Jonathan Wakely [Tue, 3 Jul 2018 21:04:45 +0000 (22:04 +0100)]
P0556R3 Integral power-of-2 operations, P0553R2 Bit operations
P0553R2 is not in the C++2a working draft yet, but is likely to be
approved soon. Neither proposal supports std::byte but this adds
overloads of each function for std::byte, assuming that will also get
added.
* include/Makefile.am: Add new header.
* include/Makefile.in: Regenerate.
* include/precompiled/stdc++.h: Include new header.
* include/std/bit: New header.
(__rotl, __rotr, __countl_zero, __countl_one, __countr_zero)
(__countr_one, __popcount, __ispow2, __ceil2, __floor2, __log2p1):
Define for C++14.
[!__STRICT_ANSI__] (rotl, rotr, countl_zero, countl_one, countr_zero)
(countr_one, popcount): Define for C++2a. Also overload for std::byte.
(ispow2, ceil2, floor2, log2p1): Define for C++2a.
[!__STRICT_ANSI__] (ispow2, ceil2, floor2, log2p1): Overload for
std::byte.
* testsuite/26_numerics/bit/bit.pow.two/ceil2.cc: New.
* testsuite/26_numerics/bit/bit.pow.two/floor2.cc: New.
* testsuite/26_numerics/bit/bit.pow.two/ispow2.cc: New.
* testsuite/26_numerics/bit/bit.pow.two/log2p1.cc: New.
* testsuite/26_numerics/bit/bitops.rot/rotl.cc: New.
* testsuite/26_numerics/bit/bitops.rot/rotr.cc: New.
* testsuite/26_numerics/bit/bitops.count/countl_one.cc: New.
* testsuite/26_numerics/bit/bitops.count/countl_zero.cc: New.
* testsuite/26_numerics/bit/bitops.count/countr_one.cc: New.
* testsuite/26_numerics/bit/bitops.count/countr_zero.cc: New.
Paolo Carlini [Tue, 3 Jul 2018 21:03:51 +0000 (21:03 +0000)]
decl.c (min_location): New.
/cp
2018-07-03 Paolo Carlini <paolo.carlini@oracle.com>
* decl.c (min_location): New.
(smallest_type_quals_location): Use the latter.
(check_concept_fn): Use DECL_SOURCE_LOCATION.
(grokdeclarator): Use accurate locations in a number of error
messages involving ds_thread, ds_storage_class, ds_virtual,
ds_constexpr, ds_typedef and ds_friend; exploit min_location.
/testsuite
2018-07-03 Paolo Carlini <paolo.carlini@oracle.com>
* config/h8300/h8300.c (h8300_insn_length_from_table): Consolidate
ADDB, ADDW and ADDL into a single ADD attribute which selects the
right table based on the size of the operand.
* config/h8300/h8300.md (length_table): Corresponding changes. All
references to "addb", "addw" and "addl" changed to "add".
(btst patterns): Merge two variants into a single pattern.
(tstqi, tsthi): Likewise.
(addhi3_incdec, addsi3_incdec): Likewise.
(subhi3_h8300hs, subsi3_h8300hs): Likewise.
(mulhi3, mulsi3): Likewise.
(udivhi3, udivsi3): Likewise.
(divhi3, divsi3): Likewise.
(andorqi3, andorhi3, andorsi3): Likewise.
The PR85694 series added a vectype argument to append_pattern_def_seq.
This patch makes more callers use it.
2018-07-03 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-patterns.c (vect_recog_rotate_pattern)
(vect_recog_vector_vector_shift_pattern, vect_recog_divmod_pattern)
(vect_recog_mixed_size_cond_pattern, adjust_bool_pattern_cast)
(adjust_bool_pattern, vect_recog_bool_pattern): Pass the vector
type to append_pattern_def_seq instead of creating a stmt_vec_info
directly.
(build_mask_conversion): Likewise. Remove vinfo argument.
(vect_add_conversion_to_patterm): Likewise, renaming to...
(vect_add_conversion_to_pattern): ...this.
(vect_recog_mask_conversion_pattern): Update call to
build_mask_conversion. Pass the vector type to
append_pattern_def_seq here too.
(vect_recog_gather_scatter_pattern): Update call to
vect_add_conversion_to_pattern.
Ensure PATTERN_DEF_SEQ is empty before recognising patterns
Various recognisers set PATTERN_DEF_SEQ to null before adding
statements to it, but it should always be null at that point anyway.
This patch asserts for that in vect_pattern_recog_1 and removes
the redundant code.
2018-07-03 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-patterns.c (new_pattern_def_seq): Delete.
(vect_recog_dot_prod_pattern, vect_recog_sad_pattern)
(vect_recog_widen_op_pattern, vect_recog_over_widening_pattern)
(vect_recog_rotate_pattern, vect_synth_mult_by_constant): Don't set
STMT_VINFO_PATTERN_DEF_SEQ to null here.
(vect_recog_pow_pattern, vect_recog_vector_vector_shift_pattern)
(vect_recog_mixed_size_cond_pattern, vect_recog_bool_pattern): Use
append_pattern_def_seq instead of new_pattern_def_seq.
(vect_recog_divmod_pattern): Do both of the above.
(vect_pattern_recog_1): Assert that STMT_VINO_PATTERN_DEF_SEQ
is null.
The PR85694 series removed the only cases in which a pattern recogniser
could attach patterns to more than one statement. I think it would be
better to avoid adding any new instances of that, since it interferes
with the normal matching order.
This patch therefore switches the interface back to passing a single
statement instead of a vector. It also gets rid of the clearing of
STMT_VINFO_RELATED_STMT on failure, since no recognisers use it now.
2018-07-03 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-patterns.c (vect_recog_dot_prod_pattern):
(vect_recog_sad_pattern, vect_recog_widen_op_pattern)
(vect_recog_widen_mult_pattern, vect_recog_pow_pattern):
(vect_recog_widen_sum_pattern, vect_recog_over_widening_pattern)
(vect_recog_average_pattern, vect_recog_cast_forwprop_pattern)
(vect_recog_widen_shift_pattern, vect_recog_rotate_pattern)
(vect_recog_vector_vector_shift_pattern, vect_synth_mult_by_constant)
(vect_recog_mult_pattern, vect_recog_divmod_pattern)
(vect_recog_mixed_size_cond_pattern, vect_recog_bool_pattern)
(vect_recog_mask_conversion_pattern): Replace vec<gimple *>
parameter with a single stmt_vec_info.
(vect_recog_func_ptr): Likewise.
(vect_recog_gather_scatter_pattern): Likewise, folding in...
(vect_try_gather_scatter_pattern): ...this.
(vect_pattern_recog_1): Remove stmts_to_replace and just pass
the stmt_vec_info of the statement to be matched. Don't clear
STMT_VINFO_RELATED_STMT.
(vect_pattern_recog): Update call accordingly.
[16/n] PR85694: Add detection of averaging operations
This patch adds detection of average instructions:
a = (((wide) b + (wide) c) >> 1);
--> a = (wide) .AVG_FLOOR (b, c);
a = (((wide) b + (wide) c + 1) >> 1);
--> a = (wide) .AVG_CEIL (b, c);
in cases where users of "a" need only the low half of the result,
making the cast to (wide) redundant. The heavy lifting was done by
earlier patches.
This showed up another problem in vectorizable_call: if the call is a
pattern definition statement rather than the main pattern statement,
the type of vectorised call might be different from the type of the
original statement.
2018-07-03 Richard Sandiford <richard.sandiford@arm.com>
gcc/
PR tree-optimization/85694
* doc/md.texi (avgM3_floor, uavgM3_floor, avgM3_ceil)
(uavgM3_ceil): Document new optabs.
* doc/sourcebuild.texi (vect_avg_qi): Document new target selector.
* internal-fn.def (IFN_AVG_FLOOR, IFN_AVG_CEIL): New internal
functions.
* optabs.def (savg_floor_optab, uavg_floor_optab, savg_ceil_optab)
(savg_ceil_optab): New optabs.
* tree-vect-patterns.c (vect_recog_average_pattern): New function.
(vect_vect_recog_func_ptrs): Add it.
* tree-vect-stmts.c (vectorizable_call): Get the type of the zero
constant directly from the associated lhs.
[15/n] PR85694: Try to split existing casts in widened patterns
The main over-widening patch can introduce quite a few extra casts,
and in many cases those casts simply "tap into" an intermediate
point in an existing extension. E.g. if we have:
unsigned char a;
int ax = (int) a;
and a later operation using ax is shortened to "unsigned short",
we would need:
unsigned short ax' = (unsigned short) a;
The a->ax extension requires one set of unpacks to get to unsigned
short and another set of unpacks to get to int. The first set are
then duplicated for ax'. If both ax and ax' are needed, the a->ax'
extension would end up counting twice during cost calculations.
This patch rewrites the original:
int ax = (int) a;
into a pattern:
unsigned short ax' = (unsigned short) a;
int ax = (int) ax';
so that each extension only counts once.
2018-07-03 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-patterns.c (vect_split_statement): New function.
(vect_convert_input): Use it to try to split an existing cast.
gcc/testsuite/
* gcc.dg/vect/vect-over-widen-5.c: Test that the extensions
get split into two for use by the over-widening pattern.
* gcc.dg/vect/vect-over-widen-6.c: Likewise.
* gcc.dg/vect/vect-over-widen-7.c: Likewise.
* gcc.dg/vect/vect-over-widen-8.c: Likewise.
* gcc.dg/vect/vect-over-widen-9.c: Likewise.
* gcc.dg/vect/vect-over-widen-10.c: Likewise.
* gcc.dg/vect/vect-over-widen-11.c: Likewise.
* gcc.dg/vect/vect-over-widen-12.c: Likewise.
* gcc.dg/vect/vect-over-widen-13.c: Likewise.
* gcc.dg/vect/vect-over-widen-14.c: Likewise.
* gcc.dg/vect/vect-over-widen-15.c: Likewise.
* gcc.dg/vect/vect-over-widen-16.c: Likewise.
* gcc.dg/vect/vect-over-widen-22.c: New test.
This patch is the main part of PR85694. The aim is to recognise at least:
signed char *a, *b, *c;
...
for (int i = 0; i < 2048; i++)
c[i] = (a[i] + b[i]) >> 1;
as an over-widening pattern, since the addition and shift can be done
on shorts rather than ints. However, it ended up being a lot more
general than that.
The current over-widening pattern detection is limited to a few simple
cases: logical ops with immediate second operands, and shifts by a
constant. These cases are enough for common pixel-format conversion
and can be detected in a peephole way.
The loop above requires two generalisations of the current code: support
for addition as well as logical ops, and support for non-constant second
operands. These are harder to detect in the same peephole way, so the
patch tries to take a more global approach.
The idea is to get information about the minimum operation width
in two ways:
(1) by using the range information attached to the SSA_NAMEs
(effectively a forward walk, since the range info is
context-independent).
(2) by back-propagating the number of output bits required by
users of the result.
As explained in the comments, there's a balance to be struck between
narrowing an individual operation and fitting in with the surrounding
code. The approach is pretty conservative: if we could narrow an
operation to N bits without changing its semantics, it's OK to do that if:
- no operations later in the chain require more than N bits; or
- all internally-defined inputs are extended from N bits or fewer,
and at least one of them is single-use.
See the comments for the rationale.
I didn't bother adding STMT_VINFO_* wrappers for the new fields
since the code seemed more readable without.
2018-06-20 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* poly-int.h (print_hex): New function.
* dumpfile.h (dump_dec, dump_hex): Declare.
* dumpfile.c (dump_dec, dump_hex): New poly_wide_int functions.
* tree-vectorizer.h (_stmt_vec_info): Add min_output_precision,
min_input_precision, operation_precision and operation_sign.
* tree-vect-patterns.c (vect_get_range_info): New function.
(vect_same_loop_or_bb_p, vect_single_imm_use)
(vect_operation_fits_smaller_type): Delete.
(vect_look_through_possible_promotion): Add an optional
single_use_p parameter.
(vect_recog_over_widening_pattern): Rewrite to use new
stmt_vec_info infomration. Handle one operation at a time.
(vect_recog_cast_forwprop_pattern, vect_narrowable_type_p)
(vect_truncatable_operation_p, vect_set_operation_type)
(vect_set_min_input_precision): New functions.
(vect_determine_min_output_precision_1): Likewise.
(vect_determine_min_output_precision): Likewise.
(vect_determine_precisions_from_range): Likewise.
(vect_determine_precisions_from_users): Likewise.
(vect_determine_stmt_precisions, vect_determine_precisions): Likewise.
(vect_vect_recog_func_ptrs): Put over_widening first.
Add cast_forwprop.
(vect_pattern_recog): Call vect_determine_precisions.
r262275 allowed pattern matching on pattern statements. Testing for
SVE on more benchmarks showed a case where this interacted badly
with 14/n.
The new over-widening detection could narrow a COND_EXPR A to another
COND_EXPR B, which mixed_size_cond could then match. This was working
as expected. However, we left B (now dead) in the pattern definition
sequence with a non-null PATTERN_DEF_SEQ. mask_conversion also
matched B, and unlike most recognisers, didn't clear PATTERN_DEF_SEQ
before adding statements to it. This meant that the statements
created by mixed_size_cond appeared in two supposedy separate
sequences, causing much confusion.
This patch removes pattern statements that are replaced by further
pattern statements. As a belt-and-braces fix, it also nullifies
PATTERN_DEF_SEQ on failure, in the same way Richard B. did recently
for RELATED_STMT.
I have patches to clean up the PATTERN_DEF_SEQ handling, but they
only apply after the complete PR85694 sequence, whereas this needs
to go in before 14/n.
2018-07-03 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-patterns.c (vect_mark_pattern_stmts): Remove pattern
statements that have been replaced by further pattern statements.
(vect_pattern_recog_1): Clear STMT_VINFO_PATTERN_DEF_SEQ on failure.
gcc/testsuite/
* gcc.dg/vect/vect-mixed-size-cond-1.c: New test.
Jeff Law [Tue, 3 Jul 2018 05:14:10 +0000 (23:14 -0600)]
h8300.md (movqi_h8300, [...]): Consolidate the H8/300, H8/300H and H8/S variants into a single pattern.
* config/h8300/h8300.md (movqi_h8300, movqi_h8300hs): Consolidate
the H8/300, H8/300H and H8/S variants into a single pattern.
(movhi_h8300, movqi_h8300hs): Similarly.
(pushqi_h8300hs, pushhi_h8300hs): Consolidate into a single pattern.
(QHI mode iterator): New.
Jonathan Wakely [Mon, 2 Jul 2018 20:31:11 +0000 (21:31 +0100)]
Optimize std::sub_match comparisons using string_view-like type
Avoid creation of unnecessary basic_string objects by using a simplified
string_view type and performing comparisons on that type instead. A
temporary basic_string object is still used when the sub_match's
iterators are not contiguous, in order to get an object that the
__string_view can reference.
* include/bits/regex.h (sub_match::operator string_type): Call str().
(sub_match::compare): Use _M_str() instead of str().
(sub_match::_M_compare): New public function.
(sub_match::__string_view): New helper type.
(sub_match::_M_str): New overloaded functions to avoid creating a
string_type object when not needed.
(operator==, operator!=, operator<, operator>, operator<=, operator>=):
Use sub_match::_M_compare instead of creating string_type objects.
Fix Doxygen comments.
* include/bits/regex_compiler.h (__has_contiguous_iter): Remove.
(__is_contiguous_normal_iter): Rename to __is_contiguous_iter and
simplify.
(__enable_if_contiguous_iter, __disable_if_contiguous_iter): Use
__enable_if_t.
* include/std/type_traits (__enable_if_t): Define for C++11.
* testsuite/28_regex/sub_match/compare.cc: New.
* testsuite/util/testsuite_iterators.h (remove_cv): Add transformation
trait.
(input_iterator_wrapper): Use remove_cv for value_type argument of
std::iterator base class.
David Malcolm [Mon, 2 Jul 2018 20:05:21 +0000 (20:05 +0000)]
selftest: introduce class auto_fix_quotes
This patch moves a workaround for locale differences from a
selftest in pretty-print.c to selftest.h/c to make it reusable; I need
this for a selftest in a followup patch.
gcc/ChangeLog:
* pretty-print.c (selftest::test_pp_format): Move save and restore
of quotes to class auto_fix_quotes, and add an instance.
* selftest.c: Include "intl.h".
(selftest::auto_fix_quotes::auto_fix_quotes): New ctor.
(selftest::auto_fix_quotes::~auto_fix_quotes): New dtor.
* selftest.h (selftest::auto_fix_quotes): New class.
re PR go/86331 (the gccgo's "go" tool looks like failing to invoke any sub go command)
PR go/86331
os: check return value as well as error from waitid
https://gcc.gnu.org/PR86331 indicates that if a signal handler runs it
is possible for syscall.Syscall6 to return a non-zero errno value even
if no error occurs. That is a problem in general, but this fix will
let us work around the general problem for the specific case of
calling waitid.
* config/aarch64/aarch64-protos.h, config/aarch64/aarch64.c
(aarch64_sve_prepare_conditional_op): Remove.
* config/aarch64/aarch64-sve.md (cond_<SVE_INT_BINARY><SVE_I>):
Allow aarch64_simd_reg_or_zero as select operand; remove
the aarch64_sve_prepare_conditional_op call.
(cond_<SVE_INT_BINARY_SD><SVE_SDI>): Likewise.
(cond_<SVE_COND_FP_BINARY><SVE_F>): Likewise.
(*cond_<SVE_INT_BINARY><SVE_I>_z): New pattern.
(*cond_<SVE_INT_BINARY_SD><SVE_SDI>_z): New pattern.
(*cond_<SVE_COND_FP_BINARY><SVE_F>_z): New pattern.
(*cond_<SVE_INT_BINARY><SVE_I>_any): New pattern.
(*cond_<SVE_INT_BINARY_SD><SVE_SDI>_any): New pattern.
(*cond_<SVE_COND_FP_BINARY><SVE_F>_any): New pattern
and a splitters to match all of the *_any patterns.
* config/aarch64/predicates.md (aarch64_sve_any_binary_operator): New.
Paul Thomas [Mon, 2 Jul 2018 07:24:43 +0000 (07:24 +0000)]
re PR fortran/45305 (Array-valued calles to elementals are not simplified)
2018-07-02 Paul Thomas <pault@gcc.gnu.org>
PR fortran/45305
* expr.c : Add a prototype for scalarize_intrinsic_call.
(gfc_simplify_expr): Use scalarize_intrinsic_call for elemental
intrinsic function calls.
(scalarize_intrinsic_call): Add 'init_flag' argument. Check if
the expression or any of the actual argument expressions are
NULL. Before calling gfc_check_init_expr, check 'init_flag'.
Only simplify the scalarized expressions if there are no errors
on the stack.
(gfc_check_init_expr): Set 'init_flag' true in the call to
scalarize_intrinsic_call.
2018-07-02 Paul Thomas <pault@gcc.gnu.org>
PR fortran/45305
* gfortran.dg/scalarize_parameter_array_2.f90: New test.
Paul Thomas [Mon, 2 Jul 2018 07:20:27 +0000 (07:20 +0000)]
re PR fortran/45305 (Array-valued calles to elementals are not simplified)
2018-07-02 Paul Thomas <pault@gcc.gnu.org>
PR fortran/45305
* expr.c : Add a prototype for scalarize_intrinsic_call.
(gfc_simplify_expr): Use scalarize_intrinsic_call for elemental
intrinsic function calls.
(scalarize_intrinsic_call): Add 'init_flag' argument. Check if
the expression or any of the actual argument expressions are
NULL. Before calling gfc_check_init_expr, check 'init_flag'.
Only simplify the scalarized expressions if there are no errors
on the stack.
(gfc_check_init_expr): Set 'init_flag' true in the call to
scalarize_intrinsic_call.
2018-07-02 Paul Thomas <pault@gcc.gnu.org>
PR fortran/45305
* gfortran.dg/scalarize_parameter_array_2.f90: New test.
David Malcolm [Mon, 2 Jul 2018 03:39:48 +0000 (03:39 +0000)]
Reinstate dump_generic_expr_loc
gcc/ChangeLog:
* dumpfile.c (dump_generic_expr_loc): Undo removal of this
function in r262149, changing "loc" param from source_location to
const dump_location_t &.
* dumpfile.h (dump_generic_expr_loc): Undo removal of this
declaration, as above.
Paul Koning [Sun, 1 Jul 2018 19:19:39 +0000 (15:19 -0400)]
Add -mgnu-asm; change -mdec-asm to generate DEC assembler
compatible output.
* common/config/pdp11/pdp11-common.c (pdp11_handle_option): Handle
-munit-asm, -mgnu-asm, -mdec-asm.
* config/pdp11/pdp11-protos.h (pdp11_gen_int_label): New.
(pdp11_output_labelref): New.
(pdp11_output_def): New.
(pdp11_output_addr_vec_elt): New.
* config/pdp11/pdp11.c: Use tab between opcode and operands. Use
%# and %@ format codes.
(pdp11_option_override): New.
(TARGET_ASM_FILE_START_FILE_DIRECTIVE): Define.
(pdp11_output_ident): New.
(pdp11_asm_named_section): New.
(pdp11_asm_init_sections): New.
(pdp11_file_start): New.
(pdp11_file_end): New.
(output_ascii): Use .ascii/.asciz for -mdec-asm.
(pdp11_asm_print_operand): Update %# and %$ for -mdec-asm. Add
%o, like %c but octal.
(pdp11_option_override): New.
* config/pdp11/pdp11.h (TEXT_SECTION_ASM_OP): Update for
-mdec-asm.
(DATA_SECTION_ASM_OP): Ditto.
(READONLY_DATA_SECTION_ASM_OP): New.
(IS_ASM_LOGICAL_LINE_SEPARATOR): New.
(ASM_GENERATE_INTERNAL_LABEL): Use new function.
(ASM_OUTPUT_LABELREF): Ditto.
(ASM_OUTPUT_DEF): Ditto.
(ASM_OUTPUT_EXTERNAL): New.
(ASM_OUTPUT_SOURCE_FILENAME): New.
(ASM_OUTPUT_ADDR_VEC_ELT): Use new function.
(ASM_OUTPUT_SKIP): Update for -mdec-asm.
* config/pdp11/pdp11.md: Use tab between opcode and operands. Use
%# and %@ format codes.
* config/pdp11/pdp11.opt (mgnu-asm): New.
(mdec-asm): Conflicts with -mgnu-asm and -munix-asm.
(munix-asm): Conflicts with -mdec-asm and -mgnu-asm.
* doc/invoke.txt (PDP-11 Options): Add -mgnu-asm.
Tom de Vries [Sun, 1 Jul 2018 18:28:24 +0000 (18:28 +0000)]
[testsuite/guality] Use @main as bp loc instead of line nrs in const-volatile.c
2018-07-01 Tom de Vries <tdevries@suse.de>
* lib/gcc-gdb-test.exp (gdb-test): Handle '@' prefix in line number
argument.
* gcc.dg/guality/const-volatile.c: Replace gdb-test line nrs 50 and 58
with @main.
Tom de Vries [Sun, 1 Jul 2018 14:56:09 +0000 (14:56 +0000)]
[testsuite] Fix get-absolute-line error handling
2018-07-01 Tom de Vries <tdevries@suse.de>
* gcc.dg-selftests/dg-final.exp (verify_call_1): Factor out of ...
(verify_call): ... here. Move to toplevel.
(verify_call_np, dg_final_directive_check_utils): New proc.
(toplevel): Call dg_final_directive_check_utils.
* lib/gcc-dg.exp (get-absolute-line): Fix typo in variable reference.
This patch reuses the same idea for casts of invariants created
during widening optimisations.
One hitch was that vect_loop_versioning asserted that the vector loop
preheader was still empty, although the cfg transformation it's doing
should be correct either way.
2018-06-30 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-patterns.c (vect_get_external_def_edge): New function,
split out from...
(vect_recog_rotate_pattern): ...here.
(vect_convert_input): Try to insert casts of invariants in the
preheader.
* tree-vect-loop-manip.c (vect_loop_versioning): Don't require the
preheader to be empty.
gcc/testsuite/
* gcc.dg/vect/vect-widen-mult-extern-1.c: New test.
[12/n] PR85694: Rework detection of widened operations
This patch adds helper functions for detecting widened operations and
generalises the existing code to handle more cases.
One of the main changes is to recognise multi-stage type conversions,
which are possible even in the original IR and can also occur as a
result of earlier pattern matching (especially after the main
over-widening patch). E.g. for:
unsigned int res = 0;
for (__INTPTR_TYPE__ i = 0; i < N; ++i)
{
int av = a[i];
int bv = b[i];
short diff = av - bv;
unsigned short abs = diff < 0 ? -diff : diff;
res += abs;
}
where the first cast establishes the sign of the promotion done
by the second cast.
vect_recog_sad_pattern didn't handle this kind of intermediate promotion
between the MINUS_EXPR and the ABS_EXPR. Sign extensions and casts from
unsigned to signed are both OK there. Unsigned promotions aren't, and
need to be rejected, but should have been folded away earlier anyway.
Also, the dot_prod and widen_sum patterns both required the promotions
to be from one signedness to the same signedness, rather than say signed
char to unsigned int. That shouldn't be necessary, since it's only the
sign of the input to the promotion that matters. Nothing requires the
narrow and wide types in a DOT_PROD_EXPR or WIDEN_SUM_EXPR to have the
same sign (and IMO that's a good thing).
Fixing these fixed an XFAIL in gcc.dg/vect/vect-widen-mult-sum.c.
vect_widened_op_tree is a bit more general than the current patch needs,
since it copes with a tree of operations rather than a single statement.
This is used by the later average-detection patch.
The patch also uses a common routine to handle both the WIDEN_MULT_EXPR
and WIDEN_LSHIFT_EXPR patterns. I hope this could be reused for other
similar operations in future.
Also, the patch means we recognise the index calculations in
vect-mult-const-pattern*.c as widening multiplications, whereas the
scan test was expecting them to be recognised as mult patterns instead.
The patch makes the tests check specifically for the multiplication we
care about.
2018-06-30 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-patterns.c (append_pattern_def_seq): Take an optional
vector type. If given, install it in the new statement's
STMT_VINFO_VECTYPE.
(vect_element_precision): New function.
(vect_unpromoted_value): New struct.
(vect_unpromoted_value::vect_unpromoted_value): New function.
(vect_unpromoted_value::set_op): Likewise.
(vect_look_through_possible_promotion): Likewise.
(vect_joust_widened_integer, vect_joust_widened_type): Likewise.
(vect_widened_op_tree, vect_convert_input): Likewise.
(vect_convert_inputs, vect_convert_output): Likewise.
(vect_recog_dot_prod_pattern): Use vect_look_through_possible_promotion
to handle the optional cast of the multiplication result and
vect_widened_op_tree to detect the widened multiplication itself.
Do not require the input and output of promotion casts to have
the same sign, but base the signedness of the operation on the
input rather than the result. If the pattern includes two
promotions, check that those promotions have the same sign.
Do not restrict the MULT_EXPR handling to a double-width result;
handle quadruple-width results and wider. Use vect_convert_inputs
to convert the inputs to the common type.
(vect_recog_sad_pattern): Use vect_look_through_possible_promotion
to handle the optional cast of the ABS result. Also allow a sign
change or a sign extension between the ABS and MINUS.
Use vect_widened_op_tree to detect the widened subtraction and use
vect_convert_inputs to convert the inputs to the common type.
(vect_handle_widen_op_by_const): Delete.
(vect_recog_widen_op_pattern): New function.
(vect_recog_widen_mult_pattern): Use it.
(vect_recog_widen_shift_pattern): Likewise.
(vect_recog_widen_sum_pattern): Use
vect_look_through_possible_promotion to handle the promoted
PLUS_EXPR operand.
gcc/testsuite/
* gcc.dg/vect/vect-widen-mult-sum.c: Remove xfail.
* gcc.dg/vect/no-scevccp-outer-6.c: Don't match widened multiplications
by 4 in the computation of a[i].
* gcc.dg/vect/vect-mult-const-pattern-1.c: Test specifically for the
main multiplication constant.
* gcc.dg/vect/vect-mult-const-pattern-2.c: Likewise.
* gcc.dg/vect/vect-widen-mult-const-s16.c: Likewise.
* gcc.dg/vect/vect-widen-mult-const-u16.c: Likewise. Expect the
pattern to cast the result to int.
* gcc.dg/vect/vect-reduc-dot-1.c: New test.
* gcc.dg/vect/vect-reduc-dot-2.c: Likewise.
* gcc.dg/vect/vect-reduc-dot-3.c: Likewise.
* gcc.dg/vect/vect-reduc-dot-4.c: Likewise.
* gcc.dg/vect/vect-reduc-dot-5.c: Likewise.
* gcc.dg/vect/vect-reduc-dot-6.c: Likewise.
* gcc.dg/vect/vect-reduc-dot-7.c: Likewise.
* gcc.dg/vect/vect-reduc-dot-8.c: Likewise.
* gcc.dg/vect/vect-reduc-sad-1.c: Likewise.
* gcc.dg/vect/vect-reduc-sad-2.c: Likewise.
* gcc.dg/vect/vect-reduc-sad-3.c: Likewise.
* gcc.dg/vect/vect-reduc-sad-4.c: Likewise.
* gcc.dg/vect/vect-reduc-sad-5.c: Likewise.
* gcc.dg/vect/vect-reduc-sad-6.c: Likewise.
* gcc.dg/vect/vect-reduc-sad-7.c: Likewise.
* gcc.dg/vect/vect-reduc-sad-8.c: Likewise.
* gcc.dg/vect/vect-widen-mult-1.c: Likewise.
* gcc.dg/vect/vect-widen-mult-2.c: Likewise.
* gcc.dg/vect/vect-widen-mult-3.c: Likewise.
* gcc.dg/vect/vect-widen-mult-4.c: Likewise.
[11/n] PR85694: Apply pattern matching to pattern definition statements
Although the first pattern match wins in the sense that no later
function can match the *old* gimple statement, it still seems worth
letting them match the *new* gimple statements, just like we would if
the original IR had included that sequence from the outset.
This is mostly true after the later patch for PR85694, where e.g. we
could recognise:
signed char a;
int ap = (int) a;
int res = ap * 3;
as the pattern:
short ap' = (short) a;
short res = ap' * 3; // S1: definition statement
int res = (int) res; // S2: pattern statement
and then apply the mult pattern to "ap' * 3". The patch needs to
come first (without its own test cases) so that the main over-widening
patch doesn't regress anything.
2018-06-30 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* gimple-iterator.c (gsi_for_stmt): Add a new overload that takes
the containing gimple_seq *.
* gimple-iterator.h (gsi_for_stmt): Declare it.
* tree-vect-patterns.c (vect_recog_dot_prod_pattern)
(vect_recog_sad_pattern, vect_recog_widen_sum_pattern)
(vect_recog_widen_shift_pattern, vect_recog_rotate_pattern)
(vect_recog_vector_vector_shift_pattern, vect_recog_divmod_pattern)
(vect_recog_mask_conversion_pattern): Remove STMT_VINFO_IN_PATTERN_P
checks.
(vect_init_pattern_stmt, vect_set_pattern_stmt): New functions,
split out from...
(vect_mark_pattern_stmts): ...here. Handle cases in which the
statement being replaced is part of an existing pattern
definition sequence, inserting the new pattern statements before
the original one.
(vect_pattern_recog_1): Don't return a bool. If the statement
is already part of a pattern, instead apply pattern matching
to the pattern definition statements. Don't clear the
STMT_VINFO_RELATED_STMT if is_pattern_stmt_p.
(vect_pattern_recog): Don't break after the first match;
continue processing the pattern definition statements instead.
Don't bail out for STMT_VINFO_IN_PATTERN_P here.
[10/n] PR85694: Split out check for vectorizable associative reductions
This patch adds an overload of vect_reassociating_reduction_p
that checks for a vectorizable associative reduction,
since the check was duplicated in three functions.
2018-06-30 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
(vect_recog_dot_prod_pattern, vect_recog_sad_pattern)
(vect_recog_widen_sum_pattern): Use it.
[9b/n] PR85694: Make vect_is_simple_use look through pattern statements
As suggested by Richard B., this patch makes vect_is_simple_use check
whether a defining statement has been replaced by a pattern statement,
and if so returns the pattern statement instead.
The reason for doing this is that the main patch for PR85694
makes over_widening handle more general cases. These over-widened
patterns can still be useful when matching later statements;
e.g. an overwidened MULT_EXPR could be the input to a DOT_PROD_EXPR.
The patch doesn't do anything with the STMT_VINFO_IN_PATTERN_P checks
in vect_recog_over_widening_pattern or vect_recog_widen_shift_pattern
since later patches rewrite them anyway.
Doing this fixed an XFAIL in vect-reduc-dot-u16b.c.
2018-06-30 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-loop.c (vectorizable_reduction): Assert that the
phi is not a pattern statement and has not been replaced by
a pattern statement.
* tree-vect-patterns.c (type_conversion_p): Don't check
STMT_VINFO_IN_PATTERN_P.
(vect_recog_vector_vector_shift_pattern): Likewise.
(vect_recog_dot_prod_pattern): Expect vect_is_simple_use to return
the pattern statement rather than the original statement; check
directly for a WIDEN_MULT_EXPR here.
* tree-vect-slp.c (vect_get_and_check_slp_defs): Expect
vect_is_simple_use to return the pattern statement rather
than the original statement; use is_pattern_stmt_p to check
for such a pattern statement.
* tree-vect-stmts.c (process_use): Expect vect_is_simple_use
to return the pattern statement rather than the original statement;
don't do the same transformation here.
(vect_is_simple_use): If the defining statement has been replaced
by a pattern statement, return the pattern statement instead.
Remove the corresponding (local) transformation from the vectype
overload.
gcc/testsuite/
* gcc.dg/vect/vect-reduc-dot-u16b.c: Remove xfail and update the
test for vectorization along the lines described in the comment.
As suggested by Richard B., this patch reorders the arguments to
vect_is_simple_use so that def_stmt comes last and is optional.
Many callers can then drop it, making it more obvious which of
the remaining calls would be affected by the next patch.
2018-06-30 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vectorizer.h (vect_is_simple_use): Move the gimple ** to the
end and default to null.
* tree-vect-loop.c (vect_create_epilog_for_reduction)
(vectorizable_reduction): Update calls accordingly, dropping the
gimple ** argument if the passed-back statement isn't needed.
* tree-vect-patterns.c (vect_get_internal_def, type_conversion_p)
(vect_recog_rotate_pattern): Likewise.
(vect_recog_mask_conversion_pattern): Likewise.
* tree-vect-slp.c (vect_get_and_check_slp_defs): Likewise.
(vect_mask_constant_operand_p): Likewise.
* tree-vect-stmts.c (is_simple_and_all_uses_invariant, process_use):
(vect_model_simple_cost, vect_get_vec_def_for_operand): Likewise.
(get_group_load_store_type, get_load_store_type): Likewise.
(vect_check_load_store_mask, vect_check_store_rhs): Likewise.
(vectorizable_call, vectorizable_simd_clone_call): Likewise.
(vectorizable_conversion, vectorizable_assignment): Likewise.
(vectorizable_shift, vectorizable_operation): Likewise.
(vectorizable_store, vect_is_simple_cond): Likewise.
(vectorizable_condition, vectorizable_comparison): Likewise.
(get_same_sized_vectype, vect_get_mask_type_for_stmt): Likewise.
(vect_is_simple_use): Rename the def_stmt argument to def_stmt_out
and move it to the end. Cope with null def_stmt_outs.
Tom de Vries [Sat, 30 Jun 2018 08:09:26 +0000 (08:09 +0000)]
[testsuite/guality] Don't use attribute used in pr45882.c for -O0
2018-06-30 Tom de Vries <tdevries@suse.de>
* gcc.dg/guality/guality.exp (guality_transform_options): New proc.
(toplevel): Apply guality_transform_options on DG_TORTURE_OPTIONS and
LTO_TORTURE_OPTIONS.
* gcc.dg/guality/prevent-optimization.h: New file.
* gcc.dg/guality/pr45882.c: Include prevent-optimization.h.
(a): Replace __attribute__((used)) with ATTRIBUTE_USED.
Tom de Vries [Sat, 30 Jun 2018 08:07:01 +0000 (08:07 +0000)]
[testsuite/guality] Use relative line numbers in gdb-test
2018-06-30 Tom de Vries <tdevries@suse.de>
* gcc.dg/guality/pr45882.c (foo): Use relative line numbers.
* lib/gcc-dg.exp (dg-final): New proc.
* lib/gcc-gdb-test.exp (gdb-test): Add and handle additional line number
argument.
Alexandre Oliva [Sat, 30 Jun 2018 02:47:29 +0000 (02:47 +0000)]
Introduce @unless/@endunless and postbootstrap Makefile targets
This patch turns dependencies of non-bootstrap targets on bootstrap
targets for bootstrap builds into dependencies on stage_last. This
arrangement gets stage1-bubble to run from stage_last if we haven't
started a bootstrap yet, and to use the current stage otherwise. This
was already the case of target libs, just not of non-bootstrapped host
modules.
In order to retain preexisting dependencies in non-bootstrap builds,
or in gcc-less builds, this introduces support for @unless/@endunless
pairs in Makefile.in.
There is a remaining possibility of problem if activating, in a tree
configured for bootstrap, a parallel build of two or more modules, at
least one bootstrapped and one not. In this case, make might decide
to build stage_current and stage_last in parallel, the latter will
start a submake to build stage1 while the initial make, having
satisfied stage_current, proceeds to build the bootstrapped module in
non-bootstrapped configurations. The two builds will overlap and will
likely conflict. This situation does NOT arise in normal settings,
however: a post-bootstrap build of all-host all-target will indeed
activate such targets concurrently, but only after building all
bootstrapped modules successfully, and it will have both stage_last
and stage_current targets already satisfied, so the potential race
between builds will not arise.
Another remaining problem, that is slightly expanded with this patch,
is that of an interrupted build in a tree configured for bootstrap,
continued with a non-bootstrapped target. Target modules that were
not bootstrapped would already fail to complete the current stage when
activated explicitly in the command line for a retry; host modules,
however, would attempt to build their bootstrapped dependencies, which
is what led to the problem of concurrent builds addressed with this
patch. An interrupted or failed build might still recover correctly,
if the non-bootstrapped target is activated in both builds, because
then make will remove stage_last when its build command is
interrupted, so that it will attempt to recreate it with stage1-bubble
in the second try. A bootstrap build, however, will not be attempting
to build stage_last, so the file will remain and the retry won't go
through stage1-bubble. We have lived with that for target modules, so
we can probably live with that for host modules too.
Another undesirable consequence of this change is that non-boostrapped
host modules, in a tree configured for bootstrap, when activated as
make all-<module>, will build all of stage1 instead of only the
module's usual dependencies. This is intentional and necessary to fix
the parallel-build problem. If it's not desirable, disabling the
unnecessary bootstrap configuration will suffice to restore the
original set of dependencies.
for ChangeLog
* configure.ac: Introduce support for @unless/@endunless.
* Makefile.tpl (dep-kind): Rewrite with cond; return
postbootstrap in some cases.
(make-postboot-dep, postboot-targets): New.
(dependencies): Do not output postbootstrap dependencies at
first. Output non-target ones changed for configure to depend
on stage_last @if gcc-bootstrap, and the original deps @unless
gcc-bootstrap.
* configure.in, Makefile.in: Rebuilt.
Kyrylo Tkachov [Fri, 29 Jun 2018 13:36:35 +0000 (13:36 +0000)]
[arm] Avoid STRD with odd register for TARGET_ARM in output_move_double
In this testcase the user forces an odd register as the starting reg for a DFmode value.
The output_move_double function tries to store that using an STRD instruction.
But for TARGET_ARM the starting register of an STRD must be an even one.
This is always the case with compiler-allocated registers for DFmode values, but the
inline assembly forced our hand here.
This patch restricts the STRD-emitting logic in output_move_double to not avoid
odd-numbered source registers in STRD.
I'm not a fan of the whole function, we should be exposing a lot of the logic in there
to RTL rather than at the final output stage, but that would need to be fixed separately.
* config/arm/arm.c (output_move_double): Don't allow STRD instructions
if starting source register is not even.
contrib: introduce Vim addon directory, add match.pd syntax plugin
* vim-gcc-dev/README: New file.
* vim-gcc-dev/ftdetect/gcc-dev.vim: New file.
* vim-gcc-dev/syntax/gcc-match.vim: New file.
* gimple.vim: Move under vim-gcc-dev/syntax/.
* gcc-rtl.vim: Likewise.
Tom de Vries [Fri, 29 Jun 2018 11:25:49 +0000 (11:25 +0000)]
[testsuite/guality] Use line number vars in gdb-test
2018-06-29 Tom de Vries <tdevries@suse.de>
* gcc.dg/guality/pr45882.c (foo): Add line number var for breakpoint
line, and use it.
* lib/gcc-dg.exp (get-absolute-line): Factor out of ...
(process-message): ... here.
* lib/gcc-gdb-test.exp (gdb-test): Use get-absolute-line.
David Malcolm [Fri, 29 Jun 2018 09:56:40 +0000 (09:56 +0000)]
dumpfile.c: add indentation via DUMP_VECT_SCOPE
This patch adds a concept of nested "scopes" to dumpfile.c's dump_*_loc
calls, and wires it up to the DUMP_VECT_SCOPE macro in tree-vectorizer.h,
so that the nested structure is shown in -fopt-info by indentation.
For example, this converts -fopt-info-all e.g. from:
test.c:8:3: note: === analyzing loop ===
test.c:8:3: note: === analyze_loop_nest ===
test.c:8:3: note: === vect_analyze_loop_form ===
test.c:8:3: note: === get_loop_niters ===
test.c:8:3: note: symbolic number of iterations is (unsigned int) n_9(D)
test.c:8:3: note: not vectorized: loop contains function calls or data references that cannot be analyzed
test.c:8:3: note: vectorized 0 loops in function
to:
test.c:8:3: note: === analyzing loop ===
test.c:8:3: note: === analyze_loop_nest ===
test.c:8:3: note: === vect_analyze_loop_form ===
test.c:8:3: note: === get_loop_niters ===
test.c:8:3: note: symbolic number of iterations is (unsigned int) n_9(D)
test.c:8:3: note: not vectorized: loop contains function calls or data references that cannot be analyzed
test.c:8:3: note: vectorized 0 loops in function
showing that the "symbolic number of iterations" message is within
the "=== analyze_loop_nest ===" (and not within the
"=== vect_analyze_loop_form ===").
This is also enabling work for followups involving optimization records
(allowing the records to directly capture the nested structure of the
dump messages).
gcc/ChangeLog:
* dumpfile.c (dump_loc): Add indentation based on scope depth.
(dump_scope_depth): New variable.
(get_dump_scope_depth): New function.
(dump_begin_scope): New function.
(dump_end_scope): New function.
* dumpfile.h (get_dump_scope_depth): New declaration.
(dump_begin_scope): New declaration.
(dump_end_scope): New declaration.
(class auto_dump_scope): New class.
(AUTO_DUMP_SCOPE): New macro.
* tree-vectorizer.h (DUMP_VECT_SCOPE): Reimplement in terms of
AUTO_DUMP_SCOPE.
David Malcolm [Thu, 28 Jun 2018 14:55:42 +0000 (14:55 +0000)]
Hide alt_dump_file within dumpfile.c
This patch removes alt_dump_file from dumpfile.h, making it static
within dumpfile.c. This allows for changing how -fopt-info is
implemented, and potentially adding other kinds of dump target, such
as remarks or optimization records.
Doing so requires changing the implementation of dump_enabled_p, so
the patch changes this to a simple lookup of a boolean global, which
is updated any time dump_file or alt_dump_file change.
gcc/ChangeLog:
* cgraph.c (cgraph_node::get_body): Replace assignments to
"dump_file" with calls to set_dump_file.
* dumpfile.c (alt_dump_file): Make static, and group with...
(alt_flags): ...this definition.
(dumps_are_enabled): New variable.
(refresh_dumps_are_enabled): New function.
(set_dump_file): New function.
(set_alt_dump_file): New function.
(gcc::dump_manager::dump_start): Replace assignments to
"dump_file" and "alt_dump_file" with calls to set_dump_file and
set_alt_dump_file.
(gcc::dump_manager::dump_finish): Likewise.
* dumpfile.h (alt_dump_file): Delete decl.
(dumps_are_enabled): New variable decl.
(set_dump_file): New function decl.
(dump_enabled_p): Rewrite in terms of new "dumps_are_enabled"
global.
* tree-nested.c (lower_nested_functions): Replace assignments to
"dump_file" with calls to set_dump_file.
Richard Biener [Thu, 28 Jun 2018 10:22:38 +0000 (10:22 +0000)]
dwarf2out.c (decl_scope_table): Remove.
2018-06-28 Richard Biener <rguenther@suse.de>
* dwarf2out.c (decl_scope_table): Remove.
(push_decl_scope): Likewise.
(pop_decl_scope): Likewise.
(gen_type_die_for_member): Do not call push/pop_decl_scope.
(gen_struct_or_union_type_die): Likewise.
(gen_tagged_type_die): Likewise.
(dwarf2out_init): Do not initialize decl_scope_table.
(dwarf2out_c_finalize): Do not free it.
projects / thirdparty / gcc.git / log
