Instead of searching the global pool multiple times: in
LookupFactoryMethodInGlobalPool, LookupInstanceMethodInGlobalPool,
CollectMultipleMethodsInGlobalPool, and AreMultipleMethodsInGlobalPool,
we now collect the method candidates in CollectMultipleMethodsInGlobalPool
only, and other functions will use the collected method set.
This commit adds parameter "Methods" to AreMultipleMethodsInGlobalPool,
and SelectBestMethod. It also changes the implementation of
CollectMultipleMethodsInGlobalPool to collect the desired kind first, if none is
found, to collect the other kind. This avoids the need to call both
LookupFactoryMethodInGlobalPool and LookupInstanceMethodInGlobalPool.
This is a mechanical move of CodeGenOptions from libFrontend to libBasic. This
fixes the layering violation introduced earlier by threading CodeGenOptions into
TargetInfo. It should also fix the modules based self-hosting builds. NFC.
[OPENMP 4.0] Parsing/sema analysis for 'simdlen' clause in 'declare simd'
construct.
OpenMP 4.0 defines '#pragma omp declare simd' construct that may have
associated 'simdlen' clause with constant positive expression as an
argument:
simdlen(<const_expr>)
Patch adds parsin and semantic analysis for simdlen clause.
This threads CodeGenOptions into the TargetInfo hierarchy. This is motivated by
ARM which can change some target information based on the EABI selected
(-meabi). Similar options exist for other platforms (e.g. MIPS) and thus is
generally useful. NFC.
[CrashReproducer] Setup 'case-sensitive' in YAML files.
The crash reproducer was not setting up case sensitivity in the
VFS yaml files, which defaults to true. Make the crash reproducer
explicitly set that flag based on the case sensitivity of the .cache
path where vfs and modules are dumped.
Tim Northover [Wed, 6 Apr 2016 19:58:07 +0000 (19:58 +0000)]
Restore slightly less dodgy diagnostic handler for inline asm
Turns out it was there mostly to prevent Clang asking people to report a bug.
This time we report something to Clang's real diagnostics handler so that it
exits with something approximating a real error and tidies up after itself.
Richard Smith [Wed, 6 Apr 2016 17:06:00 +0000 (17:06 +0000)]
Re-commit r265518 ("[modules] Continue factoring encoding of AST records out of
ASTWriter."), reverted in r265526, with a fix for an iterator invalidation bug
(thanks, MSan!).
Ed Schouten [Wed, 6 Apr 2016 15:37:06 +0000 (15:37 +0000)]
Enable PIE for CloudABI.
As we're currently working on making CloudABI executables easier to
emulate in userspace (e.g., on OS X and Windows), it makes a whole lot
of sense to build these using PIE. By using PIE, they can simply be
loaded into the existing process address space without clashes.
PIE support got added to CloudABI's C library and seems to work pretty
well. CloudABI does not make use of an ld.so, so the binary's _start()
has all the logic in it to do the relocations.
Now that all but one bug in LLD relating to PIE support have been
squashed (and a patch for that is already in code review), I'd like to
go ahead and force the use of PIE for Clang 3.9. When released, we'll
also switch over to using LLD exclusively.
[OPENMP] Parsing and Sema support for 'omp declare target' directive
Add parsing, sema analysis for 'declare target' construct for OpenMP 4.0
(4.5 support will be added in separate patch).
The declare target directive specifies that variables, functions (C, C++
and Fortran), and subroutines (Fortran) are mapped to a device. The declare
target directive is a declarative directive. In Clang declare target is
implemented as implicit attribute for the declaration.
The syntax of the declare target directive is as follows:
#pragma omp declare target
declarations-definition-seq
#pragma omp end declare target
Based on patch from Michael Wong http://reviews.llvm.org/D15321
Richard Trieu [Tue, 5 Apr 2016 21:13:54 +0000 (21:13 +0000)]
Fix a crash on invalid with template handling
This is a fix for https://llvm.org/bugs/show_bug.cgi?id=25561 which was a
crash on invalid. Change the handling of invalid decls to have a catch-all
case to prevent unexpecting decls from triggering an assertion.
Summary:
Setting this flag causes all functions are annotated with the
"nvvm-f32ftz" = "true" attribute.
In addition, we annotate the module with "nvvm-reflect-ftz" set
to 0 or 1, depending on whether -cuda-flush-denormals-to-zero is set.
This is read by the NVVMReflect pass.
Andy Gibbs [Tue, 5 Apr 2016 08:36:47 +0000 (08:36 +0000)]
Consolidate and improve the handling of built-in feature-like macros
Summary:
The parsing logic has been separated out from the macro implementation logic, leading to a number of improvements:
* Gracefully handle unexpected/invalid tokens, too few, too many and nested parameters
* Provide consistent behaviour between all built-in feature-like macros
* Simplify the implementation of macro logic
* Fix __is_identifier to correctly return '0' for non-identifiers
[CrashReproducer] Pass -I, -F and -resource-dir to the reproducer script when using modules/vfs
The reproducer should use -I/-F/-resource-dir in the same way as the
original command. The VFS already collects the right headers but without
these flags the reproducer will fail to do the right thing.
The module.modulemap file in the lib/Headers directory was missing the LLVM
copyright notice. This patch adds the copyright notice just like the rest of
the files in this directory.
Carlo Bertolli [Mon, 4 Apr 2016 15:55:02 +0000 (15:55 +0000)]
[OPENMP] Codegen for teams directive for NVPTX
This patch implements the teams directive for the NVPTX backend. It is different from the host code generation path as it:
Does not call kmpc_fork_teams. All necessary teams and threads are started upon touching the target region, when launching a CUDA kernel, and their execution is coordinated through sequential and parallel regions within the target region.
Does not call kmpc_push_num_teams even if a num_teams of thread_limit clause is present. Setting the number of teams and the thread limit is implemented by the nvptx-related runtime.
Please note that I am now passing a Clang Expr * to emitPushNumTeams instead of the originally chosen llvm::Value * type. The reason for that is that I want to avoid emitting expressions for num_teams and thread_limit if they are not needed in the target region.
John Brawn [Mon, 4 Apr 2016 14:22:58 +0000 (14:22 +0000)]
Add a PragmaHandler Registry for plugins to add PragmaHandlers to
This allows plugins which add AST passes to also define pragmas to do things
like only enable certain behaviour of the AST pass in files where a certain
pragma is used.
The objc_runtime_visible attribute deals with an odd corner case where
a particular Objective-C class is known to the Objective-C runtime
(and, therefore, accessible by name) but its symbol has been hidden
for some reason. For such classes, teach CodeGen to use
objc_lookUpClass to retrieve the Class object, rather than referencing
the class symbol directly.
Classes annotated with objc_runtime_visible have two major limitations
that fall out from places where Objective-C metadata needs to refer to
the class (or metaclass) symbol directly:
* One cannot implement a subclass of an objc_runtime_visible class.
* One cannot implement a category on an objc_runtime_visible class.
[CodeGen] Emit lifetime.end intrinsic after objects are destructed in
landing pads.
Previously, lifetime.end intrinsics were inserted only on normal control
flows. This prevented StackColoring from merging stack slots for objects
that were destroyed on the exception handling control flow since it
couldn't tell their lifetime ranges were disjoint. This patch fixes
code-gen to emit the intrinsic on both control flows.
Richard Smith [Fri, 1 Apr 2016 22:52:03 +0000 (22:52 +0000)]
[modules] Start moving the code for encoding AST records out of ASTWriter into
a separate class. The goal is for this class to have a separate lifetime from
the AST writer so that it can meaningfully track pending statement nodes and
context for more compact encoding of various types.
[CrashReproducer] Add -fmodule-cache-path to reproducer script
The cc1 invocation in the reproducer script should contain a valid path in
-fmodule-cache-path; for that reuse "<name>.cache/module" dir we already
use to dump the vfs and modules.
Benjamin Kramer [Fri, 1 Apr 2016 09:58:45 +0000 (09:58 +0000)]
[Lexer] Don't read out of bounds if a conflict marker is at the end of a file
This can happen as we look for '<<<<' while scanning tokens but then expect
'<<<<\n' to tell apart perforce from diff3 conflict markers. Just harden
the pointer arithmetic.
[OPENMP 4.5] Allow data members as loop counters in loop-based
directives.
OpenMP 4.5 allows privatization of non-static data members in non-static
member functions. Patch allows to use and implicit privatization of data
members used as counters in loop-based directives.
Mike Spertus [Fri, 1 Apr 2016 03:31:43 +0000 (03:31 +0000)]
Visual Studio native visualizers for clang::TemplateSpecializationType
Displays a template specialization as, say, A<int, double>. Does not
yet handle UncommonTemplateNameStorage, QualifiedTemplateName, or
DependentTemplateName, but still more than worthwhile
[analyzer] Prefer accessor method in extension over category in CallEvent.
In ObjCMethodCall:getRuntimeDefinition(), if the method is an accessor in a
category, and it doesn't have a self declaration, first try to find the method
in a class extension. This works around a bug in Sema where multiple accessors
are synthesized for properties in class extensions that are redeclared in a
category. The implicit parameters are not filled in for the method on the
category, which causes a crash when trying to synthesize a getter for the
property in BodyFarm. The Sema bug is tracked as rdar://problem/25481164.
Tim Northover [Thu, 31 Mar 2016 19:19:24 +0000 (19:19 +0000)]
Diagnostics: remove dodgy handler for bitcode inlineasm diagnostics.
Whatever crash it was there to present appears to have been fixed in the
backend now, and it had the nasty side-effect of causing clang to exit(0) and
leave a .o containing goodness knows what even when an error hit.
Betul Buyukkurt [Thu, 31 Mar 2016 18:41:34 +0000 (18:41 +0000)]
[PGO] Avoid instrumenting constants at value sites
Value profiling should not profile constants and/or constant
expressions when they appear as callees in call instructions.
Constant expressions form when a direct callee has bitcasts or
inttoptr(ptrtint (callee)) nests surrounding it. Value profiling
should avoid instrumenting such cases. Mostly NFC.
Akira Hatanaka [Thu, 31 Mar 2016 06:36:07 +0000 (06:36 +0000)]
[CodeGenCXX] Fix ItaniumCXXABI::getAlignmentOfExnObject to return 8-byte
alignment on Darwin.
Itanium C++ ABI specifies that _Unwind_Exception should be double-word
aligned (16B). To conform to the ABI, libraries implementing exception
handling declare the struct with __attribute__((aligned)), which aligns
the unwindHeader field (and the end of __cxa_exception) to the default
target alignment (which is typically 16-bytes).
struct __cxa_exception {
...
// struct is declared with __attribute__((aligned)).
_Unwind_Exception unwindHeader;
};
Based on the assumption that _Unwind_Exception is declared with
__attribute__((aligned)), ItaniumCXXABI::getAlignmentOfExnObject returns
the target default alignment for __attribute__((aligned)). It turns out
that libc++abi, which is used on Darwin, doesn't declare the struct with
the attribute and therefore doesn't guarantee that unwindHeader is
aligned to the alignment specified by the ABI, which in some cases
causes the program to crash because of unaligned memory accesses.
This commit avoids crashes due to unaligned memory accesses by having
getAlignmentOfExnObject return an 8-byte alignment on Darwin. I've only
fixed the problem for Darwin, but we should also figure out whether other
platforms using libc++abi need similar fixes.
Richard Trieu [Thu, 31 Mar 2016 04:18:07 +0000 (04:18 +0000)]
Fix -Wdynamic-class-memaccess to skip invalid classes.
This warning sometimes will infinitely recurse on CXXRecordDecl's from
ill-formed recursive classes that have fields of themselves. Skip processing
these classes to prevent this from happening.
Fixes https://llvm.org/bugs/show_bug.cgi?id=27142
[Sema] Fix PR27122: ICE with enable_if+ill-formed call.
In some cases, when we encounter a direct function call with an
incorrect number of arguments, we'll emit a diagnostic, and pretend that
the call to the function was valid. For example, in C:
int foo();
int a = foo(1);
Prior to this patch, we'd get an ICE if foo had an enable_if attribute,
because CheckEnableIf assumes that the number of arguments it gets
passed is valid for the function it's passed. Now, we check that the
number of args looks valid prior to checking enable_if conditions.
This fix was not done inside of CheckEnableIf because the problem
presently can only occur in one caller of CheckEnableIf (ActOnCallExpr).
Additionally, checking inside of CheckEnableIf would make us emit
multiple diagnostics for the same error (one "enable_if failed", one
"you gave this function the wrong number of arguments"), which seems
worse than just complaining about the latter.
[CrashReproducer] Add a module map callback for added headers
The current ModuleDependencyCollector has a AST listener to collect
header files present in loaded modules, but this isn't enough to collect
all headers needed in the crash reproducer. One of the reasons is that
the AST writer doesn't write symbolic link header paths in the pcm modules,
this makes the listeners on the reader only able to collect the real files.
Since the module maps could contain submodules that use headers which
are symbolic links, not collecting those forbid the reproducer scripts
to regen the modules.
<SUBMODULE_HEADER abbrevid=6/> blob data = 'pthread_impl.h'
<SUBMODULE_TOPHEADER abbrevid=7/> blob data = '/<path_to_sdk>/usr/include/pthread/pthread_impl.h'
Note that we don't have "usr/include/pthread_impl.h" which is requested
by the module.map in case we want to reconstruct the module in the
reproducer. The reason the original symbolic link path isn't used is
because the headers are kept by name and requested through the
FileManager, which unique files and returns the real path only.
To fix that, add a callback to be invoked everytime a header is added
while parsing module maps and hook that up to the module dependecy
collector. This callback is only registered when generating the
reproducer.
The VFS YAML files contain empty directory entries to describe that it's
returning from a subdirectory before describing new files in the parent.
In the future, we should properly sort and write YAML files avoiding
such empty dirs and mitigate the extra recurson cost. However, since
this is used by previous existing YAMLs, make the traversal work in
their presence.
Justin Lebar [Wed, 30 Mar 2016 23:45:38 +0000 (23:45 +0000)]
[CUDA] Add -disable-llvm-passes to CodeGenCUDA/link-device-bitcode.cu. NFC
We already have this flag in most of the file, but we need it everywhere
else, to disable the NVVMReflect pass, which we're explicitly checking
doesn't run here. (Upcoming changes to llvm will cause it to be run.)
Eric Fiselier [Wed, 30 Mar 2016 23:39:56 +0000 (23:39 +0000)]
Fix deduction of __atomic_load's parameter types.
Summary:
__atomic_load's allows it's first argument to be a pointer to a const type. However the second argument is an output parameter and must be a pointer to non-const.
This patch fixes the signature of __atomic_load generated by clang so that it respects the above requirements.
Justin Lebar [Wed, 30 Mar 2016 23:30:25 +0000 (23:30 +0000)]
[CUDA] Don't initialize the CUDA toolchain if we don't have any CUDA inputs.
Summary:
This prevents errors when you invoke clang with a flag that the NVPTX
toolchain doesn't support. For example, on x86-64,
clang -mthread-model single -x c++ /dev/null -o /dev/null
should output just one error about "invalid thread model 'single' in
'-mthread-model single' for this target"; x86-64 doesn't support
-mthread-model, but we shouldn't also instantiate a NVPTX target!
Justin Lebar [Wed, 30 Mar 2016 23:30:21 +0000 (23:30 +0000)]
[CUDA] Make unattributed constexpr functions implicitly host+device.
With this patch, by a constexpr function is implicitly host+device
unless:
a) it's a variadic function (variadic functions are not allowed on the
device side), or
b) it's preceeded by a __device__ overload in a system header.
The restriction on overloading __host__ __device__ functions on the
basis of their CUDA attributes remains in place, but we use (b) to allow
us to define __device__ overloads for constexpr functions in cmath,
which would otherwise be __host__ __device__ and thus not overloadable.
You can disable this behavior with -fno-cuda-host-device-constexpr.
Justin Lebar [Wed, 30 Mar 2016 23:30:14 +0000 (23:30 +0000)]
[CUDA] Add math forward declares to CUDA header wrapper.
Summary:
This is necessary for a future patch which will make all constexpr
functions implicitly host+device. cmath may declare constexpr
functions, but these we do *not* want to be host+device. The forward
declares added in this patch prevent this (because the rule will be,
constexpr functions become implicitly host+device unless they're
preceeded by a decl with __device__).
projects / clang / log