This re-ordering allows building f18 with shared library using and LLVM build
with static libraries.
This reordering (that also made sens form an alphabetical point of view)
works here to do such "shared+archive" compiling because the current
dependency on LLVM is simple (only one f18 lib + an executable depends on LLVM).
As soon as two f18 libraries will depend on LLVM, one will have to use an LLVM
version built with -DBUILD_SHARED_LIBS=On if one wants to use this option
to compile f18.
Original-commit: flang-compiler/f18@bbbbe9c1aa
Reviewed-on: https://github.com/flang-compiler/f18/pull/994
The Pre-FIR Tree structure is a transient data structure that
is meant to be built from the parse tree just before lowering to
FIR and that will be deleted just afterwards. It is not meant to perfrom
optimization analysis and transformations. It only provides temporary
information, such as label target information or parse tree parent nodes,
that is meant to be used to lower the parse tree structure into
FIR operations.
A PFTBuilder class builds the Pre-Fir Tree from the parse-tree.
A pretty printer is available to visualize this data structure.
- Lit tests are added to:
1. that the PFT tree structure is as expected
2. that the PFT captures all intented nodes
- Cmake changes: Prevent warnings inisde LLVM headers when compiling flang
The issue is that some LLVM headers define functions where the usage of
the parameters depend on environment ifdef. See for instance Size in:
5f940220bf/llvm/include/llvm/Support/Compiler.h (L574)
Because flang is build with -Werror and -Wunused-parameter is default in
clang, this may breaks build in some environments (like with clang9 on macos).
A solution would be to add -Wno-unused-parameter to flang CmakLists.txt,
but it is wished to keep this warning on flang sources for quality purposes.
Fixing LLVM headers is not an easy task and `[[maybe_unused]]` is C++17 and
cannot be used yet in LLVM headers.
Hence, this fix simply silence warnings coming from LLVM headers by telling
CMake they are to be considered as if they were system headers.
- drone.io changes: remove llvm 6.0 from clang config in drone.io and link
flang with libstdc++ instead of libc++
llvm-dev resolved to llvm-6.0 in clang builds on drone.io. llvm 6.0 too old.
LLVM packages are linked with libstdc++ standard library whereas libc++ was
used for flang. This caused link time failure when building clang. Change
frone.io to build flang with libc++.
Note:
This commit does not reflect an actual work log, it is a feature based split of the
changes done in the FIR experimental branch. The related work log can be found in the
commits between:
864898cbe509d032abfe1172ec367dbd3dd92bc1
and
137c23da9c64cf90584cf81fd646053a69e91f63
Other changes come from https://github.com/flang-compiler/f18/pull/959 review.
Original-commit: flang-compiler/f18@edb0943bca
Reviewed-on: https://github.com/flang-compiler/f18/pull/959
Updated CMake files accordingly, using better regex
Updated license headers to match new extension and fit within 80 columns
Updated other comments within files that referred to the old extension
Original-commit: flang-compiler/f18@ae7721e611
Reviewed-on: https://github.com/flang-compiler/f18/pull/958
Have it wrap an optional Assignment so that we can distinguish between
unanalyzed and analyzed with error.
Change analysis of PointerAssignmentStmt to proceed with bounds even
if the DataRef or Expr has an error. Otherwise any bounds expressions
won't be analyzed in that case.
In GetExpr() and GetAssignment() if we get an internal error due to an
unanalyzed expression, dump the parse tree for the expression so we have
some context for the error. They should only be called after the
expression analysis phase. At that point, every expression and assignment
should be analyzed, though some may have resulted in errors(indicated by
returning `nullptr`).
Original-commit: flang-compiler/f18@3b865d7703
Reviewed-on: https://github.com/flang-compiler/f18/pull/944
Tree-same-pre-rewrite: false
back out -Mnolargearray default temporarily
Fix C_F_POINTER(SHAPE=) argument check, it can be any kind of integer
Revert default result kind of SIZE() & al. to standard by default
Remove needless usage of -fdebug-semantics
Original-commit: flang-compiler/f18@57058a5b16
Reviewed-on: https://github.com/flang-compiler/f18/pull/907
Add `typedAssignment` to `PointerAssignmentStmt` parse tree node and
extend `evaluate::Assignment` to include pointer assignment, including
analyzed bounds. Analyze pointer assignments and fill those in.
Emit them in unparsed output and parse tree dump when present.
Change assignment checking to use analyzed expressions and assignments
rather than calling AnalyzeExpr. Check bounds in pointer assignments
for impure function calls in FORALL context.
Add `Fold` convenience function to `ExpressionAnalyzer`.
Original-commit: flang-compiler/f18@140c983423
Reviewed-on: https://github.com/flang-compiler/f18/pull/904
Make compilation of other predefined module files depend on
__fortran_builtins.mod. Currently only iso_c_binding.f90 and
iso_fortran_env.f90 depend on it but others could in the future.
Create the .f18.mod files by copying from the .mod files so that
we don't have to worry about dependencies for those.
Original-commit: flang-compiler/f18@8209ad3d32
Reviewed-on: https://github.com/flang-compiler/f18/pull/899
This changes the license information in many of the flang source files.
- Renamed LICENSE to LICENSE.txt.
- NVIDIA Copyright lines have been removed.
- Initial lines for files follow the LLVM coding convention (file name on the first line; Emacs mode information on the first line).
- License references have been replaced with the abridged LLVM text.
- License information was removed from the test files.
- No file header was placed on test files (these weren't in most LLVM test files).
- License information was added to documentation files where it was missing.
We did not add brief file summaries to the initial line.
See http://llvm.org/docs/DeveloperPolicy.html#new-llvm-project-license-framework
for a description of the new license.
See http://llvm.org/docs/CodingStandards.html#file-headers
for a description of the new LLVM standard file header.
Original-commit: flang-compiler/f18@add6cde724
Reviewed-on: https://github.com/flang-compiler/f18/pull/887
When dumping the parse tree, if there are analyzed expressions,
assignments, or call statements, include the Fortran form of those
in the dump.
This uses the same `AnalyzedObjectsAsFortran` object that unparsing
does so it applies to the same nodes.
We already do something like this for `Name` nodes as well as some
predefined types, so abstract it into an `AsFortran` function which
returns the Fortran source for a node that is to be included in the
dump.
Original-commit: flang-compiler/f18@823b044d2a
Reviewed-on: https://github.com/flang-compiler/f18/pull/878
Tree-same-pre-rewrite: false
Add `typedAssignment` to the `AssignmentStmt` node, analagous to
`typedExpr` in `Expr`. This represents the analyzed form of an assignment.
It may be a subroutine call in the case of a user-defined assignment.
Add `GetAssignment` function to get it from the parse tree node if present.
Original-commit: flang-compiler/f18@675b0b9bb7
Reviewed-on: https://github.com/flang-compiler/f18/pull/841
Tree-same-pre-rewrite: false
Update the grammar to handle logical abbreviations (e.g. `.A.` for `.AND.`)
when the feature is enabled. Only support `.X.` when both XOR and
logical abbreviations are enabled.
Fix the driver to enable logical abbreviations with the
`-flogical-abbreviations` option. This was already documented in
`documentation/Extensions.md`.
Remove `parser::Expr::XOR` from the parse tree and immediately map
`.XOR.` to `.NEQV.` if that feature is enabled. This was already being
done during expression analysis anyway.
Add `LanguageFeatureControl::GetNames` to return all of the names of
a logical or relational operator, depending on which features are
enabled. Use these in both name resolution and expression analysis.
Add `Not` to `LogicalOperator` to help in those cases.
Fix handling of BOZ literals: A numeric operation with one real or
integer operand and the other a BOZ literal is intrinsic.
Also, unary plus with a BOZ literal operand is also intrinsic.
Original-commit: flang-compiler/f18@956bd50bc7
Reviewed-on: https://github.com/flang-compiler/f18/pull/815
- -fget-definitions finds the definition of the symbol under specified source
position.
- -fget-all-symbols finds definition locations of all symbols in a document. For
symbols found in other modules, shows which module the symbol came from.
- Tests.
- New structure SourcePosition with file, line, column information.
Original-commit: flang-compiler/f18@e0099b0900
Reviewed-on: https://github.com/flang-compiler/f18/pull/698
Tree-same-pre-rewrite: false
so makefiles with "rm *.mod" will still work. Add a
new option to f18, -intrinsic-module-directory, which
supplies a directory in which f18 searches for standard
intrinsic modules. Neither the option nor the path is passed to F18_FC.
Original-commit: flang-compiler/f18@0bbda4c39e
Reviewed-on: https://github.com/flang-compiler/f18/pull/653
Add a script called flang to the bin directory that calls
f18 with the -I option pointing to the module files for the
predefined modules. The wrapper also sets the module suffix
to .fmf to avoid naming conflicts with module files generated
by other compilers. Now, the build creates and installs both
.mod and .fmf files for the predefined module files to
accommodate the flang script.
Original-commit: flang-compiler/f18@b6c30284e7
Reviewed-on: https://github.com/flang-compiler/f18/pull/653
Tree-same-pre-rewrite: false
Change the CMakeFile.txt in tools/f18 to build the predefined modules with the just-built compiler. The mod files are created in the new "include" subdirectory of the binary target directory. The mod files are installed to ${CMAKE_INSTALL_PREFIX}/include.
The f18 driver is already installed in ${CMAKE_INSTALL_PREFIX}/bin. This change change the location of the f18 binary build to a sibling directory of "include" called "bin" instead of at the top level of the binary destination directory. This change is in anticipation of changing the driver to find the include directory using a path relative to the location of f18.
Update the test scripts to find f18 in the bin subdirectory.
Remove the simple predefined module tests from test/semantics because they are compiled as part of the build and don't need to be recompiled as part of a test.
Original-commit: flang-compiler/f18@7811ccecdc
Reviewed-on: https://github.com/flang-compiler/f18/pull/526
more fixes
Access components of constant structures
Apply implicit typing to dummy args used in automatic array dimensions
SELECTED_INT_KIND and SELECTED_REAL_KIND
Finish SELECTED_{INT,REAL}_KIND and common cases of ALL()/ANY()
Original-commit: flang-compiler/f18@e9f8e53e55
Reviewed-on: https://github.com/flang-compiler/f18/pull/472
Tree-same-pre-rewrite: false
Instead of tracking just genericName_ while in a generic interface
block or generic statement, now we immediately create a symbol for it.
A parser::Name isn't good enough because a defined-operator or
defined-io-generic-spec doesn't have a name.
Change the parse tree to add a source field to GenericSpec. Use these
as names for symbols for defined-operator and defined-io-generic-spec
(e.g. "operator(+)" or "read(formatted)").
Change the source for defined-op-name to include the dots so that they
can be distinguished from normal symbols with the same name (e.g. you
can have both ".foo." and "foo"). These symbols have names in the symbol
table like ".foo.", not "operator(.foo.)", because references to them
have that form.
Add GenericKind enum to GenericDetails and GenericBindingDetails.
This allows us to know a symbol is "assignment(=)", for example,
without having to do a string comparison.
Add GenericSpecInfo to handle analyzing the various kinds of
generic-spec and generating symbol names and GenericKind for them.
Add reference to LanguageFeatureControl to SemanticsContext so that
they can be checked during semantics. For this change, if
LogicalAbbreviations is enabled, report an error if the user tries
to define an operator named ".T." or ".F.".
Add resolve-name-utils.cc to hold utility functions and classes that
don't have to be in the ResolveNamesVisitor class hierarchy. The goal
is to reduce the size of resolve-names.cc where possible.
Original-commit: flang-compiler/f18@3081f694e2
Reviewed-on: https://github.com/flang-compiler/f18/pull/338
The cmake option -DLINK_WITH_FIR=ON causes the f18 driver to be linked
against FIR and LLVM. When it is set to OFF (the default), the call into
FIR is #ifdef'd out.
The source in lib/FIR is always built; this option only affects linking.
Original-commit: flang-compiler/f18@69569edd4c
Reviewed-on: https://github.com/flang-compiler/f18/pull/329
The new SemanticsContext holds the state of semantics whose lifetime
spans all of the compilations. It contains the scope tree (and so all
symbols), the intrinsics table, messages, and the state of options that
affect semantics (default kinds, search directories, etc.)
Original-commit: flang-compiler/f18@f04ceab81f
Reviewed-on: https://github.com/flang-compiler/f18/pull/215
Tree-same-pre-rewrite: false
Refactor to create the Semantics class that is responsible for holding
state during semantics (the scope tree and messages) and managing the
logic of the various phases of semantic processing. Eliminate static
Scope::globalScope.
The messages generated during semantic processing are accumulated in a
Messages data member of Semantics so that individual phases don't need
to emit them to std::cerr. This is now done by the driver so that it has
control over where they go and eliminates other includes of iostream.
To do this, the messages object is passed in to the various semantics
operations.
Move DumpSymbols into semantics.cc: it doesn't belong in resolve-names.cc
and it depends on the global scope, so it's as good a place as any.
The call to RewriteParseTree is also moved to Semantics.
Original-commit: flang-compiler/f18@771d0e1293
Reviewed-on: https://github.com/flang-compiler/f18/pull/186
Tree-same-pre-rewrite: false
Intrinsic types are now just a TypeCategory and a int kind. If no kind
is specified the default is used so that every type has an explicit
kind. This caused changes in the expected results of some of the tests.
Add support for "double precision" and "double complex".
Intrinsic types are now stored as values in DeclTypeSpec so none of the
KindedTypeHelper machinery is needed any more.
Eliminate DerivedTypeDef, DataComponentDef, ProcComponentDef,
TypeBoundProc. The components and bindings of a derived type are now
represented by the corresponding Scope.
Original-commit: flang-compiler/f18@4ad8ffb187
Reviewed-on: https://github.com/flang-compiler/f18/pull/182
Tree-same-pre-rewrite: false
Symbols for submodules have `ModuleDetails` with `isSubmodule` set.
Scopes for submodules have `Module` kind and have a parent scope that
is also `Module` kind.
Scopes for modules now contain a mapping of submodule name to scope
so that we can find them without having to search the scope tree or
re-read their `.mod` file.
The module file for submodule `s` with ancestor module `m` is named `m-s.mod`.
The tree structure of scopes means module file writing is now recursive.
Similarly, reading the module file for a submodule may require reading
the module files of its parent and ancestor. `ResolveNames` now requires
the parent scope to be passed in -- it is not always the global scope.
`test_modfiles.sh` now handles an argument that is a filename glob so
that the test can involve multiple files. This allows `modfile09` to
test reading of `.mod` files for modules and submodules.
Original-commit: flang-compiler/f18@2e4424dbc8
Reviewed-on: https://github.com/flang-compiler/f18/pull/160
Tree-same-pre-rewrite: false
The `-module` option specifies where to write module files. It is also
the first place search for modules files. The current options name and
behavior match PGI compilers.
To support this, reorganize `ModFileWriter` to be more like `ModFileReader`.
The gives the client a way to set the output directory and fetch the
error messages. A bunch of static member functions in `ModFileWriter`
were changed to non-member functions so they didn't need to be exposed
in `mod-files.h`.
Original-commit: flang-compiler/f18@2717846e81
Reviewed-on: https://github.com/flang-compiler/f18/pull/147
When a use-stmt is encountered for a module that isn't in the global
scope, search for and read the appropriate `.mod` file. To perform the
search, pass the search directories in to ResolveNames.
For modules that were read from `.mod` files, we have to keep the cooked
source from being deleted so that the names so that references to names
stay valid. So we store the cooked source in the Scope of the module as
a `unique_ptr`.
Add `Symbol::Flag::ModFile` to distinguish module symbols that were read
from a `.mod` file rather than from the current compilation. Use it to
prevent writing those back out.
Fix test_errors.sh to run the compiler in the temp subdirectory --
otherwise tests could be affected by `.mod` files left from previous
tests.
Original-commit: flang-compiler/f18@207065999c
Reviewed-on: https://github.com/flang-compiler/f18/pull/145
Module file writing is implemented in mod-file.cc. They need to be
written after all semantic checking. Until then, for testing, write
them out whenever names are resolved.
There is a header comment in the .mod files but it is mostly a
placeholder until we can read them in and do something with it.
Rename `Symbol::details<D>` to `Symbol::get<D>`. This asserts that the
details of the symbol match D and returns that type. But we need a way
to access the details as a variant as well (not just one of its types).
`details()` is the best name for that, especially as we already have
`set_details()`. Renaming the old `details` to `get` also better matches
`has` which is used to check which variant is present.
Original-commit: flang-compiler/f18@8d14be1a16
Reviewed-on: https://github.com/flang-compiler/f18/pull/126
Tree-same-pre-rewrite: false
When `-fdebug-dump-symbols` is supplied with `-funparse`, include symbol
information in comments in the Fortran output. This will be used for
testing to verify that correct symbols are defined and references in
the right place.
In `UnparseWithSymbols()`, walk the parse tree and collect symbol
definitions and references, organized by statement. When a symbol is
defined across several statement it is associated with the first.
The definition of implicitly defined symbols is associated with the
first reference.
To write out the symbol information, a new optional argument is added to
`Unparse()`: it is a function that is called immediately before each
statement is unparsed. We pass in a function that prints out the symbol
information collected for that statement.
Add `Symbol::GetType()` to make it easier to write the symbol types
and add `Symbol::SetType()` for uniformity.
Original-commit: flang-compiler/f18@2e827de4ad
Reviewed-on: https://github.com/flang-compiler/f18/pull/112
Tree-same-pre-rewrite: false
This consists of:
- a new kind of symbols to represent them with DerivedTypeDetails
- creating symbols for derived types when they are declared
- creating a new kind of scope for the type to hold component symbols
- resolving entity declarations of objects of derived type
- resolving references to objects of derived type and to components
- handling derived types with same name as generic
Type parameters are not yet implemented.
Refactor DeclTypeSpec to be a value class wrapping an IntrinsicTypeSpec
or a DerivedTypeSpec (or neither in the TypeStar and ClassStar cases).
Store DerivedTypeSpec objects in a new structure the current scope
MakeDerivedTypeSpec so that DeclTypeSpec can just contain a pointer to
them, as it currently does for intrinsic types.
In GenericDetails, add derivedType field to handle case where generic
and derived type have the same name. The generic is in the scope and the
derived type is referenced from the generic, similar to the case where a
generic and specific have the same name. When one of these names is
mis-recognized, we sometimes have to fix up the 'occurrences' lists
of the symbols.
Assign implicit types as soon as an entity is encountered that requires
one. Otherwise implicit derived types won't work. When we see 'x%y' we
have to know the type of x in order to resolve y. Add an Implicit flag
to mark symbols that were implicitly typed
For symbols that introduce a new scope, include a pointer back to that
scope.
Add CurrNonTypeScope() for the times when we want the current scope but
ignoring derived type scopes. For example, that happens when looking for
types or parameters, or creating implicit symbols.
Original-commit: flang-compiler/f18@9bd16da020
Reviewed-on: https://github.com/flang-compiler/f18/pull/109
Add Dump() routines based on operator<< for the type so that they are
easy to call from the debugger. Overload for both pointer and reference
types and for dumping to std::cerr or a specific ostream.
Original-commit: flang-compiler/f18@ec6676eff0
Reviewed-on: https://github.com/flang-compiler/f18/pull/99
Recognize modules and open and close the corresponding scope.
Handle PUBLIC and PRIVATE statements and set the corresponding
attributes on entity declarations in the module.
Refactoring (no functional changes): Make CheckImplicitSymbol() and
GetVariableName() overloadings private and out-of-line.
Add missing option to f18 help.
Original-commit: flang-compiler/f18@d01cacca63
Reviewed-on: https://github.com/flang-compiler/f18/pull/70
Tree-same-pre-rewrite: false
Add parse-tree-mutator.h like parse-tree-visitor.h except that the Walk
functions take non-const references to parse tree nodes so the Pre and
Post methods of the mutator that are passed around can make changes to
the parse tree.
Change ExecutionPart to be a class that wraps a list so that it can be
identified during parse tree walking.
Add Symbol* field to parser::Name for the result of symbol resolution.
In parse tree dumper, dump symbol when it is there instead of just name.
Add RewriteParseTree to walk the parse tree, fill in resolved symbols in
Name nodes, and make necessary changes to the structure. Currently that
consists of rewriting statement functions as array assignments when
appropriate.
In ResolveNames, call RewriteParseTree if the resolution was successful.
Recognize a statement function that comes after a mis-identified
statement function and report an error. resolve08.f90 tests this case.
Add -fdebug-dump-symbols to dump the scope tree and symbols in each scope.
This is implemented by DumpSymbols in resolve-names.cc. Add an optional
symbol to scopes that correspond to symbols (e.g. subprograms). Remove
debug output from ResolveNamesVisitor as this option can be used instead.
Original-commit: flang-compiler/f18@9cd3372265
Reviewed-on: https://github.com/flang-compiler/f18/pull/60
Tree-same-pre-rewrite: false
"f18 -fdebug-resolve-names -fparse-only ..." is equivalent to what
test-type used to do, so the test-type executable can be eliminated.
-fdebug-dump-parse-tree does the parse-tree dumping that test-sema
does, but test-sema does more so I have left it alone.
Original-commit: flang-compiler/f18@8b3816528e
Reviewed-on: https://github.com/flang-compiler/f18/pull/55
Includes like "../dir/file.h" should sort after local includes.
This change fixes that and applies the new formatting.
Now the order (in reverse) is:
- system includes
- includes from llvm or clang (this is from the default IncludeCategories)
- includes of ../something
- everything else
Original-commit: flang-compiler/f18@324643d63a
Reviewed-on: https://github.com/flang-compiler/f18/pull/52
ImplicitRules maintains a mapping of initial character of an identifier
to its implicit type.
ImplicitRulesVisitor visits the ImplicitStmt and related parse tree
nodes and updates the rules.
Emit errors messages in these cases:
- implicit type specified more than once for a letter
- IMPLICIT statement and IMPLICIT NONE in same scope
- IMPLICIT statement and IMPLICIT NONE(TYPE) in same scope
- more than one IMPLICIT NONE in a scope
- IMPLICIT statement with empty range of letters
- IMPLICIT NONE statement after PARAMETER statement
- TYPE or EXTERNAL repeated in an IMPLICIT NONE statement
Also start emitting error messages through the parser::Messages
interface. An instance of Messages is created from the cooked source and
saved in the MessageHandler class. It tracks the source location of the
current statement and provides a way to emit messages associated with
that statement.
Original-commit: flang-compiler/f18@e12fea9850
Reviewed-on: https://github.com/flang-compiler/f18/pull/41
Tree-same-pre-rewrite: false
Work on scanning compiler directive lines.
Fix glitch in earlier change to parse-state.h.
Add ClassifyLine(), simplify some token sequence data lifetimes and interfaces.
Handle continued directives. Obey !DIR$ FIXED and FREE in prescanner.
Some refactoring of TokenSequence API, then support initial directives (FIXED, FREE, IGNORE_TKR).
Fix !DIR$ IGNORE_TKR syntax, manual was wrong.
Debugging directive scanning & parsing.
Profiling-directed speed-up - do not map cooked source locations to Provenance until a Message is emitted. Turn some non-nullable pointers into references.
Debugging.
Handle !DIR$ IVDEP too, it is in a test.
Accept directives in the execution part.
Original-commit: flang-compiler/f18@fb2ff367ec
Reviewed-on: https://github.com/flang-compiler/f18/pull/34
Tree-same-pre-rewrite: false
Jean Perier