Implement various corrections in the compatibility implementations of the
x86 vector intrinsics found after enabling 32bit mode for the associated
test cases. (Actual enablement coming in a subsequent patch.)
2018-10-26 Paul A. Clarke <pc@us.ibm.com>
gcc/ChangeLog:
* config/rs6000/mmintrin.h: Enable 32bit compilation.
* config/rs6000/xmmintrin.h: Likewise.
From-SVN: r265535
Fix asm name directive for the C version of log/syslog.syslog_c,
which didn't get included in the recent name mangling change.
Reviewed-on: https://go-review.googlesource.com/c/145017
From-SVN: r265533
For compatibility implementation of x86 vector intrinsic, _mm_extract_pi16,
adjust shift value for big-endian mode.
gcc/ChangeLog:
2018-10-25 Paul A. Clarke <pc@us.ibm.com>
* config/rs6000/xmmintrin.h (_mm_extract_pi16): Fix for big-endian.
From-SVN: r265531
* gcc_release (error, inform): Use $@ instead of $1.
(build_sources): Check for ^[[:blank:]]*GCC in both index.html
and changes.html, rather than for GCC in one and ^GCC in another one.
From-SVN: r265526
2018-10-26 Martin Liska <mliska@suse.cz>
PR testsuite/86158
* ipa-prop.c (struct ipa_vr_ggc_hash_traits): Hash with
addr_expr and not with pointers.
From-SVN: r265521
* ipa-devirt.c (warn_odr): Make static.
(types_same_for_odr): Drop strict variant.
(types_odr_comparable): Likewise.
(odr_or_derived_type_p): Look for main variants.
(odr_name_hasher::equal): Cleanup comment.
(odr_subtypes_equivalent): Add warn and warned arguments; check main
variants.
(type_variants_equivalent_p): break out from ...
(odr_types_equivalent): ... here; go for main variants where needed.
(warn_odr): ... here; turn static.
(warn_types_mismatch): Compare mangled names of main variants.
* ipa-utils.h (types_odr_comparable): Drop strict parameter.
(type_with_linkage_p): Sanity check that we look at main variant.
* lto.c (lto_read_decls): Only consider main variant to be ODR type.
* tree.h (types_same_for_odr): Drop strict argument.
From-SVN: r265519
2018-10-26 Richard Biener <rguenther@suse.de>
PR tree-optimization/87746
* tree-vect-data-refs.c (vect_update_misalignment_for_peel):
Simplify and fix WRT strided store groups with size not
equal to step in element count.
(vect_analyze_group_access_1): Dump the whole group.
* gcc.dg/pr87746.c: New testcase.
From-SVN: r265517
The current implementation of Gogo::pkgpath_for_symbol was written in
a way that allowed two distinct package paths to map to the same
symbol, which could cause collisions at link- time or compile-time.
Switch to a better mangling scheme to insure that we get a unique
packagepath symbol for each package. In the new scheme instead of having
separate mangling schemes for identifiers and package paths, the
main identifier mangler ("go_encode_id") now handles mangling of
both packagepath characters and identifier characters.
The new mangling scheme is more intrusive: "foo/bar.Baz" is mangled as
"foo..z2fbar.Baz" instead of "foo_bar.Baz". To mitigate this, this
patch also adds a demangling capability so that function names
returned from runtime.CallersFrames are converted back to their
original unmangled form.
Changing the pkgpath_for_symbol scheme requires updating a number of
//go:linkname directives and C "__asm__" directives to match the new
scheme, as well as updating the 'gotest' driver (which makes
assumptions about the correct mapping from pkgpath symbol to package
name).
Fixesgolang/go#27534.
Reviewed-on: https://go-review.googlesource.com/c/135455
From-SVN: r265510
2018-10-25 Bill Schmidt <wschmidt@linux.ibm.com>
Jinsong Ji <jji@us.ibm.com>
* config/rs6000/emmintrin.h (_mm_slli_epi16): Replace deprecated
function with vec_sl.
(_mm_slli_epi32): Likewise.
(_mm_slli_epi64): Likewise.
(_mm_srai_epi16): Replace deprecated function with vec_sra.
(_mm_srai_epi32): Likewise.
(_mm_srli_epi16): Replace deprecated function with vec_sr.
(_mm_srli_epi32): Likewise.
(_mm_srli_epi64): Likewise.
(_mm_sll_epi16): Replace deprecated function with vec_sl.
(_mm_sll_epi32): Likewise.
(_mm_sll_epi64): Likewise.
(_mm_sra_epi16): Replace deprecated function with vec_sra.
(_mm_sra_epi32): Likewise.
(_mm_srl_epi16): Replace deprecated function with vec_sr.
(_mm_srl_epi32): Likewise.
(_mm_srl_epi64): Likewise.
Co-Authored-By: Jinsong Ji <jji@us.ibm.com>
From-SVN: r265508
2018-10-25 Bill Schmidt <wschmidt@linux.ibm.com>
Jinsong Ji <jji@us.ibm.com>
* gcc/config/rs6000/emmintrin.h (_mm_sll_epi16): Replace
comparison operators with vec_cmp* for compatibility due to
unfortunate history; clean up formatting and use types more
appropriately.
(_mm_sll_epi32): Likewise.
(_mm_sll_epi64): Likewise.
(_mm_srl_epi16): Likewise.
(_mm_srl_epi32): Likewise.
(_mm_srl_epi64): Likewise.
Co-Authored-By: Jinsong Ji <jji@us.ibm.com>
From-SVN: r265507
2018-10-25 Bill Schmidt <wschmidt@linux.ibm.com>
Jinsong Ji <jji@us.ibm.com>
* config/rs6000/emmintrin.h (_mm_sll_epi64): Remove wrong cast.
* config/rs6000/xmmintrin.h (_mm_min_ps): Change m's type to
__vector __bool int. Use vec_cmpgt in preference to deprecated
function vec_vcmpgtfp.
(_mm_max_ps): Likewise.
Co-Authored-By: Jinsong Ji <jji@us.ibm.com>
From-SVN: r265506
The move constructor for the SSO string uses assign(const basic_string&)
when either:
(1) the source string is "local" and so the contents of the small string
buffer need to be copied, or
(2) the allocator does not propagate and is_always_equal is false.
Case (1) is suboptimal, because the assign member is not noexcept and
the compiler isn't smart enough to see it won't actually throw in this
case. This causes extra code in the move assignment operator so that any
exception will be turned into a call to std::terminate. This can be
fixed by copying small strings inline instead of calling assign.
Case (2) is a bug, because the specific instances of the allocators
could be equal even if is_always_equal is false. This can result in an
unnecessary deep copy (and potentially-throwing allocation) when the
storage should be moved. This can be fixed by simply checking if the
allocators are equal.
PR libstdc++/87749
* include/bits/basic_string.h [_GLIBCXX_USE_CXX11_ABI]
(basic_string::operator=(basic_string&&)): For short strings copy the
buffer inline. Only fall back to using assign(const basic_string&) to
do a deep copy when reallocation is needed.
* testsuite/21_strings/basic_string/modifiers/assign/char/87749.cc:
New test.
* testsuite/21_strings/basic_string/modifiers/assign/char/
move_assign_optim.cc: New test.
* testsuite/21_strings/basic_string/modifiers/assign/wchar_t/87749.cc:
New test.
* testsuite/21_strings/basic_string/modifiers/assign/wchar_t/
move_assign_optim.cc: New test.
From-SVN: r265493
* ipa-devirt.c (main_odr_variant): Remove.
(hash_odr_name, types_same_for_odr, types_odr_comparable,
odr_name_hasher::equal, odr_subtypes_equivalent_p):
Drop use of main_odr_variant.
(add_type_duplicate): Silence confused warnings on integer types.
(get_odr_type): Always look for main variant.
(register_odr_type): Simplify.
From-SVN: r265492
2018-10-25 Richard Biener <rguenther@suse.de>
* tree-vect-data-refs.c (vect_analyze_data_ref_accesses):
Initialize ng to silence error with release checking bootstrap.
From-SVN: r265491
Consider the following RTL:
(insn (set (mem/f/c:DI (reg/f:DI 60))
(const:DI (plus:DI (symbol_ref:DI ("*.LANCHOR0"))
(const_int 8)))))
generated by cse2 pass. It is matched to movdi_64, resulting in
the following inefficient code:
larl %r5,.L6 # Load literal pool@
lg %r1,.L7-.L6(%r5) # Load .LANCHOR0+8
stgrl %r1,.LANCHOR0
br %r14
Matching it to movdi_larl improves the code, eliminating one
instruction and the literal pool entry:
larl %r1,.LANCHOR0+8
stgrl %r1,.LANCHOR0
br %r14
Taking it one step further, there is no reason to keep movdi_64 and
movdi_larl separate, since this could potentially improve code in other
ways by giving lra one more alternative to choose from.
gcc/ChangeLog:
2018-10-25 Ilya Leoshkevich <iii@linux.ibm.com>
* config/s390/constraints.md (ZL): New constraint.
* config/s390/s390.c (legitimate_pic_operand_p): Accept LARL
operands.
* config/s390/s390.md (movdi_larl): Remove.
(movdi_64): Add the LARL alternative.
gcc/testsuite/ChangeLog:
2018-10-25 Ilya Leoshkevich <iii@linux.ibm.com>
* gcc.target/s390/global-array-almost-huge-element.c: New test.
* gcc.target/s390/global-array-almost-negative-huge-element.c: New test.
* gcc.target/s390/global-array-element-pic.c: New test.
* gcc.target/s390/global-array-even-element.c: New test.
* gcc.target/s390/global-array-huge-element.c: New test.
* gcc.target/s390/global-array-negative-huge-element.c: New test.
* gcc.target/s390/global-array-odd-element.c: New test.
From-SVN: r265490
2018-10-25 Richard Biener <rguenther@suse.de>
* tree-if-conv.c: Include tree-ssa-sccvn.h.
(tree_if_conversion): Run CSE on the if-converted loop body.
From-SVN: r265489
r264556 and r264537 changed the format of EQ_ATTR_ALT RTXs to "ww",
which also required adjusting rtx_code_size initializer. In order to
simplify things, the list of rtx_codes known to use HOST_WIDE_INTs was
replaced by the format string check. However, unlike the old one, this
new check cannot be always performed at compile time, in which case a
static constructor is generated. This may lead to a static
initialization order fiasco with respect to other static constructors
in the compiler, in case of PR87747, cselib's pool_allocator.
gcc/ChangeLog:
2018-10-25 Ilya Leoshkevich <iii@linux.ibm.com>
PR bootstrap/87747
* rtl.c (RTX_CODE_HWINT_P_1): New helper macro.
(RTX_CODE_HWINT_P): New macro.
(rtx_code_size): Use RTX_CODE_HWINT_P ().
From-SVN: r265488
* ipa-devirt.c (odr_types_equivalent_p): Do not ICE if one of types
is anonymous.
* g++.dg/lto/odr-1_0.C: New test.
* g++.dg/lto/odr-1_1.C: New test.
From-SVN: r265484
Currently, dg-cmp-results will not print anything for a test that was
not run before, even if it is a FAIL or UNRESOLVED now. This means that
when contributing a code change together with a testcase in the same
commit one must run dg-cmp-results twice: once to check for regression
on a full testsuite run and once against the new testcase with -v -v.
This also prevents using dg-cmp-results on sum files generated with
test_summary since these would not contain PASS.
This patch changes dg-cmp-results to print NA->FAIL and NA->UNRESOLVED
changes by default.
2018-10-25 Thomas Preud'homme <thomas.preudhomme@linaro.org>
contrib/
* dg-cmp-results.sh: Print NA-FAIL and NA->UNRESOLVED changes at
default verbosity.
From-SVN: r265483
gcc.dg/sibcall-9.c and gcc.dg/sibcall-10.c give execution failure
on ARM when compiled with -fPIC due to the PIC access to volatile
variable v creating an extra spill which causes the frame size of the
two recursive functions to be different. Making the variable static
solve the issue because the variable can be access in a PC-relative way
and avoid the spill, while still testing sibling call as originally
intended.
2018-10-25 Thomas Preud'homme <thomas.preudhomme@linaro.org>
gcc/testsuite/
* gcc.dg/sibcall-9.c: Make v static.
* gcc.dg/sibcall-10.c: Likewise.
From-SVN: r265482
2018-10-25 Richard Biener <rguenther@suse.de>
PR tree-optimization/87665
PR tree-optimization/87745
* tree-vectorizer.h (get_earlier_stmt): Remove.
(get_later_stmt): Pick up UID from the original non-pattern stmt.
* gfortran.dg/20181025-1.f: New testcase.
From-SVN: r265481
PR fortran/87725
* openmp.c (gfc_match_omp_clauses): Parse simd, monotonic and
nonmonotonic modifiers regardless of if they have been parsed
already or if the opposite one has. Fix up check whether
comma after modifier should be parsed.
(resolve_omp_clauses): Diagnose schedule modifier restrictions.
* c-c++-common/gomp/schedule-modifiers-1.c (bar): Separate modifier
from kind with a colon rather than comma.
* gfortran.dg/gomp/schedule-modifiers-1.f90: New test.
* gfortran.dg/gomp/schedule-modifiers-2.f90: New test.
From-SVN: r265479
Jumps are written in RTL as moves to PC. But the latter has no mode,
so we shouldn't try to use it. Since the optimization this routine
does does not really help for jumps at all, let's just skip it.
PR rtl-optimization/87720
* combine.c (make_more_copies): Skip if the dest is pc_rtx.
From-SVN: r265474
The isolate-paths pass is activated by various -f flags, but also by
-Wnull-dereference. Most of its codegen changes are conditioned on at
least one of the -f flags, but those that detect, warn about and
isolate paths that return the address of local variables are enabled
even if the pass is activated only by -Wnull-dereference.
-W flags should not cause codegen changes, so this patch makes the
codegen changes conditional on the presence of any of the -f flags
that activate the pass. Should we have a separate option to activate
only this kind of transformation?
for gcc/ChangeLog
* gimple-ssa-isolate-paths.c
(find_implicit_erroneous_behavior): Do not change code if the
pass is running for warnings only.
(find_explicit_erroneous_behavior): Likewise.
From-SVN: r265473
[gcc]
2018-10-24 Michael Meissner <meissner@linux.ibm.com>
* config/rs6000/rs6000.c (TARGET_MANGLE_DECL_ASSEMBLER_NAME):
Define as rs6000_mangle_decl_assembler_name.
(rs6000_mangle_decl_assembler_name): If the user switched from IBM
long double to IEEE long double, switch the names of the long
double built-in functions to be <func>f128 instead of <func>l.
[gcc/testsuite]
2018-10-24 Michael Meissner <meissner@linux.ibm.com>
* gcc.target/powerpc/float128-math.c: New test to make sure the
long double built-in function names use the f128 form if the user
switched from IBM long double to IEEE long double.
* gcc.target/powerpc/ppc-fortran/ieee128-math.f90: Likewise.
From-SVN: r265471
PR c++/86288
* parser.c (cp_parser_std_attribute): Canonicalize attr_ns, and when
:: is not present and attr_ns non-NULL, canonicalize also attr_id.
(cp_parser_attribute_spec): Fix comment typo.
* g++.dg/cpp0x/gen-attrs-66.C: New test.
From-SVN: r265470
Paul A. Clarke