This patch adds a missing case to the output of cycle diagnostics here a
transition from an Elaborate_Body pair may reach a destination which is
in the context of an active Elaborate_All.
2019-07-08 Hristian Kirtchev <kirtchev@adacore.com>
gcc/ada/
* bindo-diagnostics.adb (Diagnose_Cycle): Capture the presence
of an Elaborate_All edge before iterating over the edges of the
cycle.
(Output_Elaborate_Body_Transition): Update the parameter profile
and the comment on usage. Add a missing case where the edge is
within the context of an Elaborate_All.
(Output_Transition): Update the call to
Output_Elaborate_Body_Transition.
* bindo-graphs.ads, bindo-graphs.adb
(Contains_Elaborate_All_Edge): New routine.
From-SVN: r273217
GNATprove represents reads and writes via pointers as operations on a
fake __HEAP entity. This entity already had various properties set to
dummy values (e.g. Scope set to Standard_Standard), so that it can be
processed like other entities without crashing and not special-cased
everywhere. Now it also has a dummy Etype, so it can be processed with
Is_Single_Concurrent_Object.
The modified code is only executed by GNATprove; frontend is not
affected.
2019-07-08 Piotr Trojanek <trojanek@adacore.com>
gcc/ada/
* lib-xref-spark_specific.adb (Create_Heap): Set dummy Etype for
the fake __HEAP entity.
From-SVN: r273216
CommandLineToArgvW drags a dependency on SHELL32.DLL and thus GDI32.DLL.
By loading GDI32.DLL some default GDI objects are allocated. On some
Windows versions this cause the use of a lock on the graphical interface
during process termination. This can impact parallelism significantly as
termination of processes is serialized.
2019-07-08 Nicolas Roche <roche@adacore.com>
gcc/ada/
* rtinit.c (__gnat_runtime_initialize): Remove dependency on
CommandLineToArgvW.
From-SVN: r273214
Real literals of fixed-point type are expected to keep their precise
fixed-point type in GNATprove. This is now correctly enforced.
There is no impact on compilation.
2019-07-08 Yannick Moy <moy@adacore.com>
gcc/ada/
* expander.adb (Expand): Do not reset Analyzed flag always.
* sem_eval.adb (Fold_Ureal): Mark node as analyzed.
From-SVN: r273211
This patch fixes a compiler crash in the compiler on a timed entry call
whose delay expression is a type conversion, when FLoat_Overflow checks
are enabled.
2019-07-08 Ed Schonberg <schonberg@adacore.com>
gcc/ada/
* exp_ch9.adb (Expand_N_Timed_Entry_Call): Do not insert twice
the assignment statement that computes the delay value, to
prevent improper tree sharing when the value is a type
conversion and Float_Overflow checks are enabled.
gcc/testsuite/
* gnat.dg/entry1.adb, gnat.dg/entry1.ads: New testcase.
From-SVN: r273210
This patch introduces several changes to the new elaboration order
mechanism:
* The concept of "strong" and "weak" edges is introduced. Strong
edges are the byproduct of language-defined relations between
units, such as with clauses. Weak edges are the byproduct of
specilative invocations at elaboration time, which may or may not
take place depending on control flow.
* The elaboration order algorithm has been heavily modified to make
use of the strong and weak edges, and operate on units compiled
with different elaboration models.
* The elaboration order algorithm employs the following logic:
- Maintain two sets of vertices, one for all elaborable
vertices, and one for all waiting vertices.
- Pick the best elaborable vertex, and elaborate its component.
- If no such elaborable vertex is available, pick the best
weakly elaborable vertex whose unit has been compiled with the
dynamic model, and elaborate its component.
- If no such weakly elaborable vertex is available, then either
all vertices were already elaborated, or the graph contains a
cycle.
The elaboration of a component employs the same logic, with an
added step where all successors of some predecessor currently being
elaborated are notified that they have one fewer predecessor to
wait on. This may cause certain successors to become elaborable, in
which case they are moved from the set of waiting vertices to the
set of elaborable vertices.
* Three new GNATbind debug switches are introduced, -d_a, -d_b, and
-d_e, to eliminate the effects of pragmas Elaborate_All,
Elaborate_Body, and Elaborate respectively.
* The section on terminology is updated to include new entries.
2019-07-08 Hristian Kirtchev <kirtchev@adacore.com>
gcc/ada/
* bindo.adb: Update the section on terminology to include new
concepts. Update the section on switches to include new
entries.
* bindo.ads: Add type Precedence_Kind.
* bindo-builders.adb: Add with and use clauses for Debug and
Bindo.Validators. Add use clauses for
Bindo.Validators.Invocation_Graph_Validators and
Bindo.Validators.Library_Graph_Validators.
(Build_Invocation_Graph): Validate the graph immediately after
it was built.
(Build_Library_Graph): Update the parameter profile. The
creation of the graph is now elaboration model-agnostic.
Validate the graph immediately after it was built.
(Create_With_Edge): Create regular with edges for Elaborate and
Elaborate_All edges when the appropriate debug switches are in
effect.
* bindo-builders.ads (Build_Library_Graph): Update the parameter
profile.
* bindo-diagnostics.adb (Diagnose_Cycle): Track the presence of
an Elaborate_All edge throughout the inspection of the cycle's
edges.
(Output_Dynamic_Model_Suggestions): Output the suggestion only
when the cycle contains at least one weak edge where the
successor was statically elaborated.
(Output_Elaborate_Body_Transition, Output_Forced_Transition,
Output_With_Transition): Update the assertions.
* bindo-elaborators.adb: Remove use clauses for
Bindo.Validators.Invocation_Graph_Validators and
Bindo.Validators.Library_Graph_Validators. Remove strings
Add_To_All_Candidates_Msg and Add_To_Comp_Candidates_Msg.
Remove type String_Ptr.
(Add_Vertex, Add_Vertex_If_Elaborable, Create_All_Candidates_Set
Create_Component_Candidates_Set): Remove.
(Create_Component_Vertex_Sets, Create_Vertex_Sets): New routine.
(Elaborate_Component): Update the parameter profile and the
comment on usage. Reimplement the elaboration of a component.
The algorithm will now attempt to elaborate as many vertices
possible. If this is not possible, and a weakly elaborable
vertex is available use unit was compiled using the dynamic
model, the algorithm will elaborate it.
(Elaborate_Library_Graph): Reimplement the elaboration of the
graph. The algorithm will now attempt to elaborate as many
vertices along with their components as possible. If this is not
possible, and a weakly elaborable vertex is available use unit
was compiled using the dynamic model, the algorithm will
elaborate it along with its component.
(Elaborate_Units): Merge with the functionality of
Elaborate_Units_Common.
(Elaborate_Units_Common, Elaborate_Units_Dynamic,
Elaborate_Units_Static): Remove.
(Elaborate_Vertex): Update the parameter profile and the comment
on usage. Reimplemented.
(Find_Best_Candidate): Remove.
(Find_Best_Elaborable_Vertex, Find_Best_Vertex,
Find_Best_Weakly_Elaborable_Vertex, Has_Elaborable_Body,
Insert_Elaborable_Successor, Insert_Vertex): New routines.
(Is_Better_Candidate): Remove.
(Is_Better_Elaborable_Vertex,
Is_Better_Weakly_Elaborable_Vertex,
Is_Suitable_Elaborable_Vertex,
Is_Suitable_Weakly_Elaborable_Vertex): New routines.
(Trace_Candidate_Vertices): Remove.
(Trace_Component): Output the number of strong and weak
predecessors.
(Trace_Unelaborated_Vertices): Remove.
(Trace_Vertex): Output the number of strong and weak
predecessors.
(Trace_Vertices): New routine.
(Update_Successor, Update_Successors): Update the parameter
profile and the comment on usage.
* bindo-graphs.adb: Remove type Precedence_Kind.
(Add_Edge_With_Return): Update the increment of pending
predecessors.
(Add_Vertex): Provide default values for strong and weak
predecessors.
(Complementary_Vertex): Move the initial declaration to the
spec. Update the parameter profile and the comment on usage.
(Contains_Weak_Static_Successor): New routine.
(Create): Update the parameter profile. The creation of the
graph is now elaboration model-agnostic.
(Decrement_Pending_Predecessors): Update the parameter profile
and the comment on usage. Reimplemented.
(Delete_Edge): Update the decrement of pending predecesors.
(Has_Elaborate_Body): Do not treat a vertex as being subject to
Elaborate_Body when a debug switch is in effect.
(Increment_Pending_Predecessors): Update the parameter profile
and the comment on usage. Reimplemented.
(Is_Elaborable_Component): Reimplemented.
(Is_Elaborable_Vertex): Move the initial declaration to the
spec. Reimplemented.
(Is_Elaborate_Body_Pair): New routine.
(Is_Dynamically_Elaborated): Update the parameter profile.
Reimplemented.
(Is_Weakly_Elaborable_Vertex): New routine.
(Pending_Predecessors): Removed.
(Pending_Predecessors_For_Elaboration,
Pending_Strong_Predecessors, Pending_Weak_Predecessors,
Update_Pending_Predecessors): New routines.
(Update_Pending_Predecessors_Of_Components): Update the
increment of pending predecessors.
* bindo-graphs.ads: Update the components of type
Component_Attributes. Update the components of type
Library_Graph_Attributes. Update the components of type
Library_Graph_Vertex_Attributes. Update the initialization of
No_Component_Attributes. Update the initialization of
No_Library_Graph_Vertex_Attributes.
(Complementary_Vertex, Contains_Weak_Static_Successor): New
routines.
(Create): Update the parameter profile and the comment on usage.
(Decrement_Pending_Predecessors, Is_Dynamically_Elaborated):
Update the parameter profile and the comment on usage.
(Is_Elaborate_Body_Pair, Is_Weakly_Elaborable_Vertex): New
routines.
(Pending_Predecessors): Removed.
(Pending_Predecessors_For_Elaboration,
Pending_Strong_Predecessors, Pending_Weak_Predecessors): New
routines.
* bindo-writers.adb (Write_Components): Moved from the spec.
(Write_Component): Output the strong and weak predecessors.
(Write_Library_Graph): Output the components as part of the
graph.
(Write_Library_Graph_Vertex): Output the strong and weak
predecessors.
* bindo-writers.ads (Write_Components): Moved to the body.
* debug.adb: Add and document new GNATbind switches -d_a, -d_b,
-d_e.
* bindo-validators.adb: Minor reformattings.
From-SVN: r273209
This patch adds additional data rates to the GNAT.Serial_Communications
package (Linux version).
2019-07-08 Bob Duff <duff@adacore.com>
gcc/ada/
* libgnat/g-sercom.ads, libgnat/g-sercom__linux.adb (Data_Rate):
Support additional data rates.
From-SVN: r273208
Unlike, say, clones created internally by the compiler, elab procs
materialize specific user code and flagging them artificial now takes
elab code away from gcov's analysis, a regression compared to previous
releases.
On the testcase below:
package Gcov_Q is
function F (X : Integer) return Integer is (X + 1);
end;
with Gcov_Q;
package Gcov_P is
Y : Integer := Gcov_Q.F (X => 1);
end;
with Gcov_P;
procedure Gcov_Test is
begin
if Gcov_P.Y /= 2 then
raise Program_Error;
end if;
end;
After compiling with:
gnatmake -f -g gcov_test.adb \
-cargs -ftest-coverage -fprofile-arcs \
-largs -fprofile-generate
and executing with
./gcov_test
We expect
gcov gcov_p
to produce a gcov_p.ads.gcov report.
2019-07-08 Olivier Hainque <hainque@adacore.com>
gcc/ada/
* gcc-interface/trans.c (Compilation_Unit_to_gnu): Don't request
DECL_ARTIFICIAL_P on elab proc declarations.
From-SVN: r273207
This fixes a crash (or an assertion failure) during the processing done
for -gnatRj on the declaration of an extension of a private type.
Generally speaking, extension declarations are delicate in this context
because the front-end does not duplicate the structure of the parent
type, so the processing required to output the structural layout needs
to go up to the declaration of the parent type, which may or may not be
available or usable.
The change also makes the processing more robust by falling back to the
flat layout if the declaration of the parent type cannot be processed.
2019-07-08 Eric Botcazou <ebotcazou@adacore.com>
gcc/ada/
* repinfo.adb (List_Record_Info): Declare Incomplete_Layout and
Not_In_Extended_Main local exceptions.
(List_Structural_Record_Layout): For an extension, raise the
former if the parent subtype has not been built and the latter
if it is not declared in the main source unit. Fall back to the
flat layout if either exception has been raised.
From-SVN: r273206
This patch corrects a bug in the implementation of Delete in an unusual
boundary case: the RM describes the semantics of Delete as equivalent to
that of Replace_String with a null argument. As a result, deleting a
null string that starts past the end of its argument is a noop and must
not raise Index_Error.
2019-07-08 Ed Schonberg <schonberg@adacore.com>
gcc/ada/
* libgnat/a-strfix.adb (Delete): The RM describes the semantics
of Delete as equivalent to that of Replace_String with a null
argument. As a result, deleting a null string that starts past
the end of its argument is a noop and must not raise
Index_Error.
gcc/testsuite/
* gnat.dg/fixed_delete.adb: New testcase.
From-SVN: r273205
The frontend crashes processing the derivation of a tagged type whose
ultimate ancestor is an interface type I1 that has a null primitive,
implements another interface I2 derived from I2, and does not override
the null primitive.
2019-07-08 Javier Miranda <miranda@adacore.com>
gcc/ada/
* exp_disp.adb (Register_Primitive): When registering a
primitive in the secondary dispatch table, handle primitive
inherited through several levels of type derivation (required to
properly handle inherited 'null' primitive).
gcc/testsuite/
* gnat.dg/interface9.adb, gnat.dg/interface9_root-child.ads,
gnat.dg/interface9_root.ads: New testcase.
From-SVN: r273204
This patch performs a code reorganization of the implementation of
pragma Compile_Time_Error. No functional change.
No test required.
2019-07-08 Javier Miranda <miranda@adacore.com>
gcc/ada/
* gnat1drv.adb (Post_Compilation_Validation_Checks:
Validate_Compile_Time_Warning_Errors is now located in sem_prag
(instead of sem_ch13).
* sem_ch13.ads (Validate_Compile_Time_Warning_Error,
Validate_Compile_Time_Warning_Errors): Move to sem_prag.
* sem_ch13.adb
(Compile_Time_Warnings_Errors): Move to sem_prag.
(Initialize): Remove initialization of table
Compile_Time_Warning_Errors.
(Validate_Compile_Time_Warning_Error,
Validate_Compile_Time_Warning_Errors): Move to sem_prag.
* sem_prag.ads (Validate_Compile_Time_Warning_Errors): New
procedure.
* sem_prag.adb (Initialize): Initialize table
Compile_Time_Warning_Errors.
From-SVN: r273202
This patch fixes a crash on compiling the postcondtion for a generic
subprogram, when the postcondition is a call with both positional and
named parameter associations.
2019-07-08 Ed Schonberg <schonberg@adacore.com>
gcc/ada/
* sem_ch13.adb (Analyze_Aspect_Specifications): For a
pre/postcondition of a generic subprogram declaration, do not
use Relocate_Node on the aspect expression to construct the
corresponding attribute specification, to prevent tree anomalies
when the expression is a call with named actual parameters.
gcc/testsuite/
* gnat.dg/predicate9.adb: New testcase.
From-SVN: r273201
The compiler may trigger spurious errors on pragmas Compile_Time_Error
and Compile_Time_Warning when their boolean expression computes the size
of a type. After this patch the following test compiles fine.
with Interfaces; use Interfaces;
package Types is
type Arr is array (1 .. 6) of Unsigned_8
with Size => 48,
Alignment => 1;
type Rec is record
Comp_1 : Unsigned_32;
Comp_2 : Unsigned_16;
end record
with Size => 48,
Alignment => 1;
end Types;
with Types; use Types;
package Main is
pragma Compile_Time_Error
(Arr'Size = 12, "ERROR: Arr'Size is 48, not 12");
pragma Compile_Time_Error
(Arr'Size = 48, "OK: Arr");
pragma Compile_Time_Error
(Arr'Size /= 48, "ERROR: Arr'Size is 48");
pragma Compile_Time_Error
(Rec'Size = 34, "ERROR: Rec'Size is 48, not 34");
pragma Compile_Time_Error
(Rec'Size = 48, "OK: Rec");
pragma Compile_Time_Error
(Rec'Size /= 48, "ERROR: Rec'Size is 48");
end Main;
Command: gcc -c main.ads
Output: main.ads:7:07: OK: Arr
main.ads:14:07: OK: Rec
2019-07-08 Javier Miranda <miranda@adacore.com>
gcc/ada/
* sem_attr.adb (Analyze_Attribute [Attribute_Size]): For pragmas
used to report user defined compile time warning or errors
handle 'Size for types with known static RM size.
From-SVN: r273200
This patch fixes an issue whereby the creation of an enumeration within
package where Default_Scalar_Storage_Order is in effect may lead to a
crash when the attributes Image or Value are applied to objects of said
type or the type directly.
2019-07-08 Justin Squirek <squirek@adacore.com>
gcc/ada/
* exp_imgv.adb (Build_Enumeration_Image_Tables): Default SSO for
the building of image tables.
(Expand_Image_Attribute): Minor cleanup.
gcc/testsuite/
* gnat.dg/sso16.adb: New testcase.
From-SVN: r273199
This patch fixes a spurious error when verifying that the visibility of
the expression of an aspect has not changed between the freeze point of
the entity to which it applies, and the end of the enclosing declarative
part. If the entity is a composite type its components must be made
directly visible for the analysis of the expression. In a generic
context this must be done explicitly at the end of the declarative part.
2019-07-08 Ed Schonberg <schonberg@adacore.com>
gcc/ada/
* sem_ch13.adb (Check_Aspect_At_End_Of_Declarations): For an
unanalized aspect in a generic context that has not been
analyzed yet, if the aspect applies to a type, place the type on
the scope stack to make its components visible, before checking
conformance with the version of the expression analyzed at the
freeze point.
gcc/testsuite/
* gnat.dg/predicate8.adb, gnat.dg/predicate8_pkg.adb,
gnat.dg/predicate8_pkg.ads: New testcase.
From-SVN: r273197
Although BOOT_CFLAGS can be used to bootstrap with -Og, having a
dedicated build config is sometimes more convenient.
2019-07-08 Richard Sandiford <richard.sandiford@arm.com>
config/
* bootstrap-Og.mk: New file.
gcc/
* doc/install.texi (bootstrap-Og): Document.
From-SVN: r273193
guality.exp is silently skipped on Linux systems with
kernel.yama.ptrace_scope=1 because gdb fails to attach to the
sanity check test. This patch uses PR_SET_PTRACER (where available)
to avoid this.
prctl was apparently added in Linux 2.1.57, so I don't think we
need any tests other than __linux for the #include.
2019-07-08 Richard Sandiford <richard.sandiford@arm.com>
gcc/testsuite/
* gcc.dg/guality/guality.h: Include <sys/prctl.h> on Linux targets.
(main): Use PR_SET_PTRACER where available.
From-SVN: r273192
This patch is part of a series that fixes ambiguous attribute
uses in .md files, i.e. cases in which attributes didn't use
<ITER:ATTR> to specify an iterator, and in which <ATTR> could
have different values depending on the iterator chosen.
No behavioural change -- produces the same code as before.
2019-07-08 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/riscv/pic.md (*local_pic_load_s<mode>)
(*local_pic_load_u<mode>): Explicitly specify the mode iterator
referenced by <mode>, giving...
(*local_pic_load_s<SUBX:mode>, *local_pic_load_u<SUBX:mode>): ...these.
* config/riscv/riscv.md (*sge<u>_<X:mode><GPR:mode>)
(*slt<u>_<X:mode><GPR:mode>, *sle<u>_<X:mode><GPR:mode>): Explicitly
use <X:MODE> for the mode attribute.
From-SVN: r273191
Final check on PCI options; for Darwin these are not dependent on the PIE
ones, although PIE does require PIC to support it. Specifically, for Darwin,
"fPIC fno-PIE" should result in the same as "-fno-PIE -fPIC".
2019-07-07 Iain Sandoe <iain@sandoe.co.uk>
* config/darwin.c (darwin_override_options): Make a final check on PIC
options.
From-SVN: r273181
For PPC Darwin, we need the JBSR long jump code to be enabled when generating
kernel code. Now we have that handled in rs6000.c, we can drop the conflated
setting in the common code. Symbol stubs are not generated for any X86 case.
2019-07-07 Iain Sandoe <iain@sandoe.co.uk>
* config/darwin.c (darwin_override_options): Don't jam symbol stubs
on for kernel code.
From-SVN: r273180
2019-07-07 Paul Thomas <pault@gcc.gnu.org>
PR fortran/91077
* trans-array.c (gfc_conv_scalarized_array_ref) Delete code
that gave symbol backend decl for subref arrays and deferred
length variables.
2019-07-07 Paul Thomas <pault@gcc.gnu.org>
PR fortran/91077
* gfortran.dg/pointer_array_11.f90 : New test.
From-SVN: r273176
PR91068 is a case in which we have (ignoring non-LRA alternatives):
[(set (match_operand:SI 0 "register_operand" "=l,d?")
(plus:SI (mult:SI (match_operand:SI 1 "register_operand" "d,d")
(match_operand:SI 2 "register_operand" "d,d"))
(match_operand:SI 3 "register_operand" "0,d")))
(clobber (match_scratch:SI 4 "=X,l"))
(clobber (match_scratch:SI 5 "=X,&d"))]
where the first alternative is one instruction but the second is two.
This is very similar to the case that my recent IRA patches were
supposed to help. The crucial difference is that the cheap
alternative requires a single-register class while the expensive
alternative uses general registers.
This makes a difference when one of operand 0 or 3 can naturally be
allocated to LO but the other can't. If IRA makes that allocation,
both alternatives require one reload of equal cost and so the first
alternative clearly wins.
However, if we say that tying operands 0 and 3 saves the cost of a full
move, then all other things being equal, IRA will prefer to allocate
both registers to the same GPR. The registers will then naturally
fit the second alternative.
This has a more drastic effect in the MIPS case than it should because
using the GPR alternative is much more expensive there than it appears
to the RA. But that's really a separate problem and something we were
able to live with before my IRA patch.
What makes tying less useful here is the fact that the tied register is
a single-register class. I think in those circumstances it's better not
to use tied operands at all and instead use "l" for the inputs.
Allocating the input to LO, and allocating the output to LO, then depend
naturally on class costs. If we decide to allocate at least one of them
to LO, we'll use the cheap alternative, otherwise we'll (correctly) use
the expensive alternative. This effectively restores the situation
before my IRA patch, but this time making the preference on the input
register more explicit.
I originally wrote the patterns in the early days of IRA, and certainly
well before LRA. I think they were largely influened by reload rather
than RA proper (see the comment above *mul_acc_si, which is all about
the reload behaviour). LRA copes with the two-"l" case just fine.
The patch may well cause problems for -mno-lra, but I think we should
cull that option anyway.
2019-07-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
PR target/91068
* config/mips/mips.md (*mul_acc_si, *mul_acc_si_r3900, *macc)
(*msac, *msac_using_macc, *mul_sub_si): Use "l" for input operands
instead of matching them to "l" output operands.
From-SVN: r273175
While testing the fix for PR91068, I hit an rtl checking failure
while building newlib. mips_split_move was decomposing an address that
happened to be symbolic and then tried to access the REGNO of the base
register field, which wasn't initialised but which by chance pointed to
valid memory.
2019-07-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/mips/mips.c (mips_split_move): Zero-initialize addr
and check whether addr.reg is nonnull before using it.
From-SVN: r273174
* omp-low.c (lower_rec_input_clauses): For lastprivate clauses in
ctx->for_simd_scan_phase simd copy the outer var to the privatized
variable(s). For conditional lastprivate look through outer
GIMPLE_OMP_SCAN context.
(lower_omp_1): For conditional lastprivate look through outer
GIMPLE_OMP_SCAN context.
* testsuite/libgomp.c/scan-19.c: New test.
* testsuite/libgomp.c/scan-20.c: New test.
From-SVN: r273169
* omp-low.c (struct omp_context): Rename combined_into_simd_safelen0
member to combined_into_simd_safelen1.
(lower_rec_input_clauses, lower_omp_1): Adjust uses.
(lower_lastprivate_clauses): Likewise. For conditional lastprivate
clauses if ctx->combined_into_simd_safelen1 put statements after the
predicate conditionalized block rather than into it.
From-SVN: r273168
This patch is part of a series that fixes ambiguous attribute
uses in .md files, i.e. cases in which attributes didn't use
<ITER:ATTR> to specify an iterator, and in which <ATTR> could
have different values depending on the iterator chosen.
The vx-builtins.md part changes the choice of <mode> from the
implicit <VFCMP:mode> to an explicit <VF_HW:mode> (i.e. from the
mode of the comparison result to the mode of the operands being
compared). That seemed like the intended behaviour given later
patterns like vec_cmpeq<mode>_cc.
The use of BFP in the s390.md LNDFR pattern looks like a typo,
since the operand to (abs ...) has to have the same mode as the result.
The only effect before this series was to create some extra variants
that would never match, making it harmless apart from very minor code
bloat.
2019-07-06 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/s390/s390.md (*negabs<FP:mode>2_nocc): Use FP for
operand 1.
* config/s390/vx-builtins.md (*vec_cmp<insn_cmp><mode>_cconly):
Make the choice of <mode> explicit, giving...
(*vec_cmp<insn_cmp><VF_HW:mode>_cconly): ...this.
From-SVN: r273162
This patch is part of a series that fixes ambiguous attribute
uses in .md files, i.e. cases in which attributes didn't use
<ITER:ATTR> to specify an iterator, and in which <ATTR> could
have different values depending on the iterator chosen.
No behavioural change except for dropping the unused *andnot<mode>3_bcst
permutations.
2019-07-06 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/i386/i386.md (*fop_<X87MODEF:mode>_3_i387)
(l<rounding_insn><MODEF:mode><SWI48:mode>2): Fix ambiguous uses
of .md attributes.
* config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask)
(*avx512pf_gatherpf<mode>df_mask, *avx512pf_scatterpf<mode>sf_mask)
(*avx512pf_scatterpf<mode>df_mask, *avx2_gathersi<mode>)
(*avx2_gathersi<mode>_2, *avx2_gatherdi<mode>)
(*avx2_gatherdi<mode>_2, *avx2_gatherdi<mode>_3): Likewise.
(*avx2_gatherdi<mode>_4, *avx512f_gathersi<mode>): Likewise.
(*avx512f_gathersi<mode>_2, *avx512f_gatherdi<mode>): Likewise.
(*avx512f_gatherdi<mode>_2, *avx512f_scattersi<mode>): Likewise.
(*avx512f_scatterdi<mode>): Likewise.
(*andnot<mode>3_bcst): Fix VI/VI48_AVX512VL typo.
From-SVN: r273161
This patch is part of a series that fixes ambiguous attribute
uses in .md files, i.e. cases in which attributes didn't use
<ITER:ATTR> to specify an iterator, and in which <ATTR> could
have different values depending on the iterator chosen.
No behavioural change -- produces the same code as before.
2019-07-06 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/h8300/h8300.md (*push1_h8300hs_<mode>): Explicitly
specify the mode iterator referenced by <mode>, giving...
(*push1_h8300hs_<QHI:mode>): ...this.
From-SVN: r273160
This patch is part of a series that fixes ambiguous attribute
uses in .md files, i.e. cases in which attributes didn't use
<ITER:ATTR> to specify an iterator, and in which <ATTR> could
have different values depending on the iterator chosen.
I think this is a genuine bugfix for the case in which the 1REG_MODE
and 1REG_ALT are different, since previously we would use the 1REG_MODE
for both the comparison and the select, even though the operands being
compared are 1REG_ALT rather than 1REG_MODE.
2019-07-06 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/gcn/gcn-valu.md
(vcond<VEC_1REG_MODE:mode><VEC_1REG_ALT:mode>): Use
gen_vec_cmp<VEC_1REG_ALT:mode>di rather than (implicitly)
gen_vec_cmp<VEC_1REG_MODE:mode>di. Explicitly use
gen_vcond_mask_<VEC_1REG_MODE:mode>di.
(vcond<VEC_1REG_MODE:mode><VEC_1REG_ALT:mode>_exec): Likewise,
but using the _exec comparison patterns.
(vcondu<VEC_1REG_INT_MODE:mode><VEC_1REG_INT_ALT:mode>): Use
gen_vec_cmp<VEC_1REG_INT_ALT:mode>di rather than (implicitly)
gen_vec_cmp<VEC_1REG_INT_MODE:mode>di. Explicitly use
gen_vcond_mask_<VEC_1REG_INT_MODE:mode>di.
(vcondu<VEC_1REG_INT_MODE:mode><VEC_1REG_INT_ALT:mode>_exec): Likewise,
but using the _exec comparison patterns.
From-SVN: r273159
This patch is part of a series that fixes ambiguous attribute
uses in .md files, i.e. cases in which attributes didn't use
<ITER:ATTR> to specify an iterator, and in which <ATTR> could
have different values depending on the iterator chosen.
I think this is a genuine bugfix for Thumb-1, since previously the
LDREX width was taken from the SImode success result rather than the
memory mode:
-#define HAVE_atomic_compare_and_swapt1qi_1 ((TARGET_HAVE_LDREX && TARGET_HAVE_MEMORY_BARRIER) && (TARGET_THUMB1))
-#define HAVE_atomic_compare_and_swapt1hi_1 ((TARGET_HAVE_LDREX && TARGET_HAVE_MEMORY_BARRIER) && (TARGET_THUMB1))
-#define HAVE_atomic_compare_and_swapt1di_1 ((TARGET_HAVE_LDREX && TARGET_HAVE_MEMORY_BARRIER) && (TARGET_THUMB1))
+#define HAVE_atomic_compare_and_swapt1qi_1 ((TARGET_HAVE_LDREXBH && TARGET_HAVE_MEMORY_BARRIER) && (TARGET_THUMB1))
+#define HAVE_atomic_compare_and_swapt1hi_1 ((TARGET_HAVE_LDREXBH && TARGET_HAVE_MEMORY_BARRIER) && (TARGET_THUMB1))
+#define HAVE_atomic_compare_and_swapt1di_1 ((TARGET_HAVE_LDREXD && ARM_DOUBLEWORD_ALIGN \
+ && TARGET_HAVE_MEMORY_BARRIER) && (TARGET_THUMB1))
The same goes for the predicate and constraints in
@atomic_compare_and_swapt1di_1, which previously used the
SI values from the success result.
2019-07-06 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/arm/sync.md
(@atomic_compare_and_swap<CCSI:arch><NARROW:mode>_1): Use
<NARROW:sync_predtab> instead of (implicitly) <CCSI:sync_predtab>.
(@atomic_compare_and_swap<CCSI:arch><SIDI:mode>_1): Likewise
<SIDI:sync_predtab>. Use <SIDI:cas_cmp_operand> and
<SIDI:cas_cmp_str>.
From-SVN: r273158
* omp-low.c (struct omp_context): Add for_simd_scan_phase member.
(maybe_lookup_ctx): Add forward declaration.
(omp_find_scan): Likewise. Walk into body of simd if composited
with worksharing loop.
(scan_omp_simd_scan): New function.
(scan_omp_1_stmt): Call it.
(lower_rec_simd_input_clauses): Don't create rvar nor rvar2 if
ctx->for_simd_scan_phase.
(lower_rec_input_clauses): Do much less work for inscan reductions
in ctx->for_simd_scan_phase is_simd regions.
(lower_omp_scan): Set is_simd also on simd constructs composited
with worksharing loop, unless ctx->for_simd_scan_phase. Never emit
a sorry message. Don't change GIMPLE_OMP_SCAN stmts into nops and
emit their body after in simd constructs composited with worksharing
loop.
(lower_omp_for_scan): Handle worksharing loop composited with simd.
* c-c++-common/gomp/scan-4.c: Don't expect sorry message.
* testsuite/libgomp.c/scan-11.c: New test.
* testsuite/libgomp.c/scan-12.c: New test.
* testsuite/libgomp.c/scan-13.c: New test.
* testsuite/libgomp.c/scan-14.c: New test.
* testsuite/libgomp.c/scan-15.c: New test.
* testsuite/libgomp.c/scan-16.c: New test.
* testsuite/libgomp.c/scan-17.c: New test.
* testsuite/libgomp.c/scan-18.c: New test.
* testsuite/libgomp.c++/scan-9.C: New test.
* testsuite/libgomp.c++/scan-10.C: New test.
* testsuite/libgomp.c++/scan-11.C: New test.
* testsuite/libgomp.c++/scan-12.C: New test.
* testsuite/libgomp.c++/scan-13.C: New test.
* testsuite/libgomp.c++/scan-14.C: New test.
* testsuite/libgomp.c++/scan-15.C: New test.
* testsuite/libgomp.c++/scan-16.C: New test.
From-SVN: r273157
PR tree-optimization/91096
* gcc.dg/vect/vect-simd-10.c (FLT_MIN_VALUE): Define.
(bar, main): Use it instead of -__builtin_inff ().
* gcc.dg/vect/vect-simd-14.c (FLT_MIN_VALUE): Define.
(bar, main): Use it instead of -__builtin_inff ().
From-SVN: r273156
* omp-low.c (omp_find_scan): Make static.
(lower_omp_for_scan): Fix order of merge arguments in input phase of
the second loop, var2 represents the first partial sum and so needs
to go before rprivb[ivar].
From-SVN: r273155
TARGET_LINK_STACK is unused on Darwin, and only relevant to a processor on
which the port was never released.
2019-07-05 Iain Sandoe <iain@sandoe.co.uk>
* config/rs6000/rs6000-logue.c: Remove unused code.
From-SVN: r273148
