cfun->machine->max_used_stack_alignment is used to decide how stack frame
should be aligned. This is independent of any psABIs nor 32-bit vs 64-bit.
It is always safe to compute max_used_stack_alignment. We compute it only
if 128-bit aligned load/store may be generated on misaligned stack slot
which will lead to segfault.
gcc/
PR target/86386
* config/i386/i386.c (ix86_finalize_stack_frame_flags): Set
cfun->machine->max_used_stack_alignment if needed.
gcc/testsuite/
PR target/86386
* gcc.target/i386/pr86386.c: New file.
From-SVN: r263317
Using libgomp configure option --with-cuda-driver=<dir> we can indicate what
cuda driver to use to build the libgomp nvptx plugin. Without such an option,
the system cuda driver is used, if available. If not availabe, a dlopen
interface is used instead.
However, when we use --without-cuda-driver (or the equivalent
--with-cuda-driver=no) the system cuda driver is still used if available.
This patch fixes that, making sure that --without-cuda-driver selects the dlopen
interface.
Build on x86_64 with nvptx accelerator and tested libgomp testsuite, with and
without option --without-cuda-driver.
2018-08-04 Tom de Vries <tdevries@suse.de>
* plugin/configfrag.ac: For --without-cuda-driver, set
CUDA_DRIVER_INCLUDE and CUDA_DRIVER_LIB to no. Handle
CUDA_DRIVER_INCLUDE == no and CUDA_DRIVER_LIB == no.
* configure: Regenerate.
From-SVN: r263310
gcc/cp/ChangeLog:
PR c++/85523
* decl.c: Include "gcc-rich-location.h".
(add_return_star_this_fixit): New function.
(finish_function): When warning about missing return statements in
functions returning non-void, add a "return *this;" fix-it hint for
assignment operators.
gcc/testsuite/ChangeLog:
PR c++/85523
* g++.dg/pr85523.C: New test.
Co-Authored-By: Jonathan Wakely <jwakely@redhat.com>
From-SVN: r263298
If a struct contains an anonymous union and both have a field with the
same name, detect_field_duplicates_hash() will replace one of them
with NULL. If compilation doesn't stop immediately, it may later call
lookup_field() on the union, which falsely assumes the union's
LANG_SPECIFIC array is sorted, and may loop indefinitely because of
this.
2018-08-03 Bogdan Harjoc <harjoc@gmail.com>
PR c/86690
gcc/c:
* c-typeck.c (lookup_field): Do not use TYPE_LANG_SPECIFIC after
errors.
gcc/testsuite:
* gcc.dg/union-duplicate-field.c: New test.
From-SVN: r263294
This partially reverts r262482, at it broke canadian builds.
2018-08-03 Pierre-Marie de Rodat <derodat@adacore.com>
gcc/ada/
Reverts
2018-07-06 Jim Wilson <jimw@sifive.com>
* Make-generated.in (treeprs.ads): Use $(GNATMAKE) instead of gnatmake.
(einfo.h, sinfo.h, stamp-snames, stamp-nmake): Likewise.
* gcc-interface/Makefile.in (xoscons): Likewise.
From-SVN: r263291
We couldn't vectorise:
for (int j = 0; j < n; ++j)
{
for (int i = 0; i < 16; ++i)
a[i] = (b[i] + c[i]) >> 1;
a += step;
b += step;
c += step;
}
at -O3 because cunrolli unrolled the inner loop and SLP couldn't handle
AVG_FLOOR patterns (see also PR86504). The problem was some overly
strict checking of pattern statements compared to normal statements
in vect_get_and_check_slp_defs:
switch (gimple_code (def_stmt))
{
case GIMPLE_PHI:
case GIMPLE_ASSIGN:
break;
default:
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"unsupported defining stmt:\n");
return -1;
}
The easy fix would have been to add GIMPLE_CALL to the switch,
but I don't think the switch is doing anything useful. We only create
pattern statements that the rest of the vectoriser can handle, and the
other checks in this function and elsewhere check whether SLP is possible.
I'm also not sure why:
if (!first && !oprnd_info->first_pattern
/* Allow different pattern state for the defs of the
first stmt in reduction chains. */
&& (oprnd_info->first_dt != vect_reduction_def
is necessary. All that should matter is that the statements in the
node are "similar enough". It turned out to be quite hard to find a
convincing example that used a mixture of pattern and non-pattern
statements, so bb-slp-pow-1.c is the best I could come up with.
But it does show that the combination of "xi * xi" statements and
"pow (xj, 2) -> xj * xj" patterns are handled correctly.
The patch therefore just removes the whole if block.
The loop also needed commutative swapping to be extended to at least
AVG_FLOOR.
This gives +3.9% on 525.x264_r at -O3.
2018-08-03 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* internal-fn.h (first_commutative_argument): Declare.
* internal-fn.c (first_commutative_argument): New function.
* tree-vect-slp.c (vect_get_and_check_slp_defs): Remove extra
restrictions for pattern statements. Use first_commutative_argument
to look for commutative operands in calls to internal functions.
gcc/testsuite/
* gcc.dg/vect/bb-slp-over-widen-1.c: Expect AVG_FLOOR to be used
on vect_avg_qi targets.
* gcc.dg/vect/bb-slp-over-widen-2.c: Likewise.
* gcc.dg/vect/bb-slp-pow-1.c: New test.
* gcc.dg/vect/vect-avg-15.c: Likewise.
From-SVN: r263290
* src/c++11/system_error.cc
(system_error_category::default_error_condition): Add workaround for
ENOTEMPTY and EEXIST having the same value on AIX.
* testsuite/19_diagnostics/error_category/system_category.cc: Add
extra testcases for EDOM, EILSEQ, ERANGE, EEXIST and ENOTEMPTY.
From-SVN: r263289
If a target does not support exceptions, it can indicate this by returning
UI_NONE in TARGET_EXCEPT_UNWIND_INFO. Currently the compiler still emits
exception tables for such a target.
This patch makes sure that no exception tables are emitted if the target does
not support exceptions. This allows us to remove a workaround in
TARGET_ASM_BYTE_OP in the nvptx port.
Build on x86_64 with nvptx accelerator, and tested libgomp.
Build and reg-tested on x86_64.
2018-08-03 Tom de Vries <tdevries@suse.de>
* common/config/nvptx/nvptx-common.c (nvptx_except_unwind_info): Return
UI_NONE.
* config/nvptx/nvptx.c (TARGET_ASM_BYTE_OP): Remove define.
* except.c (output_function_exception_table): Do early exit if
targetm_common.except_unwind_info (&global_options) == UI_NONE.
From-SVN: r263287
There was a typo in the pipeline description where DUP was assigned to
the vector pipes for quad mode ops when it really only uses the VTOG
pipes. Fixing this does not show any noticeable difference in
performance (there's a very small bump of 1.7% in x264 but that's
about it) in my tests but is the more precise description of operations
for falkor.
* config/aarch64/falkor.md (falkor_am_1_vxvy_vxvy): Move
neon_dup_q to...
(falkor_am_1_gtov_gtov): ... a new insn reservation.
From-SVN: r263285
We were rather sloppy about handling the ownership of prefixes for
pretty_printer, and this lead to a memory leak for any time a
diagnostic_show_locus call emits multiple line spans.
This showed up in "make selftest-valgrind" as:
3,976 bytes in 28 blocks are definitely lost in loss record 632 of 669
at 0x4A0645D: malloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
by 0x1F08227: xmalloc (xmalloc.c:147)
by 0x1F083E6: xvasprintf (xvasprintf.c:58)
by 0x1E7EC7D: build_message_string(char const*, ...) (diagnostic.c:78)
by 0x1E7F438: diagnostic_get_location_text(diagnostic_context*, expanded_location) (diagnostic.c:328)
by 0x1E7FD54: default_diagnostic_start_span_fn(diagnostic_context*, expanded_location) (diagnostic.c:626)
by 0x1EB3508: selftest::test_diagnostic_context::start_span_cb(diagnostic_context*, expanded_location) (selftest-diagnostic.c:57)
by 0x1E89215: diagnostic_show_locus(diagnostic_context*, rich_location*, diagnostic_t) (diagnostic-show-locus.c:1992)
by 0x1E8ECAD: selftest::test_fixit_insert_containing_newline_2(selftest::line_table_case const&) (diagnostic-show-locus.c:3044)
by 0x1EB0606: selftest::for_each_line_table_case(void (*)(selftest::line_table_case const&)) (input.c:3525)
by 0x1E8F3F5: selftest::diagnostic_show_locus_c_tests() (diagnostic-show-locus.c:3164)
by 0x1E010BF: selftest::run_tests() (selftest-run-tests.c:88)
4,004 bytes in 28 blocks are definitely lost in loss record 633 of 669
at 0x4A0645D: malloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
by 0x1F08227: xmalloc (xmalloc.c:147)
by 0x1F083E6: xvasprintf (xvasprintf.c:58)
by 0x1E7EC7D: build_message_string(char const*, ...) (diagnostic.c:78)
by 0x1E7F438: diagnostic_get_location_text(diagnostic_context*, expanded_location) (diagnostic.c:328)
by 0x1E7FD54: default_diagnostic_start_span_fn(diagnostic_context*, expanded_location) (diagnostic.c:626)
by 0x1EB3508: selftest::test_diagnostic_context::start_span_cb(diagnostic_context*, expanded_location) (selftest-diagnostic.c:57)
by 0x1E89215: diagnostic_show_locus(diagnostic_context*, rich_location*, diagnostic_t) (diagnostic-show-locus.c:1992)
by 0x1E8B373: selftest::test_diagnostic_show_locus_fixit_lines(selftest::line_table_case const&) (diagnostic-show-locus.c:2500)
by 0x1EB0606: selftest::for_each_line_table_case(void (*)(selftest::line_table_case const&)) (input.c:3525)
by 0x1E8F3B9: selftest::diagnostic_show_locus_c_tests() (diagnostic-show-locus.c:3159)
by 0x1E010BF: selftest::run_tests() (selftest-run-tests.c:88)
This patch fixes the leaks by ensuring that the pretty_printer "owns"
the prefix if it's non-NULL, freeing it in the dtor and in pp_set_prefix.
gcc/cp/ChangeLog:
* error.c (cxx_print_error_function): Duplicate "file" before
passing it to pp_set_prefix.
(cp_print_error_function): Use pp_take_prefix when saving the
existing prefix.
gcc/ChangeLog:
* diagnostic-show-locus.c (diagnostic_show_locus): Use
pp_take_prefix when saving the existing prefix.
* diagnostic.c (diagnostic_append_note): Likewise.
* langhooks.c (lhd_print_error_function): Likewise.
* pretty-print.c (pp_set_prefix): Drop the "const" from "prefix"
param's type. Free the existing prefix.
(pp_take_prefix): New function.
(pretty_printer::pretty_printer): Drop the prefix parameter.
Rename the length parameter to match the comment.
(pretty_printer::~pretty_printer): Free the prefix.
* pretty-print.h (pretty_printer::pretty_printer): Drop the prefix
parameter.
(struct pretty_printer): Drop the "const" from "prefix" field's
type and clarify memory management.
(pp_set_prefix): Drop the "const" from the 2nd param.
(pp_take_prefix): New decl.
From-SVN: r263275
2018-07-31 Aaron Sawdey <acsawdey@linux.ibm.com>
* config/rs6000/rs6000-string.c (select_block_compare_mode): Move test
for word_mode_ok here instead of passing as argument.
(expand_block_compare): Change select_block_compare_mode() call.
(expand_strncmp_gpr_sequence): New function.
(expand_strn_compare): Make use of expand_strncmp_gpr_sequence.
From-SVN: r263273
The nvptx port can't support exceptions using sjlj, because ptx does not
support sjlj. However, default_except_unwind_info still returns UI_SJLJ, even
even if we configure with --disable-sjlj-exceptions, because UI_SJLJ is the
fallback option.
The reason default_except_unwind_info doesn't return UI_DWARF2 is because
DWARF2_UNWIND_INFO is not defined in defaults.h, because
INCOMING_RETURN_ADDR_RTX is not defined, because there's no ptx equivalent.
Testcase libgomp.c++/for-15.C currently doesn't compile unless fno-exceptions
is added because:
- it tries to generate sjlj exception handling code, and
- it tries to generate exception tables using label-addressed .byte sequence.
Ptx doesn't support generating random data at a label, nor being able to
load/write data relative to a label.
This patch fixes the first problem by using UI_TARGET for nvptx.
The second problem is worked around by generating all .byte sequences commented
out. It would be better to have a narrower workaround, and define
TARGET_ASM_BYTE_OP to "error: .byte unsupported " or some such.
This patch does not enable exceptions for nvptx, it merely allows c++ programs
to run correctly if they do no use exception handling.
Build and reg-tested on x86_64 with nvptx accelerator.
2018-08-02 Tom de Vries <tdevries@suse.de>
PR target/86660
* common/config/nvptx/nvptx-common.c (nvptx_except_unwind_info): New
function. Return UI_TARGET unconditionally.
(TARGET_EXCEPT_UNWIND_INFO): Redefine to nvptx_except_unwind_info.
* config/nvptx/nvptx.c (TARGET_ASM_BYTE_OP): Emit commented out '.byte'.
* testsuite/libgomp.oacc-c++/routine-1-auto.C: Remove -fno-exceptions.
* testsuite/libgomp.oacc-c++/routine-1-template-auto.C: Same.
* testsuite/libgomp.oacc-c++/routine-1-template-trailing-return-type.C:
Same.
* testsuite/libgomp.oacc-c++/routine-1-template.C: Same.
* testsuite/libgomp.oacc-c++/routine-1-trailing-return-type.C: Same.
* testsuite/libgomp.oacc-c-c++-common/routine-1.c: Same.
From-SVN: r263265
2018-08-02 Richard Biener <rguenther@suse.de>
PR c++/86763
* class.c (layout_class_type): Copy TYPE_TYPELESS_STORAGE
to the CLASSTYPE_AS_BASE.
* g++.dg/torture/pr86763.C: New testcase.
From-SVN: r263261
2018-08-02 Richard Biener <rguenther@suse.de>
PR tree-optimization/86816
* tree-ssa-tail-merge.c (tail_merge_valueize): New function
which checks for value availability before querying it.
(gvn_uses_equal): Use it.
(same_succ_hash): Likewise.
(gimple_equal_p): Likewise.
* g++.dg/torture/pr86816.C: New testcase.
From-SVN: r263257
Given a pattern like:
(define_insn "aarch64_frecpe<mode>" ...)
the SVE ACLE implementation wants to generate the pattern for a
particular (non-constant) mode. This patch automatically generates
helpers to do that, specifically:
// Return CODE_FOR_nothing on failure.
insn_code maybe_code_for_aarch64_frecpe (machine_mode);
// Assert that the code exists.
insn_code code_for_aarch64_frecpe (machine_mode);
// Return NULL_RTX on failure.
rtx maybe_gen_aarch64_frecpe (machine_mode, rtx, rtx);
// Assert that generation succeeds.
rtx gen_aarch64_frecpe (machine_mode, rtx, rtx);
Many patterns don't have sensible names when all <...>s are removed.
E.g. "<optab><mode>2" would give a base name "2". The new functions
therefore require explicit opt-in, which should also help to reduce
code bloat.
The (arbitrary) opt-in syntax I went for was to prefix the pattern
name with '@', similarly to the existing '*' marker.
The patch also makes config/aarch64 use the new routines in cases where
they obviously apply. This was mostly straight-forward, but it seemed
odd that we defined:
aarch64_reload_movcp<...><P:mode>
but then only used it with DImode, never SImode. If we should be
using Pmode instead of DImode, then that's a simple change,
but should probably be a separate patch.
2018-08-02 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* doc/md.texi: Expand the documentation of instruction names
to mention port-local uses. Document '@' in pattern names.
* read-md.h (overloaded_instance, overloaded_name): New structs.
(mapping): Declare.
(md_reader::handle_overloaded_name): New member function.
(md_reader::get_overloads): Likewise.
(md_reader::m_first_overload): New member variable.
(md_reader::m_next_overload_ptr): Likewise.
(md_reader::m_overloads_htab): Likewise.
* read-md.c (md_reader::md_reader): Initialize m_first_overload,
m_next_overload_ptr and m_overloads_htab.
* read-rtl.c (iterator_group): Add "type" and "get_c_token" fields.
(get_mode_token, get_code_token, get_int_token): New functions.
(map_attr_string): Add an optional argument that passes back
the associated iterator.
(overloaded_name_hash, overloaded_name_eq_p, named_rtx_p):
(md_reader::handle_overloaded_name, add_overload_instance): New
functions.
(apply_iterators): Handle '@' names. Report an error if '@'
is used without iterators.
(initialize_iterators): Initialize the new iterator_group fields.
* genopinit.c (handle_overloaded_code_for)
(handle_overloaded_gen): New functions.
(main): Use them to print declarations of maybe_code_for_* and
maybe_gen_* functions, and inline definitions of code_for_* and gen_*.
* genemit.c (print_overload_arguments, print_overload_test)
(handle_overloaded_code_for, handle_overloaded_gen): New functions.
(main): Use it to print definitions of maybe_code_for_* and
maybe_gen_* functions.
* config/aarch64/aarch64.c (aarch64_split_128bit_move): Use
gen_aarch64_mov{low,high}_di and gen_aarch64_movdi_{low,high}
instead of explicit mode checks.
(aarch64_split_simd_combine): Likewise gen_aarch64_simd_combine.
(aarch64_split_simd_move): Likewise gen_aarch64_split_simd_mov.
(aarch64_emit_load_exclusive): Likewise gen_aarch64_load_exclusive.
(aarch64_emit_store_exclusive): Likewise gen_aarch64_store_exclusive.
(aarch64_expand_compare_and_swap): Likewise
gen_aarch64_compare_and_swap and gen_aarch64_compare_and_swap_lse
(aarch64_gen_atomic_cas): Likewise gen_aarch64_atomic_cas.
(aarch64_emit_atomic_swap): Likewise gen_aarch64_atomic_swp.
(aarch64_constant_pool_reload_icode): Delete.
(aarch64_secondary_reload): Use code_for_aarch64_reload_movcp
instead of aarch64_constant_pool_reload_icode. Use
code_for_aarch64_reload_mov instead of explicit mode checks.
(rsqrte_type, get_rsqrte_type, rsqrts_type, get_rsqrts_type): Delete.
(aarch64_emit_approx_sqrt): Use gen_aarch64_rsqrte instead of
get_rsqrte_type and gen_aarch64_rsqrts instead of gen_rqrts_type.
(recpe_type, get_recpe_type, recps_type, get_recps_type): Delete.
(aarch64_emit_approx_div): Use gen_aarch64_frecpe instead of
get_recpe_type and gen_aarch64_frecps instead of get_recps_type.
(aarch64_atomic_load_op_code): Delete.
(aarch64_emit_atomic_load_op): Likewise.
(aarch64_gen_atomic_ldop): Use UNSPECV_ATOMIC_* instead of
aarch64_atomic_load_op_code. Use gen_aarch64_atomic_load
instead of aarch64_emit_atomic_load_op.
* config/aarch64/aarch64.md (aarch64_reload_movcp<GPF_TF:mode><P:mode>)
(aarch64_reload_movcp<VALL:mode><P:mode>, aarch64_reload_mov<mode>)
(aarch64_movdi_<mode>low, aarch64_movdi_<mode>high)
(aarch64_mov<mode>high_di, aarch64_mov<mode>low_di): Add a '@'
character before the pattern name.
* config/aarch64/aarch64-simd.md (aarch64_split_simd_mov<mode>)
(aarch64_rsqrte<mode>, aarch64_rsqrts<mode>)
(aarch64_simd_combine<mode>, aarch64_frecpe<mode>)
(aarch64_frecps<mode>): Likewise.
* config/aarch64/atomics.md (atomic_compare_and_swap<mode>)
(aarch64_compare_and_swap<mode>, aarch64_compare_and_swap<mode>_lse)
(aarch64_load_exclusive<mode>, aarch64_store_exclusive<mode>)
(aarch64_atomic_swp<mode>, aarch64_atomic_cas<mode>)
(aarch64_atomic_load<atomic_ldop><mode>): Likewise.
From-SVN: r263251
aarch64_float_const_representable_p was still returning false for
HFmode, so we wouldn't use 16-bit FMOV immediate. E.g. before the
patch:
__fp16 foo (void) { return 0x1.1p-3; }
gave:
mov w0, 12352
fmov h0, w0
with -march=armv8.2-a+fp16, whereas now it gives:
fmov h0, 1.328125e-1
2018-08-02 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/aarch64.c (aarch64_float_const_representable_p):
Allow HFmode constants if TARGET_FP_F16INST.
gcc/testsuite/
* gcc.target/aarch64/f16_mov_immediate_1.c: Expect fmov immediate
to be used.
* gcc.target/aarch64/f16_mov_immediate_2.c: Likewise.
* gcc.target/aarch64/f16_mov_immediate_3.c: Force +nofp16.
* gcc.target/aarch64/sve/single_1.c: Except fmov immediate to be used
for .h.
* gcc.target/aarch64/sve/single_2.c: Likewise.
* gcc.target/aarch64/sve/single_3.c: Likewise.
* gcc.target/aarch64/sve/single_4.c: Likewise.
From-SVN: r263250
