]> gcc.gnu.org Git - gcc.git/log
gcc.git
12 months agoRefine maskloadmn pattern with UNSPEC_MASKLOAD.
liuhongt [Tue, 20 Jun 2023 07:41:00 +0000 (15:41 +0800)]
Refine maskloadmn pattern with UNSPEC_MASKLOAD.

If mem_addr points to a memory region with less than whole vector size
bytes of accessible memory and k is a mask that would prevent reading
the inaccessible bytes from mem_addr, add UNSPEC_MASKLOAD to prevent
it to be transformed to vpblendd.

gcc/ChangeLog:

PR target/110309
* config/i386/sse.md (maskload<mode><avx512fmaskmodelower>):
Refine pattern with UNSPEC_MASKLOAD.
(maskload<mode><avx512fmaskmodelower>): Ditto.
(*<avx512>_load<mode>_mask): Extend mode iterator to
VI12HFBF_AVX512VL.
(*<avx512>_load<mode>): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/pr110309.c: New test.

13 months agoSSA ALIAS: Apply LEN_MASK_STORE to 'ref_maybe_used_by_call_p_1'
Ju-Zhe Zhong [Fri, 23 Jun 2023 14:15:39 +0000 (22:15 +0800)]
SSA ALIAS: Apply LEN_MASK_STORE to 'ref_maybe_used_by_call_p_1'

gcc/ChangeLog:

* tree-ssa-alias.cc (call_may_clobber_ref_p_1): Add LEN_MASK_STORE.

13 months agoSSA ALIAS: Apply LEN_MASK_{LOAD, STORE} into SSA alias analysis
Ju-Zhe Zhong [Fri, 23 Jun 2023 13:56:35 +0000 (21:56 +0800)]
SSA ALIAS: Apply LEN_MASK_{LOAD, STORE} into SSA alias analysis

gcc/ChangeLog:

* tree-ssa-alias.cc (ref_maybe_used_by_call_p_1): Apply
LEN_MASK_{LOAD,STORE}

13 months agoRISC-V:Add float16 tuple type abi
yulong [Wed, 21 Jun 2023 07:39:55 +0000 (15:39 +0800)]
RISC-V:Add float16 tuple type abi

gcc/ChangeLog:

* config/riscv/vector.md: Add float16 attr at sew、vlmul and ratio.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/base/abi-10.c: Add float16 tuple type case.
* gcc.target/riscv/rvv/base/abi-11.c: Ditto.
* gcc.target/riscv/rvv/base/abi-12.c: Ditto.
* gcc.target/riscv/rvv/base/abi-15.c: Ditto.
* gcc.target/riscv/rvv/base/abi-8.c: Ditto.
* gcc.target/riscv/rvv/base/abi-9.c: Ditto.
* gcc.target/riscv/rvv/base/abi-17.c: New test.
* gcc.target/riscv/rvv/base/abi-18.c: New test.

13 months agoDaily bump.
GCC Administrator [Sun, 25 Jun 2023 00:16:41 +0000 (00:16 +0000)]
Daily bump.

13 months agoi386: Add alternate representation for {and,or,xor}b %ah,%dh.
Roger Sayle [Sat, 24 Jun 2023 22:05:25 +0000 (23:05 +0100)]
i386: Add alternate representation for {and,or,xor}b %ah,%dh.

A patch that I'm working on to improve RTL simplifications in the
middle-end results in the regression of pr78904-1b.c, due to changes in
the canonical representation of high-byte (%ah, %bh, %ch, %dh) logic.
See also PR target/78904.

This patch avoids/prevents those failures by adding support for the
alternate representation, duplicating the existing *<code>qi_ext<mode>_2
as *<code>qi_ext<mode>_3 (the new version also replacing any_or with
any_logic to provide *andqi_ext<mode>_3 in the same pattern).  Removing
the original pattern isn't trivial, as it's generated by define_split,
but this can be investigated after the other pieces are approved.

The current representation of this instruction is:

(set (zero_extract:DI (reg/v:DI 87 [ aD.2763 ])
        (const_int 8 [0x8])
        (const_int 8 [0x8]))
    (subreg:DI (xor:QI (subreg:QI (zero_extract:DI (reg:DI 94)
                    (const_int 8 [0x8])
                    (const_int 8 [0x8])) 0)
            (subreg:QI (zero_extract:DI (reg/v:DI 87 [ aD.2763 ])
                    (const_int 8 [0x8])
                    (const_int 8 [0x8])) 0)) 0))

after my proposed middle-end improvement, we attempt to recognize:

(set (zero_extract:DI (reg/v:DI 87 [ aD.2763 ])
        (const_int 8 [0x8])
        (const_int 8 [0x8]))
    (zero_extract:DI (xor:DI (reg:DI 94)
            (reg/v:DI 87 [ aD.2763 ]))
        (const_int 8 [0x8])
        (const_int 8 [0x8])))

2023-06-24  Roger Sayle  <roger@nextmovesoftware.com>

gcc/ChangeLog
* config/i386/i386.md (*<code>qi_ext<mode>_3): New define_insn.

13 months agoFortran: ABI for scalar CHARACTER(LEN=1),VALUE dummy argument [PR110360]
Harald Anlauf [Sat, 24 Jun 2023 18:36:53 +0000 (20:36 +0200)]
Fortran: ABI for scalar CHARACTER(LEN=1),VALUE dummy argument [PR110360]

gcc/fortran/ChangeLog:

PR fortran/110360
* trans-expr.cc (gfc_conv_procedure_call): Truncate constant string
argument of length > 1 passed to scalar CHARACTER(1),VALUE dummy.

13 months agoRISC-V: Refactor the integer ternary autovec pattern
Juzhe-Zhong [Wed, 21 Jun 2023 22:38:42 +0000 (06:38 +0800)]
RISC-V: Refactor the integer ternary autovec pattern

Long time ago, I encounter ICE when trying to set clobber register as Pmode
and I forgot the reason.

So, I clobber SI scratch and PUT_MODE to make it Pmode after reload which
makes patterns look unreasonable.

According to Jeff's comments, I tried it again, it works now when we try to
set clobber register as Pmode and the patterns look more reasonable now.

The tests are all passed, Ok for trunk.

gcc/ChangeLog:

* config/riscv/autovec.md (*fma<mode>): set clobber to Pmode in expand stage.
(*fma<VI:mode><P:mode>): Ditto.
(*fnma<mode>): Ditto.
(*fnma<VI:mode><P:mode>): Ditto.

13 months agoRISC-V: Support RVV floating-point auto-vectorization
Juzhe-Zhong [Wed, 21 Jun 2023 15:53:14 +0000 (23:53 +0800)]
RISC-V: Support RVV floating-point auto-vectorization

This patch adds RVV floating-point auto-vectorization.
Also, fix attribute bug of floating-point ternary operations in vector.md.

gcc/ChangeLog:

* config/riscv/autovec.md (fma<mode>4): New pattern.
(*fma<mode>): Ditto.
(fnma<mode>4): Ditto.
(*fnma<mode>): Ditto.
(fms<mode>4): Ditto.
(*fms<mode>): Ditto.
(fnms<mode>4): Ditto.
(*fnms<mode>): Ditto.
* config/riscv/riscv-protos.h (emit_vlmax_fp_ternary_insn):
New function.
* config/riscv/riscv-v.cc (emit_vlmax_fp_ternary_insn): Ditto.
* config/riscv/vector.md: Fix attribute bug.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/autovec/ternop/ternop-1.c: Adjust tests.
* gcc.target/riscv/rvv/autovec/ternop/ternop-2.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop-3.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop-4.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop-5.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop-6.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-1.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-2.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-3.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-4.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-5.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-6.c: Ditto.
* gcc.target/riscv/rvv/autovec/ternop/ternop-10.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop-11.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop-12.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop-7.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop-8.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop-9.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-10.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-11.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-12.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-7.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-8.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run-9.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-1.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-10.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-11.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-12.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-2.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-3.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-4.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-5.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-6.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-7.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-8.c: New test.
* gcc.target/riscv/rvv/autovec/ternop/ternop_run_zvfh-9.c: New test.

13 months agoLOOP IVOPTS: Apply LEN_MASK_{LOAD,STORE}
Ju-Zhe Zhong [Fri, 23 Jun 2023 23:41:57 +0000 (07:41 +0800)]
LOOP IVOPTS: Apply LEN_MASK_{LOAD,STORE}

Hi, Jeff. I fix format as you suggested.
Ok for trunk ?

gcc/ChangeLog:

* tree-ssa-loop-ivopts.cc (get_mem_type_for_internal_fn):
Apply LEN_MASK_{LOAD,STORE}.

13 months agoIVOPTS: Add LEN_MASK_{LOAD, STORE} into 'get_alias_ptr_type_for_ptr_address'
Ju-Zhe Zhong [Fri, 23 Jun 2023 14:21:03 +0000 (22:21 +0800)]
IVOPTS: Add LEN_MASK_{LOAD, STORE} into 'get_alias_ptr_type_for_ptr_address'

gcc/ChangeLog:

* tree-ssa-loop-ivopts.cc (get_alias_ptr_type_for_ptr_address):
Add LEN_MASK_{LOAD,STORE}.

13 months agotext-art: remove explicit #include of C++ standard library headers
David Malcolm [Sat, 24 Jun 2023 01:18:15 +0000 (21:18 -0400)]
text-art: remove explicit #include of C++ standard library headers

gcc/analyzer/ChangeLog:
* access-diagram.cc: Add #define INCLUDE_VECTOR.
* bounds-checking.cc: Likewise.

gcc/ChangeLog:
* diagnostic-format-sarif.cc: Add #define INCLUDE_VECTOR.
* diagnostic.cc: Likewise.
* text-art/box-drawing.cc: Likewise.
* text-art/canvas.cc: Likewise.
* text-art/ruler.cc: Likewise.
* text-art/selftests.cc: Likewise.
* text-art/selftests.h (text_art::canvas): New forward decl.
* text-art/style.cc: Add #define INCLUDE_VECTOR.
* text-art/styled-string.cc: Likewise.
* text-art/table.cc: Likewise.
* text-art/table.h: Remove #include <vector>.
* text-art/theme.cc: Add #define INCLUDE_VECTOR.
* text-art/types.h: Check that INCLUDE_VECTOR is defined.
Remove #include of <vector> and <string>.
* text-art/widget.cc: Add #define INCLUDE_VECTOR.
* text-art/widget.h: Remove #include <vector>.

gcc/testsuite/ChangeLog:
* gcc.dg/plugin/diagnostic_plugin_test_text_art.c: Add
#define INCLUDE_VECTOR.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
13 months agoVECT: Apply LEN_MASK_{LOAD,STORE} into vectorizer
Ju-Zhe Zhong [Thu, 22 Jun 2023 23:51:12 +0000 (07:51 +0800)]
VECT: Apply LEN_MASK_{LOAD,STORE} into vectorizer

Address comments from Richard and Bernhard from V5 patch.
V6 fixed all issues according their comments.

gcc/ChangeLog:

* internal-fn.cc (expand_partial_store_optab_fn): Adapt for LEN_MASK_STORE.
(internal_load_fn_p): Add LEN_MASK_LOAD.
(internal_store_fn_p): Add LEN_MASK_STORE.
(internal_fn_mask_index): Add LEN_MASK_{LOAD,STORE}.
(internal_fn_stored_value_index): Add LEN_MASK_STORE.
(internal_len_load_store_bias):  Add LEN_MASK_{LOAD,STORE}.
* optabs-tree.cc (can_vec_mask_load_store_p): Adapt for LEN_MASK_{LOAD,STORE}.
(get_len_load_store_mode): Ditto.
* optabs-tree.h (can_vec_mask_load_store_p): Ditto.
(get_len_load_store_mode): Ditto.
* tree-vect-stmts.cc (check_load_store_for_partial_vectors): Ditto.
(get_all_ones_mask): New function.
(vectorizable_store): Apply LEN_MASK_{LOAD,STORE} into vectorizer.
(vectorizable_load): Ditto.

13 months agoDaily bump.
GCC Administrator [Sat, 24 Jun 2023 00:17:01 +0000 (00:17 +0000)]
Daily bump.

13 months agocompiler, libgo: support bootstrapping gc compiler
Ian Lance Taylor [Thu, 22 Jun 2023 00:27:39 +0000 (17:27 -0700)]
compiler, libgo: support bootstrapping gc compiler

In the Go 1.21 release the package internal/profile imports
internal/lazyregexp.  That works when bootstrapping with Go 1.17,
because that compiler has internal/lazyregep and permits importing it.
We also have internal/lazyregexp in libgo, but since it is not installed
it is not available for importing.  This CL adds internal/lazyregexp
to the list of internal packages that are installed for bootstrapping.

The Go 1.21, and earlier, releases have a couple of functions in
the internal/abi package that are always fully intrinsified.
The gofrontend recognizes and intrinsifies those functions as well.
However, the gofrontend was also building function descriptors
for references to the functions without calling them, which
failed because there was nothing to refer to.  That is OK for the
gc compiler, which guarantees that the functions are only called,
not referenced.  This CL arranges to not generate function descriptors
for these functions.

For golang/go#60913

Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/504798

13 months agoc++: provide #include hint for missing includes [PR110164]
David Malcolm [Fri, 23 Jun 2023 21:56:14 +0000 (17:56 -0400)]
c++: provide #include hint for missing includes [PR110164]

PR c++/110164 notes that in cases where we have a forward decl
of a std library type such as:

std::array<int, 10> x;

we emit this diagnostic:

error: aggregate ‘std::array<int, 10> x’ has incomplete type and cannot be defined

This patch adds this hint to the diagnostic:

note: ‘std::array’ is defined in header ‘<array>’; this is probably fixable by adding ‘#include <array>’

gcc/cp/ChangeLog:
PR c++/110164
* cp-name-hint.h (maybe_suggest_missing_header): New decl.
* decl.cc: Define INCLUDE_MEMORY.  Add include of
"cp/cp-name-hint.h".
(start_decl_1): Call maybe_suggest_missing_header.
* name-lookup.cc (maybe_suggest_missing_header): Remove "static".

gcc/testsuite/ChangeLog:
PR c++/110164
* g++.dg/diagnostic/missing-header-pr110164.C: New test.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
13 months agoc++: Add support for -std={c,gnu}++2{c,6}
Marek Polacek [Thu, 22 Jun 2023 20:56:26 +0000 (16:56 -0400)]
c++: Add support for -std={c,gnu}++2{c,6}

It seems prudent to add C++26 now that the first C++26 papers have been
approved.  I followed commit r11-6920 as well as r8-3237.

Since C++23 is essentially finished and its __cplusplus value has
settled to 202302L, I've updated cpp_init_builtins and marked
-std=c++2b Undocumented and made -std=c++23 no longer Undocumented.

As for __cplusplus, I've chosen 202400L:

  $ xg++ -std=c++26 -dM -E -x c++ - < /dev/null | grep cplusplus
  #define __cplusplus 202400L

I've verified the patch with a simple test, exercising the new
directives.  Don't forget to update your GXX_TESTSUITE_STDS!

This patch does not add -Wc++26-extensions.

gcc/c-family/ChangeLog:

* c-common.h (cxx_dialect): Add cxx26 as a dialect.
* c-opts.cc (set_std_cxx26): New.
(c_common_handle_option): Set options when -std={c,gnu}++2{c,6} is
enabled.
(c_common_post_options): Adjust comments.
* c.opt: Add options for -std=c++26, std=c++2c, -std=gnu++26,
and -std=gnu++2c.
(std=c++2b): Mark as Undocumented.
(std=c++23): No longer Undocumented.

gcc/ChangeLog:

* doc/cpp.texi (__cplusplus): Document value for -std=c++26 and
-std=gnu++26.  Document that for C++23, its value is 202302L.
* doc/invoke.texi: Document -std=c++26 and -std=gnu++26.
* dwarf2out.cc (highest_c_language): Handle GNU C++26.
(gen_compile_unit_die): Likewise.

libcpp/ChangeLog:

* include/cpplib.h (c_lang): Add CXX26 and GNUCXX26.
* init.cc (lang_defaults): Add rows for CXX26 and GNUCXX26.
(cpp_init_builtins): Set __cplusplus to 202400L for C++26.
Set __cplusplus to 202302L for C++23.

gcc/testsuite/ChangeLog:

* lib/target-supports.exp (check_effective_target_c++23): Return
1 also if check_effective_target_c++26.
(check_effective_target_c++23_down): New.
(check_effective_target_c++26_only): New.
(check_effective_target_c++26): New.
* g++.dg/cpp23/cplusplus.C: Adjust expected value.
* g++.dg/cpp26/cplusplus.C: New test.

13 months agolibcpp: allow UCS_LIMIT codepoints in UTF-8 strings
Ben Boeckel [Wed, 21 Jun 2023 18:58:20 +0000 (14:58 -0400)]
libcpp: allow UCS_LIMIT codepoints in UTF-8 strings

Fixes r14-1954 (libcpp: reject codepoints above 0x10FFFF, 2023-06-06)

libcpp/

* charset.cc: Allow `UCS_LIMIT` in UTF-8 strings.

Reported-by: Damien Guibouret <damien.guibouret@partition-saving.com>
Signed-off-by: Ben Boeckel <ben.boeckel@kitware.com>
13 months agolibstdc++: Use RAII in std::vector::_M_realloc_insert
Jonathan Wakely [Tue, 20 Jun 2023 12:39:29 +0000 (13:39 +0100)]
libstdc++: Use RAII in std::vector::_M_realloc_insert

Replace the try-block with RAII types for deallocating storage and
destroying elements.

libstdc++-v3/ChangeLog:

* include/bits/vector.tcc (_M_realloc_insert): Replace try-block
with RAII types.

13 months agoTiny phiprop compile time optimization
Jan Hubicka [Fri, 23 Jun 2023 16:45:42 +0000 (18:45 +0200)]
Tiny phiprop compile time optimization

gcc/ChangeLog:

* tree-ssa-phiprop.cc (propagate_with_phi): Compute post dominators on
demand.
(pass_phiprop::execute): Do not compute it here; return
update_ssa_only_virtuals if something changed.
(pass_data_phiprop): Remove TODO_update_ssa from todos.

13 months agoFortran: ABI for scalar CHARACTER(LEN=1),VALUE dummy argument [PR110360]
Harald Anlauf [Thu, 22 Jun 2023 20:07:41 +0000 (22:07 +0200)]
Fortran: ABI for scalar CHARACTER(LEN=1),VALUE dummy argument [PR110360]

gcc/fortran/ChangeLog:

PR fortran/110360
* trans-expr.cc (gfc_conv_procedure_call): Pass actual argument
to scalar CHARACTER(1),VALUE dummy argument by value.

gcc/testsuite/ChangeLog:

PR fortran/110360
* gfortran.dg/value_9.f90: New test.

13 months agoFix power10 fusion bug with prefixed loads, PR target/105325
Michael Meissner [Fri, 23 Jun 2023 15:32:39 +0000 (11:32 -0400)]
Fix power10 fusion bug with prefixed loads, PR target/105325

This changes fixes PR target/105325.  PR target/105325 is a bug where an
invalid lwa instruction is generated due to power10 fusion of a load
instruction to a GPR and an compare immediate instruction with the immediate
being -1, 0, or 1.

In some cases, when the load instruction is done, the GCC compiler would
generate a load instruction with an offset that was too large to fit into the
normal load instruction.

In particular, loads from the stack might originally have a small offset, so
that the load is not a prefixed load.  However, after the stack is set up, and
register allocation has been done, the offset now is large enough that we would
have to use a prefixed load instruction.

The support for prefixed loads did not consider that patterns with a fused load
and compare might have a prefixed address.  Without this support, the proper
prefixed load won't be generated.

In the original code, when the split2 pass is run after reload has finished the
ds_form_mem_operand predicate that was used for lwa and ld no longer returns
true.  When the pattern was created, ds_form_mem_operand recognized the insn as
being valid since the offset was small.  But after register allocation,
ds_form_mem_operand did not return true.  Because it didn't return true, the
insn could not be split.  Since the insn was not split and the prefix support
did not indicate a prefixed instruction was used, the wrong load is generated.

The solution involves:

    1) Don't use ds_form_mem_operand for ld and lwa, always use
non_update_memory_operand.

    2) Delete ds_form_mem_operand since it is no longer used.

    3) Use the "YZ" constraints for ld/lwa instead of "m".

    4) If we don't need to sign extend the lwa, convert it to lwz, and use
cmpwi instead of cmpdi.  Adjust the insn name to reflect the code
generate.

    5) Insure that the insn using lwa will be recognized as having a prefixed
operand (and hence the insn length will be 16 bytes instead of 8
bytes).

5a) Set the prefixed and maybe_prefix attributes to know that
    fused_load_cmpi are also load insns;

5b) In the case where we are just setting CC and not using the memory
    afterward, set the clobber to use a DI register, and put an
    explicit sign_extend operation in the split;

5c) Set the sign_extend attribute to "yes" for lwa.

5d) 5a-5c are the things that prefixed_load_p in rs6000.cc checks to
    ensure that lwa is treated as a ds-form instruction and not as
    a d-form instruction (i.e. lwz).

    6) Add a new test case for this case.

    7) Adjust the insn counts in fusion-p10-ldcmpi.c.  Because we are no
longer using ds_form_mem_operand, the ld and lwa instructions will fuse
x-form (reg+reg) addresses in addition ds-form (reg+offset or reg).

2023-06-23   Michael Meissner  <meissner@linux.ibm.com>

gcc/

PR target/105325
* config/rs6000/genfusion.pl (gen_ld_cmpi_p10_one): Fix problems that
allowed prefixed lwa to be generated.
* config/rs6000/fusion.md: Regenerate.
* config/rs6000/predicates.md (ds_form_mem_operand): Delete.
* config/rs6000/rs6000.md (prefixed attribute): Add support for load
plus compare immediate fused insns.
(maybe_prefixed): Likewise.

gcc/testsuite/

PR target/105325
* g++.target/powerpc/pr105325.C: New test.
* gcc.target/powerpc/fusion-p10-ldcmpi.c: Update insn counts.

Co-Authored-By: Aaron Sawdey <acsawdey@linux.ibm.com>
13 months agotestsuite,objective-c++: Fix imported NSObjCRuntime.h.
Iain Sandoe [Fri, 23 Jun 2023 14:29:14 +0000 (15:29 +0100)]
testsuite,objective-c++: Fix imported NSObjCRuntime.h.

We have imported some headers from the GNUStep project to allow us
to maintain the testsuite independent to changing versions of system
headers.

One of these headers has a macro that (now we have support for
__has_feature) expands to a declaration that triggers a warning.

These headers are considered part of the implementation so that, in
this case, we can suppress the warning with the system_header pragma.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
gcc/testsuite/ChangeLog:

* objc-obj-c++-shared/GNUStep/Foundation/NSObjCRuntime.h: Make
this header use pragma system_header.

13 months agoImproved SUBREG simplifications in simplify-rtx.cc's simplify_subreg.
Roger Sayle [Fri, 23 Jun 2023 14:23:20 +0000 (15:23 +0100)]
Improved SUBREG simplifications in simplify-rtx.cc's simplify_subreg.

An x86 backend improvement that I'm working results in combine attempting
to recognize:

(set (reg:DI 87 [ xD.2846 ])
     (ior:DI (subreg:DI (ashift:TI (zero_extend:TI (reg:DI 92))
                                   (const_int 64 [0x40])) 0)
             (reg:DI 91)))

where the lowpart SUBREG has difficulty seeing through the (hi<<64)
that the lowpart must be zero.  Rather than workaround this in the
backend, the better fix is to teach simplify-rtx that
lowpart((hi<<64)|lo) -> lo and highpart((hi<<64)|lo) -> hi, so that
all backends benefit.  Reducing the number of places where the
middle-end generates a SUBREG of something other than REG is a
good thing.

On x86_64-pc-linux-gnu, the testcase pr78904-1b.c FAILs with this patch,
due to changes in expected/canonical RTL, for which a backend patch to
i386.md has already been provisionally approved.

2023-06-23  Roger Sayle  <roger@nextmovesoftware.com>

gcc/ChangeLog
* simplify-rtx.cc (simplify_subreg):  Optimize lowpart SUBREGs
of ASHIFT to const0_rtx with sufficiently large shift count.
Optimize highpart SUBREGs of ASHIFT as the shift operand when
the shift count is the correct offset.  Optimize SUBREGs of
multi-word logic operations if the SUBREGs of both operands
can be simplified.

13 months agoFix initializer_constant_valid_p_1 TYPE_PRECISION use
Richard Biener [Fri, 23 Jun 2023 10:50:50 +0000 (12:50 +0200)]
Fix initializer_constant_valid_p_1 TYPE_PRECISION use

initializer_constant_valid_p_1 is letting through all conversions
of float vector types that have the same number of elements but
that's of course not valid.  The following restricts the code
to scalar floating point types as was probably intended (only
scalar integer types are handled as well).

* varasm.cc (initializer_constant_valid_p_1): Only
allow conversions between scalar floating point types.

13 months agoDeal with vector typed operands in conversions
Richard Biener [Fri, 23 Jun 2023 10:48:36 +0000 (12:48 +0200)]
Deal with vector typed operands in conversions

The following avoids using TYPE_PRECISION on VECTOR_TYPE when
looking for bit-precision changes in vectorizable_assignment.
We didn't anticipate a stmt like

  _21 = VIEW_CONVERT_EXPR<unsigned int>(vect__1.7_28);

and the following makes sure to handle that.

* tree-vect-stmts.cc (vectorizable_assignment):
Properly handle non-integral operands when analyzing
conversions.

13 months ago[aarch64/match.pd] Fix ICE observed in PR110280.
Prathamesh Kulkarni [Fri, 23 Jun 2023 09:57:17 +0000 (15:27 +0530)]
[aarch64/match.pd] Fix ICE observed in PR110280.

gcc/ChangeLog:
PR tree-optimization/110280
* match.pd (vec_perm_expr(v, v, mask) -> v): Explicitly build vector
using build_vector_from_val with the element of input operand, and
mask's type if operand and mask's types don't match.

gcc/testsuite/ChangeLog:
PR tree-optimization/110280
* gcc.target/aarch64/sve/pr110280.c: New test.

13 months agoFix tree_simple_nonnegative_warnv_p for VECTOR_TYPEs
Richard Biener [Fri, 23 Jun 2023 08:12:24 +0000 (10:12 +0200)]
Fix tree_simple_nonnegative_warnv_p for VECTOR_TYPEs

tree_simple_nonnegative_warnv_p ends up being called on VECTOR_TYPEs
which I think even gets the wrong answer here for tcc_comparison
since vector bools are signed.  The following properly guards
that with !VECTOR_TYPE_P.

* fold-const.cc (tree_simple_nonnegative_warnv_p): Guard
the truth_value_p case with !VECTOR_TYPE_P.

13 months agoProperly guard vect_look_through_possible_promotion
Richard Biener [Fri, 23 Jun 2023 08:08:50 +0000 (10:08 +0200)]
Properly guard vect_look_through_possible_promotion

The function ends up getting called on VECTOR_TYPEs which it
really isn't prepared for and with the TYPE_PRECISION checking
changes will ICE.  The following exits early when the type
to work on isn't scalar integral.

* tree-vect-patterns.cc (vect_look_through_possible_promotion):
Exit early when the type isn't scalar integral.

13 months agoUse element_precision for match.pd arith conversion optimization
Richard Biener [Fri, 23 Jun 2023 08:20:45 +0000 (10:20 +0200)]
Use element_precision for match.pd arith conversion optimization

The simplification (outertype)((innertype0)a+(innertype1)b) to
((newtype)a+(newtype)b) ends up using TYPE_PRECISION to check
whether it can elide a conversion but in some paths there can
be VECTOR_TYPEs where this instead compares the number of lanes.
The following fixes the missed optimizations and uses
element_precision in those places.

* match.pd ((outertype)((innertype0)a+(innertype1)b)
-> ((newtype)a+(newtype)b)): Use element_precision
where appropriate.

13 months agoBogus and missed folding on vector compares
Richard Biener [Fri, 23 Jun 2023 08:15:27 +0000 (10:15 +0200)]
Bogus and missed folding on vector compares

fold_binary tries to transform (double)float1 CMP (double)float2
into float1 CMP float2 but ends up using TYPE_PRECISION on the
argument types.  For vector types that compares the number of
lanes which should be always equal (so it's harmless as to
not generating wrong code).  The following instead properly
uses element_precision.

The same happens in the corresponding match.pd pattern.

* fold-const.cc (fold_binary_loc): Use element_precision
when trying (double)float1 CMP (double)float2 to
float1 CMP float2 simplification.
* match.pd: Likewise.

13 months agoOptimize vector codegen for invariant loads, fix SLP support
Richard Biener [Fri, 23 Jun 2023 07:28:34 +0000 (09:28 +0200)]
Optimize vector codegen for invariant loads, fix SLP support

The following avoids creating duplicate stmts for invariant loads
which was necessary when the vector stmts were in a linked list.
It also fixes SLP support which didn't correctly create the
appropriate number of copies.

* tree-vect-stmts.cc (vectorizable_load): Avoid useless
copies of VMAT_INVARIANT vectorized stmts, fix SLP support.

13 months agoImprove vector_vector_composition_type
Richard Biener [Thu, 22 Jun 2023 13:14:51 +0000 (15:14 +0200)]
Improve vector_vector_composition_type

We sometimes get to ask to decompose, say V2DFmode into two halves.
Currently this results in composing it from two DImode pieces
instead of the obvious two DFmode pieces.  The following adjusts
vector_vector_composition_type for this trivial case and avoids
a VIEW_CONVERT_EXPR in the initial code generation.

* tree-vect-stmts.cc (vector_vector_composition_type):
Handle composition of a vector from a number of elements that
happens to match its number of lanes.

13 months agoDaily bump.
GCC Administrator [Fri, 23 Jun 2023 00:16:38 +0000 (00:16 +0000)]
Daily bump.

13 months agorust: Update usage of TARGET_AIX to TARGET_AIX_OS
Paul E. Murphy [Thu, 22 Jun 2023 22:57:23 +0000 (17:57 -0500)]
rust: Update usage of TARGET_AIX to TARGET_AIX_OS

This was noticed when fixing the gccgo usage of the macro, the
rust usage is very similar.

TARGET_AIX is defined as a non-zero value on linux/powerpc64le
which may cause unexpected behavior.  TARGET_AIX_OS should be
used to toggle AIX specific behavior.

2023-06-22  Paul E. Murphy  <murphyp@linux.ibm.com>

gcc/rust/
* rust-object-export.cc [TARGET_AIX]: Rename and update usage to
TARGET_AIX_OS.

13 months agogo: Update usage of TARGET_AIX to TARGET_AIX_OS
Paul E. Murphy [Thu, 22 Jun 2023 22:53:46 +0000 (17:53 -0500)]
go: Update usage of TARGET_AIX to TARGET_AIX_OS

TARGET_AIX is defined to a non-zero value on linux and maybe other
powerpc64le targets.  This leads to unexpected behavior such as
dropping the .go_export section when linking a shared library
on linux/powerpc64le.

Instead, use TARGET_AIX_OS to toggle AIX specific behavior.

Fixes golang/go#60798.

2023-06-22  Paul E. Murphy  <murphyp@linux.ibm.com>

gcc/go/
* go-backend.cc [TARGET_AIX]: Rename and update usage to TARGET_AIX_OS.
* go-lang.cc: Likewise.

13 months agoconfigure: Implement --enable-host-bind-now
Marek Polacek [Thu, 22 Jun 2023 15:30:01 +0000 (11:30 -0400)]
configure: Implement --enable-host-bind-now

As promised in the --enable-host-pie patch, this patch adds another
configure option, --enable-host-bind-now, which adds -z now when linking
the compiler executables in order to extend hardening.  BIND_NOW with RELRO
allows the GOT to be marked RO; this prevents GOT modification attacks.

This option does not affect linking of target libraries; you can use
LDFLAGS_FOR_TARGET=-Wl,-z,relro,-z,now to enable RELRO/BIND_NOW.

With this patch:
$ readelf -Wd cc1{,plus,obj,gm2} f951 lto1 cpp  rust1 gnat1 | grep FLAGS
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE
 0x000000000000001e (FLAGS)              BIND_NOW
 0x000000006ffffffb (FLAGS_1)            Flags: NOW PIE

c++tools/ChangeLog:

* configure.ac (--enable-host-bind-now): New check.
* configure: Regenerate.

gcc/ChangeLog:

* configure.ac (--enable-host-bind-now): New check.  Add
-Wl,-z,now to LD_PICFLAG if --enable-host-bind-now.
* configure: Regenerate.
* doc/install.texi: Document --enable-host-bind-now.

lto-plugin/ChangeLog:

* configure.ac (--enable-host-bind-now): New check.  Link with
-z,now.
* configure: Regenerate.

13 months agoChange fma_reassoc_width tuning for ampere1
Di Zhao OS [Thu, 22 Jun 2023 14:16:57 +0000 (15:16 +0100)]
Change fma_reassoc_width tuning for ampere1

This patch enables reassociation of floating-point additions on ampere1.
This brings about 1% overall benefit on spec2017 fprate cases. (There
are minor regressions in 510.parest_r and 508.namd_r, analyzed here:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110279 .)

gcc/ChangeLog:

* config/aarch64/aarch64.cc: Change fma_reassoc_width for ampere1.

13 months agolibgomp.texi: Improve OpenMP ICV description
Tobias Burnus [Thu, 22 Jun 2023 12:57:54 +0000 (14:57 +0200)]
libgomp.texi: Improve OpenMP ICV description

Use @var{} instead of @emph{} - for semantic texinfo formatting; the result
is similar: slanted instead of italic in PDF, still italic in HTML, albeit
in info is is now uppercase instead of '_' as pre/suffix.

The patch also documents the newer _ALL/_DEV/_DEV_<no> env var suffixes
and as it refers to the ICV vars and their scope, those were added to the
OMP_ env vars for reference. For OMP_NESTING, a note that those were
deprecated was added plus a bunch of cross references. For OMP_ALLOCATOR,
add note about the lack of per-device env vars support.

A new section, consisting mostly of cross references was added to document
the implementation-defined ICV initialization, especially as OpenMP demands
that implementations document what they do for 'implementation defined'.

For nvptx, the implementation-defined used stack size was documented

libgomp/
* libgomp.texi: Use @var for ICV vars.
(OpenMP Environment Variables): Mention _ALL/_DEV/_DEV_<no> variants,
document which ICV is set and which scope the ICV has; extend/cleanup
some @ref.
(Implementation-defined ICV Initialization): New.
(nvptx): Document the implementation-defined used per-warp stack size.

13 months agotree-optimization/110332 - fix ICE with phiprop
Richard Biener [Thu, 22 Jun 2023 07:04:01 +0000 (09:04 +0200)]
tree-optimization/110332 - fix ICE with phiprop

The following fixes an ICE that occurs when we visit an edge
inserted load from the code validating correctness for inserting
an aggregate copy there.  We can simply skip those loads here.

PR tree-optimization/110332
* tree-ssa-phiprop.cc (propagate_with_phi): Always
check aliasing with edge inserted loads.

* g++.dg/torture/pr110332.C: New testcase.
* gcc.dg/torture/pr110332-1.c: Likewise.
* gcc.dg/torture/pr110332-2.c: Likewise.

13 months agoi386: Convert ptestz of pandn into ptestc.
Roger Sayle [Thu, 22 Jun 2023 06:43:07 +0000 (07:43 +0100)]
i386: Convert ptestz of pandn into ptestc.

This patch is the next installment in a set of backend patches around
improvements to ptest/vptest.  A previous patch optimized the sequence
t=pand(x,y); ptestz(t,t) into the equivalent ptestz(x,y), using the
property that ZF is set to (X&Y) == 0.  This patch performs a similar
transformation, converting t=pandn(x,y); ptestz(t,t) into the (almost)
equivalent ptestc(y,x), using the property that the CF flags is set to
(~X&Y) == 0.  The tricky bit is that this sets the CF flag instead of
the ZF flag, so we can only perform this transformation when we can
also convert the flags consumer, as well as the producer.

For the test case:

int foo (__m128i x, __m128i y)
{
  __m128i a = x & ~y;
  return __builtin_ia32_ptestz128 (a, a);
}

With -O2 -msse4.1 we previously generated:

foo: pandn   %xmm0, %xmm1
        xorl    %eax, %eax
        ptest   %xmm1, %xmm1
        sete    %al
        ret

with this patch we now generate:

foo: xorl    %eax, %eax
        ptest   %xmm0, %xmm1
        setc    %al
        ret

At the same time, this patch also provides alternative fixes for
PR target/109973 and PR target/110118, by recognizing that ptestc(x,x)
always sets the carry flag (X&~X is always zero).  This is achieved
both by recognizing the special case in ix86_expand_sse_ptest and with
a splitter to convert an eligible ptest into an stc.

2023-06-22  Roger Sayle  <roger@nextmovesoftware.com>
    Uros Bizjak  <ubizjak@gmail.com>

gcc/ChangeLog
* config/i386/i386-expand.cc (ix86_expand_sse_ptest): Recognize
expansion of ptestc with equal operands as producing const1_rtx.
* config/i386/i386.cc (ix86_rtx_costs): Provide accurate cost
estimates of UNSPEC_PTEST, where the ptest performs the PAND
or PAND of its operands.
* config/i386/sse.md (define_split): Transform CCCmode UNSPEC_PTEST
of reg_equal_p operands into an x86_stc instruction.
(define_split): Split pandn/ptestz/set{n?}e into ptestc/set{n?}c.
(define_split): Similar to above for strict_low_part destinations.
(define_split): Split pandn/ptestz/j{n?}e into ptestc/j{n?}c.

gcc/testsuite/ChangeLog
* gcc.target/i386/avx-vptest-4.c: New test case.
* gcc.target/i386/avx-vptest-5.c: Likewise.
* gcc.target/i386/avx-vptest-6.c: Likewise.
* gcc.target/i386/pr109973-1.c: Update test case.
* gcc.target/i386/pr109973-2.c: Likewise.
* gcc.target/i386/sse4_1-ptest-4.c: New test case.
* gcc.target/i386/sse4_1-ptest-5.c: Likewise.
* gcc.target/i386/sse4_1-ptest-6.c: Likewise.

13 months agoanalyzer: add text-art visualizations of out-of-bounds accesses [PR106626]
David Malcolm [Thu, 22 Jun 2023 01:49:00 +0000 (21:49 -0400)]
analyzer: add text-art visualizations of out-of-bounds accesses [PR106626]

This patch extends -Wanalyzer-out-of-bounds so that, where possible, it
will emit a text art diagram visualizing the spatial relationship between
(a) the memory region that the analyzer predicts would be accessed, versus
(b) the range of memory that is valid to access - whether they overlap,
are touching, are close or far apart; which one is before or after in
memory, the relative sizes involved, the direction of the access (read vs
write), and, in some cases, the values of data involved.  This diagram
can be suppressed using -fdiagnostics-text-art-charset=none.

For example, given:

int32_t arr[10];

int32_t int_arr_read_element_before_start_far(void)
{
  return arr[-100];
}

it emits:

demo-1.c: In function ‘int_arr_read_element_before_start_far’:
demo-1.c:7:13: warning: buffer under-read [CWE-127] [-Wanalyzer-out-of-bounds]
    7 |   return arr[-100];
      |          ~~~^~~~~~
  ‘int_arr_read_element_before_start_far’: event 1
    |
    |    7 |   return arr[-100];
    |      |          ~~~^~~~~~
    |      |             |
    |      |             (1) out-of-bounds read from byte -400 till byte -397 but ‘arr’ starts at byte 0
    |
demo-1.c:7:13: note: valid subscripts for ‘arr’ are ‘[0]’ to ‘[9]’

  ┌───────────────────────────┐
  │read of ‘int32_t’ (4 bytes)│
  └───────────────────────────┘
                ^
                │
                │
  ┌───────────────────────────┐              ┌────────┬────────┬─────────┐
  │                           │              │  [0]   │  ...   │   [9]   │
  │    before valid range     │              ├────────┴────────┴─────────┤
  │                           │              │‘arr’ (type: ‘int32_t[10]’)│
  └───────────────────────────┘              └───────────────────────────┘
  ├─────────────┬─────────────┤├─────┬──────┤├─────────────┬─────────────┤
                │                    │                     │
   ╭────────────┴───────────╮   ╭────┴────╮        ╭───────┴──────╮
   │⚠️  under-read of 4 bytes│   │396 bytes│        │size: 40 bytes│
   ╰────────────────────────╯   ╰─────────╯        ╰──────────────╯

and given:

  #include <string.h>

  void
  test_non_ascii ()
  {
    char buf[5];
    strcpy (buf, "文字化け");
  }

it emits:

demo-2.c: In function ‘test_non_ascii’:
demo-2.c:7:3: warning: stack-based buffer overflow [CWE-121] [-Wanalyzer-out-of-bounds]
    7 |   strcpy (buf, "文字化け");
      |   ^~~~~~~~~~~~~~~~~~~~~~~~
  ‘test_non_ascii’: events 1-2
    |
    |    6 |   char buf[5];
    |      |        ^~~
    |      |        |
    |      |        (1) capacity: 5 bytes
    |    7 |   strcpy (buf, "文字化け");
    |      |   ~~~~~~~~~~~~~~~~~~~~~~~~
    |      |   |
    |      |   (2) out-of-bounds write from byte 5 till byte 12 but ‘buf’ ends at byte 5
    |
demo-2.c:7:3: note: write of 8 bytes to beyond the end of ‘buf’
    7 |   strcpy (buf, "文字化け");
      |   ^~~~~~~~~~~~~~~~~~~~~~~~
demo-2.c:7:3: note: valid subscripts for ‘buf’ are ‘[0]’ to ‘[4]’

  ┌─────┬─────┬─────┬────┬────┐┌────┬────┬────┬────┬────┬────┬────┬──────┐
  │ [0] │ [1] │ [2] │[3] │[4] ││[5] │[6] │[7] │[8] │[9] │[10]│[11]│ [12] │
  ├─────┼─────┼─────┼────┼────┤├────┼────┼────┼────┼────┼────┼────┼──────┤
  │0xe6 │0x96 │0x87 │0xe5│0xad││0x97│0xe5│0x8c│0x96│0xe3│0x81│0x91│ 0x00 │
  ├─────┴─────┴─────┼────┴────┴┴────┼────┴────┴────┼────┴────┴────┼──────┤
  │     U+6587      │    U+5b57     │    U+5316    │    U+3051    │U+0000│
  ├─────────────────┼───────────────┼──────────────┼──────────────┼──────┤
  │       文        │      字       │      化      │      け      │ NUL  │
  ├─────────────────┴───────────────┴──────────────┴──────────────┴──────┤
  │                  string literal (type: ‘char[13]’)                   │
  └──────────────────────────────────────────────────────────────────────┘
     │     │     │    │    │     │    │    │    │    │    │    │     │
     │     │     │    │    │     │    │    │    │    │    │    │     │
     v     v     v    v    v     v    v    v    v    v    v    v     v
  ┌─────┬────────────────┬────┐┌─────────────────────────────────────────┐
  │ [0] │      ...       │[4] ││                                         │
  ├─────┴────────────────┴────┤│            after valid range            │
  │  ‘buf’ (type: ‘char[5]’)  ││                                         │
  └───────────────────────────┘└─────────────────────────────────────────┘
  ├─────────────┬─────────────┤├────────────────────┬────────────────────┤
                │                                   │
       ╭────────┴────────╮              ╭───────────┴──────────╮
       │capacity: 5 bytes│              │⚠️  overflow of 8 bytes│
       ╰─────────────────╯              ╰──────────────────────╯

showing that the overflow occurs partway through the UTF-8 encoding of
the U+5b57 code point.

There are lots more examples in the test suite.

It doesn't show up in this email, but the above diagrams are colorized
to constrast the valid and invalid access ranges.

gcc/ChangeLog:
PR analyzer/106626
* Makefile.in (ANALYZER_OBJS): Add analyzer/access-diagram.o.
* doc/invoke.texi (Wanalyzer-out-of-bounds): Add description of
text art.
(fanalyzer-debug-text-art): New.

gcc/analyzer/ChangeLog:
PR analyzer/106626
* access-diagram.cc: New file.
* access-diagram.h: New file.
* analyzer.h (class region_offset): Add default ctor.
(region_offset::make_byte_offset): New decl.
(region_offset::concrete_p): New.
(region_offset::get_concrete_byte_offset): New.
(region_offset::calc_symbolic_bit_offset): New decl.
(region_offset::calc_symbolic_byte_offset): New decl.
(region_offset::dump_to_pp): New decl.
(region_offset::dump): New decl.
(operator<, operator<=, operator>, operator>=): New decls for
region_offset.
* analyzer.opt
(-param=analyzer-text-art-string-ellipsis-threshold=): New.
(-param=analyzer-text-art-string-ellipsis-head-len=): New.
(-param=analyzer-text-art-string-ellipsis-tail-len=): New.
(-param=analyzer-text-art-ideal-canvas-width=): New.
(fanalyzer-debug-text-art): New.
* bounds-checking.cc: Include "intl.h", "diagnostic-diagram.h",
and "analyzer/access-diagram.h".
(class out_of_bounds::oob_region_creation_event_capacity): New.
(out_of_bounds::out_of_bounds): Add "model" and "sval_hint"
params.
(out_of_bounds::mark_interesting_stuff): Use the base region.
(out_of_bounds::add_region_creation_events): Use
oob_region_creation_event_capacity.
(out_of_bounds::get_dir): New pure vfunc.
(out_of_bounds::maybe_show_notes): New.
(out_of_bounds::maybe_show_diagram): New.
(out_of_bounds::make_access_diagram): New.
(out_of_bounds::m_model): New field.
(out_of_bounds::m_sval_hint): New field.
(out_of_bounds::m_region_creation_event_id): New field.
(concrete_out_of_bounds::concrete_out_of_bounds): Update for new
fields.
(concrete_past_the_end::concrete_past_the_end): Likewise.
(concrete_past_the_end::add_region_creation_events): Use
oob_region_creation_event_capacity.
(concrete_buffer_overflow::concrete_buffer_overflow): Update for
new fields.
(concrete_buffer_overflow::emit): Replace call to
maybe_describe_array_bounds with maybe_show_notes.
(concrete_buffer_overflow::get_dir): New.
(concrete_buffer_over_read::concrete_buffer_over_read): Update for
new fields.
(concrete_buffer_over_read::emit): Replace call to
maybe_describe_array_bounds with maybe_show_notes.
(concrete_buffer_overflow::get_dir): New.
(concrete_buffer_underwrite::concrete_buffer_underwrite): Update
for new fields.
(concrete_buffer_underwrite::emit): Replace call to
maybe_describe_array_bounds with maybe_show_notes.
(concrete_buffer_underwrite::get_dir): New.
(concrete_buffer_under_read::concrete_buffer_under_read): Update
for new fields.
(concrete_buffer_under_read::emit): Replace call to
maybe_describe_array_bounds with maybe_show_notes.
(concrete_buffer_under_read::get_dir): New.
(symbolic_past_the_end::symbolic_past_the_end): Update for new
fields.
(symbolic_buffer_overflow::symbolic_buffer_overflow): Likewise.
(symbolic_buffer_overflow::emit): Call maybe_show_notes.
(symbolic_buffer_overflow::get_dir): New.
(symbolic_buffer_over_read::symbolic_buffer_over_read): Update for
new fields.
(symbolic_buffer_over_read::emit): Call maybe_show_notes.
(symbolic_buffer_over_read::get_dir): New.
(region_model::check_symbolic_bounds): Add "sval_hint" param.  Pass
it and sized_offset_reg to diagnostics.
(region_model::check_region_bounds): Add "sval_hint" param, passing
it to diagnostics.
* diagnostic-manager.cc
(diagnostic_manager::emit_saved_diagnostic): Pass logger to
pending_diagnostic::emit.
* engine.cc: Add logger param to pending_diagnostic::emit
implementations.
* infinite-recursion.cc: Likewise.
* kf-analyzer.cc: Likewise.
* kf.cc: Likewise.  Add nullptr for new param of
check_region_for_write.
* pending-diagnostic.h: Likewise in decl.
* region-model-manager.cc
(region_model_manager::get_or_create_int_cst): Convert param from
poly_int64 to const poly_wide_int_ref &.
(region_model_manager::maybe_fold_binop): Support type being NULL
when checking for floating-point types.
Check for (X + Y) - X => Y.  Be less strict about types when folding
associative ops.  Check for (X + Y) * CST => (X * CST) + (Y * CST).
* region-model-manager.h
(region_model_manager::get_or_create_int_cst): Convert param from
poly_int64 to const poly_wide_int_ref &.
* region-model.cc: Add logger param to pending_diagnostic::emit
implementations.
(region_model::check_external_function_for_access_attr): Update
for new param of check_region_for_write.
(region_model::deref_rvalue): Use nullptr rather than NULL.
(region_model::get_capacity): Handle RK_STRING.
(region_model::check_region_access): Add "sval_hint" param; pass it to
check_region_bounds.
(region_model::check_region_for_write): Add "sval_hint" param;
pass it to check_region_access.
(region_model::check_region_for_read): Add NULL for new param to
check_region_access.
(region_model::set_value): Pass rhs_sval to
check_region_for_write.
(region_model::get_representative_path_var_1): Handle SK_CONSTANT
in the check for infinite recursion.
* region-model.h (region_model::check_region_for_write): Add
"sval_hint" param.
(region_model::check_region_access): Likewise.
(region_model::check_symbolic_bounds): Likewise.
(region_model::check_region_bounds): Likewise.
* region.cc (region_offset::make_byte_offset): New.
(region_offset::calc_symbolic_bit_offset): New.
(region_offset::calc_symbolic_byte_offset): New.
(region_offset::dump_to_pp): New.
(region_offset::dump): New.
(struct linear_op): New.
(operator<, operator<=, operator>, operator>=): New, for
region_offset.
(region::get_next_offset): New.
(region::get_relative_symbolic_offset): Use ptrdiff_type_node.
(field_region::get_relative_symbolic_offset): Likewise.
(element_region::get_relative_symbolic_offset): Likewise.
(bit_range_region::get_relative_symbolic_offset): Likewise.
* region.h (region::get_next_offset): New decl.
* sm-fd.cc: Add logger param to pending_diagnostic::emit
implementations.
* sm-file.cc: Likewise.
* sm-malloc.cc: Likewise.
* sm-pattern-test.cc: Likewise.
* sm-sensitive.cc: Likewise.
* sm-signal.cc: Likewise.
* sm-taint.cc: Likewise.
* store.cc (bit_range::contains_p): Allow "out" to be null.
* store.h (byte_range::get_start_bit_offset): New.
(byte_range::get_next_bit_offset): New.
* varargs.cc: Add logger param to pending_diagnostic::emit
implementations.

gcc/testsuite/ChangeLog:
PR analyzer/106626
* gcc.dg/analyzer/data-model-1.c (test_16): Update for
out-of-bounds working.
* gcc.dg/analyzer/out-of-bounds-diagram-1-ascii.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-1-debug.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-1-emoji.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-1-json.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-1-sarif.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-1-unicode.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-10.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-11.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-12.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-13.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-14.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-15.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-2.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-3.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-4.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-5-ascii.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-5-unicode.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-6.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-7.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-8.c: New test.
* gcc.dg/analyzer/out-of-bounds-diagram-9.c: New test.
* gcc.dg/analyzer/pattern-test-2.c: Update expected results.
* gcc.dg/analyzer/pr101962.c: Update expected results.
* gcc.dg/plugin/analyzer_gil_plugin.c:  Add logger param to
pending_diagnostic::emit implementations.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
13 months agodiagnostics: add support for "text art" diagrams
David Malcolm [Thu, 22 Jun 2023 01:49:00 +0000 (21:49 -0400)]
diagnostics: add support for "text art" diagrams

Existing text output in GCC has to be implemented by writing
sequentially to a pretty_printer instance.  This makes it
hard to implement some kinds of diagnostic output (see e.g.
diagnostic-show-locus.cc).

This patch adds more flexible ways of creating text output:
- a canvas class, which can be "painted" to via random-access (rather
that sequentially)
- a table class for 2D grid layout, supporting items that span
multiple rows/columns
- a widget class for organizing diagrams hierarchically.

The patch also expands GCC's diagnostics subsystem so that diagnostics
can have "text art" diagrams - think ASCII art, but potentially
including some Unicode characters, such as box-drawing chars.

The new code is in a new "gcc/text-art" subdirectory and "text_art"
namespace.

The patch adds a new "-fdiagnostics-text-art-charset=VAL" option, with
values:
- "none": don't emit diagrams (added to -fdiagnostics-plain-output)
- "ascii": use pure ASCII in diagrams
- "unicode": allow for conservative use of unicode drawing characters
(such as box-drawing characters).
- "emoji" (the default): as "unicode", but potentially allow for
conservative use of emoji in the output (such as U+26A0 WARNING SIGN).
I made it possible to disable emoji separately from unicode as I believe
there's a generation gap in acceptance of these characters (some older
programmers have a visceral reaction against them, whereas younger
programmers may have no problem with them).

Diagrams are emitted to stderr by default.  With SARIF output they are
captured as a location in "relatedLocations", with the diagram as a
code block in Markdown within a "markdown" property of a message.

This patch doesn't add any such diagram usage to GCC, saving that for
followups, apart from adding a plugin to the test suite to exercise the
functionality.

contrib/ChangeLog:
* unicode/gen-box-drawing-chars.py: New file.
* unicode/gen-combining-chars.py: New file.
* unicode/gen-printable-chars.py: New file.

gcc/ChangeLog:
* Makefile.in (OBJS-libcommon): Add text-art/box-drawing.o,
text-art/canvas.o, text-art/ruler.o, text-art/selftests.o,
text-art/style.o, text-art/styled-string.o, text-art/table.o,
text-art/theme.o, and text-art/widget.o.
* color-macros.h (COLOR_FG_BRIGHT_BLACK): New.
(COLOR_FG_BRIGHT_RED): New.
(COLOR_FG_BRIGHT_GREEN): New.
(COLOR_FG_BRIGHT_YELLOW): New.
(COLOR_FG_BRIGHT_BLUE): New.
(COLOR_FG_BRIGHT_MAGENTA): New.
(COLOR_FG_BRIGHT_CYAN): New.
(COLOR_FG_BRIGHT_WHITE): New.
(COLOR_BG_BRIGHT_BLACK): New.
(COLOR_BG_BRIGHT_RED): New.
(COLOR_BG_BRIGHT_GREEN): New.
(COLOR_BG_BRIGHT_YELLOW): New.
(COLOR_BG_BRIGHT_BLUE): New.
(COLOR_BG_BRIGHT_MAGENTA): New.
(COLOR_BG_BRIGHT_CYAN): New.
(COLOR_BG_BRIGHT_WHITE): New.
* common.opt (fdiagnostics-text-art-charset=): New option.
(diagnostic-text-art.h): New SourceInclude.
(diagnostic_text_art_charset) New Enum and EnumValues.
* configure: Regenerate.
* configure.ac (gccdepdir): Add text-art to loop.
* diagnostic-diagram.h: New file.
* diagnostic-format-json.cc (json_emit_diagram): New.
(diagnostic_output_format_init_json): Wire it up to
context->m_diagrams.m_emission_cb.
* diagnostic-format-sarif.cc: Include "diagnostic-diagram.h" and
"text-art/canvas.h".
(sarif_result::on_nested_diagnostic): Move code to...
(sarif_result::add_related_location): ...this new function.
(sarif_result::on_diagram): New.
(sarif_builder::emit_diagram): New.
(sarif_builder::make_message_object_for_diagram): New.
(sarif_emit_diagram): New.
(diagnostic_output_format_init_sarif): Set
context->m_diagrams.m_emission_cb to sarif_emit_diagram.
* diagnostic-text-art.h: New file.
* diagnostic.cc: Include "diagnostic-text-art.h",
"diagnostic-diagram.h", and "text-art/theme.h".
(diagnostic_initialize): Initialize context->m_diagrams and
call diagnostics_text_art_charset_init.
(diagnostic_finish): Clean up context->m_diagrams.m_theme.
(diagnostic_emit_diagram): New.
(diagnostics_text_art_charset_init): New.
* diagnostic.h (text_art::theme): New forward decl.
(class diagnostic_diagram): Likewise.
(diagnostic_context::m_diagrams): New field.
(diagnostic_emit_diagram): New decl.
* doc/invoke.texi (Diagnostic Message Formatting Options): Add
-fdiagnostics-text-art-charset=.
(-fdiagnostics-plain-output): Add
-fdiagnostics-text-art-charset=none.
* gcc.cc: Include "diagnostic-text-art.h".
(driver_handle_option): Handle OPT_fdiagnostics_text_art_charset_.
* opts-common.cc (decode_cmdline_options_to_array): Add
"-fdiagnostics-text-art-charset=none" to expanded_args for
-fdiagnostics-plain-output.
* opts.cc: Include "diagnostic-text-art.h".
(common_handle_option): Handle OPT_fdiagnostics_text_art_charset_.
* pretty-print.cc (pp_unicode_character): New.
* pretty-print.h (pp_unicode_character): New decl.
* selftest-run-tests.cc: Include "text-art/selftests.h".
(selftest::run_tests): Call text_art_tests.
* text-art/box-drawing-chars.inc: New file, generated by
contrib/unicode/gen-box-drawing-chars.py.
* text-art/box-drawing.cc: New file.
* text-art/box-drawing.h: New file.
* text-art/canvas.cc: New file.
* text-art/canvas.h: New file.
* text-art/ruler.cc: New file.
* text-art/ruler.h: New file.
* text-art/selftests.cc: New file.
* text-art/selftests.h: New file.
* text-art/style.cc: New file.
* text-art/styled-string.cc: New file.
* text-art/table.cc: New file.
* text-art/table.h: New file.
* text-art/theme.cc: New file.
* text-art/theme.h: New file.
* text-art/types.h: New file.
* text-art/widget.cc: New file.
* text-art/widget.h: New file.

gcc/testsuite/ChangeLog:
* gcc.dg/plugin/diagnostic-test-text-art-ascii-bw.c: New test.
* gcc.dg/plugin/diagnostic-test-text-art-ascii-color.c: New test.
* gcc.dg/plugin/diagnostic-test-text-art-none.c: New test.
* gcc.dg/plugin/diagnostic-test-text-art-unicode-bw.c: New test.
* gcc.dg/plugin/diagnostic-test-text-art-unicode-color.c: New test.
* gcc.dg/plugin/diagnostic_plugin_test_text_art.c: New test plugin.
* gcc.dg/plugin/plugin.exp (plugin_test_list): Add them.

libcpp/ChangeLog:
* charset.cc (get_cppchar_property): New function template, based
on...
(cpp_wcwidth): ...this function.  Rework to use the above.
Include "combining-chars.inc".
(cpp_is_combining_char): New function
Include "printable-chars.inc".
(cpp_is_printable_char): New function
* combining-chars.inc: New file, generated by
contrib/unicode/gen-combining-chars.py.
* include/cpplib.h (cpp_is_combining_char): New function decl.
(cpp_is_printable_char): New function decl.
* printable-chars.inc: New file, generated by
contrib/unicode/gen-printable-chars.py.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
13 months agotestsuite: move handle-multiline-outputs to before check for blank lines
David Malcolm [Thu, 22 Jun 2023 01:48:59 +0000 (21:48 -0400)]
testsuite: move handle-multiline-outputs to before check for blank lines

I have followup patches that require checking for multiline patterns
that have blank lines within them, so this moves the handling of
multiline patterns before the check for blank lines, allowing for such
multiline patterns.

Doing so uncovers some issues with existing multiline directives, which
the patch fixes.

gcc/testsuite/ChangeLog:
* c-c++-common/Wlogical-not-parentheses-2.c: Split up the
multiline directive.
* gcc.dg/analyzer/malloc-macro-inline-events.c: Remove redundant
dg-regexp directives.
* gcc.dg/missing-header-fixit-5.c: Split up the multiline
directives.
* lib/gcc-dg.exp (gcc-dg-prune): Move call to
handle-multiline-outputs from prune_gcc_output to here.
* lib/multiline.exp (dg-end-multiline-output): Move call to
maybe-handle-nn-line-numbers from prune_gcc_output to here.
* lib/prune.exp (prune_gcc_output): Move calls to
maybe-handle-nn-line-numbers and handle-multiline-outputs from
here to the above.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
13 months agocompiler: determine types of Slice_{value,info} expressions
Ian Lance Taylor [Wed, 21 Jun 2023 21:27:02 +0000 (14:27 -0700)]
compiler: determine types of Slice_{value,info} expressions

This fixes an accidental omission in the determine types pass.

Test case is https://go.dev/cl/505015.

Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/504797

13 months agoDaily bump.
GCC Administrator [Thu, 22 Jun 2023 00:17:09 +0000 (00:17 +0000)]
Daily bump.

13 months agofunction: Change return type of predicate function from int to bool
Uros Bizjak [Wed, 21 Jun 2023 19:55:30 +0000 (21:55 +0200)]
function: Change return type of predicate function from int to bool

Also change some internal variables to bool and some functions to void.

gcc/ChangeLog:

* function.h (emit_initial_value_sets):
Change return type from int to void.
(aggregate_value_p): Change return type from int to bool.
(prologue_contains): Ditto.
(epilogue_contains): Ditto.
(prologue_epilogue_contains): Ditto.
* function.cc (temp_slot): Make "in_use" variable bool.
(make_slot_available): Update for changed "in_use" variable.
(assign_stack_temp_for_type): Ditto.
(emit_initial_value_sets): Change return type from int to void
and update function body accordingly.
(instantiate_virtual_regs): Ditto.
(rest_of_handle_thread_prologue_and_epilogue): Ditto.
(safe_insn_predicate): Change return type from int to bool.
(aggregate_value_p): Change return type from int to bool
and update function body accordingly.
(prologue_contains): Change return type from int to bool.
(prologue_epilogue_contains): Ditto.

13 months agoc-family: implement -ffp-contract=on
Alexander Monakov [Thu, 18 May 2023 20:47:47 +0000 (23:47 +0300)]
c-family: implement -ffp-contract=on

Implement -ffp-contract=on for C and C++ without changing default
behavior (=off for -std=cNN, =fast for C++ and -std=gnuNN).

gcc/c-family/ChangeLog:

* c-gimplify.cc (fma_supported_p): New helper.
(c_gimplify_expr) [PLUS_EXPR, MINUS_EXPR]: Implement FMA
contraction.

gcc/ChangeLog:

* common.opt (fp_contract_mode) [on]: Remove fallback.
* config/sh/sh.md (*fmasf4): Correct flag_fp_contract_mode test.
* doc/invoke.texi (-ffp-contract): Update.
* trans-mem.cc (diagnose_tm_1): Skip internal function calls.

13 months agoFortran: Fix some bugs in associate [PR87477]
Paul Thomas [Wed, 21 Jun 2023 16:05:58 +0000 (17:05 +0100)]
Fortran: Fix some bugs in associate [PR87477]

2023-06-21  Paul Thomas  <pault@gcc.gnu.org>

gcc/fortran
PR fortran/87477
PR fortran/88688
PR fortran/94380
PR fortran/107900
PR fortran/110224
* decl.cc (char_len_param_value): Fix memory leak.
(resolve_block_construct): Remove unnecessary static decls.
* expr.cc (gfc_is_ptr_fcn): New function.
(gfc_check_vardef_context): Use it to permit pointer function
result selectors to be used for associate names in variable
definition context.
* gfortran.h: Prototype for gfc_is_ptr_fcn.
* match.cc (build_associate_name): New function.
(gfc_match_select_type): Use the new function to replace inline
version and to build a new associate name for the case where
the supplied associate name is already used for that purpose.
* resolve.cc (resolve_assoc_var): Call gfc_is_ptr_fcn to allow
associate names with pointer function targets to be used in
variable definition context.
* trans-decl.cc (gfc_get_symbol_decl): Unlimited polymorphic
variables need deferred initialisation of the vptr.
(gfc_trans_deferred_vars): Do the vptr initialisation.
* trans-stmt.cc (trans_associate_var): Ensure that a pointer
associate name points to the target of the selector and not
the selector itself.

gcc/testsuite/
PR fortran/87477
PR fortran/107900
* gfortran.dg/pr107900.f90 : New test

PR fortran/110224
* gfortran.dg/pr110224.f90 : New test

PR fortran/88688
* gfortran.dg/pr88688.f90 : New test

PR fortran/94380
* gfortran.dg/pr94380.f90 : New test

PR fortran/95398
* gfortran.dg/pr95398.f90 : Set -std=f2008, bump the line
numbers in the error tests by two and change the text in two.

13 months agoFortran: Seg fault passing string to type cptr dummy [PR108961].
Paul Thomas [Wed, 21 Jun 2023 16:01:57 +0000 (17:01 +0100)]
Fortran: Seg fault passing string to type cptr dummy [PR108961].

2023-06-21  Paul Thomas  <pault@gcc.gnu.org>

gcc/fortran
PR fortran/108961
* trans-expr.cc (gfc_conv_procedure_call): The hidden string
length must not be passed to a formal arg of type(cptr).

gcc/testsuite/
PR fortran/108961
* gfortran.dg/pr108961.f90: New test.

13 months agovect: Add testcases for unsigned conversions [PR110018]
Uros Bizjak [Wed, 21 Jun 2023 14:34:39 +0000 (16:34 +0200)]
vect: Add testcases for unsigned conversions [PR110018]

Also test convresions with unsigned types.

PR target/110018

gcc/testsuite/ChangeLog:

* gcc.target/i386/pr110018-1.c: Use explicit signed types.
* gcc.target/i386/pr110018-2.c: New test.

13 months agoaarch64: Avoid same input and output Z register for gather loads
Kyrylo Tkachov [Wed, 21 Jun 2023 12:43:26 +0000 (13:43 +0100)]
aarch64: Avoid same input and output Z register for gather loads

The architecture recommends that load-gather instructions avoid using the same
Z register for the load address and the destination, and the Software Optimization
Guides for Arm cores recommend that as well.
This means that for code like:

svuint64_t
food (svbool_t p, uint64_t *in, svint64_t offsets, svuint64_t a)
{
  return svadd_u64_x (p, a, svld1_gather_offset(p, in, offsets));
}

we'll want to avoid generating the current:
food:
        ld1d    z0.d, p0/z, [x0, z0.d] // Z0 reused as input and output.
        add     z0.d, z1.d, z0.d
        ret

However, we still want to avoid generating extra moves where there were
none before, so the tight aarch64-sve-acle.exp tests for load gathers
should still pass as they are.

This patch implements that recommendation for the load gather patterns by:
* duplicating the alternatives
* marking the output operand as early clobber
* Tying the input Z register operand in the original alternatives to 0
* Penalising the original alternatives with '?'

This results in a large-ish patch in terms of diff lines but the new
compact syntax (thanks Tamar) makes it quite a readable an regular change.

The benchmark numbers on a Neoverse V1 on fprate look okay:
        diff
503.bwaves_r 0.00%
507.cactuBSSN_r 0.00%
508.namd_r 0.00%
510.parest_r 0.55%
511.povray_r 0.22%
519.lbm_r 0.00%
521.wrf_r 0.00%
526.blender_r 0.00%
527.cam4_r 0.56%
538.imagick_r 0.00%
544.nab_r 0.00%
549.fotonik3d_r 0.00%
554.roms_r 0.00%
fprate         0.10%

Bootstrapped and tested on aarch64-none-linux-gnu.

gcc/ChangeLog:

* config/aarch64/aarch64-sve.md (mask_gather_load<mode><v_int_container>):
Add alternatives to prefer to avoid same input and output Z register.
(mask_gather_load<mode><v_int_container>): Likewise.
(*mask_gather_load<mode><v_int_container>_<su>xtw_unpacked): Likewise.
(*mask_gather_load<mode><v_int_container>_sxtw): Likewise.
(*mask_gather_load<mode><v_int_container>_uxtw): Likewise.
(@aarch64_gather_load_<ANY_EXTEND:optab><SVE_4HSI:mode><SVE_4BHI:mode>):
Likewise.
(@aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode><SVE_2BHSI:mode>):
Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_<ANY_EXTEND2:su>xtw_unpacked): Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_sxtw): Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_uxtw): Likewise.
(@aarch64_ldff1_gather<mode>): Likewise.
(@aarch64_ldff1_gather<mode>): Likewise.
(*aarch64_ldff1_gather<mode>_sxtw): Likewise.
(*aarch64_ldff1_gather<mode>_uxtw): Likewise.
(@aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx4_WIDE:mode>
<VNx4_NARROW:mode>): Likewise.
(@aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>): Likewise.
(*aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>_sxtw): Likewise.
(*aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>_uxtw): Likewise.
* config/aarch64/aarch64-sve2.md (@aarch64_gather_ldnt<mode>): Likewise.
(@aarch64_gather_ldnt_<ANY_EXTEND:optab><SVE_FULL_SDI:mode>
<SVE_PARTIAL_I:mode>): Likewise.

gcc/testsuite/ChangeLog:

* gcc.target/aarch64/sve/gather_earlyclobber.c: New test.
* gcc.target/aarch64/sve2/gather_earlyclobber.c: New test.

13 months agoaarch64: Convert SVE gather patterns to compact syntax
Kyrylo Tkachov [Wed, 21 Jun 2023 12:40:15 +0000 (13:40 +0100)]
aarch64: Convert SVE gather patterns to compact syntax

This patch converts the SVE load gather patterns to the new compact syntax
that Tamar introduced. This allows for a future patch I want to contribute
to add more alternatives that are better viewed in the more compact form.

The lines in some patterns are >80 long now, but I think that's unavoidable
and those patterns already had overly long constraint strings.

No functional change intended.
Bootstrapped and tested on aarch64-none-linux-gnu.

gcc/ChangeLog:

* config/aarch64/aarch64-sve.md (mask_gather_load<mode><v_int_container>):
Convert to compact alternatives syntax.
(mask_gather_load<mode><v_int_container>): Likewise.
(*mask_gather_load<mode><v_int_container>_<su>xtw_unpacked): Likewise.
(*mask_gather_load<mode><v_int_container>_sxtw): Likewise.
(*mask_gather_load<mode><v_int_container>_uxtw): Likewise.
(@aarch64_gather_load_<ANY_EXTEND:optab><SVE_4HSI:mode><SVE_4BHI:mode>):
Likewise.
(@aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode><SVE_2BHSI:mode>):
Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_<ANY_EXTEND2:su>xtw_unpacked): Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_sxtw): Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_uxtw): Likewise.
(@aarch64_ldff1_gather<mode>): Likewise.
(@aarch64_ldff1_gather<mode>): Likewise.
(*aarch64_ldff1_gather<mode>_sxtw): Likewise.
(*aarch64_ldff1_gather<mode>_uxtw): Likewise.
(@aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx4_WIDE:mode>
<VNx4_NARROW:mode>): Likewise.
(@aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>): Likewise.
(*aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>_sxtw): Likewise.
(*aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>_uxtw): Likewise.
* config/aarch64/aarch64-sve2.md (@aarch64_gather_ldnt<mode>): Likewise.
(@aarch64_gather_ldnt_<ANY_EXTEND:optab><SVE_FULL_SDI:mode>
<SVE_PARTIAL_I:mode>): Likewise.

13 months agoRevert "aarch64: Convert SVE gather patterns to compact syntax"
Kyrylo Tkachov [Wed, 21 Jun 2023 12:38:56 +0000 (13:38 +0100)]
Revert "aarch64: Convert SVE gather patterns to compact syntax"

This reverts commit bb3c69058a5fb874ea3c5c26bfb331d33d0497c3.

13 months agoMove can_vec_mask_load_store_p and get_len_load_store_mode from "optabs-query" into...
Ju-Zhe Zhong [Wed, 21 Jun 2023 11:33:38 +0000 (19:33 +0800)]
Move can_vec_mask_load_store_p and get_len_load_store_mode from "optabs-query" into "optabs-tree"

Since we want both can_vec_mask_load_store_p and get_len_load_store_mode
can see "internal_fn", move these 2 functions into optabs-tree.

gcc/ChangeLog:

* optabs-query.cc (can_vec_mask_load_store_p): Move to optabs-tree.cc.
(get_len_load_store_mode): Ditto.
* optabs-query.h (can_vec_mask_load_store_p): Move to optabs-tree.h.
(get_len_load_store_mode): Ditto.
* optabs-tree.cc (can_vec_mask_load_store_p): New function.
(get_len_load_store_mode): Ditto.
* optabs-tree.h (can_vec_mask_load_store_p): Ditto.
(get_len_load_store_mode): Ditto.
* tree-if-conv.cc: include optabs-tree instead of optabs-query

13 months agoLess strip_offset in IVOPTs
Richard Biener [Wed, 21 Jun 2023 10:08:01 +0000 (12:08 +0200)]
Less strip_offset in IVOPTs

This avoids one strip_offset use in add_iv_candidate_for_use where
we know it operates on a sizetype quantity.

* tree-ssa-loop-ivopts.cc (add_iv_candidate_for_use): Use
split_constant_offset for the POINTER_PLUS_EXPR case.

13 months agoLess strip_offset in IVOPTs
Richard Biener [Wed, 21 Jun 2023 09:38:19 +0000 (11:38 +0200)]
Less strip_offset in IVOPTs

This avoids a strip_offset use in record_group_use where we know
it operates on addresses.

* tree-ssa-loop-ivopts.cc (record_group_use): Use
split_constant_offset.

13 months agoHide IVOPTs strip_offset
Richard Biener [Wed, 21 Jun 2023 09:12:36 +0000 (11:12 +0200)]
Hide IVOPTs strip_offset

PR110243 shows strip_offset has some correctness issues, the following
avoids using it from loop distribution which can use the more correct
split_constant_offset from data-ref analysis instead.  The patch then
un-exports the function from IVOPTs.

* tree-loop-distribution.cc (classify_builtin_st): Use
split_constant_offset.
* tree-ssa-loop-ivopts.h (strip_offset): Remove.
* tree-ssa-loop-ivopts.cc (strip_offset): Make static.

13 months agoaarch64: Convert SVE gather patterns to compact syntax
Kyrylo Tkachov [Wed, 21 Jun 2023 11:03:22 +0000 (12:03 +0100)]
aarch64: Convert SVE gather patterns to compact syntax

This patch converts the SVE load gather patterns to the new compact syntax
that Tamar introduced. This allows for a future patch I want to contribute
to add more alternatives that are better viewed in the more compact form.

The lines in some patterns are >80 long now, but I think that's unavoidable
and those patterns already had overly long constraint strings.

No functional change intended.
Bootstrapped and tested on aarch64-none-linux-gnu.

gcc/ChangeLog:

* config/aarch64/aarch64-sve.md (mask_gather_load<mode><v_int_container>):
Convert to compact alternatives syntax.
(mask_gather_load<mode><v_int_container>): Likewise.
(*mask_gather_load<mode><v_int_container>_<su>xtw_unpacked): Likewise.
(*mask_gather_load<mode><v_int_container>_sxtw): Likewise.
(*mask_gather_load<mode><v_int_container>_uxtw): Likewise.
(@aarch64_gather_load_<ANY_EXTEND:optab><SVE_4HSI:mode><SVE_4BHI:mode>):
Likewise.
(@aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode><SVE_2BHSI:mode>):
Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_<ANY_EXTEND2:su>xtw_unpacked): Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_sxtw): Likewise.
(*aarch64_gather_load_<ANY_EXTEND:optab><SVE_2HSDI:mode>
<SVE_2BHSI:mode>_uxtw): Likewise.
(@aarch64_ldff1_gather<mode>): Likewise.
(@aarch64_ldff1_gather<mode>): Likewise.
(*aarch64_ldff1_gather<mode>_sxtw): Likewise.
(*aarch64_ldff1_gather<mode>_uxtw): Likewise.
(@aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx4_WIDE:mode>
<VNx4_NARROW:mode>): Likewise.
(@aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>): Likewise.
(*aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>_sxtw): Likewise.
(*aarch64_ldff1_gather_<ANY_EXTEND:optab><VNx2_WIDE:mode>
<VNx2_NARROW:mode>_uxtw): Likewise.
* config/aarch64/aarch64-sve2.md (@aarch64_gather_ldnt<mode>): Likewise.
(@aarch64_gather_ldnt_<ANY_EXTEND:optab><SVE_FULL_SDI:mode>
<SVE_PARTIAL_I:mode>): Likewise.

13 months agodocs: replace backslashchar [PR 110329].
Tamar Christina [Wed, 21 Jun 2023 09:34:54 +0000 (10:34 +0100)]
docs: replace backslashchar [PR 110329].

It seems like @blackslashchar{} is a relatively new addition
to texinfo.  Other parts of the docs use @samp{\} so use it
here too so older distros work.

gcc/ChangeLog:

PR other/110329
* doc/md.texi: Replace backslashchar.

13 months ago[i386] Reject too large vectors for partial vector vectorization
Richard Biener [Mon, 19 Jun 2023 10:28:32 +0000 (12:28 +0200)]
[i386] Reject too large vectors for partial vector vectorization

The following works around the lack of the x86 backend making the
vectorizer compare the costs of the different possible vector
sizes the backed advertises through the vector_modes hook.  When
enabling masked epilogues or main loops then this means we will
select the prefered vector mode which is usually the largest even
for loops that do not iterate close to the times the vector has
lanes.  When not using masking the vectorizer would reject any
mode resulting in a VF bigger than the number of iterations
but with masking they are simply masked out.

So this overloads the finish_cost function and matches for
the problematic case, forcing a high cost to make us try a
smaller vector size.

* config/i386/i386.cc (ix86_vector_costs::finish_cost):
Overload.  For masked main loops make sure the vectorization
factor isn't more than double the number of iterations.

* gcc.target/i386/vect-partial-vectors-1.c: New testcase.
* gcc.target/i386/vect-partial-vectors-2.c: Likewise.

13 months agox86: make VPTERNLOG* usable on less than 512-bit operands with just AVX512F
Jan Beulich [Wed, 21 Jun 2023 06:03:05 +0000 (08:03 +0200)]
x86: make VPTERNLOG* usable on less than 512-bit operands with just AVX512F

There's no reason to constrain this to AVX512VL, unless instructed so by
-mprefer-vector-width=, as the wider operation is unusable for more
narrow operands only when the possible memory source is a non-broadcast
one. This way even the scalar copysign<mode>3 can benefit from the
operation being a single-insn one (leaving aside moves which the
compiler decides to insert for unclear reasons, and leaving aside the
fact that bcst_mem_operand() is too restrictive for broadcast to be
embedded right into VPTERNLOG*).

While there also bring *<avx512>_vternlog<mode>_all's in sync with that
of the three splitters.

Along with this also request value duplication in
ix86_expand_copysign()'s call to ix86_build_signbit_mask(), eliminating
excess space allocation in .rodata.*, filled with zeros which are never
read.

gcc/

* config/i386/i386-expand.cc (ix86_expand_copysign): Request
value duplication by ix86_build_signbit_mask() when AVX512F and
not HFmode.
* config/i386/sse.md (*<avx512>_vternlog<mode>_all): Convert to
2-alternative form. Adjust "mode" attribute. Add "enabled"
attribute.
(*<avx512>_vpternlog<mode>_1): Also permit when TARGET_AVX512F
&& !TARGET_PREFER_AVX256.
(*<avx512>_vpternlog<mode>_2): Likewise.
(*<avx512>_vpternlog<mode>_3): Likewise.

gcc/testsuite/

* gcc.target/i386/avx512f-copysign.c: New test.

13 months agox86: add -mprefer-vector-width=512 to new avx512f-dupv2di.c testcase
Jan Beulich [Wed, 21 Jun 2023 06:01:32 +0000 (08:01 +0200)]
x86: add -mprefer-vector-width=512 to new avx512f-dupv2di.c testcase

This is to cover testing also being done with -march=cascadelake.

gcc/testsuite/

* gcc.target/i386/avx512f-dupv2di.c: Add
-mprefer-vector-width=512.

13 months agoUse intermiediate integer type for float_expr/fix_trunc_expr when direct optab is...
liuhongt [Wed, 31 May 2023 03:20:46 +0000 (11:20 +0800)]
Use intermiediate integer type for float_expr/fix_trunc_expr when direct optab is not existed.

We have already use intermidate type in case WIDEN, but not for NONE,
this patch extended that.

gcc/ChangeLog:

PR target/110018
* tree-vect-stmts.cc (vectorizable_conversion): Use
intermiediate integer type for float_expr/fix_trunc_expr when
direct optab is not existed.

gcc/testsuite/ChangeLog:

* gcc.target/i386/pr110018-1.c: New test.

13 months agoDaily bump.
GCC Administrator [Wed, 21 Jun 2023 00:17:14 +0000 (00:17 +0000)]
Daily bump.

13 months agogensupport: drop suppport for define_cond_exec from compact syntac
Tamar Christina [Tue, 20 Jun 2023 22:31:40 +0000 (23:31 +0100)]
gensupport: drop suppport for define_cond_exec from compact syntac

define_cond_exec does not support the special @@ syntax
and so can't support {@.  As such just remove support
for it.

gcc/ChangeLog:

PR bootstrap/110324
* gensupport.cc (convert_syntax): Explicitly check for RTX code.

13 months agolibcpp: Improve location for macro names [PR66290]
Lewis Hyatt [Wed, 3 Aug 2022 14:46:23 +0000 (10:46 -0400)]
libcpp: Improve location for macro names [PR66290]

When libcpp reports diagnostics whose locus is a macro name (such as for
-Wunused-macros), it uses the location in the cpp_macro object that was
stored by _cpp_new_macro. This is currently set to pfile->directive_line,
which contains the line number only and no column information. This patch
changes the stored location to the src_loc for the token defining the macro
name, which includes the location and range information.

libcpp/ChangeLog:

PR c++/66290
* macro.cc (_cpp_create_definition): Add location argument.
* internal.h (_cpp_create_definition): Adjust prototype.
* directives.cc (do_define): Pass new location argument to
_cpp_create_definition.
(do_undef): Stop passing inferior location to cpp_warning_with_line;
the default from cpp_warning is better.
(cpp_pop_definition): Pass new location argument to
_cpp_create_definition.
* pch.cc (cpp_read_state): Likewise.

gcc/testsuite/ChangeLog:

PR c++/66290
* c-c++-common/cpp/macro-ranges.c: New test.
* c-c++-common/cpp/line-2.c: Adapt to check for column information
on macro-related libcpp warnings.
* c-c++-common/cpp/line-3.c: Likewise.
* c-c++-common/cpp/macro-arg-count-1.c: Likewise.
* c-c++-common/cpp/pr58844-1.c: Likewise.
* c-c++-common/cpp/pr58844-2.c: Likewise.
* c-c++-common/cpp/warning-zero-location.c: Likewise.
* c-c++-common/pragma-diag-14.c: Likewise.
* c-c++-common/pragma-diag-15.c: Likewise.
* g++.dg/modules/macro-2_d.C: Likewise.
* g++.dg/modules/macro-4_d.C: Likewise.
* g++.dg/modules/macro-4_e.C: Likewise.
* g++.dg/spellcheck-macro-ordering.C: Likewise.
* gcc.dg/builtin-redefine.c: Likewise.
* gcc.dg/cpp/Wunused.c: Likewise.
* gcc.dg/cpp/redef2.c: Likewise.
* gcc.dg/cpp/redef3.c: Likewise.
* gcc.dg/cpp/redef4.c: Likewise.
* gcc.dg/cpp/ucnid-11-utf8.c: Likewise.
* gcc.dg/cpp/ucnid-11.c: Likewise.
* gcc.dg/cpp/undef2.c: Likewise.
* gcc.dg/cpp/warn-redefined-2.c: Likewise.
* gcc.dg/cpp/warn-redefined.c: Likewise.
* gcc.dg/cpp/warn-unused-macros-2.c: Likewise.
* gcc.dg/cpp/warn-unused-macros.c: Likewise.

13 months agoaarch64: Fix gcc.target/aarch64/sve/pcs failures
Richard Sandiford [Tue, 20 Jun 2023 20:48:38 +0000 (21:48 +0100)]
aarch64: Fix gcc.target/aarch64/sve/pcs failures

Several gcc.target/aarch64/sve/pcs tests started failing after
6a2e8dcbbd4, because the tests weren't robust against whether
an indirect argument register or the stack pointer was used as
the base for stores.

The patch allows either base register when there is only one
indirect argument.  It disables -fcprop-registers in cases where
there are sometimes multiple indirect arguments, since the name
of the argument register is then an important part of the test.

Disabling -fcprop-registers gives poor final register allocation,
since:

* combine's make_more_copies hack adds extra redundant moves
* code with those moves is not allocated as well as moves without them
* we often rely on -fcprop-registers to clean up the allocation later

The patch therefore disables combine in the same tests as
cprop-registers.

gcc/testsuite/
* gcc.target/aarch64/sve/pcs/args_1.c: Match moves from the stack
pointer to indirect argument registers and allow either to be used
as the base register in subsequent stores.
* gcc.target/aarch64/sve/pcs/args_8.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_2.c: Allow the store of the
indirect argument to happen via the argument register or the
stack pointer.
* gcc.target/aarch64/sve/pcs/args_3.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_4.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_bf16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_f16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_f32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_f64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_s16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_s32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_s64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_s8.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_u16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_u32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_u64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_be_u8.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_bf16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_f16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_f32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_f64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_s16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_s32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_s64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_s8.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_u16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_u32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_u64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_5_le_u8.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_bf16.c: Disable
-fcprop-registers and combine.
* gcc.target/aarch64/sve/pcs/args_6_be_f16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_f32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_f64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_s16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_s32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_s64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_s8.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_u16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_u32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_u64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_be_u8.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_bf16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_f16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_f32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_f64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_s16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_s32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_s64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_s8.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_u16.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_u32.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_u64.c: Likewise.
* gcc.target/aarch64/sve/pcs/args_6_le_u8.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_1.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_f16.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_f32.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_f64.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_s16.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_s32.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_s64.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_s8.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_u16.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_u32.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_u64.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_2_u8.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_3_nosc.c: Likewise.
* gcc.target/aarch64/sve/pcs/varargs_3_sc.c: Likewise.

13 months agoaarch64: Robustify stack tie handling
Richard Sandiford [Tue, 20 Jun 2023 20:48:38 +0000 (21:48 +0100)]
aarch64: Robustify stack tie handling

The SVE handling of stack clash protection copied the stack
pointer to X11 before the probe and set up X11 as the CFA
for unwind purposes:

    /* This is done to provide unwinding information for the stack
       adjustments we're about to do, however to prevent the optimizers
       from removing the R11 move and leaving the CFA note (which would be
       very wrong) we tie the old and new stack pointer together.
       The tie will expand to nothing but the optimizers will not touch
       the instruction.  */
    rtx stack_ptr_copy = gen_rtx_REG (Pmode, STACK_CLASH_SVE_CFA_REGNUM);
    emit_move_insn (stack_ptr_copy, stack_pointer_rtx);
    emit_insn (gen_stack_tie (stack_ptr_copy, stack_pointer_rtx));

    /* We want the CFA independent of the stack pointer for the
       duration of the loop.  */
    add_reg_note (insn, REG_CFA_DEF_CFA, stack_ptr_copy);
    RTX_FRAME_RELATED_P (insn) = 1;

-fcprop-registers is now smart enough to realise that X11 = SP,
replace X11 with SP in the stack tie, and delete the instruction
created above.

This patch tries to prevent that by making stack_tie fussy about
the register numbers.  It fixes failures in
gcc.target/aarch64/sve/pcs/stack_clash*.c.

gcc/
* config/aarch64/aarch64.md (stack_tie): Hard-code the first
register operand to the stack pointer.  Require the second register
operand to have the number specified in a separate const_int operand.
* config/aarch64/aarch64.cc (aarch64_emit_stack_tie): New function.
(aarch64_allocate_and_probe_stack_space): Use it.
(aarch64_expand_prologue, aarch64_expand_epilogue): Likewise.
(aarch64_expand_epilogue): Likewise.

13 months agotree-ssa-math-opts: Small uaddc/usubc pattern matching improvement [PR79173]
Jakub Jelinek [Tue, 20 Jun 2023 18:17:41 +0000 (20:17 +0200)]
tree-ssa-math-opts: Small uaddc/usubc pattern matching improvement [PR79173]

In the following testcase we fail to pattern recognize the least significant
.UADDC call.  The reason is that arg3 in that case is
  _3 = .ADD_OVERFLOW (...);
  _2 = __imag__ _3;
  _1 = _2 != 0;
  arg3 = (unsigned long) _1;
and while before the changes arg3 has a single use in some .ADD_OVERFLOW
later on, we add a .UADDC call next to it (and gsi_remove/gsi_replace only
what is strictly necessary and leave quite a few dead stmts around which
next DCE cleans up) and so it all of sudden isn't used just once, but twice
(.ADD_OVERFLOW and .UADDC) and so uaddc_cast fails.  While we could tweak
uaddc_cast and not require has_single_use in these uses, there is also
no vrp that would figure out that because __imag__ _3 is in [0, 1] range,
it can just use arg3 = __imag__ _3; and drop the comparison and cast.

We already search if either arg2 or arg3 is ultimately set from __imag__
of .{{ADD,SUB}_OVERFLOW,U{ADD,SUB}C} call, so the following patch just
remembers the lhs of __imag__ from that case and uses it later.

2023-06-20  Jakub Jelinek  <jakub@redhat.com>

PR middle-end/79173
* tree-ssa-math-opts.cc (match_uaddc_usubc): Remember lhs of
IMAGPART_EXPR of arg2/arg3 and use that as arg3 if it has the right
type.

* g++.target/i386/pr79173-1.C: New test.

13 months agocalls: Change return type of predicate function from int to bool
Uros Bizjak [Tue, 20 Jun 2023 17:42:21 +0000 (19:42 +0200)]
calls: Change return type of predicate function from int to bool

Also change some internal variables and some function arguments to bool.

gcc/ChangeLog:

* calls.h (setjmp_call_p): Change return type from int to bool.
* calls.cc (struct arg_data): Change "pass_on_stack" to bool.
(store_one_arg): Change return type from int to bool
and adjust function body accordingly.  Change "sibcall_failure"
variable to bool.
(finalize_must_preallocate): Ditto.  Change *must_preallocate pointer
argument  to bool.  Change "partial_seen" variable to bool.
(load_register_parameters):  Change *sibcall_failure
pointer argument to bool.
(check_sibcall_argument_overlap_1): Change return type from int to bool
and adjust function body accordingly.
(check_sibcall_argument_overlap):  Ditto.  Change
"mark_stored_args_map" argument to bool.
(emit_call_1): Change "already_popped" variable to bool.
(setjmp_call_p): Change return type from int to bool
and adjust function body accordingly.
(initialize_argument_information): Change *must_preallocate
pointer argument to bool.
(expand_call): Change "pcc_struct_value", "must_preallocate"
and "sibcall_failure" variables to bool.
(emit_library_call_value_1): Change "pcc_struct_value"
variable to bool.

13 months agoruntime: use a C function to call mmap
Ian Lance Taylor [Mon, 19 Jun 2023 21:57:54 +0000 (14:57 -0700)]
runtime: use a C function to call mmap

The final argument to mmap, of type off_t, varies.
In CL 445375 we changed it to always use the C off_t type,
but that broke 32-bit big-endian Linux systems.  On those systems,
using the C off_t type requires calling the mmap64 function.
In C this is automatically handled by the <sys/mman.h> file.
In Go, we would have to change the magic //extern comment to
call mmap64 when appropriate.  Rather than try to get that right,
we instead go through a C function that uses C implicit type
conversions to pick the right type.

Fixes PR go/110297

Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/504415

13 months agoipa-sra: Disable candidates with no known callers (PR 110276)
Martin Jambor [Tue, 20 Jun 2023 16:15:22 +0000 (18:15 +0200)]
ipa-sra: Disable candidates with no known callers (PR 110276)

In IPA-SRA we use can_be_local_p () predicate rather than just plain
local call graph flag in order to figure out whether the node is a
part of an external API that we cannot change.  Although there are
cases where this can allow more transformations, it also means we can
analyze functions which have no callers at all, which is pointless.

Moreover, it makes an assert of hint propagation trigger, which checks
that we have looked at callers before processing hints that come from
them.  This has been reported as PR 110276.

This patch simply adds a check that a node has at least one caller
into the early checks and makes the node a non-candidate for any
transformation if it does not.

gcc/ChangeLog:

2023-06-16  Martin Jambor  <mjambor@suse.cz>

PR ipa/110276
* ipa-sra.cc (struct caller_issues): New field there_is_one.
(check_for_caller_issues): Set it.
(check_all_callers_for_issues): Check it.

gcc/testsuite/ChangeLog:

2023-06-16  Martin Jambor  <mjambor@suse.cz>

PR ipa/110276
* gcc.dg/ipa/pr110276.c: New test.

13 months agoipa-cp: Avoid long linear searches through DECL_ARGUMENTS
Martin Jambor [Tue, 20 Jun 2023 16:15:22 +0000 (18:15 +0200)]
ipa-cp: Avoid long linear searches through DECL_ARGUMENTS

There have been concerns that linear searches through DECL_ARGUMENTS
that are often necessary to compute the index of a particular
PARM_DECL which is the key to results of IPA-CP can happen often
enough to be a compile time issue, especially if we plug the results
into value numbering, as I intend to do with a follow-up patch.

This patch creates a vector sorted according to PARM_DECLs to do the look-up
for all functions which have some information discovered by IPA-CP and which
have 32 parameters or more.  32 is a hard-wired magical constant here to
capture the trade-off between the memory allocation overhead and length of the
linear search.  I do not think it is worth making it a --param but if people
think it appropriate, I can turn it into one.

gcc/ChangeLog:

2023-05-31  Martin Jambor  <mjambor@suse.cz>

* ipa-prop.h (ipa_uid_to_idx_map_elt): New type.
(struct ipcp_transformation): Rearrange members according to
C++ class coding convention, add m_uid_to_idx,
get_param_index and maybe_create_parm_idx_map.
* ipa-cp.cc (ipcp_transformation::get_param_index): New function.
(compare_uids): Likewise.
(ipcp_transformation::maype_create_parm_idx_map): Likewise.
* ipa-prop.cc (ipcp_get_parm_bits): Use get_param_index.
(ipcp_update_bits): Accept TS as a parameter, assume it is not NULL.
(ipcp_update_vr): Likewise.
(ipcp_transform_function): Call, maybe_create_parm_idx_map of TS, bail
out quickly if empty, pass it to ipcp_update_bits and ipcp_update_vr.

13 months agors6000: Add builtins for IEEE 128-bit floating point values
Carl Love [Tue, 20 Jun 2023 15:40:30 +0000 (11:40 -0400)]
rs6000: Add builtins for IEEE 128-bit floating point values

Add support for the following builtins:

 __vector unsigned long long int scalar_extract_exp_to_vec (__ieee128);
 __vector unsigned __int128 scalar_extract_sig_to_vec (__ieee128);
 __ieee128 scalar_insert_exp (__vector unsigned __int128,
        __vector unsigned long long);

The instructions used in the builtins operate on vector registers.  Thus
the result must be moved to a scalar type.  There is no clean, performant
way to do this.  The user code typically needs the result as a vector
anyway.

gcc/
* config/rs6000/rs6000-builtin.cc (rs6000_expand_builtin):
Rename CODE_FOR_xsxsigqp_tf to CODE_FOR_xsxsigqp_tf_ti.
Rename CODE_FOR_xsxsigqp_kf to CODE_FOR_xsxsigqp_kf_ti.
Rename CCDE_FOR_xsxexpqp_tf to CODE_FOR_xsxexpqp_tf_di.
Rename CODE_FOR_xsxexpqp_kf to CODE_FOR_xsxexpqp_kf_di.
(CODE_FOR_xsxexpqp_kf_v2di, CODE_FOR_xsxsigqp_kf_v1ti,
CODE_FOR_xsiexpqp_kf_v2di): Add case statements.
* config/rs6000/rs6000-builtins.def
(__builtin_vsx_scalar_extract_exp_to_vec,
__builtin_vsx_scalar_extract_sig_to_vec,
__builtin_vsx_scalar_insert_exp_vqp): Add new builtin definitions.
Rename xsxexpqp_kf, xsxsigqp_kf, xsiexpqp_kf to xsexpqp_kf_di,
xsxsigqp_kf_ti, xsiexpqp_kf_di respectively.
* config/rs6000/rs6000-c.cc (altivec_resolve_overloaded_builtin):
Update case RS6000_OVLD_VEC_VSIE to handle MODE_VECTOR_INT for new
overloaded instance. Update comments.
* config/rs6000/rs6000-overload.def
(__builtin_vec_scalar_insert_exp): Add new overload definition with
vector arguments.
(scalar_extract_exp_to_vec, scalar_extract_sig_to_vec): New
overloaded definitions.
* config/rs6000/vsx.md (V2DI_DI): New mode iterator.
(DI_to_TI): New mode attribute.
Rename xsxexpqp_<mode> to sxexpqp_<IEEE128:mode>_<V2DI_DI:mode>.
Rename xsxsigqp_<mode> to xsxsigqp_<IEEE128:mode>_<VEC_TI:mode>.
Rename xsiexpqp_<mode> to xsiexpqp_<IEEE128:mode>_<V2DI_DI:mode>.
* doc/extend.texi (scalar_extract_exp_to_vec,
scalar_extract_sig_to_vec): Add documentation for new builtins.
(scalar_insert_exp): Add new overloaded builtin definition.

gcc/testsuite/
* gcc.target/powerpc/bfp/scalar-extract-exp-8.c: New test case.
* gcc.target/powerpc/bfp/scalar-extract-sig-8.c: New test case.
* gcc.target/powerpc/bfp/scalar-insert-exp-16.c: New test case.

13 months agolibstdc++: Remove redundant code in std::to_array
Jonathan Wakely [Tue, 20 Jun 2023 08:27:04 +0000 (09:27 +0100)]
libstdc++: Remove redundant code in std::to_array

libstdc++-v3/ChangeLog:

* include/std/array (to_array(T(&)[N])): Remove redundant
condition.
(to_array(T(&&)[N])): Remove redundant std::move.

13 months agoRISC-V: testsuite: Add missing -mabi=lp64d.
Robin Dapp [Tue, 20 Jun 2023 13:52:16 +0000 (15:52 +0200)]
RISC-V: testsuite: Add missing -mabi=lp64d.

This fixes more cases of missing -mabi=lp64d.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/autovec/vls-vlmax/full-vec-move1.c: Add
-mabi=lp64d.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-1.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-2.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-3.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-4.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-1.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-2.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-3.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-4.c: Dito.

13 months agoRISC-V: Set the natural size of constant vector mask modes to one RVV data vector.
Li Xu [Tue, 20 Jun 2023 06:46:18 +0000 (06:46 +0000)]
RISC-V: Set the natural size of constant vector mask modes to one RVV data vector.

If reinterpret vnx2bi as vnx16qi, vnx16qi must occupy no more of the underlying
registers than vnx2bi.

Consider this following case:
void test_vreinterpret_v_b64_i8m1 (uint8_t *in, int8_t *out)
{
  vbool64_t vmask = __riscv_vlm_v_b64 (in, 2);
  vint8m1_t vout = __riscv_vreinterpret_v_b64_i8m1 (vmask);
  __riscv_vse8_v_i8m1(out, vout, 16);
}

compiler parameters: -march=rv64gcv -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -O3
Compilation fails with:
test_vreinterpret_v_b64_i8m1during RTL pass: expand

test.c: In function 'test_vreinterpret_v_b64_i8m1':
test.c:11:22: internal compiler error: in gen_lowpart_general, at rtlhooks.cc:57
   11 |     vint8m1_t vout = __riscv_vreinterpret_v_b64_i8m1(src);
      |                      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
0xf11876 gen_lowpart_general(machine_mode, rtx_def*)
        ../.././riscv-gcc/gcc/rtlhooks.cc:57
0x191435e gen_vreinterpretvnx16qi(rtx_def*, rtx_def*)
        ../.././riscv-gcc/gcc/config/riscv/vector.md:486
0xe08858 maybe_expand_insn(insn_code, unsigned int, expand_operand*)
        ../.././riscv-gcc/gcc/optabs.cc:8213
0x1471209 riscv_vector::function_expander::generate_insn(insn_code)
        ../.././riscv-gcc/gcc/config/riscv/riscv-vector-builtins.cc:3813
0x147629c riscv_vector::function_expander::expand()
        ../.././riscv-gcc/gcc/config/riscv/riscv-vector-builtins.h:520
0x147629c riscv_vector::expand_builtin(unsigned int, tree_node*, rtx_def*)
        ../.././riscv-gcc/gcc/config/riscv/riscv-vector-builtins.cc:4103
0x9868f9 expand_builtin(tree_node*, rtx_def*, rtx_def*, machine_mode, int)
        ../.././riscv-gcc/gcc/builtins.cc:7342

gcc/ChangeLog:

* config/riscv/riscv.cc (riscv_regmode_natural_size): set the natural
size of vector mask mode to one rvv register.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/autovec/vreinterpet-fixed.c: New test.

13 months agoRISC-V: Optimize codegen of VLA SLP
Juzhe-Zhong [Tue, 20 Jun 2023 09:00:31 +0000 (17:00 +0800)]
RISC-V: Optimize codegen of VLA SLP

Add comments for Robin:
We want to create a pattern where value[ix] = floor (ix / NPATTERNS).
As NPATTERNS is always a power of two we can rewrite this as
= ix & -NPATTERNS.
`
Recently, I figure out a better approach in case of codegen for VLA stepped vector.

Here is the detail descriptions:

Case 1:
void
f (uint8_t *restrict a, uint8_t *restrict b)
{
  for (int i = 0; i < 100; ++i)
    {
      a[i * 8] = b[i * 8 + 37] + 1;
      a[i * 8 + 1] = b[i * 8 + 37] + 2;
      a[i * 8 + 2] = b[i * 8 + 37] + 3;
      a[i * 8 + 3] = b[i * 8 + 37] + 4;
      a[i * 8 + 4] = b[i * 8 + 37] + 5;
      a[i * 8 + 5] = b[i * 8 + 37] + 6;
      a[i * 8 + 6] = b[i * 8 + 37] + 7;
      a[i * 8 + 7] = b[i * 8 + 37] + 8;
    }
}

We need to generate the stepped vector:
NPATTERNS = 8.
{ 0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 8, 8, 8, 8 }

Before this patch:
vid.v    v4         ;; {0,1,2,3,4,5,6,7,...}
vsrl.vi  v4,v4,3    ;; {0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,...}
li       a3,8       ;; {8}
vmul.vx  v4,v4,a3   ;; {0,0,0,0,0,0,0,8,8,8,8,8,8,8,8,...}

After this patch:
vid.v    v4                    ;; {0,1,2,3,4,5,6,7,...}
vand.vi  v4,v4,-8(-NPATTERNS)  ;; {0,0,0,0,0,0,0,8,8,8,8,8,8,8,8,...}

Case 2:
void
f (uint8_t *restrict a, uint8_t *restrict b)
{
  for (int i = 0; i < 100; ++i)
    {
      a[i * 8] = b[i * 8 + 3] + 1;
      a[i * 8 + 1] = b[i * 8 + 2] + 2;
      a[i * 8 + 2] = b[i * 8 + 1] + 3;
      a[i * 8 + 3] = b[i * 8 + 0] + 4;
      a[i * 8 + 4] = b[i * 8 + 7] + 5;
      a[i * 8 + 5] = b[i * 8 + 6] + 6;
      a[i * 8 + 6] = b[i * 8 + 5] + 7;
      a[i * 8 + 7] = b[i * 8 + 4] + 8;
    }
}

We need to generate the stepped vector:
NPATTERNS = 4.
{ 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12, ... }

Before this patch:
li       a6,134221824
slli     a6,a6,5
addi     a6,a6,3        ;; 64-bit: 0x0003000200010000
vmv.v.x  v6,a6          ;; {3, 2, 1, 0, ... }
vid.v    v4             ;; {0, 1, 2, 3, 4, 5, 6, 7, ... }
vsrl.vi  v4,v4,2        ;; {0, 0, 0, 0, 1, 1, 1, 1, ... }
li       a3,4           ;; {4}
vmul.vx  v4,v4,a3       ;; {0, 0, 0, 0, 4, 4, 4, 4, ... }
vadd.vv  v4,v4,v6       ;; {3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12, ... }

After this patch:
li a3,-536875008
slli a3,a3,4
addi a3,a3,1
slli a3,a3,16
vmv.v.x v2,a3           ;; {3, 1, -1, -3, ... }
vid.v v4              ;; {0, 1, 2, 3, 4, 5, 6, 7, ... }
vadd.vv v4,v4,v2        ;; {3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12, ... }

gcc/ChangeLog:

* config/riscv/riscv-v.cc (expand_const_vector): Optimize codegen.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/autovec/partial/slp-1.c: Adapt testcase.
* gcc.target/riscv/rvv/autovec/partial/slp-16.c: New test.
* gcc.target/riscv/rvv/autovec/partial/slp_run-16.c: New test.

13 months agoRISC-V: testsuite: Add -Wno-psabi to vec_set/vec_extract testcases.
Robin Dapp [Tue, 20 Jun 2023 11:58:32 +0000 (13:58 +0200)]
RISC-V: testsuite: Add -Wno-psabi to vec_set/vec_extract testcases.

This fixes some fallout from the recent psabi changes.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-1.c: Add
-Wno-psabi.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-2.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-3.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-4.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_extract-run.c:
Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-1.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-2.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-3.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-4.c: Dito.
* gcc.target/riscv/rvv/autovec/vls-vlmax/vec_set-run.c: Dito.

13 months agoRISC-V: Fix compiler warning of riscv_arg_has_vector
Lehua Ding [Tue, 20 Jun 2023 09:45:17 +0000 (17:45 +0800)]
RISC-V: Fix compiler warning of riscv_arg_has_vector

Hi,

This little patch fixes a compile warning issue that my previous patch
introduced, sorry for introducing this issue.

Best,
Lehua

gcc/ChangeLog:

* config/riscv/riscv.cc (riscv_arg_has_vector): Add default
switch handler.

13 months agoRISC-V: testsuite: Fix vmul test expectation and fix -ffast-math.
Robin Dapp [Tue, 20 Jun 2023 07:12:41 +0000 (09:12 +0200)]
RISC-V: testsuite: Fix vmul test expectation and fix -ffast-math.

I forgot to check for vfmul in the multiplication tests as well as
some -ffast-math arguments.  Fix this.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/autovec/binop/vadd-run.c: Add
-ffast-math.
* gcc.target/riscv/rvv/autovec/binop/vadd-zvfh-run.c: Dito.
* gcc.target/riscv/rvv/autovec/binop/vrem-rv32gcv.c: Remove
-ffast-math
* gcc.target/riscv/rvv/autovec/binop/vmul-rv32gcv.c: Check for
vfmul.
* gcc.target/riscv/rvv/autovec/binop/vmul-rv64gcv.c: Dito.

13 months agoFortran: Fix parse-dump-tree for OpenMP ALLOCATE clause
Tobias Burnus [Tue, 20 Jun 2023 11:46:11 +0000 (13:46 +0200)]
Fortran: Fix parse-dump-tree for OpenMP ALLOCATE clause

Commit r14-1301-gd64e8e1224708e added u2.allocator to gfc_omp_namelist
for better readability and to permit to use namelist->expr for code
like the following:
  !$omp allocators allocate(align(32) : dt%alloc_comp)
    allocate (dt%alloc_comp(5))
  !$omp allocate(dt%alloc_comp2) align(64)
    allocate (dt%alloc_comp2(10))
However, for the parse-tree dump the change was incomplete.

gcc/fortran/ChangeLog:

* dump-parse-tree.cc (show_omp_namelist): Fix dump of the allocator
modifier of OMP_LIST_ALLOCATE.

13 months agoada: Minor tweaks
Eric Botcazou [Fri, 9 Jun 2023 18:24:41 +0000 (20:24 +0200)]
ada: Minor tweaks

gcc/ada/

* gcc-interface/decl.cc (gnat_to_gnu_entity) <E_Variable>: Pass
the NULL_TREE explicitly and test imported_p in lieu of
Is_Imported. <E_Function>: Remove public_flag local variable and
make extern_flag local variable a constant.

13 months agoada: Fix crash on inlining in GNATprove
Yannick Moy [Thu, 8 Jun 2023 14:52:24 +0000 (16:52 +0200)]
ada: Fix crash on inlining in GNATprove

After the recent change on detection of non-inlining, calls inside
the iterator part of a quantified expression were not considered
as preventing inlining anymore, leading to a crash later on inside
GNATprove. Now fixed.

gcc/ada/

* sem_res.adb (Resolve_Call): Fix change that replaced test for
quantified expressions by the test for potentially unevaluated
contexts. Both should be performed.

13 months agoada: Further fixes to handling of private views in instances
Eric Botcazou [Thu, 1 Jun 2023 11:25:53 +0000 (13:25 +0200)]
ada: Further fixes to handling of private views in instances

This removes more bypasses for private views in instances that are present
in type predicates (Conforming_Types, Covers, Specific_Type and Wrong_Type),
which in exchange requires additional work in Sem_Ch12 to restore the proper
view of types during the instantiation of generic bodies.

The main mechanism for this is the Has_Private_View flag, but it comes with
the limitations that 1) there must be a direct reference to the global type
in the generic construct (either a reference to a global object of this type
or the explicit declaration of a local object of this type), which is not
always the case e.g. for loop parameters and 2) it can deal with a single
type at a time, e.g. it cannot deal with an array type and its component
type if their respective views are not the same in the instance.

To overcome the second limitation, a new Has_Secondary_Private_View flag
is introduced to deal with a secondary type, which as of this writing is
either the component type of an array type or the designated type of an
access type (together they make up the vast majority of the problematic
cases for the Has_Private_View flag alone). This new mechanism subsumes
a specific treatment for them that was added in Copy_Generic_Node a few
years ago, although a specific treatment still needs to be preserved for
comparison and equality operators in a narrower case.

Additional handling is also introduced to overcome the first limitation
for loop parameters in Copy_Generic_Node, and a relaxed condition is used
in Exp_Ch7.Convert_View to generate an unchecked conversion between views.

gcc/ada/

* exp_ch7.adb (Convert_View): Detect more cases of mismatches for
private types and use Implementation_Base_Type as main criterion.
* gen_il-fields.ads (Opt_Field_Enum): Add
Has_Secondary_Private_View
* gen_il-gen-gen_nodes.adb (N_Expanded_Name): Likewise.
(N_Direct_Name): Likewise.
(N_Op): Likewise.
* sem_ch12.ads (Check_Private_View): Document the usage of second
flag Has_Secondary_Private_View.
* sem_ch12.adb (Get_Associated_Entity): New function to retrieve
the ultimate associated entity, if any.
(Check_Private_View): Implement Has_Secondary_Private_View
support.
(Copy_Generic_Node): Remove specific treatment for Component_Type
of an array type and Designated_Type of an access type. Add
specific treatment for comparison and equality operators, as well
as iterator and loop parameter specifications.
(Instantiate_Type): Implement Has_Secondary_Private_View support.
(Requires_Delayed_Save): Call Get_Associated_Entity.
(Set_Global_Type): Implement Has_Secondary_Private_View support.
* sem_ch6.adb (Conforming_Types): Remove bypass for private views
in instances.
* sem_type.adb (Covers): Return true if Is_Subtype_Of does so.
Remove bypass for private views in instances.
(Specific_Type): Likewise.
* sem_util.adb (Wrong_Type): Likewise.
* sinfo.ads (Has_Secondary_Private_View): Document new flag.

13 months agoada: Remove outdated comment
Ronan Desplanques [Fri, 9 Jun 2023 12:31:00 +0000 (14:31 +0200)]
ada: Remove outdated comment

The Preelaborate pragma the removed comment was referring to was
indeed present in AI 167, as well as in clause 5.3 of the rationale
for Ada 2012, but it never made it into the 2012 version of the
reference manual.

gcc/ada/

* libgnarl/s-mudido.ads: Remove outdated comment.

13 months agoFortran's gfc_match_char: %S to match symbol with host_assoc
Tobias Burnus [Tue, 20 Jun 2023 11:23:40 +0000 (13:23 +0200)]
Fortran's gfc_match_char: %S to match symbol with host_assoc

gfc_match ("... %s ...", ...) matches a gfc_symbol but with
host_assoc = 0. This commit adds '%S' as variant which matches
with host_assoc = 1

gcc/fortran/ChangeLog:

* match.cc (gfc_match_char): Match with '%S' a symbol
with host_assoc = 1.

13 months agoImprove DSE to handle stores before __builtin_unreachable ()
Richard Biener [Mon, 19 Jun 2023 12:19:47 +0000 (14:19 +0200)]
Improve DSE to handle stores before __builtin_unreachable ()

DSE isn't good at identifying program points that end lifetime
of variables that are not associated with virtual operands.  But
at least for those that end basic-blocks we can handle the simple
case where this ending is in the same basic-block as the definition
we want to elide.  That should catch quite some common cases already.

* tree-ssa-dse.cc (dse_classify_store): When we found
no defs and the basic-block with the original definition
ends in __builtin_unreachable[_trap] the store is dead.

* gcc.dg/tree-ssa/ssa-dse-47.c: New testcase.
* c-c++-common/asan/pr106558.c: Avoid undefined behavior
due to missing return.

13 months agoUpdate virtual SSA form manually where easily possible in phiprop
Richard Biener [Tue, 20 Jun 2023 07:51:40 +0000 (09:51 +0200)]
Update virtual SSA form manually where easily possible in phiprop

This keeps virtual SSA form up-to-date in phiprop when easily possible.
Only when we deal with aggregate copies the work would be too
heavy-handed in general.

* tree-ssa-phiprop.cc (phiprop_insert_phi): For simple loads
keep the virtual SSA form up-to-date.

13 months agoaarch64: Optimise ADDP with same source operands
Kyrylo Tkachov [Tue, 20 Jun 2023 10:03:47 +0000 (11:03 +0100)]
aarch64: Optimise ADDP with same source operands

We've been asked to optimise the testcase in this patch of a 64-bit ADDP with
the low and high halves of the same 128-bit vector. This can be done by a
single .4s ADDP followed by just reading the bottom 64 bits. A splitter for
this is quite straightforward now that all the vec_concat stuff is collapsed
by simplify-rtx.

With this patch we generate a single:
addp v0.4s, v0.4s, v0.4s
instead of:
        dup     d31, v0.d[1]
        addp    v0.2s, v0.2s, v31.2s
        ret

Bootstrapped and tested on aarch64-none-linux-gnu and aarch64_be-none-elf.

gcc/ChangeLog:

* config/aarch64/aarch64-simd.md (*aarch64_addp_same_reg<mode>):
New define_insn_and_split.

gcc/testsuite/ChangeLog:

* gcc.target/aarch64/simd/addp-same-low_1.c: New test.

13 months agoAArch64: remove test comment from *mov<mode>_aarch64
Tamar Christina [Tue, 20 Jun 2023 07:54:42 +0000 (08:54 +0100)]
AArch64: remove test comment from *mov<mode>_aarch64

I accidentally left a test comment in the final version of the patch.
This removes the comment.

gcc/ChangeLog:

* config/aarch64/aarch64.md (*mov<mode>_aarch64): Drop test comment.

13 months agoada: Fix couple of issues in documentation of overflow checking
Eric Botcazou [Thu, 1 Jun 2023 22:11:05 +0000 (00:11 +0200)]
ada: Fix couple of issues in documentation of overflow checking

There is still a mention of the defunct CHECKED mode and the Default
Settings paragraph is confusing with regard to the -gnato switch.

gcc/ada/

* doc/gnat_ugn/gnat_and_program_execution.rst (Overflows in GNAT)
<Default Settings>: Remove obsolete paragraph about -gnato.
<Implementation Notes>: Replace CHECKED with STRICT.
* gnat_ugn.texi: Regenerate.

13 months agoada: Do not issue warning on postcondition in some cases
Yannick Moy [Thu, 8 Jun 2023 09:12:25 +0000 (09:12 +0000)]
ada: Do not issue warning on postcondition in some cases

Warning on suspicious postcondition is not relevant if contract
Exceptional_Cases is present, or if contract Always_Terminates is
present with a non-statically True value, as in those cases the
postcondition can be used to indicate constraints on those pre-state
for which the subprogram might terminate normally.

gcc/ada/

* sem_util.adb (Check_Result_And_Post_State): Do not warn in cases
where the warning could be spurious.

13 months agoada: Add the ability to add error codes to error messages
Yannick Moy [Wed, 10 May 2023 14:10:54 +0000 (16:10 +0200)]
ada: Add the ability to add error codes to error messages

Add a new character sequence [] for error codes in error messages
handled by Error_Msg procedures, to use for SPARK-related errors.
Display of additional information on the error or warning based on
the error code is delegated to GNATprove.

gcc/ada/

* err_vars.ads (Error_Msg_Code): New variable for error codes.
* errout.adb (Error_Msg_Internal): Display continuation message
when an error code was present.
(Set_Msg_Text): Handle character sequence [] for error codes.
* errout.ads: Document new insertion sequence [].
(Error_Msg_Code): New renaming.
* erroutc.adb (Prescan_Message): Detect presence of error code.
(Set_Msg_Insertion_Code): Handle new insertion sequence [].
* erroutc.ads (Has_Error_Code): New variable for prescan.
(Set_Msg_Insertion_Code): Handle new insertion sequence [].
* contracts.adb (Check_Type_Or_Object_External_Properties):
Replace reference to SPARK RM section by an error code.
* sem_elab.adb (SPARK_Processor): Same.
* sem_prag.adb (Check_Missing_Part_Of): Same.
* sem_res.adb (Resolve_Actuals, Resolve_Entity_Name): Same.

13 months agoada: Fix for attribute Range in Exceptional_Cases
Piotr Trojanek [Mon, 5 Jun 2023 08:30:39 +0000 (10:30 +0200)]
ada: Fix for attribute Range in Exceptional_Cases

Attribute Range is now handled like First and Last when occurring within
the consequence of Exceptional_Cases, i.e. attribute Range is not
considered to be a read of a formal parameter that would not be allowed
in the contract.

gcc/ada/

* sem_res.adb (Resolve_Entity_Name): Handle Range like First and Last.

13 months agoada: Document partition-wide Ada signal handlers
Jose Ruiz [Tue, 30 May 2023 14:08:38 +0000 (16:08 +0200)]
ada: Document partition-wide Ada signal handlers

Indicate the signal handlers that are set by the Ada
run time, and explain how to prevent them if needed.

gcc/ada/

* doc/gnat_ugn/the_gnat_compilation_model.rst
(Partition-Wide Settings): add this subsection to document
configuration settings made by the Ada run time.
* gnat_ugn.texi: Regenerate.

13 months agoada: Fix for quantified expressions in Exceptional_Cases
Piotr Trojanek [Thu, 1 Jun 2023 07:59:40 +0000 (09:59 +0200)]
ada: Fix for quantified expressions in Exceptional_Cases

When detecting illegal uses of formal parameters of the current
subprogram in contract of its Exceptional_Cases, we relied on the
Current_Scope. However, quantified expressions introduce an implicit
scope, which we need to take into account.

gcc/ada/

* sem_res.adb (Resolve_Entity_Name): Ignore implicit loop scopes
introduced by quantified expressions.

13 months agoada: Fix bug in predicate checks with address clauses
Bob Duff [Wed, 31 May 2023 13:21:44 +0000 (09:21 -0400)]
ada: Fix bug in predicate checks with address clauses

This patch fixes a compiler bug triggered by having a type with some
defaulted components, and a predicate, and an object of that type with
an address clause. In this case, the compiler was crashing.

gcc/ada/

* sem_ch3.adb (Analyze_Object_Declaration): Remove predicate-check
generation if there is an address clause. These are unnecessary,
and cause gigi to crash.
* exp_util.ads (Following_Address_Clause): Remove obsolete "???"
comments. The suggested changes were done long ago.

13 months agoada: Fix fallout of fix to handling of private views in instances
Eric Botcazou [Wed, 31 May 2023 12:32:59 +0000 (14:32 +0200)]
ada: Fix fallout of fix to handling of private views in instances

Check_Actual_Type incorrectly switches the view of a private type declared
in the enclosing scope of a generic unit but that has a private ancestor.

gcc/ada/

* einfo.ads (Has_Private_Ancestor): Fix inaccuracy in description.
* sem_ch12.adb (Check_Actual_Type): Do not switch the view of the
type if it has a private ancestor.

This page took 0.138983 seconds and 5 git commands to generate.