]> git.ipfire.org Git - thirdparty/gcc.git/log
thirdparty/gcc.git
12 months agoc++/coroutines: correct passing *this to promise type [PR104981]
Patrick Palka [Tue, 23 Jul 2024 01:30:49 +0000 (21:30 -0400)] 
c++/coroutines: correct passing *this to promise type [PR104981]

When passing *this to the promise type ctor (or to its operator new)
(as per [dcl.fct.def.coroutine]/4), we add an explicit cast to lvalue
reference.  But this is unnecessary since *this is already always an
lvalue.  And doing so means we need to call convert_from_reference
afterward to lower the reference expression to an implicit dereference,
which we're currently neglecting to do and which causes overload
resolution to get confused when computing argument conversions.

So this patch removes this unneeded reference cast when passing *this
to the promise ctor, and removes both the cast and implicit deref when
passing *this to operator new, for consistency.  While we're here, use
cp_build_fold_indirect_ref instead of directly building INDIRECT_REF.

PR c++/104981
PR c++/115550

gcc/cp/ChangeLog:

* coroutines.cc (morph_fn_to_coro): Remove unneeded calls
to convert_to_reference and convert_from_reference when
passing *this.  Use cp_build_fold_indirect_ref instead
of directly building INDIRECT_REF.

gcc/testsuite/ChangeLog:

* g++.dg/coroutines/pr104981-preview-this.C: New test.
* g++.dg/coroutines/pr115550-preview-this.C: New test.

Reviewed-by: Iain Sandoe <iain@sandoe.co.uk>
Reviewed-by: Jason Merrill <jason@redhat.com>
12 months agoRISC-V: Implement the .SAT_TRUNC for scalar
Pan Li [Mon, 1 Jul 2024 08:36:35 +0000 (16:36 +0800)] 
RISC-V: Implement the .SAT_TRUNC for scalar

This patch would like to implement the simple .SAT_TRUNC pattern
in the riscv backend. Aka:

Form 1:
  #define DEF_SAT_U_TRUC_FMT_1(NT, WT)     \
  NT __attribute__((noinline))             \
  sat_u_truc_##WT##_to_##NT##_fmt_1 (WT x) \
  {                                        \
    bool overflow = x > (WT)(NT)(-1);      \
    return ((NT)x) | (NT)-overflow;        \
  }

DEF_SAT_U_TRUC_FMT_1(uint32_t, uint64_t)

Before this patch:
__attribute__((noinline))
uint8_t sat_u_truc_uint16_t_to_uint8_t_fmt_1 (uint16_t x)
{
  _Bool overflow;
  unsigned char _1;
  unsigned char _2;
  unsigned char _3;
  uint8_t _6;

;;   basic block 2, loop depth 0
;;    pred:       ENTRY
  overflow_5 = x_4(D) > 255;
  _1 = (unsigned char) x_4(D);
  _2 = (unsigned char) overflow_5;
  _3 = -_2;
  _6 = _1 | _3;
  return _6;
;;    succ:       EXIT

}

After this patch:
__attribute__((noinline))
uint8_t sat_u_truc_uint16_t_to_uint8_t_fmt_1 (uint16_t x)
{
  uint8_t _6;

;;   basic block 2, loop depth 0
;;    pred:       ENTRY
  _6 = .SAT_TRUNC (x_4(D)); [tail call]
  return _6;
;;    succ:       EXIT

}

The below tests suites are passed for this patch
1. The rv64gcv fully regression test.
2. The rv64gcv build with glibc

gcc/ChangeLog:

* config/riscv/iterators.md (ANYI_DOUBLE_TRUNC): Add new iterator
for int double truncation.
(ANYI_DOUBLE_TRUNCATED): Add new attr for int double truncation.
(anyi_double_truncated): Ditto but for lowercase.
* config/riscv/riscv-protos.h (riscv_expand_ustrunc): Add new
func decl for expanding ustrunc
* config/riscv/riscv.cc (riscv_expand_ustrunc): Add new func
impl to expand ustrunc.
* config/riscv/riscv.md (ustrunc<mode><anyi_double_truncated>2): Impl
the new pattern ustrunc<m><n>2 for int.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/sat_arith.h: Add test helper macro.
* gcc.target/riscv/sat_arith_data.h: New test.
* gcc.target/riscv/sat_u_trunc-1.c: New test.
* gcc.target/riscv/sat_u_trunc-2.c: New test.
* gcc.target/riscv/sat_u_trunc-3.c: New test.
* gcc.target/riscv/sat_u_trunc-run-1.c: New test.
* gcc.target/riscv/sat_u_trunc-run-2.c: New test.
* gcc.target/riscv/sat_u_trunc-run-3.c: New test.
* gcc.target/riscv/scalar_sat_unary.h: New test.

Signed-off-by: Pan Li <pan2.li@intel.com>
12 months agoDaily bump.
GCC Administrator [Tue, 23 Jul 2024 00:19:00 +0000 (00:19 +0000)] 
Daily bump.

12 months agoFix handling of ICF_NOVOPS in ipa-modref
Jan Hubicka [Mon, 22 Jul 2024 21:01:50 +0000 (23:01 +0200)] 
Fix handling of ICF_NOVOPS in ipa-modref

As shown in somewhat convoluted testcase, ipa-modref is mistreating
ECF_NOVOPS as "having no side effects".  This come from time when
modref cared only about memory accesses and thus it was possible to
shortcut on it.

This patch removes (hopefully) all those bad shortcuts.
Bootstrapped/regtested x86_64-linux, comitted.

gcc/ChangeLog:

PR ipa/109985

* ipa-modref.cc (modref_summary::useful_p): Fix handling of ECF_NOVOPS.
(modref_access_analysis::process_fnspec): Likevise.
(modref_access_analysis::analyze_call): Likevise.
(propagate_unknown_call): Likevise.
(modref_propagate_in_scc): Likevise.
(modref_propagate_flags_in_scc): Likewise.
(ipa_merge_modref_summary_after_inlining): Likewise.

12 months agoc++: Some cp-tree.def comment fixes
Jakub Jelinek [Mon, 22 Jul 2024 17:47:17 +0000 (19:47 +0200)] 
c++: Some cp-tree.def comment fixes

While reading the fold expression and concept tree comments, I found
various spots referring to non-existent macros etc.

The following patch attempts to sync that with what is actually implemented.

2024-07-22  Jakub Jelinek  <jakub@redhat.com>

* cp-tree.def (UNARY_LEFT_FOLD_EXPR): Use FOLD_EXPR_MODIFY_P instead
of FOLD_EXPR_MOD_P or FOLDEXPR_MOD_P in the comment.  Comment
formatting fixes.
(ATOMIC_CONSTEXPR): Use CONSTR_INFO instead of ATOMIC_CONSTR_INFO
and ATOMIC_CONSTR_MAP instead of ATOMIC_CONSTR_PARMS in the comment.
Comment formatting fixes.
(CONJ_CONSTR): Remove comment about third operand.  Use CONSTR_INFO
instead of CONJ_CONSTR_INFO and DISJ_CONSTR_INFO.
(CHECK_CONSTR): Use CHECK_CONSTR_ARGS instead of
CHECK_CONSTR_ARGUMENTS.

12 months agoFix modref's iteraction with store merging
Jan Hubicka [Mon, 22 Jul 2024 17:00:39 +0000 (19:00 +0200)] 
Fix modref's iteraction with store merging

Hi,
this patch fixes wrong code in case store-merging introduces load of function
parameter that was previously write-only (which happens for bitfields).
Without this, the whole store-merged area is consdered to be killed.

PR ipa/111613

gcc/ChangeLog:

* ipa-modref.cc (analyze_parms): Do not preserve EAF_NO_DIRECT_READ and
EAF_NO_INDIRECT_READ from past flags.

gcc/testsuite/ChangeLog:

* gcc.c-torture/pr111613.c: New test.

12 months agoAdd -mcpu=power11 support.
Michael Meissner [Mon, 22 Jul 2024 16:20:43 +0000 (12:20 -0400)] 
Add -mcpu=power11 support.

This patch adds the power11 option to the -mcpu= and -mtune= switches.

This patch treats the power11 like a power10 in terms of costs and reassociation
width.

This patch issues a ".machine power11" to the assembly file if you use
-mcpu=power11.

This patch defines _ARCH_PWR11 if the user uses -mcpu=power11.

This patch allows GCC to be configured with the --with-cpu=power11 and
--with-tune=power11 options.

This patch passes -mpwr11 to the assembler if the user uses -mcpu=power11.

This patch adds support for using "power11" in the __builtin_cpu_is built-in
function.

2024-07-22  Michael Meissner  <meissner@linux.ibm.com>

gcc/

* config.gcc (powerpc*-*-*): Add support for power11.
* config/rs6000/aix71.h (ASM_CPU_SPEC): Add support for -mcpu=power11.
* config/rs6000/aix72.h (ASM_CPU_SPEC): Likewise.
* config/rs6000/aix73.h (ASM_CPU_SPEC): Likewise.
* config/rs6000/driver-rs6000.cc (asm_names): Likewise.
* config/rs6000/ppc-auxv.h (PPC_PLATFORM_POWER11): New define.
* config/rs6000/rs6000-builtin.cc (cpu_is_info): Add power11.
* config/rs6000/rs6000-c.cc (rs6000_target_modify_macros): Define
_ARCH_PWR11 if -mcpu=power11.
* config/rs6000/rs6000-cpus.def (POWER11_MASKS_SERVER): New define.
(POWERPC_MASKS): Add power11.
(power11 cpu): Add power11 definition.
* config/rs6000/rs6000-opts.h (PROCESSOR_POWER11): Add power11 processor.
* config/rs6000/rs6000-string.cc (expand_compare_loop): Likewise.
* config/rs6000/rs6000-tables.opt: Regenerate.
* config/rs6000/rs6000.cc (rs6000_option_override_internal): Add power11
support.
(rs6000_machine_from_flags): Likewise.
(rs6000_reassociation_width): Likewise.
(rs6000_adjust_cost): Likewise.
(rs6000_issue_rate): Likewise.
(rs6000_sched_reorder): Likewise.
(rs6000_sched_reorder2): Likewise.
(rs6000_register_move_cost): Likewise.
(rs6000_opt_masks): Likewise.
* config/rs6000/rs6000.h (ASM_CPU_SPEC): Likewise.
* config/rs6000/rs6000.md (cpu attribute): Add power11.
* config/rs6000/rs6000.opt (-mpower11): Add internal power11 flag.
* doc/invoke.texi (RS/6000 and PowerPC Options): Document -mcpu=power11.
* config/rs6000/power10.md (all reservations): Add power11 support.

gcc/testsuite/

* gcc.target/powerpc/power11-1.c: New test.
* gcc.target/powerpc/power11-2.c: Likewise.
* gcc.target/powerpc/power11-3.c: Likewise.

12 months ago[4/n][PR rtl-optimization/115877] Correct SUBREG handling in a destination
Jeff Law [Mon, 22 Jul 2024 16:11:57 +0000 (10:11 -0600)] 
[4/n][PR rtl-optimization/115877] Correct SUBREG handling in a destination

If we encounter something during SET handling that we can not handle, the safe
thing to do is to ignore the destination and continue the loop.

We've actually been trying to do slightly better with SUBREG destinations by
iterating into SUBREG_REG.  It turns out that wasn't working as expected.

The problem is once we "continue" we lose the state that we were inside the SET
and thus we ended up ignoring the destination completely rather than tracking
the SUBREG_REG object.  This could be fixed by restarting SET processing, but I
just don't see this as all that important to handle.  So rather than leave the
code as-is, not working per design, I'm twiddling it to use the common 'skip
subrtxs and continue' idiom used elsewhere.

This is a prerequisite for another patch in this series.  Specifically I have a
patch that explicitly tracks if we skipped a destination rather than trying to
imply it from the state of LIVE_TMP.  So this is probably NFC right now, but
that's a short-lived NFC.

Bootstrapped and regression tested on x86 and also run as part of a larger kit
on the crosses in my tester.

PR rtl-optimization/115877
gcc/
* ext-dce.cc (ext_dce_process_sets): More correctly handle SUBREG
destinations.

12 months agoFix modref_eaf_analysis::analyze_ssa_name handling of values dereferenced to function...
Jan Hubicka [Mon, 22 Jul 2024 16:08:08 +0000 (18:08 +0200)] 
Fix modref_eaf_analysis::analyze_ssa_name handling of values dereferenced to function call parameters

modref_eaf_analysis::analyze_ssa_name misinterprets EAF flags.  If dereferenced
parameter is passed (to map_iterator in the testcase) it can be returned
indirectly which in turn makes it to escape into the next function call.

PR ipa/115033

gcc/ChangeLog:

* ipa-modref.cc (modref_eaf_analysis::analyze_ssa_name): Fix checking of
EAF flags when analysing values dereferenced as function parameters.

gcc/testsuite/ChangeLog:

* gcc.c-torture/execute/pr115033.c: New test.

12 months agoFix accounting of offsets in unadjusted_ptr_and_unit_offset
Jan Hubicka [Mon, 22 Jul 2024 16:05:26 +0000 (18:05 +0200)] 
Fix accounting of offsets in unadjusted_ptr_and_unit_offset

unadjusted_ptr_and_unit_offset accidentally throws away the offset computed by
get_addr_base_and_unit_offset. Instead of passing extra_offset it passes offset.

PR ipa/114207

gcc/ChangeLog:

* ipa-prop.cc (unadjusted_ptr_and_unit_offset): Fix accounting of offsets in ADDR_EXPR.

gcc/testsuite/ChangeLog:

* gcc.c-torture/execute/pr114207.c: New test.

12 months agoCompare loop bounds in ipa-icf
Jan Hubicka [Mon, 22 Jul 2024 16:01:57 +0000 (18:01 +0200)] 
Compare loop bounds in ipa-icf

Hi,
this testcase shows another poblem with missing comparators for metadata
in ICF. With value ranges available to loop optimizations during early
opts we can estimate number of iterations based on guarding condition that
can be split away by the fnsplit pass. This patch disables ICF when
number of iteraitons does not match.

Bootstrapped/regtesed x86_64-linux, will commit it shortly

gcc/ChangeLog:

PR ipa/115277
* ipa-icf-gimple.cc (func_checker::compare_loops): compare loop
bounds.

gcc/testsuite/ChangeLog:

* gcc.c-torture/compile/pr115277.c: New test.

12 months agortl-ssa: Avoid using a stale splay tree root [PR116009]
Richard Sandiford [Mon, 22 Jul 2024 15:42:16 +0000 (16:42 +0100)] 
rtl-ssa: Avoid using a stale splay tree root [PR116009]

In the fix for PR115928, I'd failed to notice that "root" was used
later in the function, so needed to be updated.

gcc/
PR rtl-optimization/116009
* rtl-ssa/accesses.cc (function_info::add_def): Set the root
local variable after removing the old clobber group.

gcc/testsuite/
PR rtl-optimization/116009
* gcc.c-torture/compile/pr116009.c: New test.

12 months agortl-ssa: Add debug routines for def_splay_tree
Richard Sandiford [Mon, 22 Jul 2024 15:42:16 +0000 (16:42 +0100)] 
rtl-ssa: Add debug routines for def_splay_tree

This patch adds debug routines for def_splay_tree, which I found
useful while debugging PR116009.

gcc/
* rtl-ssa/accesses.h (rtl_ssa::pp_def_splay_tree): Declare.
(dump, debug): Add overloads for def_splay_tree.
* rtl-ssa/accesses.cc (rtl_ssa::pp_def_splay_tree): New function.
(dump, debug): Add overloads for def_splay_tree.

12 months agoaarch64: Tighten aarch64_simd_mem_operand_p [PR115969]
Richard Sandiford [Mon, 22 Jul 2024 15:42:15 +0000 (16:42 +0100)] 
aarch64: Tighten aarch64_simd_mem_operand_p [PR115969]

aarch64_simd_mem_operand_p checked for a memory with a POST_INC
or REG address, but it didn't check what kind of register was
being used.  This meant that it allowed DImode FPRs as well as GPRs.

I wondered about rewriting it to use aarch64_classify_address,
but this one-line fix seemed simpler.  The structure then mirrors
the existing early exit in aarch64_classify_address itself:

  /* On LE, for AdvSIMD, don't support anything other than POST_INC or
     REG addressing.  */
  if (advsimd_struct_p
      && TARGET_SIMD
      && !BYTES_BIG_ENDIAN
      && (code != POST_INC && code != REG))
    return false;

gcc/
PR target/115969
* config/aarch64/aarch64.cc (aarch64_simd_mem_operand_p): Require
the operand to be a legitimate memory_operand.

gcc/testsuite/
PR target/115969
* gcc.target/aarch64/pr115969.c: New test.

12 months ago[NFC][PR rtl-optimization/115877] Avoid setting irrelevant bit groups as live in...
Jeff Law [Mon, 22 Jul 2024 14:45:10 +0000 (08:45 -0600)] 
[NFC][PR rtl-optimization/115877] Avoid setting irrelevant bit groups as live in ext-dce

Another patch to refine liveness computations.  This should be NFC and is
designed to help debugging.

In simplest terms the patch avoids setting bit groups outside the size of a
pseudo as live.  Consider a HImode pseudo, bits 16..63 for such a pseudo don't
really have meaning, yet we often set bit groups related to bits 16.63 on in
the liveness bitmaps.

This makes debugging harder than it needs to be by simply having larger bitmaps
to verify when walking through the code in a debugger.

This has been bootstrapped and regression tested on x86_64.  It's also been
tested on the crosses in my tester without regressions.

Pushing to the trunk,

PR rtl-optimization/115877
gcc/
* ext-dce.cc (group_limit): New function.
(mark_reg_live): Likewise.
(ext_dce_process_sets): Use new functions.
(ext_dce_process_uses): Likewise.
(ext_dce_init): Likewise.

12 months agoFix hash of WIDEN_*_EXPR
Richard Biener [Mon, 22 Jul 2024 09:07:28 +0000 (11:07 +0200)] 
Fix hash of WIDEN_*_EXPR

We're hashing operand 2 to the temporary hash.

* fold-const.cc (operand_compare::hash_operand): Fix hash
of WIDEN_*_EXPR.

12 months agoconstify inchash
Richard Biener [Mon, 22 Jul 2024 09:09:03 +0000 (11:09 +0200)] 
constify inchash

The following constifies parts of inchash.

* inchash.h (inchash::end): Make const.
(inchash::merge): Take const reference hash argument.
(inchash::add_commutative): Likewise.

12 months agoFix Rejects allocatable coarray passed as a dummy argument [88624]
Andre Vehreschild [Thu, 11 Jul 2024 08:07:12 +0000 (10:07 +0200)] 
Fix Rejects allocatable coarray passed as a dummy argument [88624]

Coarray parameters of procedures/functions need to be dereffed, because
they are references to the descriptor but the routine expected the
descriptor directly.

PR fortran/88624

gcc/fortran/ChangeLog:

* trans-expr.cc (gfc_conv_procedure_call): Treat
pointers/references (e.g. from parameters) correctly by derefing
them.

gcc/testsuite/ChangeLog:

* gfortran.dg/coarray/dummy_1.f90: Add calling function trough
function.
* gfortran.dg/pr88624.f90: New test.

12 months agoAArch64: implement TARGET_VECTORIZE_CONDITIONAL_OPERATION_IS_EXPENSIVE [PR115531].
Tamar Christina [Mon, 22 Jul 2024 09:28:19 +0000 (10:28 +0100)] 
AArch64: implement TARGET_VECTORIZE_CONDITIONAL_OPERATION_IS_EXPENSIVE  [PR115531].

This implements the new target hook indicating that for AArch64 when possible
we prefer masked operations for any type vs doing LOAD + SELECT or
SELECT + STORE.

Thanks,
Tamar

gcc/ChangeLog:

PR tree-optimization/115531
* config/aarch64/aarch64.cc
(aarch64_conditional_operation_is_expensive): New.
(TARGET_VECTORIZE_CONDITIONAL_OPERATION_IS_EXPENSIVE): New.

gcc/testsuite/ChangeLog:

PR tree-optimization/115531
* gcc.dg/vect/vect-conditional_store_1.c: New test.
* gcc.dg/vect/vect-conditional_store_2.c: New test.
* gcc.dg/vect/vect-conditional_store_3.c: New test.
* gcc.dg/vect/vect-conditional_store_4.c: New test.

12 months agomiddle-end: Implement conditonal store vectorizer pattern [PR115531]
Tamar Christina [Mon, 22 Jul 2024 09:26:14 +0000 (10:26 +0100)] 
middle-end: Implement conditonal store vectorizer pattern [PR115531]

This adds a conditional store optimization for the vectorizer as a pattern.
The vectorizer already supports modifying memory accesses because of the pattern
based gather/scatter recognition.

Doing it in the vectorizer allows us to still keep the ability to vectorize such
loops for architectures that don't have MASK_STORE support, whereas doing this
in ifcvt makes us commit to MASK_STORE.

Concretely for this loop:

void foo1 (char *restrict a, int *restrict b, int *restrict c, int n, int stride)
{
  if (stride <= 1)
    return;

  for (int i = 0; i < n; i++)
    {
      int res = c[i];
      int t = b[i+stride];
      if (a[i] != 0)
        res = t;
      c[i] = res;
    }
}

today we generate:

.L3:
        ld1b    z29.s, p7/z, [x0, x5]
        ld1w    z31.s, p7/z, [x2, x5, lsl 2]
        ld1w    z30.s, p7/z, [x1, x5, lsl 2]
        cmpne   p15.b, p6/z, z29.b, #0
        sel     z30.s, p15, z30.s, z31.s
        st1w    z30.s, p7, [x2, x5, lsl 2]
        add     x5, x5, x4
        whilelo p7.s, w5, w3
        b.any   .L3

which in gimple is:

  vect_res_18.9_68 = .MASK_LOAD (vectp_c.7_65, 32B, loop_mask_67);
  vect_t_20.12_74 = .MASK_LOAD (vectp.10_72, 32B, loop_mask_67);
  vect__9.15_77 = .MASK_LOAD (vectp_a.13_75, 8B, loop_mask_67);
  mask__34.16_79 = vect__9.15_77 != { 0, ... };
  vect_res_11.17_80 = VEC_COND_EXPR <mask__34.16_79, vect_t_20.12_74, vect_res_18.9_68>;
  .MASK_STORE (vectp_c.18_81, 32B, loop_mask_67, vect_res_11.17_80);

A MASK_STORE is already conditional, so there's no need to perform the load of
the old values and the VEC_COND_EXPR.  This patch makes it so we generate:

  vect_res_18.9_68 = .MASK_LOAD (vectp_c.7_65, 32B, loop_mask_67);
  vect__9.15_77 = .MASK_LOAD (vectp_a.13_75, 8B, loop_mask_67);
  mask__34.16_79 = vect__9.15_77 != { 0, ... };
  .MASK_STORE (vectp_c.18_81, 32B, mask__34.16_79, vect_res_18.9_68);

which generates:

.L3:
        ld1b    z30.s, p7/z, [x0, x5]
        ld1w    z31.s, p7/z, [x1, x5, lsl 2]
        cmpne   p7.b, p7/z, z30.b, #0
        st1w    z31.s, p7, [x2, x5, lsl 2]
        add     x5, x5, x4
        whilelo p7.s, w5, w3
        b.any   .L3

gcc/ChangeLog:

PR tree-optimization/115531
* tree-vect-patterns.cc (vect_cond_store_pattern_same_ref): New.
(vect_recog_cond_store_pattern): New.
(vect_vect_recog_func_ptrs): Use it.
* target.def (conditional_operation_is_expensive): New.
* doc/tm.texi: Regenerate.
* doc/tm.texi.in: Document it.
* targhooks.cc (default_conditional_operation_is_expensive): New.
* targhooks.h (default_conditional_operation_is_expensive): New.

12 months agotestsuite: powerpc: fix dg-do run typo
Sam James [Mon, 22 Jul 2024 01:36:08 +0000 (20:36 -0500)] 
testsuite: powerpc: fix dg-do run typo

'dg-run' is not a valid dejagnu directive, 'dg-do run' is needed here
for the test to be executed.

PR target/108699

gcc/testsuite/ChangeLog:

* gcc.target/powerpc/pr108699.c: Fix 'dg-run' typo.

Signed-off-by: Sam James <sam@gentoo.org>
12 months agoRISC-V: Rearrange the test helper files for vector .SAT_*
Pan Li [Sat, 20 Jul 2024 02:43:44 +0000 (10:43 +0800)] 
RISC-V: Rearrange the test helper files for vector .SAT_*

Rearrange the test help header files,  as well as align the name
conventions.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/rvv/autovec/binop/vec_sat_binary.h: Move to...
* gcc.target/riscv/rvv/autovec/binop/vec_sat_binary_vvv_run.h: ...here.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_binary_scalar.h: Move to...
* gcc.target/riscv/rvv/autovec/binop/vec_sat_binary_vvx_run.h: ...here.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_binary_vx.h: Move to...
* gcc.target/riscv/rvv/autovec/binop/vec_sat_binary_vx_run.h: ...here.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-1.c: Adjust
the include file names.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-10.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-11.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-12.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-13.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-14.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-15.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-16.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-17.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-18.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-19.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-2.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-20.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-21.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-22.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-23.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-24.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-25.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-26.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-27.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-28.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-29.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-3.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-30.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-31.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-32.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-4.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-5.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-6.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-7.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-8.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-9.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-1.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-10.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-11.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-12.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-13.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-14.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-15.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-16.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-17.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-18.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-19.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-2.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-20.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-21.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-22.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-23.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-24.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-25.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-26.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-27.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-28.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-29.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-3.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-30.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-31.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-32.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-4.c: Ditto.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-5.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-6.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-7.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-8.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add-run-9.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-1.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-2.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-3.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-4.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-5.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-6.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-7.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-8.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-run-1.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-run-2.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-run-3.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-run-4.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-run-5.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-run-6.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-run-7.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_add_imm-run-8.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-1.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-10.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-11.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-12.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-13.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-14.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-15.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-16.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-17.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-18.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-19.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-2.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-20.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-21.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-22.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-23.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-24.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-25.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-26.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-27.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-28.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-29.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-3.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-30.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-31.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-32.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-33.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-34.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-35.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-36.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-37.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-38.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-39.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-4.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-40.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-5.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-6.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-7.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-8.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-9.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-1.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-10.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-11.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-12.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-13.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-14.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-15.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-16.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-17.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-18.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-19.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-2.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-20.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-21.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-22.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-23.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-24.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-25.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-26.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-27.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-28.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-29.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-3.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-30.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-31.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-32.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-33.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-34.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-35.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-36.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-37.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-38.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-39.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-4.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-40.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-5.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-6.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-7.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-8.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub-run-9.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub_trunc-1.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub_trunc-2.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub_trunc-3.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub_trunc-run-1.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub_trunc-run-2.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub_trunc-run-3.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub_zip-run.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_u_sub_zip.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-1.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-2.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-3.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-4.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-5.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-6.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-run-1.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-run-2.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-run-3.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-run-4.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-run-5.c: Ditto
* gcc.target/riscv/rvv/autovec/unop/vec_sat_u_trunc-run-6.c: Ditto
* gcc.target/riscv/rvv/autovec/binop/vec_sat_arith.h: Move to...
* gcc.target/riscv/rvv/autovec/vec_sat_arith.h: ...here.

Signed-off-by: Pan Li <pan2.li@intel.com>
12 months agoDaily bump.
GCC Administrator [Mon, 22 Jul 2024 00:16:38 +0000 (00:16 +0000)] 
Daily bump.

12 months agoFortran: Fix regression caused by r14-10477 [PR59104]
Paul Thomas [Sun, 21 Jul 2024 16:48:47 +0000 (17:48 +0100)] 
Fortran: Fix regression caused by r14-10477 [PR59104]

2024-07-21  Paul Thomas  <pault@gcc.gnu.org>

gcc/fortran
PR fortran/59104
* gfortran.h : Add decl_order to gfc_symbol.
* symbol.cc : Add static next_decl_order..
(gfc_set_sym_referenced): Set symbol decl_order.
* trans-decl.cc : Include dependency.h.
(decl_order): Replace symbol declared_at.lb->location with
decl_order.

gcc/testsuite/
PR fortran/59104
* gfortran.dg/dependent_decls_3.f90: New test.

12 months ago[PR rtl-optimization/115877][2/n] Improve liveness computation for constant initializ...
Jeff Law [Sun, 21 Jul 2024 14:41:28 +0000 (08:41 -0600)] 
[PR rtl-optimization/115877][2/n] Improve liveness computation for constant initialization

While debugging pr115877, I noticed we were failing to remove the destination
register from LIVENOW bitmap when it was set to a constant value.  ie  (set
(dest) (const_int)).  This was a trivial oversight in
safe_for_live_propagation.

I don't have an example of this affecting code generation, but it certainly
could.  More importantly, by making LIVENOW more accurate it's easier to debug
when LIVENOW differs from expectations.

As with the prior patch this has been tested as part of a larger patchset with
the crosses as well as individually on x86_64.

Pushing to the trunk,

PR rtl-optimization/115877
gcc/
* ext-dce.cc (safe_for_live_propagation): Handle RTX_CONST_OBJ.

12 months ago[PR rtl-optimization/115877] Fix livein computation for ext-dce
Jeff Law [Sun, 21 Jul 2024 13:36:37 +0000 (07:36 -0600)] 
[PR rtl-optimization/115877] Fix livein computation for ext-dce

So I'm not yet sure how I'm going to break everything down, but this is easy
enough to break out as 1/N of ext-dce fixes/improvements.

When handling uses in an insn, we first determine what bits are set in the
destination which is represented in DST_MASK.  Then we use that to refine what
bits are live in the source operands.

In the source operand handling section we *modify* DST_MASK if the source
operand is a SUBREG (ugh!).  So if the first operand is a SUBREG, then we can
incorrectly compute which bit groups are live in the second operand, especially
if it is a SUBREG as well.

This was seen when testing a larger set of patches on the rl78 port
(builtin-arith-overflow-p-7 & pr71631 execution failures), so no new test for
this bugfix.

Run through my tester (in conjunction with other ext-dce changes) on the
various cross targets.  Run individually through a bootstrap and regression
test cycle on x86_64 as well.

Pushing to the trunk.

PR rtl-optimization/115877
gcc/
* ext-dce.cc (ext_dce_process_uses): Restore the value of DST_MASK
for reach operand.

12 months agogcc: stop adding -fno-common for checking builds
Sam James [Sat, 20 Jul 2024 00:21:59 +0000 (01:21 +0100)] 
gcc: stop adding -fno-common for checking builds

Originally added in r0-44646-g204250d2fcd084 and r0-44627-gfd350d241fecf6 whic
moved -fno-common from all builds to just checking builds.

Since r10-4867-g6271dd984d7f92, GCC defaults to -fno-common. There's no need
to pass it specially for checking builds.

We could keep it for older bootstrap compilers with checking but I don't see
much value in that, it was already just a bonus before.

gcc/ChangeLog:
* Makefile.in (NOCOMMON_FLAG): Delete.
(GCC_WARN_CFLAGS): Drop NOCOMMON_FLAG.
(GCC_WARN_CXXFLAGS): Drop NOCOMMON_FLAG.
* configure.ac: Ditto.
* configure: Regenerate.

gcc/d/ChangeLog:
* Make-lang.in (WARN_DFLAGS): Drop NOCOMMON_FLAG.

12 months agoSH: Fix outage caused by recently added 2nd combine pass after reg alloc
Oleg Endo [Sun, 21 Jul 2024 05:11:21 +0000 (14:11 +0900)] 
SH: Fix outage caused by recently added 2nd combine pass after reg alloc

I've also confirmed on the CSiBE set that the secondary combine pass is
actually beneficial on SH.  It does result in some code size reductions.

gcc/CHangeLog:
* config/sh/sh.md (mov_neg_si_t): Allow insn and split after
register allocation.
(*treg_noop_move): New insn.

12 months agoDaily bump.
GCC Administrator [Sun, 21 Jul 2024 00:17:52 +0000 (00:17 +0000)] 
Daily bump.

12 months agoRequire bitint575 for pr116003.c
Andrew MacLeod [Sat, 20 Jul 2024 16:49:39 +0000 (12:49 -0400)] 
Require bitint575 for pr116003.c

Require a bitint target large enough.

gcc/testsuite/
* gcc.dg/pr116003.c: Require bitint575 target.

12 months agoRevert "Add documentation for musttail attribute"
Andi Kleen [Sat, 20 Jul 2024 23:09:41 +0000 (16:09 -0700)] 
Revert "Add documentation for musttail attribute"

This reverts commit 56f824cc206ff00d466aaeb11211d8005c4668bc.

12 months agoRevert "Add tests for C/C++ musttail attributes"
Andi Kleen [Sat, 20 Jul 2024 23:09:25 +0000 (16:09 -0700)] 
Revert "Add tests for C/C++ musttail attributes"

This reverts commit 37c4703ce84722b9c24db3e8e6d57ab6d3a7b5eb.

12 months agoRevert "C: Implement musttail attribute for returns"
Andi Kleen [Sat, 20 Jul 2024 23:09:07 +0000 (16:09 -0700)] 
Revert "C: Implement musttail attribute for returns"

This reverts commit 7db47f7b915c5f5d645fa536547e26b92290afe3.

12 months agoRevert "C++: Support clang compatible [[musttail]] (PR83324)"
Andi Kleen [Sat, 20 Jul 2024 23:07:41 +0000 (16:07 -0700)] 
Revert "C++: Support clang compatible [[musttail]] (PR83324)"

This reverts commit 59dd1d7ab21ad9a7ebf641ec9aeea609c003ad2f.

12 months agoOutput CodeView function information
Mark Harmstone [Thu, 27 Jun 2024 23:36:14 +0000 (00:36 +0100)] 
Output CodeView function information

Translate DW_TAG_subprogram DIEs into CodeView LF_FUNC_ID types and
S_GPROC32_ID / S_LPROC32_ID symbols.  ld will then transform these into
S_GPROC32 / S_LPROC32 symbols, which map addresses to unmangled function
names.

gcc/
* dwarf2codeview.cc (enum cv_sym_type): Add new values.
(struct codeview_symbol): Add function to union.
(struct codeview_custom_type): Add lf_func_id to union.
(write_function): New function.
(write_codeview_symbols): Call write_function.
(write_lf_func_id): New function.
(write_custom_types): Call write_lf_func_id.
(add_function): New function.
(codeview_debug_early_finish): Call add_function.

12 months agoAdd bitint to options for testcase
Andrew MacLeod [Sat, 20 Jul 2024 15:45:16 +0000 (11:45 -0400)] 
Add bitint to options for testcase

Testcase should only be for bitint targets

gcc/testsuite/
* gcc.dg/pr116003.c : Add target bitint.

12 months agodoc: Remove documentation of two obsolete spec strings
André Maroneze [Sat, 20 Jul 2024 14:42:47 +0000 (16:42 +0200)] 
doc: Remove documentation of two obsolete spec strings

gcc:
* doc/invoke.texi (Spec Files): Remove documentation of obsolete
spec strings "predefines" and "signed_char".

12 months agoAvoid undefined behaviour in build_option_suggestions
Siddhesh Poyarekar [Fri, 19 Jul 2024 16:44:32 +0000 (12:44 -0400)] 
Avoid undefined behaviour in build_option_suggestions

The inner loop in build_option_suggestions uses OPTION to take the
address of OPTB and use it across iterations, which is undefined
behaviour since OPTB is defined within the loop.  Pull it outside the
loop to make this defined.

gcc/ChangeLog:

* opt-suggestions.cc
(option_proposer::build_option_suggestions): Pull OPTB
definition out of the innermost loop.

12 months agoAdd documentation for musttail attribute
Andi Kleen [Wed, 24 Jan 2024 07:38:23 +0000 (23:38 -0800)] 
Add documentation for musttail attribute

gcc/ChangeLog:

PR c/83324
* doc/extend.texi: Document [[musttail]]

12 months agoAdd tests for C/C++ musttail attributes
Andi Kleen [Wed, 24 Jan 2024 07:54:56 +0000 (23:54 -0800)] 
Add tests for C/C++ musttail attributes

Some adopted from the existing C musttail plugin tests.
Also extends the ability to query the sibcall capabilities of the
target.

gcc/testsuite/ChangeLog:

* lib/target-supports.exp:
(check_effective_target_struct_tail_call): New function.
* c-c++-common/musttail1.c: New test.
* c-c++-common/musttail12.c: New test.
* c-c++-common/musttail13.c: New test.
* c-c++-common/musttail2.c: New test.
* c-c++-common/musttail3.c: New test.
* c-c++-common/musttail4.c: New test.
* c-c++-common/musttail5.c: New test.
* c-c++-common/musttail7.c: New test.
* c-c++-common/musttail8.c: New test.
* g++.dg/musttail10.C: New test.
* g++.dg/musttail11.C: New test.
* g++.dg/musttail6.C: New test.
* g++.dg/musttail9.C: New test.

12 months agoC: Implement musttail attribute for returns
Andi Kleen [Wed, 24 Jan 2024 15:44:23 +0000 (07:44 -0800)] 
C: Implement musttail attribute for returns

Implement a C23 clang compatible musttail attribute similar to the earlier
C++ implementation in the C parser.

gcc/c/ChangeLog:

PR c/83324
* c-parser.cc (struct attr_state): Define with musttail_p.
(c_parser_statement_after_labels): Handle [[musttail]].
(c_parser_std_attribute): Dito.
(c_parser_handle_musttail): Dito.
(c_parser_compound_statement_nostart): Dito.
(c_parser_all_labels): Dito.
(c_parser_statement): Dito.
* c-tree.h (c_finish_return): Add musttail_p flag.
* c-typeck.cc (c_finish_return): Handle musttail_p flag.

12 months agoC++: Support clang compatible [[musttail]] (PR83324)
Andi Kleen [Wed, 24 Jan 2024 07:44:48 +0000 (23:44 -0800)] 
C++: Support clang compatible [[musttail]] (PR83324)

This patch implements a clang compatible [[musttail]] attribute for
returns.

musttail is useful as an alternative to computed goto for interpreters.
With computed goto the interpreter function usually ends up very big
which causes problems with register allocation and other per function
optimizations not scaling. With musttail the interpreter can be instead
written as a sequence of smaller functions that call each other. To
avoid unbounded stack growth this requires forcing a sibling call, which
this attribute does. It guarantees an error if the call cannot be tail
called which allows the programmer to fix it instead of risking a stack
overflow. Unlike computed goto it is also type-safe.

It turns out that David Malcolm had already implemented middle/backend
support for a musttail attribute back in 2016, but it wasn't exposed
to any frontend other than a special plugin.

This patch adds a [[gnu::musttail]] attribute for C++ that can be added
to return statements. The return statement must be a direct call
(it does not follow dependencies), which is similar to what clang
implements. It then uses the existing must tail infrastructure.

For compatibility it also detects clang::musttail

Passes bootstrap and full test

gcc/c-family/ChangeLog:

* c-attribs.cc (set_musttail_on_return): New function.
* c-common.h (set_musttail_on_return): Declare new function.

gcc/cp/ChangeLog:

PR c/83324
* cp-tree.h (AGGR_INIT_EXPR_MUST_TAIL): Add.
* parser.cc (cp_parser_statement): Handle musttail.
(cp_parser_jump_statement): Dito.
* pt.cc (tsubst_expr): Copy CALL_EXPR_MUST_TAIL_CALL.
* semantics.cc (simplify_aggr_init_expr): Handle musttail.

12 months agoAdd a musttail generic attribute to the c-attribs table
Andi Kleen [Thu, 16 May 2024 02:38:43 +0000 (19:38 -0700)] 
Add a musttail generic attribute to the c-attribs table

The actual handling is directly in the parser since the
generic mechanism doesn't support statement attributes,
but this gives basic error checking/detection on the attribute.

gcc/c-family/ChangeLog:

PR c/83324
* c-attribs.cc (handle_musttail_attribute): Add.
* c-common.h (handle_musttail_attribute): Add.

12 months agoLoongArch: Organize the code related to split move and merge the same functions.
Lulu Cheng [Fri, 12 Jul 2024 01:57:40 +0000 (09:57 +0800)] 
LoongArch: Organize the code related to split move and merge the same functions.

gcc/ChangeLog:

* config/loongarch/loongarch-protos.h
(loongarch_split_128bit_move): Delete.
(loongarch_split_128bit_move_p): Delete.
(loongarch_split_256bit_move): Delete.
(loongarch_split_256bit_move_p): Delete.
(loongarch_split_vector_move): Add a function declaration.
* config/loongarch/loongarch.cc
(loongarch_vector_costs::finish_cost): Adjust the code
formatting.
(loongarch_split_vector_move_p): Merge
loongarch_split_128bit_move_p and loongarch_split_256bit_move_p.
(loongarch_split_move_p): Merge code.
(loongarch_split_move): Likewise.
(loongarch_split_128bit_move_p): Delete.
(loongarch_split_256bit_move_p): Delete.
(loongarch_split_128bit_move): Delete.
(loongarch_split_vector_move): Merge loongarch_split_128bit_move
and loongarch_split_256bit_move.
(loongarch_split_256bit_move): Delete.
(loongarch_global_init): Remove the extra semicolon at the
end of the function.
* config/loongarch/loongarch.md (*movdf_softfloat):  Added a new
condition TARGET_64BIT.

12 months agoDaily bump.
GCC Administrator [Sat, 20 Jul 2024 00:17:53 +0000 (00:17 +0000)] 
Daily bump.

12 months agoCheck for SSA_NAME not in the IL yet.
Andrew MacLeod [Fri, 19 Jul 2024 21:39:40 +0000 (17:39 -0400)] 
Check for SSA_NAME not in the IL yet.

Check for an SSA_NAME not in the CFG before trying to create an
equivalence record in the defintion block.

PR tree-optimization/116003
gcc/
* value-relation.cc (equiv_oracle::register_initial_def): Check
if SSA_NAME is in the IL before registering.

gcc/testsuite/
* gcc.dg/pr116003.c: New.

12 months agolibgomp: Document 'GOMP_teams4'
Thomas Schwinge [Tue, 16 Jul 2024 15:09:38 +0000 (17:09 +0200)] 
libgomp: Document 'GOMP_teams4'

For reference:

  - <https://inbox.sourceware.org/20211111190313.GV2710@tucnak> "[PATCH] openmp: Honor OpenMP 5.1 num_teams lower bound"
  - <https://inbox.sourceware.org/20211112132023.GC2710@tucnak> "[PATCH] libgomp, nvptx: Honor OpenMP 5.1 num_teams lower bound"

libgomp/
* config/gcn/target.c (GOMP_teams4): Document.
* config/nvptx/target.c (GOMP_teams4): Likewise.
* target.c (GOMP_teams4): Likewise.

12 months agoGCN: Honor OpenMP 5.1 'num_teams' lower bound
Thomas Schwinge [Mon, 15 Jul 2024 09:19:28 +0000 (11:19 +0200)] 
GCN: Honor OpenMP 5.1 'num_teams' lower bound

Corresponding to commit 9fa72756d90e0d9edadf6e6f5f56476029925788
"libgomp, nvptx: Honor OpenMP 5.1 num_teams lower bound", these are the
GCN offloading changes to fix:

    PASS: libgomp.c/../libgomp.c-c++-common/teams-2.c (test for excess errors)
    [-FAIL:-]{+PASS:+} libgomp.c/../libgomp.c-c++-common/teams-2.c execution test

    PASS: libgomp.c++/../libgomp.c-c++-common/teams-2.c (test for excess errors)
    [-FAIL:-]{+PASS:+} libgomp.c++/../libgomp.c-c++-common/teams-2.c execution test

..., and omptests' 't-critical' test case.  I've cross checked that those test
cases are the ones that regress for nvptx offloading, if I locally revert the
"libgomp, nvptx: Honor OpenMP 5.1 num_teams lower bound" changes.

libgomp/
* config/gcn/libgomp-gcn.h (GOMP_TEAM_NUM): Inject.
* config/gcn/target.c (GOMP_teams4): Handle.
* config/gcn/team.c (gomp_gcn_enter_kernel): Initialize.
* config/gcn/teams.c (omp_get_team_num): Adjust.

12 months agoRewrite usage comment at the top of 'gcc/passes.def'
Thomas Schwinge [Fri, 28 Jun 2024 12:05:04 +0000 (14:05 +0200)] 
Rewrite usage comment at the top of 'gcc/passes.def'

Since Subversion r201359 (Git commit a167b052dfe9a8509bb23c374ffaeee953df0917)
"Introduce gen-pass-instances.awk and pass-instances.def", the usage comment at
the top of 'gcc/passes.def' no longer is accurate (even if that latter file
does continue to use the 'NEXT_PASS' form without 'NUM') -- and, worse, the
'NEXT_PASS' etc. in that usage comment are processed by the
'gcc/gen-pass-instances.awk' script:

    --- source-gcc/gcc/passes.def   2024-06-24 18:55:15.132561641 +0200
    +++ build-gcc/gcc/pass-instances.def    2024-06-24 18:55:27.768562714 +0200
    [...]
    @@ -20,546 +22,578 @@
     /*
      Macros that should be defined when using this file:
        INSERT_PASSES_AFTER (PASS)
        PUSH_INSERT_PASSES_WITHIN (PASS)
        POP_INSERT_PASSES ()
    -   NEXT_PASS (PASS)
    +   NEXT_PASS (PASS, 1)
        TERMINATE_PASS_LIST (PASS)
      */
    [...]

(That is, this is 'NEXT_PASS' for the first instance of pass 'PASS'.)
That's benign so far, but with another thing that I'll be extending, I'd
then run into an error while the script handles this comment block.  ;-\

gcc/
* passes.def: Rewrite usage comment at the top.

12 months agoTreat boolean vector elements as 0/-1 [PR115406]
Richard Sandiford [Fri, 19 Jul 2024 18:09:37 +0000 (19:09 +0100)] 
Treat boolean vector elements as 0/-1 [PR115406]

Previously we built vector boolean constants using 1 for true
elements and 0 for false elements.  This matches the predicates
produced by SVE's PTRUE instruction, but leads to a miscompilation
on AVX, where all bits of a boolean element should be set.

One option for RTL would be to make this target-configurable.
But that isn't really possible at the tree level, where vectors
should work in a more target-independent way.  (There is currently
no way to create a "generic" packed boolean vector, but never say
never :))  And, if we were going to pick a generic behaviour,
it would make sense to use 0/-1 rather than 0/1, for consistency
with integer vectors.

Both behaviours should work with SVE on read, since SVE ignores
the upper bits in each predicate element.  And the choice shouldn't
make much difference for RTL, since all SVE predicate modes are
expressed as vectors of BI, rather than of multi-bit booleans.

I suspect there might be some fallout from this change on SVE.
But I think we should at least give it a go, and see whether any
fallout provides a strong counterargument against the approach.

gcc/
PR middle-end/115406
* fold-const.cc (native_encode_vector_part): For vector booleans,
check whether an element is nonzero and, if so, set all of the
correspending bits in the target image.
* simplify-rtx.cc (native_encode_rtx): Likewise.

gcc/testsuite/
PR middle-end/115406
* gcc.dg/torture/pr115406.c: New test.

12 months agoarm: Update fp16-aapcs-[24].c after insn_propagation patch
Richard Sandiford [Fri, 19 Jul 2024 18:09:37 +0000 (19:09 +0100)] 
arm: Update fp16-aapcs-[24].c after insn_propagation patch

These tests used to generate:

        bl      swap
        ldr     r2, [sp, #4]
        mov     r0, r2  @ __fp16

but g:9d20529d94b23275885f380d155fe8671ab5353a means that we can
load directly into r0:

        bl      swap
        ldrh    r0, [sp, #4]    @ __fp16

This patch updates the tests to "defend" this change.

While there, the scans include:

mov\tr1, r[03]}

But if the spill of r2 occurs first, there's no real reason why
r2 couldn't be used as the temporary, instead r3.

The patch tries to update the scans while preserving the spirit
of the originals.

gcc/testsuite/
* gcc.target/arm/fp16-aapcs-2.c: Expect the return value to be
loaded directly from the stack.  Test that the swap generates
two moves out of r0/r1 and two moves in.
* gcc.target/arm/fp16-aapcs-4.c: Likewise.

12 months agoc++: xobj fn call without obj [PR115783]
Patrick Palka [Fri, 19 Jul 2024 17:48:12 +0000 (13:48 -0400)] 
c++: xobj fn call without obj [PR115783]

The code path for rejecting an object-less call to a non-static member
function should also consider xobj member functions (so that we correctly
reject the below calls with a "cannot call member function without object"
diagnostic).

PR c++/115783

gcc/cp/ChangeLog:

* call.cc (build_new_method_call): Generalize METHOD_TYPE
check to DECL_OBJECT_MEMBER_FUNCTION_P.

gcc/testsuite/ChangeLog:

* g++.dg/cpp23/explicit-obj-diagnostics11.C: New test.

Reviewed-by: Jason Merrill <jason@redhat.com>
12 months agoAVR: Support new built-in function __builtin_avr_mask1.
Georg-Johann Lay [Fri, 19 Jul 2024 16:22:26 +0000 (18:22 +0200)] 
AVR: Support new built-in function __builtin_avr_mask1.

gcc/
* config/avr/builtins.def (MASK1): New DEF_BUILTIN.
* config/avr/avr.cc (avr_rtx_costs_1): Handle rtx costs for
expressions like __builtin_avr_mask1.
(avr_init_builtins) <uintQI_ftype_uintQI_uintQI>: New tree type.
(avr_expand_builtin) [AVR_BUILTIN_MASK1]: Diagnose unexpected forms.
(avr_fold_builtin) [AVR_BUILTIN_MASK1]: Handle case.
* config/avr/avr.md (gen_mask1): New expand helper.
(mask1_0x01_split, mask1_0x80_split, mask1_0xfe_split): New
insn-and-split.
(*mask1_0x01, *mask1_0x80, *mask1_0xfe): New insns.
* doc/extend.texi (AVR Built-in Functions) <__builtin_avr_mask1>:
Document new built-in function.
gcc/testsuite/
* gcc.target/avr/torture/builtin-mask1.c: New test.

12 months agolibgomp: Remove bogus warnings from privatized-ref-2.f90.
Paul Thomas [Fri, 19 Jul 2024 15:58:33 +0000 (16:58 +0100)] 
libgomp: Remove bogus warnings from privatized-ref-2.f90.

2024-07-19  Paul Thomas  <pault@gcc.gnu.org>

libgomp/ChangeLog

* testsuite/libgomp.oacc-fortran/privatized-ref-2.f90: Cut
dg-note about 'a' and remove bogus warnings about its array
descriptor components being used uninitialized.

12 months agoFortran: character array constructor with >= 4 constant elements [PR103115]
Harald Anlauf [Thu, 18 Jul 2024 19:15:48 +0000 (21:15 +0200)] 
Fortran: character array constructor with >= 4 constant elements [PR103115]

gcc/fortran/ChangeLog:

PR fortran/103115
* trans-array.cc (gfc_trans_array_constructor_value): If the first
element of an array constructor is deferred-length character and
therefore does not have an element size known at compile time, do
not try to collect subsequent constant elements into a constructor
for optimization.

gcc/testsuite/ChangeLog:

PR fortran/103115
* gfortran.dg/string_array_constructor_4.f90: New test.

12 months agors6000: Catch unsupported ABI errors when using -mrop-protect [PR114759,PR115988]
Peter Bergner [Thu, 18 Jul 2024 23:01:46 +0000 (18:01 -0500)] 
rs6000: Catch unsupported ABI errors when using -mrop-protect [PR114759,PR115988]

2024-07-18  Peter Bergner  <bergner@linux.ibm.com>

gcc/testsuite/
PR target/114759
PR target/115988
* gcc.target/powerpc/pr114759-3.c: Catch unsupported ABI errors.

12 months agoc++: add fixed testcase [PR109464]
Patrick Palka [Fri, 19 Jul 2024 15:08:09 +0000 (11:08 -0400)] 
c++: add fixed testcase [PR109464]

Seems to be fixed by r15-521-g6ad7ca1bb90573.

PR c++/109464

gcc/testsuite/ChangeLog:

* g++.dg/template/explicit-instantiation8.C: New test.

12 months agobpf: create modifier for mem operand for xchg and cmpxchg
Cupertino Miranda [Thu, 11 Jul 2024 14:28:09 +0000 (15:28 +0100)] 
bpf: create modifier for mem operand for xchg and cmpxchg

Both xchg and cmpxchg instructions, in the pseudo-C dialect, do not
expect their memory address operand to be surrounded by parentheses.
For example, it should be output as "w0 =cmpxchg32_32(r8+8,w0,w2)"
instead of "w0 =cmpxchg32_32((r8+8),w0,w2)".

This patch implements an operand modifier 'M' which marks the
instruction templates that do not expect the parentheses, and adds it do
xchg and cmpxchg templates.

gcc/ChangeLog:
* config/bpf/atomic.md (atomic_compare_and_swap,
atomic_exchange): Add operand modifier %M to the first
operand.
* config/bpf/bpf.cc (no_parentheses_mem_operand): Create
variable.
(bpf_print_operand): Set no_parentheses_mem_operand variable if
%M operand is used.
(bpf_print_operand_address): Conditionally output parentheses.

gcc/testsuite/ChangeLog:
* gcc.target/bpf/pseudoc-atomic-memaddr-op.c: Add test.

12 months agoc++: Add [dcl.init.aggr] examples to testsuite
Jakub Jelinek [Fri, 19 Jul 2024 06:53:47 +0000 (08:53 +0200)] 
c++: Add [dcl.init.aggr] examples to testsuite

When working on the #embed optimization support, I went recently through
all of reshape_init_r* and today I read in detail all the P3106R1 changes
and I believe we implement it that way for years.
To double check that, I've added tests with the current [dcl.init.aggr]
examples but tested in all the languages from C++98 to C++26, of course
guarded as needed for constructs which require newer versions of C++.
The examples come in two tests, one is a runtime test for the non-erroneous
examples, the other is a compile time test for the diagnostics.
The former one includes mostly intact examples with runtime checking (both
to test what is written in the section exactly and to test at least
something with C++98) and then when useful also adds constexpr tests with
static_asserts for C++11 and later.

Tested on x86_64-linux and i686-linux with
GXX_TESTSUITE_STDS=98,11,14,17,20,23,26 make check-g++ RUNTESTFLAGS='dg.exp=aggr-init*.C'

Also tested on GCC 11 branch with
GXX_TESTSUITE_STDS=98,11,14,17,20,2b make check-g++ RUNTESTFLAGS='dg.exp=aggr-init*.C'
where just the " is a GCC extension" part of one error is left out,
otherwise it passes the same, ditto with clang 14 (of course with different
diagnostics, but verified it emits diagnostics on the right lines), so I
believe we can claim implementation of this DR paper, either in all versions
or at least in GCC 11+.

2024-07-19  Jakub Jelinek  <jakub@redhat.com>

PR c++/114460
* g++.dg/cpp26/aggr-init1.C: New test.
* g++.dg/cpp26/aggr-init2.C: New test.

12 months agoClose GCC 11 branch
Richard Biener [Fri, 19 Jul 2024 05:58:28 +0000 (07:58 +0200)] 
Close GCC 11 branch

Remove gcc-11 branch from updating and snapshot generating

contrib/
* gcc-changelog/git_update_version.py: Remove gcc-11 branch.

maintainer-scripts/
* crontab: Remove entry for gcc-11 branch.

12 months agoc++: Hash placeholder constraint in ctp_hasher
Seyed Sajad Kahani [Thu, 18 Jul 2024 15:01:32 +0000 (16:01 +0100)] 
c++: Hash placeholder constraint in ctp_hasher

This patch addresses a difference between the hash function and the equality
function for canonical types of template parameters (ctp_hasher). The equality
function uses comptypes (typeck.cc) (with COMPARE_STRUCTURAL) and checks
constraint equality for two auto nodes (typeck.cc:1586), while the hash
function ignores it (pt.cc:4528). This leads to hash collisions that can be
avoided by using `hash_placeholder_constraint` (constraint.cc:1150).

Note that due to the proper handling of hash collisions (hash-table.h:1059),
there is no test case that can distinguish the current implementation from the
proposed one.

* constraint.cc (hash_placeholder_constraint): Rename to
iterative_hash_placeholder_constraint.
(iterative_hash_placeholder_constraint): Rename from
hash_placeholder_constraint and add the initial val argument.
* cp-tree.h (hash_placeholder_constraint): Rename to
iterative_hash_placeholder_constraint.
(iterative_hash_placeholder_constraint): Renamed from
hash_placeholder_constraint and add the initial val argument.
* pt.cc (struct ctp_hasher): Updated to use
iterative_hash_placeholder_constraint in the case of a valid placeholder
constraint.
(auto_hash::hash): Reflect the renaming of hash_placeholder_constraint to
iterative_hash_placeholder_constraint.

12 months agoMatch: Only allow single use of MIN_EXPR for SAT_TRUNC form 2 [PR115863]
Pan Li [Thu, 18 Jul 2024 12:16:34 +0000 (20:16 +0800)] 
Match: Only allow single use of MIN_EXPR for SAT_TRUNC form 2 [PR115863]

The SAT_TRUNC form 2 has below pattern matching.
From:
  _18 = MIN_EXPR <left_8, 4294967295>;
  iftmp.0_11 = (unsigned int) _18;

To:
  _18 = MIN_EXPR <left_8, 4294967295>;
  iftmp.0_11 = .SAT_TRUNC (left_8);

But if there is another use of _18 like below,  the transform to the
.SAT_TRUNC may have no earnings.  For example:

From:
  _18 = MIN_EXPR <left_8, 4294967295>; // op_0 def
  iftmp.0_11 = (unsigned int) _18;     // op_0
  stream.avail_out = iftmp.0_11;
  left_37 = left_8 - _18;              // op_0 use

To:
  _18 = MIN_EXPR <left_8, 4294967295>; // op_0 def
  iftmp.0_11 = .SAT_TRUNC (left_8);
  stream.avail_out = iftmp.0_11;
  left_37 = left_8 - _18;              // op_0 use

Pattern recog to .SAT_TRUNC cannot eliminate MIN_EXPR as above.  Then the
backend (for example x86/riscv) will have additional 2-3 more insns
after pattern recog besides the MIN_EXPR.  Thus,  keep the normal truncation
as is should be the better choose.

The below testsuites are passed for this patch:
1. The rv64gcv fully regression tests.
2. The x86 bootstrap tests.
3. The x86 fully regression tests.

PR target/115863

gcc/ChangeLog:

* match.pd: Add single_use check for .SAT_TRUNC form 2.

gcc/testsuite/ChangeLog:

* gcc.target/i386/pr115863-1.c: New test.

Signed-off-by: Pan Li <pan2.li@intel.com>
12 months agoDaily bump.
GCC Administrator [Fri, 19 Jul 2024 00:18:20 +0000 (00:18 +0000)] 
Daily bump.

12 months agolibatomic: Handle AVX+CX16 ZHAOXIN like Intel for 16b atomic [PR104688]
mayshao [Thu, 18 Jul 2024 20:43:00 +0000 (22:43 +0200)] 
libatomic: Handle AVX+CX16 ZHAOXIN like Intel for 16b atomic [PR104688]

PR target/104688

libatomic/ChangeLog:

* config/x86/init.c (__libat_feat1_init): Don't clear
bit_AVX on ZHAOXIN CPUs.

12 months agoc++: implement DR1363 and DR1496 for __is_trivial [PR85723]
Marek Polacek [Tue, 18 Jun 2024 20:49:24 +0000 (16:49 -0400)] 
c++: implement DR1363 and DR1496 for __is_trivial [PR85723]

is_trivial was introduced in
<https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2230.html>
which split POD into is_trivial and is_standard_layout.

Later came CWG 1363.  Since

  struct A {
    A() = default;
    A(int = 42) {}
  };

cannot be default-initialized, it should not be trivial, so the definition
of what is a trivial class changed.

Similarly, CWG 1496 concluded that

  struct B {
    B() = delete;
  }:

should not be trivial either.

P0848 adjusted the definition further to say "eligible".  That means
that

  template<typename T>
  struct C {
    C() requires false = default;
  };

should not be trivial, either, since C::C() is not eligible.

Bug 85723 reports that we implement none of the CWGs.

I chose to fix this by using type_has_non_deleted_trivial_default_ctor
which uses locate_ctor which uses build_new_method_call, which would
be used by default-initialization as well.  With that, all __is_trivial
problems I could find in the Bugzilla are fixed, except for PR96288,
which may need changes to trivially-copyable, so I'm not messing with
that now.

I hope this has no ABI implications.  There's effort undergoing to
remove "trivial class" from the core language as it's not really
meaningful.  So the impact of this change should be pretty low except
to fix a few libstdc++ problems.

PR c++/108769
PR c++/58074
PR c++/115522
PR c++/85723

gcc/cp/ChangeLog:

* class.cc (type_has_non_deleted_trivial_default_ctor): Fix formatting.
* tree.cc (trivial_type_p): Instead of TYPE_HAS_TRIVIAL_DFLT, use
type_has_non_deleted_trivial_default_ctor.

gcc/testsuite/ChangeLog:

* g++.dg/warn/Wclass-memaccess.C: Add dg-warning.
* g++.dg/ext/is_trivial1.C: New test.
* g++.dg/ext/is_trivial2.C: New test.
* g++.dg/ext/is_trivial3.C: New test.
* g++.dg/ext/is_trivial4.C: New test.
* g++.dg/ext/is_trivial5.C: New test.
* g++.dg/ext/is_trivial6.C: New test.

12 months agolibbacktrace: use __has_attribute for fallthrough
Ian Lance Taylor [Thu, 18 Jul 2024 18:34:09 +0000 (11:34 -0700)] 
libbacktrace: use __has_attribute for fallthrough

Also convert some FALLTHROUGH comments to ATTRIBUTE_FALLTHROUGH.

* internal.h: Use __has_attribute to check for fallthrough
attribute.
* elf.c (elf_zstd_decompress): Use ATTRIBUTE_FALLTHROUGH rather
than a FALLTHROUGH comment.

12 months agors6000: Fix .machine cpu selection w/ altivec [PR97367]
René Rebe [Fri, 12 Jul 2024 21:17:08 +0000 (21:17 +0000)] 
rs6000: Fix .machine cpu selection w/ altivec [PR97367]

There are various non-IBM CPUs with altivec, so we cannot use that
flag to determine which .machine cpu to use, so ignore it.
Emit an additional ".machine altivec" if Altivec is enabled so
that the assembler doesn't require an explicit -maltivec option
to assemble any Altivec instructions for those targets where
the ".machine cpu" is insufficient to enable Altivec.  For example,
-mcpu=G5 emits a ".machine power4".

2024-07-18  René Rebe  <rene@exactcode.de>
    Peter Bergner  <bergner@linux.ibm.com>

gcc/
PR target/97367
* config/rs6000/rs6000.cc (rs6000_machine_from_flags): Do not consider
OPTION_MASK_ALTIVEC.
(emit_asm_machine): For Altivec compiles, emit a ".machine altivec".

gcc/testsuite/
PR target/97367
* gcc.target/powerpc/pr97367.c: New test.

Signed-off-by: René Rebe <rene@exactcode.de>
12 months agors6000, update effective target for tests builtins-10*.c and vec_perm-runnable-i128.c
Carl Love [Fri, 12 Jul 2024 18:37:36 +0000 (13:37 -0500)] 
rs6000, update effective target for tests builtins-10*.c and vec_perm-runnable-i128.c

The tests:

  tests builtins-10-runnable.c
  tests builtins-10.c
  vec_perm-runnable-i128.c

use __int128 types that are not supported on all platforms.  Update the
tests to check int128 effective target to avoid unsupported type errors
on unsupported platforms.

gcc/testsuite/ChangeLog:
* gcc.target/powerpc/builtins-10-runnable.c: Add
target int128.
* gcc.target/powerpc/builtins-10.c: Add
target int128.
* gcc.target/powerpc/vec_perm-runnable-i128.c: Add
target int128.

12 months agolibatomic: Improve cpuid usage in __libat_feat1_init
Uros Bizjak [Thu, 18 Jul 2024 14:58:09 +0000 (16:58 +0200)] 
libatomic: Improve cpuid usage in __libat_feat1_init

Check the result of __get_cpuid and process FEAT1_REGISTER only when
__get_cpuid returns success.  Use __cpuid instead of nested __get_cpuid.

libatomic/ChangeLog:

* config/x86/init.c (__libat_feat1_init): Check the result of
__get_cpuid and process FEAT1_REGISTER only when __get_cpuid
returns success.  Use __cpuid instead of nested __get_cpuid.

12 months agoeh: ICE with std::initializer_list and ASan [PR115865]
Marek Polacek [Thu, 11 Jul 2024 19:57:43 +0000 (15:57 -0400)] 
eh: ICE with std::initializer_list and ASan [PR115865]

Here we ICE with -fsanitize=address on

  std::initializer_list x = { 1, 2, 3 };

since r14-8681, which removed .ASAN_MARK calls on TREE_STATIC variables.
That means that lower_try_finally now instead of

  try
    {
      .ASAN_MARK (UNPOISON, &C.0, 12);
      x = {};
      x._M_len = 3;
      x._M_array = &C.0;
    }
  finally
    {
      .ASAN_MARK (POISON, &C.0, 12);
    }

gets:

  try
    {
      x = {};
      x._M_len = 3;
      x._M_array = &C.0;
    }
  finally
    {

    }

and we ICE on the empty finally in lower_try_finally_onedest while
getting get_eh_else.

PR c++/115865

gcc/ChangeLog:

* tree-eh.cc (get_eh_else): Check that the result of
gimple_seq_first_stmt is non-null.

gcc/testsuite/ChangeLog:

* g++.dg/asan/initlist2.C: New test.

Co-authored-by: Jakub Jelinek <jakub@redhat.com>
12 months agoDo not use caller-saved registers for COMDAT functions
LIU Hao [Mon, 15 Jul 2024 08:55:52 +0000 (16:55 +0800)] 
Do not use caller-saved registers for COMDAT functions

A reference to a COMDAT function may be resolved to another definition
outside the current translation unit, so it's not eligible for `-fipa-ra`.

In `decl_binds_to_current_def_p()` there is already a check for weak
symbols. This commit checks for COMDAT functions that are not implemented
as weak symbols, for example, on *-*-mingw32.

gcc/ChangeLog:

PR rtl-optimization/115049
* varasm.cc (decl_binds_to_current_def_p): Add a check for COMDAT
declarations too, like weak ones.

12 months agomiddle-end/115641 - invalid address construction
Richard Biener [Thu, 18 Jul 2024 11:35:33 +0000 (13:35 +0200)] 
middle-end/115641 - invalid address construction

fold_truth_andor_1 via make_bit_field_ref builds an address of
a CALL_EXPR which isn't valid GENERIC and later causes an ICE.
The following simply avoids the folding for f ().a != 1 || f ().b != 2
as it is a premature optimization anyway.  The alternative would
have been to build a TARGET_EXPR around the call.  To get this far
f () has to be const as otherwise the two calls are not semantically
equivalent for the optimization.

PR middle-end/115641
* fold-const.cc (decode_field_reference): If the inner
reference isn't something we can take the address of, fail.

* gcc.dg/torture/pr115641.c: New testcase.

12 months agoDoc: Add Standard-Names ustrunc and sstrunc for integer modes
Pan Li [Thu, 18 Jul 2024 03:30:38 +0000 (11:30 +0800)] 
Doc: Add Standard-Names ustrunc and sstrunc for integer modes

This patch would like to add the doc for the Standard-Names
ustrunc and sstrunc,  include both the scalar and vector integer
modes.

gcc/ChangeLog:

* doc/md.texi: Add Standard-Names ustrunc and sstrunc.

Signed-off-by: Pan Li <pan2.li@intel.com>
12 months agoFortran: Fix Explicit cobounds of a procedures parameter not respected [PR78466]
Andre Vehreschild [Thu, 31 Dec 2020 09:40:30 +0000 (10:40 +0100)] 
Fortran: Fix Explicit cobounds of a procedures parameter not respected [PR78466]

Explicit cobounds of class array procedure parameters were not taken
into account.  Furthermore were different cobounds in distinct
procedure parameter lists mixed up, i.e. the last definition was taken
for all.  The bounds are now regenerated when tree's and expr's bounds
do not match.

PR fortran/78466
PR fortran/80774

gcc/fortran/ChangeLog:

* array.cc (gfc_compare_array_spec): Take cotype into account.
* class.cc (gfc_build_class_symbol): Coarrays are also arrays.
* gfortran.h (IS_CLASS_COARRAY_OR_ARRAY): New macro to detect
regular and coarray class arrays.
* interface.cc (compare_components): Take codimension into
account.
* resolve.cc (resolve_symbol): Improve error message.
* simplify.cc (simplify_bound_dim): Remove duplicate.
* trans-array.cc (gfc_trans_array_cobounds): Coarrays are also
arrays.
(gfc_trans_array_bounds): Same.
(gfc_trans_dummy_array_bias): Same.
(get_coarray_as): Get the as having a non-zero codim.
(is_explicit_coarray): Detect explicit coarrays.
(gfc_conv_expr_descriptor): Create a new descriptor for explicit
coarrays.
* trans-decl.cc (gfc_build_qualified_array): Coarrays are also
arrays.
(gfc_build_dummy_array_decl): Same.
(gfc_get_symbol_decl): Same.
(gfc_trans_deferred_vars): Same.
* trans-expr.cc (class_scalar_coarray_to_class): Get the
descriptor from the correct location.
(gfc_conv_variable): Pick up the descriptor when needed.
* trans-types.cc (gfc_is_nodesc_array): Coarrays are also
arrays.
(gfc_get_nodesc_array_type): Indentation fix only.
(cobounds_match_decl): Match a tree's bounds to the expr's
bounds and return true, when they match.
(gfc_get_derived_type): Create a new type tree/descriptor, when
the cobounds of the existing declaration and expr to not
match.  This happends for class arrays in parameter list, when
there are different cobound declarations.

gcc/testsuite/ChangeLog:

* gfortran.dg/coarray/poly_run_1.f90: Activate old test code.
* gfortran.dg/coarray/poly_run_2.f90: Activate test.  It was
stopping before and passing without an error.

12 months agotestsuite: Add dg-do run to more tests
Sam James [Thu, 18 Jul 2024 08:00:17 +0000 (10:00 +0200)] 
testsuite: Add dg-do run to more tests

All of these are for wrong-code bugs.  Confirmed to be used before but
with no execution.

2024-07-18  Sam James  <sam@gentoo.org>

PR c++/53288
PR c++/57437
PR c/65345
PR libstdc++/88101
PR tree-optimization/96369
PR tree-optimization/102124
PR tree-optimization/108692
* c-c++-common/pr96369.c: Add dg-do run directive.
* gcc.dg/torture/pr102124.c: Ditto.
* gcc.dg/pr108692.c: Ditto.
* gcc.dg/atomic/pr65345-4.c: Ditto.
* g++.dg/cpp0x/lambda/lambda-return1.C: Ditto.
* g++.dg/init/lifetime4.C: Ditto.
* g++.dg/torture/builtin-clear-padding-1.C: Ditto.
* g++.dg/torture/builtin-clear-padding-2.C: Ditto.
* g++.dg/torture/builtin-clear-padding-3.C: Ditto.
* g++.dg/torture/builtin-clear-padding-4.C: Ditto.
* g++.dg/torture/builtin-clear-padding-5.C: Ditto.

12 months agoFortran: Suppress bogus used uninitialized warnings [PR108889].
Paul Thomas [Thu, 18 Jul 2024 07:51:35 +0000 (08:51 +0100)] 
Fortran: Suppress bogus used uninitialized warnings [PR108889].

2024-07-18  Paul Thomas  <pault@gcc.gnu.org>

gcc/fortran
PR fortran/108889
* gfortran.h: Add bit field 'allocated_in_scope' to gfc_symbol.
* trans-array.cc (gfc_array_allocate): Set 'allocated_in_scope'
after allocation if not a component reference.
(gfc_alloc_allocatable_for_assignment): If 'allocated_in_scope'
not set, not a component ref and not allocated, set the array
bounds and offset to give zero length in all dimensions. Then
set allocated_in_scope.

gcc/testsuite/
PR fortran/108889
* gfortran.dg/pr108889.f90: New test.

12 months agogimple-fold: consistent dump of builtin call simplifications
Rubin Gerritsen [Tue, 16 Jul 2024 19:11:24 +0000 (21:11 +0200)] 
gimple-fold: consistent dump of builtin call simplifications

Previously only simplifications of the `__st[xrp]cpy_chk`
were dumped. Now all call replacement simplifications are
dumped.

Examples of statements with corresponding dumpfile entries:

`printf("mystr\n");`:
  optimized: simplified printf to __builtin_puts
`printf("%c", 'a');`:
  optimized: simplified printf to __builtin_putchar
`printf("%s\n", "mystr");`:
  optimized: simplified printf to __builtin_puts

The below test suites passed for this patch
* The x86 bootstrap test.
* Manual testing with some small example code manually
  examining dump logs, outputting the lines mentioned above.

gcc/ChangeLog:

* gimple-fold.cc (dump_transformation): Moved definition.
(replace_call_with_call_and_fold): Calls dump_transformation.
(gimple_fold_builtin_stxcpy_chk): Removes call to
dump_transformation, now in replace_call_with_call_and_fold.
(gimple_fold_builtin_stxncpy_chk): Removes call to
dump_transformation, now in replace_call_with_call_and_fold.

Signed-off-by: Rubin Gerritsen <rubin.gerritsen@gmail.com>
12 months agotree-optimization/104515 - store motion and clobbers
Richard Biener [Wed, 17 Jul 2024 08:22:47 +0000 (10:22 +0200)] 
tree-optimization/104515 - store motion and clobbers

The following addresses an old regression when end-of-object/storage
clobbers were introduced.  In particular when there's an end-of-object
clobber in a loop but no corresponding begin-of-object we can still
perform store motion of may-aliased refs when we re-issue the
end-of-object/storage on the exits but elide it from the loop.  This
should be the safest way to deal with this considering stack-slot
sharing and it should not cause missed dead store eliminations given
DSE can now follow multiple paths in case there are multiple exits.

Note when the clobber is re-materialized only on one exit but not
on anther we are erroring on the side of removing the clobber on
such path.  This should be OK (removing clobbers is always OK).

Note there's no corresponding code to handle begin-of-object/storage
during the hoisting part of loads that are part of a store motion
optimization, so this only enables stored-only store motion or cases
without such clobber inside the loop.

PR tree-optimization/104515
* tree-ssa-loop-im.cc (execute_sm_exit): Add clobbers_to_prune
parameter and handle re-materializing of clobbers.
(sm_seq_valid_bb): end-of-storage/object clobbers are OK inside
an ordered sequence of stores.
(sm_seq_push_down): Refuse to push down clobbers.
(hoist_memory_references): Prune clobbers from the loop body
we re-materialized on an exit.

* g++.dg/opt/pr104515.C: New testcase.

12 months agoImplement a -ftrapping-math/-fsignaling-nans TODO in match.pd.
Roger Sayle [Thu, 18 Jul 2024 07:27:36 +0000 (08:27 +0100)] 
Implement a -ftrapping-math/-fsignaling-nans TODO in match.pd.

I've been investigating some (float)i == CST optimizations for match.pd,
and noticed there's already a TODO comment in match.pd that's relatively
easy to implement.  When CST is a NaN, we only need to worry about
exceptions with flag_trapping_math, and equality/inequality tests for
sNaN only behave differently to qNaN with -fsignaling-nans.  These
issues are related to PR 57371 and PR 106805 in bugzilla.

2024-07-18  Roger Sayle  <roger@nextmovesoftware.com>

gcc/ChangeLog
* match.pd ((FTYPE) N CMP CST): Only worry about exceptions with
flag_trapping_math, and about signaling NaNs with HONOR_SNANS.

gcc/testsuite/ChangeLog
* c-c++-common/pr57371-4.c: Update comment.
* c-c++-common/pr57371-5.c: Add missing testcases from pr57371-4.c
and update for -fno-signaling-nans -fno-trapping-math.

12 months agoFortran: Use char* for deferred length character arrays [PR82904]
Andre Vehreschild [Wed, 10 Jul 2024 12:37:37 +0000 (14:37 +0200)] 
Fortran: Use char* for deferred length character arrays [PR82904]

Randomly during compiling the pass IPA: inline would ICE.  This was
caused by a saved deferred length string.  The length variable was not
set, but the variable was used in the array's declaration.  Now using a
character pointer to prevent this.

PR fortran/82904

gcc/fortran/ChangeLog:

* trans-types.cc (gfc_sym_type): Use type `char*` for saved
deferred length char arrays.
* trans.cc (get_array_span): Get `.span` also for `char*` typed
arrays, i.e. for those that have INTEGER_TYPE instead of
ARRAY_TYPE.

gcc/testsuite/ChangeLog:

* gfortran.dg/deferred_character_38.f90: New test.

12 months agotestsuite: Fix up builtin-clear-padding-3.c for -funsigned-char
Jakub Jelinek [Thu, 18 Jul 2024 07:22:10 +0000 (09:22 +0200)] 
testsuite: Fix up builtin-clear-padding-3.c for -funsigned-char

As reported on gcc-regression, this test FAILs on aarch64, but my
r15-2090 change didn't change anything on the generated assembly,
just added the forgotten dg-do run directive to the test, so the
test has been failing forever, just we didn't know it.

I can actually reproduce it on x86_64 with -funsigned-char too,
s2.b.a has int type and -1 is stored to it, so we should compare
it against -1 rather than (char) -1; the latter is appropriate for
testing char fields into which we've stored -1.

2024-07-18  Jakub Jelinek  <jakub@redhat.com>

* c-c++-common/torture/builtin-clear-padding-3.c (main): Compare
s2.b.a against -1 rather than (char) -1.

12 months agoi386: Fix testcases generating invalid asm
Haochen Jiang [Wed, 17 Jul 2024 08:26:35 +0000 (16:26 +0800)] 
i386: Fix testcases generating invalid asm

For compile test, we should generate valid asm except for special purposes.
Fix the compile test that generates invalid asm.

gcc/testsuite/ChangeLog:

* gcc.target/i386/apx-egprs-names.c: Use ax for short and
al for char instead of eax.
* gcc.target/i386/avx512bw-kandnq-1.c: Do not run the test
under -m32 since kmovq with register is invalid. Use long
long to use 64 bit register instead of 32 bit register for
kmovq.
* gcc.target/i386/avx512bw-kandq-1.c: Ditto.
* gcc.target/i386/avx512bw-knotq-1.c: Ditto.
* gcc.target/i386/avx512bw-korq-1.c: Ditto.
* gcc.target/i386/avx512bw-kshiftlq-1.c: Ditto.
* gcc.target/i386/avx512bw-kshiftrq-1.c: Ditto.
* gcc.target/i386/avx512bw-kxnorq-1.c: Ditto.
* gcc.target/i386/avx512bw-kxorq-1.c: Ditto.

12 months ago[aarch64] Document rewriting of -march=native to -mcpu=native
Kyrylo Tkachov [Tue, 16 Jul 2024 11:29:42 +0000 (16:59 +0530)] 
[aarch64] Document rewriting of -march=native to -mcpu=native

Commit dd9e5f4db2debf1429feab7f785962ccef6e0dbd changed -march=native to
treat it as -mcpu=native if no other mcpu or mtune option was given.
It would make sense to document this, especially if we try to persuade
compilers like LLVM to take the same approach.
This patch documents that behaviour.

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

Signed-off-by: Kyrylo Tkachov <ktkachov@nvidia.com>
gcc/ChangeLog:

* doc/invoke.texi (AArch64 Options): Document rewriting of
-march=native to -mcpu=native.

12 months agoOptimize maskstore when mask is 0 or -1 in UNSPEC_MASKMOV
liuhongt [Tue, 16 Jul 2024 07:29:01 +0000 (15:29 +0800)] 
Optimize maskstore when mask is 0 or -1 in UNSPEC_MASKMOV

gcc/ChangeLog:

PR target/115843
* config/i386/predicates.md (const0_or_m1_operand): New
predicate.
* config/i386/sse.md (*<avx512>_store<mode>_mask_1): New
pre_reload define_insn_and_split.
(V): Add V32BF,V16BF,V8BF.
(V4SF_V8BF): Rename to ..
(V24F_128): .. this.
(*vec_concat<mode>): Adjust with V24F_128.
(*vec_concat<mode>_0): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/pr115843.c: New test.

12 months agoMark expand musttail error messages for translation
Andi Kleen [Fri, 21 Jun 2024 18:19:12 +0000 (11:19 -0700)] 
Mark expand musttail error messages for translation

The musttail error messages are reported to the user, so must be
translated.

gcc/ChangeLog:

PR c/83324
* calls.cc (initialize_argument_information): Mark messages
for translation.
(can_implement_as_sibling_call_p): Dito.
(expand_call): Dito.

12 months agoGive better error messages for musttail
Andi Kleen [Tue, 21 May 2024 14:01:57 +0000 (07:01 -0700)] 
Give better error messages for musttail

When musttail is set, make tree-tailcall give error messages
when it cannot handle a call. This avoids vague "other reasons"
error messages later at expand time when it sees a musttail
function not marked tail call.

In various cases this requires delaying the error until
the call is discovered.

Also print more information on the failure to the dump file.

gcc/ChangeLog:

PR c/83324
* tree-tailcall.cc (maybe_error_musttail): New function.
(suitable_for_tail_opt_p): Report error reason.
(suitable_for_tail_call_opt_p): Report error reason.
(find_tail_calls): Accept basic blocks with abnormal edges.
Delay reporting of errors until the call is discovered.
Move top level suitability checks to here.
(tree_optimize_tail_calls_1): Remove top level checks.

12 months agoEnable musttail tail conversion even when not optimizing
Andi Kleen [Thu, 16 May 2024 02:57:22 +0000 (19:57 -0700)] 
Enable musttail tail conversion even when not optimizing

Enable the tailcall optimization for non optimizing builds,
but in this case only checks calls that have the musttail attribute set.
This makes musttail work without optimization.

This is done with a new late musttail pass that is only active when
not optimizing. The new pass relies on tree-cfg to discover musttails.
This avoids a ~0.8% compiler run time penalty at -O0.

gcc/ChangeLog:

PR c/83324
* function.h (struct function): Add has_musttail.
* lto-streamer-in.cc (input_struct_function_base): Stream
has_musttail.
* lto-streamer-out.cc (output_struct_function_base): Dito.
* passes.def (pass_musttail): Add.
* tree-cfg.cc (notice_special_calls): Record has_musttail.
(clear_special_calls): Clear has_musttail.
* tree-pass.h (make_pass_musttail): Add.
* tree-tailcall.cc (find_tail_calls): Handle only_musttail
argument.
(tree_optimize_tail_calls_1): Pass on only_musttail.
(execute_tail_calls): Pass only_musttail as false.
(class pass_musttail): Add.
(make_pass_musttail): Add.

12 months agoFix pro_and_epilogue for sibcalls at -O0 (PR115255)
Andi Kleen [Sun, 2 Jun 2024 05:04:41 +0000 (22:04 -0700)] 
Fix pro_and_epilogue for sibcalls at -O0 (PR115255)

Some of the cfg fixups in pro_and_epilogue for sibcalls were dependent on "optimize".
Make them check cfun->tail_call_marked instead to handle the -O0 musttail
case. This fixes the musttail test cases on arm targets.

gcc/ChangeLog:

PR target/115255
* function.cc (thread_prologue_and_epilogue_insns): Check
cfun->tail_call_marked for sibcalls too.
(rest_of_handle_thread_prologue_and_epilogue): Dito.

12 months agoImprove must tail in RTL backend
Andi Kleen [Wed, 24 Jan 2024 07:42:08 +0000 (23:42 -0800)] 
Improve must tail in RTL backend

- Give error messages for all causes of non sibling call generation
- When giving error messages clear the musttail flag to avoid ICEs
- Error out when tree-tailcall failed to mark a must-tail call
sibcall. In this case it doesn't know the true reason and only gives
a vague message.

gcc/ChangeLog:

PR c/83324
* calls.cc (maybe_complain_about_tail_call): Clear must tail
flag on error.
(expand_call): Give error messages for all musttail failures.

12 months agoc++/modules: Conditionally start timer during lazy load [PR115165]
Nathaniel Shead [Sun, 7 Jul 2024 13:19:52 +0000 (23:19 +1000)] 
c++/modules: Conditionally start timer during lazy load [PR115165]

While lazy loading, instantiation of pendings can sometimes recursively
perform name lookup and begin further lazy loading.  When using the
'-ftime-report' functionality this causes ICEs as we could start an
already-running timer for the importing.

This patch fixes the issue by using the 'timevar_cond*' API instead to
support such recursive calls.

PR c++/115165

gcc/cp/ChangeLog:

* module.cc (lazy_load_binding): Use 'timevar_cond*' APIs.
(lazy_load_pendings): Likewise.

gcc/testsuite/ChangeLog:

* g++.dg/modules/timevar-1_a.H: New test.
* g++.dg/modules/timevar-1_b.C: New test.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
12 months agoc++: prev declared hidden tmpl friend inst [PR112288]
Patrick Palka [Thu, 18 Jul 2024 01:02:52 +0000 (21:02 -0400)] 
c++: prev declared hidden tmpl friend inst [PR112288]

When partially instantiating a previously declared hidden template
friend definition (at class template scope) such as slot_allocated in
the first testcase below, tsubst_friend_function needs to go through
all existing specializations thereof and make them point to the new
definition.

But when the previous declaration was also at class template scope,
old_decl is not the most general template, instead it's the partial
instantiation, and since instantiations are relative to the most general
template, old_decl's DECL_TEMPLATE_INSTANTIATIONS is empty.  So we
to consistently use the most general template here.  And when adjusting
DECL_TI_ARGS to match, only the innermost template arguments should be
preserved; the outer ones should correspond to the new definition.

Otherwise we fail a checking-only sanity check in instantiate_decl in
the first testcase, and in the second/third we end up emitting multiple
definitions of the template friend instantiation, resulting in a link
failure.

PR c++/112288

gcc/cp/ChangeLog:

* pt.cc (tsubst_friend_function): When adjusting existing
specializations after defining a previously declared template
friend, consider the most general template and correct
DECL_TI_ARGS adjustment.

gcc/testsuite/ChangeLog:

* g++.dg/template/friend80.C: New test.
* g++.dg/template/friend81.C: New test.
* g++.dg/template/friend81a.C: New test.

Reviewed-by: Jason Merrill <jason@redhat.com>
12 months agolibbacktrace: add cast to avoid warning
Ian Lance Taylor [Thu, 18 Jul 2024 00:58:56 +0000 (17:58 -0700)] 
libbacktrace: add cast to avoid warning

* print.c (print_syminfo_callback): Add cast to avoid warning.

12 months agoc++: missing -Wunused-value for !<expr> [PR114104]
Patrick Palka [Thu, 18 Jul 2024 00:57:54 +0000 (20:57 -0400)] 
c++: missing -Wunused-value for !<expr> [PR114104]

Here we're neglecting to issue a -Wunused-value warning for suitable !
operator expressions, and in turn for != operator expressions that are
rewritten as !(x == y), only because we don't call warn_if_unused_value
on TRUTH_NOT_EXPR since its class is tcc_expression.  This patch makes
us also consider warning for TRUTH_NOT_EXPR and also for ADDR_EXPR.

PR c++/114104

gcc/cp/ChangeLog:

* cvt.cc (convert_to_void): Call warn_if_unused_value for
TRUTH_NOT_EXPR and ADDR_EXPR as well.

gcc/testsuite/ChangeLog:

* g++.dg/warn/Wunused-20.C: New test.

Reviewed-by: Jason Merrill <jason@redhat.com>
12 months agoc++: diagnose failed qualified lookup into current inst
Patrick Palka [Thu, 18 Jul 2024 00:54:14 +0000 (20:54 -0400)] 
c++: diagnose failed qualified lookup into current inst

When the scope of a qualified name is the current instantiation, and
qualified lookup finds nothing at template definition time, then we
know it'll find nothing at instantiation time (unless the current
instantiation has dependent bases).  So such qualified name lookup
failure can be diagnosed ahead of time as per [temp.res.general]/6.

This patch implements that, for qualified names of the form (where
the current instantiation is A<T>):

  this->non_existent
  a.non_existent
  A::non_existent
  typename A::non_existent

It turns out we already optimistically attempt qualified lookup of
seemingly every qualified name, even when it's dependently scoped, and
then suppress issuing a lookup failure diagnostic after the fact.
So implementing this is mostly a matter of restricting the diagnostic
suppression to "dependentish" scopes (i.e. dependent scopes or the
current instantiation with dependent bases), rather than suppressing
for any dependently-typed scope as we currently do.

The cp_parser_conversion_function_id change is needed to avoid regressing
lookup/using8.C:

  using A<T>::operator typename A<T>::Nested*;

When looking up A<T>::Nested we consider it not dependently scoped since
we entered A<T> from cp_parser_conversion_function_id earlier.   But this
A<T> is the implicit instantiation A<T> not the primary template type A<T>,
and so the lookup fails which we now diagnose.  This patch works around
this by not entering the template scope of a qualified conversion
function-id in this case, i.e. if we're in an expression vs declaration
context, by seeing if the type already went through finish_template_type
with entering_scope=true.

gcc/cp/ChangeLog:

* decl.cc (make_typename_type): Restrict name lookup failure
punting to dependentish_scope_p instead of dependent_type_p.
* error.cc (qualified_name_lookup_error): Improve diagnostic
when the scope is the current instantiation.
* parser.cc (cp_parser_diagnose_invalid_type_name): Likewise.
(cp_parser_conversion_function_id): Don't call push_scope on
a template scope unless we're in a declaration context.
(cp_parser_lookup_name): Restrict name lookup failure
punting to dependentish_scope_p instead of depedent_type_p.
* semantics.cc (finish_id_expression_1): Likewise.
* typeck.cc (finish_class_member_access_expr): Likewise.

libstdc++-v3/ChangeLog:

* include/experimental/socket
(basic_socket_iostream::basic_socket_iostream): Fix typo.
* include/tr2/dynamic_bitset
(__dynamic_bitset_base::_M_is_proper_subset_of): Likewise.

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/alignas18.C: Expect name lookup error for U::X.
* g++.dg/cpp0x/forw_enum13.C: Expect name lookup error for
D3::A and D4<T>::A.
* g++.dg/parse/access13.C: Declare A::E::V to avoid name lookup
failure and preserve intent of the test.
* g++.dg/parse/enum11.C: Expect extra errors, matching the
non-template case.
* g++.dg/template/crash123.C: Avoid name lookup failure to
preserve intent of the test.
* g++.dg/template/crash124.C: Likewise.
* g++.dg/template/crash7.C: Adjust expected diagnostics.
* g++.dg/template/dtor6.C: Declare A::~A() to avoid name lookup
failure and preserve intent of the test.
* g++.dg/template/error22.C: Adjust expected diagnostics.
* g++.dg/template/static30.C: Avoid name lookup failure to
preserve intent of the test.
* g++.old-deja/g++.other/decl5.C: Adjust expected diagnostics.
* g++.dg/template/non-dependent34.C: New test.

Reviewed-by: Jason Merrill <jason@redhat.com>
12 months agolibbacktrace: better backtrace_print when no debug info
Ian Lance Taylor [Thu, 18 Jul 2024 00:36:25 +0000 (17:36 -0700)] 
libbacktrace: better backtrace_print when no debug info

Fixes https://github.com/ianlancetaylor/libbacktrace/issues/59

* print.c (print_syminfo_callback): New static function.
(print_callback): Call backtrace_syminfo if there is no function
or file name.

12 months agoDaily bump.
GCC Administrator [Thu, 18 Jul 2024 00:18:58 +0000 (00:18 +0000)] 
Daily bump.

12 months agolibbacktrace: add notes about dl_iterate_phdr to README
Ian Lance Taylor [Thu, 18 Jul 2024 00:02:56 +0000 (17:02 -0700)] 
libbacktrace: add notes about dl_iterate_phdr to README

* README: Add notes about dl_iterate_phdr.

12 months agotestsuite: Fix up pr111150* tests on i686-linux [PR111150]
Jakub Jelinek [Wed, 17 Jul 2024 21:47:17 +0000 (23:47 +0200)] 
testsuite: Fix up pr111150* tests on i686-linux [PR111150]

The tests FAIL on i686-linux due to unexpected -Wpsabi diagnostics.
Fixed as usually by adding -Wno-psabi to dg-options.

2024-07-17  Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/111150
* gcc.dg/tree-ssa/pr111150.c: Add -Wno-psabi to dg-options.
* g++.dg/tree-ssa/pr111150.C: Likewise.

12 months agoUse foreach, not lmap, for tcl <= 8.5 compat
Jørgen Kvalsvik [Sun, 14 Jul 2024 19:39:44 +0000 (21:39 +0200)] 
Use foreach, not lmap, for tcl <= 8.5 compat

lmap was introduced in tcl 8.6, and while it was released in 2012, lmap
does not really make too much of a difference to warrant the friction on
consverative (and relevant) systems.

gcc/testsuite/ChangeLog:

* lib/gcov.exp: Use foreach, not lmap, for tcl <= 8.5 compat.

12 months agortl-ssa: Fix move range canonicalisation [PR115929]
Richard Sandiford [Wed, 17 Jul 2024 18:38:12 +0000 (19:38 +0100)] 
rtl-ssa: Fix move range canonicalisation [PR115929]

In this PR, canonicalize_move_range walked off the end of a list
and triggered a null dereference.  There are multiple ways of fixing
that, but I think the approach taken in the patch should be
relatively efficient.

gcc/
PR rtl-optimization/115929
* rtl-ssa/movement.h (canonicalize_move_range): Check for null prev
and next insns and create an invalid move range for them.

gcc/testsuite/
PR rtl-optimization/115929
* gcc.dg/torture/pr115929-2.c: New test.