Address several issues involving control flow graph generation and
structured code ops.
- Fix a problem with constructs nested inside unstructured selection
constructs. This is a general problem involving branches that are
implied rather than explicit. It is addressed in the generic genFIR
"wrapper" function that calls individual statement-specific genFIR calls.
- The previous fix requires some compensating changes in IF and DO
construct code lowering.
- Streamline the code to generate explicit DO loop variable updates.
- Fix a problem with the individual detailed genFIR calls made in the
genFIR(SelectTypeConstruct) call.
- Modify control flow graph generation to support the insertion of
deallocation and finalization code when lowering most END <construct>
statements.
Addresses and properties (bounds, length parameters) of host
variables associated in an internal procedure were all passed via
an extra tuple argument of the internal procedure.
This extra tuple is in general an overhead: it must be created and
passed, and require creating thunks when taking the address of the
internal procedure.
This patch allows not using the tuple for host global variables
(from modules, common block, or local saved variables) since they can
be instantiated from the fir.global symbol in the internal procedure
instead.
Add a fir.internal_proc attribute to mlir::FuncOp for internal procedures
so that ArrayValueCopy can still detect internal procedures even if they
do not have a tuple argument.
Differential Revision: https://reviews.llvm.org/D140288
This patch fixes:
flang/lib/Lower/PFTBuilder.cpp:1042:6: error: function 'dumpScope'
is not needed and will not be emitted
[-Werror,-Wunneeded-internal-declaration]
A submodule is a program unit that may contain the implementions of procedures
declared in an ancestor module or submodule.
Processing for the equivalence groups and variables declared in a submodule
scope is similar to existing processing for the equivalence groups and
variables in module and procedure scopes. However, module and procedure scopes
are tied directly to code in the Pre-FIR Tree (PFT), whereas processing for a
submodule must have access to an ancestor module scope that is guaranteed
to be present in a .mod file, but is not guaranteed to be in the PFT. This
difference is accommodated by tying processing directly to a front end scope.
Function scopes that can be processed on the fly are done that way; the
resulting variable information is never stored. Module and submodule scopes
whose symbol information may be needed during lowering of any number of module
procedures are instead cached on first use, and reused as needed.
These changes are a direct extension of current code. All module and submodule
variables in scope are processed, whether referenced or not. A possible
alternative would be to instead process symbols only when first used. While
this could ultimately be beneficial, such an approach must account for the
presence of equivalence groups. That information is not currently available
for on-the-fly variable processing.
Some additional changes are needed to include submodules in places where
modules must be considered, and to include separate module procedures in
places where other subprogram variants are considered. There is also a fix
for a bug involving the use of variables in an equivalence group in a
namelist group, which also involves scope processing code.
Represent the select rank statement + select rank case statement
the same way the select case statement and case statement are represented.
controlSuccessor was not correctly attributed to the next type guard stmt.
Similar to D137460 for select type construct.
Reviewed By: vdonaldson
Differential Revision: https://reviews.llvm.org/D137490
Represent the select type statement + type guard statement
the same way the select case statement and case statement are represented.
controlSuccessor was not correctly attributed to the next type guard stmt.
Reviewed By: PeteSteinfeld, vdonaldson
Differential Revision: https://reviews.llvm.org/D137460
Previously, AggregateStores were created for aggregates associated with common
blocks. As a) AggregateStoreMap uses scope and offset information to search for
aggregate stores and b) variables related to common blocks have their
offsets set relative to the common block itself, if there were multiple
equivalences and at least one involved variables defined in a common block there
was an opportunity for the scope/offset pairs to match between distinct
aggregate stores. As a result, entries in AggregateStoreMap could collide,
resulting in incorrect stores being returned for a particular variable.
To prevent these collisions, skip creating AggregateStores for aggregates which
are associated with common blocks. This information was already unused as
aggregates associated with common blocks are handled by instantiateCommon.
Fixes https://github.com/llvm/llvm-project/issues/57749
Differential Revision: https://reviews.llvm.org/D134828
Currently, lowering is promoting main program array and character
variables that are not saved into static memory.
This causes issues with equivalence initial value images because
semantics is relying on IsSaved to build the initial value of variables
in static memory. It seems more robust to have IsSaved be the place
deciding if a variable needs to be in static memory (except for common
block members).
Move the logic to decide if a main program variable must be in static
memory into evaluate::IsSaved and add two options to semantics to
replace the llvm options that were used in lowering:
- SaveMainProgram (off by default): save all main program variables.
- SaveBigMainProgramVariables (on by default): save all main program
variables that are bigger than 32 bytes.
The first options is required to run a few old programs that expect all
main program variables to be in bss (and therefore zero initialized).
The second option is added to allow performance testing: placing big
arrays in static memory seems a sane default to avoid blowing up the
stack with old programs that define big local arrays in the main
program, but since it is easier to prove that an alloca does not
escape/is not modified by calls, keeping big arrays on the stack could
yield improvements.
The logic of SaveBigMainProgramVariables is slightly changed compared to what
it was doing in lowering. The old code was placing all arrays and all
explicit length characters in static memory.
The new code is placing everything bigger than 32 bytes in static
memory. This has the advantages of being a simpler logic, and covering
the cases of scalar derived type with big array components or many
components. Small strings and arrays are now left on the stack (after
all, a character(1) can fit in register).
Note: I think it could have been nicer to add a single "integer" option
to set a threshold to place main program variables in static memory so
that this can be fine tuned by the drivers (SaveMainProgram would be
implemented by setting it to zero). But the language feature options are
not meant to carry integer options. Extending it for this seems an
overkill precedent, and placing it in SemanticsContext is weird (it is
a too low level option to be a bare member of SemanticsContext in my
opinion). So I just rolled my own dices and picked 32 for the sake of
simplicity.
Differential Revision: https://reviews.llvm.org/D134735
Branch optimization in function rewriteIfGotos attempts to rewrite code
such as
<<IfConstruct>>
1 If[Then]Stmt: if(cond) goto L
2 GotoStmt: goto L
3 EndIfStmt
<<End IfConstruct>>
4 Statement: ...
5 Statement: ...
6 Statement: L ...
to eliminate a branch and a trivial basic block to get:
<<IfConstruct>>
1 If[Then]Stmt [negate]: if(cond) goto L
4 Statement: ...
5 Statement: ...
3 EndIfStmt
<<End IfConstruct>>
6 Statement: L ...
Among other requirements, this is invalid if any statement between the
GOTO and its target is an intermediate construct statement such as a
CASE or ELSE IF statement, like the CASE DEFAULT statement in:
select case(i)
case (:2)
n = i * 10
case (5:)
n = i * 1000
if (i <= 6) goto 9 ! exit over 'case default'; may not be rewritten
n = i * 10000
case default
n = i * 100
9 end select
Evaluations for the Where and Forall constructs previously did
not have their constructExit field fixed up. This could lead to
falling through to subsequent case blocks in select case
statements if either a Where or Forall construct was the final part
of one case block. Setting the constructExit field results in the
proper branching behavior.
Fixes issue: https://github.com/llvm/llvm-project/issues/56500
Differential Revision: https://reviews.llvm.org/D129879
Change-Id: Ia868df12084520a935f087524e118bcdf47f6d7a
Fixes several bugs relating to initialization expressions. An
initialization expression has no access to dynamic resources like the
stack or the heap. It must reduce to a relocatable expression that the
loader can complete at runtime.
Adds regression test.
This patch is part of the upstreaming effort from fir-dev branch.
Reviewed By: PeteSteinfeld
Differential Revision: https://reviews.llvm.org/D128380
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Character and array results are allocated on the caller side. This
require evaluating the result interface on the call site. When calling
such functions inside an internal procedure, it is possible that the
interface is defined in the host, in which case the lengths/bounds of
the function results must be captured so that they are available in
the internal function to emit the call.
To handle this case, extend the PFT symbol visit to visit the bounds and length
parameters of functions called in the internal procedure parse tree.
This patch is part of the upstreaming effort from fir-dev branch.
Reviewed By: klausler
Differential Revision: https://reviews.llvm.org/D128371
Co-authored-by: Jean Perier <jperier@nvidia.com>
Lowering was crashing with "fatal internal error: node has not been analyzed"
if a PDT with initial component value was defined inside an internal
procedure. This is because the related expression cannot be analyzed
without the component values (which happens at the instatiation).
These expression do not need to be visited (the instantiations, if any
will be). Use the form of GetExpr that tolerates the parse tree expression
to not be analyzed into an evaluate::Expr when looking through the
symbols used in an internal procedure.
Note that the PDTs TODO will then fire (it happens after the PFT
analysis) as expected if the derived type is used.
Differential Revision: https://reviews.llvm.org/D127735
This supports lowering parse-tree to MLIR for threadprivate directive
following the OpenMP 5.1 [2.21.2] standard. Take the following as an
example:
```
program m
integer, save :: i
!$omp threadprivate(i)
call sub(i)
!$omp parallel
call sub(i)
!$omp end parallel
end
```
```
func.func @_QQmain() {
%0 = fir.address_of(@_QFEi) : !fir.ref<i32>
%1 = omp.threadprivate %0 : !fir.ref<i32> -> !fir.ref<i32>
fir.call @_QPsub(%1) : (!fir.ref<i32>) -> ()
omp.parallel {
%2 = omp.threadprivate %0 : !fir.ref<i32> -> !fir.ref<i32>
fir.call @_QPsub(%2) : (!fir.ref<i32>) -> ()
omp.terminator
}
return
}
```
A threadprivate operation (omp.threadprivate) is created for all
references to a threadprivate variable. The runtime will appropriately
return a threadprivate var (%1 as above) or its copy (%2 as above)
depending on whether it is outside or inside a parallel region. For
threadprivate access outside the parallel region, the threadprivate
operation is created in instantiateVar. Inside the parallel region, it
is created in createBodyOfOp.
One new utility function collectSymbolSet is created for collecting
all the variables with a property within a evaluation, which may be one
Fortran, or OpenMP, or OpenACC construct.
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D124226
A dummy argument in an entry point of a subprogram with multiple
entry points need not be defined in other entry points. It is only
legal to reference such an argument when calling an entry point that
does have a definition. An entry point without such a definition
needs a local "substitute" definition sufficient to generate code.
It is nonconformant to reference such a definition at runtime.
Most such definitions and associated code will be deleted as dead
code at compile time. However, that is not always possible, as in
the following code. This code is conformant if all calls to entry
point ss set m=3, and all calls to entry point ee set n=3.
subroutine ss(a, b, m, d, k) ! no x, y, n
integer :: a(m), b(a(m)), m, d(k)
integer :: x(n), y(x(n)), n
integer :: k
1 print*, m, k
print*, a
print*, b
print*, d
if (m == 3) return
entry ee(x, y, n, d, k) ! no a, b, m
print*, n, k
print*, x
print*, y
print*, d
if (n /= 3) goto 1
end
integer :: xx(3), yy(5), zz(3)
xx = 5
yy = 7
zz = 9
call ss(xx, yy, 3, zz, 3)
call ss(xx, yy, 3, zz, 3)
end
Lowering currently generates fir::UndefOp's for all unused arguments.
This is usually ok, but cases such as the one here incorrectly access
unused UndefOp arguments for m and n from an entry point that doesn't
have a proper definition.
The problem is addressed by creating a more complete definition of an
unused argument in most cases. This is implemented in large part by
moving the definition of an unused argument from mapDummiesAndResults
to mapSymbolAttributes. The code in mapSymbolAttributes then chooses
one of three code generation options, depending on information
available there.
This patch deals with dummy procedures in alternate entries, and adds
a TODO for procedure pointers (the PFTBuilder is modified to analyze
procedure pointer symbol so that they are not silently ignored, and
instead hits proper TODOs).
BoxAnalyzer is also changed because assumed-sized arrays were wrongfully
categorized as constant shape arrays. This had no impact, except when
there were unused entry points.
Co-authored-by: jeanPerier <jperier@nvidia.com>
Differential Revision: https://reviews.llvm.org/D125867
Semantics is not preventing a named common block to appear with
different size in a same file (named common block should always have
the same storage size (see Fortran 2018 8.10.2.5), but it is a common
extension to accept different sizes).
Lowering was not coping with this well, since it just use the first
common block appearance, starting with BLOCK DATAs to define common
blocks (this also was an issue with the blank common block, which can
legally appear with different size in different scoping units).
Semantics is also not preventing named common from being initialized
outside of a BLOCK DATA, and lowering was dealing badly with this,
since it only gave an initial value to common blocks Globals if the
first common block appearance, starting with BLOCK DATAs had an initial
value.
Semantics is also allowing blank common to be initialized, while
lowering was assuming this would never happen, and was never creating
an initial value for it.
Lastly, semantics was not complaining if a COMMON block was initialized
in several scoping unit in a same file, while lowering can only generate
one of these initial value.
To fix this, add a structure to keep track of COMMON block properties
(biggest size, and initial value if any) at the Program level. Once the
size of a common block appearance is know, the common block appearance
is checked against this information. It allows semantics to emit an error
in case of multiple initialization in different scopes of a same common
block, and to warn in case named common blocks appears with different
sizes. Lastly, this allows lowering to use the Program level info about
common blocks to emit the right GlobalOp for a Common Block, regardless
of the COMMON Block appearances order: It emits a GlobalOp with the
biggest size, whose lowest bytes are initialized with the initial value
if any is given in a scope where the common block appears.
Lowering is updated to go emit the common blocks before anything else so
that the related GlobalOps are available when lowering the scopes where
common block appear. It is also updated to not assume that blank common
are never initialized.
Differential Revision: https://reviews.llvm.org/D124622
Lowering of FailImage statement generates a runtime call and the
unreachable operation. The unreachable operation cannot terminate
a structured operation like the IF operation, hence mark as
unstructured.
Note: This patch is part of upstreaming code from the fir-dev branch of
https://github.com/flang-compiler/f18-llvm-project.
Reviewed By: clementval
Differential Revision: https://reviews.llvm.org/D124520
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Push the ModuleLikeUnit evalutionList when entering module unit. Pop it
when exiting module unit if there is no module procedure. Otherwise, pop
it when entering the first module procedure.
Reviewed By: V Donaldson
Differential Revision: https://reviews.llvm.org/D120460
OpenMP/OpenACC declarative directives can also be used in module unit.
Add support for dump them in module.
Reviewed By: kiranchandramohan, V Donaldson
Differential Revision: https://reviews.llvm.org/D120459
This patch update the PFTBuilder to be able to lower
the construct present in semantics.
This is a building block for other lowering patches that will be posted soon.
This patch is part of the upstreaming effort from fir-dev branch.
Reviewed By: PeteSteinfeld, schweitz
Differential Revision: https://reviews.llvm.org/D120336
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: V Donaldson <vdonaldson@nvidia.com>
The PFT has been updated to support Fortran 77.
clang-tidy cleanup.
Authors: Val Donaldson, Jean Perier, Eric Schweitz, et.al.
Differential Revision: https://reviews.llvm.org/D98283
Msvc reports the following error when a ReferenceVariantBase is constructed using an r-value reference or instantiated as std::vector template parameter. The error message is:
```
PFTBuilder.h(59,1): error C2665: 'std::variant<...>::variant': none of the 2 overloads could convert all the argument types
variant(1248,1): message : could be 'std::variant<...>::variant(std::variant<...> &&) noexcept(false)'
variant(1248,1): message : or 'std::variant<...>::variant(const std::variant<...> &) noexcept(false)'
PFTBuilder.h(59,1): message : while trying to match the argument list '(common::Reference<lower::pft::ReferenceVariantBase<false,...>>)'
```
Work around the ambiguity by only taking `common::Reference` arguments in the constructor. That is, conversion to common::Reference has to be done be the caller instead of being done inside the ctor. Unfortunately, with this change clang/gcc (but not msvc) insist on that the ReferenceVariantBase is stored in a `std::initializer_list`-initialized variable before being used, like being passed to a function or returned.
This patch is part of the series to make flang compilable with MS Visual Studio <http://lists.llvm.org/pipermail/flang-dev/2020-July/000448.html>.
Reviewed By: DavidTruby
Differential Revision: https://reviews.llvm.org/D88109
In order for these files to build properly, this patch rolls up a number of changes that have been made to various files that have been upstreamed.
Implementations for the interfaces included in Bridge.h and IntrinsicCall.h will be included in a future diff.
Differential revision: https://reviews.llvm.org/D82608
structure for upstreaming to llvm-project.
These files have had many changes since they were originally upstreamed.
Some of the changes are cosmetic. Most of the functional changes were
done to support the lowering of control-flow syntax from the front-end
parse trees to the FIR dialect.
This patch is meant to be a reviewable size. The functionality it
provides will be used by code yet to be upstreamed in lowering.
review comments:
[review D80449][NFC] make PFT ParentVariant a ReferenceVariant
ReferenceVariant had to be slightly updated to also support
non constant references (which is required for ParentType).
[review D80449] extend Variable implementation beyond a comment
