Functions that do not contain any code still have to be emitted. This
occurs on AArch64 where functions can consist only of a constant island.
To support fragment semantics in code emission, this commits adds a
guaranteed main fragment to function layout. This fragment might be
empty, but allows us omit checks whether the function is empty in most
places.
Reviewed By: rafauler
Differential Revision: https://reviews.llvm.org/D130051
TragetLowering had two last InstructionCost related `getTypeLegalizationCost()`
and `getScalingFactorCost()` members, but all other costs are processed in TTI.
E.g. it is not comfortable to use other TTI members in these two functions
overrided in a target.
Minor refactoring: `getTypeLegalizationCost()` now doesn't need DataLayout
parameter - it was always passed from TTI.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D117723
_HAS_EXCEPTIONS=0 allows disabling the exception parts of the MS STL
and vcruntime, and e.g. compiler-rt/lib/fuzzer sets this define (to
work around issues with MS STL). If using libc++ instead of MS STL,
this define previously broke the libc++ headers.
If _HAS_EXCEPTIONS is set to 0, the vcruntime_exception.h header
doesn't define the ABI base class std::exception. If no exceptions
are going to be thrown, this probably is fine (although it also
breaks using subclasses of it as regular objects that aren't thrown),
but it requires ifdeffing out all subclasses of all exception/error
derived objects (which are sprinkled throughout the headers).
Instead, libc++ will supply an ABI compatible definition when
_HAS_EXCEPTIONS is set to 0, which will make the class hierarchies
complete.
In this build configuration, one can still create instances of
exception subclasses, and those objects will be ABI incompatible
with the ones from when _HAS_EXCEPTIONS isn't defined to 0 - but
one may argue that's a pathological/self-imposed problem in that case.
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D103947
The restrictions added in D131704 were not sufficient to avoid all non-constant
expression contexts. In particular constant initialization cases.
We need to check EvaluatingDecl to detect if the variable we are initializing is
constexpr or not.
At this point it looks like this is the remaining case affecting various projects
with this diagnostic.
Differential Revision: https://reviews.llvm.org/D131874
The `arith.addi_carry` op implements integer addition with overflows. The carry is returned via the second result, as `i1`.
Reviewed By: antiagainst, bondhugula
Differential Revision: https://reviews.llvm.org/D131893
The current code may not always work correctly, e.g.:
https://github.com/llvm/llvm-project/issues/57201
I added 'enable-experimental' pass option so that SUM simplification
may be enabled in LIT tests, but it is not enabled when the pass
is added to the passes pipeline.
Differential Revision: https://reviews.llvm.org/D131640
Before, NativePDB uses scoped range as a workaround for value range, that causes
problems (e.g. a variable's value can only have one range, but usually a
variable's value is located at different address ranges, each at different
locations, in optimized build).
This patch let NativePDB switch to DWARFLocationList so a variable's value can
be described at multiple non-overlapped address ranges and each range maps to a
location.
Because overlapping ranges exists, here's peference when choosing ranges:
1. Always prefer whole value locations. Suppose a variable size is 8 bytes, one record is that for range [1, 5) first 4 bytes is at ecx, and another record is that for range [2, 8) the 8 bytes value is at rdx. This results: [1, 2) has first 4 bytes at ecx, [2, 8) has the whole value at rdx.
2. Always prefer the locations parsed later. Suppose first record is that for range [1, 5) value is at ecx, second record is that for range [2, 6) value is at eax. This results: [1, 2) -> ecx, [2, 6) -> eax.
Differential Revision: https://reviews.llvm.org/D130796
This is a potentially better alternative to D131452 that also
should avoid the infinite loop bug from:
issue #56403
This is again a minimal fix to reduce merging pain for the
release. But if this makes sense, then we might want to guard
all of the RTLIB generation (and other libcalls?) with a
similar name check.
Differential Revision: https://reviews.llvm.org/D131521
These are useful in builders of Ops taking DenseArrayAttrs or for use in Rewriter, to create constant instances.
Differential Revision: https://reviews.llvm.org/D132067
Improve copy statistics:
- Count copies from or to physical registers: They are used to model function parameters and calling conventions and the register allocator optimizes for them.
- Check physical registers assigned to virtual registers and stop counting "identity" `COPY`s where source and destination is the same physical registers; they will be removed in the `virtregmap` pass anyway.
Differential Revision: https://reviews.llvm.org/D131932
The current machine function splitter is reliant on profile data to do profile summary analysis to split blocks into cold section. This may sometimes limit the usage of machine function splitter especially in cases where we could do some form of static analysis to split out cold blocks if profile data is absent or profile data which may be faulty (Consider Sample PGO).
Of all code that could statically be marked cold Exception handling blocks are one of them (In fact BFI framework also tends to mark them as cold), and the most in size contribution. In my experiments I found out Exception handling pads and all code reachable from there account for up to 6-8% of the .text section on modern production binaries. This patch introduces a flag to split out all Exception handling blocks and blocks only reachable from Exceptional Handling pad to cold section. This flag has shown to give a performance win of up to 0.1% in terms of average cycles and instructions executed on internal facebook search service.
Reviewed By: snehasish
Differential Revision: https://reviews.llvm.org/D131824
For now, this only builds the dylib, so using `-fsanitize=undefined`
with `-static-libsan` or `fsanitize-minimal-runtime` still won't
work -- but the common case does work.
Differential Revision: https://reviews.llvm.org/D131969
Add the `-enable-post-pgo-loop-rotation` option to enable or disable the loop rotation transformation [1]. With some instrumentations, e.g., function entry coverage [2], loop rotation is not necessary and can lead to some surprise differences in codegen, even for functions where instrumentation is blocked with `noprofile` or `skipprofile`. The default value is `true` so the default behavior does not change.
[1] https://www.llvm.org/docs/LoopTerminology.html#loop-terminology-loop-rotate
[2] https://reviews.llvm.org/D116180
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D131817
Previously the specification was in a table in paragraph 14 of [container.requirements.general] but it has since been given its own sub section header, 24.2.2.4 [container.opt.reqmts].
This happened in 93ff092d1c
This update also includes misc. other cleanups of the spaceship projects sheet.
Reviewed By: Mordante, avogelsgesang, #libc
Differential Revision: https://reviews.llvm.org/D131928
Adding more supported storage classes in pointer type
Removing support for arithmetic ops as this is not available on the tensor hardware
Reviwed By: antiagainst
Differential Revision: https://reviews.llvm.org/D132039
An issue I often see in codebases compiled for unusual platforms is
that header search paths are specified manually and are subtly wrong.
For example, people will manually add `-isystem <some-toolchain>/usr/include`,
which ends up messing up the layering of header search paths required by
libc++ (because the C Standard Library now appears *before* libc++ in
the search paths). Without this patch, this will end up causing
compilation errors that are pretty inscrutable. This patch aims to
improve the user experience by diagnosing this issue explicitly.
In all cases I can think of, I would expect that a compilation error
occur if these header search paths are not layered properly. This
should only provide an explicit diagnostic instead of failing due
to seemingly unrelated compilation errors.
Differential Revision: https://reviews.llvm.org/D131441
llvm-objdump takes foo-bar style flags, while llvm-otool takes foo_bar style
flags. dyld_info was the only exception to that.
Add a -dyld_info flag to llvm-otool instead.
(Both in llvm-objdump and llvm-otool, the flag doesn't really do anything
yet.)
Differential Revision: https://reviews.llvm.org/D131897
When there are no section headers section information printed by llvm-readelf is not useful and unnecessarily verbose. When there are no program headers there's a similar verbose output shown when section mapping is requested. Simplifying the message shown in these cases to match the behavior of `GNU binuntils readelf`.
Reviewed By: jhenderson, MaskRay
Differential Revision: https://reviews.llvm.org/D130670
Extracted from D131729 where we handled C==0. It's now generalized
to more constants.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D132000
This refactors the code into a separate function with early returns.
D132000 adds an additional operation to the if/else that selects
NewLHS, but can otherwise share the rest of the code.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D132002
This commit adds tests to ensure that queue-specific breakpoints
work as expected, as this feature wasn't being tested before.
Differential Revision: https://reviews.llvm.org/D131605
`add_llvm_library()` would cause the proto target to be added in `LLVMExports.cmake` and could fail an external configuration
of compiler-rt with error "The following imported targets are referenced, but are missing: grpc++ protobuf::libprotobuf"
Differential Revision: https://reviews.llvm.org/D132033
Done according to @phosek's comments in D117537, but not done then to
separate pure refactor (that) from possible behavior change (this).
Wasn't working before, but I think that was due to an issue of mismatched variable names fixed in D110005.
Reviewed By: phosek, #libunwind, #libc_abi
Differential Revision: https://reviews.llvm.org/D117833
This reverts commit 8c4aea438c310816bb4e4f9a32d783381ef3182e.
Needed because buildbot failures (warnings) gave a clue that there was
a functional bug in the ARM rejection logic.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D132037
Implement the new sparse_tensor.reduce operation which
accepts a starting identity value and a code block
describing how to perform the reduction.
Reviewed by: aartbik
Differential Revision: https://reviews.llvm.org/D130573
Remove a test that relied on the underlying instruction latency modeling.
Such dependency blocks efforts such as D79483 to improve this cost modeling.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D132029
There are some of this supported in various places, but the
basic conversion of single operations to LLVM was not supported.
Adding this to allow Flang to use these.
Reviewed By: bondhugula
Differential Revision: https://reviews.llvm.org/D131912
While working on D129964 I noticed some code hadn't been uglyfied, this
rectifies the issue.
Depends on D129964
Reviewed By: #libc, philnik
Differential Revision: https://reviews.llvm.org/D131834
Since we branched LLVM install Clang 16 and remove Clang 12.
Currently our Docker installs 4 versions of Clang so our CI can use the
same image for both the main and the release branch. This wasn't done for
the other Clang tools so they always use the same version for testing
the main and the release branch. Instead install 2 versions for the
tools.
However it seems the default for Clang and its tools were the latest
released version instead of the ToT. To lessen the risk of breaking the
release CI, version 14 is installed hard-coded as a temporary solution.
Updating the main branch to use the Clang 16 compiler will be done in a
separate patch.
Reviewed By: ldionne, #libc
Differential Revision: https://reviews.llvm.org/D131324
This interface allows a target to reject a proposed
SMS schedule. For Hexagon/PowerPC, all schedules
are accepted, leaving behavior unchanged. For ARM,
schedules which exceed register pressure limits are
rejected.
Also, two RegisterPressureTracker methods now need to be public so
that register pressure can be computed by more callers.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D128941
This test fails for Clang versions < 14.0 for `dsym` variants.
`dsymutil` strips debug info for classes with only static members.
Thus move the failing assertions into the XFAIL test case.
Differential Revision: https://reviews.llvm.org/D132004
In D131869 we noticed that we jump through some hoops because we parse the
tolerance option used in MisExpect.cpp into a 64-bit integer. This is
unnecessary, since the value can only be in the range [0, 100).
This patch changes the underlying type to be 32-bit from where it is
parsed in Clang through to it's use in LLVM.
Reviewed By: jloser
Differential Revision: https://reviews.llvm.org/D131935
If a function only has a few instructions, instrumentation can significantly increase the size and performance overhead of that function. Add the `-pgo-function-size-threshold` option to select a size threshold so these small functions are not instrumented.
A similar option `-fxray-instruction-threshold=<N>` is used for XRay to reduce binary size overhead [1].
[1] https://www.llvm.org/docs/XRay.html
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D131816
This reverts commit 258c3aee54e11bc5c5d8ac137eb15e8d5bbcc7e4.
This should fix the libc++ issue that caused the revert, by re-designing
slightly how we determined when we should evaluate the constraints.
Additionally, many of the other components to the original patch (the
NFC parts) were committed separately to shrink the size of this patch
for review.
Differential Revision: https://reviews.llvm.org/D126907
