During ICE reporting, sometimes rtl_dump_bb is called on partially
initialized basic blocks. This produces another ICE, obscuring the
original problem.
Fix by checking that that basic blocks are initialized before touching
their bb_infos.
gcc/ChangeLog:
2020-11-25 Ilya Leoshkevich <iii@linux.ibm.com>
* cfgrtl.c (rtl_bb_info_initialized_p): New function.
(rtl_dump_bb): Use rtl_bb_info_initialized_p before accessing bb
insns.
This recently submitted test was found to fail on some Cortex-M
targets. This was because codegen on these CPUs would emit a ldr
instead of a movw/movt pair, resulting in an overall smaller test
(i.e. the branch wasn't as far) and the behaviour being tested
for not being triggered.
This commit doubles the size of the test to account for this.
gcc/testsuite/ChangeLog:
* gcc.target/arm/pr91816.c: New test.
As discussed in the PR, e.g. on x86_64 (both -m32 and -m64) there is no
double-word modulo and so we expand it to a __{,u}mod[dt]i3 call.
For certain constant divisors we can do better. E.g. consider
32-bit word-size, 0x100000000ULL % 3 == 1, so we can use partly the Hacker's
delight modulo by summing digits approach and optimize
unsigned long long foo (unsigned long long x) { return x % 3; }
as
unsigned long long foo (unsigned long long x) {
unsigned int sum, carry;
carry = __builtin_add_overflow ((unsigned int) x, (unsigned int) (x >> 32), &sum);
sum += carry;
return sum % 3;
}
Similarly, 0x10000000ULL % 5 == 1 (note, 1 << 28), so
unsigned long long bar (unsigned long long x) { return x % 5; }
as
unsigned long long bar (unsigned long long x) {
unsigned int sum = x & ((1 << 28) - 1);
sum += (x >> 28) & ((1 << 28) - 1);
sum += (x >> 56);
return sum % 5;
}
etc.
And we can do also signed modulo,
long long baz (long long x) { return x % 5; }
as
long long baz (long long x) {
unsigned int sum = x & ((1 << 28) - 1);
sum += ((unsigned long long) x >> 28) & ((1 << 28) - 1);
sum += ((unsigned long long) x >> 56);
/* Sum adjustment for negative x. */
sum += (x >> 63) & 3;
unsigned int rem = sum % 5;
/* And finally adjust it to the right interval for negative values. */
return (int) (rem + ((x >> 63) & -4));
}
2020-11-30 Jakub Jelinek <jakub@redhat.com>
PR rtl-optimization/97459
* internal-fn.h (expand_addsub_overflow): Declare.
* internal-fn.c (expand_addsub_overflow): No longer static.
* optabs.c (expand_doubleword_mod): New function.
(expand_binop): Optimize double-word mod with constant divisor.
* gcc.dg/pr97459-1.c: New test.
* gcc.dg/pr97459-2.c: New test.
- Define MULTILIB_DEFAULTS can reduce the total number of multilib if
the default arch and ABI are listed in the multilib config.
- This also simplify the implementation of --with-multilib-list.
gcc/ChangeLog:
* config.gcc (riscv*-*-*): Add TARGET_RISCV_DEFAULT_ABI and
TARGET_RISCV_DEFAULT_ARCH to tm_defines.
Remove including riscv/withmultilib.h for --with-multilib-list.
* config/riscv/riscv.h (STRINGIZING): New.
(__STRINGIZING): Ditto.
(MULTILIB_DEFAULTS): Ditto.
* config/riscv/withmultilib.h: Remove.
This issue is found when we try to always define MULTILIB_DEFAULTS for -march
and -mabi for RISC-V back-end, however `-print-multi-lib` will skip
multi-lib setting if match any one of flag in MULTILIB_DEFAULTS, even
some options are specified in the option list with '!'.
e.g.
We have default march=rv32i and mabi=ilp32.
And we have following multi-lib set:
". !march=rv32i !march=rv32im !march=rv32imafc !mabi=ilp32 !mabi=ilp32f;"
"rv32i/ilp32 march=rv32i !march=rv32im !march=rv32imafc mabi=ilp32 !mabi=ilp32f;"
"rv32im/ilp32 !march=rv32i march=rv32im !march=rv32imafc mabi=ilp32 !mabi=ilp32f;"
"rv32imafc/ilp32f !march=rv32i !march=rv32ic !march=rv32im march=rv32imafc !mabi=ilp32 mabi=ilp32f;"
`-print-multi-lib` willl show `.` and `rv32imafc/ilp32f` only, because
the mabi=ilp32 is matched, however there is `!march=rv32i` in `rv32im/ilp32`, so
`rv32im/ilp32` should keep, because it reject march=rv32i.
Note: This can be reproduced via following configure options with patch [1]:
gcc/configure --target=riscv64-elf --with-arch=rv32i --with-abi=ilp32 \
--with-multilib-generator="rv32i-ilp32--;rv32im-ilp32--;rv32imafc-ilp32f--"
[1] https://gcc.gnu.org/pipermail/gcc-patches/2020-November/559707.html
gcc/ChangeLog:
* gcc.c (print_multilib_info): Check default arguments not
appeared in multi-lib option list with '!'
When substituting a parameter variable of type character, the character
length was reset to 1. Fix this by copying the length.
gcc/fortran/ChangeLog:
* expr.c (simplify_parameter_variable): Fix up character length
after copying an array-valued expression.
gcc/testsuite/ChangeLog:
* gfortran.dg/pr98017.f90: New test.
PR middle-end/92936 - missing warning on a past-the-end store to a PHI
PR middle-end/92940 - incorrect offset and size in -Wstringop-overflow for out-of-bounds store into VLA and two offset ranges
PR middle-end/89428 - missing -Wstringop-overflow on a PHI with variable offset
gcc/ChangeLog:
PR middle-end/92936
PR middle-end/92940
PR middle-end/89428
* builtins.c (access_ref::access_ref): Initialize member.
(access_ref::phi): New function.
(access_ref::get_ref): New function.
(access_ref::add_offset): Remove duplicate assignment.
(maybe_warn_for_bound): Add "maybe" kind of warning messages.
(warn_for_access): Same.
(inform_access): Rename...
(access_ref::inform_access): ...to this. Print PHI arguments. Format
offset the same as size and simplify. Improve printing of allocation
functions and VLAs.
(check_access): Adjust to the above.
(gimple_parm_array_size): Change argument.
(handle_min_max_size): New function.
* builtins.h (class ssa_name_limit_t): Move class here from
tree-ssa-strlen.c.
(struct access_ref): Declare new members.
(gimple_parm_array_size): Change argument.
* tree-ssa-strlen.c (maybe_warn_overflow): Use access_ref and simplify.
(handle_builtin_memcpy): Correct argument passed to maybe_warn_overflow.
(handle_builtin_memset): Same.
(class ssa_name_limit_t): Move class to builtins.{h,c}.
gcc/testsuite/ChangeLog:
PR middle-end/92936
PR middle-end/92940
PR middle-end/89428
* c-c++-common/Wstringop-overflow-2.c: Adjust text of expected
informational notes.
* g++.dg/warn/Wstringop-overflow-3.C: Same.
* g++.dg/warn/Wplacement-new-size.C: Remove a test for a no longer
issued warning.
* gcc.dg/Warray-bounds-43.c: Removed unused declarations.
* gcc.dg/Wstringop-overflow-11.c: Remove xfails.
* gcc.dg/Wstringop-overflow-12.c: Same.
* gcc.dg/Wstringop-overflow-17.c: Adjust text of expected messages.
* gcc.dg/Wstringop-overflow-27.c: Same. Remove xfails.
* gcc.dg/Wstringop-overflow-28.c: Adjust text of expected messages.
* gcc.dg/Wstringop-overflow-29.c: Same.
* gcc.dg/Wstringop-overflow-37.c: Same.
* gcc.dg/Wstringop-overflow-46.c: Same.
* gcc.dg/Wstringop-overflow-47.c: Same.
* gcc.dg/Wstringop-overflow-54.c: Same.
* gcc.dg/warn-strnlen-no-nul.c: Add expected warning.
* gcc.dg/Wstringop-overflow-7.c: New test.
* gcc.dg/Wstringop-overflow-58.c: New test.
* gcc.dg/Wstringop-overflow-59.c: New test.
* gcc.dg/Wstringop-overflow-60.c: New test.
* gcc.dg/Wstringop-overflow-61.c: New test.
* gcc.dg/Wstringop-overflow-62.c: New test.
* gcc.dg/Wstringop-overflow-63.c: New test.
* gcc.dg/Wstringop-overflow-64.c: New test.
gcc/ChangeLog:
* config.gcc (*-*-darwin*): Set d_target_objs and target_has_targetdm.
* config/elfos.h (TARGET_D_MINFO_SECTION): New macro.
(TARGET_D_MINFO_START_NAME): New macro.
(TARGET_D_MINFO_END_NAME): New macro.
* config/t-darwin: Add darwin-d.o.
* doc/tm.texi: Regenerate.
* doc/tm.texi.in (D language and ABI): Add @hook for
TARGET_D_MINFO_SECTION, TARGET_D_MINFO_START_NAME, and
TARGET_D_MINFO_END_NAME.
* config/darwin-d.c: New file.
gcc/d/ChangeLog:
* d-target.def (d_minfo_section): New hook.
(d_minfo_start_name): New hook.
(d_minfo_end_name): New hook.
* modules.cc: Include d-target.h.
(register_moduleinfo): Update to use new targetdm hooks.
The bootstrap has been confirmed to be succeeding, there's no need to
set it as an unsupported language.
ChangeLog:
PR d/87788
* configure.ac: Don't disable D for *-*-darwin*.
* configure: Regenerate.
2020-11-29 John David Anglin <danglin@gcc.gnu.org>
gcc/ada/
PR ada/97504
* Makefile.rtl (LIBGNAT_TARGET_PAIRS) <hppa*-*-hpux*>: Use wraplf
version of Aux_Long_Long_Float.
* ipa-modref.c (modref_lattice::merge): Do nothing if F is EAF_UNUSED.
(analyze_parms): Detect unused params.
(modref_merge_call_site_flags): Merge correct EAF_UNUSED.
The typedef for d_size_t assumes that the implementation of the
front-end is written in D itself, where size_t can map only to uint32_t
or uint64_t. As that is not the case for the current D front-end, the
typedef should be removed. This would fix the bootstrap on targets where
in C++ size_t is a long, such as darwin i386.
Reviewed-on: https://github.com/dlang/dmd/pull/12008
gcc/d/ChangeLog:
PR d/87788
* dmd/MERGE: Merge upsream dmd 45fa6cfd2.
This fixes an issue with nested structures and adds an Alignment clause
to counter the effect of the Pack aspect.
gcc/c-family/ChangeLog:
* c-ada-spec.c (dump_nested_type) <RECORD_TYPE>: Remove obsolete code.
(dump_ada_structure): Also deal with convention, unchecked union and
bit-field for nested types. In the latter case, print an Alignment
aspect along with the Pack aspect.
The little dance around 4096 that add/sub instructions do on the SPARC
needs to be taken into account for the overflow arithmetic operations.
It cannot be done for unsigned overflow, but it can be done for signed
overflow.
gcc/ChangeLog:
PR target/97939
* config/sparc/predicates.md (arith_double_add_operand): Comment.
* config/sparc/sparc.md (uaddvdi4): Use arith_double_operand.
(addvdi4): Use arith_double_add_operand.
(addsi3): Remove useless attributes.
(addvsi4): Use arith_add_operand.
(*cmp_ccv_plus): Likewise and add second alternative accordingly.
(*cmp_ccxv_plus): Likewise.
(*cmp_ccv_plus_set): Likewise.
(*cmp_ccxv_plus_set): Likewise.
(*cmp_ccv_plus_sltu_set): Likewise.
(usubvdi4): Use arith_double_operand.
(subvdi4): Use arith_double_add_operand.
(subsi3): Remove useless attributes.
(subvsi4): Use arith_add_operand.
(*cmp_ccv_minus): Likewise and add second alternative accordingly.
(*cmp_ccxv_minus): Likewise.
(*cmp_ccv_minus_set): Likewise.
(*cmp_ccxv_minus_set): Likewise.
(*cmp_ccv_minus_sltu_set): Likewise.
(negsi2): Use register_operand.
(unegvsi3): Likewise.
(negvsi3) Likewise.
(*cmp_ccnz_neg): Likewise.
(*cmp_ccxnz_neg): Likewise.
(*cmp_ccnz_neg_set): Likewise.
(*cmp_ccxnz_neg_set): Likewise.
(*cmp_ccc_neg_set): Likewise.
(*cmp_ccxc_neg_set): Likewise.
(*cmp_ccc_neg_sltu_set): Likewise.
(*cmp_ccv_neg): Likewise.
(*cmp_ccxv_neg): Likewise.
(*cmp_ccv_neg_set): Likewise.
(*cmp_ccxv_neg_set): Likewise.
(*cmp_ccv_neg_sltu_set): Likewise.
gcc/testsuite/ChangeLog:
* gcc.target/sparc/overflow-6.c: New test.
This patch works by generating the correct fn spec from the
actual arguments in specific_intrinsic_symbol, and by passing this
down via gfc_get_extern_function_decl to gfc_get_function_type.
CSHIFT remains to be done.
gcc/fortran/ChangeLog:
PR fortran/97454
* trans-decl.c (gfc_get_extern_function_decl): Add argument
fnspec.
* trans-intrinsic.c (MAX_SPEC_ARG): Define.
(intrinsic_fnspec): New function.
(ADD_CHAR): Define.
(specific_intrinsic_symbol): Adjust comment. Pass
fn spec to gfc_get_extern_function_decl.
(gfc_conv_intrinsic_funcall): Add ANY, ALL, NORM2, PRODUCT
and SUM intrnisic. Add FIXME for cshift et al.
* trans-types.c (gfc_get_function_type): Add fnspec argument,
handle it.
* trans-types.h (gfc_get_function_type): Add optinal fnspec
argument.
* trans.h (gfc_get_extern_function_decl): Likewise.
The preprocessor check for overflow (of linenum_type = unsigned int)
when reading the line number in a #line directive is incomplete; it
checks "reg < reg_prev" which doesn't cover all cases where
multiplying by 10 overflowed. Fix this by checking for overflow
before rather than after it occurs (using essentially the same logic
as used by e.g. glibc printf when reading width and precision values
from strings).
Bootstrapped with no regressions for x86_64-pc-linux-gnu.
libcpp/
2020-11-27 Joseph Myers <joseph@codesourcery.com>
PR preprocessor/97602
* directives.c (strtolinenum): Check for overflow before it
occurs. Correct comment.
gcc/testsuite/
2020-11-27 Joseph Myers <joseph@codesourcery.com>
PR preprocessor/97602
* gcc.dg/cpp/line9.c, gcc.dg/cpp/line10.c: New tests.
When libphobos is configured with --enable-cet, this adds extra fields
to the Fiber class to support the ucontext_t fallback implementation.
These fields get omitted when compiling user code unless they also used
`-fversion=CET' to build their project, which resulted in data being
overwritten from within swapcontext().
On reviewing the ucontext_t definitions, it was found that the shadow
stack fields were missing, and the struct size didn't match up on X32.
This has been fixed in upstream druntime and merged down here.
Reviewed-on: https://github.com/dlang/druntime/pull/3293
libphobos/ChangeLog:
PR d/98025
* Makefile.in: Regenerate.
* configure: Regenerate.
* configure.ac (DCFG_ENABLE_CET): Substitute.
* libdruntime/MERGE: Merge upstream druntime 0fe7974c.
* libdruntime/Makefile.in: Regenerate.
* libdruntime/core/thread.d: Import gcc.config.
(class Fiber): Add ucontext_t fields when GNU_Enable_CET is true.
* libdruntime/gcc/config.d.in (GNU_Enable_CET): Define.
* src/Makefile.in: Regenerate.
* testsuite/Makefile.in: Regenerate.
Adds support for FreeBSD/x86 53-bit precision reals, updates bindings
for FreeBSD 12.x, and removes all support code and tests for the
extern(Pascal) calling convention.
Reviewed-on: https://github.com/dlang/druntime/pull/3286https://github.com/dlang/druntime/pull/3287
libphobos/ChangeLog:
* libdruntime/MERGE: Merge upstream druntime d37ef985.
* libdruntime/Makefile.am (DRUNTIME_DSOURCES_FREEBSD): Add
core/sys/freebsd/config.d
* libdruntime/Makefile.in: Regenerate.
For the math intrinsics: cos, fabs, ldexp, rint, rndtol, and sin, new
overloads have been added to the core.math module for matching float and
double types. These have been implemented in the compiler.
A recent change to dump_function_to_file started triggering some
scan-tree-dump tests to FAIL, these have been adjusted as well when
updating the test.
gcc/d/ChangeLog:
* intrinsics.cc (maybe_expand_intrinsic): Handle new intrinsics.
* intrinsics.def (INTRINSIC_COS): Add float and double overloads.
(INTRINSIC_FABS): Likewise.
(INTRINSIC_LDEXP): Likewise.
(INTRINSIC_RINT): Likewise.
(INTRINSIC_RNDTOL): Likewise.
(INTRINSIC_SIN): Likewise.
(INTRINSIC_TOPREC): Adjust signature.
libphobos/ChangeLog:
* libdruntime/MERGE: Merge upstream druntime 5e4492c4.
gcc/testsuite/ChangeLog:
* gdc.dg/intrinsics.d: Adjust patterns in scan-tree-dump.
This introduces two new procs to replace boilerplate in the
effective-target checks.
libstdc++-v3/ChangeLog:
* testsuite/lib/libstdc++.exp (v3_try_preprocess): Define
new proc to preprocess a chunk of code.
(v3_check_preprocessor_condition): Define new proc to test
a preprocessor condition depending on GCC or libstdc++ macros.
(check_v3_target_debug_mode, check_v3_target_normal_mode):
Use v3_try_preprocess.
(check_v3_target_normal_namespace)
(check_v3_target_parallel_mode, check_v3_target_cstdint)
(check_v3_target_cmath, check_v3_target_atomic_builtins)
(check_v3_target_gthreads, check_v3_target_gthreads_timed)
(check_v3_target_sleep, check_v3_target_sched_yield)
(check_v3_target_string_conversions, check_v3_target_swprintf)
(check_v3_target_binary_io, check_v3_target_nprocs): Use
v3_check_preprocessor_condition.
(check_effective_target_cxx11): Likewise.
(check_effective_target_random_device): Likewise.
(check_effective_target_tbb-backend): Likewise.
(check_effective_target_futex): Likewise.
(check_v3_target_little_endian) Call check_effective_target_le.
(check_effective_target_atomic-builtins): New proc to define
new effective-target keyword.
(check_effective_target_gthreads-timed): Likewise.
Also, instead of just examining the first statement of inner 'GIMPLE_BIND' (via
'inner_sequence' being a 'typedef gimple *gimple_seq'), in fact examine all
statements contained therein, which I suppose must've been the intention here.
This "fixes" the testcase 'c-c++-common/goacc/kernels-decompose-ice-2.c' (which
now runs into the same ICE as 'c-c++-common/goacc/kernels-decompose-ice-1.c',
etc.).
gcc/
* omp-oacc-kernels-decompose.cc (flatten_binds): Don't choke on
empty GIMPLE sequence, and examine all statements contained in
inner 'GIMPLE_BIND'.
gcc/testsuite/
* c-c++-common/goacc/kernels-decompose-ice-1.c: Adjust.
* c-c++-common/goacc/kernels-decompose-ice-2.c: Likewise.
It handles the following:
(cherry picked from commit c0c7270cc4 (testsuite changes only))
contrib/ChangeLog:
* gcc-changelog/git_commit.py: Use regex for cherry pick prefix.
* gcc-changelog/test_email.py: Test it.
* gcc-changelog/test_patches.txt: Likewise.
The changes in r11-5314 are broken, because it means we don't use
__gthread_once for the first few initializations, but after the program
becomes multi-threaded we will repeat the initialization, using
__gthread_once once this time. This leads to memory errors.
The use of __is_single_threaded() in locale:🆔:_M_id() is OK, because
the side effects are the same either way.
libstdc++-v3/ChangeLog:
* src/c++98/locale.cc (locale::facet::_S_get_c_locale()):
Revert change to use __is_single_threaded.
* src/c++98/locale_init.cc (locale::_S_initialize()):
Likewise.
The change to make PRE insertion iterate less had a typo in checking
successors RPO state. Fixing this exposes that regular PRE insertion
when facing a value that is the same on all edges inserts an
assignment in place of a PHI node but fails to set up things so that
this insertion is not repeated (it correctly does not return
'new_stuff'). But with the new iteration scheme this causes us
to repeatedly insert such assignment and change AVAIL_OUT over to
the newly inserted expression. The fix is to treat this as PHI
and insert into PHI_GEN, avoiding repetitive insertion.
2020-11-27 Richard Biener <rguenther@suse.de>
PR tree-optimization/98024
* tree-ssa-pre.c (insert): Fix successor RPO order check.
(do_pre_regular_insertion): When inserting an assignment
in place of an all-same-value PHI still record that into
PHI_GEN.
My recent match.pd change required quite a lot of code due to the separate
need to handle INTEGER_CSTs and SSA_NAMEs, and after all, I didn't even
handle one case there, when in x * y / y the x is INTEGER_CST and y is
SSA_NAME.
The following patch allows to simplify it, by allowing non-SSA_NAME argument
to get_range_info, for INTEGER_CSTs it will return VR_RANGE with *min == *max
equal to the constnat, and for non-INTEGER_CST/SSA_NAMEs it will just return
VR_VARYING.
This allows not to simplify just the match.pd, but some other spots too.
2020-11-27 Jakub Jelinek <jakub@redhat.com>
* tree-ssanames.c (get_range_info): Handle INTEGER_CST by returning
VR_RANGE with both *min and *max set to the wide_int value of the
INTEGER_CST. Return VR_VARYING for non-SSA_NAMEs.
* match.pd ((t * 2) / 2) -> t): Handle also @0 being INTEGER_CST.
Simplify by calling get_range_info on everything.
* tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Simplify by calling
get_range_info on everything.
* tree-scalar-evolution.c (iv_can_overflow_p): Likewise.
When playing with __builtin_bit_cast, I have noticed __builtin_clear_padding
ICE on the G class below. The artificial field with D type has offset 0
and size 8 bytes, but the following artificial field with E type has offset
0 and size 0, so it triggers the asserts that we don't move current position
backwards. Fixed by ignoring is_empty_type (TREE_TYPE (field)) fields, all
of their bits are padding which is what is added when skipping over to next
field anyway.
2020-11-27 Jakub Jelinek <jakub@redhat.com>
PR libstdc++/88101
* gimple-fold.c (clear_padding_type): Ignore fields with is_empty_type
types.
* g++.dg/torture/builtin-clear-padding-3.C: New test.
gcc/ChangeLog:
PR c/97880
* omp-expand.c (expand_oacc_collapse_init, expand_oacc_collapse_vars):
Use now passed diff_type.
(expand_oacc_for): Take largest type for diff_type, taking tiling
and collapsing into account.
gcc/testsuite/ChangeLog:
PR c/97880
* gcc.dg/goacc/tile-1.c: New test.
This is a patch that introduces the aarch64-autovec-preference that can
take values from 0 - 4, 0 being the default.
It can be used to override the autovectorisation preferences in the
backend:
0 - use default scheme
1 - only use Advanced SIMD
2 - only use SVE
3 - use Advanced SIMD and SVE, prefer Advanced SIMD in the event of a
tie (as determined by costs)
4 - use Advanced SIMD and SVE, prefer SVE in the event of a tie (as
determined by costs)
It can valuable for experimentation when comparing SVE and Advanced SIMD
autovectorisation strategies.
It achieves this adjusting the order of the interleaved SVE and Advanced
SIMD modes in aarch64_autovectorize_vector_modes.
It also adjusts aarch64_preferred_simd_mode to use the new comparison
function to pick Advanced SIMD or SVE to start with.
Bootstrapped and tested on aarch64-none-linux-gnu.
gcc/
* config/aarch64/aarch64.opt
(-param=aarch64-autovec-preference): Define.
* config/aarch64/aarch64.c (aarch64_override_options_internal):
Set aarch64_sve_compare_costs to 0 when preferring only Advanced
SIMD.
(aarch64_cmp_autovec_modes): Define.
(aarch64_preferred_simd_mode): Adjust to use the above.
(aarch64_autovectorize_vector_modes): Likewise.
* doc/invoke.texi: Document aarch64-autovec-preference param.
gcc/ada/
* libgnat/s-valuef.adb (Integer_To_Fixed): Take into account the
extra digit when scaling up the input.
* libgnat/s-valuer.adb (Scan_Decimal_Digits): Restrict previous
change to fixed-point types.
(Scan_Integral_Digits): Likewise.
