This patch uses global rather than member operators for wide-int.h,
so that the first operand can be a non-wide-int type.
The patch also removes the and_not and or_not member functions.
It was already inconsistent to have member functions for these
two operations (one of which was never used) and not other wi::
ones like udiv. After the operator change, we'd have the additional
inconsistency that "non-wi & wi" would work but "non-wi.and_not (wi)"
wouldn't.
2017-10-09 Richard Sandiford <richard.sandiford@linaro.org>
Martin Jambor [Mon, 9 Oct 2017 09:41:44 +0000 (11:41 +0200)]
[PR 82416] Do not extend operands to at least 32 bits
2017-10-09 Martin Jambor <mjambor@suse.cz>
PR hsa/82416
gcc/
* hsa-common.h (hsa_op_with_type): New method extend_int_to_32bit.
* hsa-gen.c (hsa_extend_inttype_to_32bit): New function.
(hsa_type_for_scalar_tree_type): Use it. Always force min32int for
COMPLEX types.
(hsa_fixup_mov_insn_type): New function.
(hsa_op_with_type::get_in_type): Use it.
(hsa_build_append_simple_mov): Likewise. Allow sub-32bit
immediates in an assert.
(hsa_op_with_type::extend_int_to_32bit): New method.
(gen_hsa_insns_for_bitfield): Fixup instruction and intermediary
types. Convert to dest type if necessary.
(gen_hsa_insns_for_bitfield_load): Fixup load type if necessary.
(reg_for_gimple_ssa): Pass false as min32int to
hsa_type_for_scalar_tree_type.
(gen_hsa_addr): Fixup type when creating addresable temporary.
(gen_hsa_cmp_insn_from_gimple): Extend operands if necessary.
(gen_hsa_unary_operation): Extend operands and convert to dest type if
necessary. Call hsa_fixup_mov_insn_type.
(gen_hsa_binary_operation): Changed operand types to hsa_op_with_type,
extend operands and convert to dest type if necessary.
(gen_hsa_insns_for_operation_assignment): Extend operands and convert
to dest type if necessary.
(set_output_in_type): Call hsa_fixup_mov_insn_type. Just ude dest
if conversion nt necessary and size matches.
(gen_hsa_insns_for_load): Call hsa_fixup_mov_insn_type, convert
to dest type if necessary.
(gen_hsa_insns_for_store): Call hsa_fixup_mov_insn_type.
(gen_hsa_insns_for_switch_stmt): Likewise. Also extend operands if
necessary.
(gen_hsa_clrsb): Likewise.
(gen_hsa_ffs): Likewise.
(gen_hsa_divmod): Extend operands and convert to dest type if
necessary.
(gen_hsa_atomic_for_builtin): Change type of op to hsa_op_with_type.
libgomp/
* testsuite/libgomp.hsa.c/pr82416.c: New test.
This cleans up formatting of many jump patterns. Mostly it is writing
output statements as brace blocks, and deleting default arguments. It
also makes indentation a little bit more consistent, and uses b%T0
(instead of bctr/blr alternatives) in two more places.
* config/rs6000/rs6000.md (conditional branch): Clean up formatting.
Remove empty default arguments. Use a brace block as output
statement.
(conditional return): Ditto.
(jump): Ditto.
(indirect_jump): Ditto. Use b%T0 instead of bctr/blr.
(tablejump, tablejumpsi, tablejumpdi, *tablejump<mode>_internal1):
Ditto.
(group_ending_nop): Ditto.
(doloop_end): Ditto.
(ctr<mode>, ctr<mode>_internal1, ctr<mode>_internal2): Ditto.
(splitters for those): Ditto.
rs6000: Update conditional jump patterns, no more (pc) (label_ref)
Currently we have two patterns for every conditional jump: one jumping
if some condition is met, and the other jumping if it is not met. We
don't need that second form because all our conditions have an opposite
already, and the generic code can deal with that just fine.
This patch deletes that second form. We generated it directly in one
case; adjusted here.
* config/rs6000/rs6000-string.c (expand_strncmp_align_check): Invert
a conditional jump (and the compare for it) so that pc_rtx is the
last operand.
* config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Adjust
for the deleted and renamed ctr<mode>_internal[234] patterns.
* config/rs6000/rs6000.md: Delete second conditional branch pattern.
Delete second conditional return pattern.
(ctr<mode>_internal2): Delete this second bdnz pattern.
(ctr<mode>_internal3): Rename to ctr<mode>_internal2.
(ctr<mode>_internal4): Delete this second bdz pattern.
* tree-outof-ssa.h (ssaexpand): Add partitions_for_undefined_values.
(always_initialized_rtx_for_ssa_name_p): New predicate.
* tree-outof-ssa.c (remove_ssa_form): Initialize new field of SA.
(finish_out_of_ssa): Free new field of SA.
* tree-ssa-coalesce.h (get_undefined_value_partitions): Declare.
* tree-ssa-coalesce.c: Include tree-ssa.h.
(get_parm_default_def_partitions): Remove extern keyword.
(get_undefined_value_partitions): New function.
* expr.c (expand_expr_real_1) <expand_decl_rtl>: For a SSA_NAME, do
not set SUBREG_PROMOTED_VAR_P on the sub-register if it may contain
uninitialized bits.
* loop-iv.c (iv_get_reaching_def): Disqualify all subregs.
Paul Thomas [Sun, 8 Oct 2017 15:23:24 +0000 (15:23 +0000)]
re PR fortran/82375 (PDT components in PDT declarations fail to compile)
2017-10-08 Paul Thomas <pault@gcc.gnu.org>
PR fortran/82375
* module.c : Bump up MOD_VERSION to 15.
(mio_component): Edit comment about PDT specification list.
(mio_expr, mio_symbol): Include the expression and symbol PDT
specification lists in the same way as in mio_component.
Thomas Koenig [Sun, 8 Oct 2017 12:54:40 +0000 (12:54 +0000)]
dump_prase_tree (show_symbol): Output list of variables in NAMELIST.
2017-10-08 Thomas Koenig <tkoenig@gcc.gnu.org>
* dump_prase_tree (show_symbol): Output list of variables in
NAMELIST.
(show_code_node): Add new line for ELSE and END DO for DO
CONCURRENT.
* invoke.texi: Document that the output of
-fdump-fortran-original, -fdump-fortran-optimized and
-fdump-parse-tree is unsable and may lead to ICEs.
Olivier Hainque [Sun, 8 Oct 2017 08:20:25 +0000 (08:20 +0000)]
arm.c (arm_set_return_address): Use MEM_VOLATILE_P on the target mem instead of RTX_FRAME_RELATED_P on the insn...
2017-10-08 Olivier Hainque <hainque@adacore.com>
* config/arm/arm.c (arm_set_return_address): Use MEM_VOLATILE_P
on the target mem instead of RTX_FRAME_RELATED_P on the insn to
prevent DSE.
(thumb_set_return_address): Likewise.
Eric Botcazou [Sat, 7 Oct 2017 22:17:16 +0000 (22:17 +0000)]
builtins.def (BUILT_IN_SETJMP): Declare as library builtin instead of GCC builtin if...
* builtins.def (BUILT_IN_SETJMP): Declare as library builtin instead
of GCC builtin if DONT_USE_BUILTIN_SETJMP is defined.
* except.c (sjlj_emit_function_enter): If DONT_USE_BUILTIN_SETJMP is
defined, force the creation of a new block for a dispatch label.
Paul Thomas [Sat, 7 Oct 2017 21:14:06 +0000 (21:14 +0000)]
re PR fortran/82375 (PDT components in PDT declarations fail to compile)
2017-10-07 Paul Thomas <pault@gcc.gnu.org>
PR fortran/82375
* class.c (gfc_find_derived_vtab): Return NULL for a passed
pdt template to prevent bad procedures from being written.
* decl.c (gfc_get_pdt_instance): Do not use the default
initializer for pointer and allocatable pdt type components. If
the component is allocatbale, set the 'alloc_comp' attribute of
'instance'.
* module.c : Add a prototype for 'mio_actual_arglist'. Add a
boolean argument 'pdt'.
(mio_component): Call it for the parameter list of pdt type
components with 'pdt' set to true.
(mio_actual_arg): Add the boolean 'pdt' and, if it is set, call
mio_integer for the 'spec_type'.
(mio_actual_arglist): Add the boolean 'pdt' and use it in the
call to mio_actual_arg.
(mio_expr, mio_omp_udr_expr): Call mio_actual_arglist with
'pdt' set false.
* resolve.c (get_pdt_spec_expr): Add the parameter name to the
KIND parameter error.
(get_pdt_constructor): Check that cons->expr is non-null.
* trans-array.c (structure_alloc_comps): For deallocation of
allocatable components, ensure that parameterized components
are deallocated first. Likewise, when parameterized components
are allocated, nullify allocatable components first. Do not
recurse into pointer or allocatable pdt components while
allocating or deallocating parameterized components. Test that
parameterized arrays or strings are allocated before freeing
them.
(gfc_trans_pointer_assignment): Call the new function. Tidy up
a minor whitespace issue.
trans-decl.c (gfc_trans_deferred_vars): Set 'tmp' to NULL_TREE
to prevent the expression from being used a second time.
2017-10-07 Paul Thomas <pault@gcc.gnu.org>
PR fortran/82375
* gfortran.dg/pdt_13.f03 : New test.
* gfortran.dg/pdt_14.f03 : New test.
* gfortran.dg/pdt_15.f03 : New test.
Thomas Koenig [Sat, 7 Oct 2017 11:43:58 +0000 (11:43 +0000)]
gfortran.h (async_io_dt): Add external reference.
2017-10-07 Thomas Koenig <tkoenig@gcc.gnu.org>
* gfortran.h (async_io_dt): Add external reference.
* io.c (async_io_dt): Add variable.
(compare_to_allowed_values): Add prototyte. Add optional argument
num. If present, set it to the number of the entry that was
matched.
(check_io_constraints): If this is for an asynchronous I/O
statement, set async_io_dt and set the asynchronous flag for
a SIZE tag.
* resolve.c (resolve_transfer): If async_io_dt is set, set
the asynchronous flag on the variable.
(resolve_fl_namelist): If async_io_dt is set, set the asynchronous
flag on all elements of the namelist.
Jakub Jelinek [Fri, 6 Oct 2017 15:43:17 +0000 (17:43 +0200)]
P0704R1 - fixing const-qualified pointers to members
P0704R1 - fixing const-qualified pointers to members
* typeck2.c (build_m_component_ref): For -std=c++2a allow
pointer to const & qualified method on rvalue.
Wilco Dijkstra [Fri, 6 Oct 2017 11:54:51 +0000 (11:54 +0000)]
PR82396 workaround
r253236 broke AArch64 bootstrap. This is a temporary workaround that
disables qsort checking in the scheduler to enable continued development
and testing on AArch64. This will be removed once the autopref scheduling
code has been fixed.
Jakub Jelinek [Fri, 6 Oct 2017 09:39:07 +0000 (11:39 +0200)]
re PR tree-optimization/82434 (-fstore-merging does not work reliably.)
PR tree-optimization/82434
* fold-const.h (can_native_encode_type_p,
can_native_encode_string_p): Remove.
* fold-const.c (native_encode_int): Formatting fixes. If ptr is NULL,
don't encode anything, just return what would be otherwise returned.
(native_encode_fixed, native_encode_complex, native_encode_vector):
Likewise.
(native_encode_string): Likewise. Inline by hand
can_native_encode_string_p.
(can_native_encode_type_p): Remove.
(can_native_encode_string_p): Remove.
* tree-vect-stmts.c (vectorizable_store): Instead of testing just
STRING_CSTs using can_native_encode_string_p, test all
CONSTANT_CLASS_P values using native_encode_expr with NULL ptr.
* gimple-ssa-store-merging.c (encode_tree_to_bitpos): Remove last
argument from native_encode_expr.
(rhs_valid_for_store_merging_p): Use native_encode_expr with NULL ptr.
(pass_store_merging::execute): Don't unnecessarily look for 3 stmts,
but just 2.
Richard Biener [Fri, 6 Oct 2017 09:27:09 +0000 (09:27 +0000)]
re PR tree-optimization/82397 (qsort comparator non-negative on sorted output: 1 in vect_analyze_data_ref_accesses)
2017-10-06 Richard Biener <rguenther@suse.de>
PR tree-optimization/82397
* tree-vect-data-refs.c (dr_group_sort_cmp): Do not use
operand_equal_p but rely on data_ref_compare_tree for detecting
equalities.
(vect_analyze_data_ref_accesses): Use data_ref_compare_tree
to match up with dr_group_sort_cmp.
Andreas Krebbel [Fri, 6 Oct 2017 09:19:46 +0000 (09:19 +0000)]
PR82322: S/390: Fix vec_ceil and friends
vec_ceil and friends are expanded by vecintrin.h to __builtin_s390_vfi
which is an overloaded builtin being replaced by either
__builtin_s390_vfisb or __builtin_s390_vfidb depending on its argument
types.
The problem in this PR was that the overloaded builtin definition of
__builtin_s390_vfi was missing in s390-builtins.def.
gcc/ChangeLog:
2017-10-06 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
Andreas Krebbel [Fri, 6 Oct 2017 09:18:37 +0000 (09:18 +0000)]
PR82317: S/390: Fix vec_min/vec_max builtins for IBM z13.
With IBM z14 a hardware instruction for floating point min and max has
been added while for IBM z13 we emulated min/max for vector double with
compare and select. This testcase makes sure that we fall back to the
emulated variant when compiling for z13.
gcc/ChangeLog:
2017-10-06 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
PR target/82317
* config/s390/s390-builtin-types.def: Regenerate.
* config/s390/s390-builtins.def (s390_vfmaxdb_4, s390_vfmindb_4):
Change flag from B_VXE to B_VX.
(s390_vec_min_dbl): Remove B_VXE flag.
gcc/testsuite/ChangeLog:
2017-10-06 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
PR target/82317
* gcc.target/s390/zvector/pr82317.c: New test.
Richard Biener [Fri, 6 Oct 2017 07:03:23 +0000 (07:03 +0000)]
re PR tree-optimization/82436 (465.tonto ICE in vect_get_slp_vect_defs, at tree-vect-slp.c:3410)
2017-10-06 Richard Biener <rguenther@suse.de>
PR tree-optimization/82436
* tree-vect-slp.c (vect_supported_load_permutation_p): More
conservatively choose the vectorization factor when checking
whether we can perform the required load permutation.
(vect_transform_slp_perm_load): Assert when we may not fail.
compiler: drop special handling of unexported func/var assembler names
For example, for the package math/big, we used to generate unexported
names as `big.trim` and exported names as `math_big.NewInt`. After
this change we will use `math_big` consistently.
Consolidate all symbol name handling into the new file names.cc. This
is intended to define all the names that will appear in the generated
assembly code. Names that will not appear in the assembly code, such
as local variable names or label names, remain where they are.
This consolidation is not intended to change any of the existing
symbol names. Tested by building without and without this patch and
comparing the libgo symbol table.
elf.c (elf_zlib_fetch): Change pval argument to uint64_t *.
* elf.c (elf_zlib_fetch): Change pval argument to uint64_t *.
Read a four byte integer.
(elf_zlib_inflate): Change val to uint64_t. Align pin to a 32-bit
boundary before ever calling elf_zlib_fetch.
* ztest.c (test_large): Simplify print statements a bit.
Jan Hubicka [Thu, 5 Oct 2017 14:56:32 +0000 (16:56 +0200)]
i386.c (ix86_size_cost, [...]): Set reassociation width to 1.
* i386.c (ix86_size_cost, i386_cost, i486_cost, pentium_cost,
lakemont_cost, pentiumpro_cost, geode_cost, k6_cost,
athlon_cost, k8_cost, amdfam10_cost, btver1_cost, btver2_cost,
pentium4_cost, nocona_cost): Set reassociation width to 1.
(bdver1_cost, bdver2_cost, bdver3_cost, bdver4_cost): Set reassociation
width to 2 for fp operations and 1 otherwise.
(znver1_cost): Set scalar reassoc width to 4 and vector to 3 and 6
for int and fp.
(atom_cost): Set reassociation width to 2.
(slm_cost, generic_cost): Set fp reassociation width to 2 and 1 otherwise.
(intel_cost): Set fp reassociation width to 4 and 1 otherwise.
(core_cost): Set fp reassociation width to 4 and vector to 2.
(ix86_reassociation_width): Rewrite using cost table; special case
plus/minus on Zen; honor X86_TUNE_SSE_SPLIT_REGS
and TARGET_AVX128_OPTIMAL.
* i386.h (processor_costs): Add
reassoc_int, reassoc_fp, reassoc_vec_int, reassoc_vec_fp.
(TARGET_VECTOR_PARALLEL_EXECUTION, TARGET_REASSOC_INT_TO_PARALLEL,
TARGET_REASSOC_FP_TO_PARALLEL): Remove.
* x86-tune.def (X86_TUNE_REASSOC_INT_TO_PARALLEL): Remove.
(X86_TUNE_REASSOC_FP_TO_PARALLEL): Remove.
(X86_TUNE_VECTOR_PARALLEL_EXECUTION): Remove.
Paolo Carlini [Wed, 4 Oct 2017 21:25:20 +0000 (21:25 +0000)]
re PR c++/80471 ((gcc extension) Forwarding-reference `auto` function parameters do not follow the same rules as template functions or generic lambdas)
2017-10-04 Paolo Carlini <paolo.carlini@oracle.com>
Nathan Sidwell [Wed, 4 Oct 2017 17:54:59 +0000 (17:54 +0000)]
[C++ PATCH] give builtin types consistent name
https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00201.html
Give builtin types the correct name.
* name-lookup.c (set_global_binding): Assert name is DECL_NAME.
* decl.c (record_builtin_type): Reimplement, use new TYPE_DECL for
rname.
Sudakshina Das [Wed, 4 Oct 2017 16:59:40 +0000 (16:59 +0000)]
[PATCH][AArch64] Add BIC-imm and ORR-imm SIMD pattern
This patch adds the support for BIC (vector, immediate) and
ORR (vector, immediate) SIMD patterns to the AArch64 backend.
Committed on behalf of Sudi Das.
Reviewed-by: Richard Earnshaw <Richard.Earnshaw@arm.com> Reviewed-by: James Greenhalgh <james.greenhalgh@arm.com>
gcc/
* config/aarch64/aarch64-protos.h (enum simd_immediate_check): New
check type for aarch64_simd_valid_immediate.
(aarch64_output_simd_mov_immediate): Update prototype.
(aarch64_simd_valid_immediate): Update prototype.
* config/aarch64/aarch64-simd.md (orr<mode>3): modified pattern to add
support for ORR-immediate.
(and<mode>3): modified pattern to add support for BIC-immediate.
* config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Function
now checks for valid immediate for BIC and ORR based on new enum
argument.
(aarch64_output_simd_mov_immediate): Function now used to output
BIC/ORR imm as well based on new enum argument.
* config/aarch64/constraints.md (Do): New vector immediate constraint.
(Db) : Likewise.
* config/aarch64/predicates.md (aarch64_reg_or_orr_imm): New predicate.
(aarch64_reg_or_bic_imm): Likewise.
gcc/testsuite/
* gcc.target/aarch64/bic_imm_1.c: New test.
* gcc.target/aarch64/orr_imm_1.c: Likewise.
Andreas Krebbel [Wed, 4 Oct 2017 16:43:09 +0000 (16:43 +0000)]
S/390: Fix mode in vector merge pattern.
vec_unpacks_hi_v4sf/vec_unpacks_lo_v4sf expand vec_mergeh and vec_mergel
patterns also for z13 with V4SF modes so the patterns should better
accept this. Fixed by changing the mode iterator to V_128_NOSINGLE
which accepts V4SF unconditionally.
gcc/ChangeLog:
2017-10-04 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
* config/s390/vx-builtins.md ("vec_mergeh<mode>")
("vec_mergel<mode>"): Change mode iterator to V_128_NOSINGLE.
Petr Ovtchenkov [Wed, 4 Oct 2017 16:15:18 +0000 (16:15 +0000)]
2017-10-04 Petr Ovtchenkov <ptr@void-ptr.info>
François Dumont <fdumont@gcc.gnu.org>
* include/bits/streambuf_iterator.h
(istreambuf_iterator<>::operator*()): Do not capture iterator state
in Debug assertion.
(istreambuf_iterator<>::operator++()): Likewise and remove _M_sbuf check.
(istreambuf_iterator<>::operator++(int)): Likewise.
(istreambuf_iterator<>::_M_get()): Remove _M_c assignment.
(istreambuf_iterator<>::_S_is_eof()): New.
(istreambuf_iterator<>::_M_at_eof()): Adapt, use latter.
(find(istreambuf_iterator<>, istreambuf_iterator<>, _CharT)):
Return an iterator with _M_c set to eof to capture streambuf state
on evaluation.
(testsuite/24_iterators/istreambuf_iterator/2.cc): Add checks.
Co-Authored-By: François Dumont <fdumont@gcc.gnu.org>
From-SVN: r253417
David Malcolm [Wed, 4 Oct 2017 14:10:59 +0000 (14:10 +0000)]
C: underline parameters in mismatching function calls
In r253096
("C++: underline parameters in mismatching function calls"
aka 5d78d423a5f7a1d135c7bb678e82007678d1313c
https://gcc.gnu.org/ml/gcc-patches/2017-09/msg01546.html )
I updated the C++ FE's handling of mismatched types in function calls
so that it underlines the pertinent param of the callee, rather than
just the function name.
The following patch does the same for the C frontend.
Given e.g. this type mismatch:
extern int callee (int one, const char *two, float three);
int caller (int first, int second, float third)
{
return callee (first, second, third);
}
the C FE currently emits (trunk):
test.c: In function 'caller':
test.c:5:25: warning: passing argument 2 of 'callee' makes pointer from
integer without a cast [-Wint-conversion]
return callee (first, second, third);
^~~~~~
test.c:1:12: note: expected 'const char *' but argument is of type 'int'
extern int callee (int one, const char *two, float three);
^~~~~~
whereas with this patch the note underlines the pertinent param of
the callee:
test.c: In function 'caller':
test.c:5:25: warning: passing argument 2 of 'callee' makes pointer from
integer without a cast [-Wint-conversion]
return callee (first, second, third);
^~~~~~
test.c:1:41: note: expected 'const char *' but argument is of type 'int'
extern int callee (int one, const char *two, float three);
~~~~~~~~~~~~^~~
making the problem more obvious to the user.
As with the C++ patch, the patch:
(a) updates the locations of the params to cover the range of all
of their tokens, putting the caret on the first character of the
param name (if present), otherwise at the start of the first token
(doing so requires adding a last_token_location to the c_parser, so
we can determine the location of the last consumed token).
(b) updates the "note" to use the param location, rather than the
fndecl location
gcc/c/ChangeLog:
* c-decl.c (push_parm_decl): Store c_parm's location into the
PARAM_DECL.
(build_c_parm): Add "loc" param and store it within the c_parm.
* c-parser.c (struct c_parser): Add "last_token_location" field.
(c_parser_consume_token): Store location of the token into the
new field.
(c_parser_declaration_or_fndef): Store params into DECL_ARGUMENTS
when handling a FUNCTION_DECL, if it doesn't already have them.
(c_parser_parameter_declaration): Generate a location for the
parameter, and pass it to the call to build_c_parm.
* c-tree.h (struct c_parm): Add field "loc".
(build_c_parm): Add location_t param.
* c-typeck.c (get_fndecl_argument_location): New function.
(inform_for_arg): New function.
(convert_for_assignment): Use inform_for_arg when dealing with
ic_argpass.
gcc/testsuite/ChangeLog:
* gcc.dg/diagnostic-range-bad-called-object.c: Update expected
underlining for param.
* gcc.dg/param-type-mismatch.c: Update expected results to reflect
highlighting of parameters; add test coverage for trivial
parameter decls, and for callback parameters.
* gcc.dg/pr68533.c: Update location of two errors to reflect
location of params.
/* Build a vector rvalue from an array of elements.
"vec_type" should be a vector type, created using gcc_jit_type_get_vector.
This API entrypoint was added in LIBGCCJIT_ABI_10; you can test for its
presence using
#ifdef LIBGCCJIT_HAVE_gcc_jit_context_new_rvalue_from_vector
*/
extern gcc_jit_rvalue *
gcc_jit_context_new_rvalue_from_vector (gcc_jit_context *ctxt,
gcc_jit_location *loc,
gcc_jit_type *vec_type,
size_t num_elements,
gcc_jit_rvalue **elements);
gcc/jit/ChangeLog:
* docs/cp/topics/expressions.rst (Vector expressions): New
section.
* docs/topics/compatibility.rst (LIBGCCJIT_ABI_10): New ABI tag.
* docs/topics/expressions.rst (Vector expressions): New section.
* docs/topics/types.rst (gcc_jit_type_get_vector): Add link to
gcc_jit_context_new_rvalue_from_vector.
* jit-common.h (gcc::jit:recording::vector_type): New forward
decl.
* jit-playback.c
(gcc::jit::playback::context::new_rvalue_from_vector): New method.
* jit-playback.h
(gcc::jit::playback::context::new_rvalue_from_vector): New method.
* jit-recording.c: In namespace gcc::jit::
(class comma_separated_string): New class.
(comma_separated_string::comma_separated_string): New ctor,
adapted from recording::call::make_debug_string.
(comma_separated_string::~comma_separated_string): New dtor.
In namespace gcc::jit::recording::
(context::new_rvalue_from_vector): New method.
(type::get_vector): Update for renaming of memento_of_get_vector.
(class memento_of_get_vector): Rename to...
(class vector_type): ..this.
(memento_of_new_rvalue_from_vector::memento_of_new_rvalue_from_vector):
New ctor.
(memento_of_new_rvalue_from_vector::replay_into): New method.
(memento_of_new_rvalue_from_vector::visit_children): New method.
(memento_of_new_rvalue_from_vector::make_debug_string): New
method.
(memento_of_new_rvalue_from_vector::write_reproducer): New method.
(call::make_debug_string): Split out arg-printing code into ctor
for comma_separated_string.
* jit-recording.h: In namespace gcc::jit::recording::
(context::new_rvalue_from_vector): New method.
(type::dyn_cast_vector_type): New virtual function.
(class memento_of_get_vector): Rename to...
(class vector_type): ...this.
(vector_type::unqualified): Remove this vfunc override in favor
of...
(vector_type::get_element_type): ...this new method.
(vector_type::get_num_units): New method.
(vector_type::dyn_cast_vector_type): New vfunc override.
(class memento_of_new_rvalue_from_vector): New class.
* libgccjit++.h (gccjit::context::new_rvalue): Add overload for
vector of rvalue.
* libgccjit.c (gcc_jit_context_new_binary_op): Strip off type
qualifications when checking that both operands have same type.
(gcc_jit_context_new_rvalue_from_vector): New API entrypoint.
* libgccjit.h
(LIBGCCJIT_HAVE_gcc_jit_context_new_rvalue_from_vector): New
macro.
(gcc_jit_context_new_rvalue_from_vector): New API entrypoint.
* libgccjit.map (LIBGCCJIT_ABI_10): New ABI tag.
gcc/testsuite/ChangeLog:
* jit.dg/test-expressions.c (make_test_of_vectors): New function.
(create_code): Call it.
* jit.dg/test-vector-rvalues.cc: New test case.
2017-10-04 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
* g++.dg/vect/slp-pr56812.cc: xfail for targets which don't want
vector loop peeling.
* lib/target-supports.exp (check_effective_target_vect_peeling_profitable):
New proc.
Andreas Krebbel [Wed, 4 Oct 2017 11:59:30 +0000 (11:59 +0000)]
New target check for double<->int conversions
gcc/ChangeLog:
2017-10-04 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
* doc/sourcebuild.texi: Document vect_intdouble_cvt and
vect_doubleint_cvt.
gcc/testsuite/ChangeLog:
2017-10-04 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
* gcc.dg/vect/pr66251.c: Replace vect_floatint_cvt with
vect_doubleint_cvt and vect_intfloat_cvt with vect_intdouble_cvt.
* gcc.dg/vect/vect-floatint-conversion-2.c: Replace
vect_floatint_cvt with vect_doubleint_cvt.
* gcc.dg/vect/vect-intfloat-conversion-3.c: Replace
vect_intfloat_cvt with vect_intdouble_cvt.
* gfortran.dg/vect/pr60510.f: Require vect_intdouble_cvt.
* gfortran.dg/vect/vect-8.f90: Make number of vectorized loops
depend on vect_intdouble_cvt.
* lib/target-supports.exp
(check_effective_target_vect_doubleint_cvt)
(check_effective_target_vect_intdouble_cvt): New procs.
Andreas Krebbel [Wed, 4 Oct 2017 11:56:28 +0000 (11:56 +0000)]
Enable vect_float with S/390 VXE and adjust testcases
The target supports routines provide vect_double and vect_float but
these do not appear to be used consequently in the vect testcases.
With z13 we only have support for vector double but with z14 also for
vector float. This patch adds vect_float to the testcases using the
float data type and make the vect_float target check to return 1 only
on z14.
gcc/testsuite/ChangeLog:
2017-10-04 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
PR82413: Mismatched precisions in build_range_check
build_range_check explicitly allows LOW and HIGH to be a different type
from EXP, so we need to use w::to_widest when comparing a value based on
HIGH with a value based on EXP's type.
2017-10-04 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
PR tree-optimization/82413
* fold-const.c (build_range_check): Use widest_int when comparing
the maximum ETYPE value with HIGH.
gcc/testsuite/
PR tree-optimization/82413
* g++.dg/pr82413.C: New test.
git://gcc.gnu.org / gcc.git / log