David Tenty [Fri, 7 Jun 2019 15:45:25 +0000 (15:45 +0000)]
Build with _XOPEN_SOURCE defined on AIX
Summary:
It is useful to build with _XOPEN_SOURCE defined on AIX, enabling X/Open
and POSIX compatibility mode, to work around stray macros and other
bugs in the headers provided by the system and build compiler.
This patch adds the config to cmake to build with _XOPEN_SOURCE defined
on AIX with a few exceptions. Google Test internals require access to
platform specific thread info constructs on AIX so in that case we build
with _ALL_SOURCE defined instead. Libclang also uses header which needs
_ALL_SOURCE on AIX so we leave that as is as well.
We also add building on AIX with the large file API and doing CMake
header checks with X/OPEN definitions so the results are consistent with
the environment that will be present in the build.
Sjoerd Meijer [Fri, 7 Jun 2019 15:20:56 +0000 (15:20 +0000)]
[ARM] Add ACLE feature macros for MVE
If MVE is present at all, then the macro __ARM_FEATURE_MVE is defined
to a value which has bit 0 set for integer MVE, and bit 1 set for
floating-point MVE.
(Floating-point MVE implies integer MVE, so if this macro is defined
at all then it will be set to 1 or 3, never 2.)
Simon Tatham [Fri, 7 Jun 2019 12:42:54 +0000 (12:42 +0000)]
[ARM] Fix bugs introduced by the fp64/d32 rework.
Change D60691 caused some knock-on failures that weren't caught by the
existing tests. Firstly, selecting a CPU that should have had a
restricted FPU (e.g. `-mcpu=cortex-m4`, which should have 16 d-regs
and no double precision) could give the unrestricted version, because
`ARM::getFPUFeatures` returned a list of features including subtracted
ones (here `-fp64`,`-d32`), but `ARMTargetInfo::initFeatureMap` threw
away all the ones that didn't start with `+`. Secondly, the
preprocessor macros didn't reliably match the actual compilation
settings: for example, `-mfpu=softvfp` could still set `__ARM_FP` as
if hardware FP was available, because the list of features on the cc1
command line would include things like `+vfp4`,`-vfp4d16` and clang
didn't realise that one of those cancelled out the other.
I've fixed both of these issues by rewriting `ARM::getFPUFeatures` so
that it returns a list that enables every FP-related feature
compatible with the selected FPU and disables every feature not
compatible, which is more verbose but means clang doesn't have to
understand the dependency relationships between the backend features.
Meanwhile, `ARMTargetInfo::handleTargetFeatures` is testing for all
the various forms of the FP feature names, so that it won't miss cases
where it should have set `HW_FP` to feed into feature test macros.
That in turn caused an ordering problem when handling `-mcpu=foo+bar`
together with `-mfpu=something_that_turns_off_bar`. To fix that, I've
arranged that the `+bar` suffixes on the end of `-mcpu` and `-march`
cause feature names to be put into a separate vector which is
concatenated after the output of `getFPUFeatures`.
Another side effect of all this is to fix a bug where `clang -target
armv8-eabi` by itself would fail to set `__ARM_FEATURE_FMA`, even
though `armv8` (aka Arm v8-A) implies FP-Armv8 which has FMA. That was
because `HW_FP` was being set to a value including only the `FPARMV8`
bit, but that feature test macro was testing only the `VFP4FPU` bit.
Now `HW_FP` ends up with all the bits set, so it gives the right
answer.
Changes to tests included in this patch:
* `arm-target-features.c`: I had to change basically all the expected
results. (The Cortex-M4 test in there should function as a
regression test for the accidental double-precision bug.)
* `arm-mfpu.c`, `armv8.1m.main.c`: switched to using `CHECK-DAG`
everywhere so that those tests are no longer sensitive to the order
of cc1 feature options on the command line.
* `arm-acle-6.5.c`: been updated to expect the right answer to that
FMA test.
* `Preprocessor/arm-target-features.c`: added a regression test for
the `mfpu=softvfp` issue.
Sam McCall [Fri, 7 Jun 2019 09:45:17 +0000 (09:45 +0000)]
[CodeComplete] Improve overload handling for C++ qualified and ref-qualified methods.
Summary:
- when a method is not available because of the target value kind (e.g. an &&
method on a Foo& variable), then don't offer it.
- when a method is effectively shadowed by another method from the same class
with a) an identical argument list and b) superior qualifiers, then don't
offer it.
Richard Smith [Thu, 6 Jun 2019 23:24:15 +0000 (23:24 +0000)]
Convert MemberExpr creation and serialization to work the same way as
most / all other Expr subclasses.
This reinstates r362551, reverted in r362597, with a fix to a bug that
caused MemberExprs to sometimes have a null FoundDecl after a round-trip
through an AST file.
Summary:
Other macros are used to declare namespaces, and should thus be handled
similarly. This is the case for crpcut's TESTSUITE macro, or for
unittest-cpp's SUITE macro:
This patch deals with this cases by introducing a new option to specify
lists of namespace macros. Internally, it re-uses the system already in
place for foreach and statement macros, to ensure there is no impact on
performance.
[LibTooling] Add insert/remove convenience functions for creating `ASTEdit`s.
Summary: `change()` is an all purpose function; the revision adds simple shortcuts for the specific operations of inserting (before/after) or removing source.
Pengfei Wang [Thu, 6 Jun 2019 08:28:42 +0000 (08:28 +0000)]
[X86] Add ENQCMD instructions
For more details about these instructions, please refer to the latest
ISE document:
https://software.intel.com/en-us/download/intel-architecture-instruction-set-extensions-programming-reference.
Tim Northover [Wed, 5 Jun 2019 21:12:14 +0000 (21:12 +0000)]
LLVM IR: Generate new-style byval-with-Type from Clang
LLVM IR recently added a Type parameter to the byval Attribute, so that
when pointers become opaque and no longer have an element type the
information will still be present in IR.
For now the Type parameter is optional (which is why Clang didn't need
this change at the time), but it will become mandatory soon.
Erich Keane [Wed, 5 Jun 2019 14:10:39 +0000 (14:10 +0000)]
Avoid using NoThrow Exception Specifier in non-C++ Modes.
As reported in https://bugs.llvm.org/show_bug.cgi?id=42113, there are a
number of locations in Clang where it is assumed that exception
specifications are only valid in C++ mode. Since the original
justification for the NoThrow Exception Specifier Type was C++ related,
this patch just makes C mode use the attribute-based nothrow handling.
Additionally, I noticed that the handling of non-prototype functions
regressed the behavior of the nothrow attribute, in part because it is
was listed in the function type macro(which I did in the previous
patch). In reality, it should only be doing so in a conditional nature,
so this patch removes it there and puts it directly in the switch to be
handled correctly.
[Sema] Prevent binding incompatible addr space ref to temporaries
References to arbitrary address spaces can't always be bound to
temporaries. This change extends the reference binding logic to
check that the address space of a temporary can be implicitly
converted to the address space in a reference when temporary
materialization is performed.
Sjoerd Meijer [Wed, 5 Jun 2019 13:12:01 +0000 (13:12 +0000)]
[ARM] Allow "-march=foo+fp" to vary with foo
Now, when clang processes an argument of the form "-march=foo+x+y+z",
then instead of calling getArchExtFeature() for each of the extension
names "x", "y", "z" and appending the returned string to its list of
low-level subtarget features, it will call appendArchExtFeatures()
which does the appending itself.
The difference is that appendArchExtFeatures can add _more_ than one
low-level feature name to the output feature list if it has to, and
also, it gets told some information about what base architecture and
CPU the extension is going to go with, which means that "+fp" can now
mean something different for different CPUs. Namely, "+fp" now selects
whatever the _default_ FPU is for the selected CPU and/or
architecture, as defined in the ARM_ARCH or ARM_CPU_NAME macros in
ARMTargetParser.def.
On the clang side, I adjust DecodeARMFeatures to call the new
appendArchExtFeatures function in place of getArchExtFeature. This
means DecodeARMFeatures needs to be passed a CPU name and an ArchKind,
which meant changing its call sites to make those available, and also
sawing getLLVMArchSuffixForARM in half so that you can get an ArchKind
enum value out of it instead of a string.
Also, I add support here for the extension name "+fp.dp", which will
automatically look through the FPU list for something that looks just
like the default FPU except for also supporting double precision.
Michal Gorny [Wed, 5 Jun 2019 08:21:42 +0000 (08:21 +0000)]
[clang] [test] Add a (xfailing) test for PR41027
Add a test for tracking PR41027 (8.0 regression breaking assembly code
relying on __builtin_constant_p() to identify compile-time constants).
Mark it as expected to fail everywhere.
Petr Hosek [Wed, 5 Jun 2019 03:17:11 +0000 (03:17 +0000)]
[Clang] Disable new PM for tests that use optimization level -O1, -O2 and -O3
Tests that use -O1, -O2 and -O3 would often produce different results
with the new pass manager which makes these tests fail. Disable new PM
explicitly for these tests.
Initial support for vectorization using MASSV (IBM MASS vector library)
Part 2 (the Clang portion) of D59881.
This patch (first of two patches) enables the vectorizer to recognize the
IBM MASS vector library routines. This patch specifically adds support for
recognizing the -vector-library=MASSV option, and defines mappings from IEEE
standard scalar math functions to generic PowerPC MASS vector counterparts.
For instance, the generic PowerPC MASS vector entry for double-precision
cbrt function is __cbrtd2_massv.
The second patch will further lower the generic PowerPC vector entries to
PowerPC subtarget-specific entries.
For instance, the PowerPC generic entry cbrtd2_massv is lowered to
cbrtd2_P9 for Power9 subtarget.
The overall support for MASS vector library is presented as such in two patches
for ease of review.
Nico Weber [Tue, 4 Jun 2019 23:27:40 +0000 (23:27 +0000)]
msabi: Fix exponential mangling time for even more contrived inputs
This is a follow-up to r362293 which fixed exponential time needed
for mangling certain templates. This fixes the same issue if that
template pattern happens in template arguments > 10: The first
ten template arguments can use back references, and r362293 added
caching for back references. For latter arguments, we have to add
a cache for the mangling itself instead.
This patch allows current users of Value::stripPointerCasts() to force
the result of the function to have the same representation as the value
it was called on. This is useful in various cases, e.g., (non-)null
checks.
In this patch only a single call site was adjusted to fix an existing
misuse that would cause nonnull where they may be wrong. Uses in
attribute deduction and other areas, e.g., D60047, are to be expected.
Richard Smith [Tue, 4 Jun 2019 17:17:20 +0000 (17:17 +0000)]
PR42104: Support instantiations of lambdas that implicitly capture
packs.
Two changes:
* Track odr-use via FunctionParmPackExprs to properly handle dependent
odr-uses of packs in generic lambdas.
* Do not instantiate implicit captures; instead, regenerate them by
instantiating the body of the lambda. This is necessary to
distinguish between cases where only one element of a pack is
captured and cases where the entire pack is captured.
This reinstates r362358 (reverted in r362375) with a fix for an
uninitialized variable use in UpdateMarkingForLValueToRValue.
Ilya Biryukov [Tue, 4 Jun 2019 09:26:08 +0000 (09:26 +0000)]
[CodeComplete] Include more text into typed chunks of pattern completions
Summary:
To allow filtering on any of the words in the editors.
In particular, the following completions were changed:
- 'using namespace <#name#>'
Typed text before: 'using', after: 'using namespace'.
- 'else if (#<condition#>)'
Before: 'else', after: 'else if'.
- 'using typename <#qualifier#>::<#name#>'
Before: 'using', after: 'using typename'.
Alex Lorenz [Mon, 3 Jun 2019 22:59:17 +0000 (22:59 +0000)]
Add clang source minimizer that reduces source to directives
that might affect the dependency list for a compilation
This commit introduces a dependency directives source minimizer to clang
that minimizes header and source files to the minimum necessary preprocessor
directives for evaluating includes. It reduces the source down to #define, #include,
The source minimizer works by lexing the input with a custom fast lexer that recognizes
the preprocessor directives it cares about, and emitting those directives in the minimized source.
It ignores source code, comments, and normalizes whitespace. It gives up and fails if seems
any directives that it doesn't recognize as valid (e.g. #define 0).
In addition to the source minimizer this patch adds a
-print-dependency-directives-minimized-source CC1 option that allows you to invoke the minimizer
from clang directly.
Erich Keane [Mon, 3 Jun 2019 18:36:33 +0000 (18:36 +0000)]
Make NoThrow FunctionLike, make FunctionLike include references, fix
prettyprint
__declspec(nothrow) should work on function pointers as well as function
references, so this changes it to FunctionLike. Additionally,
FunctionLike needed to be modified to permit function references.
Finally, the TypePrinter didn't properly print the NoThrow exception
specifier, so make sure we get that right as well.
[OpenCL] Use long instead of long long in x86 builtins
Summary: According to C99 standard long long is at least 64 bits in
size. However, OpenCL C defines long long as 128 bit signed
integer. This prevents one to use x86 builtins when compiling OpenCL C
code for x86 targets. The patch changes long long to long for OpenCL
only.
Patch by: Alexander Batashev <alexander.batashev@intel.com>
Simon Tatham [Mon, 3 Jun 2019 11:02:53 +0000 (11:02 +0000)]
[ARM] Fix recent breakage of -mfpu=none.
The recent change D60691 introduced a bug in clang when handling
option combinations such as `-mcpu=cortex-m4 -mfpu=none`. Those
options together should select Cortex-M4 but disable all use of
hardware FP, but in fact, now hardware FP instructions can still be
generated in that mode.
The reason is because the handling of FPUVersion::NONE disables all
the same feature names it used to, of which the base one is `vfp2`.
But now there are further features below that, like `vfp2d16fp` and
(following D60694) `fpregs`, which also need to be turned off to
disable hardware FP completely.
Added a tiny test which double-checks that compiling a simple FP
function doesn't access the FP registers.
Simon Pilgrim [Mon, 3 Jun 2019 09:56:09 +0000 (09:56 +0000)]
Revert rL362358 : PR42104: Support instantiations of lambdas that implicitly capture packs.
Two changes:
* Track odr-use via FunctionParmPackExprs to properly handle dependent
odr-uses of packs in generic lambdas.
* Do not instantiate implicit captures; instead, regenerate them by
instantiating the body of the lambda. This is necessary to
distinguish between cases where only one element of a pack is
captured and cases where the entire pack is captured.
........
Fixes http://lab.llvm.org:8011/builders/llvm-clang-x86_64-expensive-checks-win buildbot failures
This patch adds a `-fdeclare-opencl-builtins` command line option to
the clang frontend. This enables clang to verify OpenCL C builtin
function declarations using a fast StringMatcher lookup, instead of
including the opencl-c.h file with the `-finclude-default-header`
option. This avoids the large parse time penalty of the header file.
This commit only adds the basic infrastructure and some of the OpenCL
builtins. It does not cover all builtins defined by the various OpenCL
specifications. As such, it is not a replacement for
`-finclude-default-header` yet.
Richard Smith [Mon, 3 Jun 2019 06:02:10 +0000 (06:02 +0000)]
PR42104: Support instantiations of lambdas that implicitly capture
packs.
Two changes:
* Track odr-use via FunctionParmPackExprs to properly handle dependent
odr-uses of packs in generic lambdas.
* Do not instantiate implicit captures; instead, regenerate them by
instantiating the body of the lambda. This is necessary to
distinguish between cases where only one element of a pack is
captured and cases where the entire pack is captured.
Brian Gesiak [Mon, 3 Jun 2019 00:47:32 +0000 (00:47 +0000)]
[coroutines][PR41909] Don't build dependent coroutine statements for generic lambda
Summary:
https://bugs.llvm.org/show_bug.cgi?id=41909 describes an issue in which
a generic lambda that takes a dependent argument `auto set` causes the
template instantiation machinery for coroutine body statements to crash
with an ICE. The issue is two-fold:
1. The paths taken by the template instantiator contain several asserts
that the coroutine promise must not have a dependent type.
2. The template instantiator unconditionally builds corotuine statements
that depend on the promise type, which cannot be dependent.
To work around the issue, prevent the template instantiator from building
dependent coroutine statements if the coroutine promise type is dependent.
Since we only expect this to occur in the case of a generic lambda, limit
the workaround behavior to just that case.
This is a utility to improve readability and generally manipulate
GraphViz dumps of the analysis graph. Such dumps are often huge and
not only hard to read, but also often hang the viewer apps with their
mere size. Such script should significantly improve debugging experience.
Don Hinton [Sun, 2 Jun 2019 15:53:43 +0000 (15:53 +0000)]
[test] Fix plugin tests
Recommit of r361790 that was temporarily reverted in r361793 due to bot breakage.
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:
Nico Weber [Sat, 1 Jun 2019 10:12:07 +0000 (10:12 +0000)]
msabi: Fix exponential mangling time for certain pathological inputs
Template back references used to be recursively recomputed, add a
memoization cache to cut down on this.
Since there are now two different types of argument maps, rename the
existing TypeBackReferences to FunArgBackReferences, and rename
mangleArgumentType() to mangleFunctionArgumentType().
Fixes PR42091, the input there now takes 50ms instead of 7s to compile.
Aaron Puchert [Fri, 31 May 2019 21:27:39 +0000 (21:27 +0000)]
Clarify when fix-it hints on warnings are appropriate
Summary:
This is not a change in the rules, it's meant as a clarification about
warnings. Since the recovery from warnings is a no-op, the fix-it hints
on warnings shouldn't change anything. Anything that doesn't just
suppress the warning and changes the meaning of the code (even if it's
for the better) should be on an additional note.
Erich Keane [Fri, 31 May 2019 15:56:27 +0000 (15:56 +0000)]
Suppress nothrow/exception spec conflict warning when ES is parsed.
The previously added warning ended up causing false positives when
nothrow was used on member functions, where the exception specification
wasn't yet parsed. So, throw() and noexcept(true) both were incorrectly
warning. There doesn't seem to be a good way to force these to be parsed
to identify which they are (and likely should not be), so suppress the warning.
For now, unevaluated/uninstantiated are left as warnings as I am not
creative enough to find a reproducer that causes a false positive for
either.
Erich Keane [Fri, 31 May 2019 14:26:19 +0000 (14:26 +0000)]
Fix for PR42089, regression from r362119
The implementation of the NoThrow ExceptionSpecificationType missed a
switch statement for forming the diagnostic when an out-of-line member
redeclaration misses the exception specification. This patch adds the
correct case statement.
Zi Xuan Wu [Fri, 31 May 2019 04:42:13 +0000 (04:42 +0000)]
re-commit r361928: [PowerPC] [Clang] Port SSE intrinsics to PowerPC
Port xmmintrin.h which include Intel SSE intrinsics implementation to PowerPC platform (using Altivec).
The new headers containing those implemenations are located into a directory named ppc_wrappers
which has higher priority when the platform is PowerPC on Linux. They are mainly developed by Steven Munroe,
with contributions from Paul Clarke, Bill Schmidt, Jinsong Ji and Zixuan Wu.
Patched by: Qiu Chaofan <qiucf@cn.ibm.com>
Reviewed By: Jinsong Ji
Petr Hosek [Fri, 31 May 2019 01:34:51 +0000 (01:34 +0000)]
[CMake] Provide an option to use relative paths in debug info
CMake always uses absolute file paths in the generated compiler
invocation which results in absolute file paths being embedded in debug
info. This is undesirable when building a toolchain e.g. on bots as the
debug info may embed the bot source checkout path which is meaningless
anywhere else.
This change introduces the LLVM_USE_RELATIVE_PATHS_IN_DEBUG_INFO which uses
-fdebug-prefix-map (where supported) options to rewrite paths embedded
into debug info with relative ones. Additionally, LLVM_SOURCE_PREFIX can
be used to override the path to source directory with a different one.
Richard Smith [Fri, 31 May 2019 01:25:16 +0000 (01:25 +0000)]
PR39728: When completing a class, complete the destructor first.
We need to know whether the destructor is trivial in order to tell
whether other parts of the class are valid (in particular, this affects
whether the type is a literal type, which affects whether defaulted
special members can be declared constexpr or are implicitly constexpr).
John McCall [Fri, 31 May 2019 01:21:36 +0000 (01:21 +0000)]
Fix the predefined exponent limit macros for the 16-bit IEEE format.
The magnitude range of normalized _Float16 is 2^-14 (~6e-5) to
(2-2^-10)*2^15 (65504). You might think, then, that the code is
correct to defne FLT16_MIN_EXP and FLT16_MAX_EXP to be -14 and 15
respectively. However, for some reason the C specification actually
specifies a bias for these macros:
C11 5.2.4.2.2:
- minimum negative integer such that FLT_RADIX raised to one less than
that power is a normalized floating-point number, e_min:
FLT_MIN_EXP
DBL_MIN_EXP
LDBL_MIN_EXP
- maximum integer such that FLT_RADIX raised to one less than that
power is a representable finite floating-point number, e_max:
FLT_MAX_EXP
DBL_MAX_EXP
LDBL_MAX_EXP
FLT16_MIN_EXP and FLT16_MAX_EXP should clearly be biased the same way,
and other compilers do in fact do so, as do our OpenCL headers for `half`.
Richard Smith [Fri, 31 May 2019 01:17:04 +0000 (01:17 +0000)]
Defer building 'this' captures until we have left the capturing region
and returned to the context in which 'this' should be captured.
This means we now always mark 'this' referenced from the context in
which it's actually referenced, rather than potentially from some
context nested within that.
Richard Smith [Fri, 31 May 2019 00:45:10 +0000 (00:45 +0000)]
Defer capture initialization for captured regions until after we've left
the captured region scope.
This removes a case where we would build expressions (and mark
declarations odr-used) in the wrong scope.
Remove the now-unused 'capture initializer' field on sema::Capture
(except for 'this' captures, which still need to be cleaned up).
No functionality change intended (except that we now very slightly more
precisely determine whether we need to use a capture or not when another
captured region encloses an OpenMP captured region).
Richard Smith [Thu, 30 May 2019 23:21:14 +0000 (23:21 +0000)]
Refactor OpenMP stack management.
Instead of duplicating access to the directive stack throughout
SemaOpenMP.cpp, consolidate it to a few methods and call those
everywhere else. In passing, simplify adjacent code where possible.
projects / clang / log