Philip Herron [Mon, 13 Feb 2023 17:51:19 +0000 (17:51 +0000)]
gccrs: Support for Sized builtin marker trait
When implementing general bounds checking as part of unify calls, we did
not check associated types on bounds which lead to alot of missed error
checking. This now recursively checks the bounds and the associated types
with a decent error message. This also required us to implement the Sized
marker trait to keep existing test-cases happy.
Fixes #1725
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
gcc/rust/ChangeLog:
* typecheck/rust-hir-trait-reference.cc (TraitReference::clear_associated_types): make const
(TraitReference::clear_associated_type_projections): new interface
* typecheck/rust-hir-trait-reference.h:
* typecheck/rust-hir-trait-resolve.cc (TraitResolver::resolve_trait): refactor
(TraitItemReference::associated_type_reset): reset projections
* typecheck/rust-hir-type-bounds.h:
* typecheck/rust-hir-type-check-expr.cc (TypeCheckExpr::visit): fix bounds
* typecheck/rust-tyty-bounds.cc (TypeBoundsProbe::TypeBoundsProbe): refactor into cc file
(TypeBoundsProbe::Probe): refactor
(TypeBoundsProbe::is_bound_satisfied_for_type): likewise
(TypeBoundsProbe::assemble_sized_builtin): add builtin for Sized
(TypeCheckBase::get_predicate_from_bound): refactor
(TypeBoundPredicate::lookup_associated_type): refactor
* typecheck/rust-tyty-subst.cc (SubstitutionRef::lookup_associated_impl)
(SubstitutionRef::prepare_higher_ranked_bounds): new interface to clear hanging bounds
(SubstitutionRef::monomorphize): refactor
* typecheck/rust-tyty-subst.h:
* typecheck/rust-tyty.cc (BaseType::get_locus): helper
(BaseType::satisfies_bound): ensure bounds are satisfied and assoicated types
(ParamType::ParamType): new field in constructor
(ParamType::clone): update clone
(ParamType::set_implicit_self_trait): new interface
(ParamType::is_implicit_self_trait): likewise
* typecheck/rust-tyty.h: cleanup
* util/rust-hir-map.cc (Mappings::Mappings): builtin marker
(Mappings::~Mappings): delete marker
(Mappings::lookup_builtin_marker): lookup
* util/rust-hir-map.h: update header
gcc/testsuite/ChangeLog:
* rust/compile/issue-1725-1.rs: New test.
* rust/compile/issue-1725-2.rs: New test.
RAIIFile constructor was accepting directory filename. This lead to
unattended directory opening in some part of the code (load_file_bytes)
wich resulted in ice. Since RAIIFile are used for the lexer, removing
the ability to open directories with RAIIFile fixes those issues and
prevent future mistakes.
Owen Avery [Fri, 3 Feb 2023 15:19:32 +0000 (10:19 -0500)]
gccrs: Simplify WildcardPattern let statement handling
gcc/rust/ChangeLog:
* backend/rust-compile-pattern.cc
(CompilePatternLet::visit): Simplify WildcardPattern compilation for let statements.
* backend/rust-compile-var-decl.h:
(CompileVarDecl::visit): Remove variable declaration for WildcardPattern.
* resolve/rust-ast-resolve-pattern.h:
(PatternDeclaration::visit): Remove name resolution for WildcardPattern.
Arthur Cohen [Wed, 18 Jan 2023 11:23:03 +0000 (12:23 +0100)]
gccrs: macros: Perform macro expansion in a fixed-point fashion.
This commit changes our macro expansion system from an eager and recursive
macro expansion to a fixed-point like system. Instead of, when seeing
a macro invocation, expanding it and all of the macros within it, we
now perform multiple passes of expansion on the entire crate.
This, however, leads to a problem. Rust macros are expanded lazily, but
Rust builtin macros should be expanded eagerly. Due to this, we must
work around the lazy expansion in builtin macros and perform eager
expansion for each pass of the fixed-point, before finally expanding
the builtin when there are no longer any inner macro invocations.
To perform proper macro scoping, the ENR now keeps track of the current
scope (`current_scope` member) and resolves macros accordingly.
This is done through the use of the `scoped` method, which creates a new
scope, runs a specified lambda and then exits the scope. This prevents
pushing/popping errors that we've seen happen already in similar
contexts.
We might think about generalizing it to other classes, providing a
`Scoped<EntryFn, ExitFn>` class or similar
gcc/rust/ChangeLog:
* ast/rust-macro.cc: New file.
* Make-lang.in: Add `rust-macro.o` object
* ast/rust-ast-fragment.cc (Fragment::Fragment): Change API around
the construction of AST fragments.
(Fragment::operator=): Correct `Fragment::operator=` to take into
account the fragment tokens.
(Fragment::create_error): Use new constructor.
(Fragment::complete): Remove in favor of new constructor.
(Fragment::unexpanded): Remove as that Fragment type is no longer used
or possible.
(Fragment::get_tokens): Add helper to access a fragment's tokens.
* ast/rust-ast-fragment.h (enum class): Remove `FragmentKind::Unused`
* ast/rust-ast.cc (MacroInvocation::as_string): Display
builtin macro invocations properly.
* ast/rust-ast.h: Fix `DelimTokenTree` class copy constructors and
handling of its token vector.
* ast/rust-macro.h (class MacroMatcher): Format.
(class MetaItemSeq): Likewise.
(builtin_macro_from_string): Get a `BuiltinMacroKind` from a given
string, i.e the name of the macro (`assert!`, `cfg!` and so on).
* expand/rust-attribute-visitor.cc (AttrVisitor::visit): Do not expand
macros recursively anymore.
(AttrVisitor::maybe_expand_expr): Likewise.
(AttrVisitor::maybe_expand_type): Likewise.
* expand/rust-attribute-visitor.h: Likewise, and remove
`expand_macro_fragment_recursively` function.
* expand/rust-macro-builtins.cc (make_token): Add shorthand for
returning `std::unique_ptr<AST::Token>`s.
(make_macro_invocation): Add shorthand for returning fragments
containing builtin macro invocations.
(try_expand_macro_expression): Do not expand macros recursively.
(try_expand_single_string_literal): Likewise.
(try_expand_many_expr): Likewise.
(parse_single_string_literal): Error out more appropriately.
(MacroBuiltin::compile_error_handler): Add explanation for eager
invocation
(MacroBuiltin::file_handler): Return the proper tokens associated with
macro invocation, and builtin macros in the case of necessary eager
expansion.
(MacroBuiltin::column_handler): Likewise.
(MacroBuiltin::include_bytes_handler): Likewise.
(MacroBuiltin::include_str_handler): Likewise.
(MacroBuiltin::concat_handler): Likewise.
(MacroBuiltin::env_handler): Likewise.
(MacroBuiltin::cfg_handler): Likewise.
(MacroBuiltin::include_handler): Likewise.
(MacroBuiltin::line_handler): Likewise.
* expand/rust-macro-expand.cc (MacroExpander::expand_eager_invocations):
Add function to expand eager invocations *once* in the fixed point
pipeline.
(MacroExpander::expand_invoc): Call into `expand_eager_invocations` for
builtin macro invocations.
(MacroExpander::expand_crate): Use new `AttrVisitor` API.
(parse_many): Return tokens in `AST::Fragment`.
(transcribe_expression): Likewise.
(transcribe_type): Likewise.
* expand/rust-macro-expand.h (struct MacroExpander): Add `has_changed`
flag for fixed point checking.
* resolve/rust-early-name-resolver.cc (EarlyNameResolver::EarlyNameResolver):
Keep track of the current macro scope.
(EarlyNameResolver::go): Use `scoped` API.
(EarlyNameResolver::visit): Likewise.
* resolve/rust-early-name-resolver.h: Add `scoped` API.
* rust-session-manager.cc (Session::expansion): Perform macro expansion
in a fixed-point fashion.
gcc/testsuite/ChangeLog:
* rust/compile/macro17.rs: Fix testsuite for new recursion errors.
* rust/compile/macro44.rs: Fix invalid testcase assertions.
* rust/compile/builtin_macro_recurse.rs: Fix invalid test.
* rust/compile/builtin_macro_recurse2.rs: New test.
* rust/compile/macro46.rs: New test.
Philip Herron [Tue, 31 Jan 2023 18:52:33 +0000 (18:52 +0000)]
gccrs: Fix higher ranked trait bounds computation of self
This updates the higher ranked trait bounds computation to handle ambigious
cases. When we have a slice for example:
let slice = &a[1..3];
This works by reusing the Index operator overload from libcore, so when the
index range of 1..3 is computed, the type system needs to compute what the
types of index are; this works by integer inference variables
Range<<integer>> that need to be unified with the impl Index for
Range<Usize> which computes the real type of usize for the index. This is
fine but what happens when we have the Copy and Clone traits bounds which
have implementations for all the primitive types i8, i16, i32, i64...
which is valid for any integer inference variable so the code prior to this
patch would have grabbed the first impl it would have found and used it
which is incorrect. When we have integer or float inference variables we
need to look for their respective defaults or emit an ambigious type bound
error.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
gcc/rust/ChangeLog:
Philip Herron [Tue, 31 Jan 2023 14:27:49 +0000 (14:27 +0000)]
gccrs: Fix nullptr dereference
When we check if this is concrete the guard checks to ensure the argument
is non null but the check here is wrongly returning early when the check
is non null meaning when it is null and therefore not concrete it will
end up doing a null dereference.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
gcc/rust/ChangeLog:
Philip Herron [Mon, 30 Jan 2023 18:19:07 +0000 (18:19 +0000)]
gccrs: Refactor the type unification code
This refactors the unification systems to be a consistent interface using
switch statements and simple functions instead of the old clunky visitor
system. This is more maintainable as it is harder to miss cases when we
can take advantages of switch statements.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
gcc/rust/ChangeLog:
Philip Herron [Fri, 27 Jan 2023 15:38:58 +0000 (15:38 +0000)]
gccrs: Remove monomorphization hack to setup possible associated types
During CallExpr argument type checking we may be calling a default
implementation of a trait function this will require any possible
associated types to be resolved and setup. This monomoprhization call does
this but it will premtivly do extra unification of types which will throw
off type checking later on. This fix is required for my work into type
bounds checking.
Fixes #1773
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
gcc/rust/ChangeLog:
Philip Herron [Fri, 27 Jan 2023 18:31:11 +0000 (18:31 +0000)]
gccrs: Clear the substitution callbacks when copying ArgumentMappings
When we set the callback on substitutions this is not safe to be copied
around since we store the used argument mappings and can reuse them in
different contexts. This clears the callback on copy's to make it safer.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
gcc/rust/ChangeLog:
Philip Herron [Fri, 27 Jan 2023 18:28:06 +0000 (18:28 +0000)]
gccrs: Refactor handle_substitutions to take a reference
This patch changes the recusive substitution code to take a reference
instead of a copy. This is important as the callback field is going to be
made non-copyable in a future patch and this pipeline is for recursive
substitutions so its ok to reuse the same mappings here.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
gcc/rust/ChangeLog:
The parser now recursively tries to parse a reference type after seeing
a `&` or `&&` token.
gcc/rust/ChangeLog:
* parse/rust-parse-impl.h (Parser::parse_type): Handle double ampersan
properly
(Parser::parse_reference_type): Call into `parse_reference_type_inner`
and wrap double reference types in another `AST::ReferenceType` node
(Parser::parse_reference_type_inner): Add parsing implementation
which does not care about the leading token (& or &&)
(Parser::parse_type_no_bounds): Handle double ampersand properly
* parse/rust-parse.h: Declare `parse_reference_type_inner`
Owen Avery [Wed, 11 Jan 2023 17:05:39 +0000 (12:05 -0500)]
gccrs: Change how CompileVarDecl outputs Bvariable's
This allows patterns to declare multiple/no variables
gcc/rust/ChangeLog:
* backend/rust-compile-base.cc (HIRCompileBase::compile_locals_for_block):
Allow patterns to declare zero or multiple variables.
* backend/rust-compile-var-decl.h: Change function declaration.
mxlol233 [Wed, 11 Jan 2023 12:36:13 +0000 (20:36 +0800)]
gccrs: diagnostics: Add underline for tokens in diagnostics.
Currently, the diagnostics only point to the corresponding token's
start position by carets, and lack of underlines for full token. This
commit add support for such underlines in diagnostics by encoding range
information in location_t.
mxlol233 [Wed, 11 Jan 2023 15:24:07 +0000 (23:24 +0800)]
gccrs: Add get_locus function for abstract class MetaItemInner.
This commit adds virtual function get_locus to base class MetaItemInner,
which is helpful when we need to print diagnostics on some sub-classes of
MetaItemInner.
Philip Herron [Sat, 7 Jan 2023 14:41:12 +0000 (14:41 +0000)]
gccrs: Support associated type bound arguments
This patch adds support for the GenercArgsBinding type, where you can
specify the associated types of a trait bound using `<Foo=i32>` style
syntax. Note that the type-resolution relys on the i32 impl for Add
as type resolution will resolve the `a+a` to the core::ops::Add method
so code generation will require this to exist.
I have ameded testsuite/rust/compile/bounds.rs as this code is wrongly
creating an HIR::GenericArgs with a trait-object type and causing issues.
the parsing is still correct but we dont have the mechanism to represent
this in AST and HIR properly. I think we will need a new HIR::GenericArgs
AssociatedTypeBindingBound or something similar. We are still lacking
bounds checking during are type coercions and unifications so running this
example using an f32 will wrongly pass type checking, this will need
addressed next.
Fixes #1720
gcc/rust/ChangeLog:
* hir/tree/rust-hir-path.h: Add const get_identifier and get_type method.
* typecheck/rust-hir-path-probe.h: Use new SubstitutionArgumentMappings constructor.
* typecheck/rust-hir-trait-resolve.cc: Likewise.
* typecheck/rust-hir-type-check-expr.cc (TypeCheckExpr::visit): Likewise.
* typecheck/rust-tyty-bounds.cc (TypeCheckBase::get_predicate_from_bound):
Do not assert failure on size mismatch anymore.
(TypeBoundPredicate::TypeBoundPredicate): Use new SubstitutionArgumentMappings constructor.
(TypeBoundPredicate::operator=): Likewise.
(TypeBoundPredicate::apply_generic_arguments): Likewise.
(TypeBoundPredicateItem::get_tyty_for_receiver): Likewise.
(TypeBoundPredicate::get_num_associated_bindings): Likewise.
(TypeBoundPredicate::lookup_associated_type): Fix implementation for new system.
(TypeBoundPredicate::get_associated_type_items): Likewise.
* typecheck/rust-tyty.cc (SubstitutionRef::get_mappings_from_generic_args): Add new
behavior.
(SubstitutionRef::infer_substitions): Use new constructor and add comment.
(SubstitutionRef::solve_missing_mappings_from_this): Use new constructor.
* typecheck/rust-tyty.h: Define new constructors.
gcc/testsuite/ChangeLog:
* rust/compile/bounds.rs: change to use -fsyntax-only
* rust/execute/torture/issue-1720.rs: New test.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
Philip Herron [Sat, 7 Jan 2023 17:15:23 +0000 (17:15 +0000)]
gccrs: Add name resolution to generic argument associated item bindings
When specifying generic arguments to Traits we can also specify the
associated types using `<BindingName=i32>` syntax we need to add
name resolution to the type argument here and rely on the type
resolution pass to ensure the associated type exists and to setup the
associated types accordingly.
Addresses #1720
gcc/rust/ChangeLog:
* resolve/rust-ast-resolve-type.cc (ResolveGenericArgs::go): Add name resolution to
Trait items.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
* rust/compile/decl_macro1.rs: New test.
* rust/compile/decl_macro2.rs: New test.
* rust/compile/decl_macro3.rs: New test.
* rust/compile/decl_macro4.rs: New test.
* rust/compile/decl_macro5.rs: New test.
* rust/compile/decl_macro6.rs: New test.
* rust/compile/decl_macro7.rs: New test.
* rust/execute/torture/decl_macro1.rs: New test.
* rust/execute/torture/decl_macro2.rs: New test.
* rust/execute/torture/decl_macro3.rs: New test.
* rust/execute/torture/decl_macro4.rs: New test.
Prajwal S N [Sat, 31 Dec 2022 07:19:02 +0000 (12:49 +0530)]
gccrs: unsafe: check use of `target_feature` attribute
The `target_feature` attribute is for conditional compilation and may or
may not compile on all platforms. Using it requires an unsafe function
or block.
gcc/rust/ChangeLog:
* checks/errors/rust-unsafe-checker.cc (check_target_attr): New function.
(UnsafeChecker::check_function_attr): Call into `check_target_attr`.
(UnsafeChecker::visit): Check for target_feature attributes.
* checks/errors/rust-unsafe-checker.h: Add declarations.
* util/rust-attributes.cc: Add attribute.
gcc/testsuite/ChangeLog:
* rust/compile/unsafe11.rs: New test.
Signed-off-by: Prajwal S N <prajwalnadig21@gmail.com>
[testsuite] enable -maltivec like vect_int for signbit-2.c
Explicitly enable altivec if it's supported. vect_int tests for
powerpc_altivec_ok, but without the explicit option, if altivec is not
enabled by default, we end up expecting vectorization that doesn't
occur.
testsuite: fix proc unsupported overriding in modules.exp [PR108899]
The overrider of proc unsupported in modules.exp had two problems
reported by Thomas Schwinge, even after Jakub Jelínek's fix:
- it remained in effect while running other dejagnu testsets
- it didn't quote correctly the argument list passed to it, which
caused test names to be surrounded by curly braces, as in:
UNSUPPORTED: {...}
This patch fixes both issues, obsoleting and reverting Jakub's change,
by dropping the overrider and renaming the saved proc back, and by
using uplevel's argument list splicing.
Co-authored-by: Thomas Schwinge <thomas@codesourcery.com>
for gcc/testsuite/ChangeLog
PR testsuite/108899
* g++.dg/modules/modules.exp (unsupported): Drop renaming.
Fix quoting.
Gaius Mulley [Wed, 5 Apr 2023 22:07:46 +0000 (23:07 +0100)]
PR modula2/109423 cc1gm2 ICE if an INCL or EXCL is performed on an unknown set
This patch fixes an ICE if attempting to INCL or EXCL on an unknown
set. The fix was to correct an error format string. Also included in
the patch are patches to remove unused variables. The patch also
marks a variable as written in BuildAdr.
gcc/m2/ChangeLog:
PR modula2/109423
* gm2-compiler/M2Base.def (Unbounded): Remove.
* gm2-compiler/M2Error.def (ErrorAbort0): Add noreturn
attribute.
* gm2-compiler/M2Quads.mod (BuildInclProcedure): Correct
error format string.
(BuildExceptProcedure): Correct error format string.
(BuildAdrFunction): Call PutWriteQuad when taking the
address of a variable.
* gm2-libs-ch/SysExceptions.c (_M2_SysExceptions_init): Add
parameters.
* gm2-libs-ch/wrapc.c (_M2_wrapc_init): Add parameters.
* gm2-libs/DynamicStrings.mod (DumpStringInfo): Remove t.
(PopAllocationExemption): Remove f.
* gm2-libs/FIO.mod (BufferedWrite): Remove result.
* gm2-libs/FormatStrings.mod (Copy): Remove endpos and
afterperc.
(HandlePercent): Remove result.
* gm2-libs/Indexing.mod (RemoveIndiceFromIndex): Remove k.
* gm2-libs/M2Dependent.mod (CreateModule): Remove p0
and p1.
(DumpModuleData): Remove mptr.
(ConstructModules): Remove nulp.
* gm2-libs/RTExceptions.mod (PopHandler): Remove i.
* gm2-libs/RTint.mod (Listen): Remove b4s, b4m, afs
and afm.
* gm2-libs/SFIO.mod (ReadS): Remove c.
* gm2-libs/StringConvert.mod (doDecimalPlaces): Remove
whole and fraction.
gcc/testsuite/ChangeLog:
* gm2/pim/fail/setunknown.mod: New test.
PR modula2/109423
* gm2/pim/fail/setunknown2.mod: New test.
libstdc++: Downgrade DEBUG to ASSERTIONS when !HOSTED
Supporting the debug mode in freestanding is a non-trivial job, so
instead, as a best-effort, enable assertions, which are light and easy.
libstdc++-v3/ChangeLog:
* include/bits/c++config: When __STDC_HOSTED__ is zero,
disable _GLIBCXX_DEBUG and, if it was set, enable
_GLIBCXX_ASSERTIONS.
* testsuite/lib/libstdc++.exp (check_v3_target_debug_mode):
Include <bits/c++config.h> when determining whether debug is
set, in order to inherit the logic from above
Arsen Arsenović [Wed, 8 Mar 2023 16:01:24 +0000 (17:01 +0100)]
libstdc++: Add a test for <version> FTM redefinitions
This test detects redefinitions by compiling stdc++.h and <version>, by
disabling the system_header pragma on the latter, to allow warnings in
it. Thanks Patrick Palka for the suggestion.
libstdc++-v3/ChangeLog:
* testsuite/17_intro/versionconflict.cc: New test.
* include/std/version: Allow disabling the system_header pragma
via _GLIBCXX_TESTING_SYSHDR.
Arsen Arsenović [Wed, 8 Mar 2023 11:04:11 +0000 (12:04 +0100)]
libstdc++: Harmonize <version> and other headers
Due to recent, large changes in libstdc++, the feature test macros
declared in <version> got out of sync with the other headers that
possibly declare them.
libstdc++-v3/ChangeLog:
* include/bits/unique_ptr.h (__cpp_lib_constexpr_memory):
Synchronize the definition block with...
* include/bits/ptr_traits.h (__cpp_lib_constexpr_memory):
... this one here. Also define the 202202L value, rather than
leaving it up to purely unique_ptr.h, so that the value is
synchronized across all headers.
(__gnu_debug::_Safe_iterator_base): Move into new conditional
block.
* include/std/memory (__cpp_lib_atomic_value_initialization):
Define on freestanding under the same conditions as in
atomic_base.h.
* include/std/version (__cpp_lib_robust_nonmodifying_seq_ops):
Also define on freestanding.
(__cpp_lib_to_chars): Ditto.
(__cpp_lib_gcd): Ditto.
(__cpp_lib_gcd_lcm): Ditto.
(__cpp_lib_raw_memory_algorithms): Ditto.
(__cpp_lib_array_constexpr): Ditto.
(__cpp_lib_nonmember_container_access): Ditto.
(__cpp_lib_clamp): Ditto.
(__cpp_lib_constexpr_char_traits): Ditto.
(__cpp_lib_constexpr_string): Ditto.
(__cpp_lib_sample): Ditto.
(__cpp_lib_lcm): Ditto.
(__cpp_lib_constexpr_iterator): Ditto.
(__cpp_lib_constexpr_char_traits): Ditto.
(__cpp_lib_interpolate): Ditto.
(__cpp_lib_constexpr_utility): Ditto.
(__cpp_lib_shift): Ditto.
(__cpp_lib_ranges): Ditto.
(__cpp_lib_move_iterator_concept): Ditto.
(__cpp_lib_constexpr_numeric): Ditto.
(__cpp_lib_constexpr_functional): Ditto.
(__cpp_lib_constexpr_algorithms): Ditto.
(__cpp_lib_constexpr_tuple): Ditto.
(__cpp_lib_constexpr_memory): Ditto.
Jeff Law [Wed, 5 Apr 2023 15:16:29 +0000 (09:16 -0600)]
[RFA][Bug target/108892 ][13 regression] Force re-recognition after changing RTL structure of an insn
So as mentioned in the PR the underlying issue here is combine changes the form of an existing insn, but fails to force re-recognition. As a result other parts of the compiler blow up.
> /* Temporarily replace the set's source with the
> contents of the REG_EQUAL note. The insn will
> be deleted or recognized by try_combine. */
> rtx orig_src = SET_SRC (set); rtx orig_dest = SET_DEST (set); if (GET_CODE (SET_DEST (set)) == ZERO_EXTRACT)
> SET_DEST (set) = XEXP (SET_DEST (set), 0);
> SET_SRC (set) = note;
> i2mod = temp;
> i2mod_old_rhs = copy_rtx (orig_src);
> i2mod_new_rhs = copy_rtx (note);
> next = try_combine (insn, i2mod, NULL, NULL,
> &new_direct_jump_p, last_combined_insn);
> i2mod = NULL;
> if (next)
> {
> statistics_counter_event (cfun, "insn-with-note combine", 1);
> goto retry;
> } SET_SRC (set) = orig_src;
> SET_DEST (set) = orig_dest;
This code replaces the SET_SRC of an insn in the RTL stream with the contents of a REG_EQUAL note. So given an insn like this:
It replaces the (ior ...) with a (const_int ...). The resulting insn is passed to try_combine which will try to recognize it, then use it in a combination attempt. Recognition succeeds with the special define_insn_and_split pattern in the risc-v backend resulting in:
This is as-expected. Now assume we were unable to combine anything, so try_combine returns NULL_RTX. The quoted code above restores SET_SRC (and SET_DEST) resulting in:
Martin Jambor [Wed, 5 Apr 2023 14:36:49 +0000 (16:36 +0200)]
ipa: Avoid another ICE when dealing with type-incompatibilities (PR 108959)
PR 108959 shows one more example where undefined code with type
incompatible accesses to stuff passed in parameters can cause an ICE
because we try to create a VIEW_CONVERT_EXPR of mismatching sizes:
1. IPA-CP tries to push one type from above,
2. IPA-SRA (correctly) decides the parameter is useless because it is
only used to construct an argument to another function which does not
use it and so the formal parameter should be removed,
3. but the code reconciling IPA-CP and IPA-SRA transformations still
wants to perform the IPA-CP and overrides the built-in DCE of
useless statements and tries to stuff constants into them
instead, constants of a type with mismatching type and size.
This patch avoids the situation in IPA-SRA by purging the IPA-CP
results from any "aggregate" constants that are passed in parameters
which are detected to be useless. It also removes IPA value range and
bits information associated with removed parameters stored in the same
structure so that the useless information is not streamed later on.
gcc/ChangeLog:
2023-03-22 Martin Jambor <mjambor@suse.cz>
PR ipa/108959
* ipa-sra.cc (zap_useless_ipcp_results): New function.
(process_isra_node_results): Call it.
Jakub Jelinek [Tue, 4 Apr 2023 14:13:06 +0000 (16:13 +0200)]
range-op-float: Fix reverse ops of comparisons [PR109386]
I've missed one of my recent range-op-float.cc changes (likely the
r13-6967 one) caused
FAIL: libphobos.phobos/std/math/algebraic.d execution test
FAIL: libphobos.phobos_shared/std/math/algebraic.d execution test
regressions, distilled into a C testcase below.
In the testcase, we have
!(u >= v)
condition where both u and v are results of fabs*, which guards
t1 = u u<= __FLT_MAX__;
and
t2 = v u<= __FLT_MAX__;
comparisons. From false u >= v where u and v have [0.0, +Inf] NAN
ranges we (incorrectly deduce that one of them is [nextafterf (0.0, 1.0), +Inf] NAN
and the other is [0.0, nextafterf (+Inf, 0.0)] NAN and from that deduce that
one of the comparisons is always true, because UNLE_EXPR with the maximum
representable number are false only if the value is +Inf and our ranges tell
that is not possible.
The bug is that the u >= v comparison determines a sensible range only when
it is true - we then know neither operand can be NAN and it behaves
correctly. But when the comparison is false, our current code gives
sensible answers only if the other op can't be NAN. If it can be NAN,
whenever it is NAN, the comparison is always false regardless of the other
value, so the other value needs to be VARYING.
Now, foperator_unordered_lt::op1_range etc. had code to deal with that
for op?.known_nan (), but as the testcase shows, it is enough if it may be a
NAN at runtime to make it VARYING.
So, the following patch replaces for all those BRS_FALSE cases of the normal
non-equality comparisons if (opOTHER.known_isnan ()) r.set_varying (type);
to do it also if maybe_isnan ().
For the unordered or ... comparisons, it is similar for BRS_TRUE. Those
comparisons are true whenever either operand is NAN, or if neither is NAN,
the corresponding normal comparison. So, if those comparisons are true
and other operand might be a NAN, we can't tell (VARYING), if it is false,
currently handling is correct.
2023-04-04 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/109386
* range-op-float.cc (foperator_lt::op1_range, foperator_lt::op2_range,
foperator_le::op1_range, foperator_le::op2_range,
foperator_gt::op1_range, foperator_gt::op2_range,
foperator_ge::op1_range, foperator_ge::op2_range): Make r varying for
BRS_FALSE case even if the other op is maybe_isnan, not just
known_isnan.
(foperator_unordered_lt::op1_range, foperator_unordered_lt::op2_range,
foperator_unordered_le::op1_range, foperator_unordered_le::op2_range,
foperator_unordered_gt::op1_range, foperator_unordered_gt::op2_range,
foperator_unordered_ge::op1_range, foperator_unordered_ge::op2_range):
Make r varying for BRS_TRUE case even if the other op is maybe_isnan,
not just known_isnan.
* gcc.c-torture/execute/ieee/pr109386.c: New test.
wrongly eliding the overflow. This kind of problems is usually
avoided by using TYPE_OVERFLOW_SANITIZED in the appropriate place.
The first match.pd hunk in the patch fixes it. I've constructed
a testcase for each of the surrounding cases as well. Then I
noticed that fold_binary_loc/associate has the same problem, so I've
added a TYPE_OVERFLOW_SANITIZED there as well (it may be too coarse,
sorry). Then I found yet another problem, but instead of fixing it
now I've opened 109134. I could probably go on and find a dozen more.
PR sanitizer/109107
gcc/ChangeLog:
* fold-const.cc (fold_binary_loc): Use TYPE_OVERFLOW_SANITIZED
when associating.
* match.pd: Use TYPE_OVERFLOW_SANITIZED.
gcc/testsuite/ChangeLog:
* c-c++-common/ubsan/pr109107-1.c: New test.
* c-c++-common/ubsan/pr109107-2.c: New test.
* c-c++-common/ubsan/pr109107-3.c: New test.
* c-c++-common/ubsan/pr109107-4.c: New test.
From the initial merge of the MVE backend, the vcreate
intrinsic has had the vector lanes mixed up, compared
to the intended (as per the ACLE) definition. This is
also a discrepancy with clang:
https://godbolt.org/z/4n93e5aqj
This patches simply switches the operands around and
makes the tests more specific on the input registers
(I do not touch the output Q regs as they vary based
on softfp/hardfp or the input registers when the input
is a constant, since, in that case, a single register
is loaded with a constant and then the same register is
used twice as "vmov q0[2], q0[0], r2, r2" and the reg
num might not always be guaranteed).
projects / thirdparty / gcc.git / log
