When translating offset info from the callee at a call site, first check if the
offset is Unknown. Any offset in the caller should be added only if the callee
offset is valid.
Differential Revision: https://reviews.llvm.org/D137011
Loop Fusion (Function Pass) requires loops in simplified form. With
legacy-pm, loop-simplify pass is added as a dependency for loop-fusion.
But the new pass manager does not always ensure this format. This patch
tries to invoke simplifyLoop() on loops that are not in simplified form
only for new PM.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D136781
Updates the aarch64 pointer jump stub sequence to:
ADRP x16, <ptr>@page21
LDR x16, [x16, <ptr>@pageoff12]
BR x16
from:
LDR x16, <ptr>@ldrimm19
BR x16
The old sequence can only reference pointers within +/-1Mb, which may not be
enough for large object files. The new sequence can reach pointers within
+/-4Gb. (A future pre-fixup-pass could apply range-based optimizations to
turn this into an ldr-br-nop sequence.)
Also adds createPointerJumpStubBlock and createAnonymousPointerJumpStub
utilities along the same lines as their x86-64 counterparts.
The PLTTableManager is updated to use the new utility functions.
This patch fix issues found during call to deallocate.
fir.class can be handled the same way as fir.box in fir.store operation
code generation. In MutableBox::createNewFirBox, the fir.class is
also already a boxed entity.
Reviewed By: PeteSteinfeld
Differential Revision: https://reviews.llvm.org/D137079
A part of the unit test CommandLineTest/BadResponseFile, added in
the commit fd3d7a9f8cbb (Handle errors in expansion of response files)
fails on AIX. The test checks if an error is issued when `file` in
`@file` is a path to directory. This change excludes that check on
AIX platform.
Differential Revision: https://reviews.llvm.org/D136090
Re-apply of: 3d0e9edd8e53fb72e85084f4170513159212839a
Reverted in: 0cb65b0a585c8b3d4a8a2aefe994a8fc907934f8
A function parameter was using the wrong type 'llvm::TargetRegion' instead of
'const llvm:: TargetRegion&', which caused the error in the address sanitizer.
The correct type is now used.
This patch puts the individual target region information attributes into a
struct so that the nested mappings are not needed and passing the information
around is simplified.
Reviewed By: jdoerfert, mikerice
Differential Revision: https://reviews.llvm.org/D136601
When multiple threads invoke checkDeviceAndCtors, both of them may read true
from the shared variable Device.HasPendingGlobals, and then invoke initLibrary
redundantly. Therefore only protecting the access to Device.HasPendingGlobals
is not sufficient to guarantee that initLibrary is invoked just once.
To fix this race condition, we move the invocation of initLibrary into the
critical section, and remove the same lock inside initLibrary.
Differential Revision: https://reviews.llvm.org/D136952
This patch moves intrinsic evaluate::Constant<T> lowering into its own
unit outside of ScalarExpr and genarr lowering so that it can
be used by the new lowering without any changes.
DerivedType lowering cannot be shared at that stage because it is too
correlated with the current lowering (requires structure constructor
and designator lowering).
The code had to be refactored quite a bit so that it could be carved
out, but the only "functional" change is that the length of character
arrays lowered by genarr is now `index` instead of `i64` (see test change).
One non-functional benefit of the change is that `toEvExpr` is not
needed anymore and some compile time copies of big constant arrays
that it was causing are removed (see old calls in previous genarr code),
although I am not sure any compile time speed-ups are visible here.
Differential Revision: https://reviews.llvm.org/D136955
Update docs to describe the behaviour when a default-valued attribute is
used as an optional-group's anchor after D134993.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D136987
PragmaIncludes captures the pragma-based header-mapping information, it is used
in the "Location => Header" step to determine the final spelling header for a
symbol (rather than the header directive).
The structure is by design to be used inside the include-cleaner library
and clangd.
Differential Revision: https://reviews.llvm.org/D136071
The existing way of creating the predicate in the guard blocks uses
a boolean value per outgoing block. This increases the number of live
booleans as the number of outgoing blocks increases. The new way added
in this change is to store one integer to represent the outgoing block
we want to branch to, then at each guard block, an integer equality
check is performed to decide which a specific outgoing block is taken.
Using an integer reduces the number of live values and decreases
register pressure especially in cases where there are a large number
of outgoing blocks. The integer based approach is used when the
number of outgoing blocks crosses a threshold, which is currently set
to 32.
Patch by Ruiling Song.
Differential review: https://reviews.llvm.org/D127831
Exclude building googletest and LLVMTestingSupport libraries from
the `all` target. If unittests are being built, these libraries will
be built as a dependency anyway. If they are not being built, building
them makes little sense as they are not installed or used otherwise.
This will also make standalone builds of other projects easier, as it
makes it possible to include these directories without having to cover
them with additional conditions to prevent them from being built
unconditionally.
Differential Revision: https://reviews.llvm.org/D137035
The runtime function `AllocatableDeallocate/PointerDeallocate` is reponsible to
deallocate and finalize it. This patch just adds tests to make sure the correct
runtime function is called.
Reviewed By: PeteSteinfeld
Differential Revision: https://reviews.llvm.org/D137068
This patch adds the assembly/disassembly for the following instructions:
sqrshrn : Signed saturating rounding shift right narrow by immediate and interleave
sqrshrun : Signed saturating rounding shift right unsigned narrow by immediate and interleave
uqrshrn : Unsigned saturating rounding shift right narrow by immediate and interleave
The reference can be found here:
https://developer.arm.com/documentation/ddi0602/2022-09
Differential Revision: https://reviews.llvm.org/D136690
This is a sibling to:
6064e92b0a84
...but we canonicalize the shl+add to shl+xor,
so the pattern is different than I expected:
https://alive2.llvm.org/ce/z/8CX16e
I have not found any patterns that are safe
to propagate no-wrap, so that is not included
here.
The sme2 predicate was as AssemblerPredicate, not
AssemblerPredicateWithAll like all the other features, meaning it wasn't
included in +all. This fixes that inconsistency, allowing the
instructions to be decoded by default.
Differential Revision: https://reviews.llvm.org/D137016
This patch implements the `GetFunctionDisplayName` API which gets
used by the frame-formatting code to decide how to print a
function name.
Currently this API trivially returns `false`, so we try to parse
the demangled function base-name by hand. We try find the closing
parenthesis by doing a forward scan through the demangled name. However,
for arguments that contain parenthesis (e.g., function pointers)
this would leave garbage in the frame function name.
By re-using the `CPlusPlusLanguage` parser for this we offload the
need to parse function names to a component that knows how to do this
already.
We leave the existing parsing code in `FormatEntity` since it's used
in cases where a language-plugin is not available (and is not
necessarily C++ specific).
**Example**
For following function:
```
int foo(std::function<int(void)> const& func) { return 1; }
```
Before patch:
```
frame #0: 0x000000010000151c a.out`foo(func= Function = bar() )> const&) at sample.cpp:11:49
```
After patch:
```
frame #0: 0x000000010000151c a.out`foo(func= Function = bar() ) at sample.cpp:11:49
```
**Testing**
* Added shell test
This patch adds a way to extract the return type out
of the `CPlusPlusNameParser`. This will be useful
for cases where we want a function's basename *and* the
return type but not the function arguments; this is
currently not possible (the parser either gives us the
full name or just the basename). Since the parser knows
how to handle return types already we should just expose
this to users that need it.
**Testing**
* Added unit-tests
Differential Revision: https://reviews.llvm.org/D136935
Hopefully makes the code more readable and allows
us to re-use argument pretty-printing code from
the `CPlusPlusLanguage` plugin in a follow-up commit.
Differential Revision: https://reviews.llvm.org/D136934
Only add the option processing and store the result. No attributes are
added to FIR yet.
Only the "off" and "fast" options are supported. "fast-honor-pragmas" is not applicable because we do not implement `#pragma clang fp contract()` in Fortran [1]. "on" is not supported because it is unclear how to fuse only within individual statements. gfortran also does not implement "on": treating it as an "off".
Currently the default value is "off" to preserve existing behavior. gfortran uses "fast" by default and that may be the right thing for flang-new after further discussion in the future, but that can be changed separately. gfortran's documentation is available [[ https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html | here ]].
[1] https://clang.llvm.org/docs/LanguageExtensions.html#extensions-to-specify-floating-point-flags
Reviewed By: vzakhari, awarzynski
Differential Revision: https://reviews.llvm.org/D136080
This aligns with what was done for SVE, which consistently uses 32bit
immediates at the LLVM IR level.
Additionally, this change forces the intrinsic operands to be immediates
using ImmArg<>, which subsequently requires the codegenerator to match
TargetConstants instead of Constants.
Reviewed By: paulwalker-arm
Differential Revision: https://reviews.llvm.org/D136933
When streaming mode is enabled, lower some operations and disable some code paths;
to force generateing code compatible to streaming mode.
Add streaming-mode flag for new sve-fixed-length testing files:
build_vector.ll
concat.ll
extract-subvector.ll
extract-vector-elt.ll
int-shifts.ll
loads.ll
shuffle.ll
stores.ll
Differential Revision: https://reviews.llvm.org/D135564
The OpenMP standard specifies the region inside to be a structured
block. This means that branches into or out of the region is not
permitted but branches are permitted inside the OpenMP single region.
This is currently modeled by having a region of kind `AnyRegion` for
all the OpenMP ops. In future we can have a more strictly defined
region.
Fixes#58396.
Reviewed By: peixin
Differential Revision: https://reviews.llvm.org/D137023
Fortran standard 7.3.2.3 point 7 mentions that a diassociated
pointer dynamic type is its declared type.
in 9.7.2 note 1, when a NULLIFY statement is applied to a polymorphic pointer,
its dynamic type becomes the same as its declared type.
This patch enforce these standard points by calling the runtime function
`PointerNullifyDerived` with the declared type descriptor.
Reviewed By: jeanPerier
Differential Revision: https://reviews.llvm.org/D136948
Follow-up to D135524, to replace two more deprecated C functions
in the OCaml bindings. const_in_bounds_gep now accepts the source
element type as argument, and const_element has been changed into
aggregate_element, which works on a wider range of constants and
returns an option.
Differential Revision: https://reviews.llvm.org/D136914
programUndefinedIfUndefOrPoison used to eagerly propagate the fact that
a value is poison to the users of the value. The problem is that if the
value has a lot of uses (orders of magnitude more than the scanning
limit we use in this function), then we spend the bulk of our time in
eagerly propagating the poison property, which we will mostly never use
later anyway due to the scanning limit.
I have a test case (of ~50k lines of machine generated C++), where this
results in ~60% of 35s compilation time being spent doing just this
eager propagation.
This patch changes programUndefinedIfUndefOrPoison to only propagate to
instructions actually visited, looking back to see if their operands are
poison. This should be equivalent and no functional change is intended,
but we regain virtually all of the 60% compilation time spent in this
function in my test case (i.e.: a 2.5x total compilation speedup).
Differential Revision: https://reviews.llvm.org/D137027
