haifa-sched exports an insn_cost function, but it is only used in a
few places and specialised to scheduling. This patch renames it to
insn_sched_cost.
* haifa-sched.c: Rename insn_cost to insn_sched_cost.
* sched-rgn.c: Ditto.
* sel-sched-ir.c: Ditto.
From-SVN: r253072
PR sanitizer/81715
* tree-inline.c (expand_call_inline): Emit clobber stmts for
VAR_DECLs to which addressable non-volatile parameters are mapped
and for id->retvar after the return value assignment. Clear
id->retval and id->retbnd after inlining.
* g++.dg/tree-ssa/pr8781.C (noop): Change argument type from
const predicate to const predicate & to avoid UB.
* g++.dg/opt/pr81715.C: New test.
From-SVN: r253065
2017-09-21 Richard Biener <rguenther@suse.de>
PR tree-optimization/82276
PR tree-optimization/82244
* tree-vrp.c (build_assert_expr_for): Set
SSA_NAME_OCCURS_IN_ABNORMAL_PHI if the variable we assert on
has it set.
(remove_range_assertions): Revert earlier change.
* gcc.dg/torture/pr82276.c: New testcase.
From-SVN: r253062
As described in PR71951, if libgcc is built with -fomit-frame-pointer,
unwinding crashes, for example while doing a backtrace. The underlying
reason is the Dwarf unwinder does not setup the frame pointer register
in the initialization code. When later unwinding a function that uses
the frame pointer, it tries to read FP using _Unwind_GetGR, and this
crashes if has never restored FP. To unwind correctly the first frame
must save and restore FP (it is unwound in a special way so that it
uses SP instead of FP). This is done by adding -fno-omit-frame-pointer.
gcc/
PR target/71951
* config/aarch64/aarch64.h (LIBGCC2_UNWIND_ATTRIBUTE): Define.
From-SVN: r253061
2017-09-21 Richard Biener <rguenther@suse.de>
* graphite-isl-ast-to-gimple.c (graphite_regenerate_ast_isl):
Restore valid IL after code generation errors.
* graphite.c (graphite_transform_loops): Diagnose code
generation issues as MSG_MISSED_OPTIMIZATION and continue
with processing SCOPs.
From-SVN: r253060
This patch replaces various places that previously used:
if (GET_CODE (y) == CONST_INT)
... plus_constant (..., x, [-]INTVAL (y)) ...
else
... gen_rtx_PLUS/MINUS (..., x, y) ...
with single calls to simplify_gen_binary. This allows
them to handle polynomial integers as well as constants.
2017-09-21 Richard Sandiford <richard.sandiford@linaro.org>
Alan Hayward <alan.hayward@arm.com>
David Sherwood <david.sherwood@arm.com>
gcc/
* calls.c (compute_argument_addresses): Use simplify_gen_binary
rather than choosing between plus_constant and gen_rtx_<CODE>.
* expr.c (emit_push_insn): Likewise.
(expand_expr_real_2): Likewise.
Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r253059
This is needed if the step is an unshared constant, like many
(const ...)s are. Without this patch, libgfortran would fail
to build for SVE.
2017-09-21 Richard Sandiford <richard.sandiford@linaro.org>
Alan Hayward <alan.hayward@arm.com>
David Sherwood <david.sherwood@arm.com>
gcc/
* loop-unroll.c (split_iv): Call copy_rtx on the step.
Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r253058
This patch uses tree_fits_uhwi_p to protect a previously unguarded
use of tree_to_uhwi. Previously we would ICE for variable-sized types.
2017-09-20 Richard Sandiford <richard.sandiford@linaro.org>
Alan Hayward <alan.hayward@arm.com>
David Sherwood <david.sherwood@arm.com>
gcc/
* tree.c (find_atomic_core_type): Check tree_fits_uhwi_p before
calling tree_to_uhwi.
Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r253057
bit_value_unop and bit_value_binop require constant values
to be INTEGER_CSTs:
gcc_assert ((rval.lattice_val == CONSTANT
&& TREE_CODE (rval.value) == INTEGER_CST)
|| wi::sext (rval.mask, TYPE_PRECISION (TREE_TYPE (rhs))) == -1);
However, when deciding whether to record a constant value,
the for_bits_p handling in get_value_for_expr used a negative
test for ADDR_EXPR:
else if (is_gimple_min_invariant (expr)
&& (!for_bits_p || TREE_CODE (expr) != ADDR_EXPR))
This patch uses a positive test for INTEGER_CST instead.
Existing tests showed the need for this once polynomial constants
are added.
2017-09-21 Richard Sandiford <richard.sandiford@linaro.org>
Alan Hayward <alan.hayward@arm.com>
David Sherwood <david.sherwood@arm.com>
gcc/
* tree-ssa-ccp.c (get_value_for_expr): Use a positive test for
INTEGER_CST rather than a negative test for ADDR_EXPR.
Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r253056
The BIT_AND_EXPR handling in extract_range_from_binary_expr_1
was using value_range_constant_singleton without first checking
whether the range was a constant. The earlier handling was correctly
guarded:
/* If either input range contains only non-negative values
we can truncate the result range maximum to the respective
maximum of the input range. */
if (int_cst_range0 && tree_int_cst_sgn (vr0.min) >= 0)
wmax = wi::min (wmax, vr0.max, TYPE_SIGN (expr_type));
if (int_cst_range1 && tree_int_cst_sgn (vr1.min) >= 0)
wmax = wi::min (wmax, vr1.max, TYPE_SIGN (expr_type));
so this patch uses the same guards again.
Existing tests showed the need for this once polynomial constants
are added.
2017-09-21 Richard Sandiford <richard.sandiford@linaro.org>
Alan Hayward <alan.hayward@arm.com>
David Sherwood <david.sherwood@arm.com>
gcc/
* tree-vrp.c (extract_range_from_binary_expr_1): Check
int_cst_rangeN before calling value_range_constant_singleton (&vrN).
Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r253054
* testsuite/25_algorithms/clamp/1.cc: Fix order of arguments and
expected results when using predicate defining reverse order.
* testsuite/25_algorithms/clamp/constexpr.cc: Likewise.
From-SVN: r253051
PR target/82260
* config/i386/i386.md (*movqi_internal): Replace (=q,q) alternative
with (=Q,Q), (=R,R) and (=r,r) alternatives, only enable the
latter two for 64-bit, renumber alternatives, for -Os imov (=q,n)
alternative always use QI mode, for -Os imov (=R,R) alternative
always use SI mode, for imov (=Q,Q) or (=r,r) alternatives
ignore -Os.
* gcc.target/i386/pr82260-1.c: New test.
* gcc.target/i386/pr82260-2.c: New test.
From-SVN: r253050
* config/s390/s390.c (MIN_UNROLL_PROBES): Define.
(allocate_stack_space): New function, partially extracted from
s390_emit_prologue.
(s390_emit_prologue): Track offset to most recent stack probe.
Code to allocate space moved into allocate_stack_space.
Dump actions when no stack is allocated.
(s390_prologue_plus_offset): New function.
(s390_emit_stack_probe): Likewise.
* gcc.dg/stack-check-5.c: Add argument for s390.
* lib/target-supports.exp:
(check_effective_target_supports_stack_clash_protection): Enable for
s390/s390x targets.
Co-Authored-By: Jeff Law <law@redhat.com>
From-SVN: r253049
This patch that adds -g to the set of negatable prefixes along with -f,
-m and -W. Besides the mapping from -gno- to negated -g in option_map
and adding g to the [fmW] matches for negatable options, I had to
introduce gno- as an remapping prefix, for the option searching
machinery to backtrack to and recognize as a remapping prefix, instead
of backtracking to -g and stopping at it as if no-* was its Joined
argument. Adding such remapping prefixes to preempt further
backtracking can be accomplished by introducing the prefix as an
Undocumented option with a Joined argument and without Driver, Target,
Common, or any language-specific option. Whenever we match such a fake
options prefix, we abandon further backtracking (it matches, after all),
but find_opt returns the same code it would if it hadn't found any
match, so that we resort to option mapping.
I've arranged for such remapping prefixes to not be considered when
looking for and suggesting a correct spelling for misspelled options.
While testing that, I found a few -W-started options that were not
marked as RejectNegative but should (-Wno-a, is not something we'd like
to suggest ;-) I've also marked as such -g-started options that
it makes no sense to negate, and removed the explicit -gno- ones,
allowing their opposites to be negated.
for gcc/ChangeLog
* common.opt (Wa, Wl, Wp, g, gz=): Add
RejectNegative.
(gno-column-info): Remove.
(gcolumn-info): Drop RejectNegative.
(gno-): New prefix.
(gno-record-gcc-switches): Remove.
(grecord-gcc-switches): Drop RejectNegative.
(gno-split-dwarf): Remove.
(gsplit-dwarf): Drop RejectNegative.
(gno-strict-dwarf): Remove.
(gstrict-dwarf): Drop RejectNegative.
* config/darwin.opt (gfull, gused): Add RejectNegative.
* dwarf2out.c (gen_producer_string): Drop
gno-record-gcc-switches handler.
* optc-gen.awk: Add g to prefixes with negative forms.
* opts-common.c (remapping_prefix_p): New.
(find_opt): Check it.
(generate_canonical_option): Test g prefix.
(option_map): Add -gno- mapping.
(add_misspelling_candidates): Check remapping_prefix_p.
for gcc/ada/ChangeLog
* gcc-interface/lang.opt (gant, gnatO, gnat): Add
RejectNegative.
for gcc/c-family/ChangeLog
* c.opt (gen-decls): Add RejectNegative.
From-SVN: r253047
With -enable-static=no we don't build non-pic objects, but libgotool.a
is built from non-pic objects. Build the packages that go into
libgotool.a in static mode in all cases.
Also ensure that internal test packages are built, since nothing
explicitly depended on them.
Reviewed-on: https://go-review.googlesource.com/65050
From-SVN: r253042
PR libstdc++/79162
* include/bits/basic_string.h [!_GLIBCXX_USE_CXX11_ABI]
(basic_string::_If_sv): Remove from the overload set when the
argument is derived from basic_string.
From-SVN: r253035
* explow.c (compute_stack_clash_protection_loop_data): Use
CONST_INT_P instead of explicit test. Verify object is a
CONST_INT_P before looking at INTVAL.
(anti_adjust_stack_and_probe_stack_clash): Use CONST_INT_P
instead of explicit test.
* gcc.target/i386/stack-check-11.c: Update test and regexp
so that it works for both i?86 and x86_64.
From-SVN: r253034
With the 32-bit SVR4 ABI we don't have a red zone, so we have to restore
the callee-saved registers before we restore the stack pointer.
The previous fix for this PR failed in two ways, for huge frames: first,
we use a negative offset from r11 in that case, so the (mem:BLK 11) access
does no good; second, sched does not handle accesses to mem:BLK correctly
in this case (does not make dependencies).
This patch fixes it by doing a store to (mem:BLK (scratch)) instead.
This means no unrelated (not to stack) loads/stores can be moved over the
stack restore either, but so be it.
PR target/77687
* config/rs6000/rs6000.md (stack_restore_tie): Store to a scratch
address instead of to r1 and r11.
gcc/testsuite/
PR target/77687
* gcc.target/powerpc/pr77687.c: New testcase.
From-SVN: r253033
PR libstdc++/79162
* include/bits/basic_string.h (basic_string::_If_sv): Remove from the
overload set when the argument is derived from basic_string.
* testsuite/21_strings/basic_string/cons/char/moveable2_c++17.cc: New
test.
* testsuite/21_strings/basic_string/cons/wchar_t/moveable2_c++17.cc:
New test.
From-SVN: r253024
PR libstdc++/82262
* include/std/optional (__optional_hash_call_base): Add template
parameter for remove_const_t<_Tp> and use it consistently.
* testsuite/20_util/optional/hash.cc: Test optional<const T>.
From-SVN: r253010
2017-09-20 Richard Biener <rguenther@suse.de>
PR tree-optimization/80213
* graphite-scop-detection.c (trivially_empty_bb_p): Labels
are allowed in empty BBs as well.
(canonicalize_loop_closed_ssa): Also look for other complex
edges.
(scop_detection::get_sese): Include the loop-closed PHI block
in loop SESEs.
(scop_detection::merge_sese): Remove code adding extra blocks.
(scop_detection::region_has_one_loop): Adjust for get_sese changes.
(build_scops): Assert the final returned scop is invalid.
From-SVN: r253008
2017-09-20 Richard Biener <rguenther@suse.de>
PR tree-optimization/82264
* tree-ssa-sccvn.c (vn_phi_eq): Use safe_dyn_cast to check
for GIMPLE_CONDs.
(vn_phi_lookup): Likewise.
(vn_phi_insert): Likewise.
* gcc.dg/torture/pr82264.c: New testcase.
From-SVN: r253005
* dwarf2out.c (tree_add_const_value_attribute): For INTEGER_CST
that fits into uhwi or shwi, add DW_AT_const_value regardless
of early_dwarf without going through RTL, using add_AT_unsigned
or add_AT_int.
* g++.dg/debug/dwarf2/template-params-13.C: New test.
* g++.dg/debug/dwarf2/template-params-5.C: Adjust regexps so that
it doesn't match newlines.
* g++.dg/debug/dwarf2/template-params-3.C: Likewise.
* g++.dg/debug/dwarf2/template-func-params-3.C: Likewise.
* g++.dg/debug/dwarf2/lang-cpp98.C: Likewise.
* g++.dg/debug/dwarf2/template-func-params-2.C: Likewise.
* g++.dg/debug/dwarf2/template-func-params-1.C: Likewise.
* g++.dg/debug/dwarf2/template-func-params-5.C: Likewise.
* g++.dg/debug/dwarf2/template-params-1.C: Likewise.
* g++.dg/debug/dwarf2/template-params-2.C: Likewise.
* g++.dg/debug/dwarf2/lang-cpp14.C: Likewise.
* g++.dg/debug/dwarf2/lang-cpp11.C: Likewise.
From-SVN: r253004