Yannick Moy [Mon, 12 Aug 2019 08:59:42 +0000 (08:59 +0000)]
[Ada] More precise handling of Size/Object_Size in GNATprove
GNATprove does a partial expansion which did not allow getting the
most precise value for attributes Size/Object_Size. Now fixed.
There is no impact on compilation.
2019-08-12 Yannick Moy <moy@adacore.com>
gcc/ada/
* exp_attr.adb, exp_attr.ads (Expand_Size_Attribute): New
procedure to share part of the attribute expansion with
GNATprove mode.
(Expand_N_Attribute_Reference): Extract part of the
Size/Object_Size expansion in the new procedure
Expand_Size_Attribute.
* exp_spark.adb (Expand_SPARK_N_Attribute_Reference): Expand
Size/Object_Size attributes using the new procedure
Expand_Size_Attribute.
Justin Squirek [Mon, 12 Aug 2019 08:59:33 +0000 (08:59 +0000)]
[Ada] Inconsistent compile time Constraint_Error warning
This patch corrects several bugs within the compiler which led to
inconsistent handling of compile time Constraint_Errors. Notibly,
subtype out of range checks which are only out of range of the subtype
must be warnings while out of range checks where the value is out of
range of the base type must be an error. Also, type conversions and
qualified expressions on literals constitute errors on any out of range
value. The compiler needed many of these cases clarified.
------------
-- Source --
------------
-- main.ads
with System;
package Main is
type T_Enum is (Enum_1, Enum_2, Unknown)
with Default_Value => Unknown;
subtype T_Valid_Enum is T_Enum range Enum_1 .. Enum_2;
Value : T_Valid_Enum; -- WARNING
generic
type T_Element is (<>);
Init : T_Element;
package Generic_Test is
Value : T_Element := Init;
end;
package titi is new Generic_Test (T_Valid_Enum, Unknown); -- WARNING
$ gnatmake -q -gnatw_a main.adb
main.ads:9:12: warning: value not in range of type "T_Valid_Enum" defined at
line 7
main.ads:9:12: warning: "Constraint_Error" will be raised at run time
main.ads:18:52: warning: value not in range of type "T_Element" defined at
line 12, instance at line 18
main.ads:18:52: warning: "Constraint_Error" will be raised at run time
main.ads:25:23: value not in range of type "My_Float" defined at line 20
main.ads:25:23: static expression fails Constraint_Check
main.ads:27:19: value not in range of type "Standard.Positive"
main.ads:27:19: static expression fails Constraint_Check
main.ads:28:19: value not in range of type "Standard.Positive"
main.ads:28:19: static expression fails Constraint_Check
main.ads:29:19: warning: value not in range of type "Standard.Positive"
main.ads:29:19: warning: "Constraint_Error" will be raised at run time
main.ads:30:19: value not in range of type "Standard.Positive"
main.ads:30:19: static expression fails Constraint_Check
main.ads:31:27: value not in range of type "Standard.Positive"
main.ads:31:27: static expression fails Constraint_Check
main.ads:32:27: value not in range of type "Standard.Positive"
main.ads:32:27: static expression fails Constraint_Check
gnatmake: "main.ads" compilation error
2019-08-12 Justin Squirek <squirek@adacore.com>
gcc/ada/
* sem_eval.adb (Check_Non_Static_Context): Add a condition to
determine if a range violation constitues a warning or an error.
(Out_Of_Range): Add a condition to determine if a range
violation constitues a warning or an error.
Eric Botcazou [Mon, 12 Aug 2019 08:59:28 +0000 (08:59 +0000)]
[Ada] Eliminate redundant range checks on conversions
This gets rid of redundant range checks generated in 5 out of the 9
cases of scalar conversions, i.e. (integer, fixed-point, floating-point)
converted to (integer, fixed-point, floating-point).
The problem is that the Real_Range_Check routine rewrites the conversion
node into a conversion to the base type so, when its parent node is
analyzed, a new conversion to the subtype may be introduced, depending
on the context, giving rise to a second range check against the subtype
bounds.
This change makes Real_Range_Check rewrite the expression of the
conversion node instead of the node, so that the type of the node is
preserved and no new conversion is introduced. As a matter of fact,
this is exactly what happens in the float-to-float case which goes to
the Generate_Range_Check circuit instead and does not suffer from the
duplication of range checks.
For the following procedure, the compiler must now generate exactly one
range check per nested function:
procedure P is
type I1 is new Integer range -100 .. 100;
type I2 is new Integer range -200 .. 200;
type D1 is delta 0.5 range -100.0 .. 100.0;
type D2 is delta 0.5 range -200.0 .. 200.0;
type F1 is new Long_Float range -100.0 .. 100.0;
type F2 is new Long_Float range -200.0 .. 200.0;
function Conv (A : I2) return I1 is
begin
return I1 (A);
end;
function Conv (A : D2) return I1 is
begin
return I1 (A);
end;
function Conv (A : F2) return I1 is
begin
return I1 (A);
end;
function Conv (A : I2) return D1 is
begin
return D1 (A);
end;
function Conv (A : D2) return D1 is
begin
return D1 (A);
end;
function Conv (A : F2) return D1 is
begin
return D1 (A);
end;
function Conv (A : I2) return F1 is
begin
return F1 (A);
end;
function Conv (A : D2) return F1 is
begin
return F1 (A);
end;
function Conv (A : F2) return F1 is
begin
return F1 (A);
end;
begin
null;
end;
2019-08-12 Eric Botcazou <ebotcazou@adacore.com>
gcc/ada/
* exp_ch4.adb (Real_Range_Check): Do not rewrite the conversion
node but its expression instead, after having fetched its
current value. Clear the Do_Range_Check flag on entry. Return
early for a rewritten float-to-float conversion. Remove
redundant local variable. Suppress all checks when inserting
the temporary and do not reanalyze the node.
Eric Botcazou [Mon, 12 Aug 2019 08:59:18 +0000 (08:59 +0000)]
[Ada] Fix leak of Do_Range_Check flag in -gnatVa mode
This fixes a small glitch in Insert_Valid_Check, which needs to
propagate the Do_Range_Check flag onto the rewritten expression, but
uses its Original_Node as the source of the copy. Now Original_Node
does not necessarily point to the node that was just rewritten, but to
the ultimately original node, which is not the same node if the
expression was rewritten multiple times. The end result is that a
stalled Do_Range_Check flag can be wrongly resintated and leak to the
code generator.
2019-08-12 Eric Botcazou <ebotcazou@adacore.com>
gcc/ada/
* checks.adb (Insert_Valid_Check): Do not retrieve the
Do_Range_Check flag from the Original_Node but from the
Validated_Object. Remove useless bypass for floating-point
types.
Yannick Moy [Mon, 12 Aug 2019 08:59:13 +0000 (08:59 +0000)]
[Ada] Extended traversal subprograms for GNATprove
GNATprove needs traversal subprograms that do not simply traverse
syntactic nodes like Atree.Traverse_Func and Atree.Traverse_Proc, but
also traverse semantic nodes which are logically children of the nodes.
Now available through Sem_Util.Traverse_More_Func and
Sem_Util.Traverse_More_Proc.
There is no impact on compilation.
2019-08-12 Yannick Moy <moy@adacore.com>
gcc/ada/
* sem_util.adb, sem_util.ads (Traverse_More_Func,
Traverse_More_Proc): New traversal subprograms.
Eric Botcazou [Mon, 12 Aug 2019 08:58:57 +0000 (08:58 +0000)]
[Ada] Add special bypass for obsolete code pattern
This change prevents the analysis phase of the front-end from setting
the Do_Range_Check flag in the very peculiar case of the source of a
conversion whose result is passed by reference to a "valued procedure",
because the expansion phase would not be able to generate the check.
This pattern appears in the ancient DEC Starlet package and it doesn't
seem to be useful at this point to change the expander to deal with it,
so instead the analysis phase is adjusted. Morever the compiler already
issues a warning in this case so this is probably good enough.
2019-08-12 Eric Botcazou <ebotcazou@adacore.com>
gcc/ada/
* sem_res.adb: Add with & use clause for Sem_Mech and
alphabetize.
(Resolve_Actuals): Do not apply a scalar range check for the
source of a conversion whose result is passed by reference to a
valued procedure.
Eric Botcazou [Mon, 12 Aug 2019 08:58:52 +0000 (08:58 +0000)]
[Ada] Fix missing range check for In/Out parameter with -gnatVa
This plugs another small loophole in the front-end which fails to
generate a range check for a scalar In/Out parameter when -gnatVa is
specified. This also fixes a few more leaks of the Do_Range_Check flag
on actual parameters, both in regular and -gnatVa modes, as well as a
leak specific to expression function in -gnatp mode.
2019-08-12 Eric Botcazou <ebotcazou@adacore.com>
gcc/ada/
* checks.adb (Insert_Valid_Check): Reset the Do_Range_Check flag
on the validated object.
* exp_ch6.adb (Add_Call_By_Copy_Code): Reset the Do_Range_Check
flag on the actual here, as well as on the Expression if the
actual is a N_Type_Conversion node.
(Add_Validation_Call_By_Copy_Code): Generate the incoming range
check if needed and reset the Do_Range_Check flag on the
Expression if the actual is a N_Type_Conversion node.
(Expand_Actuals): Do not reset the Do_Range_Check flag here.
Generate the incoming range check for In parameters here instead
of...
(Expand_Call_Helper): ...here. Remove redudant condition.
* sem_res.adb (Resolve_Actuals): Use local variable A_Typ and
remove obsolete comments.
(Resolve_Type_Conversion): Do not force the Do_Range_Check flag
on the operand if range checks are suppressed.
Eric Botcazou [Mon, 12 Aug 2019 08:58:46 +0000 (08:58 +0000)]
[Ada] Fix incorrect Do_Range_Check on type conversion
This gets rid of another leak of the Do_Range_Check flag to the back-end
which is specific to expression functions. No functional changes.
2019-08-12 Eric Botcazou <ebotcazou@adacore.com>
gcc/ada/
* checks.adb (Activate_Range_Check): Remove redundant argument.
(Generate_Range_Check): Likewise.
(Apply_Float_Conversion_Check): Reset the Do_Range_Check flag on
entry and remove redundant condition.
Janne Blomqvist [Sun, 11 Aug 2019 09:42:41 +0000 (12:42 +0300)]
PR fortran/91413 Generate warning when making array static
When moving a local variable from the stack to static storage, the
procedure is no longer safe to be called recursively or concurrently
from multiple threads. Thus generate a warning when this is done.
Also double the default limit for switching from stack to static.
Regtested on x86_64-pc-linux-gnu.
gcc/fortran/ChangeLog:
2019-08-11 Janne Blomqvist <jb@gcc.gnu.org>
PR fortran/91413
* invoke.texi (-fmax-stack-var-size): Document increased default.
* options.c (gfc_post_options): Increase default stack var size to
65536 bytes.
* trans-decl.c (gfc_finish_var_decl): Generate warning when local
array moved to static storage.
Iain Buclaw [Sun, 11 Aug 2019 06:53:14 +0000 (06:53 +0000)]
d: Fix ICE: gimplification failed (gimplify.c at 13436)
The expression that caused the ICE
++(a += 1.0);
The D front-end rewrites and applies implicit type conversions so the
expression gets simplified as
(int)((double) a += 1.0) += 1
The codegen pass would subsequently generate the following invalid code
(int)(double) a = (int)((double) a + 1.0) + 1
The LHS expression `(int)(double) a', represented as a FIX_TRUNC_EXPR
being what trips as it is not a valid lvalue for assignment.
While LHS casts are stripped away, convert_expr adds a double cast
because it converts the expression to its original type before
converting it to its target type. There is no valid reason why this is
done, so it has been removed.
gcc/d/ChangeLog:
PR d/90601
* d-convert.cc (convert_expr): Don't convert an expression to its
original front-end type before converting to its target type.
Steven G. Kargl [Sat, 10 Aug 2019 18:26:13 +0000 (18:26 +0000)]
decl.c (match_old_style_init): Use a clearer error message.
2019-08-10 Steven G. Kargl <kargl@gcc.gnu.org>
* decl.c (match_old_style_init): Use a clearer error message.
* expr.c (gfc_check_assign): Update BOZ checking to provide a stricter
adherence to the Fortran standard. Use gfc_invalid_boz () to
relax errors into warnings.
* gfortran.h (gfc_isym_id): Add new ids GFC_ISYM_DFLOAT,
GFC_ISYM_FLOAT, GFC_ISYM_REALPART, and GFC_ISYM_SNGL
* intrinsic.c (add_functions): Use new ids to split REAL generic into
REAL, FLOAT, DFLOAT, SNGL, and REALPART generics.
(gfc_intrinsic_func_interface): Allow new intrinsics in an
initialization expression
* resolve.c (resolve_operator): Deal with BOZ as operands.
Use gfc_invalid_boz to allow for errors or warnings via the
-fallow-invalid-boz option. A BOZ cannot be an operand to an
unary operator. Both operands of a binary operator cannot be BOZ.
For binary operators, convert a BOZ operand into the type and
kind of the other operand for REAL or INTEGER operand.
* trans-intrinsic.c: Use new ids to cause conversions to happen.
2019-08-10 Steven G. Kargl <kargl@gcc.gnu.org>
* gfortran.dg/boz_8.f90: Adjust error messages.
* gfortran.dg/nan_4.f90: Ditto.
* gfortran.dg/boz_1.f90: Add -fallow-invalid-boz to dg-options,
and test for warnings.
* gfortran.dg/boz_3.f90: Ditto.
* gfortran.dg/boz_4.f90: Ditto.
* gfortran.dg/dec_structure_6.f90: Ditto.
* gfortran.dg/ibits.f90: Ditto.
This fixes two minor problems with the new testcases. The first is
that almost all other tests, including all vec* tests, for powerpc use
names with dashes, not underscores. The more important one is the the
vec-rotate-1.c and vec-rotate-3.c tests need the -maltivec flag.
[aarch64] PR target/91386 Use copy_rtx to avoid modifying original insns in peep2 pattern
PR target/91386 is a situation where a peephole2 pattern substitution
is discarded late because the selected instructions contain
frame-related notes that we cannot redistribute (because the pattern
has more than one insn in the output). Unfortunately, the original
insns were being modified during the generation, so after the undo we
are left with corrupt RTL.
We avoid this by ensuring that the modifications are always made on a
copy, so that the original insns are never changed.
PR target/91386
* config/aarch64/aarch64.c (aarch64_gen_adjusted_ldpstp): Use copy_rtx
to preserve the contents of the original insns.
[arm] Recognize thumb2 16-bit variants of the add and compare instructions
The addsi3_compare_op[12] patterns currently only have constraints to
pick the 32-bit variants of the instructions. Although the assembler
may sometimes opportunistically match a 16-bit t2 instruction, there's
no real control over that within the compiler. Consequently we might
emit a 32-bit adds instruction with a 16-bit subs instruction would
serve equally well. We do, of course still have to be careful about
the small number of boundary cases by controlling the order quite
carefully.
This patch adds the constraints and templates to match the t2 16-bit
variants of these instructions. Now, for example, we can generate
Alexandre Oliva [Fri, 9 Aug 2019 09:20:58 +0000 (09:20 +0000)]
skip Cholesky decomposition in is>>n_mv_dist
normal_mv_distribution maintains the variance-covariance matrix param
in Cholesky-decomposed form. Existing param_type constructors, when
taking a full or lower-triangle varcov matrix, perform Cholesky
decomposition to convert it to the internal representation. This
internal representation is visible both in the varcov() result, and in
the streamed-out representation of a normal_mv_distribution object.
The problem is that when that representation is streamed back in, the
read-back decomposed varcov matrix is used as a lower-triangle
non-decomposed varcov matrix, and it undergoes Cholesky decomposition
again. So, each cycle of stream-out/stream-in changes the varcov
matrix to its "square root", instead of restoring the original
params.
This patch includes Corentin's changes that introduce verification in
testsuite/ext/random/normal_mv_distribution/operators/serialize.cc and
other similar tests that the object read back in compares equal to the
written-out object: the modified tests pass only if (u == v).
This patch also fixes the error exposed by his change, introducing an
alternate private constructor for param_type, used only by operator>>.
for libstdc++-v3/ChangeLog
* include/ext/random
(normal_mv_distribution::param_type::param_type): New private
ctor taking a decomposed varcov matrix, for use by...
(operator>>): ... this, befriended.
* include/ext/random.tcc (operator>>): Use it.
(normal_mv_distribution::param_type::_M_init_lower): Adjust
member function name in exception message.
for libstdc++-v3/ChangeLog
from Corentin Gay <gay@adacore.com>
* testsuite/ext/random/beta_distribution/operators/serialize.cc,
testsuite/ext/random/hypergeometric_distribution/operators/serialize.cc,
testsuite/ext/random/normal_mv_distribution/operators/serialize.cc,
testsuite/ext/random/triangular_distribution/operators/serialize.cc,
testsuite/ext/random/von_mises_distribution/operators/serialize.cc:
Add call to `VERIFY`.
Martin Liska [Fri, 9 Aug 2019 08:33:05 +0000 (10:33 +0200)]
Add -flto=auto option value.
2019-08-09 Martin Liska <mliska@suse.cz>
* doc/invoke.texi: Document the option value.
* lto-wrapper.c (run_gcc): Set auto_parallel
only with -flto=auto.
2019-08-09 Martin Liska <mliska@suse.cz>
Martin Liska [Fri, 9 Aug 2019 07:50:12 +0000 (09:50 +0200)]
Implement -fopt-info support for IPA ICF.
2019-08-09 Martin Liska <mliska@suse.cz>
* ipa-icf.c (sem_function::merge): Define AUTO_DUMP_SCOPE and
use dump_printf to report optimization.
(sem_variable::merge): Likwise.
(sem_item_optimizer::merge_classes): Use dump_printf to report
ICF hits.
2019-08-09 Martin Liska <mliska@suse.cz>
Jakub Jelinek [Fri, 9 Aug 2019 07:24:51 +0000 (09:24 +0200)]
c-parser.c (check_no_duplicate_clause): Simplify using omp_find_clause.
c/
* c-parser.c (check_no_duplicate_clause): Simplify using
omp_find_clause.
(c_parser_omp_clause_if): Fix up printing of target {enter,exit} data
directive name modifiers.
(c_parser_omp_clause_proc_bind): Check for duplicate proc_bind clause.
cp/
* parser.c (check_no_duplicate_clause): Simplify using
omp_find_clause.
(cp_parser_omp_clause_if): Fix up printing of target {enter,exit} data
directive name modifiers.
testsuite/
* c-c++-common/gomp/if-4.c: New test.
* c-c++-common/gomp/clause-dups-1.c: New test.
Jakub Jelinek [Fri, 9 Aug 2019 07:23:03 +0000 (09:23 +0200)]
re PR c/91401 (schedule + dist_schedule clauses rejected on distribute parallel for)
PR c/91401
c/
* c-parser.c (c_parser_omp_clause_dist_schedule): Fix up typos in the
check_no_duplicate_clause call. Comment it out, instead emit a
warning for duplicate dist_schedule clauses.
cp/
* parser.c (cp_parser_omp_clause_dist_schedule): Comment out the
check_no_duplicate_clause call, instead emit a warning for duplicate
dist_schedule clauses.
testsuite/
* c-c++-common/gomp/pr91401-1.c: New test.
* c-c++-common/gomp/pr91401-2.c: New test.
Alexandre Oliva [Fri, 9 Aug 2019 03:48:53 +0000 (03:48 +0000)]
use rand instead of random
rand is in ISO C, whereas random is only in POSIX, so it makes sense
to use the more portable function everywhere instead of falling back
from one to the other on systems that miss the less portable one.
for gcc/testsuite/ChangeLog
* gcc.target/i386/sse2-mul-1.c: Use rand. Drop fallback.
* gcc.target/i386/sse4_1-blendps-2.c: Likewise.
* gcc.target/i386/sse4_1-blendps.c: Likewise.
* gcc.target/i386/xop-vshift-1.c: Likewise.
* gcc.target/powerpc/direct-move.h: Likewise.
I noticed some patterns in dfp.md could use the D64_D128 iterator but
don't yet. This converts all remaining simple cases.
* config/rs6000/dfp.md (D64_D128): Move earlier in the file.
(dfp_suffix): Ditto.
(adddd3, addtd3): Merge to ...
(add<mode>3 for D64_D128): ... this.
(subdd3, subtd3): Merge to ...
(sub<mode>3 for D64_D128): ... this.
(muldd3, multd3): Merge to ...
(mul<mode>3 for D64_D128): ... this.
(divdd3, divtd3): Merge to ...
(div<mode>3 for D64_D128): ... this.
(*cmpdd_internal1, *cmptd_internal1): Merge to ...
(*cmp<mode>_internal1 for D64_D128): ... this.
(ftruncdd2, ftrunctd2): Merge to ...
(ftrunc<mode>2 for D64_D128): ... this.
(fixdddi2, fixtddi2): Merge to ...
(fix<mode>di2 for D64_D128): ... this.
Jim Wilson [Thu, 8 Aug 2019 19:04:56 +0000 (19:04 +0000)]
RISC-V: Fix C ABI for flattened struct with 0-length bitfield.
gcc/
PR target/91229
* config/riscv/riscv.c (riscv_flatten_aggregate_field): New arg
ignore_zero_width_bit_field_p. Skip zero size bitfields when true.
Pass into recursive call.
(riscv_flatten_aggregate_argument): New arg. Pass to
riscv_flatten_aggregate_field.
(riscv_pass_aggregate_in_fpr_pair_p): New local warned. Call
riscv_flatten_aggregate_argument twice, with false and true as last
arg. Process result twice. Compare results and warn if different.
(riscv_pass_aggregate_in_fpr_and_gpr_p): Likewise.
gcc/testsuite/
* gcc.target/riscv/flattened-struct-abi-1.c: New test.
* gcc.target/riscv/flattened-struct-abi-2.c: New test.
we emit a bogus error about nested definitions before the real error:
foo.c:2:6: error: nested redefinition of ‘enum a’
2 | enum a { B };
| ^
foo.c:2:6: error: redeclaration of ‘enum a’
foo.c:1:6: note: originally defined here
1 | enum a { A };
| ^
This is because we weren't clearing C_TYPE_BEING_DEFINED once the
definition was over.
I think it's OK to clear C_TYPE_BEING_DEFINED even for a definition
that actually is nested (and so whose outer definition is still open),
since we'll already have given an error by then. It means that second
and subsequent attempts to define a nested enum will usually get the
redeclaration error instead of the nested error, but that seems just
as accurate (nested_first and nested_second in the test). The only
exception is if the first nested enum was also invalid by being empty,
but then the enum as a whole has already produced two errors
(nested_empty in the test).
2019-08-08 Richard Sandiford <richard.sandiford@arm.com>
Jonathan Wakely [Thu, 8 Aug 2019 10:18:53 +0000 (11:18 +0100)]
P0325R4 to_array from LFTS with updates
As an extension to what the standard requires, this also adds
conditional noexcept-specifiers to the std::to_array functions.
P0325R4 to_array from LFTS with updates
* include/experimental/array (to_array): Qualify call to __to_array.
* include/std/array (__cpp_lib_to_array, to_array): Define for C++20.
* include/std/version (__cpp_lib_to_array): Likewise.
* testsuite/23_containers/array/creation/1.cc: New test.
* testsuite/23_containers/array/creation/2.cc: New test.
* testsuite/23_containers/array/creation/3_neg.cc: New test.
* testsuite/23_containers/array/tuple_interface/tuple_element_neg.cc:
Use zero for dg-error line number.
Jakub Jelinek [Thu, 8 Aug 2019 06:39:02 +0000 (08:39 +0200)]
gimplify.c (omp_add_variable): Use GOVD_PRIVATE | GOVD_EXPLICIT for VLA helper variables on target data even if...
* gimplify.c (omp_add_variable): Use GOVD_PRIVATE | GOVD_EXPLICIT
for VLA helper variables on target data even if not GOVD_FIRSTPRIVATE.
(gimplify_scan_omp_clauses): For OMP_CLAUSE_USE_DEVICE_* use just
GOVD_EXPLICIT flags.
(gimplify_omp_workshare): For OMP_TARGET_DATA move all
OMP_CLAUSE_USE_DEVICE_* clauses to the end of clauses chain.
* omp-low.c (scan_sharing_clauses): For OMP_CLAUSE_USE_DEVICE_*
call install_var_field with mask 11 instead of 3.
(lower_omp_target): For OMP_CLAUSE_USE_DEVICE_* use pass
(splay_tree_key) &DECL_UID (var) to build_sender_ref instead of var.
gcc/c/
* c-typeck.c (c_finish_omp_clauses): For C_ORT_OMP
OMP_CLAUSE_USE_DEVICE_* clauses use oacc_reduction_head bitmap
instead of generic_head to track duplicates.
gcc/cp/
* semantics.c (finish_omp_clauses): For C_ORT_OMP
OMP_CLAUSE_USE_DEVICE_* clauses use oacc_reduction_head bitmap
instead of generic_head to track duplicates.
libgomp/
* target.c (gomp_map_vars_internal): For GOMP_MAP_USE_DEVICE_PTR
perform the lookup in the first loop only if !not_found_cnt, otherwise
perform lookups for it in the second loop guarded with
if (not_found_cnt || has_firstprivate).
* testsuite/libgomp.c/target-37.c: New test.
* testsuite/libgomp.c++/target-22.C: New test.
This patch makes the SVE CLASTB GPR alternative more expensive than the
FPR alternative in order to avoid unnecessary cross-file moves. It also
fixes the prefix used to print the FPR; <vw> only handles 32-bit and
64-bit elements.
2019-08-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/aarch64-sve.md (fold_extract_last_<mode>):
Disparage the GPR alternative relative to the FPR one.
Fix handling of 8-bit and 16-bit FPR values.
gcc/testsuite/
* gcc.target/aarch64/sve/clastb_8.c: New test.
The reorg showed that we had an unnecessary separation between
the bitwise and max/min reductions for integers, and the
addition and max/min reductions for fp.
2019-08-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/iterators.md (BITWISEV): Delete.
(SVE_INT_REDUCTION, SVE_FP_REDUCTION): New int iterators.
(optab): Handle UNSPEC_UMAXV, UNSPEC_UMINV, UNSPEC_SMAXV,
UNSPEC_SMINV, UNSPEC_FADDV, UNSPEC_FMAXNMV, UNSPEC_FMAXV,
UNSPEC_FMINNMV, UNSPEC_FMINV.
(bit_reduc_op): Delete.
(sve_int_op): New int attribute.
(sve_fp_op): Handle UNSPEC_FADDV, UNSPEC_FMAXNMV, UNSPEC_FMAXV,
UNSPEC_FMINNMV, UNSPEC_FMINV.
* config/aarch64/aarch64-sve.md
(reduc_<MAXMINV:maxmin_uns>_scal_<SVE_I:mode>)
(*reduc_<MAXMINV:maxmin_uns>_scal_<SVE_I:mode>)
(reduc_<BITWISEV:optab>_scal_<SVE_I:mode>)
(*reduc_<BITWISEV:optab>_scal_<SVE_I:mode>): Merge into...
(reduc_<SVE_INT_REDUCTION:optab>_scal_<SVE_I:mode>)
(*reduc_<SVE_INT_REDUCTION:optab>_scal_<SVE_I:mode>): ...these
new patterns.
(reduc_plus_scal_<SVE_F:mode>, *reduc_plus_scal_<SVE_I:mode>)
(reduc_<FMAXMINV:optab>_scal_<SVE_F:mode>)
(*reduc_<FMAXMINV:optab>_scal_<SVE_F:mode>): Merge into...
(reduc_<SVE_FP_REDUCTION:optab>_scal_<SVE_F:mode>)
(*reduc_<SVE_FP_REDUCTION:optab>_scal_<SVE_F:mode>): ...these
new patterns.
This patch combines the four individual fused multiply-add optabs
into one pattern and uses unspecs instead of rtx codes. This is
part of a series of patches that change the SVE FP patterns so that
they can describe cases in which the predicate isn't all-true.
2019-08-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/aarch64-sve.md (fma<mode>4, *fma<mode>4)
(fnma<mode>4, *fnma<mode>4, fnms<mode>4, *fnms<mode>4)
(fms<mode>4, *fms<mode>4): Replace with...
(<SVE_COND_FP_TERNARY:optab><SVE_F:mode>4)
(*<SVE_COND_FP_TERNARY:optab><SVE_F:mode>4): ...these new patterns.
Use unspecs instead of rtx codes.
(cond_<optab><mode>, *cond_<optab><mode>_2, *cond_<optab><mode>_4)
(*cond_<optab><mode>_any): Add the predicate to SVE_COND_FP_TERNARY.
This patch makes us use the same define_insn for both the smax/smin
and fmax/fmin optabs. It also continues the process started by
the earlier FP unary patch of moving predicated FP patterns from
rtx codes to unspecs.
There's no need to handle the FMAX and FMIN instructions until
the ACLE patch, since we only use FMAXNM and FMINNM at present.
2019-08-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/iterators.md (SVE_COND_FP_MAXMIN_PUBLIC): New
int iterator.
(maxmin_uns_op): Handle UNSPEC_COND_FMAXNM and UNSPEC_COND_FMINNM.
* config/aarch64/aarch64-sve.md
(<FMAXMIN:su><FMAXMIN:maxmin><SVE_F:mode>3): Rename to...
(<SVE_COND_FP_MAXMIN_PUBLIC:optab><SVE_F:mode>3): ...this and
use a single unspec for the rhs.
(*<su><maxmin><mode>3): Delete.
(<maxmin_uns><SVE_F:mode>3): Use a single unspec for the rhs.
This patch merges the SVE FP rounding patterns with the other SVE
FP unary patterns.
At the moment, we only generate unary FP operations for full vectors,
so we can use (sqrt:VNx4SF ...) etc. in the rtl pattern. With the ACLE,
it's also possible to generate predicated operations on partial vectors
without specifying a value for inactive lanes. (sqrt:VNx4SF ...) would
then have different faulting behaviour from the instruction that the
pattern generates.
This patch therefore uses unspecs to represent the operations instead.
Later patches make this change for other patterns.
2019-08-07 Richard Sandiford <richard.sandiford@arm.com>
*pred_fold_left_plus_<mode> could no longer match anything, since
UNSPEC_FADDA now takes three operands. Predicated FADDAs should
now go through mask_fold_left_plus_<mode> instead.
2019-08-07 Richard Sandiford <richard.sandiford@arm.com>
[AArch64] Make SVE UNSPEC_COND_*s match the insn mnemonic
This patch makes the UNSPEC_COND* names match the instruction mnemonics,
rather than having the previous mixture in which some used instructions
while others used operator names.
2019-08-07 Richard Sandiford <richard.sandiford@arm.com>
aarch64-sve.md was getting a bit jumbled, with related operations
separated by unrelated operations. Also, many SVE instructions need
to have several patterns due to the various ways in which predication
can be used, and it was getting hard to tell which instructions had a
complete set of patterns.
This patch therefore tries to use an explicit grouping scheme.
Banner comments are usually frowned on, but md files have been
a traditional exception (e.g. when using Mike's template for
new targets).
The patch also lists the instructions implemented by each section,
so that it's possible to search the file by instruction mnemonic.
I wouldn't be surprised if I end up having to rip out the contents
section, but I found it useful for the month or so that that I've
been using it locally. The patch checks that the contents remain
up-to-date by running a checking script during an early stage of
the build.
No functional change intended.
2019-08-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/aarch64-sve.md: Reorganize contents and add
banner comments.
* config/aarch64/check-sve-md.awk: New file.
* config/aarch64/t-aarch64 (s-check-sve-md): New rule.
(insn-conditions.md): Depend on it.
PR driver/91130 Use CL_DRIVER when handling of COLLECT_GCC_OPTIONS in lto-wrapper.c
Some options are handled differently by the main driver (gcc, g++,
etc) from the back-end compiler programs (cc1, cc1plus, etc) in that
in the driver they do not take an additional argument, while in the
compiler programs they do. The processing option option CL_DRIVER
controls this alternative interpretation of the options.
The environment variable COLLECT_GCC_OPTIONS is the list of options to
add to a compile if the compiler re-invokes itself at some point. As
such, the options are driver options, so CL_DRIVER should be used when
processing this list. Currently lto-wrapper is doing this
incorrectly.
PR driver/91130
* lto-wrapper.c (find_and_merge_options): Use CL_DRIVER when
processing COLLECT_GCC_OPTIONS.
(run_gcc): Likewise.
Marek Polacek [Wed, 7 Aug 2019 14:20:40 +0000 (14:20 +0000)]
PR c++/91346 - Implement P1668R1, allow unevaluated asm in constexpr.
* constexpr.c (cxx_eval_constant_expression): Handle ASM_EXPR.
(potential_constant_expression_1) <case ASM_EXPR>: Allow.
* cp-tree.h (finish_asm_stmt): Adjust.
* parser.c (cp_parser_asm_definition): Grab the locaion of "asm" and
use it. Change an error to a pedwarn. Allow asm in C++2a, warn
otherwise.
* pt.c (tsubst_expr): Pass a location down to finish_asm_stmt.
* semantics.c (finish_asm_stmt): New location_t parameter. Use it.
* g++.dg/cpp2a/inline-asm1.C: New test.
* g++.dg/cpp2a/inline-asm2.C: New test.
* g++.dg/cpp1y/constexpr-neg1.C: Adjust dg-error.
Martin Liska [Wed, 7 Aug 2019 12:45:57 +0000 (14:45 +0200)]
Replace int with boolean in predicate functions.
2019-08-07 Martin Liska <mliska@suse.cz>
* fold-const.c (twoval_comparison_p): Replace int
with bool as a return type.
(simple_operand_p): Likewise.
(operand_equal_p): Replace int with bool as a return type.
* fold-const.h (operand_equal_p): Likewise.
* tree-core.h (enum omp_clause_code): Adjust OMP_CLAUSE_USE_DEVICE_PTR
OpenMP description. Add OMP_CLAUSE_USE_DEVICE_ADDR clause.
* tree.c (omp_clause_num_ops, omp_clause_code_name): Add entries
for OMP_CLAUSE_USE_DEVICE_ADDR clause.
(walk_tree_1): Handle OMP_CLAUSE_USE_DEVICE_ADDR.
* tree-pretty-print.c (dump_omp_clause): Likewise.
* tree-nested.c (convert_nonlocal_omp_clauses,
convert_local_omp_clauses): Likewise.
* gimplify.c (gimplify_scan_omp_clauses, gimplify_adjust_omp_clauses):
Likewise.
* omp-low.c (scan_sharing_clauses, lower_omp_target): Likewise.
Treat OMP_CLAUSE_USE_DEVICE_ADDR like OMP_CLAUSE_USE_DEVICE_PTR
clause with array or reference to array types, no matter what type
except for reference it has.
gcc/c-family/
* c-pragma.h (enum pragma_omp_clause): Add
PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR. Set PRAGMA_OACC_CLAUSE_USE_DEVICE
equal to PRAGMA_OMP_CLAUSE_USE_DEVICE_PTR instead of being a separate
enumeration value.
gcc/c/
* c-parser.c (c_parser_omp_clause_name): Parse use_device_addr clause.
(c_parser_omp_clause_use_device_addr): New function.
(c_parser_omp_all_clauses): Handle PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.
(OMP_TARGET_DATA_CLAUSE_MASK): Add PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.
(c_parser_omp_target_data): Handle PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR
like PRAGMA_OMP_CLAUSE_USE_DEVICE_PTR, adjust diagnostics about no
map or use_device_* clauses.
* c-typeck.c (c_finish_omp_clauses): For OMP_CLAUSE_USE_DEVICE_PTR
in OpenMP, require pointer type rather than pointer or array type.
Handle OMP_CLAUSE_USE_DEVICE_ADDR.
gcc/cp/
* parser.c (cp_parser_omp_clause_name): Parse use_device_addr clause.
(cp_parser_omp_all_clauses): Handle PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.
(OMP_TARGET_DATA_CLAUSE_MASK): Add PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.
(cp_parser_omp_target_data): Handle PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR
like PRAGMA_OMP_CLAUSE_USE_DEVICE_PTR, adjust diagnostics about no
map or use_device_* clauses.
* semantics.c (finish_omp_clauses): For OMP_CLAUSE_USE_DEVICE_PTR
in OpenMP, require pointer or reference to pointer type rather than
pointer or array or reference to pointer or array type. Handle
OMP_CLAUSE_USE_DEVICE_ADDR.
* pt.c (tsubst_omp_clauses): Handle OMP_CLAUSE_USE_DEVICE_ADDR.
gcc/testsuite/
* c-c++-common/gomp/target-data-1.c (foo): Use use_device_addr clause
instead of use_device_ptr clause where required by OpenMP 5.0, add
further tests for both use_device_ptr and use_device_addr clauses.
libgomp/
* testsuite/libgomp.c/target-18.c (struct S): New type.
(foo): Use use_device_addr clause instead of use_device_ptr clause
where required by OpenMP 5.0, add further tests for both use_device_ptr
and use_device_addr clauses.
* testsuite/libgomp.c++/target-9.C (struct S): New type.
(foo): Use use_device_addr clause instead of use_device_ptr clause
where required by OpenMP 5.0, add further tests for both use_device_ptr
and use_device_addr clauses. Add t and u arguments.
(main): Adjust caller.
Kewen Lin [Wed, 7 Aug 2019 07:11:14 +0000 (07:11 +0000)]
rs6000: support vector int type rotatert
gcc/ChangeLog
2019-08-07 Kewen Lin <linkw@gcc.gnu.org>
* config/rs6000/vector.md (vrotr<mode>3): New define_expand.
gcc/testsuite/ChangeLog
2019-08-07 Kewen Lin <linkw@gcc.gnu.org>
* gcc.target/powerpc/vec_rotate-1.c: New test.
* gcc.target/powerpc/vec_rotate-2.c: New test.
* gcc.target/powerpc/vec_rotate-3.c: New test.
* gcc.target/powerpc/vec_rotate-4.c: New test.
Alexandre Oliva [Wed, 7 Aug 2019 06:35:39 +0000 (06:35 +0000)]
wrap math.h for M_PI et al in target/i386 tests
Most but not all of the tests that expect M_PI, M_PI_2 and/or M_PI_4
to be defined in math.h explicitly exclude one target system that does
not satisfy this non-standard assumption.
This patch introduces a wrapper header that includes math.h and then
conditionally supplies the missing non-standard macro definitions.
With that, we can drop the dg-skip-if "no M_PI" exclusions.
Jonathan Wakely [Tue, 6 Aug 2019 15:57:55 +0000 (16:57 +0100)]
P1651R0 bind_front should not unwrap reference_wrapper
P1651R0 bind_front should not unwrap reference_wrapper
* include/std/functional (bind_front): Don't unwrap reference_wrapper.
* include/std/version (__cpp_lib_bind_front): Update value.
* testsuite/20_util/function_objects/bind_front/1.cc: Fix test for
feature test macro.
* testsuite/20_util/function_objects/bind_front/2.cc: New test.
Jason Merrill [Tue, 6 Aug 2019 14:07:59 +0000 (10:07 -0400)]
PR c++/91378 - ICE with noexcept and auto return type.
Here, since the call to g is not type-dependent, we call mark_used on it to
determine its return type. This also wants to instantiate the
noexcept-expression. But since nothing in maybe_instantiate_noexcept was
calling push_to_top_level, we substituted b.i with processing_template_decl
set, so we left it unresolved for later access checking. As a result, the
type of C::g<int> remained instantiation-dependent, leading to an ICE in
type_dependent_expression_p on the assert that the type of a function
template with no dependent template arguments must be non-dependent.
Jakub Jelinek [Tue, 6 Aug 2019 07:26:32 +0000 (09:26 +0200)]
tree.h (OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV): Rename to ...
* tree.h (OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV): Rename to ...
(OMP_CLAUSE_LASTPRIVATE_LOOP_IV): ... this. Adjust comment.
* gimplify.c (gimple_add_tmp_var): In SIMD contexts, turn addressable
new vars into GOVD_PRIVATE rather than GOVD_LOCAL.
(gimplify_omp_for): Don't do C++ random access iterator clause
adjustments on combined constructs from OMP_LOOP. For OMP_LOOP,
don't predetermine the artificial iterator in case of C++ random
access iterators as lastprivate, but private. For OMP_LOOP, force
bind expr around simd body and force for_pre_body before the
construct. Use OMP_CLAUSE_LASTPRIVATE_LOOP_IV instead of
OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV.
(gimplify_omp_loop): Add firstprivate clauses on OMP_PARALLEL for
diff var of C++ random access iterators. Handle
OMP_CLAUSE_FIRSTPRIVATE. For OMP_CLAUSE_LASTPRIVATE_LOOP_IV, if
not outermost also add OMP_CLAUSE_FIRSTPRIVATE, and in both cases
clear OMP_CLAUSE_LASTPRIVATE_LOOP_IV on the lastprivate clause
on the OMP_FOR and OMP_DISTRIBUTE constructs if any.
* omp-low.c (lower_rec_input_clauses): For
OMP_CLAUSE_LASTPRIVATE_LOOP_IV on simd copy construct the private
variables instead of default constructing them.
(lower_lastprivate_clauses): Use OMP_CLAUSE_LASTPRIVATE_LOOP_IV
instead of OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV and move the
is_taskloop_ctx check from the assert to the guarding condition.
gcc/cp/
* parser.c (cp_parser_omp_for_loop): For OMP_LOOP, ignore parallel
clauses and predetermine iterator as lastprivate.
* semantics.c (handle_omp_for_class_iterator): Use
OMP_CLAUSE_LASTPRIVATE_LOOP_IV instead of
OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV, set it for lastprivate also
on OMP_LOOP construct. If a clause is missing for class iterator
on OMP_LOOP, add firstprivate clause, and if there is private
clause, turn it into firstprivate too.
(finish_omp_for): Formatting fix. For OMP_LOOP, adjust
OMP_CLAUSE_LASTPRIVATE_LOOP_IV clause CP_CLAUSE_INFO, so that it
uses copy ctor instead of default ctor.
* cp-gimplify.c (cp_gimplify_expr): Handle OMP_LOOP like
OMP_DISTRIBUTE etc.
(cp_fold_r): Likewise.
(cp_genericize_r): Likewise.
(cxx_omp_finish_clause): Also finish lastprivate clause with
OMP_CLAUSE_LASTPRIVATE_LOOP_IV flag.
* pt.c (tsubst_omp_clauses): Handle OMP_CLAUSE_BIND.
(tsubst_omp_for_iterator): For OMP_LOOP, ignore parallel
clauses and predetermine iterator as lastprivate.
* constexpr.c (potential_constant_expression_1): Handle OMP_LOOP
like OMP_DISTRIBUTE etc.
libgomp/
* testsuite/libgomp.c++/loop-13.C: New test.
* testsuite/libgomp.c++/loop-14.C: New test.
* testsuite/libgomp.c++/loop-15.C: New test.
projects / thirdparty / gcc.git / log
