Similar for sqrt/sqrtl.
gcc/ChangeLog:
PR target/102464
* match.pd: Simplify (_Float16) sqrtf((float) a) to .SQRT(a)
when direct_internal_fn_supported_p, similar for sqrt/sqrtl.
gcc/testsuite/ChangeLog:
PR target/102464
* gcc.target/i386/pr102464-sqrtph.c: New test.
* gcc.target/i386/pr102464-sqrtsh.c: New test.
This fixes a latent issue exposed by now allowing VN_TOP in PHI
arguments. We may only use optimistic equality when merging values on
different edges, not when merging values on the same edge - in particular
we may not choose the undef value on any edge when there's a not undef
value as well.
2021-10-25 Richard Biener <rguenther@suse.de>
PR tree-optimization/102920
* tree-ssa-sccvn.h (expressions_equal_p): Add argument
controlling VN_TOP matching behavior.
* tree-ssa-sccvn.c (expressions_equal_p): Likewise.
(vn_phi_eq): Do not optimistically match VN_TOP.
* gcc.dg/torture/pr102920.c: New testcase.
This patch is to support transform in fast-math something like
_mm512_add_ph(x1, _mm512_fmadd_pch(a, b, _mm512_setzero_ph())) to
_mm512_fmadd_pch(a, b, x1).
And support transform _mm512_add_ph(x1, _mm512_fmul_pch(a, b))
to _mm512_fmadd_pch(a, b, x1).
gcc/ChangeLog:
* config/i386/sse.md (fma_<mode>_fadd_fmul): Add new
define_insn_and_split.
(fma_<mode>_fadd_fcmul):Likewise
(fma_<complexopname>_<mode>_fma_zero):Likewise
gcc/testsuite/ChangeLog:
* gcc.target/i386/avx512fp16-complex-fma.c: New test.
The behavior of the -mdisable-fpregs is confusing in that it doesn't
disable the use of the floating-point registers in all situations.
The -msoft-float disables the use of the floating-point registers in
all situations. The Linux kernel only needs to disable use of the
xmpyu instruction to avoid using the floating-point registers.
This change revises the -mdisable-fpregs option to disable the use of
the floating-point registers in all situations. It is now equivalent
to the -msoft-float option. A new -msoft-mult option is added to
disable use of the xmpyu instruction. The libgcc library can be
compiled with the -msoft-mult option to avoid using hardware integer
multiplication.
2021-10-24 John David Anglin <danglin@gcc.gnu.org>
gcc/ChangeLog:
* config/pa/pa-d.c (pa_d_handle_target_float_abi): Don't check
TARGET_DISABLE_FPREGS.
* config/pa/pa.c (fix_range): Use MASK_SOFT_FLOAT instead of
MASK_DISABLE_FPREGS.
(hppa_rtx_costs): Don't check TARGET_DISABLE_FPREGS. Adjust
cost of hardware integer multiplication.
(pa_conditional_register_usage): Don't check TARGET_DISABLE_FPREGS.
* config/pa/pa.h (INT14_OK_STRICT): Likewise.
* config/pa/pa.md: Don't check TARGET_DISABLE_FPREGS. Check
TARGET_SOFT_FLOAT in patterns that use xmpyu instruction.
* config/pa/pa.opt (mdisable-fpregs): Change target mask to
SOFT_FLOAT. Revise comment.
(msoft-float): New option.
The 'G' constraint only matches a float zero.
2021-10-24 John David Anglin <danglin@gcc.gnu.org>
gcc/ChangeLog:
* config/pa/pa.md: Don't use 'G' constraint in integer move patterns.
This patch cures the testsuite failure of bfin/20090914-3.c, which
currently FAILs on bfin-elf with "(test for excess errors)" due to:
20090914-3.c:3:1: warning: return type defaults to 'int' [-Wimplicit-int]
which is obviously not what this code was intended to test. Fixed by
turning the code into a function returning the final "fract32" result,
as simply specifying an "int" return type for main, results in the
entire function being optimized away, as the result is unused.
2021-10-24 Roger Sayle <roger@nextmovesoftware.com>
gcc/testsuite/ChangeLog
* gcc.target/bfin/20090914-3.c: Tweak test case.
Move bind-c-intent-out-2.f90 to gfortran.dg/ubsan for -fsanitize=undefined.
PR fortran/9262
* gfortran.dg/bind-c-intent-out-2.f90: Moved to ...
* gfortran.dg/ubsan/bind-c-intent-out-2.f90
On x86_64, V1TI mode holds a 128-bit integer value in a (vector) SSE
register (where regular TI mode uses a pair of 64-bit general purpose
scalar registers). This patch improves the implementation of AND, IOR,
XOR and NOT on these values.
The benefit is demonstrated by the following simple test program:
typedef unsigned __int128 v1ti __attribute__ ((__vector_size__ (16)));
v1ti and(v1ti x, v1ti y) { return x & y; }
v1ti ior(v1ti x, v1ti y) { return x | y; }
v1ti xor(v1ti x, v1ti y) { return x ^ y; }
v1ti not(v1ti x) { return ~x; }
For which GCC currently generates the rather large:
and: movdqa %xmm0, %xmm2
movq %xmm1, %rdx
movq %xmm0, %rax
andq %rdx, %rax
movhlps %xmm2, %xmm3
movhlps %xmm1, %xmm4
movq %rax, %xmm0
movq %xmm4, %rdx
movq %xmm3, %rax
andq %rdx, %rax
movq %rax, %xmm5
punpcklqdq %xmm5, %xmm0
ret
ior: movdqa %xmm0, %xmm2
movq %xmm1, %rdx
movq %xmm0, %rax
orq %rdx, %rax
movhlps %xmm2, %xmm3
movhlps %xmm1, %xmm4
movq %rax, %xmm0
movq %xmm4, %rdx
movq %xmm3, %rax
orq %rdx, %rax
movq %rax, %xmm5
punpcklqdq %xmm5, %xmm0
ret
xor: movdqa %xmm0, %xmm2
movq %xmm1, %rdx
movq %xmm0, %rax
xorq %rdx, %rax
movhlps %xmm2, %xmm3
movhlps %xmm1, %xmm4
movq %rax, %xmm0
movq %xmm4, %rdx
movq %xmm3, %rax
xorq %rdx, %rax
movq %rax, %xmm5
punpcklqdq %xmm5, %xmm0
ret
not: movdqa %xmm0, %xmm1
movq %xmm0, %rax
notq %rax
movhlps %xmm1, %xmm2
movq %rax, %xmm0
movq %xmm2, %rax
notq %rax
movq %rax, %xmm3
punpcklqdq %xmm3, %xmm0
ret
with this patch we now generate the much more efficient:
and: pand %xmm1, %xmm0
ret
ior: por %xmm1, %xmm0
ret
xor: pxor %xmm1, %xmm0
ret
not: pcmpeqd %xmm1, %xmm1
pxor %xmm1, %xmm0
ret
For my first few attempts at this patch I tried adding V1TI to the
existing VI and VI12_AVX_512F mode iterators, but these then have
dependencies on other iterators (and attributes), and so on until
everything ties itself into a knot, as V1TI mode isn't really a
first-class vector mode on x86_64. Hence I ultimately opted to use
simple stand-alone patterns (as used by the existing TF mode support).
2021-10-23 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
* config/i386/sse.md (<any_logic>v1ti3): New define_insn to
implement V1TImode AND, IOR and XOR on TARGET_SSE2 (and above).
(one_cmplv1ti2): New define expand.
gcc/testsuite/ChangeLog
* gcc.target/i386/sse2-v1ti-logic.c: New test case.
* gcc.target/i386/sse2-v1ti-logic-2.c: New test case.
std::make_any should be constrained so it can only be called if the
construction of the return value would be valid.
libstdc++-v3/ChangeLog:
PR libstdc++/102894
* include/std/any (make_any): Add SFINAE constraint.
* testsuite/20_util/any/102894.cc: New test.
This was not defined in the spec and not consistent in the
implementation causing incosistent behavior. After review we have
updated the CPU implementations and proposed the spec be updated to
specific that FPU tininess checks check for tininess before roudning.
Architecture change draft:
https://openrisc.io/proposals/p18-fpu-tininess
libgcc/ChangeLog:
* config/or1k/sfp-machine.h (_FP_TININESS_AFTER_ROUNDING):
Change to 0.
PR testsuite/102742
libbacktrace/ChangeLog:
* btest.c (MIN_DESCRIPTOR): New.
(MAX_DESCRIPTOR): Likewise.
(check_available_files): Likewise.
(check_open_files): Check only file descriptors that
were not available at the entry.
(main): Call check_available_files.
The following fixes the test for an exit edge I put in place for
the fix for PR45178 where I somehow misunderstood how the cyclic
list works.
2021-10-22 Richard Biener <rguenther@suse.de>
PR tree-optimization/102893
* tree-ssa-dce.c (find_obviously_necessary_stmts): Fix the
test for an exit edge.
* gcc.dg/tree-ssa/ssa-dce-9.c: New testcase.
The equivalence oracle creates a new equiv set at each def point,
killing any incoming equivalences, however in the path sensitive
oracle we create brand new equivalences at each PHI:
BB4:
BB8:
x_5 = PHI <y_8(4)>
Here we note that x_5 == y_8 at the end of the path.
The current code is intersecting this new equivalence with previously
known equivalences coming into the path. This is incorrect, as this
is a new definition. This patch kills any known equivalence before we
register a new one.
This hasn't caused problems so far, but upcoming changes to the
pipeline has us threading more aggressively and triggering corner
cases where this causes incorrect code.
I have tested this patch with the usual regstrap cycle. I have also
hacked a compiler comparing the old and new behavior to see if we were
previously threading paths where the decision was made due to invalid
equivalences. Luckily, there were no such paths, but there were 22
paths in a set of .ii files where disregarding incoming relations
allowed us to thread the path. This is a miniscule improvement,
but we moved a handful of thredable paths earlier in the pipeline,
which is always good.
Tested on x86-64 Linux.
Co-authored-by: Andrew MacLeod <amacleod@redhat.com>
gcc/ChangeLog:
* gimple-range-path.cc (path_range_query::compute_phi_relations):
Kill any global relations we may know before registering a new
one.
* value-relation.cc (path_oracle::killing_def): New.
* value-relation.h (path_oracle::killing_def): New.
The PR shows that we fail to CSE PHIs containing (different)
default definitions due to the fact on how we now handle
on-demand build of VN_INFO. The following fixes this in the
same way the PHI visitation code does.
On gcc.dg/ubsan/pr81981.c this causes one expected warning to be
elided since the uninit pass sees the change
<bb 4> [local count: 1073741824]:
# u$0_2 = PHI <u$0_5(D)(3), i_3(D)(5)>
- # cstore_11 = PHI <t$0_6(D)(3), i_3(D)(5)>
v = u$0_2;
- return cstore_11;
+ return u$0_2;
and thus only one of the conditionally uninitialized uses (the
other became dead). I have XFAILed the missing diagnostic,
I don't see a way to preserve that.
2021-10-22 Richard Biener <rguenther@suse.de>
PR bootstrap/102681
* tree-ssa-sccvn.c (vn_phi_insert): For undefined SSA args
record VN_TOP.
(vn_phi_lookup): Likewise.
* gcc.dg/tree-ssa/ssa-fre-97.c: New testcase.
* gcc.dg/ubsan/pr81981.c: XFAIL one case.
The man pages generated by Doxygen show internal header files, not the
standard headers that users actually care about. The run_doxygen script
uses the doc/doxygen/stdheader.cc program to address that, but it
doesn't work. It only tries to fix headers with underscores in the
names, which doesn't work for <bits/align.h> or <bits/fsteam.tcc>. It
isn't prepared for the strings like "bits/stl_set\&.h" that are produced
by Doxygen. It doesn't know about many headers that have been added
since it was written. And the run_doxygen script fails to use its output
correctly to modify the man pages. Additionally, run_doxygen doesn't
know about new nested namespaces like std::filesystem and std::ranges.
This change rewrites the stdheader.cc program to do a better job of
finding the right header. The run_doxygen script now uses the just-built
compiler to build stdheader.cc and actually uses its output. And the
script now knows about other nested namespaces.
The stdheader.cc program might be unnecessary if we consistently used
@headername tags in the Doxygen comments, but we don't (and probably
never will).
A problem that remains after this change is that all the free function
defined in namespace std get dumped into a single man page for std(3),
without detailed descriptions. We don't even install that std(3) page,
but remove it before installation. That means only classes are
documented in man pages (including many internal ones that should not be
publicly documented such as _Deque_base and _Tuple_impl).
libstdc++-v3/ChangeLog:
* doc/doxygen/stdheader.cc: Refactor. Use C++23. Add new
headers.
* scripts/run_doxygen: Fix post-processing of #include
directives in man pages. Use new xg++ to compile helper program.
If the listen call fails then 'goto fail' will jump to that label and
use freeaddrinfo again. Set the pointer to null to prevent that.
libcody/ChangeLog:
* netserver.cc (ListenInet6): Set pointer to null after
deallocation.
libstdc++-v3/ChangeLog:
* include/std/ranges (istream_view): Replace this function
template with an alias template as per P2432R1.
(wistream_view): Define as per P2432R1.
(views::_Istream, views::istream): Likewise.
* testsuite/std/ranges/istream_view.cc (test07): New test.
This implements P1739R4 along with the resolution for LWG 3407 which
corrects the paper's wording.
libstdc++-v3/ChangeLog:
* include/bits/ranges_util.h (views::_Drop): Forward declare.
(subrange): Befriend views::_Drop.
(subrange::_S_store_size): Declare constexpr instead of just
const, remove obsolete comment.
* include/std/ranges (views::__detail::__is_empty_view): Define.
(views::__detail::__is_basic_string_view): Likewise.
(views::__detail::__is_subrange): Likewise.
(views::__detail::__is_iota_view): Likewise.
(views::__detail::__can_take_view): Rename template parm _Tp to _Dp.
(views::_Take): Rename template parm _Tp to _Dp, make it non-deducible
and fix it to range_difference_t<_Range>. Implement P1739R4 and
LWG 3407 changes.
(views::__detail::__can_drop_view): Rename template parm _Tp to _Dp.
(views::_Drop): As with views::_Take.
(views::_Counted): Implement P1739R4 changes.
* include/std/span (__detail::__is_std_span): Rename to ...
(__detail::__is_span): ... this and turn it into a variable
template.
(__detail::__is_std_array): Turn it into a variable template.
(span::span): Adjust uses of __is_std_span and __is_std_array
accordingly.
* testsuite/std/ranges/adaptors/p1739.cc: New test.
These vector type definitions are an artifact from the initial commit
that added the AArch64 port.
gcc/ChangeLog:
2021-10-15 Jonathan Wright <jonathan.wright@arm.com>
* config/aarch64/arm_neon.h (__STRUCTN): Delete function
macro and all invocations.
The HF vector move have been updated to align with HI vector,
adjust according testcase for _Float16 vector load and store.
gcc/testsuite/ChangeLog:
* gcc.target/i386/avx512fp16-13.c: Adjust scan-assembler for
xmm/ymm load/store.
With Ranger being used in more than EVRP, the debug output should no longer
be tied up with the EVRP mode flag.
* doc/invoke.texi (ranger-debug): Document.
* flag-types.h (enum ranger_debug): New.
(enum evrp_mode): Remove debug values.
* gimple-range-cache.cc (DEBUG_RANGE_CACHE): Use new debug flag.
* gimple-range-gori.cc (gori_compute::gori_compute): Ditto.
* gimple-range.cc (gimple_ranger::gimple_ranger): Ditto.
* gimple-ssa-evrp.c (hybrid_folder::choose_value): Ditto.
(execute_early_vrp): Use evrp-mode directly.
* params.opt (enum evrp_mode): Remove debug values.
(ranger-debug): New.
(ranger-logical-depth): Relocate to be in alphabetical order.
Add 2 new params to select between VRP and RANGER to be used for each pass.
* doc/invoke.texi: (vrp1-mode, vrp2-mode): Document.
* flag-types.h: (enum vrp_mode): New.
* params.opt: (vrp1-mode, vrp2-mode): New.
* tree-vrp.c (vrp_pass_num): New.
(pass_vrp::pass_vrp): Set pass number.
(pass_vrp::execute): Choose which VRP mode to execute.
Consolidate the RVRP folder into a single "execute_vrp" routine that mimics
the format used by the vrp1 and vrp2 passes. Relocate into the tree-vrp file.
* gimple-ssa-evrp.c (class rvrp_folder): Move to tree-vrp.c.
(execute_early_vrp): For ranger only mode, invoke ranger_vrp.
* tree-vrp.c (class rvrp_folder): Relocate here.
(execute_ranger_vrp): New.
* tree-vrp.h (execute_ranger_vrp): Export.
PR debug/102585
PR bootstrap/102766
gcc/ChangeLog:
* opts.c (finish_options): Process flag_var_tracking* options
here as they can be adjusted by optimize attribute.
Process also flag_syntax_only and flag_gtoggle.
* toplev.c (process_options): Remove it here.
* common.opt: Make debug_nonbind_markers_p as PerFunction
attribute as it depends on optimization level.
gcc/testsuite/ChangeLog:
* gcc.dg/pr102585.c: New test.
PR 102505 is a situation where of SRA takes its initial top-level
access size from a get_ref_base_and_extent called on a COMPONENT_REF,
and thus derived frm the FIELD_DECL, which however does not include a
virtual base. Total scalarization then goes on traversing the type,
which however has virtual base past the non-virtual bits, tricking SRA
to create sub-accesses outside of the supposedly encompassing
accesses, which in turn triggers the verifier within the pass.
The patch below fixes that by failing total scalarization when this
situation is detected.
gcc/ChangeLog:
2021-10-20 Martin Jambor <mjambor@suse.cz>
PR tree-optimization/102505
* tree-sra.c (totally_scalarize_subtree): Check that the
encountered field fits within the acces we would like to put it
in.
gcc/testsuite/ChangeLog:
2021-10-20 Martin Jambor <mjambor@suse.cz>
PR tree-optimization/102505
* g++.dg/torture/pr102505.C: New test.
I've tested this patch on the wrong tree, and picked up the test changes
in a pending patch, without which this patch is no longer obvious.
Plus, it causes a regression in an invalid test I've recommended we remove.
I'm reverting this patch until the dependencies are reviewed.
Sorry for the noise.
gcc/ChangeLog:
* tree-ssa-threadbackward.c
(back_threader::maybe_register_path): Remove circular paths check.
I just realized that when I moved the langhook call I failed to
move the initial debug_hooks setting whose comment mentions the
langhook as reason.
2021-10-21 Richard Biener <rguenther@suse.de>
* toplev.c (process_options): Move the initial debug_hooks
setting ...
(toplev::main): ... before the call of the post_options
langhook.
