After implemented this hook, we will model the memory dependency in the scheduling dependency graph more precise,
and will have more opportunity to reorder the load/stores, as they didn't have the dependency at some condition
Differential Revision: https://reviews.llvm.org/D63804
llvm-svn: 364886
For a given floating point load / store pair, if the load value isn't used by any other operations,
then consider transforming the pair to integer load / store operations if the target deems the transformation profitable.
And we can exploiting much more when there are other operation nodes with chain operand between the load/store pair
so long as we keep the chain ordering original. We only replace the register used to load/store from float to integer.
I only add testcase in ARM because the TLI.isDesirableToTransformToIntegerOp hook is only enabled in ARM target.
Differential Revision: https://reviews.llvm.org/D60601
llvm-svn: 364883
They appear to fail to link in various 32-bit configurations for unknown
reasons. This change was already reverted, and it seems preferable to me
to make forward progress and remove this once the problems are fully
understood.
llvm-svn: 364877
Summary:
In cross mode, we build a separate NATIVE tblgen that runs on the
host and is used during the build. Separately, we have a flag that
disables building all executables in utils/. Of course generally,
this doesn't turn off tblgen, since we need that during the build.
In cross mode, however, that tblegen is useless since we never
actually use it. Furthermore, it can be actively problematic if the
cross toolchain doesn't like building executables for whatever reason.
And even if building executables works fine, we can at least save
compile time by omitting it from the target build. There's two changes
needed to make this happen:
- Stop creating a dependency from the native tool to the target tool.
No such dependency is required for a correct build, so I'm not entirely
sure why it was there in the first place.
- If utils were disabled on the CMake command line and we're in cross mode,
respect that by excluding it from the install target (using EXCLUDE_FROM_ALL).
Reviewers: smeenai
Differential Revision: https://reviews.llvm.org/D64032
llvm-svn: 364872
Due to RVO the target region of a function that returns an object by
value isn't necessarily a temporary object region; it may be an
arbitrary memory region. In particular, it may be a field of a bigger
object.
Make sure we don't invalidate the bigger object when said function is
evaluated conservatively.
Differential Revision: https://reviews.llvm.org/D63968
llvm-svn: 364870
The NonnullGlobalConstants checker models the rule "it doesn't make sense
to make a constant global pointer and initialize it to null"; it makes sure
that whatever it's initialized with is known to be non-null.
Ironically, annotating the type of the pointer as _Nonnull breaks the checker.
Fix handling of the _Nonnull annotation so that it was instead one more reason
to believe that the value is non-null.
Differential Revision: https://reviews.llvm.org/D63956
llvm-svn: 364869
This patch uses the new CDF_MaybeBuiltin flag to handle C library functions.
It's mostly an NFC/refactoring pass, but it does fix a bug in handling memset()
when it expands to __builtin___memset_chk() because the latter has
one more argument and memset() handling code was trying to match
the exact number of arguments. Now the code is deduplicated and there's
less room for mistakes.
Differential Revision: https://reviews.llvm.org/D62557
llvm-svn: 364868
When matching C standard library functions in the checker, it's easy to forget
that they are often implemented as macros that are expanded to builtins.
Such builtins would have a different name, so matching the callee identifier
would fail, or may sometimes have more arguments than expected, so matching
the exact number of arguments would fail, but this is fine as long as we have
all the arguments that we need in their respective places.
This patch adds a set of flags to the CallDescription class so that to handle
various special matching rules, and adds the first flag into this set,
which enables a more fuzzy matching for functions that
may be implemented as compiler builtins.
Differential Revision: https://reviews.llvm.org/D62556
llvm-svn: 364867
It encapsulates the procedure of figuring out whether a call event
corresponds to a function that's modeled by a checker.
Checker developers no longer need to worry about performance of
lookups into their own custom maps.
Add unittests - which finally test CallDescription itself as well.
Differential Revision: https://reviews.llvm.org/D62441
llvm-svn: 364866
Slightly cleanup emission of horizontal lines and unhardcode the title
for generic maps.
Differential Revision: https://reviews.llvm.org/D64041
llvm-svn: 364865
It looks like when this test was added, it was based on TestHelloWorld and it ended up with the same class name. This is an issue because the logs associated with the tests use the class name as the identifier for the file and if two tests have the same name their logs overwrite each other. On non-windows, this just means we lose one of the logs, but on Windows this means that one of the tests will fail occasionally because the file are locked by the other test.
llvm-svn: 364860
Use both one bit and signbit shifting to check for one bit merge.
Reviewers: lebedev.ri, spatel, efriedma, craig.topper
Reviewed By: lebedev.ri
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D63903
llvm-svn: 364857
Extends the transform from:
rL364341
...to include another (more common?) pattern that tests whether a
value is a power-of-2 (including or excluding zero).
llvm-svn: 364856
D60974 added tests which incorrectly assume that llvm-readelf is available. This is a bad assumption, it should instead declare the dependency explicitly in the tests.
llvm-svn: 364855
I'm not able to reproduce the reproducer flakiness we're seeing on
GreenDragon. I want to add this assert to find out if the GDB remote
packets are somehow getting out of sync when this happens.
llvm-svn: 364852
v2i64 vzload defines a 64-bit memory access. It doesn't look like
we have any coverage for this either way.
Also remove some vzload usages where the instruction loads only
16-bits.
llvm-svn: 364851
Summary:
Instead of falling back to ObjCLanguageRuntime, we should be falling
back to every loaded language runtime. This makes ValueObject more
language agnostic.
Reviewers: labath, compnerd, JDevlieghere, davide
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D63240
llvm-svn: 364845
If the requested source type an be used as a merge source type, create
a merge of merges. This avoids creating large, illegal extensions and
bit-ops directly to the result type.
llvm-svn: 364841
These instructions only read 64-bits of memory so we shouldn't
allow a full vector width load to be pattern matched in case it
is marked volatile.
Instead allow vzload or scalar_to_vector+load.
Also add a DAG combine to turn full vector loads into vzload when
used by one of these instructions if the load isn't volatile.
This fixes another case for PR42079
llvm-svn: 364838
Craig Topper