2019-02-18 Martin Jambor <mjambor@suse.cz>
PR tree-optimization/89209
* tree-sra.c (create_access_replacement): New optional parameter
reg_tree. Use it as a type if non-NULL and access type is not of
a register type.
(get_repl_default_def_ssa_name): New parameter REG_TYPE, pass it
to create_access_replacement.
(sra_modify_assign): Pass LHS type to get_repl_default_def_ssa_name.
Check lacc is non-NULL before attempting to re-create it on the RHS.
testsuite/
* gcc.dg/tree-ssa/pr89209.c: New test.
From-SVN: r268980
2019-02-18 Martin Liska <mliska@suse.cz>
PR ipa/89306
* cgraph.c (symbol_table::create_edge): Set m_summary_id to -1
by default.
(symbol_table::free_edge): Recycle m_summary_id.
* cgraph.h (get_summary_id): New.
(symbol_table::release_symbol): Set m_summary_id to -1
by default.
(symbol_table::allocate_cgraph_symbol): Recycle m_summary_id.
* ipa-fnsummary.c (ipa_fn_summary_t): Switch from
function_summary to fast_function_summary.
* ipa-fnsummary.h (ipa_fn_summary_t): Likewise.
* ipa-pure-const.c (class funct_state_summary_t):
Switch from function_summary to fast_function_summary.
* ipa-reference.c (class ipa_ref_var_info_summary_t): Likewise.
(class ipa_ref_opt_summary_t): Switch from function_summary
to fast_function_summary.
* symbol-summary.h (class function_summary_base): New class
that is created from base of former function_summary.
(function_summary_base::unregister_hooks): New.
(class function_summary): Inherit from function_summary_base.
(class call_summary_base): New class
that is created from base of former call_summary.
(class call_summary): Inherit from call_summary_base.
(struct is_same): New.
(class fast_function_summary): New summary class.
(class fast_call_summary): New summary class.
* vec.h (vec_safe_grow_cleared): New function.
From-SVN: r268979
2019-02-18 Martin Liska <mliska@suse.cz>
* config/i386/i386.c (ix86_get_multilib_abi_name): New function.
(TARGET_GET_MULTILIB_ABI_NAME): New macro defined.
* doc/tm.texi: Document new target hook.
* doc/tm.texi.in: Likewise.
* target.def: Add new target macro.
* gcc.c (find_fortran_preinclude_file): Do not search multilib
suffixes.
2019-02-18 Martin Liska <mliska@suse.cz>
* decl.c (gfc_match_gcc_builtin): Add support for filtering
of builtin directive based on multilib ABI name.
2019-02-18 Martin Liska <mliska@suse.cz>
* gfortran.dg/simd-builtins-7.f90: New test.
* gfortran.dg/simd-builtins-7.h: New test.
From-SVN: r268978
2019-02-17 Harald Anlauf <anlauf@gmx.de>
PR fortran/88299
* resolve.c (resolve_common_blocks,resolve_common_vars): Move
check for obsolent COMMON feature in F2018 to better place.
PR fortran/88299
* gfortran.dg/pr88299.f90: New test.
From-SVN: r268974
2019-02-17 Harald Anlauf <anlauf@gmx.de>
PR fortran/89077
* decl.c (gfc_set_constant_character_len): Clear original string
representation after padding has been performed to target length.
PR fortran/89077
* gfortran.dg/transfer_simplify_12.f90: New test.
From-SVN: r268973
* constexpr.c (unshare_constructor): No longer static.
* cp-tree.h (unshare_constructor): Declare.
* semantics.c (finish_compound_literal): When dealing with a
non-dependent expression in a template, return the original
expression. Pass LOOKUP_NO_NARROWING to digest_init_flags.
* g++.dg/cpp0x/range-for37.C: New test.
From-SVN: r268969
This patch fixes a bug that can result in "insn does not satisfy its
constraints" if these splitters fire due to not getting ctr for the
jump insn. Since the jump insn can have any of r,m,d,wi,c,l as the
decremented count output, it's not sufficient to check for
gpc_reg_operand (which matches VSX regs for example). Seen after
correcting register_move_cost when the cost of gpr <-> vsx is much
lower. Since this is a prerequisite to fixing PR89271, I'm mentioning
that PR in the ChangeLog.
The <bd>tf_<mode> split had a further bug in that it wouldn't match
if the count output was m,d,wi, or l.
PR target/89271
* config/rs6000/rs6000.md (<bd>_<mode> split): Check for an int
output reg on add insn.
(<bd>tf_<mode> split): Likewise. Match predicates with insn.
From-SVN: r268968
There is no V4HI pmulhrsw in AVX512BW and V4HI/V8HI pmulhrsw don't require
AVX2.
PR target/89372
* config/i386/sse.md (ssedoublemode): Remove V4HI.
(PMULHRSW): Likewise.
(<ssse3_avx2>_pmulhrsw<mode>3): Require TARGET_SSSE3, not
TARGET_AVX2.
(ssse3_pmulhrswv4hi3): New expander.
From-SVN: r268964
The second and third alternatives in *vec_extractv2si_zext_mem don't
require MMX. But the second one requires SSE2.
* config/i386/mmx.md (*vec_extractv2si_zext_mem): Doesn't require
MMX. Add isa attribute.
From-SVN: r268963
PR middle-end/88074
* simplify.c (simplify_transformation_to_array): Run post_op
immediately after processing corresponding row, rather than at the
end.
(norm2_scale): New variable.
(add_squared): Rename to ...
(norm2_add_squared): ... this. Scale down operand and/or result
if needed.
(do_sqrt): Rename to ...
(norm2_do_sqrt): ... this. Handle the result == e case. Scale up
result and clear norm2_scale.
(gfc_simplify_norm2): Clear norm2_scale. Change add_squared to
norm2_add_squared and &do_sqrt to norm2_do_sqrt. Scale up result
and clear norm2_scale again.
From-SVN: r268962
PR c++/88680 reports excess warnings from -Wtype-limits after the C++
FE's use of location wrappers was extended in r267272 for cases such as:
const unsigned n = 8;
static_assert (n >= 0 && n % 2 == 0, "");
t.C:3:18: warning: comparison of unsigned expression >= 0 is always true
[-Wtype-limits]
3 | static_assert (n >= 0 && n % 2 == 0, "");
| ~~^~~~
The root cause is that the location wrapper around "n" breaks the
suppression of the warning for the "if OP0 is a constant that is >= 0"
case.
This patch fixes it by calling fold_for_warn on OP0, extracting the
constant.
gcc/c-family/ChangeLog:
PR c++/88680
* c-common.c (shorten_compare): Call fold_for_warn on op0 when
implementing -Wtype-limits.
gcc/testsuite/ChangeLog:
PR c++/88680
* g++.dg/wrappers/pr88680.C: New test.
From-SVN: r268961
2019-02-17 Thomas Koenig <tkoenig@gcc.gnu.org>
PR fortran/71066
* trans-decl.c (generate_coarray_sym_init): For an array
constructor in a DATA statement of a coarray variable, set the
rank to 1 to avoid confusion later on. If the constructor
contains only one value, use that for initiailizig.
2019-02-17 Thomas Koenig <tkoenig@gcc.gnu.org>
PR fortran/71066
* gfortran.dg/coarray_data_1.f90: New test.
From-SVN: r268960
PR rtl-optimization/66152
* builtins.h (c_readstr): Declare.
* builtins.c (c_readstr): Remove forward declaration. Add
null_terminated_p argument, if false, read all bytes from the
string instead of stopping after '\0'.
* expr.c (string_cst_read_str): New function.
(store_expr): Use string_cst_read_str instead of
builtin_strncpy_read_str. Try to store by pieces the whole
exp_len first, and only if that fails, split it up into
store by pieces followed by clear_storage. Formatting fix.
* gcc.target/i386/pr66152.c: New test.
From-SVN: r268957
Currently, the compiler lowers runtime.getcallersp to
__builtin_frame_address(1). In the C side of the runtime,
getcallersp is defined as __builtin_frame_address(0). They don't
match. Further, neither of them actually returns the caller's SP.
On AMD64, __builtin_frame_address(0) just returns the frame
pointer. __builtin_frame_address(1) returns the memory content
where the frame pointer points to, which is typically the
caller's frame pointer but can also be garbage if the frame
pointer is not enabled.
This CL changes it to use __builtin_dwarf_cfa(), which returns
the caller's SP at the call site. This matches the SP we get
from unwinding the stack.
Currently getcallersp is not used for anything real. It will be
used for precise stack scan (a new version of CL 159098).
Reviewed-on: https://go-review.googlesource.com/c/162905
* go-gcc.cc (Gcc_backend::Gcc_backend): Define __builtin_dwarf_cfa
instead of __builtin_frame_address.
From-SVN: r268952
PR go/89368
compiler: write barrier check nil-check policy tweak
Tweak the recipe for generating writeBarrier loads to insure that the
dereference expr is marked as not requiring a nil check (not needed
for gccgo, but needed for gollvm).
Fixes https://gcc.gnu.org/PR89368
Reviewed-on: https://go-review.googlesource.com/c/162904
From-SVN: r268948
There's a bit of a disconnect between the feature flags that don't test the fpu
and ones that do when the test itself also forces an architecture. The forcing
of the architecture would change the defaults and without explicitly giving the
correct fpu again the test would fail.
I don't see a good way to solve this problem, really the feature tests should
ideally contain the extra options the test adds too, but for this specific case
it can be solved by always testing the fpu explicitly.
Committed under the GCC obvious
gcc/testsuite/ChangeLog:
* lib/target-supports.exp
(check_effective_target_arm_neon_softfp_fp16_ok_nocache): Drop non-fpu
checking alternative.
From-SVN: r268943
* c-c++-common/patchable_function_entry-decl.c: Do not run on Visium.
* c-c++-common/patchable_function_entry-default.c: Likewise.
* c-c++-common/patchable_function_entry-definition.c: Likewise.
* gcc.dg/tree-ssa/pr84859.c: Add -ftree-cselim switch.
From-SVN: r268932
libgcc/
* config/visium/lib2funcs.c (__set_trampoline_parity): Replace
TRAMPOLINE_SIZE with __LIBGCC_TRAMPOLINE_SIZE__.
gcc/
* final.c (insn_current_reference_address): Replace test on JUMP_P
with test on jump_to_label_p.
* config/visium/visium-passes.def: New file.
* config/visium/t-visium (PASSES_EXTRA): Define.
* config/visium/visium-protos.h (make_pass_visium_reorg): Declare.
* config/visium/visium.h (TRAMPOLINE_SIZE): Adjust.
(TRAMPOLINE_ALIGNMENT): Define.
* config/visium/visium.c (visium_option_override): Do not register
the machine-specific reorg pass here.
(visium_trampoline_init): Align the BRA insn on a 64-bit boundary
for the GR6.
(output_branch): Adjust threshold for long branch instruction.
* config/visium/visium.md (cpu): Move around.
(length): Adjust for the GR6.
From-SVN: r268931
PR other/69006
PR testsuite/88920
* lib/gcc-dg.exp: If llvm_binutils effective target, set
allow_blank_lines to 2 during initialization.
(dg-allow-blank-lines-in-output): Set allow_blank_lines to 1 only if
it was previously zero.
(gcc-dg-prune): Don't check for llvm_binutils effective target here.
Clear allow_blank_lines afterwards whenever it was 1.
* gdc.test/gdc-test.exp (dmd2dg): Don't call
dg-allow-blank-lines-in-output here.
(gdc-do-test): Set allow_blank_lines to 3 if it is 0 before running
the tests and restore it back at the end.
From-SVN: r268930
* c-c++-common/ubsan/opts-1.c: New test.
* c-c++-common/ubsan/opts-2.c: New test.
* c-c++-common/ubsan/opts-3.c: New test.
* c-c++-common/ubsan/opts-4.c: New test.
From-SVN: r268929
PR tree-optimization/89278
* tree-loop-distribution.c: Include tree-eh.h.
(generate_memset_builtin, generate_memcpy_builtin): Call
rewrite_to_non_trapping_overflow on builtin->size before passing it
to force_gimple_operand_gsi.
* gcc.dg/pr89278.c: New test.
Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r268927
PR c/89340
* c-decl.c (start_function): Clear TREE_PUBLIC on nested functions
before c_decl_attributes rather than after it.
* gcc.dg/pr89340.c: New test.
* gcc.dg/torture/pr57036-2.c (jpgDecode_convert): Expect a warning
that leaf attribute on nested function is useless.
From-SVN: r268926
Compiling with LTO revealed a number of cases in the runtime and
standard library where C and Go disagreed about the type of an object or
function (or where Go and code generated by the compiler disagreed). In
all cases the underlying representation was the same (e.g., uintptr vs.
void*), so this wasn't causing actual problems, but it did result in a
number of annoying warnings when compiling with LTO.
Reviewed-on: https://go-review.googlesource.com/c/160700
From-SVN: r268923
PR go/89168
libgo: change gotest to run examples with output
Change the gotest script to act like "go test" and run examples that
have "output" comments. This is not done with full generality, but
just enough to run the libgo tests. Other packages should be tested
with "go test" as usual.
While we're here clean up some old bits of gotest, and only run
TestXXX functions that are actually in *_test.go files. The latter
change should fix https://gcc.gnu.org/PR89168.
Reviewed-on: https://go-review.googlesource.com/c/162139
From-SVN: r268922
* config/i386/i386.h (TARGET_SUBTARGET64_ISA_DEFAULT):
Enable MMX, SSE and SSE2 by default.
* config/i386/i386.c (ix86_option_override_internal): Do not
explicitly set MMX, SSE and SSE2 flags for TARGET_64BIT here.
From-SVN: r268917
PR rtl-optimization/89354
* combine.c (make_extraction): Punt if extraction_mode is narrower
than len bits.
* gcc.dg/pr89354.c: New test.
From-SVN: r268913
Make the option handling code parse the -flag-init-integer value as a
C long type, allowing a larger range on systems where long is a larger
type than int. Document the behavior.
Regtested on x86_64-pc-linux-gnu, committed as obvious.
2019-02-14 Janne Blomqvist <jb@gcc.gnu.org>
PR fortran/81552
* gfortran.h (gfc_option_t): Make flag_init_integer_value a long.
* options.c (gfc_handle_option): Use strtol instead of atoi.
* invoke.texi: Document -finit-integer behavior in more detail
From-SVN: r268906
