gcc.git
3 years ago[Ada] Small adjustments to fixed-point I/O units
Eric Botcazou [Mon, 23 Nov 2020 10:39:05 +0000 (11:39 +0100)]
[Ada] Small adjustments to fixed-point I/O units

gcc/ada/

* libgnat/a-tifiio.adb: Adjust documentation.
(OK_Get_32): Compare the object size of the base type.
(OK_Put_32): Likewise.
(OK_Get_64): Likewise.
(OK_Put_64): Likewise.
* libgnat/a-tifiio__128.adb: Adjust documentation.
(OK_Get_32): Compare the object size of the base type.
(OK_Put_32): Likewise.
(OK_Get_64): Likewise.
(OK_Put_64): Likewise.
(OK_Get_128): Likewise.
(OK_Put_128): Likewise.
* libgnat/a-wtfiio.adb (OK_Get_32): Likewise.
(OK_Put_32): Likewise.
(OK_Get_64): Likewise.
(OK_Put_64): Likewise
* libgnat/a-wtfiio__128.adb (OK_Get_32): Likewise.
(OK_Put_32): Likewise.
(OK_Get_64): Likewise.
(OK_Put_64): Likewise.
(OK_Get_128): Likewise.
(OK_Put_128): Likewise.
* libgnat/a-ztfiio.adb (OK_Get_32): Likewise.
(OK_Put_32): Likewise.
(OK_Get_64): Likewise.
(OK_Put_64): Likewise
* libgnat/a-ztfiio__128.adb (OK_Get_32): Likewise.
(OK_Put_32): Likewise.
(OK_Get_64): Likewise.
(OK_Put_64): Likewise.
(OK_Get_128): Likewise.
(OK_Put_128): Likewise.

3 years ago[Ada] Small cleanup in the Ada.Text_IO hierarchy
Eric Botcazou [Mon, 23 Nov 2020 07:50:10 +0000 (08:50 +0100)]
[Ada] Small cleanup in the Ada.Text_IO hierarchy

gcc/ada/

* libgnat/a-tifiio.adb (Get): Replace Current_Input with Current_In.
* libgnat/a-tifiio__128.adb: (Get): Likewise.
* libgnat/a-wtcoio.adb (Get): Likewise.
(Put): Replace Current_Output with Current_Out.
* libgnat/a-wtdeio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtdeio__128.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtenio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtfiio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtfiio__128.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtflio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtinio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtinio__128.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtmoio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-wtmoio__128.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztcoio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztdeio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztdeio__128.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztenio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztfiio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztfiio__128.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztflio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztinio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztinio__128.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztmoio.adb (Get): Likewise.
(Put): Likewise.
* libgnat/a-ztmoio__128.adb (Get): Likewise.
(Put): Likewise.

3 years ago[Ada] Incorrect accessibility level on type in formal package
Justin Squirek [Fri, 20 Nov 2020 13:11:12 +0000 (08:11 -0500)]
[Ada] Incorrect accessibility level on type in formal package

gcc/ada/

* sem_util.adb, sem_util.ads (In_Generic_Formal_Package):
Created to identify type declarations occurring within generic
formal packages.
* sem_res.adb (Resolve_Allocator): Add condition to avoid
emitting an error for allocators when the type being allocated
is class-wide and from a generic formal package.

3 years ago[Ada] Fix alignment warning in System.Fat_Gen unit
Eric Botcazou [Sun, 22 Nov 2020 11:00:02 +0000 (12:00 +0100)]
[Ada] Fix alignment warning in System.Fat_Gen unit

gcc/ada/

* libgnat/s-fatgen.adb (Tiny80): Add alignment clause.

3 years ago[Ada] Add annotation after recent compiler changes
Arnaud Charlet [Sun, 22 Nov 2020 10:34:54 +0000 (05:34 -0500)]
[Ada] Add annotation after recent compiler changes

gcc/ada/

* exp_util.adb (Process_Current_Value_Condition): Add assertion.
* libgnat/s-fatgen.adb (Scaling): Add annotation.

3 years ago[Ada] Fix compile time evaluation of shift intrinsics
Arnaud Charlet [Fri, 20 Nov 2020 15:18:22 +0000 (10:18 -0500)]
[Ada] Fix compile time evaluation of shift intrinsics

gcc/ada/

* sem_eval.adb (Fold_Shift): Compute values using the base type.

3 years ago[Ada] Adjust again previous change to System.Fat_Gen
Eric Botcazou [Sat, 21 Nov 2020 23:54:18 +0000 (00:54 +0100)]
[Ada] Adjust again previous change to System.Fat_Gen

gcc/ada/

* libgnat/s-fatgen.adb: Add with clause for Interfaces and use
type clause for Interfaces.Unsigned_64.
(Small): Comment out.
(Tiny): Likewise.
(Tiny16): New integer constant.
(Tiny32): Likewise.
(Tiny64): Likewise.
(Tiny80): New integer array constant.
(Pred): Declare a local overlay for Tiny.
(Succ): Likewise.

3 years ago[Ada] Fix internal error on bit-packed array in Volatile_Full_Access record
Eric Botcazou [Fri, 20 Nov 2020 20:29:13 +0000 (21:29 +0100)]
[Ada] Fix internal error on bit-packed array in Volatile_Full_Access record

gcc/ada/

* exp_pakd.adb (Expand_Bit_Packed_Element_Set): Fix again packed
array type in complex cases where array is Volatile.
* exp_util.adb (Remove_Side_Effects): Do not force a renaming to
be handled by the back-end.

3 years ago[Ada] Adjust previous change to System.Fat_Gen
Eric Botcazou [Fri, 20 Nov 2020 18:33:21 +0000 (19:33 +0100)]
[Ada] Adjust previous change to System.Fat_Gen

gcc/ada/

* libgnat/s-fatgen.adb: Remove use clause for
System.Unsigned_Types.
(Scaling): Add renaming of System.Unsigned_Types and use type
clause for Long_Long_Unsigned.

3 years ago[Ada] Implement part of System.Fat_Gen more efficiently
Eric Botcazou [Thu, 19 Nov 2020 14:35:35 +0000 (15:35 +0100)]
[Ada] Implement part of System.Fat_Gen more efficiently

gcc/ada/

* libgnat/s-fatgen.ads (Compose): Add pragma Inline.
(Copy_Sign): Likewise.
(Exponent): Likewise.
(Fraction): Likewise.
* libgnat/s-fatgen.adb: Remove with clause for System, add
with and use clauses for System.Unsigned_Types.
Add pragma Warnings (Off) for non-static constants.
Remove precomputed tables of powers of radix and add a few
constants describing the floating-point format.
(Gradual_Scaling): Delete.
(Copy_Sign): Reimplement directly.
(Decompose): Likewise.
(Scaling): Likewise.
(Pred): Speed up.
(Succ): Likewise.
(Truncation): Tidy up.
(Valid): Move constants to library level.

3 years ago[Ada] Remove double initialization of the known value cache
Piotr Trojanek [Fri, 6 Nov 2020 11:01:52 +0000 (12:01 +0100)]
[Ada] Remove double initialization of the known value cache

gcc/ada/

* sem_eval.adb (CV_Cache): Remove initialization at elaboration.

3 years ago[Ada] Fix documentation of -gnatw.K switch (activates => disables)
Gary Dismukes [Thu, 19 Nov 2020 20:18:39 +0000 (15:18 -0500)]
[Ada] Fix documentation of -gnatw.K switch (activates => disables)

gcc/ada/

* doc/gnat_ugn/building_executable_programs_with_gnat.rst:
Correct documentation of the -gnatw.K switch to say that it
disables rather than activates the warning.
* gnat_ugn.texi: Regenerate.

3 years ago[Ada] armhf-linux: symbolic tracebacks
Doug Rupp [Thu, 19 Nov 2020 16:35:47 +0000 (08:35 -0800)]
[Ada] armhf-linux: symbolic tracebacks

gcc/ada/

* tracebak.c: Add a section for ARM Linux.

3 years ago[Ada] Implement AI12-0398-1/03
Ghjuvan Lacambre [Wed, 18 Nov 2020 10:23:05 +0000 (11:23 +0100)]
[Ada] Implement AI12-0398-1/03

gcc/ada/

* par-ch3.adb (P_Discriminant_Part_Opt): Parse aspects, update
documentation.
* par-ch6.adb (P_Return_Statement): Likewise.
* par-ch9.adb (P_Entry_Index_Specification): Likewise.

3 years ago[Ada] Additional fixes for Default_Initial_Condition
Gary Dismukes [Wed, 18 Nov 2020 23:06:14 +0000 (18:06 -0500)]
[Ada] Additional fixes for Default_Initial_Condition

gcc/ada/

* exp_aggr.adb (Build_Array_Aggr_Code.Gen_Assign): Move
generation of the call for DIC check past the optional
generation of calls to controlled Initialize procedures.
* exp_ch3.adb
(Build_Array_Init_Proc.Init_One_Dimension.Possible_DIC_Call):
Suppress generation of a DIC call when the array component type
is controlled.  The call will now be generated later inside the
array's DI (Deep_Initialize) procedure.
* exp_ch7.adb
(Make_Deep_Array_Body.Build_Initialize_Statements): Generate a
DIC call (when needed by the array component type) after any
call to the component type's controlled Initialize procedure, or
generate the DIC call by itself if there's no Initialize to
call.
* sem_aggr.adb (Resolve_Record_Aggregate.Add_Association):
Simplify condition to only test Is_Box_Init_By_Default (previous
condition was overkill, as well as incorrect in some cases).
* sem_elab.adb (Active_Scenarios.Output_Call): For
Default_Initial_Condition, suppress call to
Output_Verification_Call when the subprogram is a partial DIC
procedure.

3 years ago[Ada] Fix couple of bugs in the implementation of Round attribute
Eric Botcazou [Wed, 18 Nov 2020 20:42:18 +0000 (21:42 +0100)]
[Ada] Fix couple of bugs in the implementation of Round attribute

gcc/ada/

* exp_attr.adb (Expand_N_Attribute_Reference) <Attribute_Round>:
Adjust commentary and set the Rounded_Result flag on the type
conversion node when the node is needed.
* exp_ch4.adb (Expand_N_Type_Conversion): Minor tweak.
(Fixup_Universal_Fixed_Operation): Look through the type conversion
only when it is to Universal_Real.
* exp_fixd.adb: Remove with and use clauses for Snames.
(Build_Divide): Remove redundant test.
(Expand_Convert_Float_To_Fixed): Use Rounded_Result flag on the
node to set the truncation parameter.

3 years ago[Ada] Simplify prevention of cascaded errors for Refined_State
Piotr Trojanek [Wed, 18 Nov 2020 14:52:55 +0000 (15:52 +0100)]
[Ada] Simplify prevention of cascaded errors for Refined_State

gcc/ada/

* sem_prag.adb (Analyze_Refinement_Clause): Simplify recently
added code for preventing cascaded errors.

3 years ago[Ada] Compiler crash on call to function instance
Bob Duff [Mon, 16 Nov 2020 20:02:47 +0000 (15:02 -0500)]
[Ada] Compiler crash on call to function instance

gcc/ada/

* exp_ch6.adb (Is_Build_In_Place_Result_Type): Further narrow
the conditions under which we enable build-in-place for
controlled types.

3 years ago[Ada] Fix warning control character for message on IN OUT parameter
Yannick Moy [Tue, 17 Nov 2020 15:04:34 +0000 (16:04 +0100)]
[Ada] Fix warning control character for message on IN OUT parameter

gcc/ada/

* sem_warn.adb (Output_Non_Modified_In_Out_Warnings): Use right
warning control character 'k' in both comment and call to
Errout_Msg_N.

3 years ago[Ada] Refine error messages on illegal Refined_State in SPARK
Yannick Moy [Mon, 16 Nov 2020 20:50:17 +0000 (21:50 +0100)]
[Ada] Refine error messages on illegal Refined_State in SPARK

gcc/ada/

* sem_prag.adb (Analyze_Refined_State_In_Decl_Part): Refine the
error message for missing Part_Of on constituent. Avoid
cascading error.

3 years ago[Ada] Avoid reanalysis of malformed dependency relations
Piotr Trojanek [Mon, 16 Nov 2020 20:40:56 +0000 (21:40 +0100)]
[Ada] Avoid reanalysis of malformed dependency relations

gcc/ada/

* sem_prag.adb (Analyze_Depends_In_Decl_Part): Replace early
returns with goto Leave.
(Collect_Subprogram_Inputs_Outputs): Fix style in comment.

3 years ago[Ada] Tidy up implementation of System.Fat_Gen.Valid and inline it again
Eric Botcazou [Tue, 17 Nov 2020 08:21:19 +0000 (09:21 +0100)]
[Ada] Tidy up implementation of System.Fat_Gen.Valid and inline it again

gcc/ada/

* libgnat/s-fatgen.ads (Valid): Add again pragma Inline.
* libgnat/s-fatgen.adb (Valid): Improve commentary, tidy up left
and right, and remove superfluous trick for denormalized numbers.

3 years ago[Ada] Fix analysis of access objects in Depends contracts
Piotr Trojanek [Mon, 16 Nov 2020 14:21:20 +0000 (15:21 +0100)]
[Ada] Fix analysis of access objects in Depends contracts

gcc/ada/

* sem_prag.adb (Find_Role): Constant object of
access-to-constant and access-to-subprogram types are not
writable.
(Collect_Subprogram_Inputs_Outputs): In-parameters of
access-to-variable type can act as outputs of the Depends
contracts.

3 years ago[Ada] Update comment for processing of pragma Assertion_Policy
Piotr Trojanek [Thu, 5 Nov 2020 09:14:36 +0000 (10:14 +0100)]
[Ada] Update comment for processing of pragma Assertion_Policy

gcc/ada/

* sa_messages.ads: Reference Subprogram_Variant in the comment
for Assertion_Check.
* sem_prag.adb (Analyze_Pragma): Add Subprogram_Variant as an
ID_ASSERTION_KIND; move Default_Initial_Condition as an
RM_ASSERTION_KIND.

3 years ago[Ada] Correctly mark subprogram as not always inlined in GNATprove mode
Yannick Moy [Mon, 16 Nov 2020 11:06:32 +0000 (12:06 +0100)]
[Ada] Correctly mark subprogram as not always inlined in GNATprove mode

gcc/ada/

* inline.adb (Cannot_Inline): Add No_Info parameter to disable
info message.
* inline.ads (Cannot_Inline): When No_Info is set to True, do
not issue info message in GNATprove mode, but still mark the
subprogram as not always inlined.
* sem_res.adb (Resolve_Call): Always call Cannot_Inline inside
an assertion expression.

3 years ago[Ada] Adjust documentation of System.Img_Util.Set_Decimal_Digits
Eric Botcazou [Mon, 16 Nov 2020 11:03:31 +0000 (12:03 +0100)]
[Ada] Adjust documentation of System.Img_Util.Set_Decimal_Digits

gcc/ada/

* libgnat/s-imguti.ads (Set_Decimal_Digits): Adjust documentation.

3 years agopreprocessor: Deferred macro support
Nathan Sidwell [Mon, 14 Dec 2020 15:21:49 +0000 (07:21 -0800)]
preprocessor: Deferred macro support

For deferred macros we also need a new field on the macro itself, so
that the module machinery can determine the macro was imported.  Also
the documentation for the hashnode's deferred field was incomplete.

libcpp/
* include/cpplib.h (struct cpp_macro): Add imported_p field.
(struct cpp_hashnode): Tweak deferred field documentation.
* macro.c (_cpp_new_macro): Clear new field.
(cpp_get_deferred_macro, get_deferred_or_lazy_macro): Assert
more.

3 years agomiddle-end: Exclude TOP permute from blend considerations
Tamar Christina [Mon, 14 Dec 2020 15:20:05 +0000 (15:20 +0000)]
middle-end: Exclude TOP permute from blend considerations

Similarly to UNKNOWN permutes, TOP needs to be excluded from being considered
for blends because it produces no permute to check.

gcc/ChangeLog:

PR middle-end/98264
* tree-vect-slp-patterns.c (linear_loads_p): Exclude TOP permute.

gcc/testsuite/ChangeLog:

PR middle-end/98264
* gcc.target/i386/pr98264.c: New test.

3 years agoLimit perf data buffer during feature checking
Ilya Leoshkevich [Tue, 8 Dec 2020 12:17:58 +0000 (13:17 +0100)]
Limit perf data buffer during feature checking

Commit 2ead1ab91123 ("Limit perf data buffer during profiling") added
-m8 to perf invocations during running tests, but the same problem
exists for checking whether perf is working in the first place.

gcc/testsuite/ChangeLog:

2020-12-08  Ilya Leoshkevich  <iii@linux.ibm.com>

* lib/target-supports.exp(check_profiling_available): Limit
perf data buffer.

3 years ago.dir-locals.el: Set 'fill-column' to 80 for c-mode
Andrea Corallo [Fri, 11 Dec 2020 14:35:25 +0000 (15:35 +0100)]
.dir-locals.el: Set 'fill-column' to 80 for c-mode

ChangeLog

2020-12-11  Andrea Corallo  <andrea.corallo@arm.com>

* .dir-locals.el (c-mode): Set 'fill-column' to 80 columns.

3 years agoarm: Auto-vectorization for MVE: vneg
Christophe Lyon [Mon, 7 Dec 2020 14:43:18 +0000 (14:43 +0000)]
arm: Auto-vectorization for MVE: vneg

This patch enables MVE vneg instructions for auto-vectorization.  MVE
vnegq insns in mve.md are modified to use 'neg' instead of unspec
expression.  The neg<mode>2 expander is added to vec-common.md.

Existing patterns in neon.md are prefixed with neon_.
It's not clear why we have different patterns for VDQW
and VH in neon.md, when WDQWH handles both, and patterns
with VDQ have provision for attributes for FP modes.

Another question is why <absneg_str><mode>2 always sets
neon_abs<q> type when it also handles neon_neq<q> cases.

2020-12-11  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/arm/mve.md (mve_vnegq_f): Use 'neg' instead of unspec.
(mve_vnegq_s): Likewise.
* config/arm/neon.md (neg<mode>2): Rename into neon_neg<mode>2.
(<absneg_str><mode>2): Rename into neon_<absneg_str><mode>2.
(neon_v<absneg_str><mode>): Call gen_neon_<absneg_str><mode>2.
(vashr<mode>3): Call gen_neon_neg<mode>2.
(vlshr<mode>3): Call gen_neon_neg<mode>2.
(neon_vneg<mode>): Call gen_neon_neg<mode>2.
* config/arm/unspecs.md (VNEGQ_F, VNEGQ_S): Remove.
* config/arm/vec-common.md (neg<mode>2): New expander.

gcc/testsuite/
* gcc.target/arm/simd/mve-vneg.c: Add tests for vneg.

3 years agoarm: Auto-vectorization for MVE: vmvn
Christophe Lyon [Wed, 2 Dec 2020 12:20:02 +0000 (12:20 +0000)]
arm: Auto-vectorization for MVE: vmvn

This patch enables MVE vmvnq instructions for auto-vectorization.  MVE
vmvnq insns in mve.md are modified to use 'not' instead of unspec
expression to support one_cmpl<mode>2.  The one_cmpl<mode>2 expander
is added to vec-common.md.

2020-12-11  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/arm/iterators.md (VDQNOTM2): New mode iterator.
(supf): Remove VMVNQ_S and VMVNQ_U.
(VMVNQ): Remove.
* config/arm/mve.md (mve_vmvnq_u<mode>): New entry for vmvn
instruction using expression not.
(mve_vmvnq_s<mode>): New expander.
* config/arm/neon.md (one_cmpl<mode>2): Renamed into
one_cmpl<mode>2_neon.
* config/arm/unspecs.md (VMVNQ_S, VMVNQ_U): Remove.
* config/arm/vec-common.md (one_cmpl<mode>2): New expander.

gcc/testsuite/
* gcc.target/arm/simd/mve-vmvn.c: Add tests for vmvn.

3 years agoarm: Auto-vectorization for MVE: vbic
Christophe Lyon [Wed, 25 Nov 2020 11:08:48 +0000 (11:08 +0000)]
arm: Auto-vectorization for MVE: vbic

This patch enables MVE vbic instructions for auto-vectorization.  MVE
vbicq insns in mve.md are modified to use 'and not' instead of unspec
expression.

2020-12-11  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/arm/iterators.md (supf): Remove VBICQ_S and VBICQ_U.
(VBICQ): Remove.
* config/arm/mve.md (mve_vbicq_u<mode>): New entry for vbic
instruction using expression and not.
(mve_vbicq_s<mode>): New expander.
(mve_vbicq_f<mode>): Replace use of unspec by 'and not'.
* config/arm/unspecs.md (VBICQ_S, VBICQ_U, VBICQ_F): Remove.

gcc/testsuite/
* gcc.target/arm/simd/mve-vbic.c: Add tests for vbic.

3 years agoarm: Auto-vectorization for MVE: veor
Christophe Lyon [Fri, 13 Nov 2020 13:05:43 +0000 (13:05 +0000)]
arm: Auto-vectorization for MVE: veor

This patch enables MVE veorq instructions for auto-vectorization.  MVE
veorq insns in mve.md are modified to use xor instead of unspec
expression to support xor<mode>3.  The xor<mode>3 expander is added to
vec-common.md

2020-12-11  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/arm/iterators.md (supf): Remove VEORQ_S and VEORQ_U.
(VEORQ): Remove.
* config/arm/mve.md (mve_veorq_u<mode>): New entry for veor
instruction using expression xor.
(mve_veorq_s<mode>): New expander.
(mve_veorq_f<mode>): Use 'xor' code instead of unspec.
* config/arm/neon.md (xor<mode>3): Renamed into xor<mode>3_neon.
* config/arm/unspecs.md (VEORQ_S, VEORQ_U, VEORQ_F): Remove.
* config/arm/vec-common.md (xor<mode>3): New expander.

gcc/testsuite/
* gcc.target/arm/simd/mve-veor.c: Add tests for veor.

3 years agoarm,testsuite: Fix vect-half-floats.c test
Christophe Lyon [Mon, 14 Dec 2020 10:40:45 +0000 (10:40 +0000)]
arm,testsuite: Fix vect-half-floats.c test

This patch fixes typos in effective targets which otherwise lead to
DejaGnu errors.

It also replaces dg-additional-options with dg-options to avoid
compiling with -ansi -pedantic-errors, resulting in
error: ISO C does not support the '_Float16' type [-Wpedantic]

2020-12-14  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/testsuite/
* gcc.target/arm/vect-half-floats.c: Fix typos.

3 years agosanitizer: do not ICE for pointer cmp/sub
Martin Liska [Wed, 9 Dec 2020 15:21:18 +0000 (16:21 +0100)]
sanitizer: do not ICE for pointer cmp/sub

gcc/c/ChangeLog:

PR sanitizer/98204
* c-typeck.c (pointer_diff): Do not emit a top-level
sanitization.
(build_binary_op): Likewise.

gcc/testsuite/ChangeLog:

PR sanitizer/98204
* c-c++-common/asan/pr98204.c: New test.

3 years agoaarch64: Add support for Cortex-A78C
Przemyslaw Wirkus [Mon, 14 Dec 2020 09:22:47 +0000 (09:22 +0000)]
aarch64: Add support for Cortex-A78C

This patch adds support for -mcpu=cortex-a78c command line option.
For more information about this processor, see [0]:

[0] https://developer.arm.com/ip-products/processors/cortex-a/cortex-a78c

gcc/ChangeLog:

* config/aarch64/aarch64-cores.def (AARCH64_CORE): Add Cortex-A78C core.
* config/aarch64/aarch64-tune.md: Regenerate.
* doc/invoke.texi: Update docs.

3 years ago-fgo-dump-spec: skip typedefs that match struct tag
Nikhil Benesch [Mon, 14 Dec 2020 07:37:11 +0000 (23:37 -0800)]
-fgo-dump-spec: skip typedefs that match struct tag

gcc/:
* godump.c (go_output_typedef): Suppress typedefs whose name
matches the tag of the underlying struct, union, or enum.
Output declarations for enums that do not appear in typedefs.
gcc/testsuite:
* gcc.misc-tests/godump-1.c: Add test cases.

3 years agolibstdc++: Fix several _GLIBCXX_DEBUG tests
François Dumont [Sat, 12 Dec 2020 17:02:47 +0000 (18:02 +0100)]
libstdc++: Fix several _GLIBCXX_DEBUG tests

libstdc++-v3/ChangeLog:

* testsuite/23_containers/array/debug/back2_neg.cc: target c++14 because assertion
for constexpr is disabled in C++11.
* testsuite/23_containers/array/debug/front2_neg.cc: Likewise.
* testsuite/23_containers/array/debug/square_brackets_operator2_neg.cc: Likewise.
* testsuite/23_containers/vector/debug/multithreaded_swap.cc: Include <memory>
for shared_ptr.

3 years agoDaily bump.
GCC Administrator [Mon, 14 Dec 2020 00:16:29 +0000 (00:16 +0000)]
Daily bump.

3 years agoVAX: Unify push operation selection
Maciej W. Rozycki [Sun, 13 Dec 2020 20:46:17 +0000 (20:46 +0000)]
VAX: Unify push operation selection

Avoid the possibility of code discrepancies like one fixed with the
previous change and improve the structure of code by selecting between
push and non-push operations in a single place in `vax_output_int_move'.

The PUSHAB/MOVAB address moves are never actually produced from this
code as the SImode invocation of this function is guarded with the
`nonsymbolic_operand' predicate, but let's not mess up with this code
too much on this occasion and keep the piece in place.

* config/vax/vax.c (vax_output_int_move): Unify push operation
selection.

3 years agoVAX: Check the correct operand for constant 0 push operation
Maciej W. Rozycki [Sun, 13 Dec 2020 20:46:17 +0000 (20:46 +0000)]
VAX: Check the correct operand for constant 0 push operation

Check the output operand for representing pushing a value onto the stack
rather than the constant 0 input in determining whether to use the PUSHL
or the CLRL instruction for a SImode move.  The latter actually works by
means of using the predecrement addressing mode with the SP register and
the machine code produced even takes the same number of bytes, however
at least with some VAX implementations it incurs a performance penalty.
Besides, we don't want to check the wrong operand anyway and have code
that works by chance only.

Add a test case covering push operations; for operands different from
constant zero there is actually a code size advantage for using PUSHL
rather than the equivalent MOVL instruction.

gcc/
* config/vax/vax.c (vax_output_int_move): Check the correct
operand for constant 0 push operation.

gcc/testsuite/
* gcc.target/vax/push.c: New test.

3 years agoVAX: Handle subtracting from self with QMATH DImode add/sub
Maciej W. Rozycki [Sun, 13 Dec 2020 20:46:17 +0000 (20:46 +0000)]
VAX: Handle subtracting from self with QMATH DImode add/sub

Remove an assertion the failure of which has not been actually observed,
but which appears clearly dangerous, for when the QMATH DImode add/sub
handler is invoked with the subtrahend and the minuend both the same.
Instead handle the operation by emitting a move of constant 0 to the
output operand.  Adjust the relevant inline comment accordingly.

gcc/
* config/vax/vax.c (vax_expand_addsub_di_operands): Handle equal
input operands with subtraction.

3 years agoVAX: Handle constant 0 with QMATH DImode add/sub
Maciej W. Rozycki [Sun, 13 Dec 2020 20:46:17 +0000 (20:46 +0000)]
VAX: Handle constant 0 with QMATH DImode add/sub

Handle constant 0 passed to the QMATH DImode add/sub handler such as
with:

#2  0x0000000011d409b0 in gen_adddi3 (operand0=0x7ffff5c0a128,
    operand1=0x7ffff5c60480, operand2=0x7ffff5c60470)
    at .../gcc/config/vax/vax.md:755
755   "vax_expand_addsub_di_operands (operands, PLUS); DONE;")
(gdb) pr operand0
(reg:DI 31)
(gdb) pr operand1
(const_int 0 [0])
(gdb) pr operand2
(const_int -1 [0xffffffffffffffff])
(gdb)

causing an assertion in `vax_expand_addsub_di_operands':

      gcc_assert (operands[1] != const0_rtx || code == MINUS);

to trigger:

during RTL pass: expand
.../gcc/testsuite/gcc.c-torture/compile/sync-1.c: In function 'test_op_ignore':
.../gcc/testsuite/gcc.c-torture/compile/sync-1.c:33:10: internal compiler error: in vax_expand_addsub_di_operands, at config/vax/vax.c:2080
0x11815003 vax_expand_addsub_di_operands(rtx_def**, rtx_code)
.../gcc/config/vax/vax.c:2080
0x11d409af gen_adddi3(rtx_def*, rtx_def*, rtx_def*)
.../gcc/config/vax/vax.md:755
0x10ea2763 rtx_insn* insn_gen_fn::operator()<rtx_def*, rtx_def*, rtx_def*>(rtx_def*, rtx_def*, rtx_def*) const
.../gcc/recog.h:304
0x10f7fc8f maybe_gen_insn(insn_code, unsigned int, expand_operand*)
.../gcc/optabs.c:7402
0x10f67f8b expand_binop_directly
.../gcc/optabs.c:1122
0x10f684cf expand_binop(machine_mode, optab_tag, rtx_def*, rtx_def*, rtx_def*, int, optab_methods)
.../gcc/optabs.c:1209
0x10f6fb4f expand_unop(machine_mode, optab_tag, rtx_def*, rtx_def*, int)
.../gcc/optabs.c:3013
0x10f6c493 expand_simple_unop(machine_mode, rtx_code, rtx_def*, rtx_def*, int)
.../gcc/optabs.c:2200
0x10f7e2f3 expand_atomic_fetch_op(rtx_def*, rtx_def*, rtx_def*, rtx_code, memmodel, bool)
.../gcc/optabs.c:7021
0x107f7523 expand_builtin_sync_operation
.../gcc/builtins.c:7605
0x107ff547 expand_builtin(tree_node*, rtx_def*, rtx_def*, machine_mode, int)
.../gcc/builtins.c:9430
0x10acda63 expand_expr_real_1(tree_node*, rtx_def*, machine_mode, expand_modifier, rtx_def**, bool)
.../gcc/expr.c:11249
0x10abeb9f expand_expr_real(tree_node*, rtx_def*, machine_mode, expand_modifier, rtx_def**, bool)
.../gcc/expr.c:8486
0x1085606b expand_expr
.../gcc/expr.h:282
0x1086157f expand_call_stmt
.../gcc/cfgexpand.c:2709
0x10865ab7 expand_gimple_stmt_1
.../gcc/cfgexpand.c:3713
0x108662fb expand_gimple_stmt
.../gcc/cfgexpand.c:3877
0x10870387 expand_gimple_basic_block
.../gcc/cfgexpand.c:5918
0x10872b6b execute
.../gcc/cfgexpand.c:6602
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.
compiler exited with status 1
FAIL: gcc.c-torture/compile/sync-1.c   -O0  (internal compiler error)

causing numerous failures in regression testing.

While requesting an addition operation to be produced for the constant
operands of 0 and -1 may seem silly, technically there is nothing wrong
with it, and non-QMATH code (as with the `-mno-qmath' option) has no
issues with that, so neither should QMATH code.  This operation will
normally be folded in later passes anyway.

Observe then, that adding or subtracting constant 0 amounts to a move
(and we even have a machine instruction available to do that with a
single operation) so handle the case explicitly, swapping the addends if
so required, removing the assertion failure and along with that 70 test
suite failures like:

FAIL: gcc.c-torture/compile/sync-1.c   -O0  (internal compiler error)
FAIL: gcc.c-torture/compile/sync-1.c   -O0  fetch_and_nand (test for warnings, line )
FAIL: gcc.c-torture/compile/sync-1.c   -O0  nand_and_fetch (test for warnings, line )
FAIL: gcc.c-torture/compile/sync-1.c   -O0  (test for excess errors)
FAIL: gcc.c-torture/compile/sync-2.c   -O0  (internal compiler error)
FAIL: gcc.c-torture/compile/sync-2.c   -O0   (test for warnings, line )
FAIL: gcc.c-torture/compile/sync-2.c   -O0  (test for excess errors)
FAIL: gcc.c-torture/compile/sync-3.c   -O0  (internal compiler error)
FAIL: gcc.c-torture/compile/sync-3.c   -O0   (test for warnings, line )
FAIL: gcc.c-torture/compile/sync-3.c   -O0  (test for excess errors)

and similarly across all the other optimization levels and compilation
options covered.

gcc/
* config/vax/vax.c (vax_expand_addsub_di_operands): Handle the
addition or subtraction of 0.

3 years agoVAX: Remove unused register allocation from QMATH DImode add/sub handler
Maciej W. Rozycki [Sun, 13 Dec 2020 20:46:16 +0000 (20:46 +0000)]
VAX: Remove unused register allocation from QMATH DImode add/sub handler

An allocation is made for a temporary register, however it is unneeded,
as actually explained in the comment preceding the conditional block in
question, and consequently never used, so remove it.  The `temp' rtx is
already used elsewhere in the function, which is possibly why this dead
assignment has not been warned about.

gcc/
* config/vax/vax.c (vax_expand_addsub_di_operands): Remove
unused register allocation.

3 years agoVAX: Fix lower bound adjustment with `casesi'
Maciej W. Rozycki [Sun, 13 Dec 2020 20:46:16 +0000 (20:46 +0000)]
VAX: Fix lower bound adjustment with `casesi'

Fix an issue with the `casesi' expander using `GEN_INT' to produce the
constant rtx for lower bound adjustment.  This generates a VOIDmode
value which may overflow the SImode range required for the operand to
stay within to satisfy `general_operand', resulting in an ICE like:

.../gcc/testsuite/gcc.c-torture/compile/pr46934.c: In function 'caller':
.../gcc/testsuite/gcc.c-torture/compile/pr46934.c:17:1: error: unrecognizable insn:
(insn 5 2 6 2 (set (reg:SI 25)
        (plus:SI (mem/c:SI (reg/f:SI 17 virtual-incoming-args) [1 reg_type+0 S4 A32])
            (const_int 2147483648 [0x80000000]))) -1
     (nil))
during RTL pass: vregs
.../gcc/testsuite/gcc.c-torture/compile/pr46934.c:17:1: internal compiler error: in extract_insn, at recog.c:2315
0x110d4673 _fatal_insn(char const*, rtx_def const*, char const*, int, char const*)
.../gcc/rtl-error.c:108
0x110d46eb _fatal_insn_not_found(rtx_def const*, char const*, int, char const*)
.../gcc/rtl-error.c:116
0x1106578b extract_insn(rtx_insn*)
.../gcc/recog.c:2315
0x10b63f73 instantiate_virtual_regs_in_insn
.../gcc/function.c:1609
0x10b65b2f instantiate_virtual_regs
.../gcc/function.c:1979
0x10b65ca7 execute
.../gcc/function.c:2028
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.
compiler exited with status 1
FAIL: gcc.c-torture/compile/pr46934.c   -O0  (internal compiler error)

Use `gen_int_mode' to produce the rtx instead, requesting a SImode value
so that the constant gets correctly truncated:

@@ -199,7 +199,7 @@ caller (unsigned int reg_type)

 (insn 5 4 6 (set (reg:SI 25)
         (plus:SI (mem/c:SI (reg/f:SI 17 virtual-incoming-args) [1 reg_type+0 S4 A32])
-            (const_int 2147483648 [0x80000000]))) -1
+            (const_int -2147483648 [0xffffffff80000000]))) -1
      (nil))

 (jump_insn 6 5 7 (set (pc)

removing these test suite failures:

FAIL: gcc.c-torture/compile/pr46934.c   -O0  (internal compiler error)
FAIL: gcc.c-torture/compile/pr46934.c   -O0  (test for excess errors)

with the `vax-netbsdelf' target.

gcc/
* config/vax/vax.md (casesi): Use `gen_int_mode' rather than
`GEN_INT' for the immediate used for lower bound adjustment.

3 years agowidening_mul: Fix a > ~b to .ADD_OVERFLOW optimization [PR98256]
Jakub Jelinek [Sun, 13 Dec 2020 18:25:33 +0000 (19:25 +0100)]
widening_mul: Fix a > ~b to .ADD_OVERFLOW optimization [PR98256]

Unfortunately, my latest tree-ssa-math-opts.c patch broke the following
testcase.  The problem is that the code is adding .ADD_OVERFLOW or
.SUB_OVERFLOW before or after the stmt on which the function has been
called, which is normally a addition or subtraction that has all the
operands.
But in the a > ~b optimization that stmt is the ~b stmt and the other
comparison operand might be defined only after that ~b stmt, so we can't
insert the .ADD_OVERFLOW next to ~b that we want to delete, but need to
insert it before the a > temp comparison that uses it; and in that case
when removing the BIT_NOT_EXPR stmt we need to ensure the caller doesn't do
gsi_next because gsi_remove already points the iterator to the next stmt.

2020-12-13  Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/98256
* tree-ssa-math-opts.c (match_uaddsub_overflow): For BIT_NOT_EXPR,
only handle a single use, and insert .ADD_OVERFLOW before the
comparison rather than after the BIT_NOT_EXPR.  Return true iff
it is BIT_NOT_EXPR and it has been removed.
(math_opts_dom_walker::after_dom_children) <case BIT_NOT_EXPR>:
If match_uaddsub_overflow returned true, continue instead of break.

* gcc.c-torture/compile/pr98256.c: New test.

3 years agoRevert "Arm: Add NEON and MVE RTL patterns for Complex Addition, Multiply and FMA."
Tamar Christina [Sun, 13 Dec 2020 16:49:55 +0000 (16:49 +0000)]
Revert "Arm: Add NEON and MVE RTL patterns for Complex Addition, Multiply and FMA."

This reverts commit 3b8a82f97dd48e153ce93b317c44254839e11461.

Has a dependency on the AArch64 patch which hasn't been approved yet.

3 years agovarasm: Reject soft frame or arg pointer registers for register vars [PR92469]
Jakub Jelinek [Sun, 13 Dec 2020 16:08:08 +0000 (17:08 +0100)]
varasm: Reject soft frame or arg pointer registers for register vars [PR92469]

The following patch rejects frame, argp and retarg registers (unless they are equal
to hard frame pointer registers or if they aren't eliminable) from local or global
register vars.
These are just internal implementation details eliminated later into hard
frame pointer or stack pointer and using them as register variable leads
to numerous ICEs.

2020-12-13  Jakub Jelinek  <jakub@redhat.com>

PR target/92469
* varasm.c (eliminable_regno_p): New function.
(make_decl_rtl): Reject asm vars for frame and argp
if they are different from hard frame pointer.

* gcc.target/i386/pr92469.c: New test.
* gcc.target/i386/pr79804.c: Adjust expected diagnostics.
* gcc.target/i386/pr88178.c: Expect an error.

3 years agoArm: Add NEON and MVE RTL patterns for Complex Addition, Multiply and FMA.
Tamar Christina [Sun, 13 Dec 2020 13:59:33 +0000 (13:59 +0000)]
Arm: Add NEON and MVE RTL patterns for Complex Addition, Multiply and FMA.

This adds implementation for the optabs for complex additions.  With this the
following C code:

  void f90 (float complex a[restrict N], float complex b[restrict N],
    float complex c[restrict N])
  {
    for (int i=0; i < N; i++)
      c[i] = a[i] + (b[i] * I);
  }

generates

  f90:
  add     r3, r2, #1600
  .L2:
  vld1.32 {q8}, [r0]!
  vld1.32 {q9}, [r1]!
  vcadd.f32       q8, q8, q9, #90
  vst1.32 {q8}, [r2]!
  cmp     r3, r2
  bne     .L2
  bx      lr

instead of

  f90:
  add     r3, r2, #1600
  .L2:
  vld2.32 {d24-d27}, [r0]!
  vld2.32 {d20-d23}, [r1]!
  vsub.f32 q8, q12, q11
  vadd.f32 q9, q13, q10
  vst2.32 {d16-d19}, [r2]!
  cmp     r3, r2
  bne     .L2
  bx      lr

gcc/ChangeLog:

* config/arm/arm_mve.h (__arm_vcaddq_rot90_u8, __arm_vcaddq_rot270_u8,
, __arm_vcaddq_rot90_s8, __arm_vcaddq_rot270_s8,
__arm_vcaddq_rot90_u16, __arm_vcaddq_rot270_u16, __arm_vcaddq_rot90_s16,
__arm_vcaddq_rot270_s16, __arm_vcaddq_rot90_u32,
__arm_vcaddq_rot270_u32, __arm_vcaddq_rot90_s32,
__arm_vcaddq_rot270_s32, __arm_vcmulq_rot90_f16,
__arm_vcmulq_rot270_f16, __arm_vcmulq_rot180_f16,
__arm_vcmulq_f16, __arm_vcaddq_rot90_f16, __arm_vcaddq_rot270_f16,
__arm_vcmulq_rot90_f32, __arm_vcmulq_rot270_f32,
__arm_vcmulq_rot180_f32, __arm_vcmulq_f32, __arm_vcaddq_rot90_f32,
__arm_vcaddq_rot270_f32, __arm_vcmlaq_f16, __arm_vcmlaq_rot180_f16,
__arm_vcmlaq_rot270_f16, __arm_vcmlaq_rot90_f16, __arm_vcmlaq_f32,
__arm_vcmlaq_rot180_f32, __arm_vcmlaq_rot270_f32,
__arm_vcmlaq_rot90_f32): Update builtin calls.
* config/arm/arm_mve_builtins.def (vcaddq_rot90_u, vcaddq_rot270_u,
vcaddq_rot90_s, vcaddq_rot270_s, vcaddq_rot90_f, vcaddq_rot270_f,
vcmulq_f, vcmulq_rot90_f, vcmulq_rot180_f, vcmulq_rot270_f,
vcmlaq_f, vcmlaq_rot90_f, vcmlaq_rot180_f, vcmlaq_rot270_f): Removed.
(vcaddq_rot90, vcaddq_rot270, vcmulq, vcmulq_rot90, vcmulq_rot180,
vcmulq_rot270, vcmlaq, vcmlaq_rot90, vcmlaq_rot180, vcmlaq_rot270):
New.
* config/arm/constraints.md (Dz): Include MVE.
* config/arm/iterators.md (mve_rotsplit1, mve_rotsplit2): New.
(rot): Add UNSPEC_VCMLS, UNSPEC_VCMUL and UNSPEC_VCMUL180.
(rot_op, rotsplit1, rotsplit2, fcmac1, VCMLA_OP, VCMUL_OP): New.
* config/arm/mve.md (VCADDQ_ROT270_S, VCADDQ_ROT90_S, VCADDQ_ROT270_U,
VCADDQ_ROT90_U, VCADDQ_ROT270_F, VCADDQ_ROT90_F, VCMULQ_F,
VCMULQ_ROT180_F, VCMULQ_ROT270_F, VCMULQ_ROT90_F, VCMLAQ_F,
VCMLAQ_ROT180_F, VCMLAQ_ROT90_F, VCMLAQ_ROT270_F, VCADDQ_ROT270_S,
VCADDQ_ROT270, VCADDQ_ROT90): Removed.
(mve_rot, VCMUL): New.
(mve_vcaddq_rot270_<supf><mode, mve_vcaddq_rot90_<supf><mode>,
mve_vcaddq_rot270_f<mode>, mve_vcaddq_rot90_f<mode>, mve_vcmulq_f<mode,
mve_vcmulq_rot180_f<mode>, mve_vcmulq_rot270_f<mode>,
mve_vcmulq_rot90_f<mode>, mve_vcmlaq_f<mode>, mve_vcmlaq_rot180_f<mode>,
mve_vcmlaq_rot270_f<mode>, mve_vcmlaq_rot90_f<mode>): Removed.
(mve_vcmlaq<mve_rot><mode>, mve_vcmulq<mve_rot><mode>,
mve_vcaddq<mve_rot><mode>, cadd<rot><mode>3, mve_vcaddq<mve_rot><mode>):
New.
(cmul<rot_op><mode>3): Exclude MVE types.
* config/arm/unspecs.md (UNSPEC_VCMUL90, UNSPEC_VCMUL270): New.
* config/arm/vec-common.md (cadd<rot><mode>3, cmul<rot_op><mode>3,
arm_vcmla<rot><mode>, cml<fcmac1><rot_op><mode>4): New.
* config/arm/unspecs.md (UNSPEC_VCMUL, UNSPEC_VCMUL180, UNSPEC_VCMLS,
UNSPEC_VCMLS180): New.
* config/arm/neon.md (cmul<rot_op><mode>3): New.

3 years agoArm: Add support for auto-vectorization using HF mode.
Tamar Christina [Sun, 13 Dec 2020 13:56:30 +0000 (13:56 +0000)]
Arm: Add support for auto-vectorization using HF mode.

This adds support to the auto-vectorizer to support HFmode vectorization for
AArch32.  This is supported when +fp16 is used.  I wonder if I should disable
the returning of the type if the option isn't enabled.

At the moment it will be returned but the vectorizer will try and fail to use
it.  It wastes a few compile cycles but doesn't result in bad code.

gcc/ChangeLog:

* config/arm/arm.c (arm_preferred_simd_mode): Add E_HFmode.

gcc/testsuite/ChangeLog:

* gcc.target/arm/vect-half-floats.c: New test.

3 years agomiddle-end: Support complex Addition
Tamar Christina [Sun, 13 Dec 2020 13:54:48 +0000 (13:54 +0000)]
middle-end: Support complex Addition

This patch adds support for

  * Complex Addition with rotation of 90 and 270.

  Addition with rotation of the second argument around the Argand plane.
    Supported rotations are 90 and 180.

    c = a + (b * I) and c = a + (b * I * I * I)

gcc/ChangeLog:

* tree-vect-slp-patterns.c: New file.
* Makefile.in: Add it.
* doc/passes.texi: Document it.
* internal-fn.def (COMPLEX_ADD_ROT90, COMPLEX_ADD_ROT270): New.
* optabs.def (cadd90_optab, cadd270_optab): New.
* doc/md.texi: Document them.
* tree-vect-loop.c (vect_analyze_loop_2): Add dissolve code.
* tree-vect-slp.c:
(vect_free_slp_instance, vect_create_new_slp_node): Export.
(vect_match_slp_patterns_2, vect_match_slp_patterns): New.
(vect_analyze_slp): Use it.
* tree-vectorizer.h (vect_free_slp_tree): Export.
(enum _complex_operation): Forward declare.
(class vect_pattern): New

gcc/testsuite/ChangeLog:

* lib/target-supports.exp
(check_effective_target_arm_v8_3a_complex_neon_ok_nocache): Fix it.
(check_effective_target_vect_complex_add_byte
,check_effective_target_vect_complex_add_int
,check_effective_target_vect_complex_add_short
,check_effective_target_vect_complex_add_long
,check_effective_target_vect_complex_add_half
,check_effective_target_vect_complex_add_float
,check_effective_target_vect_complex_add_double): New.
* gcc.dg/vect/complex/bb-slp-complex-add-pattern-byte.c: New test.
* gcc.dg/vect/complex/bb-slp-complex-add-pattern-int.c: New test.
* gcc.dg/vect/complex/bb-slp-complex-add-pattern-long.c: New test.
* gcc.dg/vect/complex/bb-slp-complex-add-pattern-short.c: New test.
* gcc.dg/vect/complex/bb-slp-complex-add-pattern-unsigned-byte.c: New test.
* gcc.dg/vect/complex/bb-slp-complex-add-pattern-unsigned-int.c: New test.
* gcc.dg/vect/complex/bb-slp-complex-add-pattern-unsigned-long.c: New test.
* gcc.dg/vect/complex/bb-slp-complex-add-pattern-unsigned-short.c: New test.
* gcc.dg/vect/complex/complex-add-pattern-template.c: New test.
* gcc.dg/vect/complex/complex-add-template.c: New test.
* gcc.dg/vect/complex/complex-operations-run.c: New test.
* gcc.dg/vect/complex/complex-operations.c: New test.
* gcc.dg/vect/complex/complex.exp: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-double.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-float.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-half-float.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-pattern-double.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-pattern-float.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-add-pattern-half-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-add-double.c: New test.
* gcc.dg/vect/complex/fast-math-complex-add-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-add-half-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-add-pattern-double.c: New test.
* gcc.dg/vect/complex/fast-math-complex-add-pattern-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-add-pattern-half-float.c: New test.
* gcc.dg/vect/complex/vect-complex-add-pattern-byte.c: New test.
* gcc.dg/vect/complex/vect-complex-add-pattern-int.c: New test.
* gcc.dg/vect/complex/vect-complex-add-pattern-long.c: New test.
* gcc.dg/vect/complex/vect-complex-add-pattern-short.c: New test.
* gcc.dg/vect/complex/vect-complex-add-pattern-unsigned-byte.c: New test.
* gcc.dg/vect/complex/vect-complex-add-pattern-unsigned-int.c: New test.
* gcc.dg/vect/complex/vect-complex-add-pattern-unsigned-long.c: New test.
* gcc.dg/vect/complex/vect-complex-add-pattern-unsigned-short.c: New test.

3 years agomiddle-end: Refactor and expose some vectorizer helper functions.
Tamar Christina [Sun, 13 Dec 2020 13:53:48 +0000 (13:53 +0000)]
middle-end: Refactor and expose some vectorizer helper functions.

This is a small refactoring which exposes some helper functions in the
vectorizer so they can be used in other places.

gcc/ChangeLog:

* tree-vect-patterns.c (vect_mark_pattern_stmts): Remove static inline.
* tree-vect-slp.c (vect_create_new_slp_node): Remove static and only
set smts if valid.
* tree-vectorizer.c (vec_info::add_pattern_stmt): New.
(vec_info::set_vinfo_for_stmt): Optionally enforce read-only.
* tree-vectorizer.h (struct _slp_tree): Use new types.
(lane_permutation_t, lane_permutation_t): New.
(vect_create_new_slp_node, vect_mark_pattern_stmts): New.

3 years agoShow coarrays on parse tree dump, implement debug for array references.
Thomas Koenig [Sun, 13 Dec 2020 11:54:57 +0000 (12:54 +0100)]
Show coarrays on parse tree dump, implement debug for array references.

gcc/fortran/ChangeLog:

* dump-parse-tree.c (show_array_ref): Also show coarrays.
(debug): Implement for array reference.

3 years agotestsuite: Fix various scan-assembler-symbol-section issues
Rainer Orth [Sun, 13 Dec 2020 11:48:26 +0000 (12:48 +0100)]
testsuite: Fix various scan-assembler-symbol-section issues

This patch addresses some of the issues that I found when looking into the
failures of the scan-assembler-symbol-section tests on Solaris/SPARC.

* The first issue was that on Solaris/SPARC, section names are
  double-quoted, both with as and gas:

        .section        ".text"

  When using as, the section flag and type syntax is completely
  different from other ELF targets:

        .section        "my_named_section",#alloc,#execinstr,#progbits

  This patch fixes this by stripping double quotes from section names.

* However, this didn't work initially (only the leading quote was
  stripped), which is due to David's recent AIX patch: with the
  introduction of the new capturing group to handle both .section (ELF)
  and .csect (XCOFF), $full_section_directive would never be empty on
  ELF and Mach-O targets, so the extraction of the section name didn't
  work any longer.  This had also broken the Darwin tests completely.

* With working double quote stripping, all but one of the tests PASSed
  on Solaris/SPARC, the exception being:

FAIL: gcc.dg/20021029-1.c scan-assembler-symbol-section symbol ar (found __sparc_get_pc_thunk.l7) has section ^\\\\.(const|rodata)|\\\\[RO\\\\] (found .text.__sparc_get_pc_thunk.l7%__sparc_get_pc_thunk.l7)

  This is due to the symbol name (ar) not being anchored in the test and
  unexpectedly matchting __sparc_get_pc_thunk.l7.

* Next, I ran the tests on Darwin 11 and found two failing tests:

FAIL: gcc.dg/darwin-sections.c scan-assembler-symbol-section symbol ^_a\$ (symbol not found) has section \\\\.data
FAIL: gcc.dg/darwin-sections.c scan-assembler-symbol-section symbol ^_b\$ (symbol not found) has section \\\\.data

  is due to Iain's recent "Darwin : Begin rework of zero-fill sections."
  patch which emits

        .globl _a
        .zerofill __DATA,__common,_a,1,0

  This is already scanned for, so the two scans above can just go.

  The other failing test is

FAIL: g++.dg/gomp/tls-5.C  -std=c++14  scan-assembler-symbol-section symbol ^_?_ZGR2ir_\$ (symbol not found) has section ^\\\\.tdata|\\\\[TL\\\\]
FAIL: g++.dg/gomp/tls-5.C  -std=c++14  scan-assembler-symbol-section symbol ^_?ir\$ (symbol not found) has section ^\\\\.tbss|\\\\[TL\\\\]

  Other scans are guarded by target tls_native, and indeed the assembler
  output has

___emutls_v._ZGR2ir_:
___emutls_t._ZGR2ir_:

___emutls_v.ir:

  Unfortunately scan-assembler-symbol-section doesn't support selects
  yet, which this test implements both for the benefit of this test and
  for symmetry.

With those changes, test results are clean now on sparc-sun-solaris2.11,
i386-pc-solaris2.11, i386-apple-darwin11.4.2, and
powerpc-ibm-aix7.2.4.0.

2020-12-03  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>

gcc:
* doc/sourcebuild.texi (Commands for use in dg-final, Scan the
assembly output, scan-assembler-symbol-section): Document.
(scan-symbol-section): Document.

gcc/testsuite:
* lib/scanasm.exp (scan-symbol-section): Pass args to
dg-scan-symbol-section.
(scan-assembler-symbol-section): Likewise.
(dg-scan-symbol-section): Handle selector from orig_args.
Get patterns from orig_args.
(parse_section_of_symbols): Fix section_pattern.
Strip double quotes from section name.

* g++.dg/gomp/tls-5.C: Restrict ir, _ZGR2ir_ scans to tls_native.
* gcc.dg/20021029-1.c: Anchor ar symbol.
* gcc.dg/darwin-sections.c: Remove obsolete scans for _a, _b in
.data.

3 years agoTweak the way that is_a is implemented
Richard Sandiford [Sun, 13 Dec 2020 10:41:09 +0000 (10:41 +0000)]
Tweak the way that is_a is implemented

At the moment, class hierarchies that use is_a are expected
to define specialisations like:

  template <>
  template <>
  inline bool
  is_a_helper <cgraph_node *>::test (symtab_node *p)
  {
    return p->type == SYMTAB_FUNCTION;
  }

But this doesn't scale well to larger hierarchies, because it only
defines ::test for an argument that is exactly “symtab_node *”
(and not for example “const symtab_node *” or something that
comes between cgraph_node and symtab_node in the hierarchy).

For example:

  struct A { int x; };
  struct B : A {};
  struct C : B {};

  template <>
  template <>
  inline bool
  is_a_helper <C *>::test (A *a)
  {
    return a->x == 1;
  }

  bool f(B *b) { return is_a<C *> (b); }

gives:

  warning: inline function ‘static bool is_a_helper<T>::test(U*) [with U = B; T = C*]’ used but never defined

and:

  bool f(const A *a) { return is_a<const C *> (a); }

gives:

  warning: inline function ‘static bool is_a_helper<T>::test(U*) [with U = const A; T = const C*]’ used but never defined

This patch instead allows is_a to be implemented by specialising
is_a_helper as a whole, for example:

  template<>
  struct is_a_helper<C *> : static_is_a_helper<C *>
  {
    static inline bool test (const A *a) { return a->x == 1; }
  };

It also adds a general specialisation of is_a_helper for const
pointers.  Together, this makes both of the above examples work.

gcc/
* is-a.h (reinterpret_is_a_helper): New class.
(static_is_a_helper): Likewise.
(is_a_helper): Inherit from reinterpret_is_a_helper.
(is_a_helper<const T *>): New specialization.

3 years agoMove iterator_range to a new iterator-utils.h file
Richard Sandiford [Sun, 13 Dec 2020 10:41:08 +0000 (10:41 +0000)]
Move iterator_range to a new iterator-utils.h file

A later patch will add more iterator-related utilities.  Rather than
putting them all directly in coretypes.h, it seemed better to add a
new header file, here called "iterator-utils.h".  This preliminary
patch moves the existing iterator_range class there too.

I used the same copyright date range as coretypes.h “just to be sure”.

gcc/
* coretypes.h (iterator_range): Move to...
* iterator-utils.h: ...this new file.

3 years agortlanal: Remove noop_move_p REG_EQUAL condition
Richard Sandiford [Sun, 13 Dec 2020 10:41:07 +0000 (10:41 +0000)]
rtlanal: Remove noop_move_p REG_EQUAL condition

noop_move_p currently keeps any instruction that has a REG_EQUAL
note, on the basis that the equality might be useful in future.
But this creates a perverse incentive not to add potentially-useful
REG_EQUAL notes, in case they prevent an instruction from later being
removed as dead.

The condition originates from flow.c:life_analysis_1 and predates
the changes tracked by the current repository (1992).  It probably
made sense when most optimisations were done on RTL rather than FE
trees, but it seems counterproductive now.

gcc/
* rtlanal.c (noop_move_p): Don't check for REG_EQUAL notes.

3 years agovec: Silence clang warning
Richard Sandiford [Sun, 13 Dec 2020 10:41:06 +0000 (10:41 +0000)]
vec: Silence clang warning

I noticed during compatibility testing that clang warns that this
operator won't be implicitly const in C++14 onwards.

gcc/
* vec.h (vnull::operator vec<T, A, L>): Make const.

3 years agoDaily bump.
GCC Administrator [Sun, 13 Dec 2020 00:16:19 +0000 (00:16 +0000)]
Daily bump.

3 years agolibstdc++: Fix _GLIBCXX_DEBUG mode constexpr compatibility
François Dumont [Fri, 11 Dec 2020 06:48:06 +0000 (07:48 +0100)]
libstdc++: Fix _GLIBCXX_DEBUG mode constexpr compatibility

The __glibcxx_check_can_[increment|decrement]_range macros are using the
_GLIBCXX_DEBUG_VERIFY_COND_AT macro which is not constexpr compliant and will produce nasty
diagnostics rather than the std::__failed_assertion dedicated to constexpr. Replace it with
correct _GLIBCXX_DEBUG_VERIFY_AT_F.

libstdc++-v3/ChangeLog:

* include/debug/macros.h (__glibcxx_check_can_increment_range): Replace
_GLIBCXX_DEBUG_VERIFY_COND_AT usage with _GLIBCXX_DEBUG_VERIFY_AT_F.
(__glibcxx_check_can_decrement_range): Likewise.
* testsuite/25_algorithms/copy_backward/constexpr.cc (test03): New.
* testsuite/25_algorithms/copy/debug/constexpr_neg.cc: New test.
* testsuite/25_algorithms/copy_backward/debug/constexpr_neg.cc: New test.
* testsuite/25_algorithms/equal/constexpr_neg.cc: New test.
* testsuite/25_algorithms/equal/debug/constexpr_neg.cc: New test.

3 years agoFortran: Enable inquiry references in data statements [PR98022].
Paul Thomas [Sat, 12 Dec 2020 14:01:08 +0000 (14:01 +0000)]
Fortran: Enable inquiry references in data statements [PR98022].

2020-12-12  Paul Thomas  <pault@gcc.gnu.org>

gcc/fortran
PR fortran/98022
* data.c (gfc_assign_data_value): Handle inquiry references in
the data statement object list.

gcc/testsuite/
PR fortran/98022
* gfortran.dg/data_inquiry_ref.f90: New test.

3 years agomatch.pd: Add ~(X - Y) -> ~X + Y simplification [PR96685]
Jakub Jelinek [Sat, 12 Dec 2020 13:49:57 +0000 (14:49 +0100)]
match.pd: Add ~(X - Y) -> ~X + Y simplification [PR96685]

This patch adds the ~(X - Y) -> ~X + Y simplification requested
in the PR (plus also ~(X + C) -> ~X + (-C) for constants C that can
be safely negated.

The first two simplify blocks is what has been requested in the PR
and that makes the first testcase pass.
Unfortunately, that change also breaks the second testcase, because
while the same expressions appearing in the same stmt and split
across multiple stmts has been folded (not really) before, with
this optimization fold-const.c optimizes ~X + Y further into
(Y - X) - 1 in fold_binary_loc associate: code, but we have nothing
like that in GIMPLE and so end up with different expressions.

The last simplify is an attempt to deal with just this case,
had to rule out there the Y == -1U case, because then we
reached infinite recursion as ~X + -1U was canonicalized by
the pattern into (-1U - X) + -1U but there is a canonicalization
-1 - A -> ~A that turns it back.  Furthermore, had to make it #if
GIMPLE only, because it otherwise resulted in infinite recursion
when interacting with the associate: optimization.
The end result is that we pass all 3 testcases and thus canonizalize
the 3 possible forms of writing the same thing.

2020-12-12  Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/96685
* match.pd (~(X - Y) -> ~X + Y): New optimization.
(~X + Y -> (Y - X) - 1): Likewise.

* gcc.dg/tree-ssa/pr96685-1.c: New test.
* gcc.dg/tree-ssa/pr96685-2.c: New test.
* gcc.dg/tree-ssa/pr96685-3.c: New test.

3 years agowidening_mul: Recognize another form of ADD_OVERFLOW [PR96272]
Jakub Jelinek [Sat, 12 Dec 2020 13:48:47 +0000 (14:48 +0100)]
widening_mul: Recognize another form of ADD_OVERFLOW [PR96272]

The following patch recognizes another form of hand written
__builtin_add_overflow (this time _p), in particular when
the code does unsigned
if (x > ~0U - y)
or
if (x <= ~0U - y)
it can be optimized (if the subtraction turned into ~y is single use)
into
if (__builtin_add_overflow_p (x, y, 0U))
or
if (!__builtin_add_overflow_p (x, y, 0U))
and generate better code, e.g. for the first function in the testcase:
-       movl    %esi, %eax
        addl    %edi, %esi
-       notl    %eax
-       cmpl    %edi, %eax
-       movl    $-1, %eax
-       cmovnb  %esi, %eax
+       jc      .L3
+       movl    %esi, %eax
+       ret
+.L3:
+       orl     $-1, %eax
        ret
on x86_64.  As for the jumps vs. conditional move case, that is some CE
issue with complex branch patterns we should fix up no matter what, but
in this case I'm actually not sure if branchy code isn't better, overflow
is something that isn't that common.

2020-12-12  Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/96272
* tree-ssa-math-opts.c (uaddsub_overflow_check_p): Add OTHER argument.
Handle BIT_NOT_EXPR.
(match_uaddsub_overflow): Optimize unsigned a > ~b into
__imag__ .ADD_OVERFLOW (a, b).
(math_opts_dom_walker::after_dom_children): Call match_uaddsub_overflow
even for BIT_NOT_EXPR.

* gcc.dg/tree-ssa/pr96272.c: New test.

3 years agoopenmp, openacc: Fix up handling of data regions [PR98183]
Jakub Jelinek [Sat, 12 Dec 2020 07:36:02 +0000 (08:36 +0100)]
openmp, openacc: Fix up handling of data regions [PR98183]

While the data regions (target data and OpenACC counterparts) aren't
standalone directives, unlike most other OpenMP/OpenACC constructs
we allow (apparently as an extension) exceptions and goto out of
the block. During gimplification we place an *end* call into a finally
block so that it is reached even on exceptions or goto out etc.).
During omplower pass we then add paired #pragma omp return for them,
but due to the exceptions because the region is not SESE we can end up
with #pragma omp return appearing only conditionally in the CFG etc.,
which the ompexp pass can't handle.
For the ompexp pass, we actually don't care about the end part or about
target data nesting, so we can treat it as standalone directive.

2020-12-12  Jakub Jelinek  <jakub@redhat.com>

PR middle-end/98183
* omp-low.c (lower_omp_target): Don't add OMP_RETURN for
data regions.
* omp-expand.c (expand_omp_target): Don't try to remove
OMP_RETURN for data regions.
(build_omp_regions_1, omp_make_gimple_edges): Don't expect
OMP_RETURN for data regions.

* gcc.dg/gomp/pr98183.c: New test.
* gcc.dg/goacc/pr98183.c: New test.

3 years agoDaily bump.
GCC Administrator [Sat, 12 Dec 2020 00:16:35 +0000 (00:16 +0000)]
Daily bump.

3 years agoc++: Avoid considering some conversion ops [PR97600]
Jason Merrill [Fri, 11 Dec 2020 19:37:09 +0000 (14:37 -0500)]
c++: Avoid considering some conversion ops [PR97600]

Patrick's earlier patch to check convertibility before constraints for
conversion ops wasn't suitable because checking convertibility can also lead
to unwanted instantiations, but it occurs to me that there's a smaller check
we can do to avoid doing normal consideration of the conversion ops in this
case: since we're in the middle of a user-defined conversion, we can exclude
from consideration any conversion ops that return a type that would need an
additional user-defined conversion to reach the desired type: namely, a type
that differs in class-ness from the desired type.

[temp.inst]/9 allows optimizations like this: "If the function selected by
overload resolution can be determined without instantiating a class template
definition, it is unspecified whether that instantiation actually takes
place."

gcc/cp/ChangeLog:

PR libstdc++/97600
* call.c (build_user_type_conversion_1): Avoid considering
conversion functions that return a clearly unsuitable type.

gcc/testsuite/ChangeLog:

* g++.dg/cpp2a/concepts-conv3.C: New test.

3 years agoc++: Fix build with --enable-gather-detailed-mem-stats.
Jason Merrill [Fri, 11 Dec 2020 19:35:20 +0000 (14:35 -0500)]
c++: Fix build with --enable-gather-detailed-mem-stats.

Nathan's recent patch added make_binding_vec defined with MEM_STAT_DECL, but
didn't add the parallel decoration to the forward declaration.

gcc/cp/ChangeLog:

* cp-tree.h (make_binding_vec): Add CXX_MEM_STAT_INFO.

3 years agoc++: Final module preparations
Nathan Sidwell [Fri, 11 Dec 2020 19:10:40 +0000 (11:10 -0800)]
c++: Final module preparations

This adds the final few preparations to drop modules in.  I'd missed a
couple of changes to core compiler -- a new pair of preprocessor
options, and marking the boundary of fixed and lazy global trees.

For C++, we need to add module.cc to the GTY scanner.  Parsing final
cleanups needs a few tweaks for modules.  Lambdas used to initialize a
global (for instance) get an extra scope, but we now need to point
that object to the lambda too.  Finally template instantiation needs
to do lazy loading before looking at the available instantiations and
specializations.

gcc/
* gcc.c (cpp_unique_options): Add Mmodules, Mno-modules.
* tree-core.h (enum tree_index): Add TI_MODULE_HWM.
gcc/cp/
* config-lang.in (gtfiles): Add cp/module.cc.
* decl2.c (c_parse_final_cleanups): Add module support.
* lambda.c (record_lambda_scope): Call maybe_attach_decl.
* module.cc (maybe_attach_decl, lazy_load_specializations): Stubs.
(finish_module_procesing): Stub.
* pt.c (lookup_template_class_1): Lazy load specializations.
(instantiate_template_1): Likewise.

3 years agoc++: Refactor final cleanup
Nathan Sidwell [Fri, 11 Dec 2020 18:05:41 +0000 (10:05 -0800)]
c++: Refactor final cleanup

This is a small refactor of the end of decl processing, into which
dropping module support will be simpler.

gcc/cp/
* decl2.c (c_parse_final_cleanups): Refactor loop.

3 years agoAdd missing varasm DECL_P check.
Jim Wilson [Thu, 10 Dec 2020 02:57:32 +0000 (18:57 -0800)]
Add missing varasm DECL_P check.

This fixes a riscv64-linux bootstrap failure.

get_constant_section calls the select_section target hook, and select_section
calls get_named_section which calls get_section.  So it is possible to have
a constant not a decl in both of these functions.  They already call DECL_P
checks everywhere except for the new code HJ recently added.  This adds the
missing DECL_P check.

gcc/
* varasm.c (get_section): Add DECL_P check before DECL_PRESERVE_P.

3 years agoDaily bump.
GCC Administrator [Fri, 11 Dec 2020 18:22:52 +0000 (18:22 +0000)]
Daily bump.

3 years agocompiler: encode user visible names if necessary
Ian Lance Taylor [Fri, 11 Dec 2020 05:07:27 +0000 (21:07 -0800)]
compiler: encode user visible names if necessary

Avoid putting weird characters into the user visible name.
It breaks stabs in particular, and may also cause debugger problems.
Instead, encode those names, and use a "g." prefix to tell the debugger.

Also dereference the type for the name of a recover thunk, to avoid a
pointless '*' that gets encoded.

Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/277232

3 years agoarm: Auto-vectorization for MVE clean condition for vand and vorr expanders
Christophe Lyon [Fri, 11 Dec 2020 16:46:26 +0000 (16:46 +0000)]
arm: Auto-vectorization for MVE clean condition for vand and vorr expanders

The patch restores the unconditional definition of the VDQ iterator,
and changes the conditions of the vand and vorr expanders to use
ARM_HAVE_<MODE>_ARITH.

2020-12-11  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/arm/iterators.md (VDQ): Remove TARGET_HAVE_MVE
conditions.
* config/arm/vec-common.md (and<mode>3): Use
ARM_HAVE_<MODE>_ARITH.
(ior<mode>3): Likewise.

3 years agoarc: Update ARC700 cache hazard detection.
Claudiu Zissulescu [Fri, 11 Dec 2020 16:27:21 +0000 (18:27 +0200)]
arc: Update ARC700 cache hazard detection.

Replace/update ARC700 cache hazard detection. The next situations are
handled:

- There are 2 stores back2back, then 3 loads in next 3 or 4 instructions.

    if 3 loads in 3 instructions then we insert 2 nops after stores.
    if 3 loads in 4 instructions then we insert 1 nop after stores

- 2 back to back stores, followed by at least 3 loads in next 4 instructions.
        st st ld ld ld ##
        st st ## ld ld ld
        st st ld ## ld ld
        st st ld ld ## ld
        ## - any instruction

- store between non-store instructions, followed by 3 loads
        $$ st SS ld ld ld
        $$ - non-store instruction, even load.

gcc/
2020-12-11  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (arc_active_insn): Ignore all non essential
instructions when getting the next active instruction.
(check_store_cacheline_hazard): Update.
(workaround_arc_anomaly): Remove obsolete cache hazard code.

Signed-off-by: Claudiu Zissulescu <claziss@gmail.com>
3 years agoarc: Avoid generating brcc instructions with limm
Claudiu Zissulescu [Fri, 11 Dec 2020 16:27:21 +0000 (18:27 +0200)]
arc: Avoid generating brcc instructions with limm

BRcc instructions are generated quite late in the compilation
process. These instructions combines a compare with a regular
conditional branch if the result of the compare is not used
anylonger. However, when compiling for size, it is better to avoid
BRcc instructions which are introducing a 32-bit long immediate.

gcc/
2020-12-11  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (arc_reorg): Avoid limm in BRcc.

3 years agoarc: Refurbish adc/sbc patterns
Claudiu Zissulescu [Fri, 11 Dec 2020 16:27:21 +0000 (18:27 +0200)]
arc: Refurbish adc/sbc patterns

The adc/sbc patterns were unecessary spliting, remove that and
associated functions.

gcc/
2020-12-11  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc-protos.h (arc_scheduling_not_expected): Remove
it.
(arc_sets_cc_p): Likewise.
(arc_need_delay): Likewise.
* config/arc/arc.c (arc_sets_cc_p): Likewise.
(arc_need_delay): Likewise.
(arc_scheduling_not_expected): Likewise.
* config/arc/arc.md: Convert adc/sbc patterns to simple
instruction definitions.

Signed-off-by: Claudiu Zissulescu <claziss@synopsys.com>
3 years agoc++: module test harness
Nathan Sidwell [Fri, 11 Dec 2020 16:26:27 +0000 (08:26 -0800)]
c++: module test harness

Here is the module test harness -- but no tests.

gcc/testsuite/
* g++.dg/modules/modules.exp: New.

3 years agoc++: cp_tree_equal tweaks
Nathan Sidwell [Fri, 11 Dec 2020 16:22:57 +0000 (08:22 -0800)]
c++: cp_tree_equal tweaks

When comparing streamed trees we can encounter NON_LVALUE_EXPR and
VIEW_CONVERT_EXPRs with null types.  Also, when checking a potential
duplicate we don't want to reject PARM_DECLs with different contexts,
if those two contexts are the two decls of interest.

gcc/cp/
* cp-tree.h (map_context_from, map_context_to): Declare.
* module.cc (map_context_from, map_context_to): Define.
* tree.c (cp_tree_equal): Check map_context_{from,to} for parm
context difference.  Allow NON_LVALUE_EXPR and VIEW_CONVERT_EXPR
with null types.

3 years agoarm: Auto-vectorization for MVE: vorr
Christophe Lyon [Fri, 13 Nov 2020 12:34:12 +0000 (12:34 +0000)]
arm: Auto-vectorization for MVE: vorr

This patch enables MVE vorrq instructions for auto-vectorization.  MVE
vorrq insns in mve.md are modified to use ior instead of unspec
expression to support ior<mode>3.  The ior<mode>3 expander is added to
vec-common.md

2020-12-03  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/arm/iterators.md (supf): Remove VORRQ_S and VORRQ_U.
(VORRQ): Remove.
* config/arm/mve.md (mve_vorrq_s<mode>): New entry for vorr
instruction using expression ior.
(mve_vorrq_u<mode>): New expander.
(mve_vorrq_f<mode>): Use ior code instead of unspec.
* config/arm/neon.md (ior<mode>3): Renamed into ior<mode>3_neon.
* config/arm/predicates.md (imm_for_neon_logic_operand): Enable
for MVE.
* config/arm/unspecs.md (VORRQ_S, VORRQ_U, VORRQ_F): Remove.
* config/arm/vec-common.md (ior<mode>3): New expander.

gcc/testsuite/
* gcc.target/arm/simd/mve-vorr.c: Add vorr tests.

3 years agoarc: Use separate predicated patterns for mpyd(u)
Claudiu Zissulescu [Tue, 28 Jul 2020 10:15:48 +0000 (13:15 +0300)]
arc: Use separate predicated patterns for mpyd(u)

The compiler can match mpyd.eq r0,r1,r0 as a predicated instruction,
which is incorrect. The mpyd(u) instruction takes as input two 32-bit
registers, returning into a double 64-bit even-odd register pair.  For
the predicated case, the ARC instruction decoder expects the
destination register to be the same as the first input register. In
the big-endian case the result is swaped in the destination register
pair, however, the instruction encoding remains the same.  Refurbish
the mpyd(u) patterns to take into account the above observation.

gcc/
2020-12-11  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (mpyd<su_optab>_arcv2hs): New template
pattern.
(*pmpyd<su_optab>_arcv2hs): Likewise.
(*pmpyd<su_optab>_imm_arcv2hs): Likewise.
(mpyd_arcv2hs): Moved into above template.
(mpyd_imm_arcv2hs): Moved into above template.
(mpydu_arcv2hs): Likewise.
(mpydu_imm_arcv2hs): Likewise.
(su_optab): New optab prefix for sign/zero-extending operations.

gcc/testsuite/
2020-12-11  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/pmpyd.c: New test.
* gcc.target/arc/tmac-1.c: Update.

Signed-off-by: Claudiu Zissulescu <claziss@gmail.com>
3 years agox86: Update user interrupt handler stack frame
H.J. Lu [Thu, 10 Dec 2020 02:14:24 +0000 (18:14 -0800)]
x86: Update user interrupt handler stack frame

User interrupt handler stack frame is similar to exception interrupt
handler stack frame.  Instead of error code, the second argument is
user interrupt request register vector.

gcc/

PR target/98219
* config/i386/uintrintrin.h (__uintr_frame): Remove uirrv.

gcc/testsuite/

PR target/98219
* gcc.dg/guality/pr98219-1.c: New test.
* gcc.dg/guality/pr98219-2.c: Likewise.
* gcc.dg/torture/pr98219-1.c: Likewise.
* gcc.dg/torture/pr98219-2.c: Likewise.
* gcc.target/i386/uintr-2.c: Scan "add[lq] $8, %[er]sp".
(uword_t): New.
(foo): Add a uword_t argument.
(UINTR_hanlder): Likewise.
* gcc.target/i386/uintr-3.c: Scan "add[lq] $8, %[er]sp".
(uword_t): New.
(UINTR_hanlder): Add a uword_t argument.
* gcc.target/i386/uintr-4.c (uword_t): New.
(UINTR_hanlder): Add a uword_t argument.
* gcc.target/i386/uintr-5.c (uword_t): New.
(UINTR_hanlder): Add a uword_t argument.

3 years agoc++: Module lang hook overriding
Nathan Sidwell [Fri, 11 Dec 2020 14:42:26 +0000 (06:42 -0800)]
c++: Module lang hook overriding

This installs stub lang hooks for modules and creates the module dump file.

gcc/cp/
* cp-lang.c (LANG_HOOKS_PREPROCESS_MAIN_FILE): Override.
(LANG_HOOKS_PREPROCESS_OPTIONS): Override.
(LANG_HOOKS_PREPROCESS_TOKEN): Override.
* cp-objcp-common.c (cp_register_dumps): Add module dump.
(cp_handle_option): New.
* cp-objcp-common.h (cp_handle_option): Declare.
(LANG_HOOKS_HANDLE_OPTION): Override.
* cp-tree.h (module_dump_id): Declare.
* module.cc (module_dump_id): Define.
(module_begin_main_file, handle_module_option)
(module_preproces_options): Stubs.

3 years agoc++: name lookup API for modules
Nathan Sidwell [Thu, 10 Dec 2020 21:05:44 +0000 (13:05 -0800)]
c++: name lookup API for modules

This adds a set of calls to name lookup that are needed by modules.
Generally installing imported bindings, or walking the current TU's
bindings.  One note about template instantiations though.  When we're
about to instantiate a template we have to know about all the
maybe-partial specializations that exist.  These can be in any
imported module -- not necesarily the module defining the template.
Thus we key such foreign templates to the innermost namespace and
identifier of the containing entitity -- that's the only thing we have
a handle on.  That's why we note and load pending specializations here.

gcc/cp/
* module.cc (lazy_specializations_p): Stub.
* name-lookup.h (append_imported_binding_slot)
(mergeable_namespacE_slots, lookup_class_binding)
(walk_module_binding, import_module_binding, set_module_binding)
(note_pending_specializations, load_pending_specializations)
(add_module_decl, add_imported_namespace): Declare.
(get_cxx_dialect_name): Declare.
(enum WMB_flags): New.
* name-lookup.c (append_imported_binding_slot)
(mergeable_namespacE_slots, lookup_class_binding)
(walk_module_binding, import_module_binding, set_module_binding)
(note_pending_specializations, load_pending_specializations)
(add_module_decl, add_imported_namespace): New.
(get_cxx_dialect_name): Make extern.

3 years agoc++: missing SFINAE with pointer subtraction [PR78173]
Patrick Palka [Fri, 11 Dec 2020 14:40:58 +0000 (09:40 -0500)]
c++: missing SFINAE with pointer subtraction [PR78173]

This fixes a missed SFINAE when subtracting pointers to an incomplete
type.

gcc/cp/ChangeLog:

PR c++/78173
* typeck.c (pointer_diff): Use complete_type_or_maybe_complain
instead of complete_type_or_else.

gcc/testsuite/ChangeLog:

PR c++/78173
* g++.dg/cpp2a/concepts-pr78173.C: New test.

3 years agoarm: Improve documentation for effective target 'arm_softfloat'
Andrea Corallo [Tue, 1 Dec 2020 10:21:33 +0000 (11:21 +0100)]
arm: Improve documentation for effective target 'arm_softfloat'

gcc/ChangeLog

2020-12-01  Andrea Corallo  <andrea.corallo@arm.com>

* doc/sourcebuild.texi (arm_softfloat): Improve documentation.

gcc/testsuite/ChangeLog

2020-12-01  Andrea Corallo  <andrea.corallo@arm.com>

* lib/target-supports.exp (check_effective_target_arm_softfloat):
Improve documentation.

3 years agoarm: [testsuite] fix lob tests for -mfloat-abi=hard
Andrea Corallo [Thu, 26 Nov 2020 11:33:18 +0000 (12:33 +0100)]
arm: [testsuite] fix lob tests for -mfloat-abi=hard

2020-11-26  Andrea Corallo  <andrea.corallo@arm.com>

* gcc.target/arm/lob2.c: Use '-march=armv8.1-m.main+fp'.
* gcc.target/arm/lob3.c: Skip with '-mfloat-abi=hard'.
* gcc.target/arm/lob4.c: Likewise.
* gcc.target/arm/lob5.c: Use '-march=armv8.1-m.main+fp'.

3 years agotestsuite/98244 - amend gcc.dg/vect/vect-live-6.c
Richard Biener [Fri, 11 Dec 2020 12:45:55 +0000 (13:45 +0100)]
testsuite/98244 - amend gcc.dg/vect/vect-live-6.c

Committed.

2020-12-11  Richard Biener  <rguenther@suse.de>

PR testsuite/98244
* gcc.dg/vect/vect-live-6.c: Require vect_condition.

3 years agotestsuite/98242 - amend gcc.dg/vect/bb-slp-subgroups-3.c
Richard Biener [Fri, 11 Dec 2020 12:31:24 +0000 (13:31 +0100)]
testsuite/98242 - amend gcc.dg/vect/bb-slp-subgroups-3.c

Committed.

2020-12-11  Richard Biener  <rguenther@suse.de>

PR testsuite/98242
* gcc.dg/vect/bb-slp-subgroups-3.c: Require vect_int_mult.

3 years agotestsuite/98240 - amend gcc.dg/vect/pr97678.c
Richard Biener [Fri, 11 Dec 2020 12:23:21 +0000 (13:23 +0100)]
testsuite/98240 - amend gcc.dg/vect/pr97678.c

Committed.

2020-12-11  Richard Biener  <rguenther@suse.de>

PR testsuite/98240
* gcc.dg/vect/pr97678.c: Require vect_int_mult and
vect_pack_trunc.

3 years agotestsuite/98239 - require vect_condition for gcc.dg/vect/bb-slp-69.c
Richard Biener [Fri, 11 Dec 2020 12:13:28 +0000 (13:13 +0100)]
testsuite/98239 - require vect_condition for gcc.dg/vect/bb-slp-69.c

Committed.

2020-12-11  Richard Biener  <rguenther@suse.de>

PR testsuite/98239
* gcc.dg/vect/bb-slp-69.c: Require vect_condition.

3 years agoexpand: Fix up expand_doubleword_mod on 32-bit targets [PR98229]
Jakub Jelinek [Fri, 11 Dec 2020 11:47:52 +0000 (12:47 +0100)]
expand: Fix up expand_doubleword_mod on 32-bit targets [PR98229]

As the testcase shows, for 32-bit word size we can end up with op1
up to 0xffffffff (0x100000000 % 0xffffffff == 1 and so we use bit == 32
for that), but the CONST_INT we got from caller is for DImode in that case
and not valid for SImode operations.

The following patch canonicalizes the two spots where the constant needs
canonicalization.

2020-12-10  Jakub Jelinek  <jakub@redhat.com>

PR rtl-optimization/98229
* optabs.c (expand_doubleword_mod): Canonicalize op1 and
1 - INTVAL (op1) as word_mode constants when used in
word_mode arithmetics.

* gcc.c-torture/compile/pr98229.c: New test.

3 years agotree-optimization/98235 - limit SLP discovery
Richard Biener [Fri, 11 Dec 2020 09:52:58 +0000 (10:52 +0100)]
tree-optimization/98235 - limit SLP discovery

With following backedges and the SLP discovery cache not being
permute aware we have to put some discovery limits in place again.
That's also the opportunity to ditch the separate limit on the
number of permutes we try, so the patch limits the overall work
done (as in vect_build_slp_tree cache misses) to what we compute
as max_tree_size which is based on the number of scalar stmts in
the vectorized region.

Note the limit is global and there's no attempt to divide the
allowed work evenly amongst opportunities, so one degenerate
can eat it all up.  That's probably only relevant for BB
vectorization where the limit is based on up to the size of the
whole function.

2020-12-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/98235
* tree-vect-slp.c (vect_build_slp_tree): Exchange npermutes
for limit.  Decrement that for each cache miss and fail
discovery when it reaches zero.
(vect_build_slp_tree_2): Remove npermutes handling and
simply pass down limit.
(vect_build_slp_instance): Use pass down limit.
(vect_analyze_slp_instance): Likewise.
(vect_analyze_slp): Base the SLP discovery limit on
max_tree_size and pass it down.

* gcc.dg/torture/pr98235.c: New testcase.

3 years agoexpansion: Sign or zero extend on MEM_REF stores into SUBREG with SUBREG_PROMOTED_VAR...
Jakub Jelinek [Fri, 11 Dec 2020 10:10:17 +0000 (11:10 +0100)]
expansion: Sign or zero extend on MEM_REF stores into SUBREG with SUBREG_PROMOTED_VAR_P [PR98190]

Some targets decide to promote certain scalar variables to wider mode,
so their DECL_RTL is a SUBREG with SUBREG_PROMOTED_VAR_P.
When storing to such vars, store_expr takes care of sign or zero extending,
but if we store e.g. through MEM_REF into them, no sign or zero extension
happens and that leads to wrong-code e.g. on the following testcase on
aarch64-linux.

The following patch uses store_expr if we overwrite all the bits and it is
not reversed storage order, i.e. something that store_expr handles normally,
and otherwise (if the most significant bit is (or for pdp11 might be, but
pdp11 doesn't promote) being modified), the code extends manually.

2020-12-11  Jakub Jelinek  <jakub@redhat.com>

PR middle-end/98190
* expr.c (expand_assignment): If to_rtx is a promoted SUBREG,
ensure sign or zero extension either through use of store_expr
or by extending manually.

* gcc.dg/pr98190.c: New test.

3 years agoira.c: Fix ICE in ira-color [PR97092]
Andrea Corallo [Wed, 9 Dec 2020 16:59:12 +0000 (17:59 +0100)]
ira.c: Fix ICE in ira-color [PR97092]

gcc/ChangeLog

2020-12-10  Andrea Corallo  <andrea.corallo@arm.com>

PR rtl-optimization/97092
* ira-color.c (update_costs_from_allocno): Do not carry over mode
between subsequent iterations.

gcc/testsuite/ChangeLog

2020-12-10  Andrea Corallo  <andrea.corallo@arm.com>

* gcc.target/aarch64/sve/pr97092.c: New test.

3 years agotree-optimization/95582 - fix vector pattern with bool conversions
Richard Biener [Fri, 11 Dec 2020 09:07:10 +0000 (10:07 +0100)]
tree-optimization/95582 - fix vector pattern with bool conversions

The pattern recognizer fends off against recognizing conversions
from VECT_SCALAR_BOOLEAN_TYPE_P to precision one types but what
it really needs to fend off is conversions between
VECT_SCALAR_BOOLEAN_TYPE_P types - the Ada FE uses an 8 bit
boolean type that satisfies this predicate.

2020-12-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/95582
* tree-vect-patterns.c (vect_recog_bool_pattern): Check
for VECT_SCALAR_BOOLEAN_TYPE_P, not just precision one.

3 years agoFix feature check for HRESET/AVX_VNNI/UINTR
Hongyu [Wed, 9 Dec 2020 19:18:41 +0000 (19:18 +0000)]
Fix feature check for HRESET/AVX_VNNI/UINTR

gcc/ChangeLog:
* common/config/i386/cpuinfo.h (get_available_features):
Move check for HRESET/AVX_VNNI/UINTR out of avx512_usable.

3 years agodojump: Fix up probabilities splitting in dojump.c comparison splitting [PR98212]
Jakub Jelinek [Thu, 10 Dec 2020 23:36:21 +0000 (00:36 +0100)]
dojump: Fix up probabilities splitting in dojump.c comparison splitting [PR98212]

When compiling:
void foo (void);
void bar (float a, float b) { if (__builtin_expect (a != b, 1)) foo (); }
void baz (float a, float b) { if (__builtin_expect (a == b, 1)) foo (); }
void qux (float a, float b) { if (__builtin_expect (a != b, 0)) foo (); }
void corge (float a, float b) { if (__builtin_expect (a == b, 0)) foo (); }
on x86_64, we get (unimportant cruft removed):
bar:    ucomiss %xmm1, %xmm0
        jp      .L4
        je      .L1
.L4:    jmp     foo
.L1:    ret
baz:    ucomiss %xmm1, %xmm0
        jp      .L6
        jne     .L6
        jmp     foo
.L6:    ret
qux:    ucomiss %xmm1, %xmm0
        jp      .L13
        jne     .L13
        ret
.L13:   jmp     foo
corge:  ucomiss %xmm1, %xmm0
        jnp     .L18
.L14:   ret
.L18:   jne     .L14
        jmp     foo
(note for bar and qux that changed with a patch I've posted earlier today).
This is all reasonable, except the last function, the overall jump to
the tail call is predicted unlikely (10%), so it is good jmp foo isn't on
the straight line path, but NaNs are (or should be) considered very unlikely
in the programs, so IMHO the right code (and one emitted with the following
patch) is:
corge:  ucomiss %xmm1, %xmm0
        jp      .L14
        je      .L18
.L14:   ret
.L18:   jmp     foo

Let's discuss the probabilities in the above testcase:
for !and_them it looks all correct, so for
bar we split
if (a != b) goto t; // prob 90%
goto f;
into:
if (a unord b) goto t; // first_prob = prob * cprob = 90% * 1% = 0.9%
if (a ltgt b) goto t; // adjusted prob = (prob - first_prob) / (1 - first_prob) = (90% - 0.9%) / (1 - 0.9%) = 89.909%
and for qux we split
if (a != b) goto t; // prob 10%
goto f;
into:
if (a unord b) goto t; // first_prob = prob * cprob = 10% * 1% = 0.1%
if (a ltgt b) goto t; // adjusted prob = (prob - first_prob) / (1 - first_prob) = (10% - 0.1%) / (1 - 0.1%) = 9.910%
Now, the and_them cases should be probability wise exactly the same
if we swap the f and t labels, because baz
if (a == b) goto t; // prob 90%
goto f;
is equivalent to:
if (a != b) goto f; // prob 10%
goto t;
which is in qux.  This means we could expand baz as:
if (a unord b) goto f; // 0.1%
if (a ltgt b) goto f; // 9.910%
goto t;
But we don't expand it exactly that way, but instead (as the comment says)
as:
if (a ord b) ; else goto f; // first_prob as probability of ;
if (a uneq b) goto t; // adjusted prob
goto f;
So, first_prob.invert () should be 0.1% and adjusted prob should be
1 - 9.910%.
Thus, the right thing is 4 inverts:
prob = prob.invert (); // baz is equivalent to qux with swap(t, f) and thus inverted original prob
first_prob = prob.split (cprob.invert ()).invert ();
// cprob.invert because by doing if (cond) ; else goto f; we effectively invert the condition
// the second invert because first_prob is probability of ; rather than goto f
prob = prob.invert (); // lastly because adjusted prob we want is
// probability of goto t;, while the one from corresponding !and_them case
// would be if (...) goto f; goto t;

2020-12-11  Jakub Jelinek  <jakub@redhat.com>

PR rtl-optimization/98212
* dojump.c (do_compare_rtx_and_jump): Change computation of
first_prob for and_them.  Add comment explaining and_them case.

* gcc.dg/predict-8.c: Adjust expected probability.

3 years agolibstdc++: Remove redundant branches in countl_one and countr_one [PR 98226]
Jonathan Wakely [Thu, 10 Dec 2020 21:57:42 +0000 (21:57 +0000)]
libstdc++: Remove redundant branches in countl_one and countr_one [PR 98226]

There's no need to explicitly check for the maximum value, because the
function we call handles it correctly anyway.

libstdc++-v3/ChangeLog:

PR libstdc++/98226
* include/std/bit (__countl_one, __countr_one): Remove redundant
branches.

3 years agoReduce memory requirements for ranger
Andrew MacLeod [Thu, 10 Dec 2020 19:59:14 +0000 (14:59 -0500)]
Reduce memory requirements for ranger

Calculate block exit info upfront, and then any SSA_NAME which is never
used in an outgoing range calculation is a pure global and can bypass the
on-entry cache.

PR tree-optimization/98174
* gimple-range-cache.cc (ranger_cache::ssa_range_in_bb): Only push
poor values to be examined if it isn't a pure global.
(ranger_cache::block_range): Don't process pure globals.
(ranger_cache::fill_block_cache): Adjust has_edge_range call.
* gimple-range-gori.cc (gori_map::all_outgoing): New bitmap.
(gori_map::gori_map): Allocate all_outgoing.
(gori_map::is_export_p): No specified BB returns global context.
(gori_map::calculate_gori): Accumulate each block into global.
(gori_compute::gori_compute): Preprocess each block for exports.
(gori_compute::has_edge_range_p): No edge returns global context.
* gimple-range-gori.h (has_edge_range_p): Provide default parameter.