Ian Lance Taylor [Sat, 31 Aug 2019 03:07:48 +0000 (03:07 +0000)]
compiler: check for notinheap struct at each struct field
When generating write barriers, we were only checking for a notinheap
struct at the outermost struct. That mishandled the case of setting a
pointer to a notinheap struct as a field of another struct that is not
notinheap. This caused an invalid write barrier error when building
the 1.13 version of the runtime.
Ian Lance Taylor [Sat, 31 Aug 2019 03:01:15 +0000 (03:01 +0000)]
compiler, runtime: support and use single argument go:linkname
The gc compiler has started permitting go:linkname comments with a
single argument to mean that a function should be externally visible
outside the package. Implement this in the Go frontend.
Change the libgo runtime package to use it, rather than repeating the
name just to export a function.
Remove a couple of unnecessary go:linkname comments on declarations.
Ian Lance Taylor [Sat, 31 Aug 2019 02:56:55 +0000 (02:56 +0000)]
compiler: don't report runtime escapes if we've seen errors
If we get errors during compilation, we skip the escape analysis pass.
If we are compiling the runtime package, we report an error if a bound
method expression escapes. The effect is that if we get an error
while compiling the runtime package, we would report confusing and
meaningless errors about bound method expressions escaping.
Ian Lance Taylor [Sat, 31 Aug 2019 02:55:09 +0000 (02:55 +0000)]
runtime: always build panic32.go
Avoids problems with arm64 ILP32 mode. We might want to handle that
mode better in general, but always building panic32.go is simple and
fixes the build.
Jim Wilson [Fri, 30 Aug 2019 23:32:52 +0000 (23:32 +0000)]
RISC-V: Disable -msave-restore for shared libraries.
This was noticed while trying to test -msave-restore support. The
save/restore routines use the alternate return register t0/x5 which is
clobbered by the PLT header, so we can't use them in shared libraries.
This patch disables -msave-restore when -fpic (and -mplt), and emits a
warning if the user explicitly turned on -msave-restore.
gcc/
* config/riscv/riscv.c (riscv_option_override): If -msave-restore
and -fpic and -mplt then disable -msave-restore and warn.
Ian Lance Taylor [Fri, 30 Aug 2019 21:49:49 +0000 (21:49 +0000)]
compile, runtime: permit anonymous and empty fields in C header
Permit putting structs with anonymous and empty fields in the C header
file runtime.inc that is used to build the C runtime code. This is
required for upcoming 1.13 support, as the m struct has picked up an
anonymous field.
Doing this lets the C header contain all the type descriptor structs,
so start using those in the C code. This cuts the number of copies of
type descriptor definitions from 3 to 2.
Martin Sebor [Fri, 30 Aug 2019 17:49:17 +0000 (17:49 +0000)]
PR middle-end/91599 - GCC does not say where warning is happening
gcc/ChangeLog:
PR middle-end/91599
* tree-ssa-strlen.c (handle_store): Use a fallback location if
the statement doesn't have one.
* gimple-pretty-print.c (percent_G_format): Same.
gcc/testsuite/ChangeLog:
PR middle-end/91599
* gcc.dg/Wstringop-overflow-16.c: New test.
Antony Polukhin [Fri, 30 Aug 2019 16:25:24 +0000 (16:25 +0000)]
Optimize std::to_chars
Bunch of micro optimizations for std::to_chars:
* For base == 8 replacing the lookup in __digits table with arithmetic
computations leads to a same CPU cycles for a loop (exchanges two movzx
with 3 bit ops). However this saves 129 bytes of data and totally avoids
a chance of cache misses on __digits.
* For base == 16 replacing the lookup in __digits table with
arithmetic computations leads to a few additional instructions, but
totally avoids a chance of cache misses on __digits (- ~9 cache misses
for worst case) and saves 513 bytes of const data.
* Replacing __first[pos] and __first[pos - 1] with __first[1] and
__first[0] on final iterations saves ~2% of code size.
* Removing trailing '\0' from arrays of digits allows the linker to
merge the symbols (so that "0123456789abcdefghijklmnopqrstuvwxyz" and
"0123456789abcdef" could share the same address). This improves data
locality and reduces binary sizes.
* Using __detail::__to_chars_len_2 instead of a generic
__detail::__to_chars_len makes the operation O(1) instead of O(N). It
also makes the code two times shorter.
In sum: this significantly reduces the size of a binary (for about 4KBs
only for base-8 conversion), deals with latency (CPU cache misses)
without changing the iterations count and without adding costly
instructions into the loops.
2019-08-30 Antony Polukhin <antoshkka@gmail.com>
* include/std/charconv (__detail::__to_chars_8)
__detail::__to_chars_16): Replace array of precomputed digits with
arithmetic operations to avoid CPU cache misses. Remove zero
termination from array of digits to allow symbol merge with generic
implementation of __detail::__to_chars. Replace final offsets with
constants. Use __detail::__to_chars_len_2 instead of a generic
__detail::__to_chars_len.
(__detail::__to_chars): Remove zero termination from array of digits.
(__detail::__to_chars_2): Leading digit is always '1'.
Eric Botcazou [Fri, 30 Aug 2019 15:32:51 +0000 (15:32 +0000)]
decl.c (maybe_saturate_size): New function.
* gcc-interface/decl.c (maybe_saturate_size): New function.
(gnat_to_gnu_entity): Invoke it on the Esize of types before sending
it for back-annotations.
* gcc-interface/trans.c: Fix typo.
Eric Botcazou [Fri, 30 Aug 2019 15:22:34 +0000 (15:22 +0000)]
ada-tree.h (DECL_FORCED_BY_REF_P): New macro.
* gcc-interface/ada-tree.h (DECL_FORCED_BY_REF_P): New macro.
* gcc-interface/decl.c (gnat_to_gnu_param): Set it on parameters
whose mechanism was forced to by-reference.
* gcc-interface/trans.c (Call_to_gnu): Do not issue a warning about a
misaligned actual parameter if it is based on a CONSTRUCTOR. Remove
obsolete warning for users of Starlet. Issue a warning if a temporary
is make around the call for a parameter with DECL_FORCED_BY_REF_P set.
(addressable_p): Return true for REAL_CST and ADDR_EXPR.
* gcc-interface/gigi.h (aggregate_type_contains_array_p): Declare.
* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Record_Type>: For an
extension, test Has_Record_Rep_Clause instead of Has_Specified_Layout.
(adjust_packed): Return 0 if the type of the field is an aggregate
type that contains (or is) a self-referential array.
(type_has_variable_size): Delete.
* gcc-interface/utils.c (inish_record_type): Constify a variable.
(aggregate_type_contains_array_p): Add parameter SELF_REFERENTIAL.
<RECORD_TYPE>: Pass it in the recursive call.
<ARRAY_TYPE>: If it is true, return true only if the array type is
self-referential.
(create_field_decl): Streamline the setting of the alignment on the
field. Pass false to aggregate_type_contains_array_p.
Jonathan Wakely [Fri, 30 Aug 2019 13:54:49 +0000 (14:54 +0100)]
PR libstdc++/89164 enforce constraints for uninitialized algos
The memmove optimizations for std::uninitialized_copy/fill/_n will
compile even if the type is not copy constructible, because std::copy
doesn't require copy construction to work. But the uninitialized
algorithms do require it.
This adds explicit static assertions to ensure we don't allow ill-formed
initializations.
PR libstdc++/89164
* include/bits/stl_algobase.h (__copy_move): Give descriptive names
to template parameters.
* include/bits/stl_uninitialized.h (uninitialized_copy)
(uninitialized_fill, uninitialized_fill_n): Add static assertions to
diagnose invalid uses.
* testsuite/20_util/specialized_algorithms/uninitialized_copy/1.cc:
Adjust expected error.
* testsuite/20_util/specialized_algorithms/uninitialized_copy/89164.cc:
New test.
* testsuite/20_util/specialized_algorithms/uninitialized_copy_n/
89164.cc: New test.
* testsuite/20_util/specialized_algorithms/uninitialized_fill/89164.cc:
New test.
* testsuite/20_util/specialized_algorithms/uninitialized_fill_n/
89164.cc: New test.
* testsuite/23_containers/vector/cons/89164.cc: New test.
* testsuite/23_containers/vector/cons/89164_c++17.cc: New test.
Eric Botcazou [Fri, 30 Aug 2019 13:50:34 +0000 (13:50 +0000)]
gigi.h (gigi_checking_assert): New macro.
* gcc-interface/gigi.h (gigi_checking_assert): New macro.
* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Modular_Integer_Type>:
Remove redundant test and adjust comments. Minor tweaks.
* gcc-interface/trans.c (Call_to_gnu): Do not generate range checks,
instead assert that the Do_Range_Check flag is not set. Adjust call
to convert_with_check.
(gnat_to_gnu): Likewise.
(assoc_to_constructor): Likewise.
(pos_to_constructor): Likewise. Remove GNAT_COMPONENT_TYPE parameter.
(emit_range_check): Delete.
(convert_with_check): Remove RANGE_P parameter and adjust. Do a single
overflow check for modular types.
* config/arm/arm.md (unaligned_loaddi,
unaligned_storedi): New unspec insn patterns.
* config/arm/neon.md (unaligned_storev8qi): Likewise.
* config/arm/arm.c (gen_cpymem_ldrd_strd): Use unaligned_loaddi
and unaligned_storedi for 4-byte aligned memory.
(arm_block_set_aligned_vect): Use unaligned_storev8qi for
4-byte aligned memory.
Martin Jambor [Fri, 30 Aug 2019 08:08:42 +0000 (10:08 +0200)]
[PR 91579] Avoid creating redundant PHI nodes in tail-call pass
2019-08-30 Martin Jambor <mjambor@suse.cz>
tree-optimization/91579
* tree-tailcall.c (tailr_arg_needs_copy): New variable.
(find_tail_calls): Allocate tailr_arg_needs_copy and set its bits as
appropriate.
(arg_needs_copy_p): Removed.
(eliminate_tail_call): Test tailr_arg_needs_copy instead of calling
arg_needs_copy_p.
(tree_optimize_tail_calls_1): Likewise. Free tailr_arg_needs_copy.
Uros Bizjak [Thu, 29 Aug 2019 18:15:13 +0000 (20:15 +0200)]
i386.c (inline_secondary_memory_needed): Return true for moves between SSE and non-general registers and between mask...
* config/i386/i386.c (inline_secondary_memory_needed): Return true
for moves between SSE and non-general registers and between
mask and non-general registers.
(ix86_register_move_cost): Remove stalled comment.
Nathan Sidwell [Thu, 29 Aug 2019 14:06:32 +0000 (14:06 +0000)]
[preprocessor] Include stacking
https://gcc.gnu.org/ml/gcc-patches/2019-08/msg01971.html
* internal.h (enum include_type): Add IT_MAIN, IT_DIRECTIVE_HWM,
IT_HEADER_HWM.
(_cpp_stack_file): Take include_type, not a bool.
* files.c (_cpp_find_file): Refactor to not hide an if inside a
for conditional.
(should_stack_file): Break apart to ...
(is_known_idempotent_file, has_unique_contents): ... these.
(_cpp_stack_file): Replace IMPORT boolean with include_type enum.
Refactor to use new predicates. Do linemap compensation here ...
(_cpp_stack_include): ... not here.
* init.c (cpp_read_main_file): Pass IT_MAIN to _cpp_stack_file.
Richard Biener [Thu, 29 Aug 2019 10:30:48 +0000 (10:30 +0000)]
re PR bootstrap/91580 (i686-{darwin, linux} bootstrap fails after r274926)
2019-08-29 Richard Biener <rguenther@suse.de>
PR bootstrap/91580
* config/i386/i386-features.c (general_scalar_chain::convert_insn):
Do not emit scalar copies for debug-insns, instead replace
their uses with the reg copy used in the chain or reset them
if there is a reaching definition outside of the chain as well.
Jakub Jelinek [Thu, 29 Aug 2019 09:20:54 +0000 (11:20 +0200)]
re PR tree-optimization/91351 (-fstrict-enums generates incorrect code)
PR tree-optimization/91351
* tree-cfg.c (generate_range_test): Use range_check_type instead of
unsigned_type_for.
* tree-cfgcleanup.c (convert_single_case_switch): Punt if
range_check_type returns NULL.
* tree-switch-conversion.c (switch_conversion::build_one_array):
Use range_check_type instead of unsigned_type_for, don't perform
linear opt if it returns NULL.
(bit_test_cluster::find_bit_tests): Formatting fix.
(bit_test_cluster::emit): Use range_check_type instead of
unsigned_type_for.
(switch_decision_tree::try_switch_expansion): Punt if range_check_type
returns NULL.
Richard Biener [Thu, 29 Aug 2019 08:07:35 +0000 (08:07 +0000)]
re PR tree-optimization/91568 (internal compiler error: in vect_schedule_slp_instance, at tree-vect-slp.c:3922)
2019-08-29 Richard Biener <rguenther@suse.de>
PR tree-optimization/91568
* tree-vectorizer.h (_slp_tree::max_nunits): Add.
(vect_update_max_nunits): Add overload for poly_uint64.
* tree-vect-slp.c (vect_create_new_slp_node): Initialize it.
(vect_build_slp_tree): Record max_nunits into the subtree
and merge it upwards.
(vect_print_slp_tree): Print max_nunits.
Marek Polacek [Thu, 29 Aug 2019 03:11:50 +0000 (03:11 +0000)]
Implement P1152R4: Deprecating some uses of volatile.
PR c++/91361
* c-opts.c (c_common_post_options): Enable -Wvolatile by
default for C++2a, unless -Wno-deprecated.
* c.opt (Wvolatile): New warning.
* cp-gimplify.c (cp_fold): Set TREE_THIS_VOLATILE.
* decl.c (grokdeclarator): Warn about a volatile-qualified structured
binding and return type.
(grokparms): Warn about a volatile-qualified function parameter.
* expr.c (mark_use) <case MODIFY_EXPR>: Emit a -Wvolatile warning.
* typeck.c (cp_build_unary_op): Emit a -Wvolatile warning for pre and
post ++/-- on a volatile operand.
(genericize_compound_lvalue): Use a better location. Don't lose
TREE_THIS_VOLATILE.
(cp_build_modify_expr): Emit a -Wvolatile warning for a compound
assignment whose LHS is volatile-qualified. Build the assignment with
a more precise location.
Ian Lance Taylor [Wed, 28 Aug 2019 20:39:32 +0000 (20:39 +0000)]
runtime: move osinit to Go
This is a step toward updating libgo to 1.13. This adds the 1.13
version of the osinit function to Go code, and removes the
corresponding code from the C runtime. This should simplify future updates.
Some additional 1.13 code was brought in to simplify this change.
* cp-tree.h (TINFO_VAR_DECLARED_CONSTINIT): Define.
(LOOKUP_CONSTINIT): Define.
(enum cp_decl_spec): Add ds_constinit.
* decl.c (check_tag_decl): Give an error for constinit in type
declarations.
(check_initializer): Also check LOOKUP_CONSTINIT.
(cp_finish_decl): Add checking for a constinit declaration. Set
TINFO_VAR_DECLARED_CONSTINIT.
(grokdeclarator): Add checking for a declaration with the constinit
specifier.
* lex.c (init_reswords): Handle D_CXX20.
* parser.c (cp_lexer_get_preprocessor_token): Pass a better location
to warning_at. Warn about C++20 keywords.
(cp_keyword_starts_decl_specifier_p): Handle RID_CONSTINIT.
(cp_parser_diagnose_invalid_type_name): Add an inform about constinit.
(cp_parser_decomposition_declaration): Maybe pass LOOKUP_CONSTINIT to
cp_finish_decl.
(cp_parser_decl_specifier_seq): Handle RID_CONSTINIT.
(cp_parser_init_declarator): Maybe pass LOOKUP_CONSTINIT to
cp_finish_decl.
(set_and_check_decl_spec_loc): Add "constinit".
* pt.c (tsubst_decl): Set TINFO_VAR_DECLARED_CONSTINIT.
(instantiate_decl): Maybe pass LOOKUP_CONSTINIT to cp_finish_decl.
* typeck2.c (store_init_value): If a constinit variable wasn't
initialized using a constant initializer, give an error.
* doc/invoke.texi: Document -Wc++20-compat.
* g++.dg/cpp2a/constinit1.C: New test.
* g++.dg/cpp2a/constinit2.C: New test.
* g++.dg/cpp2a/constinit3.C: New test.
* g++.dg/cpp2a/constinit4.C: New test.
* g++.dg/cpp2a/constinit5.C: New test.
* g++.dg/cpp2a/constinit6.C: New test.
* g++.dg/cpp2a/constinit7.C: New test.
* g++.dg/cpp2a/constinit8.C: New test.
* g++.dg/cpp2a/constinit9.C: New test.
* g++.dg/cpp2a/constinit10.C: New test.
* g++.dg/cpp2a/constinit11.C: New test.
* g++.dg/cpp2a/constinit12.C: New test.
Nathan Sidwell [Wed, 28 Aug 2019 18:43:37 +0000 (18:43 +0000)]
[Preprocessor] small cleanups
https://gcc.gnu.org/ml/gcc-patches/2019-08/msg01904.html
* directives-only.c (_cpp_preprocess_dir_only): Use false, not
zero for _cpp_handle_directive call.
* directives.c (_cpp_handle_directive): Indented is bool.
* files.c (struct _cpp_file): Make bools 1 bit bitfields.
* internal.h (enum include_type): Reformat and comment.
(struct cpp_buffer): Make flags 1 bit bitfields.
(_cpp_handle_directive): Indented is bool.
PR tree-optimization/91457
* c-c++-common/Wstringop-overflow-2.c: New test.
* g++.dg/warn/Warray-bounds-8.C: New test.
* g++.dg/warn/Wstringop-overflow-3.C: New test.
* gcc.dg/Wstringop-overflow-15.c: New test.
Martin Jambor [Wed, 28 Aug 2019 14:26:45 +0000 (16:26 +0200)]
[PR 91468] Small fixes in ipa-cp.c and ipa-prop.c
2019-08-28 Martin Jambor <mjambor@suse.cz>
PR ipa/91468
* ipa-cp.c (merge_agg_lats_step): Removed redundant test, made a
checking assert a normal assert to test it really is redundant.
* ipa-prop.c (compute_complex_assign_jump_func): Removed
redundant test.
(update_jump_functions_after_inlining): Removed combining unary
arithmetic operations with an ancestor jump function.
(ipcp_modif_dom_walker::before_dom_children): Fix wrong use of rhs
instead of t.
Mark Eggleston [Wed, 28 Aug 2019 10:21:31 +0000 (10:21 +0000)]
Stated standards in documentation
Correct the stated standards in documentation and for intrinsics and
specific intrinsics. In C_SIZEOF the printed value is T not .TRUE..
In IPARITY example wrap BOZ constants in calls to INT.
Martin Sebor [Tue, 27 Aug 2019 23:31:44 +0000 (23:31 +0000)]
PR tree-optimization/91567 - Spurious -Wformat-overflow warnings building glibc (32-bit only)
gcc/ChangeLog:
PR tree-optimization/91567
* gimple-ssa-sprintf.c (get_string_length): Handle more forms of lengths
of unknown strings.
* vr-values.c (vr_values::extract_range_basic): Set strlen upper bound
to PTRDIFF_MAX - 2.
gcc/testsuite/ChangeLog:
PR tree-optimization/91567
* gcc.dg/tree-ssa/builtin-snprintf-6.c: Xfail a subset of assertions
on targets other than x86_64 to work around PR 83543.
* gcc.dg/tree-ssa/builtin-sprintf-warn-22.c: New test.
Uros Bizjak [Tue, 27 Aug 2019 17:23:59 +0000 (19:23 +0200)]
re PR target/91528 (ICE in ix86_expand_prologue at i386.c:7844 since r274481)
PR target/91528
* config/i386/i386-features.c (convert_scalars_to_vector):
Update crtl->stack_realign_needed, crtl->stack_realign_tried and
crtl->stack_realign_processed. Update crtl->drap_reg by calling
targetm.calls.get_drap_rtx. If drap_rtx is non-null then
Update crtl->args.internal_arg_pointer and call fixup_tail_calls.
testsuite/ChangeLog:
PR target/91528
* gcc.target/i386/pr91528.c: New test.
Mark Eggleston [Tue, 27 Aug 2019 14:46:23 +0000 (14:46 +0000)]
Options documentation fixes.
Ensure that options lists fit cleanly inside the margins of a PDF page.
Reword description of option -ffrontend-loop-interchange so that it fits
cleanly inside the margins of a PDF page. Add options to those enabled
by -fdec.
Martin Liska [Tue, 27 Aug 2019 13:36:15 +0000 (15:36 +0200)]
Share a prevailing name for remove debug info symbols w/ LTO.
2019-08-27 Martin Liska <mliska@suse.cz>
PR lto/91478
* simple-object-elf.c (simple_object_elf_copy_lto_debug_sections):
First find a WEAK HIDDEN symbol in symbol table that will be
preserved. Later, use the symbol name for all removed symbols.
Richard Biener [Tue, 27 Aug 2019 12:46:07 +0000 (12:46 +0000)]
2019-08-27 Richard Biener <rguenther@suse.de>
* config/i386/i386-features.h
(general_scalar_chain::~general_scalar_chain): Add.
(general_scalar_chain::insns_conv): New bitmap.
(general_scalar_chain::n_sse_to_integer): New.
(general_scalar_chain::n_integer_to_sse): Likewise.
(general_scalar_chain::make_vector_copies): Adjust signature.
* config/i386/i386-features.c
(general_scalar_chain::general_scalar_chain): Outline,
initialize new members.
(general_scalar_chain::~general_scalar_chain): New.
(general_scalar_chain::mark_dual_mode_def): Record insns
we need to insert conversions at and count them.
(general_scalar_chain::compute_convert_gain): Account
for conversion instructions at chain boundary.
(general_scalar_chain::make_vector_copies): Generate a single
copy for a def by a specific insn.
(general_scalar_chain::convert_registers): First populate
defs_map, then make copies at out-of chain insns.
Jakub Jelinek [Tue, 27 Aug 2019 12:37:30 +0000 (14:37 +0200)]
re PR c++/91415 (Invalid warning for C++17 sequencing of shift operator E1<<E2.)
PR c++/91415
* c-common.c (verify_tree): For LSHIFT_EXPR, RSHIFT_EXPR,
COMPONENT_REF and ARRAY_REF in cxx_dialect >= cxx17 mode handle it
like COMPOUND_EXPR rather than normal expression.
Richard Earnshaw [Tue, 27 Aug 2019 10:05:51 +0000 (10:05 +0000)]
[arm/aarch64] Add comments warning that stack-protector initializer insns shouldn't be split
Following the publication of https://kb.cert.org/vuls/id/129209/ I've
been having a look at GCC's implementation for Arm and AArch64. I
haven't identified any issues yet, but it's a bit early to be
completely sure.
One observation, however, is that the instruction sequence that
initializes the stack canary might be vulnerable to producing a
reusable value if it were ever split early. I don't think we ever
would, because the memory locations involved with the stack protector
are all marked volatile to ensure that the values are only loaded at
the point in time when the test is intended to happen, and that also
has the effect of making it unlikely that the value would be reused
without reloading. Nevertheless, defence in depth is probably
warranted here.
So this patch just adds some comments warning that the patterns should
not be split.
Martin Liska [Tue, 27 Aug 2019 07:39:34 +0000 (09:39 +0200)]
Fix new clang warnings.
2019-08-27 Martin Liska <mliska@suse.cz>
* cgraph.c (cgraph_node::remove): Remove dead assignment before
loop.
* config/i386/i386-features.c (scalar_chain::emit_conversion_insns):
Enclose in anonymous namespace.
* config/i386/x86-tune-costs.h (struct processor_costs): Wrap
hard_register initialization in braces.
* tree-vrp.h (value_range_base::supports_type_p): Return false
for function with boolean return type.
Ian Lance Taylor [Mon, 26 Aug 2019 19:55:41 +0000 (19:55 +0000)]
compiler: generalize cleanup of unresolved placeholder pointer types
This change extends the work in https://golang.org/cl/51131 to include
placeholder pointer types created for Go function types, which can
also be left dangling/unresolved in some instances. This fixes an
assert in Llvm_backend::materializeComposite.
Test case can be found in https://golang.org/cl/191743.
Tejas Joshi [Mon, 26 Aug 2019 12:32:29 +0000 (12:32 +0000)]
Builtin function roundeven folding implementation
2019-08-26 Tejas Joshi <tejasjoshi9673@gmail.com>
* builtins.c (mathfn_built_in_2): Added CASE_MATHFN_FLOATN
for ROUNDEVEN.
* builtins.def: Added function definitions for roundeven function
variants.
* fold-const-call.c (fold_const_call_ss): Added case for roundeven
function call. Adjust condition for floor, ceil, trunc and round.
* fold-const.c (negate_mathfn_p): Added case for roundeven function.
(tree_call_nonnegative_warnv_p): Added case for roundeven function.
(integer_valued_real_call_p): Added case for roundeven function.
* real.c (is_even): New function. Returns true if real number is even,
otherwise returns false.
(is_halfway_below): New function. Returns true if real number is
halfway between two integers, else return false.
(real_roundeven): New function. Round real number to nearest integer,
rounding halfway cases towards even.
* real.h (real_value): Added descriptive comments. Added function
declaration for roundeven function.
* doc/extend.texi (Other Builtins): List roundeven variants among
functions which can be handled as builtins.
gcc/testsuite/ChangeLog:
2019-08-26 Tejas Joshi <tejasjoshi9673@gmail.com>
* gcc.dg/torture/builtin-round-roundeven.c: New test.
* gcc.dg/torture/builtin-round-roundevenf128.c: New test.
Richard Biener [Mon, 26 Aug 2019 10:35:59 +0000 (10:35 +0000)]
re PR target/91522 (STV is slow)
2019-08-26 Richard Biener <rguenther@suse.de>
PR target/91522
PR target/91527
* config/i386/i386-features.h (general_scalar_chain::defs_map):
New member.
(general_scalar_chain::replace_with_subreg): Remove.
(general_scalar_chain::replace_with_subreg_in_insn): Likewise.
(general_scalar_chain::convert_reg): Adjust signature.
* config/i386/i386-features.c (scalar_chain::add_insn): Do not
iterate over all defs of a reg.
(general_scalar_chain::replace_with_subreg): Remove.
(general_scalar_chain::replace_with_subreg_in_insn): Likewise.
(general_scalar_chain::make_vector_copies): Populate defs_map,
place copy only after defs that are used as vectors in the chain.
(general_scalar_chain::convert_reg): Emit a copy for a specific
def in a specific instruction.
(general_scalar_chain::convert_op): All reg uses are converted here.
(general_scalar_chain::convert_insn): Emit copies for scalar
uses of defs here. Replace uses with the copies we created.
Replace and convert the def. Adjust REG_DEAD notes, remove
REG_EQUIV/EQUAL notes.
(general_scalar_chain::convert_registers): Only handle copies
into the chain here.
* gcc.dg/tree-ssa/copy-headers-5.c: Do not run vrp pass.
* gcc.dg/tree-ssa/copy-headers-7.c: Do not run vrp pass.
* gcc.dg/tree-ssa/loop-15.c: Remove XFAIL.
* gcc.dg/tree-ssa/pr23744.c: Change search pattern.
* gcc.dg/wrapped-binop-simplify.c: New test.
Robin Dapp [Mon, 26 Aug 2019 10:18:24 +0000 (10:18 +0000)]
[PATCH 1/2] Allow folding all statements.
This patch allows users of the substitute_and_fold_engine to
enable folding all statements. It is now enabled for VRP which
is needed for the match.pd pattern in patch 2/2.
The loop versioning pass was missing one case when
deconstructing addresses that would only be triggered after
this patch for me:
It could handle addition and subsequent convert/nop but not
a convert/nop directly. This would cause the hash to be
calculated differently and, in turn, cause the pass to miss
a versioning opportunity. Fixed this by adding the missing
case.
--
gcc/ChangeLog:
2019-08-26 Robin Dapp <rdapp@linux.ibm.com>
* gimple-loop-versioning.cc (loop_versioning::record_address_fragment):
Add nop_convert case.
* tree-ssa-propagate.c (substitute_and_fold_dom_walker::before_dom_children):
Fold all statements if requested.
* tree-ssa-propagate.h (class substitute_and_fold_engine):
Allow to fold all statements.
* tree-vrp.c (class vrp_folder):
Let substitute_and_fold_engine fold all statements.
Richard Biener [Mon, 26 Aug 2019 09:29:07 +0000 (09:29 +0000)]
re PR c/91526 (Unnecessary SSE and other instructions generated when compiling in C mode (vs. C++ mode))
2019-08-26 Richard Biener <rguenther@suse.de>
PR tree-optimization/91526
* passes.def: Note that after late FRE we do TODO_update_address_taken.
* tree-ssa-sccvn.c (pass_fre::execute): In late mode schedule
TODO_update_address_taken.
projects / thirdparty / gcc.git / log
