Haojian Wu [Mon, 1 Oct 2018 12:16:38 +0000 (12:16 +0000)]
Build clang-headers when building clang tools.
Summary:
clang tools require clang headers to work on real project, e.g. when we
build clangd via `ninja clangd`, we expect the binary can run on
real-world project (without running another command `ninja clang-headers`).
Fangrui Song [Sun, 30 Sep 2018 21:41:11 +0000 (21:41 +0000)]
Use the container form llvm::sort(C, ...)
There are a few leftovers of rC343147 that are not (\w+)\.begin but in
the form of ([-[:alnum:]>.]+)\.begin or spanning two lines. Change them
to use the container form in this commit. The 12 occurrences have been
inspected manually for safety.
Eric Fiselier [Sun, 30 Sep 2018 18:05:39 +0000 (18:05 +0000)]
Fix linkage error on ProgramPoint's dump method.
Currently, ProgramPoint::dump calls the out-of-line function ProgramPoint::print. This causes
libraries which include ProgramPoint.h to become dependent on libclangAnalysis, which in turn
causes missing symbol link error when building with -DBUILD_SHARED_LIBS=ON -DLLVM_ENABLE_MODULES=ON.
The breakage was introduced in r343160.
This patch fixes the issues by moving ProgramPoint::dump's declaration out of line.
[OPENMP]Fix PR39084: Check datasharing attributes of reduction variables only.
According to OpenMP, the reduction item must be shared in parent region.
But the item can be an array section or array subscript. In this case,
we should not check for the datasharing of the base declaration.
[X86] Add the movbe instruction intrinsics from icc.
These intrinsics exist in icc. They can be found on the Intel Intrinsics Guide website.
All the backend support is in place to pattern match a load+bswap or a bswap+store pattern to the MOVBE instructions. So we just need to get the frontend to emit the correct IR. The pointer arguments in icc are declared as void so I had to jump through a packed struct to forcing a specific alignment on the load/store. Same trick we use in the unaligned vector load/store intrinsics
AST: add missing ObjC extensions to MS style name decoration
Add support for encoding type arguments for lightweight generics in
Objective-C++ mode. Additionally, add support for the `__kindof` modifier.
This should complete the coverage of the ObjC extensions that clang currently
supports under the MS style name decoration scheme.
This is implemented similar to the Objective-C lifetime qualifiers decoration:
a template specialization in the `__ObjC` namespace so that we can interoperate
with Microsoft's tools as well as ensure that we do not accidentally collide
with new features in the Microsoft implementation.
Since the `__kindof` appertains to the type and not the pointer, we apply the
template specialization to the underlying type instead of the pointer type.
Unfortunately, until D52581 is resolved, the generated name is not really
compatible with the MS tools as well as breaks interoperability with
Objective-C++ and C++.
[DRIVER][OFFLOAD] Do not invoke unbundler on unsupported file types.
clang-offload-bundler should not be invoked with the unbundling action
when the input file type does not match the action type. For example,
.so files should be unbundled during linking phase and should be linked
only with the host code.
Peter Smith [Fri, 28 Sep 2018 09:04:31 +0000 (09:04 +0000)]
[ARM] Alter test to account for change to armv6k default CPU
Review D52594 will change the default in llvm for armv6k from the
non-existent cpu arm1176jf-s to mpcore. The tests in arm-cortex-cpus.c
need to be updated to account for this change.
Richard Smith [Fri, 28 Sep 2018 01:16:43 +0000 (01:16 +0000)]
[cxx2a] P0641R2: (Some) type mismatches on defaulted functions only
render the function deleted instead of rendering the program ill-formed.
This change also adds an enabled-by-default warning for the case where
an explicitly-defaulted special member function of a non-template class
is implicitly deleted by the type checking rules. (This fires either due
to this language change or due to pre-C++20 reasons for the member being
implicitly deleted). I've tested this on a large codebase and found only
bugs (where the program means something that's clearly different from
what the programmer intended), so this is enabled by default, but we
should revisit this if there are problems with this being enabled by
default.
Erik Pilkington [Thu, 27 Sep 2018 20:36:28 +0000 (20:36 +0000)]
NFC: Fix some darwin linker warnings introduced in r338385
The darwin linker was complaining about Toolchains/RISCV.cpp and
Toolchains/Arch/RISCV.cpp had the same name. Fix is to just rename
Toolchains/RISCV.cpp to Toolchains/RISCVToolchain.cpp.
[OpenMP] Make default parallel for schedule in NVPTX target regions in SPMD mode achieve coalescing
Summary: Set default schedule for parallel for loops to schedule(static, 1) when using SPMD mode on the NVPTX device offloading toolchain to ensure coalescing.
[OpenMP] Make default distribute schedule for NVPTX target regions in SPMD mode achieve coalescing
Summary: For the OpenMP NVPTX toolchain choose a default distribute schedule that ensures coalescing on the GPU when in SPMD mode. This significantly increases the performance of offloaded target code and reduces the number of registers used on the GPU side.
Jonas Hahnfeld [Thu, 27 Sep 2018 17:27:48 +0000 (17:27 +0000)]
Fix greedy FileCheck expression in test/Driver/mips-abi.c
'ld{{.*}}"' seems to match the complete line for me which is failing
the test. Only allow an optional '.exe' for Windows systems as most
other tests do.
Another possibility would be to collapse the greedy expression with
the next check to avoid matching the full line.
Jonas Hahnfeld [Thu, 27 Sep 2018 16:12:32 +0000 (16:12 +0000)]
[OpenMP] Improve search for libomptarget-nvptx
When looking for the bclib Clang considered the default library
path first while it preferred directories in LIBRARY_PATH when
constructing the invocation of nvlink. The latter actually makes
more sense because during development it allows using a non-default
runtime library. So change the search for the bclib to start
looking in directories given by LIBRARY_PATH.
Additionally add a new option --libomptarget-nvptx-path= which
will be searched first. This will be handy for testing purposes.
Martin Probst [Thu, 27 Sep 2018 06:48:13 +0000 (06:48 +0000)]
clang-format: [JS] conditional types.
Summary:
This change adds some rudimentary support for conditional types.
Specifically it avoids breaking before `extends` and `infer` keywords,
which are subject to Automatic Semicolon Insertion, so breaking before
them creates incorrect syntax.
The actual formatting of the type expression is odd, but there is as of
yet no clear idea on how to format these.
See https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types.
Simon Atanasyan [Thu, 27 Sep 2018 05:04:50 +0000 (05:04 +0000)]
[driver][mips] Adjust target triple accordingly to provided ABI name
Explicitly selected MIPS ABI using the `-mabi` option implies
corresponding target triple. For 'O32' ABI it's a 32-bit target triple
like `mips-linux-gnu`. For 'N32' and 'N64' ABIs it's a 64-bit target
triple like `mips64-linux-gnu`. This patch adjusts target triple
accordingly these rules like we do for pseudo-target flags '-m64',
'-m32' etc already.
We don't expect useful value there unless it's "ambiguous".
However we use read it for copying and moving, so we need either init the field
add login to avoid reading invalid values. Such reads trigger ubsan errors.
[X86] For lzcnt/tzcnt intrinsics use cttz/ctlz intrinsics with zero_undef flag set to false.
Previously we used a select and the zero_undef=true intrinsic. In -O2 this pattern will get optimized to zero_undef=false. But in -O0 this optimization won't happen. This results in a compare and cmov being wrapped around a tzcnt/lzcnt instruction.
By using the zero_undef=false intrinsic directly without the select, we can improve the -O0 codegen to just an lzcnt/tzcnt instruction.
Roman Lebedev [Wed, 26 Sep 2018 13:08:44 +0000 (13:08 +0000)]
[analyzer] scan-build: if --status-bugs is passed, don't forget about the exit status of the actual build
Summary:
This has been bothering me for a while, but only now i have actually looked into this.
I'm using one CI job for static analysis - clang static analyzers as compilers + clang-tidy via cmake.
And i'd like for the build to fail if at least one of those finds issues.
If clang-tidy finds issues, it will fail the build since the warnings-as-errors is set.
If static analyzer finds anything, since --status-bugs is set, it will fail the build.
But if clang-tidy find anything, but static analyzer does not, the build succeeds :/
Fangrui Song [Wed, 26 Sep 2018 06:28:14 +0000 (06:28 +0000)]
[Frontend] Tidy up -print-decl-contexts conditions
The existing conditions are not consistent. Some have braces and define a temporary Decl while others simply call `<< *cast<XXXDecl>(I)` (mostly the NamedDecl overload of operator<<).
Just use the latter for consistency and brevity.
[analyzer] Fix a crash on casting symbolic pointers to derived classes.
Commit r340984 causes a crash when a pointer to a completely unrelated type
UnrelatedT (eg., opaque struct pattern) is being casted from base class BaseT to
derived class DerivedT, which results in an ill-formed region
Derived{SymRegion{$<UnrelatedT x>}, DerivedT}.
[analyzer] Add a testing facility for testing relationships between symbols.
Tests introduced in r329780 was disabled in r342317 because these tests
were accidentally testing dump infrastructure, when all they cared about was
how symbols relate to each other. So when dump infrastructure changed,
tests became annoying to maintain.
Add a new feature to ExprInspection: clang_analyzer_denote() and
clang_analyzer_explain(). The former adds a notation to a symbol, the latter
expresses another symbol in terms of previously denoted symbols.
It's currently a bit wonky - doesn't print parentheses and only supports
denoting atomic symbols. But it's even more readable that way.
[clang-check-codegen][cfstring] Accept either @ or % for progbits to make ppc64be bots happy.
PPC64BE bots use % instead of @ for directives like progbits. Since CFString tests also
check asm output, they fail on the following:
cfstring3.c:44:19: error: CHECK-ASM-ELF: expected string not found in input
// CHECK-ASM-ELF: .section cfstring,"aw",@progbits
<stdin>:30:2: note: possible intended match here
.section cfstring,"aw",%progbits
Updating that check with a {{[@%]}}progbits regex to make those bots happy.
Reland "[Clang][CodeGen][ObjC]: Fix CoreFoundation on ELF with `-fconstant-cfstrings`"
Relanding rL342883 with more fragmented tests to test ELF-specific
section emission separately from broad-scope CFString tests. Now this
tests the following separately
1). CoreFoundation builds and linkage for ELF while building it.
2). CFString ELF section emission outside CF in assembly output.
3). Broad scope `cfstring3.c` tests which cover all object formats at
bitcode level and assembly level (including ELF).
This fixes non-bridged CoreFoundation builds on ELF targets
that use -fconstant-cfstrings. The original changes from differential
for a similar patch to PE/COFF (https://reviews.llvm.org/D44491) did not
check for an edge case where the global could be a constant which surfaced
as an issue when building for ELF because of different linkage semantics.
This patch addresses several issues with crashes related to CF builds on ELF
as well as improves data layout by ensuring string literals that back
the actual CFConstStrings end up in .rodata in line with Mach-O.
Change itself tested with CoreFoundation on Linux x86_64 but should be valid
for BSD-like systems as well that use ELF as the native object format.
Combine the two constructor overrides into a single ArrayRef constructor
to allow easier brace initializations and simplify how the respective field
is used internally.
[analyzer] NFC: Legalize state manager factory injection.
When a checker maintains a program state trait that isn't a simple list/set/map, but is a combination of multiple lists/sets/maps (eg., a multimap - which may be implemented as a map from something to set of something), ProgramStateManager only contains the factory for the trait itself. All auxiliary lists/sets/maps need a factory to be provided by the checker, which is annoying.
So far two checkers wanted a multimap, and both decided to trick the
ProgramStateManager into keeping the auxiliary factory within itself
by pretending that it's some sort of trait they're interested in,
but then never using this trait but only using the factory.
Make this trick legal. Define a convenient macro.
One thing that becomes apparent once all pieces are put together is that
these two checkers are in fact using the same factory, because the type that
identifies it, ImmutableMap<const MemRegion *, ImmutableSet<SymbolRef>>,
is the same. This situation is different from two checkers registering similar
primitive traits.
[DRIVER][OFFLOAD] Do not invoke unbundler on unsupported file types.
clang-offload-bundler should not be invoked with the unbundling action
when the input file type does not match the action type. For example,
.so files should be unbundled during linking phase and should be linked
only with the host code.
Hans Wennborg [Tue, 25 Sep 2018 14:10:26 +0000 (14:10 +0000)]
[clang-cl] Provide separate flags for all the /O variants
This provides better help text in "clang-cl /?".
Also it cleans things up a bit: previously "/Od" could be handled either
as a separate flag aliased to "-O0", or by the main optimization flag
processing in TranslateOptArg. With this patch, all the flags get
aliased back to /O so they're handled by TranslateOptArg.
Bruno Ricci [Tue, 25 Sep 2018 13:43:25 +0000 (13:43 +0000)]
[AST] Squeeze some bits in LinkageComputer::QueryType
Replace the pair std::pair<const NamedDecl *, unsigned> where the
unsigned represents an LVComputationKind by a PointerIntPair.
This saves a pointer per entry in the map LinkageComputer::CachedLinkageInfo.
Driver: render arguments for the embedded bitcode correctly
When embedding bitcode, only a subset of the arguments should be recorded into
the bitcode compilation commandline. The frontend job is split into two jobs,
one which will generate the bitcode. Ensure that the arguments for the
compilation to bitcode is properly stripped so that the embedded arguments are
the permitted subset.
Richard Smith [Mon, 24 Sep 2018 23:17:44 +0000 (23:17 +0000)]
P0962R1: only use the member form of 'begin' and 'end' in a range-based
for loop if both members exist.
This resolves a DR whereby an errant 'begin' or 'end' member in a base
class could result in a derived class not being usable as a range with
non-member 'begin' and 'end'.
George Karpenkov [Mon, 24 Sep 2018 21:20:30 +0000 (21:20 +0000)]
[analyzer] Prevent crashes in FindLastStoreBRVisitor
This patch is a band-aid. A proper solution would be too change
trackNullOrUndefValue to only try to dereference the pointer when it is
relevant to the problem.
Revert "We allow implicit function declarations as an extension in all C dialects. Remove OpenCL special case."
Discussed on cfe-commits (Week-of-Mon-20180820), this change leads to
the generation of invalid IR for OpenCL without giving an error.
Therefore, the conclusion was to revert.
[Clang][CodeGen][ObjC]: Fix CoreFoundation on ELF with `-fconstant-cfstrings`.
[Clang][CodeGen][ObjC]: Fix non-bridged CoreFoundation builds on ELF targets
that use `-fconstant-cfstrings`. The original changes from differential
for a similar patch to PE/COFF (https://reviews.llvm.org/D44491) did not
check for an edge case where the global could be a constant which surfaced
as an issue when building for ELF because of different linkage semantics.
This patch addresses several issues with crashes related to CF builds on ELF
as well as improves data layout by ensuring string literals that back
the actual CFConstStrings end up in .rodata in line with Mach-O.
Change itself tested with CoreFoundation on Linux x86_64 but should be valid
for BSD-like systems as well that use ELF as the native object format.
projects / clang / log