This fixes a series of mis-compiles by SimpleLoopUnswitch.
My measurements showed no performance regression with -O3 on AArch64
in SPEC2006, SPEC2017 and a set of internal benchmarks.
Fixes#50387, #50430
Depends on D124251.
Reviewed By: nikic, aqjune
Differential Revision: https://reviews.llvm.org/D124252
DBG_PHI instructions can refer to stack slots, to indicate that multiple
values merge together on control flow joins in that slot. This is fine --
however the slot might be merged at a later date with a slot of a different
size. In doing so, we lose information about the size the eliminated PHI.
Later analysis passes have to guess.
Improve this by attaching an optional "bit size" operand to DBG_PHI, which
only gets added for stack slots, to let us know how large a size the value
on the stack is.
Differential Revision: https://reviews.llvm.org/D124184
After building a PCH, the vector capacity is on average ~1/3 unused.
If we're going to keep it in memory for a while, reallocate to the right size.
Take care to do this once clang is destroyed so that we can reuse its
memory rather than requesting more.
Differential Revision: https://reviews.llvm.org/D124242
Image.sample instruction can be forced to return half type instead of
float when d16 flag is enabled.
This patch adds new pattern in InstCombine to detect if output of
image.sample is used later only by fptrunc which converts the type
from float to half. If pattern is detected then fptrunc and image.sample
are combined to single image.sample which is returning half type.
Later in Lowering part d16 flag is added to image sample intrinsic.
Differential Revision: https://reviews.llvm.org/D124232
This patch adds 2 missing items required for `flang-new` to be able to
generate executables:
1. The Fortran_main runtime library, which implements the main entry
point into Fortran's `PROGRAM` in Flang,
2. Extra linker flags to include Fortran runtime libraries (e.g.
Fortran_main).
Fortran_main is the bridge between object files generated by Flang and
the C runtime that takes care of program set-up at system-level. For
every Fortran `PROGRAM`, Flang generates the `_QQmain` function.
Fortran_main implements the C `main` function that simply calls
`_QQmain`.
Additionally, "<driver-path>/../lib" directory is added to the list of
search directories for libraries. This is where the required runtime
libraries are currently located. Note that this the case for the build
directory. We haven't considered installation directories/targets yet.
With this change, you can generate an executable that will print `hello,
world!` as follows:
```bash
$ cat hello.f95
PROGRAM HELLO
write(*, *) "hello, world!"
END PROGRAM HELLO
$ flang-new -flang-experimental-exec hello.f95
./a.out
hello, world!
```
NOTE 1: Fortran_main has to be a static library at all times. It invokes
`_QQmain`, which is the main entry point generated by Flang for the
given input file (you can check this with `flang-new -S hello.f95 -o - |
grep "Qmain"`). This means that Fortran_main has an unresolved
dependency at build time. The linker will allow this for a static
library. However, if Fortran_main was a shared object, then the linker
will produce an error: `undefined symbol: `_QQmain`.
NOTE 2: When Fortran runtime libraries are generated as shared libraries
(excluding Fortran_main, which is always static), you will need to
tell the dynamic linker (by e.g. tweaking LD_LIBRARY_PATH) where to look
for them when invoking the executables. For example:
```bash
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<flang-build-dir>/lib/ ./a.out
```
NOTE 3: This feature is considered experimental and currently guarded
with a flag: `-flang-experimental-exec`.
Differential Revision: https://reviews.llvm.org/D122008
[1] https://github.com/flang-compiler/f18-llvm-project
CREDITS: Fortran_main was originally written by Eric Schweitz, Jean
Perier, Peter Klausler and Steve Scalpone in the fir-dev` branch in [1].
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Peter Klausler <pklausler@nvidia.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: Steve Scalpone <sscalpone@nvidia.com
When enabling freeze-loop-unswitch-cond the inserted freeze instruction
may block unswitching of parent loops if they get inserted in a block in
the parent loop (as the llvm::Loop-based invariance check only checks
whether an instruction is in a loop block or not).
In the optimization pipeline, LICM is responsible to hoist out loop
invariant instructions to enable further unswitching. Also run LICM for
nested unswitching tests in preparation for flipping the default of
freeze-loop-unswitch-cond.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D124251
Logic in this pass assumes that all users of loop instructions are
either in the same loop or are LCSSA Phis. In fact, there can also
be users in unreachable blocks that currently break assertions.
Such users don't need to go to the next round of simplifications.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D124368
This patch unifies the wording used for readnone, readonly and writeonly
attributes. The definitions now more specifically refer to memory visible
outside the function
The motivation for the clarification is D123473.
Reviewed By: nlopes
Differential Revision: https://reviews.llvm.org/D124124
The verifier of llvm.mlir.addressof did not properly account for opaque pointers, that is, the pointer type not having an element type equal to the type of the referenced global or function. This patch fixes that by skipping the test for the element type if the pointer is opaque.
Differential Revision: https://reviews.llvm.org/D124333
- add zip_view and views::zip for C++23
- added unit tests
- implemented section 5.6 (zip) in P2321R2
I used clang-format to format the files but they look nothing like the rest of the code base. Manually indenting each line to match the styles sounds like an impossible task. Is there any clang-format file which can format it reasonable similar to the rest of the code base so that I can manually format the rest lines that look weird?
Reviewed By: ldionne, #libc, philnik, var-const
Spies: Mordante, philnik, libcxx-commits, mgorny
Differential Revision: https://reviews.llvm.org/D122806
When exprs are written unabbreviated:
- these were encoded as 5 x vbr6 = 30 bits
- now they fit exactly into a one-chunk vbr = 6 bits
clangd --check=clangd/AST.cpp reports ~1% reduction in PCH size
(42826720->42474460)
Differential Revision: https://reviews.llvm.org/D124250
After https://reviews.llvm.org/D119743 added the `AutomaticAllocationScope`
trait to loop-like constructs, the vector transfer full/partial splitting pass
started inserting allocations for temporaries within the closest loop rather
than the closest function (or other allocation scope such as `async.execute`).
While this is correct as long as the lowered code takes care of automatic
deallocation at the end of each iteration of the loop, this interferes with
downstream optimizations that expect `alloca`s to be at the function level.
Step over loops when looking for the closest allocation scope in vector
transfer full/partial splitting pass thus restoring the original behavior.
Reviewed By: hanchung
Differential Revision: https://reviews.llvm.org/D124366
The lowering code was mistakenly assuming that the second argument
in the signature provided by semantics is the DIM argument. This
caused calls with a KIND argument but no DIM to be lowered as if the
KIND argument was DIM.
Differential Revision: https://reviews.llvm.org/D124243
This is likely preferable to having it crash if one were to specify an opaque pointer type, and the actual element type is unused either way.
Differential Revision: https://reviews.llvm.org/D124334
This patch add a function foldSelectWithFCmpToFabs, and do more combine for
fneg-of-fabs.
With 'nsz':
fold (X < +/-0.0) ? X : -X or (X <= +/-0.0) ? X : -X to -fabs(x)
fold (X > +/-0.0) ? X : -X or (X >= +/-0.0) ? X : -X to -fabs(x)
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D123830
There was some duplicate code in determineStarAmpUsage and
determinePlusMinusCaretUsage
Now a `-` or `+` following `;`, `sizeof`, `co_await`, or `delete` is
regarded as a unary operator.
Now a `*` or `&` following `case` is also a unary operator.
Reviewed By: curdeius, MyDeveloperDay, HazardyKnusperkeks
Differential Revision: https://reviews.llvm.org/D121754
It was probably accidentally added there, see the discussion on change
D97625. It is certainly without effect, to quote gitattributes(5):
When deciding what attributes are assigned to a path, Git consults
[...], `.gitattributes` file in the same directory as the path in
question, and its parent directories up to the toplevel of the work
tree [...]
Running `git check-attr -a` on the files in question shows that now the
settings are indeed effective whereas before they were not.
Lastly, lit ignores the file like any dotfile, see getTestsInDirectory
of FileBasedTest in llvm/utils/lit/lit/formats/base.py. This can be
verified with `llvm-lit --show-tests clang-tools-extra/test`.
In some case the lowering of `ichar` is generating an `arith.extui` operation
with the same from/to type. This operation do not accept from/to types to be
the same. If the from/to types are identical, we do not generate the extra
operation.
Reviewed By: jeanPerier
Differential Revision: https://reviews.llvm.org/D124107
`find_compilation_database` checked only for "/" as exit point, but on Windows, this root is impossible.
Fixes#53642
Authored By: Febbe
Reviewed By: JonasToth
Differential Revision: https://reviews.llvm.org/D119481
Removing these is extremely unhelpful and just adds extra hassle. This
is really finding out whether your test script uses -mtriple or
not. You can't meaningfully delete these fields, and the resulting
module defaults to the host.
Avoid repeating CMake checks across runtimes by unifying names of
variables used for results to leverage CMake caching.
Differential Revision: https://reviews.llvm.org/D110005
This change adds an option to detect all null dereferences for
non-default address spaces, except for address spaces 256, 257 and 258.
Those address spaces are special since null dereferences are not errors.
All address spaces can be considered (except for 256, 257, and 258) by
using -analyzer-config
core.NullDereference:DetectAllNullDereferences=true. This option is
false by default, retaining the original behavior.
A LIT test was enhanced to cover this case, and the rst documentation
was updated to describe this behavior.
Reviewed By: steakhal
Differential Revision: https://reviews.llvm.org/D122841
To fix llvm-mca's error of 'found an unsupported instruction
in the input assembly sequence.' caused by the lack of
scheduling info.
Pseudo function call instructions will be expanded to `auipc`
and `jalr`, so their scheduling info are the combination of
two.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D123578
Before this,
clang empty.cc -target x86_64-apple-ios13.1-macabi \
-framework CoreServices -fuse-ld=lld
would error out with
ld64.lld: error: path/to/MacOSX.sdk/System/Library/Frameworks/
CoreServices.framework/Versions/A/Frameworks/CarbonCore.framework/
Versions/A/CarbonCore.tbd(
/System/Library/Frameworks/
CoreServices.framework/Versions/A/Frameworks/CarbonCore.framework/
Versions/A/CarbonCore) is incompatible with x86_64 (macCatalyst)
Now it works, like with ld64.
Differential Revision: https://reviews.llvm.org/D124336
This patch set LastCallToStaticBonus based on check, it has
no noticeable size reduction on an internal workload and linux kernel
with Os/Oz.
Differential Revision: https://reviews.llvm.org/D124233
Florian Hahn