This is required in hermetic testing downstream. It is not complete, and
will not work on hardware, however it runs on QEMU, and can report a
pass/fail on our tests.
The `sycl_kernel_entry_point` attribute only appertains to functions.
This change adds missing tests to validate that an appropriate
diagnostic is issued for likely unintentional application of the
attribute to a null declaration or a null statement; perhaps due to an
unintended semicolon.
In this patch:
1. Adapt the output of printDelinearization() to match what the script
expects to see, "Printing analysis 'Delinearization' for function".
2. Register the "Delinearization" pass in the update script.
3. Automatically remove the existing CHECK statements.
4. Update the tests with a run of:
```
$ ~/llvm-project/llvm/utils/update_analyze_test_checks.py
--opt=~/llvm-project/build/bin/opt
~/llvm-project/llvm/test/Analysis/Delinearization/*.ll
```
5. Drop llvm/test/Analysis/Delinearization/undef.ll because the test
does not pass the github-actions "undef deprecator."
6. Only run delinearization pretty-print pass on the most internal loop.
7. Only run delinearization pretty-print pass on load and store
instructions.
We can generate fpr->fpr instructions for G_SITOFP and G_UITOFP. It was
previously marking the instructions as FPR but then generating GPR
instructions and introducing a copy.
This pr implements the `remove-section` option for a `DXContainer`
object in `llvm-objcopy`.
It implements a base `removeParts` to the minimal `object`
representation of a `DXContainerObject`.
This is the second step to implement
https://github.com/llvm/llvm-project/issues/150275 as a compiler actions
that invokes llvm-objcopy for functionality.
The addition of v2f32 as a legal type, supported by the B64 register
class, has caused performance regressions, broken inline assembly, and
resulted in a couple (now fixed) mis-compilations. In order to mitigate
these issues, only mark this as a legal type when there exist operations
that support it, since for targets where this is not the case it serves
no purpose. To enable further debugging, add an option to disable v2f32.
In order to allow for a target-dependent set of legal types,
ComputePTXValueVTs has been fully re-written to take advantage of
TargetLowering call-lowering APIs.
We do not gain any additional security from having PIE on the test
binaries, and this does have an impact on test times. It is about 10%
from my measurements. Not as much as previous reports
(https://reviews.llvm.org/D140880), but still significant. This is a
similar gain to finish enabling the lit internal shell, which will take
significantly more effort. I also do not think this is a meaningful
enough difference from the "default" configuration to keep it the
default, especially given the option is mostly set by distros.
Originally commited in 362b9d78b4ee9107da2b5e90b3764b0f0fa610fe and then
reverted in cb63b75e32a415c9bfc298ed7fdcd67e8d9de54c.
This re-lands a subset of the changes to
dap_server.py/DebugCommunication and addresses the python3.10
compatibility issue.
This includes less type annotations since those were the reason for the
failures on that specific version of python.
I've done additional testing on python3.8, python3.10 and python3.13 to
further validate these changes.
Static analysis flagged these classes as having implemented the
destructor but not applying rule of three. This could lead to accidental
misuse and so it makes sense to apply it.
This patch adds a new set of randomized conformance tests for
single-precision bivariate math functions.
The functions included in this set were selected based on the following
criteria:
- An implementation exists in `libc/src/math/generic` (i.e., it is not
just a wrapper around a compiler built-in).
- The corresponding LLVM CPU libm implementation is correctly rounded.
- The function is listed in Table 65 of the OpenCL C Specification
v3.0.19.
- Add PermlaneSwapOp that lowers to `rocdl.permlane16.swap` and
`rocdl.permlane32.swap`
---------
Co-authored-by: Jakub Kuderski <kubakuderski@gmail.com>
This PR upstreams the GotoSolver pass.
It works by walking the function and matching each label to a goto. If a
label is not matched to a goto, it is removed and not lowered.
This patch fixes:
third-party/unittest/googletest/include/gtest/gtest.h:1379:11:
error: comparison of integers of different signs: 'const unsigned
long' and 'const int' [-Werror,-Wsign-compare]
Replace inttoptr (add (ptrtoint %B), %O) with (getelementptr i8, %B, %o)
if all users are ICmp instruction, which in turn means only the address
value is compared. We should be able to do this, if the src pointer,
the integer type and the destination pointer types have the same
bitwidth and address space.
A common source of such (inttoptr (add (ptrtoint %B), %O)) is from
various iterations in libc++.
In practice this triggers in a number of files in Clang and various open
source projects, including cppcheck, diamond, llama and more.
Alive2 Proof with constant offset: https://alive2.llvm.org/ce/z/K_5N_B
PR: https://github.com/llvm/llvm-project/pull/153421
Rename the pass `LinalgNamedOpConversionPass` to
`SimplifyDepthwiseConvPass` to avoid conflating it with the new
morphisms we are creating between the norms.
The default behaviour in DXC is to allow all extesions the compiler
knows about. We did the same in clang: all extensions that clang knows
about. However, this causes the shader to use different extensions
because the two compilers have different sets of extensions.
To avoid using a new extension when moving from DXC to Clang, we add the
special DXC suboptions to `-fspv-extension`. If `-fspv-extension=DXC` is
used, then the available extensions will be those available in DXC.
---------
Co-authored-by: Chris B <beanz@abolishcrlf.org>
We used to vectorize these scalably but after #147026 they were split
out from RecurKind::Add into their own RecurKinds, and we didn't mark
them as supported in isLegalToVectorizeReduction.
This caused the loop vectorizer to drop the scalable VPlan because it
thinks the reductions will be scalarized.
This fixes it by just marking them as supported.
Fixes#154554
These functions turned out to have the same bug that was in wcstok()
(fixed by 4fc9801), so add the missing tests and fix the code in a way
that matches wcstok().
Also fix incorrect test expectations in existing tests.
Also update the BUILD.bazel files to actually build the strsep() test.
The C++ standard requires stddef.h to declare all of its contents in the
global namespace. We were only doing it when trying to be compatible
with Microsoft extensions. Now we expose in C++11 or later, in addition
to exposing it in Microsoft extensions mode.
Fixes#154577
... builtins. We used to access the I'th index of the output vector, but
that doesn't work since the output vector is only half the size of the
input vector.
I tracked down the document which changed the way variables are handled
in for loops for C99, it was the same document that allowed mixing code
and declarations but the editor's report made it seem like the features
came from different papers.
This is an extension we backported to C89 but it's sufficiently distinct
in the tracking page so I've added it explicitly to the backported
features documentation.
