Yaxun Liu [Tue, 28 May 2019 21:18:59 +0000 (21:18 +0000)]
[CUDA][HIP] Emit dependent libs for host only
Recently D60274 was introduced to allow lld to handle dependent libs. However current
usage of dependent libs (e.g. pragma comment(lib, *) in windows header files) are intended
for host only. Emitting the metadata in device IR causes link error in device path.
Until there is a way to different it dependent libs for device or host, metadata for dependent
libs should be emitted for host only. This patch enforces that.
As for other floating-point rounding builtins that can be optimized
when build with -fno-math-errno, this patch adds support for lrint
and llrint. It currently only optimize for AArch64 backend.
Joel E. Denny [Tue, 28 May 2019 19:27:19 +0000 (19:27 +0000)]
[OpenMP] Set pragma start loc to `#pragma` loc
This patch adjusts `PragmaOpenMPHandler` to set the location of
`tok::annot_pragma_openmp` to the `#pragma` location instead of the
`omp` location so that the former becomes the start location of the
OpenMP AST node. This can be useful when, for example, rewriting a
directive using Clang's Rewrite facility. Most of this patch updates
tests for changes to locations in diagnostics and `-ast-dump` output.
Update the `cl` emulation to support the `/Zc:char8_t[-]?` options as per the
MSVC 2019.1 toolset. These are aliases for `-fchar8_t` and `-fno-char8_t`.
Simon Tatham [Tue, 28 May 2019 16:13:20 +0000 (16:13 +0000)]
[ARM] Replace fp-only-sp and d16 with fp64 and d32.
Those two subtarget features were awkward because their semantics are
reversed: each one indicates the _lack_ of support for something in
the architecture, rather than the presence. As a consequence, you
don't get the behavior you want if you combine two sets of feature
bits.
Each SubtargetFeature for an FP architecture version now comes in four
versions, one for each combination of those options. So you can still
say (for example) '+vfp2' in a feature string and it will mean what
it's always meant, but there's a new string '+vfp2d16sp' meaning the
version without those extra options.
A lot of this change is just mechanically replacing positive checks
for the old features with negative checks for the new ones. But one
more interesting change is that I've rearranged getFPUFeatures() so
that the main FPU feature is appended to the output list *before*
rather than after the features derived from the Restriction field, so
that -fp64 and -d32 can override defaults added by the main feature.
Adam Balogh [Tue, 28 May 2019 13:07:09 +0000 (13:07 +0000)]
[Analyzer] Replace `CXXSelfAssignmentBRVisitor` with `NoteTags`
The `cplusplus.SelfAssignment` checker has a visitor that is added
to every `BugReport` to mark the to branch of the self assignment
operator with e.g. `rhs == *this` and `rhs != *this`. With the new
`NoteTag` feature this visitor is not needed anymore. Instead the
checker itself marks the two branches using the `NoteTag`s.
Hans Wennborg [Tue, 28 May 2019 12:19:38 +0000 (12:19 +0000)]
Re-commit r357452 (take 2): "SimplifyCFG SinkCommonCodeFromPredecessors: Also sink function calls without used results (PR41259)"
This was reverted in r360086 as it was supected of causing mysterious test
failures internally. However, it was never concluded that this patch was the
root cause.
> The code was previously checking that candidates for sinking had exactly
> one use or were a store instruction (which can't have uses). This meant
> we could sink call instructions only if they had a use.
>
> That limitation seemed a bit arbitrary, so this patch changes it to
> "instruction has zero or one use" which seems more natural and removes
> the need to special-case stores.
>
> Differential revision: https://reviews.llvm.org/D59936
Don Hinton [Tue, 28 May 2019 06:26:58 +0000 (06:26 +0000)]
[test] Fix plugin tests
Summary:
The following changes were required to fix these tests:
1) Change LLVM_ENABLE_PLUGINS to an option and move it to
llvm/CMakeLists.txt with an appropriate default -- which matches
the original default behavior.
2) Move the plugins directory from clang/test/Analysis
clang/lib/Analysis. It's not enough to add an exclude to the
lit.local.cfg file because add_lit_testsuites recurses the tree and
automatically adds the appropriate `check-` targets, which don't
make sense for the plugins because they aren't tests and don't
have `RUN` statements.
Here's a list of the `clang-check-anlysis*` targets with this
change:
Petr Hosek [Mon, 27 May 2019 23:23:50 +0000 (23:23 +0000)]
[Driver] Change layout of per-target runtimes to resemble multiarch
This is a follow up to r361432, changing the layout of per-target
runtimes to more closely resemble multiarch. While before, we used
the following layout:
Volodymyr Sapsai [Mon, 27 May 2019 19:15:30 +0000 (19:15 +0000)]
[Preprocessor] Fix crash emitting note with framework location for "file not found" error.
A filename can be remapped with a header map to point to a framework
header and we can find the corresponding framework without the header.
But if the original filename doesn't have a remapped framework name,
we'll fail to find its location and will dereference a null pointer
during diagnostics emission.
Fix by tracking remappings better and emit the note only if a framework
is found before any of the remappings.
Yaxun Liu [Mon, 27 May 2019 11:19:07 +0000 (11:19 +0000)]
[OpenCL] Fix file-scope const sampler variable for 2.0
OpenCL spec v2.0 s6.13.14:
Samplers can also be declared as global constants in the program
source using the following syntax.
const sampler_t <sampler name> = <value>
This works fine for OpenCL 1.2 but fails for 2.0, because clang duduces
address space of file-scope const sampler variable to be in global address
space whereas spec v2.0 s6.9.b forbids file-scope sampler variable to be
in global address space.
The fix is not to deduce address space for file-scope sampler variables.
Balazs Keri [Mon, 27 May 2019 09:36:00 +0000 (09:36 +0000)]
[ASTImporter] Added visibility context check for CXXRecordDecl.
Summary:
ASTImporter makes now difference between classes with same name in different
translation units if these are not visible outside. These classes are not linked
into one decl chain.
Petr Hosek [Sun, 26 May 2019 03:39:07 +0000 (03:39 +0000)]
[Driver] Update handling of c++ and runtime directories
This is a follow up to r361432 and r361504 which addresses issues
introduced by those changes. Specifically, it avoids duplicating
file and runtime paths in case when the effective triple is the
same as the cannonical one. Furthermore, it fixes the broken multilib
setup in the Fuchsia driver and deduplicates some of the code.
Artem Dergachev [Fri, 24 May 2019 23:37:11 +0000 (23:37 +0000)]
[analyzer] Add a prunable note for skipping vbase inits in subclasses.
When initialization of virtual base classes is skipped, we now tell the user
about it, because this aspect of C++ isn't very well-known.
The implementation is based on the new "note tags" feature (r358781).
In order to make use of it, allow note tags to produce prunable notes,
and move the note tag factory to CoreEngine.
Artem Dergachev [Fri, 24 May 2019 23:37:08 +0000 (23:37 +0000)]
[CFG] Add branch to skip vbase inits when they're handled by superclass.
This patch adds the run-time CFG branch that would skip initialization of
virtual base classes depending on whether the constructor is called from a
superclass constructor or not. Previously the Static Analyzer was already
skipping virtual base-class initializers in such constructors, but it wasn't
skipping their arguments and their potential side effects, which was causing
pr41300 (and was generally incorrect). The previous skipping behavior is
now replaced with a hard assertion that we're not even getting there due
to how our CFG works.
The new CFG element is under a CFG build option so that not to break other
consumers of the CFG by this change. Static Analyzer support for this change
is implemented.
Richard Smith [Fri, 24 May 2019 21:08:12 +0000 (21:08 +0000)]
Default arguments are potentially constant evaluated.
We need to eagerly instantiate constexpr functions used in them even if
the default argument is never actually used, because we might evaluate
portions of it when performing semantic checks.
Shafik Yaghmour [Fri, 24 May 2019 16:53:44 +0000 (16:53 +0000)]
[ASTImporter] Call to HandleNameConflict in VisitRecordDecl mistakeningly using Name instead of SearchName
Summary:
https://reviews.llvm.org/D51633 added error handling to the ASTNodeImporter::VisitRecordDecl for the conflicting names case. This could lead to erroneous return of an error in that case since we should have been using SearchName. Name may be empty in the case where we find the name via D->getTypedefNameForAnonDecl()->getDeclName().
This fix is very similar to https://reviews.llvm.org/D59665
[LibTooling] Add Explanation parameter to `makeRule`.
Summary:
Conceptually, a single-case RewriteRule has a matcher, edit(s) and an (optional)
explanation. `makeRule` previously only took the matcher and edit(s). This
change adds (optional) support for the explanation.
Ilya Biryukov [Fri, 24 May 2019 10:18:39 +0000 (10:18 +0000)]
[CodeComplete] Filter override completions by function name
Summary:
We put only part of the signature starting with a function name into "typed text"
chunks now, previously the whole signature was "typed text".
This leads to meaningful fuzzy match scores, giving better signals to
compare with other completion items.
Ideally, we would not display the result type to the user, but that requires adding
a new kind of completion chunk.
Igor Kudrin [Fri, 24 May 2019 04:46:22 +0000 (04:46 +0000)]
Do not resolve directory junctions for `-fdiagnostics-absolute-paths` on Windows.
If the source file path contains directory junctions, and we resolve them when
printing diagnostic messages, these paths look independent for an IDE.
For example, both Visual Studio and Visual Studio Code open separate editors
for such paths, which is not only inconvenient but might even result in losing
changes made in one of them.
Artem Dergachev [Fri, 24 May 2019 01:34:22 +0000 (01:34 +0000)]
[CFG] NFC: Remove implicit conversion from CFGTerminator to Stmt *.
Turn it into a variant class instead. This conversion does indeed save some code
but there's a plan to add support for more kinds of terminators that aren't
necessarily based on statements, and with those in mind it becomes more and more
confusing to have CFGTerminators implicitly convertible to a Stmt *.
Alexey Bataev [Thu, 23 May 2019 22:30:43 +0000 (22:30 +0000)]
[OPENMP]Do not crash for const firstprivates.
If the variable is a firstprivate variable and it was not emitted beause
this a constant variable with the constant initializer, we can use the
initial value instead of the variable itself. It also fixes the problem
with the compiler crash in this case.
Kristof Umann [Thu, 23 May 2019 21:46:51 +0000 (21:46 +0000)]
[analyzer] List checkers in 3 categories: released, alpha, developer
Previously, the only way to display the list of available checkers was
to invoke the analyzer with -analyzer-checker-help frontend flag. This
however wasn't really great from a maintainer standpoint: users came
across checkers meant strictly for development purposes that weren't to
be tinkered with, or those that were still in development. This patch
creates a clearer division in between these categories.
From now on, we'll have 3 flags to display the list checkers. These
lists are mutually exclusive and can be used in any combination (for
example to display both stable and alpha checkers).
-analyzer-checker-help: Displays the list for stable, production ready
checkers.
-analyzer-checker-help-alpha: Displays the list for in development
checkers. Enabling is discouraged
for non-development purposes.
-analyzer-checker-help-developer: Modeling and debug checkers. Modeling
checkers shouldn't be enabled/disabled
by hand, and debug checkers shouldn't
be touched by users.
Craig Topper [Thu, 23 May 2019 21:34:36 +0000 (21:34 +0000)]
[X86] Split multi-line chained assignments into single lines to avoid making clang-format create triangle shaped indentation. Simplify one if statement to remove a bunch of string matches. NFCI
We had an if statement that checked over every avx512* feature to see if it should enabled avx512f. Since they are all prefixed with avx512 just check for that instead.
Reid Kleckner [Thu, 23 May 2019 18:35:43 +0000 (18:35 +0000)]
[Driver] Move the "-o OUT -x TYPE SRC.c" flags to the end of -cc1
New -cc1 arguments, such as -faddrsig, have started appearing after the
input name. I personally find it convenient for the input to be the last
argument to the compile command line, since I often need to edit it when
running crash reproduction scripts.
Nico Weber [Thu, 23 May 2019 17:58:33 +0000 (17:58 +0000)]
lld-link, clang: Treat non-existent input files as possible spellos for option flags
OptTable treats arguments starting with / that aren't a known option
as filenames. This means lld-link's and clang-cl's typo correction for
unknown flags didn't do spell checking for misspelled options that start
with /.
I first tried changing OptTable, but that got pretty messy, see PR41787
comments 2 and 3.
Instead, let lld-link's and clang's (including clang-cl's) "file not
found" diagnostic check if a non-existent file looks like it could be a
mis-spelled option, and if so add a "did you mean" suggestion to the
"file not found" diagnostic.
While here, make formatting of a few diagnostics a bit more
self-consistent.
[LibTooling] Fix dangling references in RangeSelector.
Summary:
RangeSelector had a number of cases of capturing a StringRef in a lambda, which
lead to dangling references. This change converts all uses in the API of
`StringRef` to `std::string` to avoid this problem. `std::string` in the API is
a reasonable choice, because the combinators are always storing the string
beyond the life of the combinator construction.
Erich Keane [Thu, 23 May 2019 16:05:21 +0000 (16:05 +0000)]
Ensure builtins use the target default Calling Convention
r355317 changed builtins/allocation functions to use the default calling
convention in order to support platforms that use non-cdecl calling
conventions by default.
However the default calling convention is overridable on Windows 32 bit
implementations with some of the /G options. The intent is to permit the
user to set the calling convention of normal functions, however it
should NOT apply to builtins and C++ allocation functions.
This patch ensures that the builtin/allocation functions always use the
Target specific Calling Convention, ignoring the user overridden version
of said default.
Paul Robinson [Thu, 23 May 2019 15:07:46 +0000 (15:07 +0000)]
Work around a Visual C++ bug.
Using a static function as a template parameter gets a bogus compile-time
error with Visual Studio 2017, prior to version 15.8. Our current
minimum-version requirement is a particular update to VS2015, and we
assume all Visual Studio 2017 versions are usable. This patch makes the
code buildable with older versions of VS2017, and can be reverted after
we upgrade the minimum version sometime in the future.
Description of the Microsoft bug:
https://developercommunity.visualstudio.com/content/problem/25334/error-code-c2971-when-specifying-a-function-as-the.html
Petr Hosek [Thu, 23 May 2019 02:35:12 +0000 (02:35 +0000)]
[CMake][Fuchsia] Use cannonical triples for runtimes
This ensures that whether the user uses short or cannonical version
of the triple, Clang will still find the runtimes under the cannonical
triple name.
Petr Hosek [Wed, 22 May 2019 21:08:33 +0000 (21:08 +0000)]
[runtimes] Move libunwind, libc++abi and libc++ to lib/$target/c++ and include/c++
This change is a consequence of the discussion in "RFC: Place libs in
Clang-dedicated directories", specifically the suggestion that
libunwind, libc++abi and libc++ shouldn't be using Clang resource
directory. Tools like clangd make this assumption, but this is
currently not true for the LLVM_ENABLE_PER_TARGET_RUNTIME_DIR build.
This change addresses that by moving the output of these libraries to
lib/$target/c++ and include/c++ directories, leaving resource directory
only for compiler-rt runtimes and Clang builtin headers.
James Y Knight [Wed, 22 May 2019 20:39:51 +0000 (20:39 +0000)]
Add back --sysroot support for darwin header search.
Before e97b5f5cf37e ([clang][Darwin] Refactor header search path logic
into the driver), both --sysroot and -isysroot worked to specify where
to look for system and C++ headers on Darwin. However, that change
caused clang to start ignoring --sysroot.
This fixes the regression, and adds tests.
(I also note that on all other platforms, clang seems to almost
completely ignore -isysroot, but that's another issue...)
David Blaikie [Wed, 22 May 2019 20:36:06 +0000 (20:36 +0000)]
Modules: Code generation of enum constants for merged enum definitions
Found in a bootstrap of LLVM with implicit modules, resulting in a
deadlock of some Orc unit tests with libstdc++ 8.1. An enum was used as
part of the implementation of std::recursive_mutex and this bug resulted
in the constant initialization of zero instead of the desired non-zero
value. => Badness.
Richard Smith tells me neither of these fields are necessarily canonical
& so using declaresSamEntity is the right solution here (rather than
changing both of these Fields to be canonical by construction/from their
source)
Add support for creating a `StencilPart` from any `RangeSelector`, which
broadens the scope of `Stencil`.
Correspondingly, deprecate Stencil's specialized combinators `node` and `sNode`
in favor of using the new `selection` combinator directly (with the appropriate
range selector).
[LibTooling] Update Transformer to use RangeSelector instead of NodePart enum.
Transformer provides an enum to indicate the range of source text to be edited.
That support is now redundant with the new (and more general) RangeSelector
library, so we remove the custom enum support in favor of supporting any
RangeSelector.
John Brawn [Wed, 22 May 2019 11:42:54 +0000 (11:42 +0000)]
[ARM][AArch64] Fix incorrect handling of alignment in va_arg code generation
Overaligned and underaligned types (i.e. types where the alignment has been
increased or decreased using the aligned and packed attributes) weren't being
correctly handled in all cases, as the unadjusted alignment should be used.
This patch also adjusts getTypeUnadjustedAlign to correctly handle typedefs of
non-aggregate types, which it appears it never had to handle before.
projects / clang / log