The forwarding header is left in place because of its use in
`polly/lib/External/isl/interface/extract_interface.cc`, but I have
added a GCC warning about the fact it is deprecated, because it is used
in `isl` from where it is included by Polly.
This reverts commit e1354763b6e641e45cc2371270883fcd26edf705.
There is a build failure in m68k-linux testing bot
(https://lab.llvm.org/buildbot/#/builders/192/builds/267), which is
weird. Revert this for now and look at the reasons.
Required in https://reviews.llvm.org/D137534.
The build systems needs the information to know that "header X changed,
scanning may have changed, so please rerun scanning". Although it is
possible to get the information by running clang-scan-deps for the
second time with make format, it is not user friendly clearly.
Reviewed By: jansvoboda11
Differential Revision: https://reviews.llvm.org/D139168
In a private chat with @ben.boeckel , we get in consensus it would be
great for cmake if the invocation of clang-scan-deps can get rid of
compilation database. Due to the compilation database can't do very well
for the files which are not existed yet. @ben.boeckel may have more
context to add here.
This patch should be innocent for others usages.
Reviewed By: jansvoboda11
Differential Revision: https://reviews.llvm.org/D137534
This adds a release note that was accidentally dropped, and moves the
symbol from LLVM 16 to LLVM 17 in the module map.
Amends 0a51bc731bcc2c27e4fe97957a83642d93d989be
When reseting modular options, propagate the values from certain options
that have ImpliedBy relations instead of setting to the default. Also,
verify in clang-scan-deps that the command line produced round trips
exactly.
Ideally we would automatically derive the set of options that need this
kind of propagation, but for now there aren't very many impacted.
rdar://105148590
Differential Revision: https://reviews.llvm.org/D143446
Summary:
Offloading is not supported on 32-bit applications. We already disable
this for 32-bit cross-compiling but we also need to disable it for
32-bit native machines as well.
The interfaces for designators (i.e. C99 designated initializers) was
done in two slightly different ways. This was rather wasteful as the
differences could be combined into one.
Reviewed By: rsmith
Differential Revision: https://reviews.llvm.org/D140584
D140176 introduced new `FullDeps` API that's not thread-safe, breaking the class invariant. This was causing race condition when `clang-scan-deps` was run with multiple threads.
Reviewed By: steven_wu, akyrtzi
Differential Revision: https://reviews.llvm.org/D143428
Close https://github.com/llvm/llvm-project/issues/60544.
The root cause for the issue is that when we compile a module unit, the
file manager (and proprocessor and source manager) are owned by AST
instead of the compilaton instance. So the file manager may be invalid
when we want to create a time-report file for -ftime-trace when we are
compiling a module unit.
This patch tries to recreate the file manager for -ftime-trace if we
find the file manager is not valid.
For users of the C++ API, the return type of `getFullDependencies` doesn't make sense when asking for dependencies of a module. In the returned `FullDependenciesResult` instance, only `DiscoveredModules` is useful (the graph of modular dependecies). The `FullDeps` member is trying to describe a translation unit it was never given. Its command line also refers to a file in the in-memory VFS we create in the scanner, leaking the implementation detail.
This patch splits the API and improves layering and naming of the return types.
Depends on D140175.
Reviewed By: artemcm
Differential Revision: https://reviews.llvm.org/D140176
This API is no longer necessary, so let's remove it to simplify the internal APIs.
Reviewed By: benlangmuir, artemcm
Differential Revision: https://reviews.llvm.org/D140175
Summary:
This code passed the value of `-rpath` directly to the clang invocation.
If we're using the linker then it'll be fine. However, if the linker is
`gcc` as is the case when doing `-fopenmp-targets=x86_64` then this will
cause problems. This patch adds the `-Wl,-rpath,` to feed it to the
linker correctly.
There were intemittent errors in the linker wrapper when using the
sanitizers in parallel. First, this is because the `TempFiles` global
was not guarded when creating a new file. Second, even though the `Args`
list is passed as const, the internal state is mutable when adding a
string. So that needs to be guarded too.
Fixes https://github.com/llvm/llvm-project/issues/60437
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D142985
Summary:
We offer almost no support for offloading on 32-bit systems. This causes
some problems when cross-compiling for 32-bit machines as it will find
the CUDA from the host that is incompatible. Instead we force these to
always use the dynamically loaded version, which should always compile.
This patch introduces a new type __externref_t that denotes a WebAssembly opaque
reference type. It also implements builtin __builtin_wasm_ref_null_extern(),
that returns a null value of __externref_t. This lays the ground work
for further builtins and reference types.
Differential Revision: https://reviews.llvm.org/D122215
Summary:
This patch adds support for `--[no-]whole-archive` to the linker
wrapper. This allows us to bypass the symbol resolution logic that is
normally used for static archives. For multi-architecture binaries this
also allows us to build for every single member.
The new method is a wrapper of `CXXConstructorDecl::isExplicit` and
`CXXConversionDecl::isExplicit`, allowing the user to recognize whether
the declaration pointed to by a cursor was marked with the explicit
specifier.
An export for the function, together with its documentation, was added
to "clang/include/clang-c/Index.h" with an implementation provided in
"clang/tools/libclang/CIndex.cpp".
The implementation is based on similar `clang_CXXMethod`
implementations, returning a falsy unsigned value when the cursor is not
a declaration, is not a declaration for a constructor or conversion
function or is not a relevant declaration that was marked with the
`explicit` specifier.
The new symbol was added to "clang/tools/libclang/libclang.map" to be
exported, under the LLVM16 tag.
"clang/tools/c-index-test/c-index-test.c" was modified to print a
specific tag, "(explicit)", for cursors that are recognized by
`clang_CXXMethod_isExplicit`.
Two new regression files, "explicit-constructor.cpp" and
"explicit-conversion-function.cpp", were added to "clang/test/Index", to
ensure that the behavior of the new function is correct for constructors
and conversion functions, respectively.
The "get-cursor.cpp", "index-file.cpp" and
"recursive-cxx-member-calls.cpp" regression files in "clang/test/Index"
were updated as they were affected by the new "(explicit)" tag.
A binding for the new function was added to libclang's python's
bindings, in "clang/bindings/python/clang/cindex.py", as the
"is_explicit_method" method under `Cursor`.
An accompanying test was added to
"clang/bindings/python/tests/cindex/test_cursor.py", mimicking the
regression tests for the C side.
The current release note for Clang, "clang/docs/ReleaseNotes.rst" was
modified to report the new addition under the "libclang" section.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D140756
Summary:
Clang doesn't warn on `-B` options passed to it. This one is not
forwarded to the linker which results in some tests failing when
offloading to x86_64 with the `bfd` linker.
The `OpenMPOpt` pass is pivotal to the performance of many OpenMP
offloading programs. When we perform non-LTO builds with OpenMP we used
to link the OpenMP deviceRTL individually for each TU. This lead to us
getting an additional attributor run on the combined runtime and user
code. When we used LTO we lost a run and suffered a large performance
degradation. This patch simply adds in the extra `OpenMPOpt` pass that
we miss into the LTO pipeline. This patch fixes the performance
regression shown in applications that used OpenMP offloading in LTO
mode.
Previously, this wasn't legal to do as we could emit new runtime calls
into the module. That was fixed by D142646.
Depends on D142646
Fixes https://github.com/llvm/llvm-project/issues/60300
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D142650
Since D137724 and the LLVM 17 release we have updated to CMake version
3.20. This means we can use `find_package(CUDAToolkit)` instead as the
legacy module is deprecated.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D142570
Currently, we pull in every single static archive member as long as we
have an offloading architecture that requires it. This goes against the
standard sematnics of static libraries that only pull in symbols that
define currently undefined symbols. In order to support this we roll
some custom symbol resolution logic to check if a static library is
needed. Because of offloading semantics, this requires an extra check
for externally visibile symbols. E.g. if a static member defines a
kernel we should import it.
The main benefit to this is that we can now link against the
`libomptarget.devicertl.a` library unconditionally. This removes the
requirement for users to specify LTO on the link command. This will also
allow us to stop using the `amdgcn` bitcode versions of the libraries.
```
clang foo.c -fopenmp --offload-arch=gfx1030 -foffload-lto -c
clang foo.o -fopenmp --offload-arch=gfx1030 -foffload-lto
```
Reviewed By: tra
Differential Revision: https://reviews.llvm.org/D142484
Add free functions llvm::CodeGenOpt::{getLevel,getID,parseLevel} to
provide common implementations for functionality that has been
duplicated in many places across the codebase.
Differential Revision: https://reviews.llvm.org/D141968
The new method is a wrapper of `CXXConstructorDecl::isExplicit` and
`CXXConversionDecl::isExplicit`, allowing the user to recognize whether
the declaration pointed to by a cursor was marked with the explicit
specifier.
An export for the function, together with its documentation, was added
to "clang/include/clang-c/Index.h" with an implementation provided in
"clang/tools/libclang/CIndex.cpp".
The implementation is based on similar `clang_CXXMethod`
implementations, returning a falsy unsigned value when the cursor is not
a declaration, is not a declaration for a constructor or conversion
function or is not a relevant declaration that was marked with the
`explicit` specifier.
The new symbol was added to "clang/tools/libclang/libclang.map" to be
exported, under the LLVM16 tag.
"clang/tools/c-index-test/c-index-test.c" was modified to print a
specific tag, "(explicit)", for cursors that are recognized by
`clang_CXXMethod_isExplicit`.
Two new regression files, "explicit-constructor.cpp" and
"explicit-conversion-function.cpp", were added to "clang/test/Index", to
ensure that the behavior of the new function is correct for constructors
and conversion functions, respectively.
The "get-cursor.cpp", "index-file.cpp" and
"recursive-cxx-member-calls.cpp" regression files in "clang/test/Index"
were updated as they were affected by the new "(explicit)" tag.
A binding for the new function was added to libclang's python's
bindings, in "clang/bindings/python/clang/cindex.py", as the
"is_explicit_method" method under `Cursor`.
An accompanying test was added to
"clang/bindings/python/tests/cindex/test_cursor.py", mimicking the
regression tests for the C side.
The current release note for Clang, "clang/docs/ReleaseNotes.rst" was
modified to report the new addition under the "libclang" section.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D140756
Dynamic memory allows users to allocate fast shared memory when a kernel
is launched. We support a single size for all kernels via the
`LIBOMPTARGET_SHARED_MEMORY_SIZE` environment variable but now we can
control it per kernel invocation, hence allow computed values.
Note: Only the nextgen plugins will allocate memory based on the clause,
the old plugins will silently miscompile.
Differential Revision: https://reviews.llvm.org/D141233
This reverts commit c5abe893120b115907376359a5809229a9f9608a.
This reverts commit a033dbbe5c43247b60869b008e67ed86ed230eaa.
This broke the build with -DLLVM_LINK_LLVM_DYLIB=ON. Reverting while I
investigate.
Right now in the linker wrapper we manually invoke a lot of the
toolchain programs. This reproduces a lot of logic that is already
handled in clang. Since D140158 we can now target all supported
toolchains directly via cross-compilation.
This patch changes the linker wrapper to consolidate all the alternate
linking and assembler steps into a generic call to `clang` and let clang
handle the argument handling. This heavily simplifies the interface.
Reviewed By: tra, JonChesterfield
Differential Revision: https://reviews.llvm.org/D142133
Every Clang instance uses an internal FileSystemStatCache to avoid
stating the same content multiple times. However, different instances
of Clang will contend for filesystem access for their initial stats
during HeaderSearch or module validation.
On some workloads, the time spent in the kernel in these concurrent
stat calls has been measured to be over 20% of the overall compilation
time. This is extremly wassteful when most of the stat calls target
mostly immutable content like a SDK.
This commit introduces a new tool `clang-stat-cache` able to generate
an OnDiskHashmap containing the stat data for a given filesystem
hierarchy.
The driver part of this has been modeled after -ivfsoverlay given
the similarities with what it influences. It introduces a new
-ivfsstatcache driver option to instruct Clang to use a stat cache
generated by `clang-stat-cache`. These stat caches are inserted at
the bottom of the VFS stack (right above the real filesystem).
Differential Revision: https://reviews.llvm.org/D136651
Summary:
These tools are built unconditionally now. However, there seemed to be
problems where the headers would be found during cross compilation, but
no libraries present. To combat this we should elect to make the CMake
indicate whether or not we should use the dynamic library method or link
it directly rather than using `__has_include`.
Much like the changes in D141859, this patch allows the `nvptx-arch`
tool to be built and provided with every distrubition of LLVM / Clang.
This will make it more reliable for our toolchains to depend on. The
changes here configure a version that dynamically loads CUDA if it was
not found at build time.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D141861
We use the `amdgpu-arch` tool to query the installed GPUs at runtime.
One problem is that this tool is currently not build if the person
building the LLVM binary does not have the HSA runtime on their system.
This means that if someone built and distrubted an installation of LLVM
without HSA, then the user will not be able to use it even if they have
it on their system.
This patch makes us build this tool unconditionally and adds extra logic
to dynamically load HSA if it's present.
Reviewed By: JonChesterfield
Differential Revision: https://reviews.llvm.org/D141859
