
* value-relation.cc (dom_oracle::register_relation): If the 2
ssa names are the same, don't register any relation.
gcc/testsuite/
* gcc.dg/pr103003.c: New.
During the generation of the epilogue of aarch64(aarch64_expand_epilogue),
the value of crtl->calls_eh_return does not need to be checked again.
This value has been checked during aarch64_return_address_signing_enabled.
gcc/ChangeLog:
* config/aarch64/aarch64.c (aarch64_expand_epilogue): Remove
redundant check for calls_eh_return.
* config/aarch64/aarch64.md (*do_return): Likewise.
Signed-off-by: Dan Li <ashimida@linux.alibaba.com>
loop_version currently does lv_adjust_loop_entry_edge
before it loopifys the copy inserted on the header. This patch moves
the condition generation later and thus we have four pieces to help
understanding of how the adjustment works:
1) duplicating the loop on the entry edge.
2) loopify the duplicated new loop.
3) adjusting the CFG to insert a condition branching to either loop
with lv_adjust_loop_entry_edge.
4) From loopify extract the scale_loop_frequencies bits.
Also removed some piece of code seems obviously useless:
- redirect_all_edges since it is false and loopify only called once.
- extract_cond_bb_edges and lv_flush_pending_stmts (false_edge) as the
edge is not redirected actually.
gcc/ChangeLog:
2021-11-01 Xionghu Luo <luoxhu@linux.ibm.com>
* cfgloopmanip.c (loop_version): Refactor loopify to
loop_version. Move condition generation after loopify.
(loopify): Delete.
* cfgloopmanip.h (loopify): Delete.
while preparing testcase for return slot tracking I noticed that both
ipa-pure-const and modref treat return slot writes as non-local which prevents
detecting functions as pure or not modifying global state. Fixed by making
points_to_local_or_readonly_memory_p to special case return slot. This is bit
of a side case, but presently at all uses of
points_to_local_or_readonly_memory_p we want to handle return slot this way.
I also noticed that we handle gimple copy unnecesarily pesimistically. This
does not make difference right now since we do no not track non-scalars, but
I fixed it anyway.
Bootstrapped/regtested x86_64-linux, comitted.
gcc/ChangeLog:
* ipa-fnsummary.c: Include tree-dfa.h.
(points_to_local_or_readonly_memory_p): Return true on return
slot writes.
* ipa-modref.c (analyze_ssa_name_flags): Fix handling of copy
statement.
gcc/testsuite/ChangeLog:
* g++.dg/ipa/modref-1.C: New test.
The _Unwind_Exception struct had its alignment adjusted to 16-bytes,
however malloc() on Solaris X86 is not guaranteed to allocate memory
aligned to 16-bytes as well.
PR d/102837
libphobos/ChangeLog:
* libdruntime/gcc/deh.d (ExceptionHeader.free): Use memset to reset
contents of internal EH storage.
The value used to initialize the integer field in the union didn't
account for BigEndian targets running this code.
PR d/102959
gcc/testsuite/ChangeLog:
* gdc.dg/torture/pr96435.d: Adjust for BigEndian.
The teams construct only permits omp_get_num_teams and omp_get_team_num
as API call in strictly nested regions - check for it.
Additionally, for Fortran, using DECL_NAME does not show the mangled
name, hence, DECL_ASSEMBLER_NAME had to be used to.
Finally, 'target device(ancestor:1)' wrongly rejected non-API calls
as well.
PR middle-end/102972
gcc/ChangeLog:
* omp-low.c (omp_runtime_api_call): Use DECL_ASSEMBLER_NAME to get
internal Fortran name; new permit_num_teams arg to permit
omp_get_num_teams and omp_get_team_num.
(scan_omp_1_stmt): Update call to it, add missing call for
reverse offload, and check for strictly nested API calls in teams.
gcc/testsuite/ChangeLog:
* c-c++-common/gomp/target-device-ancestor-3.c: Add non-API
routine test.
* gfortran.dg/gomp/order-6.f90: Add missing bind(C).
* c-c++-common/gomp/teams-3.c: New test.
* gfortran.dg/gomp/teams-3.f90: New test.
* gfortran.dg/gomp/teams-4.f90: New test.
libgomp/ChangeLog:
* testsuite/libgomp.c-c++-common/icv-3.c: Nest API calls inside
parallel construct.
* testsuite/libgomp.c-c++-common/icv-4.c: Likewise.
* testsuite/libgomp.c/target-3.c: Likewise.
* testsuite/libgomp.c/target-5.c: Likewise.
* testsuite/libgomp.c/target-6.c: Likewise.
* testsuite/libgomp.c/target-teams-1.c: Likewise.
* testsuite/libgomp.c/teams-1.c: Likewise.
* testsuite/libgomp.c/thread-limit-2.c: Likewise.
* testsuite/libgomp.c/thread-limit-3.c: Likewise.
* testsuite/libgomp.c/thread-limit-4.c: Likewise.
* testsuite/libgomp.c/thread-limit-5.c: Likewise.
* testsuite/libgomp.fortran/icv-3.f90: Likewise.
* testsuite/libgomp.fortran/icv-4.f90: Likewise.
* testsuite/libgomp.fortran/teams1.f90: Likewise.
We did not free global symbols. For a simplified abstract_type_3.f90
valgrind reports:
96 bytes in 1 blocks are still reachable in loss record 461 of 602
at 0x48377D5: calloc (vg_replace_malloc.c:711)
by 0x21257C3: xcalloc (xmalloc.c:162)
by 0x98611B: gfc_get_gsymbol(char const*) (symbol.c:4341)
by 0x932C58: parse_module() (parse.c:5912)
by 0x9336F8: gfc_parse_file() (parse.c:6236)
by 0x991449: gfc_be_parse_file() (f95-lang.c:204)
by 0x11D8EDE: compile_file() (toplev.c:455)
by 0x11DB9C3: do_compile() (toplev.c:2170)
by 0x11DBCAF: toplev::main(int, char**) (toplev.c:2305)
by 0x2045D37: main (main.c:39)
This patch reduces this to
LEAK SUMMARY:
definitely lost: 344 bytes in 1 blocks
indirectly lost: 3,024 bytes in 4 blocks
possibly lost: 0 bytes in 0 blocks
- still reachable: 1,576,174 bytes in 2,277 blocks
+ still reachable: 1,576,078 bytes in 2,276 blocks
suppressed: 0 bytes in 0 blocks
gcc/fortran/ChangeLog:
2018-10-21 Bernhard Reutner-Fischer <aldot@gcc.gnu.org>
* parse.c (clean_up_modules): Free gsym.
Improve the wording on optimizations that prevent compare hardening,
so as to also cover cases in which explicit compares get combined into
operations with implied compares.
for gcc/ada/ChangeLog
* doc/gnat_rm/security_hardening_features.rst: Mention
optimization to operations with implied compares.
We weren't diagnosing the
The loop iteration variable may not appear in a threadprivate directive.
restriction which used to be in 5.0 just among the Worksharing-Loop
restrictions but in 5.1 it is among Canonical Loop Nest Form restrictions.
This patch diagnoses those.
2021-10-30 Jakub Jelinek <jakub@redhat.com>
* gimplify.c (gimplify_omp_for): Diagnose threadprivate iterators.
* c-c++-common/gomp/loop-10.c: New test.
The sharing of the COMPLEX_MUL node makes it so it's
more efficient to not generate both a MUL and FMA
in this node.
Because the shape for a normal FMA is not different
the FMA is no longer detected here which results in
better codegen so update the testcase.
gcc/testsuite/ChangeLog:
* g++.dg/vect/pr99149.cc: Update case.
Both #pragma and _Pragma ended up as CPP_PRAGMA. Presumably since
r131819 (2008, GCC 4.3) for PR34692, pragmas are not expanded in
macro arguments but are output as is before. From the old bug report,
that was to fix usage like
FOO (
#pragma GCC diagnostic
)
However, that change also affected _Pragma such that
BAR (
"1";
_Pragma("omp ..."); )
yielded
#pragma omp ...
followed by what BAR expanded too, possibly including '"1";'.
This commit adds a flag, PRAGMA_OP, to tokens to make the two
distinguishable - and include again _Pragma in the expanded arguments.
libcpp/ChangeLog:
PR c++/102409
* directives.c (destringize_and_run): Add PRAGMA_OP to the
CPP_PRAGMA token's flags to mark is as coming from _Pragma.
* include/cpplib.h (PRAGMA_OP): #define, to be used with token flags.
* macro.c (collect_args): Only handle CPP_PRAGMA special if PRAGMA_OP
is set.
gcc/testsuite/ChangeLog:
* c-c++-common/gomp/pragma-1.c: New test.
* c-c++-common/gomp/pragma-2.c: New test.
Adding newlines so that the two strings line up makes string equality
failures considerably easier to read.
gcc/ChangeLog:
* selftest.c (assert_streq): Add newlines when emitting non-equal
non-NULL strings.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
compiling gfortran.dg/typebound_proc_31.f90 leaked the type-bound
structs:
56 bytes in 1 blocks are definitely lost.
at 0x4C2CC05: calloc (vg_replace_malloc.c:711)
by 0x151EA90: xcalloc (xmalloc.c:162)
by 0x8E3E4F: gfc_get_typebound_proc(gfc_typebound_proc*) (symbol.c:4945)
by 0x84C095: match_procedure_in_type (decl.c:10486)
by 0x84C095: gfc_match_procedure() (decl.c:6696)
...
gcc/fortran/ChangeLog:
2017-12-06 Bernhard Reutner-Fischer <aldot@gcc.gnu.org>
* symbol.c (free_tb_tree): Free type-bound procedure struct.
(gfc_get_typebound_proc): Use explicit memcpy for clarity.
Since registering a kill means removing all references to it from the
path oracle list, make sure we don't look back to the root oracle
either.
Tested on x86-64 Linux.
Co-authored-by: Andrew MacLeod <amacleod@redhat.com>
gcc/ChangeLog:
* value-relation.cc (path_oracle::killing_def): Add a
self-equivalence so we don't look to the root oracle.
This patch upgrades the pre-VRP threading passes to fully resolving
backward threaders, and removes the post-VRP threading passes altogether.
With it, we reduce the number of threaders in our pipeline from 9 to 7.
This will leave DOM as the only forward threader client. When the ranger
can handle floats, we should be able to upgrade the pre-DOM threaders to
fully resolving threaders and kill the embedded DOM threader.
The numbers are as follows:
prev: # threads in backward + vrp-threaders = 92624
now: # threads in backward threaders = 94275
Gain: +1.78%
prev: # total threads: 189495
now: # total threads: 193714
Gain: +2.22%
The numbers are not as great as my initial proposal, but I've
recently pushed all the work that got us to this point ;-).
And... the compilation improves by 1.32%!
There's a regression on uninit-pred-7_a.c that I've yet to look at. I
want to make sure it's not a missing thread. If it is, I'll create a PR
and own it.
Also, the tree-ssa/phi_on_compare-*.c tests have all regressed. This
seems to be some special case the forward threader handles that the
backward threader does not (edge_forwards_cmp_to_conditional_jump*).
I haven't dug deep to see if this is solveable within our
infrastructure, but a cursory look shows that even though the VRP
threader threads this, the *.optimized dump ends with more conditional
jumps than without the optimization. I'd like to punt on this for
now, because DOM actually catches this through its lone use of the
forward threader (I've adjusted the tests). However, we will need to
address this sooner or later, if indeed it's still improving the final
assembly.
gcc/ChangeLog:
* passes.def: Replace the pass_thread_jumps before VRP* with
pass_thread_jumps_full. Remove all pass_vrp_threader instances.
* tree-ssa-threadbackward.c (pass_data_thread_jumps_full):
Remove hyphen from "thread-full" name.
libgomp/ChangeLog:
* testsuite/libgomp.graphite/force-parallel-4.c: Adjust for threading changes.
* testsuite/libgomp.graphite/force-parallel-8.c: Same.
gcc/testsuite/ChangeLog:
* gcc.dg/loop-unswitch-2.c: Adjust for threading changes.
* gcc.dg/old-style-asm-1.c: Same.
* gcc.dg/tree-ssa/phi_on_compare-1.c: Same.
* gcc.dg/tree-ssa/phi_on_compare-2.c: Same.
* gcc.dg/tree-ssa/phi_on_compare-3.c: Same.
* gcc.dg/tree-ssa/phi_on_compare-4.c: Same.
* gcc.dg/tree-ssa/pr20701.c: Same.
* gcc.dg/tree-ssa/pr21001.c: Same.
* gcc.dg/tree-ssa/pr21294.c: Same.
* gcc.dg/tree-ssa/pr21417.c: Same.
* gcc.dg/tree-ssa/pr21559.c: Same.
* gcc.dg/tree-ssa/pr21563.c: Same.
* gcc.dg/tree-ssa/pr49039.c: Same.
* gcc.dg/tree-ssa/pr59597.c: Same.
* gcc.dg/tree-ssa/pr61839_1.c: Same.
* gcc.dg/tree-ssa/pr61839_3.c: Same.
* gcc.dg/tree-ssa/pr66752-3.c: Same.
* gcc.dg/tree-ssa/pr68198.c: Same.
* gcc.dg/tree-ssa/pr77445-2.c: Same.
* gcc.dg/tree-ssa/pr77445.c: Same.
* gcc.dg/tree-ssa/ranger-threader-1.c: Same.
* gcc.dg/tree-ssa/ranger-threader-2.c: Same.
* gcc.dg/tree-ssa/ranger-threader-4.c: Same.
* gcc.dg/tree-ssa/ssa-dom-thread-1.c: Same.
* gcc.dg/tree-ssa/ssa-dom-thread-11.c: Same.
* gcc.dg/tree-ssa/ssa-dom-thread-12.c: Same.
* gcc.dg/tree-ssa/ssa-dom-thread-14.c: Same.
* gcc.dg/tree-ssa/ssa-dom-thread-16.c: Same.
* gcc.dg/tree-ssa/ssa-dom-thread-2b.c: Same.
* gcc.dg/tree-ssa/ssa-dom-thread-7.c: Same.
* gcc.dg/tree-ssa/ssa-thread-14.c: Same.
* gcc.dg/tree-ssa/ssa-thread-backedge.c: Same.
* gcc.dg/tree-ssa/ssa-vrp-thread-1.c: Same.
* gcc.dg/tree-ssa/vrp02.c: Same.
* gcc.dg/tree-ssa/vrp03.c: Same.
* gcc.dg/tree-ssa/vrp05.c: Same.
* gcc.dg/tree-ssa/vrp06.c: Same.
* gcc.dg/tree-ssa/vrp07.c: Same.
* gcc.dg/tree-ssa/vrp08.c: Same.
* gcc.dg/tree-ssa/vrp09.c: Same.
* gcc.dg/tree-ssa/vrp33.c: Same.
* gcc.dg/uninit-pred-9_b.c: Same.
* gcc.dg/uninit-pred-7_a.c: xfail.
Occasionally I've been seeing failures with the multi-line diagnostics. It's never been clear what's causing the spurious failures, though I have long suspected a greedy regexp match.
It happened again yesterday with a local change that in no way should affect diagnostics, so I finally went searching and found that sure enough the multi-line diagnostics had a ".*" in their regexp. According to the comments, the .* is primarily to catch any dg directives that may appear -- ie it should eat to EOL, but not multiple lines. But a .* can indeed match a newline and cause it to eat multiple lines.
The fix is simple. [^\r\n]* will eat to EOL, but not further.
Regression tested on x86_64 and on our internal target.
gcc/testsuite
* lib/multiline.exp (_build_multiline_regex): Use a better
regexp than .* to match up to EOL.
Propagation is automatically done by the temporal cache when defs are
out of date from the names on the RHS, but a gcond has no LHS, and any
updates on the RHS are never propagated. Always propagate them.
gcc/
PR tree-optimization/102983
* gimple-range-cache.h (propagate_updated_value): Make public.
* gimple-range.cc (gimple_ranger::range_of_stmt): Propagate exports
when processing gcond stmts.
gcc/testsuite/
* gcc.dg/pr102983.c: New.
Extend modref and tree-ssa-structalias to handle retslot flags.
Since retslot it essentially a hidden argument that is known to be write-only
we can do pretty much the same stuff as we do for regular parameters.
I plan to add static chain handling similar way.
We do not handle IPA propagation of retslot flags (where return slot is
initialized via return slot of other function). For this ipa-prop needs
to be extended to understand retslot as well.
Bootstrapped/regtested x86_64-linux, OK for the gimple bits?
Honza
gcc/ChangeLog:
* gimple.c (gimple_call_retslot_flags): New function.
* gimple.h (gimple_call_retslot_flags): Declare.
* ipa-modref.c: Include tree-cfg.h.
(struct escape_entry): Turn parm_index to signed.
(modref_summary_lto::modref_summary_lto): Add retslot_flags.
(modref_summary::modref_summary): Initialize retslot_flags.
(struct modref_summary_lto): Likewise.
(modref_summary::useful_p): Check retslot_flags.
(modref_summary_lto::useful_p): Likewise.
(modref_summary::dump): Dump retslot_flags.
(modref_summary_lto::dump): Likewise.
(struct escape_point): Add hidden_args enum.
(analyze_ssa_name_flags): Ignore return slot return;
use gimple_call_retslot_flags.
(record_escape_points): Break out from ...
(analyze_parms): ... here; handle retslot_flags.
(modref_summaries::duplicate): Duplicate retslot_flags.
(modref_summaries_lto::duplicate): Likewise.
(modref_write_escape_summary): Stream parm_index as signed.
(modref_read_escape_summary): Likewise.
(modref_write): Stream retslot_flags.
(read_section): Likewise.
(struct escape_map): Fix typo in comment.
(update_escape_summary_1): Fix whitespace.
(ipa_merge_modref_summary_after_inlining): Drop retslot_flags.
(modref_merge_call_site_flags): Merge retslot_flags.
* ipa-modref.h (struct modref_summary): Add retslot_flags.
* tree-ssa-structalias.c (handle_rhs_call): Handle retslot_flags.
The layout of the SLP tree has changed in GCC 12 which
broke the detection of complex FMA and FMS.
This patch updates the detection to the new tree shape
and by necessity merges the complex MUL and FMA detection
into one.
This does not yet address the wrong code-gen PR which I
will fix in a different patch as that needs backporting.
gcc/ChangeLog:
PR tree-optimization/102977
* tree-vect-slp-patterns.c (vect_match_call_p): Remove.
(vect_detect_pair_op): Add crosslane check.
(vect_match_call_complex_mla): Remove.
(class complex_mul_pattern): Update comment.
(complex_mul_pattern::matches): Update detection.
(class complex_fma_pattern): Remove.
(complex_fma_pattern::matches): Remove.
(complex_fma_pattern::recognize): Remove.
(complex_fma_pattern::build): Remove.
(class complex_fms_pattern): Update comment.
(complex_fms_pattern::matches): Remove.
(complex_operations_pattern::recognize): Remove complex_fma_pattern
As mentioned yesterday, gimple_fold_builtin_memset doesn't preserve
locus which means e.g. the -Wstringop-overflow warnings are emitted as:
In function 'test_max':
cc1: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
The function emits up to 2 new statements, but the latter (asgn) is added
through gsi_replace and therefore the locus is copied over from the call.
But store is emitted before the call and optionally the call removed
afterwards, so locus needs to be copied over manually.
2021-10-29 Jakub Jelinek <jakub@redhat.com>
* gimple-fold.c (gimple_fold_builtin_memset): Copy over location from
call to store.
* gcc.dg/Wstringop-overflow-62.c: Adjust expected diagnostics.
This forces -fexcess-precision=standard since the testcase is
otherwise prone to fail with x87 math.
2021-10-29 Richard Biener <rguenther@suse.de>
* gcc.dg/torture/fp-uint64-convert-double-1.c: Add
-fexcess-precision=standard.
Here is an implementation of DR2351 - void{} - where void{} after
pack expansion is considered valid and the same thing as void().
For templates, if CONSTRUCTOR_NELTS is 0, the CONSTRUCTOR is not dependent
and we can return void_node right away, if it is dependent and contains
only packs, then it is potentially zero element and so we need to build
CONSTRUCTOR_IS_DEPENDENT CONSTRUCTOR, while if it contains any non-pack
elts, we can diagnose it right away.
2021-10-29 Jakub Jelinek <jakub@redhat.com>
PR c++/102820
* semantics.c (maybe_zero_constructor_nelts): New function.
(finish_compound_literal): Implement DR2351 - void{}.
If type is cv void and compound_literal has no elements, return
void_node. If type is cv void and compound_literal might have no
elements after expansion, handle it like other dependent compound
literals.
* g++.dg/cpp0x/dr2351.C: New test.
If the second operand of __builtin_shuffle is const vector 0, and with
specific mask, it can be optimized to vspltisw+xxpermdi instead of lxv.
gcc/ChangeLog:
PR target/102868
* config/rs6000/rs6000.c (altivec_expand_vec_perm_const): Add
patterns match and emit for VSX xxpermdi.
gcc/testsuite/ChangeLog:
PR target/102868
* gcc.target/powerpc/pr102868.c: New test.
Same thing as the relational change. Walk any equivalences that have
been registered on the path, and remove the name being killed. The
only reason we had added the equivalence with itself earlier is so we
wouldn't search any further in the equivalency list. So if we are
removing all references to it, then we no longer need to add a "kill"
record.
Will push pending tests on x86-64 Linux.
Co-authored-by: Andrew MacLeod <amacleod@redhat.com>
gcc/ChangeLog:
* value-relation.cc (path_oracle::killing_def): Walk the
equivalency list and remove SSA from any equivalencies.
Tamar Christina
David Malcolm
Jonathan Wakely
Aldy Hernandez
Andrew MacLeod
Dan Li
Xionghu Luo
Martin Liska
GCC Administrator
Bernhard Reutner-Fischer
Jan Hubicka
Iain Buclaw
Tobias Burnus
Harald Anlauf
Steve Kargl
Alexandre Oliva
Jakub Jelinek
Jeff Law
Richard Biener
liuhongt