Ed Schonberg [Thu, 11 Jan 2018 08:51:47 +0000 (08:51 +0000)]
[Ada] Handling of pragma Predicate
This patch fixes an omission in the processing of pragma Predicate, which
should have the same semantics as the corresponding aspect, which is more
commonly used.
Executing
gnatmake -q -gnata predicate
predicate
must yield:
Even1 violated
Even2 violated
----
with Text_IO; use Text_IO;
procedure Predicate is
begin
begin
declare
subtype Even1 is Integer;
pragma Predicate (Even1, Even1 mod 2 = 0);
X1 : constant Even1 := 1; -- This should fail first
begin
null;
end;
exception
when Others => Put_Line ("Even1 violated");
end;
begin
declare
subtype Even2 is Integer with Predicate => Even2 mod 2 = 0;
X2 : constant Even2 := 1; -- This should fail later, if reached
begin
null;
end;
exception
when Others => Put_Line ("Even2 violated");
end;
end;
2018-01-11 Ed Schonberg <schonberg@adacore.com>
gcc/ada/
* sem_prag.adb (Analyze_Pragma, case Predicate): Indicate that the type
has a delayed aspect which must be processed at the point the type is
frozen. This mimics what is done when the predicate is provided by a
source aspect.
[Ada] Different runtime behavior of Predicate_Failure
This patch corrects the generation of predicate checks to handle the case where
Predicate_Failure appears as a pragma.
------------
-- Source --
------------
-- main.adb
with Ada.Assertions; use Ada.Assertions;
with Ada.Exceptions; use Ada.Exceptions;
with Ada.Text_IO; use Ada.Text_IO;
procedure Main is
subtype Even_Asp is Integer
with Predicate => Even_Asp mod 2 = 0,
Predicate_Failure => "Even_Asp failed";
subtype Even_Prag is Integer
with Predicate => Even_Prag mod 2 = 0;
pragma Predicate_Failure (Even_Prag, "Even_Prag failed");
begin
begin
declare
Val : constant Even_Asp := 1;
begin
Put_Line ("ERROR: Even_Asp: did not fail");
end;
exception
when AE : Assertion_Error => Put_Line (Exception_Message (AE));
when others => Put_Line ("ERROR: Even_Asp: raised unexpected error");
end;
begin
declare
Val : constant Even_Prag := 3;
begin
Put_Line ("ERROR: Even_Prag: did not fail");
end;
exception
when AE : Assertion_Error => Put_Line (Exception_Message (AE));
when others => Put_Line ("ERROR: Even_Prag: raised unexpected error");
end;
end Main;
----------------------------
-- Compilation and output --
----------------------------
Patrick Bernardi [Thu, 11 Jan 2018 08:51:23 +0000 (08:51 +0000)]
[Ada] Integer overflow in SS_Allocate
This patch imposes a new check and rewrites existing ones to ensure operations
involving SS_Ptr do not cause an Integer overflow. The Default_Sec_Stack_Size
function was removed in the process to simplify System.Parameter.
SS_Ptr was derived from the integer System.Parameters.Size_Type to ease the
creation of objects of type SS_Stack by the binder and imposes a maximum
secondary stack size of 2GB. In most cases, the user will not hit this limit as
they cannot specify task stack sizes of more than 2GB via the Storage_Size and
Secondary_Stack_Size pragmas. Additionally, most operating systems limit the
primary stack size to less than 2GB, with defaults under 10MB. Linux is the
rare exception where the user can unbound the primary stack.
Executing the following:
gnatmake -q overflow
./overflow
package String_Pack is
function Return_Big_String return String;
end String_Pack;
-- string_pack.adb:
with Ada.Strings.Fixed; use Ada.Strings.Fixed;
package body String_Pack is
function Return_Big_String return String is
begin
return Integer'Last * "P";
end Return_Big_String;
S : String := Return_Big_String;
end String_Pack;
2018-01-11 Patrick Bernardi <bernardi@adacore.com>
gcc/ada/
* libgnat/s-parame*.adb, libgnat/s-parame*.ads: Remove unneeded
Default_Sec_Stack_Size.
* libgnat/s-secsta.adb (SS_Allocate): Handle the fixed secondary stack
limit check so that the integer index does not overflow. Check the
dynamic stack allocation does not cause the secondary stack pointer to
overflow.
(SS_Info): Align colons.
(SS_Init): Cover the case when bootstraping with an old compiler that
does not set Default_SS_Size.
Ed Schonberg [Thu, 11 Jan 2018 08:51:18 +0000 (08:51 +0000)]
[Ada] Illegal protected calls in inherited pre/postconditions
AI12-0166 specifies that it is illegal for a pre/postcondition of a
protected operation to contain an internal call to a protected function.
This patch completes the implementation of this rule in the case the
condition is inherited from a classwide condition of an abstract operation
of an interface type.
Compiling inheritpo.adb must yield:
inheritpo.ads:9:04: instantiation error at line 6
inheritpo.ads:9:04: internal call to "F" cannot appear
in inherited precondition of protected operation "P"
inheritpo.ads:9:04: instantiation error at line 7
inheritpo.ads:9:04: internal call to "F" cannot appear
in inherited precondition of protected operation "P"
--
package InheritPO is
type T is limited interface;
function F (X : T) return Boolean is abstract;
procedure P (X : in out T) is abstract with
Pre'Class => X.F,
Post'Class => X.F;
protected type PT is new T with
overriding function F return Boolean;
overriding procedure P;
end PT;
end InheritPO;
----
package body InheritPO is
protected body PT is
function F return Boolean is begin return True; end;
procedure P is begin null; end;
end PT;
end InheritPO;
2018-01-11 Ed Schonberg <schonberg@adacore.com>
gcc/ada/
* sem_ch3.adb (Add_Internal_Interface_Entities): When checking the
legality of an inherited operation that may require overriding, ignore
primitive_wrappers that correspond to explicit operations that override
an interface primitive.
* exp_util.adb (Build_Class_Wide_Expression, Replace_Entity): If the
operation to which the class-wide expression applies is a protected op.
with a primitive_wrapper, verify that the updated inherited expression
does not contain an internal call to a protected function. This
completes the implementation of AI12-0166.
* ali.adb: Document the remaining letters available for ALI lines.
(Scan_ALI): A with clause is internal when it is encoded on a 'Z' line.
* ali.ads: Update type With_Record. Field
Implicit_With_From_Instantiation is no longer in use. Add field
Implicit_With.
* csinfo.adb (CSinfo): Remove the setup for attribute
Implicit_With_From_Instantiation.
* lib-writ.adb (Collect_Withs): Correct the logic which marks a unit as
either implicitly or explicitly withed.
(Is_Implicit_With_Clause): New routine.
(Write_ALI): Rename array Implicit_With to Has_Implicit_With to avoid
confusion with the with clause attribute by the same name.
(Write_With_Lines): Update the emission of 'W', 'Y', and 'Z' headers.
* rtsfind.adb (Maybe_Add_With): Code cleanup.
* sem_ch8.adb (Present_System_Aux): Code cleanup.
* sem_ch10.adb (Expand_With_Clause): Mark the with clause as generated
for a parent unit.
(Implicit_With_On_Parent): Mark the with clause as generated for a
parent unit.
* sem_ch12.adb (Inherit_Context): With clauses inherited by an
instantiation are no longer marked as Implicit_With_From_Instantiation
because they are already marked as implicit.
* sem_elab.adb (Ensure_Prior_Elaboration_Static): Remove the kludge
which marks implicit with clauses as related to an instantiation.
* sinfo.adb (Implicit_With_From_Instantiation): Removed.
(Parent_With): New routine.
(Set_Implicit_With_From_Instantiation): Removed.
(Set_Parent_With): New routine.
* sinfo.ads: Update the documentation of attribute Implicit_With.
Remove attribute Implicit_With_From_Instantiation along with
occurrences in nodes. Add attribute Parent_With along with occurrences
in nodes.
(Implicit_With_From_Instantiation): Removed along with pragma Inline.
(Parent_With): New routine along with pragma Inline.
(Set_Implicit_With_From_Instantiation): Removed along with pragma Inline.
(Set_Parent_With): New routine along with pragma Inline.
This patch modifies routine Find_Enclosing_Scope which obtains the scope of an
arbitrary node to return the unique defining entity of an enclosing body. This
automatically takes care of the following corner cases:
* The body is a subprogram body which does not complete a previous
declaration. In this case the proper scope is the entity of the
body.
* The body is an entry body. Due to a limitation in the AST, the
entry body does not store its correcponsing spec, but utilizes a
roundabout way of obtaining it. Regardless of the limitation, the
proper scope is the entity of the entry declaration.
The issue was discovered during the development of the GNATprove tool and
is not visible to end users. No simple test is available because this would
require a debug session.
Patrick Bernardi [Thu, 11 Jan 2018 08:50:43 +0000 (08:50 +0000)]
[Ada] Aspect/pragma Secondary_Stack_Size can evaluate non-literals as zero
This patch fixes the problem of aspect/pragma Secondary_Stack_Size expressions
with non-literals evaluating as zero in static secondary stacks allocations.
The aspect Secondary_Stack_Size is now converted to a pragma instead of an
attribute as the attribute does not have visibility on the discriminant.
Additionally, the discriminant of the corresponding record type is now
referenced if the pragma expression contains a discriminant.
No simple test available as the problem only impacts programs when
System.Parameters.Sec_Stack_Dynamic = False
2018-01-11 Patrick Bernardi <bernardi@adacore.com>
gcc/ada/
* exp_ch9.adb (Expand_N_Task_Type_Declaration): Simplified
Secondary_Stack_Size handling as a pragma is now generated for the
corresponding aspect instead of an attribute. Pragma expression is
relocated instead of evaluated. Discriminant of the corresponding
record type is referenced rather than the type discriminant.
(Create_Secondary_Stack_For_Task, Make_Task_Create_Call): Update
Secondary_Stack_Size rep item checks to only look for the pragma rep.
* sem_ch13.adb (Analyze_One_Aspect): Transform
Aspect_Secondary_Stack_Size into a pragma instead of an attribute
because the attribute does not have visibility on a task type's
discriminants when the type's definition is expanded.
(Analyze_Attribute_Definition_Clause): Remove handling of
Attribute_Secondary_Stack_Size.
* snames.adb-tmpl, snames.ads-tmpl: Remove
Attribute_Secondary_Stack_Size, no longer used.
This patch modifies the processing of controlled transient objects within case
expressions represented by an Expression_With_Actions node. The inspection of
an individual action must continue in case it denotes a complex expression,
such as a case statement, which in turn may contain additional transients.
------------
-- Source --
------------
-- pack.ads
with Ada.Finalization; use Ada.Finalization;
package Pack is
function Next_Id return Natural;
type Ctrl is new Controlled with record
Id : Natural := 0;
end record;
procedure Adjust (Obj : in out Ctrl);
procedure Finalize (Obj : in out Ctrl);
procedure Initialize (Obj : in out Ctrl);
function New_Ctrl return Ctrl;
Empty : constant Ctrl := (Controlled with Id => 1);
type Enum is (One, Two, Three);
type Ctrl_Rec is record
Comp : Ctrl;
Kind : Enum;
end record;
procedure Proc (Obj : Ctrl_Rec);
end Pack;
-- pack.adb
with Ada.Text_IO; use Ada.Text_IO;
package body Pack is
Id_Gen : Natural := 1;
procedure Adjust (Obj : in out Ctrl) is
Old_Id : constant Natural := Obj.Id;
New_Id : Natural;
begin
if Old_Id = 0 then
Put_Line (" adj: ERROR already finalized");
else
New_Id := Old_Id * 100;
Put_Line (" adj: " & Old_Id'Img & " ->" & New_Id'Img);
Obj.Id := New_Id;
end if;
end Adjust;
procedure Finalize (Obj : in out Ctrl) is
Old_Id : constant Natural := Obj.Id;
begin
if Old_Id = 0 then
Put_Line (" fin: ERROR already finalized");
else
Put_Line (" fin: " & Old_Id'Img);
Obj.Id := 0;
end if;
end Finalize;
procedure Initialize (Obj : in out Ctrl) is
New_Id : constant Natural := Next_Id;
begin
Put_Line (" ini: " & New_Id'Img);
Obj.Id := New_Id;
end Initialize;
procedure Proc (Obj : Ctrl_Rec) is
begin
Put_Line ("proc : " & Obj.Comp.Id'Img);
end Proc;
function Next_Id return Natural is
begin
Id_Gen := Id_Gen + 1;
return Id_Gen;
end Next_Id;
function New_Ctrl return Ctrl is
Obj : Ctrl;
begin
return Obj;
end New_Ctrl;
end Pack;
-- main.adb
with Ada.Text_IO; use Ada.Text_IO;
with Pack; use Pack;
procedure Main is
procedure Proc_Case_Expr (Mode : Enum) is
begin
Put_Line ("proc_case_expr: " & Mode'Img);
Proc (case Mode is
when One => (Kind => Two, Comp => Empty),
when Two => (Kind => Three, Comp => Empty),
when Three => (Kind => One, Comp => New_Ctrl));
end Proc_Case_Expr;
procedure Proc_If_Expr (Mode : Enum) is
begin
Put_Line ("proc_if_expr: " & Mode'Img);
Proc ((if Mode = One then (Kind => Two, Comp => Empty)
elsif Mode = Two then (Kind => Three, Comp => Empty)
else (Kind => One, Comp => New_Ctrl)));
end Proc_If_Expr;
begin
Proc_Case_Expr (One);
Proc_Case_Expr (Two);
Proc_Case_Expr (Three);
Proc_If_Expr (One);
Proc_If_Expr (Two);
Proc_If_Expr (Three);
end Main;
----------------------------
-- Compilation and output --
----------------------------
* exp_ch4.adb (Process_Action): Do not abandon the inspection of an
individual action because the action may denote a complex expression,
such as a case statement, which in turn may contain additional
transient objects.
Ed Schonberg [Thu, 11 Jan 2018 08:50:29 +0000 (08:50 +0000)]
[Ada] Crash on iterated_component_association in expression function
This patch improves on the handling of the Ada2020 construct Iterated_
Component_Association in various contexts, when the expression involved
is a record or array aggregate.
----
with Text_IO; use Text_IO;
with Exfor; use Exfor;
procedure Main is
Map : String := Table_ASCII;
begin
Put_Line (Map (50..91));
end;
----
package Exfor is
function Table_ASCII return String is
(for I in 1 .. Character'Pos (Character'Last) + 1 => Character'Val(I-1));
end Exfor;
2018-01-11 Ed Schonberg <schonberg@adacore.com>
gcc/ada/
* sem_aggr.adb (Resolve_Iterated_Component_Association): Perform
analysis on a copy of the expression with a copy of the index variable,
because full expansion will rewrite construct into a loop with the
original loop variable.
* exp_aggr.adb (Gen_Assign): Defer analysis and resolution if the
expression is an iterated component association. Full analysis takes
place when construct is rewritten as a loop.
(In_Place_Assign_OK, Safe_Component): An iterated_component_association
is not safe for in-place assignment.
* sem_util.adb (Remove_Entity): Handle properly the case of an isolated
entity with no homonym and no other entity in the scope.
Justin Squirek [Thu, 11 Jan 2018 08:50:25 +0000 (08:50 +0000)]
[Ada] Wrong column in sloc of "expect name" error for loop variant
This patch corrects error messages printed when using the pragma Loop_Variant
without a named argument from having an incorrect column number in some cases.
2018-01-11 Justin Squirek <squirek@adacore.com>
gcc/ada/
* sem_prag.adb (Analyze_Pragma:Pragma_Loop_Variant): Modify error
message to be printed on the pragma argument identifier.
[Ada] Inconsistent scope chain due to quantified expression
This patch corrects the minor decoration performed on invariant procedures in
case the procedure is not inserted into the tree and analyzed. The decoration
now constructs a proper first/next/last entity chain containing the single
formal parameter which represents the object of the related type. The chain
then ensures that any other entities generated by the [pre]analysis of the
invariant expression will be properly added to the chain.
The issue was discovered during the development of the GNATprove tool and
is not visible to end users. No simple test is available because this would
require a debug session.
* exp_util.adb (Build_Invariant_Procedure_Declaration): Set the last
entity of the generated invariant procedure in order to construct a
proper entity chain.
PR middle-end/83575
* cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
when in layout mode.
(cfg_layout_finalize): Do not verify cfg before we are out of layout.
* cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
partition fixup.
2018-01-10 Michael Collison <michael.collison@arm.com>
* config/aarch64/aarch64-builtins.c:
(aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
* config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
(__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
* config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
(AARCH64_ISA_SHA3): New ISA flag.
(TARGET_SHA3): New feature flag for sha3.
* config/aarch64/iterators.md (sha512_op): New int attribute.
(CRYPTO_SHA512): New int iterator.
(UNSPEC_SHA512H): New unspec.
(UNSPEC_SHA512H2): Ditto.
(UNSPEC_SHA512SU0): Ditto.
(UNSPEC_SHA512SU1): Ditto.
* config/aarch64/aarch64-simd-builtins.def
(aarch64_crypto_sha512hqv2di): New builtin.
(aarch64_crypto_sha512h2qv2di): Ditto.
(aarch64_crypto_sha512su0qv2di): Ditto.
(aarch64_crypto_sha512su1qv2di): Ditto.
(aarch64_eor3qv8hi): Ditto.
(aarch64_rax1qv2di): Ditto.
(aarch64_xarqv2di): Ditto.
(aarch64_bcaxqv8hi): Ditto.
* config/aarch64/aarch64-simd.md:
(aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
(aarch64_crypto_sha512su0qv2di): Ditto.
(aarch64_crypto_sha512su1qv2di): Ditto.
(aarch64_eor3qv8hi): Ditto.
(aarch64_rax1qv2di): Ditto.
(aarch64_xarqv2di): Ditto.
(aarch64_bcaxqv8hi): Ditto.
* config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
(vsha512h2q_u64): Ditto.
(vsha512su0q_u64): Ditto.
(vsha512su1q_u64): Ditto.
(veor3q_u16): Ditto.
(vrax1q_u64): Ditto.
(vxarq_u64): Ditto.
(vbcaxq_u16): Ditto.
* config/arm/types.md (crypto_sha512): New type attribute.
(crypto_sha3): Ditto.
(doc/invoke.texi): Document new sha3 option.
2018-01-10 Michael Collison <michael.collison@arm.com>
* config/aarch64/aarch64-builtins.c:
(aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
* config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
(__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
(__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
* config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
(AARCH64_ISA_SM4): New ISA flag.
(TARGET_SM4): New feature flag for sm4.
* config/aarch64/aarch64-simd-builtins.def
(aarch64_sm3ss1qv4si): Ditto.
(aarch64_sm3tt1aq4si): Ditto.
(aarch64_sm3tt1bq4si): Ditto.
(aarch64_sm3tt2aq4si): Ditto.
(aarch64_sm3tt2bq4si): Ditto.
(aarch64_sm3partw1qv4si): Ditto.
(aarch64_sm3partw2qv4si): Ditto.
(aarch64_sm4eqv4si): Ditto.
(aarch64_sm4ekeyqv4si): Ditto.
* config/aarch64/aarch64-simd.md:
(aarch64_sm3ss1qv4si): Ditto.
(aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
(aarch64_sm3partw<sm3part_op>qv4si): Ditto.
(aarch64_sm4eqv4si): Ditto.
(aarch64_sm4ekeyqv4si): Ditto.
* config/aarch64/iterators.md (sm3tt_op): New int iterator.
(sm3part_op): Ditto.
(CRYPTO_SM3TT): Ditto.
(CRYPTO_SM3PART): Ditto.
(UNSPEC_SM3SS1): New unspec.
(UNSPEC_SM3TT1A): Ditto.
(UNSPEC_SM3TT1B): Ditto.
(UNSPEC_SM3TT2A): Ditto.
(UNSPEC_SM3TT2B): Ditto.
(UNSPEC_SM3PARTW1): Ditto.
(UNSPEC_SM3PARTW2): Ditto.
(UNSPEC_SM4E): Ditto.
(UNSPEC_SM4EKEY): Ditto.
* config/aarch64/constraints.md (Ui2): New constraint.
* config/aarch64/predicates.md (aarch64_imm2): New predicate.
* config/arm/types.md (crypto_sm3): New type attribute.
(crypto_sm4): Ditto.
* config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
(vsm3tt1aq_u32): Ditto.
(vsm3tt1bq_u32): Ditto.
(vsm3tt2aq_u32): Ditto.
(vsm3tt2bq_u32): Ditto.
(vsm3partw1q_u32): Ditto.
(vsm3partw2q_u32): Ditto.
(vsm4eq_u32): Ditto.
(vsm4ekeyq_u32): Ditto.
(doc/invoke.texi): Document new sm4 option.
2018-01-10 Michael Collison <michael.collison@arm.com>
* config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
* config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
(AARCH64_FL_FOR_ARCH8_4): New.
(AARCH64_FL_V8_4): New flag.
(doc/invoke.texi): Document new armv8.4-a option.
2018-01-10 Michael Collison <michael.collison@arm.com>
* config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
(__ARM_FEATURE_AES): Define if TARGET_AES is true.
(__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
* config/aarch64/aarch64-option-extension.def: Add
AARCH64_OPT_EXTENSION of 'sha2'.
(aes): Add AARCH64_OPT_EXTENSION of 'aes'.
(crypto): Disable sha2 and aes if crypto disabled.
(crypto): Enable aes and sha2 if enabled.
(simd): Disable sha2 and aes if simd disabled.
* config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
New flags.
(AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
(TARGET_SHA2): New feature flag for sha2.
(TARGET_AES): New feature flag for aes.
* config/aarch64/aarch64-simd.md:
(aarch64_crypto_aes<aes_op>v16qi): Make pattern
conditional on TARGET_AES.
(aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
(aarch64_crypto_sha1hsi): Make pattern conditional
on TARGET_SHA2.
(aarch64_crypto_sha1hv4si): Ditto.
(aarch64_be_crypto_sha1hv4si): Ditto.
(aarch64_crypto_sha1su1v4si): Ditto.
(aarch64_crypto_sha1<sha1_op>v4si): Ditto.
(aarch64_crypto_sha1su0v4si): Ditto.
(aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
(aarch64_crypto_sha256su0v4si): Ditto.
(aarch64_crypto_sha256su1v4si): Ditto.
(doc/invoke.texi): Document new aes and sha2 options.
Eric Botcazou [Wed, 10 Jan 2018 23:36:02 +0000 (23:36 +0000)]
decl.c (gnat_to_gnu_component_type): Apply the check for atomic access once the component size is taken into account...
* gcc-interface/decl.c (gnat_to_gnu_component_type): Apply the check
for atomic access once the component size is taken into account and
also do it if the component type is Atomic or Volatile_Full_Access.
* c-ada-spec.c (dump_number): Add FLOAT_P parameter.
Skip 'f' and 'F' characters if it is true.
(store_ada_macro): Minor tweak.
(dump_ada_macros) <CPP_COMMENT>: Likewise.
<CPP_WSTRING>: Likewise.
<CPP_STRING>: Output '&' in the buffer if not the first string.
<CPP_NUMBER>: Adjust calls to dump_number.
Michael Meissner [Wed, 10 Jan 2018 22:02:31 +0000 (22:02 +0000)]
rs6000.c (is_complex_IBM_long_double): Explicitly check for 128-bit long double before checking TCmode.
2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
* config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
check for 128-bit long double before checking TCmode.
* config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
128-bit long doubles before checking TFmode or TCmode.
(FLOAT128_IBM_P): Likewise.
Martin Sebor [Wed, 10 Jan 2018 21:40:14 +0000 (21:40 +0000)]
PR tree-optimization/83671 - Fix for false positive reported by -Wstringop-overflow does not work with inlining
gcc/testsuite/ChangeLog:
PR tree-optimization/83671
* gcc.dg/strlenopt-40.c: New test.
* gcc.dg/strlenopt-41.c: New test.
gcc/ChangeLog:
PR tree-optimization/83671
* builtins.c (c_strlen): Unconditionally return zero for the empty
string.
Use -Warray-bounds for warnings.
* gimple-fold.c (get_range_strlen): Handle non-constant lengths
for non-constant array indices with COMPONENT_REF, arrays of
arrays, and pointers to arrays.
(gimple_fold_builtin_strlen): Determine and set length range for
non-constant character arrays.
David Malcolm [Wed, 10 Jan 2018 19:40:55 +0000 (19:40 +0000)]
Preserving locations for variable-uses and constants (PR c++/43486)
This patch implements location wrapper nodes, preserving source locations
of the uses of variables and constants in various places in the
C++ frontend: at the arguments at callsites, and for typeid, alignof,
sizeof, and offsetof.
For example, it allows the C++ FE to underline the pertinent argument
for mismatching calls, for such expressions, improving:
extern int callee (int one, const char *two, float three);
int caller (int first, int second, float third)
{
return callee (first, second, third);
}
from
test.cc: In function 'int caller(int, int, float)':
test.cc:5:38: error: invalid conversion from 'int' to 'const char*' [-fpermissive]
return callee (first, second, third);
^
test.cc:1:41: note: initializing argument 2 of 'int callee(int, const char*, float)'
extern int callee (int one, const char *two, float three);
~~~~~~~~~~~~^~~
to:
test.cc: In function 'int caller(int, int, float)':
test.cc:5:25: error: invalid conversion from 'int' to 'const char*' [-fpermissive]
return callee (first, second, third);
^~~~~~
test.cc:1:41: note: initializing argument 2 of 'int callee(int, const char*, float)'
extern int callee (int one, const char *two, float three);
~~~~~~~~~~~~^~~
This is the combination of the following patches:
"[PATCH 01/14] C++: preserve locations within build_address"
https://gcc.gnu.org/ml/gcc-patches/2017-11/msg00883.html
"[PATCH v2.4 of 02/14] Support for adding and stripping location_t wrapper nodes"
https://gcc.gnu.org/ml/gcc-patches/2018-01/msg00591.html
"[PATCH] Eliminate location wrappers in tree_nop_conversion/STRIP_NOPS"
https://gcc.gnu.org/ml/gcc-patches/2017-12/msg01330.html
"[PATCH v4 of 03/14] C++: add location_t wrapper nodes during parsing (minimal impl)"
https://gcc.gnu.org/ml/gcc-patches/2018-01/msg00660.html
"[PATCH 04/14] Update testsuite to show improvements"
https://gcc.gnu.org/ml/gcc-patches/2017-11/msg00891.html
"[v3 of 05/14] C++: handle locations wrappers when calling warn_for_memset"
https://gcc.gnu.org/ml/gcc-patches/2017-12/msg01378.html
"[PATCH 07/14] reject_gcc_builtin: strip any location wrappers"
https://gcc.gnu.org/ml/gcc-patches/2017-11/msg00886.html
"[v3 of PATCH 08/14] cp/tree.c: strip location wrappers in lvalue_kind"
https://gcc.gnu.org/ml/gcc-patches/2017-12/msg01433.html
"[PATCH 09/14] Strip location wrappers in null_ptr_cst_p"
https://gcc.gnu.org/ml/gcc-patches/2017-11/msg00888.html
"[PATCH 11/14] Handle location wrappers in string_conv_p"
https://gcc.gnu.org/ml/gcc-patches/2017-11/msg00890.html
"[v3 of PATCH 13/14] c-format.c: handle location wrappers"
https://gcc.gnu.org/ml/gcc-patches/2017-12/msg01494.html
"[PATCH 14/14] pp_c_cast_expression: don't print casts for location wrappers"
https://gcc.gnu.org/ml/gcc-patches/2017-11/msg00893.html
"[v3 of PATCH 15/14] Use fold_for_warn in get_atomic_generic_size"
https://gcc.gnu.org/ml/gcc-patches/2017-12/msg01380.html
"[PATCH] Add selftest for "fold_for_warn (error_mark_node)""
https://gcc.gnu.org/ml/gcc-patches/2017-12/msg01385.html
gcc/c-family/ChangeLog:
PR c++/43486
* c-common.c: Include "selftest.h".
(get_atomic_generic_size): Perform the test for integral type
before the range test for any integer constant, fixing indentation
of braces. Call fold_for_warn before testing for an INTEGER_CST.
(reject_gcc_builtin): Strip any location wrapper from EXPR.
(selftest::test_fold_for_warn): New function.
(selftest::c_common_c_tests): New function.
(selftest::c_family_tests): Call it, and
selftest::c_pretty_print_c_tests.
* c-common.h (selftest::c_pretty_print_c_tests): New decl.
* c-format.c (check_format_arg): Convert VAR_P check to a
fold_for_warn.
* c-pretty-print.c: Include "selftest.h".
(pp_c_cast_expression): Don't print casts for location wrappers.
(selftest::assert_c_pretty_printer_output): New function.
(ASSERT_C_PRETTY_PRINTER_OUTPUT): New macro.
(selftest::test_location_wrappers): New function.
(selftest::c_pretty_print_c_tests): New function.
* c-warn.c (warn_for_memset): Call fold_for_warn on the arguments.
gcc/cp/ChangeLog:
PR c++/43486
* call.c (null_ptr_cst_p): Strip location wrappers when
converting from '0' to a pointer type in C++11 onwards.
(conversion_null_warnings): Replace comparison with null_node with
call to null_node_p.
(build_over_call): Likewise.
* cp-gimplify.c (cp_fold): Remove the early bailout when
processing_template_decl.
* cp-lang.c (selftest::run_cp_tests): Call
selftest::cp_pt_c_tests and selftest::cp_tree_c_tests.
* cp-tree.h (cp_expr::maybe_add_location_wrapper): New method.
(selftest::run_cp_tests): Move decl to bottom of file.
(null_node_p): New inline function.
(selftest::cp_pt_c_tests): New decl.
(selftest::cp_tree_c_tests): New decl.
* cvt.c (build_expr_type_conversion): Replace comparison with
null_node with call to null_node_p.
* error.c (args_to_string): Likewise.
* except.c (build_throw): Likewise.
* mangle.c (write_expression): Skip location wrapper nodes.
* parser.c (literal_integer_zerop): New function.
(cp_parser_postfix_expression): Call maybe_add_location_wrapper on
the result for RID_TYPEID. Pass true for new "wrap_locations_p"
param of cp_parser_parenthesized_expression_list. When calling
warn_for_memset, replace integer_zerop calls with
literal_integer_zerop, eliminating the double logical negation
cast to bool. Eliminate the special-casing for CONST_DECL in
favor of the fold_for_warn within warn_for_memset.
(cp_parser_parenthesized_expression_list): Add "wrap_locations_p"
param, defaulting to false. Convert "expr" to a cp_expr, and call
maybe_add_location_wrapper on it when wrap_locations_p is true.
(cp_parser_unary_expression): Call maybe_add_location_wrapper on
the result for RID_ALIGNOF and RID_SIZEOF.
(cp_parser_builtin_offsetof): Likewise.
* pt.c: Include "selftest.h".
(tsubst_copy): Handle location wrappers.
(tsubst_copy_and_build): Likewise.
(build_non_dependent_expr): Likewise.
(selftest::test_build_non_dependent_expr): New function.
(selftest::cp_pt_c_tests): New function.
* tree.c: Include "selftest.h".
(lvalue_kind): Handle VIEW_CONVERT_EXPR location wrapper nodes.
(selftest::test_lvalue_kind): New function.
(selftest::cp_tree_c_tests): New function.
* typeck.c (string_conv_p): Strip any location wrapper from "exp".
(cp_build_binary_op): Replace comparison with null_node with call
to null_node_p.
(build_address): Use location of operand when building address
expression.
gcc/testsuite/ChangeLog:
PR c++/43486
* g++.dg/diagnostic/param-type-mismatch.C: Update expected results
to reflect that the arguments are correctly underlined.
* g++.dg/plugin/diagnostic-test-expressions-1.C: Add test coverage
for globals, params, locals and literals.
(test_sizeof): Directly test the location of "sizeof", rather than
when used in compound expressions.
(test_alignof): Likewise for "alignof".
(test_string_literals): Likewise for string literals.
(test_numeric_literals): Likewise for numeric literals.
(test_builtin_offsetof): Likewise for "__builtin_offsetof".
(test_typeid): Likewise for typeid.
(test_unary_plus): New.
* g++.dg/warn/Wformat-1.C: Add tests of pointer arithmetic on
format strings.
gcc/ChangeLog:
PR c++/43486
* tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
"public_flag".
* tree.c (tree_nop_conversion): Return true for location wrapper
nodes.
(maybe_wrap_with_location): New function.
(selftest::check_strip_nops): New function.
(selftest::test_location_wrappers): New function.
(selftest::tree_c_tests): Call it.
* tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
(maybe_wrap_with_location): New decl.
(EXPR_LOCATION_WRAPPER_P): New macro.
(location_wrapper_p): New inline function.
(tree_strip_any_location_wrapper): New inline function.
Ian Lance Taylor [Wed, 10 Jan 2018 18:02:33 +0000 (18:02 +0000)]
Add missing .a files.
Add missing .a files. These should have been committed with the
update to go1.10beta1, but were skipped because by default Subversion
ignores all files matching *.a.
Jakub Jelinek [Wed, 10 Jan 2018 16:59:09 +0000 (17:59 +0100)]
re PR c++/81327 (cast to void* does not suppress -Wclass-memaccess)
PR c++/81327
* call.c (maybe_warn_class_memaccess): Add forward declaration.
Change last argument from tree * to const vec<tree, va_gc> *, adjust
args uses and check number of operands too. Don't strip away any
nops. Use maybe_constant_value when looking for INTEGER_CST args.
Deal with src argument not having pointer type. Check
tree_fits_uhwi_p before calling tree_to_uhwi. Remove useless
test.
(build_over_call): Call maybe_warn_class_memaccess here on the
original arguments.
(build_cxx_call): Rather than here on converted arguments.
* g++.dg/Wclass-memaccess-2.C: Don't expect a warning when explicitly
cast to void *.
H.J. Lu [Wed, 10 Jan 2018 15:37:49 +0000 (15:37 +0000)]
i386: Also adjust stack frame for stack slot alignment
We should also adjust stack_realign_offset for the largest alignment of
stack slot actually used when stack realignment isn't needed. This is
required to keep stack frame properly aligned to satisfy the largest
alignment of stack slots.
Tested on Linux/i686 and Linux/x86-64.
gcc/
PR target/83735
* config/i386/i386.c (ix86_compute_frame_layout): Always adjust
stack_realign_offset for the largest alignment of stack slot
actually used.
(ix86_find_max_used_stack_alignment): New function.
(ix86_finalize_stack_frame_flags): Use it. Set
max_used_stack_alignment if we don't realign stack.
* config/i386/i386.h (machine_function): Add
max_used_stack_alignment.
gcc/testsuite/
PR target/83735
* gcc.target/i386/pr83735.c: New test.
Ian Lance Taylor [Wed, 10 Jan 2018 15:18:55 +0000 (15:18 +0000)]
cmd/go: check for another GCC error message
GCC always recognizes the -fsplit-stack option, but then tests whether
it is supported by the selected target. If not, it reports
cc1: error: ‘-fsplit-stack’ is not supported by this compiler configuration
Check for that error message when deciding whether a compiler option works.
rs6000: Wrap diff of immediates in const (PR83629)
In various of our 32-bit load_toc patterns we take the difference of
two immediates (labels) as a term to something bigger; but this isn't
canonical RTL, it needs to be wrapped in CONST.
PR target/83629
* config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
testsuite/
PR target/83629
* gcc.target/powerpc/pr83629.c: New testcase.
Don't use permutes for single-element accesses (PR83753)
After cunrolling the inner loop, the remaining loop in the testcase
has a single 32-bit access and a group of 64-bit accesses. We first
try to vectorise at 128 bits (VF 4), but decide not to for cost reasons.
We then try with 64 bits (VF 2) instead. This means that the group
of 64-bit accesses uses a single-element vector, which is deliberately
supported as of r251538. We then try to create "permutes" for these
single-element vectors and fall foul of:
for (i = 0; i < 6; i++)
sel[i] += exact_div (nelt, 2);
in vect_grouped_store_supported, since nelt==1.
Maybe we shouldn't even be trying to vectorise statements in the
single-element case, and instead just copy the scalar statement
for each member of the group. But until then, this patch treats
non-strided grouped accesses as VMAT_CONTIGUOUS if no permutation
is necessary.
2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
PR tree-optimization/83753
* tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
for non-strided grouped accesses if the number of elements is 1.
gcc/testsuite/
PR tree-optimization/83753
* gcc.dg/torture/pr83753.c: New test.
Martin Liska [Wed, 10 Jan 2018 10:54:20 +0000 (11:54 +0100)]
Clean up partitioning in try_optimize_cfg (PR bootstrap/82831).
2018-01-10 Martin Liska <mliska@suse.cz>
PR bootstrap/82831
* basic-block.h (CLEANUP_NO_PARTITIONING): New define.
* bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
partitioning.
* cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
CLEANUP_NO_PARTITIONING is not set.
r254296 added support for (const ...) wrappers around vectors,
but in the end the agreement was to use a variable-length
encoding of CONST_VECTOR (and VECTOR_CST) instead. This patch
therefore reverts the bits that are no longer needed.
The rtl.texi part isn't a full revert, since r254296 also updated the
documentation to mention unspecs in address calculations, and to relax
the requirement that the mode had to be Pmode.
2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
* doc/rtl.texi: Remove documentation of (const ...) wrappers
for vectors, as a partial revert of r254296.
* rtl.h (const_vec_p): Delete.
(const_vec_duplicate_p): Don't test for vector CONSTs.
(unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
* expmed.c (make_tree): Likewise.
Revert:
* common.md (E, F): Use CONSTANT_P instead of checking for
CONST_VECTOR.
* emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
checking for CONST_VECTOR.
Ian Lance Taylor [Wed, 10 Jan 2018 05:26:29 +0000 (05:26 +0000)]
runtime: work around escaping closure in export_test.go
When compiling runtime, it is not allowed for local variables
and closures to be heap allocated. In one test, there is a go
statement with a closure. In the gc compiler, it distinguishes
capturing variable by value vs. by address, and rewrites it to
passing the captured values as arguments. Currently we don't
have this, and the escape analysis decides to heap allocate the
closure and also the captured variables, which is not allowed.
Work around it by passing the variables explicitly.
This is in preparation of turning on escape analysis for the
runtime.
Ian Lance Taylor [Wed, 10 Jan 2018 05:15:52 +0000 (05:15 +0000)]
runtime: noescape some functions/variables
This is in preparation of turning on escape analysis for the
runtime.
- In gccgo, systemstack is implemented with mcall, which is not
go:noescape. Wrap the closure in noescape so the escape analysis
does not think it escapes.
- Mark some C functions go:noescape. They do not leak arguments.
- Use noescape function to make a few local variables' addresses
not escape. The escape analysis cannot figure out because they
are assigned to pointer indirections.
compiler: use temporary variable for stack allocation
Currently, allocation expression that can be allocated on stack
is implemented with __builtin_alloca, which turns into
__morestack_allocate_stack_space, which may call C malloc. This
may be slow. Also if this happens during certain runtime
functions (e.g. write barrier), bad things might happen (when
the escape analysis is enabled for the runtime). Make a temporary
variable on stack for the allocation instead.
Also remove the write barrier in the assignment in building heap
expression when it is stack allocated.
compiler: set captured variable address to nonescape until further flooded
The escape analysis models closures by flowing captured variable
address to the closure node. However, the escape state for the
address expressions remained unset as ESCAPE_UNKNOWN. This
caused later passes to conclude that the address escapes. Fix this by
setting its escape state to ESCAPE_NONE first. If it escapes
(because the closure escapes), the flood phase will set its
escape state properly.
compiler: move some escape check to Mark_address_taken
Move some check of escape state earlier, from get_backend to
Mark_address_taken. So we can reclaim escape analysis Nodes
before kicking off the backend (not done in this CL). Also it
makes it easier to check variables and closures do not escape
when the escape analysis is run for the runtime package (also
not done in this CL).
CL 83876 added support of go:noescape pragma, but it only works
for functions called from the same package. The pragma did not
take effect for exported functions that are not called from
the same package. The reason is that top level function
declarations are not traversed, and only reached from calls
from other functions. This CL adds this support. The Traverse
class is extended with a mode to traverse function declarations.
Arrays that are sliced are set to escape in type checking, very
early in compilation. The escape analysis runs later but cannot
undo it. This CL changes it to not escape in the early stage.
Later the escape analysis will make it escape when needed.
Jonathan Wakely [Tue, 9 Jan 2018 21:46:13 +0000 (21:46 +0000)]
PR libstdc++/80276 fix pretty printers for array smart pointers
PR libstdc++/80276
* python/libstdcxx/v6/printers.py (SharedPointerPrinter)
(UniquePointerPrinter): Print correct template argument, not type of
the pointer.
(TemplateTypePrinter._recognizer.recognize): Handle failure to lookup
a type.
* testsuite/libstdc++-prettyprinters/cxx11.cc: Test unique_ptr of
array type.
* testsuite/libstdc++-prettyprinters/cxx17.cc: Test shared_ptr and
weak_ptr of array types.
Cherry Zhang [Tue, 9 Jan 2018 21:33:59 +0000 (21:33 +0000)]
compiler: make top-level decl for address-taken non-escaping locals
If a local variable's address is taken and passed out of its
lexical scope, GCC backend may reuse the stack slot for the
variable, not knowing it is still live through a pointer. In
this case, we create a top-level temporary variable and let the
user-defined variable refer to the temporary variable as its
storage location. As the temporary variable is declared at the
top level, its stack slot will remain live throughout the
function.
Carl Love [Tue, 9 Jan 2018 20:58:04 +0000 (20:58 +0000)]
altivec.md (p8_vmrgow): Add support for V2DI, V2DF, V4SI, V4SF types.
gcc/ChangeLog:
2018-01-09 Carl Love <cel@us.ibm.com>
* config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
V4SI, V4SF types.
(p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
* config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
* config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
* config/rs6000/rs6000-protos.h: Add extern defition for
rs6000_generate_float2_double_code.
* config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
function.
* config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
(float2_v2df): Add define_expand.
gcc/testsuite/ChangeLog:
2017-01-09 Carl Love <cel@us.ibm.com>
* gcc.target/powerpc/builtins-1.c (main): Add tests for vec_mergee and
vec_mergeo builtins with float, double, long long, unsigned long long,
bool long long arguments.
* gcc.target/powerpc/builtins-3-runnable.c (main): Add test for
vec_float2 with double arguments.
* gcc.target/powerpc/builtins-mergew-mergow.c: New runable test for the
vec_mergew and vec_mergow builtins.
Cherry Zhang [Tue, 9 Jan 2018 20:42:08 +0000 (20:42 +0000)]
compiler: add escape analysis debug hash
Add a flag -fgo-debug-escape-hash for debugging escape analysis.
It takes a binary string, optionally led by a "-", as argument.
When specified, the escape analysis runs only on functions whose
name is hashed to a value with matching suffix. The "-" sign
negates the match, i.e. the analysis runs only on functions with
non-matching hash.
Juraj Oršulić [Tue, 9 Jan 2018 18:49:57 +0000 (18:49 +0000)]
PR libstdc++/59253 Improve pretty printers for smart pointers
PR libstdc++/59253 (partial)
* python/libstdcxx/v6/printers.py (SmartPtrIterator): Common iterator
type for pointer stored by shared_ptr, weak_ptr and unique_ptr.
(SharedPointerPrinter, UniquePointerPrinter): Treat stored values as
children.
* testsuite/libstdc++-prettyprinters/cxx11.cc: Update expected output
of unique_ptr printer.
* testsuite/libstdc++-prettyprinters/shared_ptr.cc: Update expected
output of shared_ptr printer.
* config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
function.
(do_ifelse): New function.
(do_isel): New function.
(do_sub3): New function.
(do_add3): New function.
(do_load_mask_compare): New function.
(do_overlap_load_compare): New function.
(expand_compare_loop): New function.
(expand_block_compare): Call expand_compare_loop() when appropriate.
* config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
option description.
(-mblock-compare-inline-loop-limit): New option.
This patch makes the AArch64 vec_perm_const code use the new
vec_perm_indices routines, instead of checking each element individually.
This means that they extend naturally to variable-length vectors.
Also, aarch64_evpc_dup was the only function that generated rtl when
testing_p is true, and that looked accidental. The patch adds the
missing check and then replaces the gen_rtx_REG/start_sequence/
end_sequence stuff with an assert that no rtl is generated.
2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
* config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
instead of checking each element individually.
(aarch64_evpc_uzp): Likewise.
(aarch64_evpc_zip): Likewise.
(aarch64_evpc_ext): Likewise.
(aarch64_evpc_rev): Likewise.
(aarch64_evpc_dup): Test the encoding for a single duplicated element,
instead of checking each element individually. Return true without
generating rtl if
(aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
whether all selected elements come from the same input, instead of
checking each element individually. Remove calls to gen_rtx_REG,
start_sequence and end_sequence and instead assert that no rtl is
generated.
The aarch64_legitimate_constant_p tests for HIGH and CONST seem
to be the wrong way round: (high (const ...)) is valid rtl that
could be passed in, but (const (high ...)) isn't. As it stands,
we disallow anchor+offset but allow (high anchor+offset).
2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
* config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
order of HIGH and CONST checks.
As mentioned in https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01575.html ,
the scatter handling in vectorizable_store seems to be dead code at the
moment. Enabling it with the vect_analyze_data_ref_access part of
that patch triggered an ICE in the avx512f-scatter-*.c tests (which
previously didn't use scatters). The problem was that the NARROW
and WIDEN handling uses permute_vec_elements to marshal the inputs,
and permute_vec_elements expected the lhs of the stmt to be an SSA_NAME,
which of course it isn't for stores.
This patch makes permute_vec_elements create a fresh variable in this case.
2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
* tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
if the destination isn't an SSA_NAME.
Richard Biener [Tue, 9 Jan 2018 13:35:43 +0000 (13:35 +0000)]
re PR tree-optimization/83668 (wrong code with -O -fno-tree-dominator-opts -fgraphite-identity)
2018-01-09 Richard Biener <rguenther@suse.de>
PR tree-optimization/83668
* graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
move prologue...
(canonicalize_loop_form): ... here, renamed from ...
(canonicalize_loop_closed_ssa_form): ... this and amended to
swap successor edges for loop exit blocks to make us use
the RPO order we need for initial schedule generation.
Joseph Myers [Tue, 9 Jan 2018 13:25:38 +0000 (13:25 +0000)]
Fix folding of Inf/NaN comparisons for -ftrapping-math (PR tree-optimization/64811).
The folding of comparisons against Inf (to constants or comparisons
with the maximum finite value) has various cases where it introduces
or loses "invalid" exceptions for comparisons with NaNs.
Folding x > +Inf to 0 should not be about HONOR_SNANS - ordered
comparisons of both quiet and signaling NaNs should raise invalid.
x <= +Inf is not the same as x == x, because again that loses an
exception (equality comparisons don't raise exceptions except for
signaling NaNs).
x == +Inf is not the same as x > DBL_MAX, and a similar issue applies
with the x != +Inf case - that transformation causes a spurious
exception.
This patch fixes the conditionals on the folding to avoid such
introducing or losing exceptions.
Bootstrapped with no regressions on x86_64-pc-linux-gnu (where the
cases involving spurious exceptions wouldn't have failed anyway before
GCC 8 because of unordered comparisons wrongly always having formerly
been used by the back end). Also tested for powerpc-linux-gnu
soft-float that this fixes many glibc math/ test failures that arose
in that configuration because this folding affected the IBM long
double support in libgcc (no such failures appeared for hard-float
because of the bug of powerpc hard-float always using unordered
comparisons) - some failures remain, but I believe them to be
unrelated.
PR tree-optimization/64811
gcc:
* match.pd: When optimizing comparisons with Inf, avoid
introducing or losing exceptions from comparisons with NaN.
gcc/testsuite:
* gcc.dg/torture/inf-compare-1.c, gcc.dg/torture/inf-compare-2.c,
gcc.dg/torture/inf-compare-3.c, gcc.dg/torture/inf-compare-4.c,
gcc.dg/torture/inf-compare-5.c, gcc.dg/torture/inf-compare-6.c,
gcc.dg/torture/inf-compare-7.c, gcc.dg/torture/inf-compare-8.c:
New tests.
* gcc.c-torture/execute/ieee/fp-cmp-7.x: New file.
projects / thirdparty / gcc.git / log
