2019-11-15 Richard Biener <rguenther@suse.de>
PR tree-optimization/92324
* tree-vect-loop.c (vect_create_epilog_for_reduction): Fix
singedness of SLP reduction epilouge operations. Also reduce
the vector width for SLP reductions before doing elementwise
operations if possible.
* gcc.dg/vect/pr92324-4.c: New testcase.
From-SVN: r278289
2019-11-15 Paul Thomas <pault@gcc.gnu.org>
PR fortran/69654
* trans-expr.c (gfc_trans_structure_assign): Move assignment to
'cm' after treatment of C pointer types and test that the type
has been completely built before it. Add an assert that the
backend_decl for each component exists.
2019-11-15 Paul Thomas <pault@gcc.gnu.org>
PR fortran/69654
* gfortran.dg/derived_init_6.f90: New test.
From-SVN: r278287
Set global epilogue_completed when skipping pro_and_epilogue pass
When compiling RTL functions marked to start at a pass after the reload
pass, `skip_pass` is used to mark the reload pass as having completed
since many patterns use the `reload_completed` variable to determine
whether to run or not.
Here we do the same for the `epilogue_completed` variable and the
pro_and_epilogue pass.
Also include a testcase that relies on the availability of a
define_split in the aarch64 backend that is conditioned on this
`epilogue_completed` variable.
regtest done on native aarch64
regtest done on native x64_86
gcc/ChangeLog:
2019-11-15 Matthew Malcomson <matthew.malcomson@arm.com>
* passes.c (skip_pass): Set epilogue_completed if skipping the
pro_and_epilogue pass.
gcc/testsuite/ChangeLog:
2019-11-15 Matthew Malcomson <matthew.malcomson@arm.com>
* gcc.dg/rtl/aarch64/test-epilogue-set.c: New test.
From-SVN: r278285
2019-11-15 Andrew Stubbs <ams@codesourcery.com>
libgomp/
* testsuite/libgomp.c/target-print-1.c: New file.
* testsuite/libgomp.fortran/target-print-1.f90: New file.
* testsuite/libgomp.oacc-c/print-1.c: New file.
* testsuite/libgomp.oacc-fortran/print-1.f90: New file.
From-SVN: r278284
Hi there,
When compiling an __RTL function that has an invalid "startwith" pass we
currently don't run the dfinish cleanup pass. This means we ICE on the next
function.
This change ensures that all state is cleaned up for the next function
to run correctly.
As an example, before this change the following code would ICE when compiling
the function `foo2` because the "peephole2" pass is not run at optimisation
level -O0.
When compiled with
./aarch64-none-linux-gnu-gcc -O0 -S missed-pass-error.c -o test.s
```
int __RTL (startwith ("peephole2")) badfoo ()
{
(function "badfoo"
(insn-chain
(block 2
(edge-from entry (flags "FALLTHRU"))
(cnote 3 [bb 2] NOTE_INSN_BASIC_BLOCK)
(cinsn 101 (set (reg:DI x19) (reg:DI x0)))
(cinsn 10 (use (reg/i:SI x19)))
(edge-to exit (flags "FALLTHRU"))
) ;; block 2
) ;; insn-chain
) ;; function "foo2"
}
int __RTL (startwith ("final")) foo2 ()
{
(function "foo2"
(insn-chain
(block 2
(edge-from entry (flags "FALLTHRU"))
(cnote 3 [bb 2] NOTE_INSN_BASIC_BLOCK)
(cinsn 101 (set (reg:DI x19) (reg:DI x0)))
(cinsn 10 (use (reg/i:SI x19)))
(edge-to exit (flags "FALLTHRU"))
) ;; block 2
) ;; insn-chain
) ;; function "foo2"
}
```
Now it silently ignores the __RTL function and successfully compiles foo2.
regtest done on aarch64
regtest done on x86_64
OK for trunk?
gcc/ChangeLog:
2019-11-15 Matthew Malcomson <matthew.malcomson@arm.com>
* passes.c (should_skip_pass_p): Always run "dfinish".
gcc/testsuite/ChangeLog:
2019-11-15 Matthew Malcomson <matthew.malcomson@arm.com>
* gcc.dg/rtl/aarch64/missed-pass-error.c: New test.
From-SVN: r278283
2019-11-15 Richard Biener <rguenther@suse.de>
* ipa-inline.c (inline_small_functions): Move assignment
to next before call destroying edge.
From-SVN: r278282
2019-11-15 Richard Biener <rguenther@suse.de>
PR tree-optimization/92039
PR tree-optimization/91975
* tree-ssa-loop-ivcanon.c (constant_after_peeling): Revert
previous change, treat invariants consistently as non-constant.
(tree_estimate_loop_size): Ternary ops with just the first op
constant are not optimized away.
* gcc.dg/tree-ssa/cunroll-2.c: Revert to state previous to
unroller adjustment.
* g++.dg/tree-ssa/ivopts-3.C: Likewise.
From-SVN: r278281
* gimplify.c (gimplify_call_expr): Don't call
omp_resolve_declare_variant after gimplification.
* omp-general.c (omp_context_selector_matches): For isa that might
match in some other function, defer if in declare simd function.
(omp_context_compute_score): Don't look for " score" in construct
trait set. Set *score to -1 if it can't ever match.
(omp_resolve_declare_variant): If any variants need to be deferred,
don't punt immediately, but compute scores of all variants and if
ther eis a score winner that doesn't need to be deferred, return that.
* c-c++-common/gomp/declare-variant-13.c: New test.
From-SVN: r278280
next is initialized only in the loop before, it is never updated
in it's own loop.
gcc/ChangeLog:
2019-11-15 Xiong Hu Luo <luoxhu@linux.ibm.com>
* ipa-inline.c (inline_small_functions): Update iterator of next.
From-SVN: r278277
When the compiler writes an inlinable function to the export data,
parameter names are written out (in Export::write_name) using the
Gogo::message_name as opposed to a raw/encoded name. This means that
sink parameters (those named "_") get created with the name "_"
instead of "._" (the name created by the lexer/parser). This confuses
Gogo::is_sink_name, which looks for the latter sequence and not just
"_". This can cause issues later on if an inlinable function is
imported and fed through the rest of the compiler (things that are
sinks are no recognized as such). To fix these issues, change
Gogo::is_sink_name to return true for either variants ("_" or "._").
Fixesgolang/go#35586.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/207259
From-SVN: r278275
* include/Makefile.am: Add <stop_token> header.
* include/Makefile.in: Regenerate.
* include/std/condition_variable: Add overloads for stop_token support
to condition_variable_any.
* include/std/stop_token: New file.
* include/std/thread: Add jthread type.
* include/std/version (__cpp_lib_jthread): New value.
* testsuite/30_threads/condition_variable_any/stop_token/1.cc: New test.
* testsuite/30_threads/condition_variable_any/stop_token/2.cc: New test.
* testsuite/30_threads/condition_variable_any/stop_token/wait_on.cc: New test.
* testsuite/30_threads/jthread/1.cc: New test.
* testsuite/30_threads/jthread/2.cc: New test.
* testsuite/30_threads/jthread/jthread.cc: New test.
* testsuite/30_threads/stop_token/1.cc: New test.
* testsuite/30_threads/stop_token/2.cc: New test.
* testsuite/30_threads/stop_token/stop_token.cc: New test.
From-SVN: r278274
When adding C2x attribute support, some [[fallthrough]] support
appeared as a side-effect because of code for that attribute going
through separate paths from the normal attribute handling.
However, going through those paths without the normal attribute
handlers meant that certain checks, such as for the invalid usage
[[fallthrough()]], did not operate. This patch improves checks by
adding this attribute to the standard attribute table, so that the
parser knows it expects no arguments, along with adding an explicit
check for "[[fallthrough]];" attribute-declarations at top level. As
with other attributes, there are still cases where warnings should be
pedwarns because C2x constraints are violated, but this patch improves
the attribute handling.
Bootstrapped with no regressions on x86_64-pc-linux-gnu.
gcc/c:
* c-decl.c (std_attribute_table): Add fallthrough.
* c-parser.c (c_parser_declaration_or_fndef): Diagnose fallthrough
attribute at top level.
gcc/c-family:
* c-attribs.c (handle_fallthrough_attribute): Remove static.
* c-common.h (handle_fallthrough_attribute): Declare.
gcc/testsuite:
* gcc.dg/c2x-attr-fallthrough-2.c,
gcc.dg/c2x-attr-fallthrough-3.c: New tests.
From-SVN: r278273
This patch adds support for the C2x [[deprecated]] attribute. All the
actual logic for generating warnings can be identical to the GNU
__attribute__ ((deprecated)), as can the attribute handler, so this is
just a matter of wiring things up appropriately and adding the checks
specified in the standard. Unlike for C++, this patch gives
"deprecated" an entry in a table of standard attributes rather than
remapping it internally to the GNU attribute, as that seems a cleaner
approach to me.
Specifically, the only form of arguments to the attribute permitted in
the standard is (string-literal); empty parentheses are not permitted
in the case of no arguments, and a string literal (which includes
concatenated adjacent string literals, because concatenation is an
earlier phase of translation) cannot have further redundant
parentheses around it. For the case of empty parentheses, this patch
makes the C parser disallow them for all known attributes using the
[[]] syntax, as done for C++. For string literals (where the C++
front end is missing the check to avoid redundant parentheses, 92521
filed for that issue), a special case is inserted in the C parser.
A known issue that I think can be addressed later as a bug fix is that
the warnings for the attribute being ignored in certain cases
(attribute declarations, statements, most uses on types) ought to be
pedwarns, as those usages are constraint violations.
Bad handling of wide string literals with this attribute is also a
pre-existing bug (91182 - although that's filed as a C++ bug, the code
in question is language-independent, in tree.c).
Bootstrapped with no regressions on x86_64-pc-linux-gnu.
gcc/c:
* c-decl.c (std_attribute_table): New.
(c_init_decl_processing): Register attributes from
std_attribute_table.
* c-parser.c (c_parser_attribute_arguments): Add arguments
require_string and allow_empty_args. All callers changed.
(c_parser_std_attribute): Set require_string argument for
"deprecated" attribute.
gcc/c-family:
* c-attribs.c (handle_deprecated_attribute): Remove static.
* c-common.h (handle_deprecated_attribute): Declare.
gcc/testsuite:
* gcc.dg/c2x-attr-deprecated-1.c, gcc.dg/c2x-attr-deprecated-2.c,
gcc.dg/c2x-attr-deprecated-3.c: New tests.
From-SVN: r278268
2019-11-14 Kwok Cheung Yeung <kcy@codesourcery.com>
gcc/
* lra-spills.c (assign_spill_hard_regs): Check that the spill
register is suitable for the mode.
From-SVN: r278267
C2x adds u8'' character constants to C. This patch adds the
corresponding GCC support.
Most of the support was already present for C++ and just needed
enabling for C2x. However, in C2x these constants have type unsigned
char, which required corresponding adjustments in the compiler and the
preprocessor to give them that type for C.
For C, it seems clear to me that having type unsigned char means the
constants are unsigned in the preprocessor (and thus treated as having
type uintmax_t in #if conditionals), so this patch implements that. I
included a conditional in the libcpp change to avoid affecting
signedness for C++, but I'm not sure if in fact these constants should
also be unsigned in the preprocessor for C++ in which case that
!CPP_OPTION (pfile, cplusplus) conditional would not be needed.
Bootstrapped with no regressions on x86_64-pc-linux-gnu.
gcc/c:
* c-parser.c (c_parser_postfix_expression)
(c_parser_check_literal_zero): Handle CPP_UTF8CHAR.
* gimple-parser.c (c_parser_gimple_postfix_expression): Likewise.
gcc/c-family:
* c-lex.c (lex_charconst): Make CPP_UTF8CHAR constants unsigned
char for C.
gcc/testsuite:
* gcc.dg/c11-utf8char-1.c, gcc.dg/c2x-utf8char-1.c,
gcc.dg/c2x-utf8char-2.c, gcc.dg/c2x-utf8char-3.c,
gcc.dg/gnu2x-utf8char-1.c: New tests.
libcpp:
* charset.c (narrow_str_to_charconst): Make CPP_UTF8CHAR constants
unsigned for C.
* init.c (lang_defaults): Set utf8_char_literals for GNUC2X and
STDC2X.
From-SVN: r278265
gcc.dg/vect/bb-slp-40.c was failing on some targets because the
explicit dg-options overrode things like -maltivec. This patch
uses dg-additional-options instead.
Also, it seems safer not to require exactly 1 instance of each message,
since that depends on the target vector length.
gcc.dg/vect/bb-slp-41.c contained invariant constructors that are
vectorised on AArch64 (foo) and constructors that aren't (bar).
This meant that the number of times we print "Found vectorizable
constructor" depended on how many vector sizes we try, since we'd
print it for each failed attempt.
In foo, we create invariant { b[0], ... } and { b[1], ... },
and the test is making sure that the two separate invariant vectors
can be fed from the same vector load at b. This is a different case
from bb-slp-40.c, where the constructors are naturally separate.
(The expected count is 4 rather than 2 because we can vectorise the
epilogue too.)
However, due to limitations in the loop vectoriser, we still do the
addition of { b[0], ... } and { b[1], ... } in the loop. Hopefully
that'll be fixed at some point, so this patch adds an alternative test
that directly needs 4 separate invariant constructors. E.g. with Joel's
SLP optimisation, the new test generates:
ldr q4, [x1]
dup v7.4s, v4.s[0]
dup v6.4s, v4.s[1]
dup v5.4s, v4.s[2]
dup v4.4s, v4.s[3]
instead of the somewhat bizarre:
ldp s6, s5, [x1, 4]
ldr s4, [x1, 12]
ld1r {v7.4s}, [x1]
dup v6.4s, v6.s[0]
dup v5.4s, v5.s[0]
dup v4.4s, v4.s[0]
The patch then disables vectorisation of the original foo in
bb-vect-slp-41.c, so that we get the same correctness testing
for bar but don't need to test for specific counts.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/testsuite/
PR testsuite/92366
* gcc.dg/vect/bb-slp-40.c: Use dg-additional-options instead
of dg-options. Remove expected counts.
* gcc.dg/vect/bb-slp-41.c: Remove dg-options and explicit
dg-do run. Suppress vectorization of foo.
* gcc.dg/vect/bb-slp-42.c: New test.
From-SVN: r278262
2019-11-14 Andrew MacLeod <amacleod@redhat.com>
PR tree-optimization/92506
* range-op.cc (range_operator::fold_range): Start with range undefined.
(operator_abs::wi_fold): Fix wrong line copy... With wrapv, abs with
overflow is varying.
From-SVN: r278259
This is a follow-up to
https://gcc.gnu.org/ml/gcc-patches/2019-11/msg00919.html (r278095).
Dominance info is deleted even if we don't perform jump threading.
Since the whole point of this pass is to perform jump threading (other
cleanups are not valuable at this point), skip it completely when
flag_thread_jumps is not set.
gcc/ChangeLog:
2019-11-14 Ilya Leoshkevich <iii@linux.ibm.com>
PR rtl-optimization/92430
* cfgcleanup.c (pass_jump_after_combine::gate): New function.
(pass_jump_after_combine::execute): Perform jump threading
unconditionally.
From-SVN: r278254
2019-11-13 Jerome Lambourg <lambourg@adacore.com>
Doug Rupp <rupp@adacore.com>
Olivier Hainque <hainque@adacore.com>
gcc/
* config.gcc: Collapse the arm-vxworks entries into
a single arm-wrs-vxworks7* one, bpabi based. Update
the default cpu from arm8 to armv7-a
* config/arm/vxworks.h (CC1_SPEC): Simplify, knowing that
we always use ARM_UNWIND_INFO.
(DWARF2_UNWIND_INFO): Remove redefinition.
(ARM_TARGET2_DWARF_FORMAT): Likewise.
(VXWORKS_PERSONALITY): Define, to "llvm".
(VXWORKS_EXTRA_LIBS_RTP): Define, to "-lllvm".
libgcc/
* config.host: Collapse the arm-vxworks entries into
a single arm-wrs-vxworks7* one.
* config/arm/unwind-arm-vxworks.c: Update comments. Provide
__gnu_Unwind_Find_exidx and a weak dummy __cxa_type_match for
kernel modules, to be overriden by libstdc++ when we link with
it. Rely on externally provided __exidx_start/end.
Co-Authored-By: Doug Rupp <rupp@adacore.com>
Co-Authored-By: Olivier Hainque <hainque@adacore.com>
From-SVN: r278253
2019-11-14 Jerome Lambourg <lambourg@adacore.com>
* config/arm/vxworks.h (TARGET_OS_CPP_BUILTINS): Use
_VX_CPU instead of CPU and handle arm_arch8.
From-SVN: r278252
2019-11-12 Olivier Hainque <hainque@adacore.com>
libgcc/
* config/t-gthr-vxworksae: New file, add all the gthr-vxworks
sources except the cxx0x support to LIB2ADDEH. We don't support
cxx0x on AE/653.
* config/t-vxworksae: New file.
* config.host: Handle *-*-vxworksae: Add the two aforementioned
Makefile fragment files at their expected position in the tmake_file
list, in accordance with what is done for other VxWorks variants.
From-SVN: r278250
2019-11-12 Corentin Gay <gay@adacore.com>
Jerome Lambourg <lambourg@adacore.com>
Olivier Hainque <hainque@adacore.com>
libgcc/
* config/t-gthr-vxworks: New file, add all the gthr-vxworks
sources to LIB2ADDEH.
* config/t-vxworks: Remove adjustments to LIB2ADDEH.
* config/t-vxworks7: Likewise.
* config.host: Append a block at the end of the file to add the
t-gthr files to the tmake_file list for VxWorks after everything
else.
* config/vxlib.c: Rename as gthr-vxworks.c.
* config/vxlib-tls.c: Rename as gthr-vxworks-tls.c.
* config/gthr-vxworks.h: Simplify a few comments. Expose a TAS
API and a basic error checking API, both internal. Simplify the
__gthread_once_t type definition and initializers. Add sections
for condition variables support and for the C++0x thread support,
conditioned against Vx653 for the latter.
* config/gthr-vxworks.c (__gthread_once): Simplify comments and
implementation, leveraging the TAS internal API.
* config/gthr-vxworks-tls.c: Introduce an internal TLS data access
API, leveraging the general availability of TLS services in VxWorks7
post SR6xxx.
(__gthread_setspecific, __gthread_setspecific): Use it.
(tls_delete_hook): Likewise, and simplify the enter/leave dtor logic.
* config/gthr-vxworks-cond.c: New file. GTHREAD_COND variable
support based on VxWorks primitives.
* config/gthr-vxworks-thread.c: New file. GTHREAD_CXX0X support
based on VxWorks primitives.
Co-Authored-By: Jerome Lambourg <lambourg@adacore.com>
Co-Authored-By: Olivier Hainque <hainque@adacore.com>
From-SVN: r278249
2019-11-06 Jerome Lambourg <lambourg@adacore.com>
Olivier Hainque <hainque@adacore.com>
libgcc/
* config/vxcrtstuff.c: New file.
* config/t-vxcrtstuff: New Makefile fragment.
* config.host: Append t-vxcrtstuff to the tmake_file list
on all VxWorks ports using dwarf for table based EH.
gcc/
* config/vx-common.h (USE_TM_CLONE_REGISTRY): Remove
definition, pointless with a VxWorks specific version
of crtstuff.
(DWARF2_UNWIND_INFO): Conditionalize on !ARM_UNWIND_INFO.
* config/vxworks.h (VX_CRTBEGIN_SPEC, VX_CRTEND_SPEC):
New local macros, controlling the addition of vxworks specific
crtstuff objects depending on the EH mechanism and kind of
module being linked.
(VXWORKS_STARTFILE_SPEC, VXWORKS_ENDFILE_SPEC): Use them.
Co-Authored-By: Olivier Hainque <hainque@adacore.com>
From-SVN: r278248
2019-11-06 Pat Bernardi <bernardi@adacore.com>
Jerome Lambourg <lambourg@adacore.com>
Olivier Hainque <hainque@adacore.com>
gcc/
* config.gcc: Add comment to introduce the TARGET_VXWORKS
commong macro definitions, conveying VXWORKS7 or 64bit general
variations. Add a block to set gcc_cv_initfini_array
unconditionally to "yes" for VxWorks7.
config/vx-common.h (VXWORKS_CC1_SPEC): New macro, empty string
by default. Update some comments.
config/vxworks.h (VXWORKS_EXTRA_LIBS_RTP): New macro, empty by
default, to be added the end of VXWORKS_LIBS_RTP.
(VXWORKS_LIBS_RTP): Replace hardcoded part by VXWORKS_BASE_LIBS_RTP
and append VXWORKS_EXTRA_LIBS_RTP, both of which specific ports may
redefine.
(VXWORKS_NET_LIBS_RTP): Account for VxWorks7 specificities.
(VXWORKS_CC1_SPEC): Common base definition, with VxWorks7 variation
to account for the now available TLS abilities.
(TARGET_LIBC_HAS_FUNCTION): Account for VxWorks7 abilities.
(VXWORKS_HAVE_TLS): Likewise.
Co-Authored-By: Jerome Lambourg <lambourg@adacore.com>
Co-Authored-By: Olivier Hainque <hainque@adacore.com>
From-SVN: r278247
If the statements in an SLP node aren't similar enough to be vectorised,
or aren't something the vectoriser has code to handle, the BB vectoriser
tries building the vector from scalars instead. This patch does the
same thing if we're able to build a viable-looking tree but fail later
during the analysis phase, e.g. because the target doesn't support a
particular vector operation.
This is needed to avoid regressions with a later patch.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-slp.c (vect_contains_pattern_stmt_p): New function.
(vect_slp_convert_to_external): Likewise.
(vect_slp_analyze_node_operations): If analysis fails, try building
the node from scalars instead.
gcc/testsuite/
* gcc.dg/vect/bb-slp-div-2.c: New test.
From-SVN: r278246
This patch adds AArch64 patterns for converting between 64-bit and
128-bit integer vectors, and makes the vectoriser and expand pass
use them.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-cfg.c (verify_gimple_assign_unary): Handle conversions
between vector types.
* tree-vect-stmts.c (vectorizable_conversion): Extend the
non-widening and non-narrowing path to handle standard
conversion codes, if the target supports them.
* expr.c (convert_move): Try using the extend and truncate optabs
for vectors.
* optabs-tree.c (supportable_convert_operation): Likewise.
* config/aarch64/iterators.md (Vnarroqw): New iterator.
* config/aarch64/aarch64-simd.md (<optab><Vnarrowq><mode>2)
(trunc<mode><Vnarrowq>2): New patterns.
gcc/testsuite/
* gcc.dg/vect/bb-slp-pr69907.c: Do not expect BB vectorization
to fail for aarch64 targets.
* gcc.dg/vect/no-scevccp-outer-12.c: Expect the test to pass
on aarch64 targets.
* gcc.dg/vect/vect-double-reduc-5.c: Likewise.
* gcc.dg/vect/vect-outer-4e.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_5.c: New test.
* gcc.target/aarch64/vect_mixed_sizes_6.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_7.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_8.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_9.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_10.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_11.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_12.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_13.c: Likewise.
From-SVN: r278245
Although a previous patch allowed mixed vector sizes within a vector
region, we generally still required equal vector sizes within a vector
stmt. Specifically, vect_get_vector_types_for_stmt computes two vector
types: the vector type corresponding to STMT_VINFO_VECTYPE and the
vector type that determines the minimum vectorisation factor for the
stmt ("nunits_vectype"). It then required these two types to be
the same size.
There doesn't seem to be any need for that restriction though. AFAICT,
all vectorizable_* functions either do their own compatibility checks
or don't need to do them (because gimple guarantees that the scalar
types are compatible).
It should always be the case that nunits_vectype has at least as many
elements as the other vectype, but that's something we can assert for.
I couldn't resist a couple of other tweaks while there:
- there's no need to compute nunits_vectype if its element type is
the same as STMT_VINFO_VECTYPE's.
- it's useful to distinguish the nunits_vectype from the main vectype
in dump messages
- when reusing the existing STMT_VINFO_VECTYPE, it's useful to say so
in the dump, and say what the type is
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-stmts.c (vect_get_vector_types_for_stmt): Don't
require vectype and nunits_vectype to have the same size;
instead assert that nunits_vectype has at least as many
elements as vectype. Don't compute a separate nunits_vectype
if the scalar type is obviously the same as vectype's.
Tweak dump messages.
From-SVN: r278244
This patch makes the vectoriser try mixtures of 64-bit and 128-bit
vector modes on AArch64. It fixes some existing XFAILs and allows
kernel 24 from the Livermore Loops test to be vectorised (by using
a mixture of V2DF and V2SI).
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/aarch64.c (aarch64_vectorize_related_mode): New
function.
(aarch64_autovectorize_vector_modes): Also add V4HImode and V2SImode.
(TARGET_VECTORIZE_RELATED_MODE): Define.
gcc/testsuite/
* gcc.dg/vect/vect-outer-4f.c: Expect the test to pass on aarch64
targets.
* gcc.dg/vect/vect-outer-4g.c: Likewise.
* gcc.dg/vect/vect-outer-4k.c: Likewise.
* gcc.dg/vect/vect-outer-4l.c: Likewise.
* gfortran.dg/vect/vect-8.f90: Expect kernel 24 to be vectorized
for aarch64.
* gcc.target/aarch64/vect_mixed_sizes_1.c: New test.
* gcc.target/aarch64/vect_mixed_sizes_2.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_3.c: Likewise.
* gcc.target/aarch64/vect_mixed_sizes_4.c: Likewise.
From-SVN: r278243
A later patch makes the AArch64 port add four entries to
autovectorize_vector_modes. Each entry describes a different
vector mode assignment for vector code that mixes 8-bit, 16-bit,
32-bit and 64-bit elements. But if (as usual) the vector code has
fewer element sizes than that, we could end up trying the same
combination of vector modes multiple times. This patch adds a
check to prevent that.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vectorizer.h (vec_info::mode_set): New typedef.
(vec_info::used_vector_mode): New member variable.
(vect_chooses_same_modes_p): Declare.
* tree-vect-stmts.c (get_vectype_for_scalar_type): Record each
chosen vector mode in vec_info::used_vector_mode.
(vect_chooses_same_modes_p): New function.
* tree-vect-loop.c (vect_analyze_loop): Use it to avoid trying
the same vector statements multiple times.
* tree-vect-slp.c (vect_slp_bb_region): Likewise.
From-SVN: r278242
After previous patches, it's now possible to make the vectoriser
support multiple vector sizes in the same vector region, using
related_vector_mode to pick the right vector mode for a given
element mode. No port yet takes advantage of this, but I have
a follow-on patch for AArch64.
This patch also seemed like a good opportunity to add some more dump
messages: one to make it clear which vector size/mode was being used
when analysis passed or failed, and another to say when we've decided
to skip a redundant vector size/mode.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* machmode.h (opt_machine_mode::operator==): New function.
(opt_machine_mode::operator!=): Likewise.
* tree-vectorizer.h (vec_info::vector_mode): Update comment.
(get_related_vectype_for_scalar_type): Delete.
(get_vectype_for_scalar_type_and_size): Declare.
* tree-vect-slp.c (vect_slp_bb_region): Print dump messages to say
whether analysis passed or failed, and with what vector modes.
Use related_vector_mode to check whether trying a particular
vector mode would be redundant with the autodetected mode,
and print a dump message if we decide to skip it.
* tree-vect-loop.c (vect_analyze_loop): Likewise.
(vect_create_epilog_for_reduction): Use
get_related_vectype_for_scalar_type instead of
get_vectype_for_scalar_type_and_size.
* tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Replace
with...
(get_related_vectype_for_scalar_type): ...this new function.
Take a starting/"prevailing" vector mode rather than a vector size.
Take an optional nunits argument, with the same meaning as for
related_vector_mode. Use related_vector_mode when not
auto-detecting a mode, falling back to mode_for_vector if no
target mode exists.
(get_vectype_for_scalar_type): Update accordingly.
(get_same_sized_vectype): Likewise.
* tree-vectorizer.c (get_vec_alignment_for_array_type): Likewise.
From-SVN: r278240
As explained in the comment, vectorizable_call needs more work to
support mixtures of sizes. This avoids testsuite fallout for
later SVE patches.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-stmts.c (vectorizable_call): Require the types
to have the same size.
From-SVN: r278239
Some callers of get_same_sized_vectype were dealing with operands that
are constant or defined externally, and so have no STMT_VINFO_VECTYPE
available. Under the current model, using get_same_sized_vectype for
that case is equivalent to using get_vectype_for_scalar_type, since
get_vectype_for_scalar_type always returns vectors of the same size,
once a size is fixed.
Using get_vectype_for_scalar_type is arguably more obvious though:
if we're using the same scalar type as we would for internal
definitions, we should use the same vector type too. (Constant and
external definitions sometimes let us change the original scalar type
to a "nicer" scalar type, but that isn't what's happening here.)
This is a prerequisite to supporting multiple vector sizes in the same
vec_info.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-stmts.c (vectorizable_call): If an operand is
constant or external, use get_vectype_for_scalar_type
rather than get_same_sized_vectype to get its vector type.
(vectorizable_conversion, vectorizable_shift): Likewise.
(vectorizable_operation): Likewise.
From-SVN: r278238
This patch replaces vec_info::vector_size with vec_info::vector_mode,
but for now continues to use it as a way of specifying a single
vector size. This makes it easier for later patches to use
related_vector_mode instead.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vectorizer.h (vec_info::vector_size): Replace with...
(vec_info::vector_mode): ...this new field.
* tree-vect-loop.c (vect_update_vf_for_slp): Update accordingly.
(vect_analyze_loop, vect_transform_loop): Likewise.
* tree-vect-loop-manip.c (vect_do_peeling): Likewise.
* tree-vect-slp.c (can_duplicate_and_interleave_p): Likewise.
(vect_make_slp_decision, vect_slp_bb_region): Likewise.
* tree-vect-stmts.c (get_vectype_for_scalar_type): Likewise.
* tree-vectorizer.c (try_vectorize_loop_1): Likewise.
gcc/testsuite/
* gcc.dg/vect/vect-tail-nomask-1.c: Update expected epilogue
vectorization message.
From-SVN: r278237
This is another patch in the series to remove the assumption that
all modes involved in vectorisation have to be the same size.
Rather than have the target provide a list of vector sizes,
it makes the target provide a list of vector "approaches",
with each approach represented by a mode.
A later patch will pass this mode to targetm.vectorize.related_mode
to get the vector mode for a given element mode. Until then, the modes
simply act as an alternative way of specifying the vector size.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* target.h (vector_sizes, auto_vector_sizes): Delete.
(vector_modes, auto_vector_modes): New typedefs.
* target.def (autovectorize_vector_sizes): Replace with...
(autovectorize_vector_modes): ...this new hook.
* doc/tm.texi.in (TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES):
Replace with...
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES): ...this new hook.
* doc/tm.texi: Regenerate.
* targhooks.h (default_autovectorize_vector_sizes): Delete.
(default_autovectorize_vector_modes): New function.
* targhooks.c (default_autovectorize_vector_sizes): Delete.
(default_autovectorize_vector_modes): New function.
* omp-general.c (omp_max_vf): Use autovectorize_vector_modes instead
of autovectorize_vector_sizes. Use the number of units in the mode
to calculate the maximum VF.
* omp-low.c (omp_clause_aligned_alignment): Use
autovectorize_vector_modes instead of autovectorize_vector_sizes.
Use a loop based on related_mode to iterate through all supported
vector modes for a given scalar mode.
* optabs-query.c (can_vec_mask_load_store_p): Use
autovectorize_vector_modes instead of autovectorize_vector_sizes.
* tree-vect-loop.c (vect_analyze_loop, vect_transform_loop): Likewise.
* tree-vect-slp.c (vect_slp_bb_region): Likewise.
* config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
Replace with...
(aarch64_autovectorize_vector_modes): ...this new function.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES): Delete.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES): Define.
* config/arc/arc.c (arc_autovectorize_vector_sizes): Replace with...
(arc_autovectorize_vector_modes): ...this new function.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES): Delete.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES): Define.
* config/arm/arm.c (arm_autovectorize_vector_sizes): Replace with...
(arm_autovectorize_vector_modes): ...this new function.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES): Delete.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES): Define.
* config/i386/i386.c (ix86_autovectorize_vector_sizes): Replace with...
(ix86_autovectorize_vector_modes): ...this new function.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES): Delete.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES): Define.
* config/mips/mips.c (mips_autovectorize_vector_sizes): Replace with...
(mips_autovectorize_vector_modes): ...this new function.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES): Delete.
(TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES): Define.
From-SVN: r278236
The validation phase of vectorizable_shift used TYPE_MODE to check
whether the shift amount vector was compatible with the shifted vector:
if ((op1_vectype == NULL_TREE
|| TYPE_MODE (op1_vectype) != TYPE_MODE (vectype))
&& (!slp_node
|| SLP_TREE_DEF_TYPE
(SLP_TREE_CHILDREN (slp_node)[1]) != vect_constant_def))
But the generation phase was stricter and required the element types to
be equivalent:
&& !useless_type_conversion_p (TREE_TYPE (vectype),
TREE_TYPE (op1)))
This difference led to an ICE with a later patch.
The first condition seems a bit too lax given that the function
supports vect_worthwhile_without_simd_p, where two different vector
types could have the same integer mode. But it seems too strict
to reject signed shifts by unsigned amounts or unsigned shifts by
signed amounts; verify_gimple_assign_binary is happy with those.
This patch therefore goes for a middle ground of checking both TYPE_MODE
and TYPE_VECTOR_SUBPARTS, using the same condition in both places.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-stmts.c (vectorizable_shift): Check the number
of vector elements as well as the type mode when deciding
whether an op1_vectype is compatible. Reuse the result of
this check when generating vector statements.
From-SVN: r278235
Except for one case, get_vectype_for_scalar_type_and_size calculates
what the vector mode should be and then calls build_vector_type,
which recomputes the mode from scratch. This patch makes it use
build_vector_type_for_mode instead.
The exception mentioned above is when preferred_simd_mode returns
an integer mode, which it does if no appropriate vector mode exists.
The integer mode in question is usually word_mode, although epiphany
can return a doubleword mode in some cases.
There's no guarantee that this integer mode is appropriate, since for
example the scalar type could be a float. The traditional behaviour is
therefore to use the integer mode to determine a size only, and leave
mode_for_vector to pick the TYPE_MODE. (Note that it can actually end
up picking a vector mode if the target defines a disabled vector mode.
We therefore still need to check TYPE_MODE after building the type.)
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): If
targetm.vectorize.preferred_simd_mode returns an integer mode,
use mode_for_vector to decide what the vector type's mode
should actually be. Use build_vector_type_for_mode instead
of build_vector_type.
From-SVN: r278234
This patch passes the data vector mode to get_mask_mode, rather than its
size and nunits. This is a bit simpler and allows targets to distinguish
between modes that happen to have the same size and number of elements.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* target.def (get_mask_mode): Take a vector mode itself as argument,
instead of properties about the vector mode.
* doc/tm.texi: Regenerate.
* targhooks.h (default_get_mask_mode): Update to reflect new
get_mode_mask interface.
* targhooks.c (default_get_mask_mode): Likewise. Use
related_int_vector_mode.
* optabs-query.c (can_vec_mask_load_store_p): Update call
to get_mask_mode.
* tree-vect-stmts.c (check_load_store_masking): Likewise, checking
first that the original mode really is a vector.
* tree.c (build_truth_vector_type_for): Likewise.
* config/aarch64/aarch64.c (aarch64_get_mask_mode): Update for new
get_mode_mask interface.
(aarch64_expand_sve_vcond): Update call accordingly.
* config/gcn/gcn.c (gcn_vectorize_get_mask_mode): Update for new
get_mode_mask interface.
* config/i386/i386.c (ix86_get_mask_mode): Likewise.
From-SVN: r278233
build_same_sized_truth_vector_type was confusingly named, since for
SVE and AVX512 the returned vector isn't the same byte size (although
it does have the same number of elements). What it really returns
is the "truth" vector type for a given data vector type.
The more general truth_type_for provides the same thing when passed
a vector and IMO has a more descriptive name, so this patch replaces
all uses of build_same_sized_truth_vector_type with that. It does
the same for a call to build_truth_vector_type, leaving truth_type_for
itself as the only remaining caller.
It's then more natural to pass build_truth_vector_type the original
vector type rather than its size and nunits, especially since the
given size isn't the size of the returned vector. This in turn allows
a future patch to simplify the interface of get_mask_mode. Doing this
also fixes a bug in which truth_type_for would pass a size of zero for
BLKmode vector types.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree.h (build_truth_vector_type): Delete.
(build_same_sized_truth_vector_type): Likewise.
* tree.c (build_truth_vector_type): Rename to...
(build_truth_vector_type_for): ...this. Make static and take
a vector type as argument.
(truth_type_for): Update accordingly.
(build_same_sized_truth_vector_type): Delete.
* tree-vect-generic.c (expand_vector_divmod): Use truth_type_for
instead of build_same_sized_truth_vector_type.
* tree-vect-loop.c (vect_create_epilog_for_reduction): Likewise.
(vect_record_loop_mask, vect_get_loop_mask): Likewise.
* tree-vect-patterns.c (build_mask_conversion): Likeise.
* tree-vect-slp.c (vect_get_constant_vectors): Likewise.
* tree-vect-stmts.c (vect_get_vec_def_for_operand): Likewise.
(vect_build_gather_load_calls, vectorizable_call): Likewise.
(scan_store_can_perm_p, vectorizable_scan_store): Likewise.
(vectorizable_store, vectorizable_condition): Likewise.
(get_mask_type_for_scalar_type, get_same_sized_vectype): Likewise.
(vect_get_mask_type_for_stmt): Use truth_type_for instead of
build_truth_vector_type.
* config/aarch64/aarch64-sve-builtins.cc (gimple_folder::convert_pred):
Use truth_type_for instead of build_same_sized_truth_vector_type.
* config/rs6000/rs6000-call.c (fold_build_vec_cmp): Likewise.
gcc/c/
* c-typeck.c (build_conditional_expr): Use truth_type_for instead
of build_same_sized_truth_vector_type.
(build_vec_cmp): Likewise.
gcc/cp/
* call.c (build_conditional_expr_1): Use truth_type_for instead
of build_same_sized_truth_vector_type.
* typeck.c (build_vec_cmp): Likewise.
gcc/d/
* d-codegen.cc (build_boolop): Use truth_type_for instead of
build_same_sized_truth_vector_type.
From-SVN: r278232
Callers of vect_halve_mask_nunits and vect_double_mask_nunits
already know what mode the resulting vector type should have,
so we might as well create the vector type directly with that mode,
just like build_vector_type_for_mode lets us build normal vectors
with a known mode. This avoids the current awkwardness of having
to recompute the mode starting from vec_info::vector_size, which
hard-codes the assumption that all vectors have to be the same size.
A later patch gets rid of build_truth_vector_type and
build_same_sized_truth_vector_type, so the net effect of the
series is to reduce the number of type functions by one.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree.h (build_truth_vector_type_for_mode): Declare.
* tree.c (build_truth_vector_type_for_mode): New function,
split out from...
(build_truth_vector_type): ...here.
(build_opaque_vector_type): Fix head comment.
* tree-vectorizer.h (supportable_narrowing_operation): Remove
vec_info parameter.
(vect_halve_mask_nunits): Replace vec_info parameter with the
mode of the new vector.
(vect_double_mask_nunits): Likewise.
* tree-vect-loop.c (vect_halve_mask_nunits): Likewise.
(vect_double_mask_nunits): Likewise.
* tree-vect-loop-manip.c: Include insn-config.h, rtl.h and recog.h.
(vect_maybe_permute_loop_masks): Remove vinfo parameter. Update call
to vect_halve_mask_nunits, getting the required mode from the unpack
patterns.
(vect_set_loop_condition_masked): Update call accordingly.
* tree-vect-stmts.c (supportable_narrowing_operation): Remove vec_info
parameter and update call to vect_double_mask_nunits.
(vectorizable_conversion): Update call accordingly.
(simple_integer_narrowing): Likewise. Remove vec_info parameter.
(vectorizable_call): Update call accordingly.
(supportable_widening_operation): Update call to
vect_halve_mask_nunits.
* config/aarch64/aarch64-sve-builtins.cc (register_builtin_types):
Use build_truth_vector_type_mode instead of build_truth_vector_type.
From-SVN: r278231
mode_for_int_vector, like mode_for_vector, can sometimes return
an integer mode or an unsupported vector mode. But no callers
are interested in that case, and only want supported vector modes.
This patch therefore replaces mode_for_int_vector with
related_int_vector_mode, which gives the target a chance to pick
its preferred vector mode for the given element mode and size.
2019-11-14 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* machmode.h (mode_for_int_vector): Delete.
(related_int_vector_mode): Declare.
* stor-layout.c (mode_for_int_vector): Delete.
(related_int_vector_mode): New function.
* optabs.c (expand_vec_perm_1): Use related_int_vector_mode
instead of mode_for_int_vector.
(expand_vec_perm_const): Likewise.
* config/aarch64/aarch64.c (aarch64_emit_approx_sqrt): Likewise.
(aarch64_evpc_sve_tbl): Likewise.
* config/s390/s390.c (s390_expand_vec_compare_cc): Likewise.
(s390_expand_vcond): Likewise.
From-SVN: r278230
