[Sema] Teach overload resolution about unaddressable functions.
Given an expression like `(&Foo)();`, we perform overload resolution as
if we are calling `Foo` directly. This causes problems if `Foo` is a
function that can't have its address taken. This patch teaches overload
resolution to ignore functions that can't have their address taken in
such cases.
John McCall [Wed, 6 Jan 2016 23:34:20 +0000 (23:34 +0000)]
Properly bind up any cleanups in an ExprWithCleanups after
instantiating a default argument expression.
This was previously just working implicitly by reinstantiating
in the current context, but caching means that we weren't
registering cleanups in subsequent uses.
Richard Smith [Wed, 6 Jan 2016 22:49:11 +0000 (22:49 +0000)]
PR26048, PR26050: put non-type template parameters and indirect field decls
into IDNS_Tag in C++, because they conflict with redeclarations of tags. (This
doesn't affect elaborated-type-specifier lookup, which looks for IDNS_Type in
C++).
John McCall [Wed, 6 Jan 2016 22:34:54 +0000 (22:34 +0000)]
Only instantiate a default argument once.
By storing the instantiated expression back in the ParmVarDecl,
we remove the last need for separately storing the sub-expression
of a CXXDefaultArgExpr. This makes PCH/Modules merging quite
simple: CXXDefaultArgExpr records are serialized as references
to the ParmVarDecl, and we ignore redundant attempts to overwrite
the instantiated expression.
This has some extremely marginal impact on user-facing semantics.
However, the major effect is that it avoids IRGen errors about
conflicting definitions due to lambdas in the argument being
instantiated multiple times while sharing the same mangling.
It should also slightly improve memory usage and module file size.
Richard Trieu [Wed, 6 Jan 2016 21:11:18 +0000 (21:11 +0000)]
Improve conditional checking during template instantiation.
When the condition in an if statement, while statement, or for loop is created
during template instantiation, it calls MakeFullExpr with only the condition
expression. However, when these conditions are created for non-templated
code in the Parser, an additional SourceLocation is passed to MakeFullExpr.
The impact of this was that non-dependent templated code could produce
diagnostics that the same code outside templates would not. Adding the missing
SourceLocation makes diagnostics consistent between templated and non-templated
code.
Nico Weber [Wed, 6 Jan 2016 20:55:00 +0000 (20:55 +0000)]
Add -Wfor-loop-analysis to -Wall.
This warning seems to have 0 false positives and some true positives in
practice, without a measurable compile time cost. It should be in -Wall, and
possibly even become a default warning at some point.
Dan Gohman [Wed, 6 Jan 2016 19:43:32 +0000 (19:43 +0000)]
[WebAssembly] Add --gc-sections to the link line.
This will eventually be accompanied with a change to enable -ffunction-sections
and -fdata-sections by default, which is currently delayed by some development
process issues.
Erik Verbruggen [Wed, 6 Jan 2016 15:12:51 +0000 (15:12 +0000)]
Show inclusions from a preamble in clang_getInclusions.
When reparsing a translation unit with preamble generation turned on,
no includes are found. This is due to the fact that all SLocs from
AST/PCH files are skipped as they are 'loaded', and inclusions from a
preamble are also 'loaded'. So, in case a file has a preamble, it first
needs to process those loaded inclusions, and then check for any local
inclusions. This latter one is for any includes that are not part of the
preamble, like includes half-way through a file.
Dimitry Andric [Wed, 6 Jan 2016 07:42:18 +0000 (07:42 +0000)]
Add -fno-movt frontend option, to disable movt/movw on ARM
Summary:
In rL256641, @davide turned off movt generation by default for FreeBSD.
This was because our ld is very old, and did not support the relocations
for it. However, Ian Lepore added the support very recently, so we
would like to revert rL256641, and replace it with a new `-fno-movt`
frontend option. This way, it can be turned off when needed.
Change the set of actions built for external gcc tools.
A gcc tool has an "integrated" assembler (usually gas) that it
will call to produce an object. Let it use that assembler so
that we don't have to deal with assembly syntax incompatibilities.
Richard Smith [Wed, 6 Jan 2016 03:52:10 +0000 (03:52 +0000)]
[modules] When a tag type that was imported from a module is referenced via an
elaborated-type-specifier, create a declaration of it to track that the current
module makes it visible too.
Anna Zaks [Wed, 6 Jan 2016 00:32:52 +0000 (00:32 +0000)]
[analyzer] Suppress reports coming from std::__independent_bits_engine
The analyzer reports a shift by a negative value in the constructor. The bug can
be easily triggered by calling std::random_shuffle on a vector
(<rdar://problem/19658126>).
(The shift by a negative value is reported because __w0_ gets constrained to
63 by the conditions along the path:__w0_ < _WDt && __w0_ >= _WDt-1,
where _WDt is 64. In normal execution, __w0_ is not 63, it is 1 and there is
no overflow. The path is infeasible, but the analyzer does not know about that.)
After configuration the following additional targets will be generated:
stage2-instrumented:
Builds a stage1 x86 compiler, runtime, and required tools (llvm-config, llvm-profdata) then uses that compiler to build an instrumented stage2 compiler.
stage2-instrumented-generate-profdata:
Depends on "stage2-instrumented" and will use the instrumented compiler to generate profdata based on the training files in <clang>/utils/perf-training
stage2:
Depends on "stage2-instrumented-generate-profdata" and will use the stage1 compiler with the stage2 profdata to build a PGO-optimized compiler.
stage2-check-llvm:
Depends on stage2 and runs check-llvm using the stage3 compiler.
stage2-check-clang:
Depends on stage2 and runs check-clang using the stage3 compiler.
stage2-check-all:
Depends on stage2 and runs check-all using the stage3 compiler.
stage2-test-suite:
Depends on stage2 and runs the test-suite using the stage3 compiler (requires in-tree test-suite).
Oleg Ranevskyy [Tue, 5 Jan 2016 19:54:39 +0000 (19:54 +0000)]
[Clang/Support/Windows/Unix] Command lines created by clang may exceed the command length limit set by the OS
Summary:
LLVM part of the patch is D15831.
When clang runs an external tool such as a linker it may create a command line that exceeds the length limit.
Clang uses the llvm::sys::argumentsFitWithinSystemLimits function to check if command line length fits the OS
limitation. There are two problems in this function that may cause exceeding of the limit:
1. It ignores the length of the program path in its calculations. On the other hand, clang adds the program
path to the command line when it runs the program.
2. It assumes no space character is inserted after the last argument, which is not true for Windows. The flattenArgs function adds the trailing space for *each* argument. The result of this is that the terminating NULL character is not counted and may be placed beyond the length limit if the command line is exactly 32768 characters long. The WinAPI's CreateProcess does not find the NULL character and fails.
[PGO] Enable clang to pass compiler-rt profile support library to linker on Windows
Summary: This change enables clang to automatically link binaries built with the -fprofile-instr-generate against the clang_rt.profile-i386.lib library.
Samuel Antao [Tue, 5 Jan 2016 16:23:04 +0000 (16:23 +0000)]
[OpenMP] Offloading descriptor registration and device codegen.
Summary:
In order to offloading work properly two things need to be in place:
- a descriptor with all the offloading information (device entry functions, and global variable) has to be created by the host and registered in the OpenMP offloading runtime library.
- all the device functions need to be emitted for the device and a convention has to be in place so that the runtime library can easily map the host ID of an entry point with the actual function in the device.
This patch adds support for these two things. However, only entry functions are being registered given that 'declare target' directive is not yet implemented.
About offloading descriptor:
The details of the descriptor are explained with more detail in http://goo.gl/L1rnKJ. Basically the descriptor will have fields that specify the number of devices, the pointers to where the device images begin and end (that will be defined by the linker), and also pointers to a the begin and end of table whose entries contain information about a specific entry point. Each entry has the type:
```
struct __tgt_offload_entry{
void *addr;
char *name;
int64_t size;
};
```
and will be implemented in a pre determined (ELF) section `.omp_offloading.entries` with 1-byte alignment, so that when all the objects are linked, the table is in that section with no padding in between entries (will be like a C array). The code generation ensures that all `__tgt_offload_entry` entries are emitted in the same order for both host and device so that the runtime can have the corresponding entries in both host and device in same index of the table, and efficiently implement the mapping.
The resulting descriptor is registered/unregistered with the runtime library using the calls `__tgt_register_lib` and `__tgt_unregister_lib`. The registration is implemented in a high priority global initializer so that the registration happens always before any initializer (that can potentially include target regions) is run.
The driver flag -omptargets= was created to specify a comma separated list of devices the user wants to support so that the new functionality can be exercised. Each device is specified with its triple.
About target codegen:
The target codegen is pretty much straightforward as it reuses completely the logic of the host version for the same target region. The tricky part is to identify the meaningful target regions in the device side. Unlike other programming models, like CUDA, there are no already outlined functions with attributes that mark what should be emitted or not. So, the information on what to emit is passed in the form of metadata in host bc file. This requires a new option to pass the host bc to the device frontend. Then everything is similar to what happens in CUDA: the global declarations emission is intercepted to check to see if it is an "interesting" declaration. The difference is that instead of checking an attribute, the metadata information in checked. Right now, there is only a form of metadata to pass information about the device entry points (target regions). A class `OffloadEntriesInfoManagerTy` was created to manage all the information and queries related with the metadata. The metadata looks like this:
```
!omp_offload.info = !{!0, !1, !2, !3, !4, !5, !6}
!0 = !{i32 0, i32 52, i32 77426347, !"_ZN2S12r1Ei", i32 479, i32 13, i32 4}
!1 = !{i32 0, i32 52, i32 77426347, !"_ZL7fstatici", i32 461, i32 11, i32 5}
!2 = !{i32 0, i32 52, i32 77426347, !"_Z9ftemplateIiET_i", i32 444, i32 11, i32 6}
!3 = !{i32 0, i32 52, i32 77426347, !"_Z3fooi", i32 99, i32 11, i32 0}
!4 = !{i32 0, i32 52, i32 77426347, !"_Z3fooi", i32 272, i32 11, i32 3}
!5 = !{i32 0, i32 52, i32 77426347, !"_Z3fooi", i32 127, i32 11, i32 1}
!6 = !{i32 0, i32 52, i32 77426347, !"_Z3fooi", i32 159, i32 11, i32 2}
```
The fields in each metadata entry are (in sequence):
Entry 1) an ID of the type of metadata - right now only zero is used meaning "OpenMP target region".
Entry 2) a unique ID of the device where the input source file that contain the target region lives.
Entry 3) a unique ID of the file where the input source file that contain the target region lives.
Entry 4) a mangled name of the function that encloses the target region.
Entries 5) and 6) line and column number where the target region was found.
Entry 7) is the order the entry was emitted.
Entry 2) and 3) are required to distinguish files that have the same function name.
Entry 4) is required to distinguish different instances of the same declaration (usually templated ones)
Entries 5) and 6) are required to distinguish the particular target region in body of the function (it is possible that a given target region is not an entry point - if clause can evaluate always to zero - and therefore we need to identify the "interesting" target regions. )
This patch replaces http://reviews.llvm.org/D12306.
[OpenCL] Disallow taking an address of a function.
An undecorated function designator implies taking the address of a function,
which is illegal in OpenCL. Implementing a check for this earlier to allow
the error to be reported even in the presence of other more obvious errors.
Easwaran Raman [Mon, 4 Jan 2016 23:32:28 +0000 (23:32 +0000)]
Remove setting of inlinehint and cold attributes based on profile data
NFC. These hints are only used for inlining and the inliner now uses
the same criteria to identify hot and cold callees and set appropriate
thresholds without relying on these hints. Hence this removed code is
superfluous.
Dimitry Andric [Mon, 4 Jan 2016 10:17:48 +0000 (10:17 +0000)]
Convert test/CXX/lex/lex.literal/lex.string/p4.cpp back to DOS line
endings, since the file is supposed to have them, according to its
comments. Also set its svn:eol-style property. Noticed by Nico Weber.
[PGO] Cleanup: Use covmap header definition in the template file
This is one last remaining instrumentatation related structure
that needs to be migrate to use the centralized template
definition. With this change, instrumentation code
related to coverage module header will be kept in sync
with the coverage mapping reader. The remaining code
which makes implicit assumption about covmap control
structure layout in the the lowering pass will cleaned
up in a different patch. This patch is not intended to
have no functional change.
Alexey Bataev [Thu, 31 Dec 2015 06:52:34 +0000 (06:52 +0000)]
[OPENMP 4.5] Codegen for 'schedule' clause with monotonic/nonmonotonic modifiers.
OpenMP 4.5 adds support for monotonic/nonmonotonic modifiers in 'schedule' clause. Add codegen for these modifiers.
David Majnemer [Thu, 31 Dec 2015 05:36:46 +0000 (05:36 +0000)]
[MSVC Compat] Diagnose multiple default ctors for dllexport'd classes
The MS ABI emits a special default constructor closure thunk if a
default constructor has a weird calling convention or default arguments.
The MS ABI has a quirk: there can be only one such thunk because the
mangling scheme does not have room for distinct manglings. We must
raise a diagnostic in this eventuality.
N.B. MSVC sorta gets this right. Multiple default constructors result
in the default constructor closure getting emitted but they seem to
get confused by which default constructors are reasonable to reference
from the closure. We try to be a little more careful which results in
mild differences in behavior.
Richard Smith [Thu, 31 Dec 2015 02:02:54 +0000 (02:02 +0000)]
Improve diagnostic for the case where a function template candidate is rejected
by overload resolution because deduction succeeds, but the substituted
parameter type for some parameter (with deduced type) doesn't exactly match the
corresponding adjusted argument type.
Aaron Ballman [Wed, 30 Dec 2015 14:26:07 +0000 (14:26 +0000)]
When performing an implicit from float to bool, the floating point value must be *exactly* zero in order for the conversion to result in 0. This does not involve a conversion through an integer value, and so truncation of the value is not performed.
Davide Italiano [Wed, 30 Dec 2015 13:53:25 +0000 (13:53 +0000)]
Disable generating movt on FreeBSD.
It's sort of an hack, but we have no choice.
The linker in the base system doesn't handle that correctly (yet).
Once FreeBSD will import lld, this can be backed out.
Chandler Carruth [Wed, 30 Dec 2015 06:21:02 +0000 (06:21 +0000)]
[ptr-traits] Some compilers (older Clang? unsure) on build bots need
this constructor to be out of line in order to not require the complete
type of TemplateDecl.
David Majnemer [Wed, 30 Dec 2015 05:40:32 +0000 (05:40 +0000)]
[MS ABI] Add variable templates to the NameBackReferences
Only function template specializations are exempt from being added to
the NameBackReferences. Redundant variable template specializations
should be appropriately substituted.
David Majnemer [Wed, 30 Dec 2015 05:13:03 +0000 (05:13 +0000)]
[MS ABI] Improve our mangling of pass_object_size
We didn't add the artificial pass_object_size arguments to the
backreference map which bloated the size of manglings which involved
pass_object_size with duplicate types.
This lets us go from:
?qux@PassObjectSize@@YAHQAHW4__pass_object_size1@__clang@@0W4__pass_object_size1@3@@Z
to:
?qux@PassObjectSize@@YAHQAHW4__pass_object_size1@__clang@@01@Z
James Y Knight [Wed, 30 Dec 2015 03:58:33 +0000 (03:58 +0000)]
Fix alignment issue in CodeGenFunction::PopCleanupBlock.
It was copying an EHCleanupStack::Cleanup object into a
SmallVector<char>, with a comment saying that SmallVector's alignment is
always large enough. Unfortunately, that isn't actually true after
r162331 in 2012.
Expand the code (somewhat distastefully) to get a stack allocation with
a correct alignment.
Chandler Carruth [Wed, 30 Dec 2015 03:40:23 +0000 (03:40 +0000)]
[ptr-traits] Add #includes of headers rather than forward declarations
for types which are used as pointees in PointerUnions, PointerIntPairs,
and DenseMap pointer keys.
This is part of a series of patches to allow LLVM to check for complete
pointee types when computing its pointer traits. This is absolutely
necessary to get correct (or reproducible) results for things like how
many low bits are guaranteed to be zero.
I think this is the last patch for getting Clang clean here!!!
Chandler Carruth [Wed, 30 Dec 2015 03:33:22 +0000 (03:33 +0000)]
[ptr-traits] Move a class definition up to the top of this header so it
can be referenced as part of a PointerIntPair.
This is part of a series of patches to allow LLVM to check for complete
pointee types when computing its pointer traits. This is absolutely
necessary to get correct (or reproducible) results for things like how
many low bits are guaranteed to be zero.
Chandler Carruth [Wed, 30 Dec 2015 03:24:14 +0000 (03:24 +0000)]
[ptr-traits] Move methods manipulating PointerUnions, DenseMap pointer
keys, and PointerIntPairs where the pointee types are incomplete
out-of-line to where we have the complete type.
This is the standard pattern used throughout the AST library to address
the inherently mutually cross referenced nature of the AST.
This is part of a series of patches to allow LLVM to check for complete
pointee types when computing its pointer traits. This is absolutely
necessary to get correct (or reproducible) results for things like how
many low bits are guaranteed to be zero.
Chandler Carruth [Wed, 30 Dec 2015 03:09:25 +0000 (03:09 +0000)]
[ptr-traits] Switch the Redeclarable template to using a void pointer in
its PointerUnion rather than an ASTContext pointer.
Using pointers with PointerUnion really should be checking the pointee
type's alignment, and we can't do this without the complete type. The
Redeclarable template inherently can't know the complete type of
ASTContext because of its layering, and because it is a template its
methods can't reasonably be out-of-line the way we traditionally solve
circular references within the AST library.
After discussing this with Richard Smith, his suggestion which I have
implemented here was to just drop to a void* for the PointerUnion. This
essentially documents that we're going to completely ignore the type
(including its potential alignment consequences) for this code. There
are still of course dynamic guards that this ended up working correctly,
and because of the way the code is factored this doesn't leak outside of
the very narrow implementation guts of Redeclarable.
This is part of a series of patches to allow LLVM to check for complete
pointee types when computing its pointer traits. This is absolutely
necessary to get correct (or reproducible) results for things like how
many low bits are guaranteed to be zero.
Chandler Carruth [Wed, 30 Dec 2015 03:00:23 +0000 (03:00 +0000)]
[ptr-traits] Switch from a really wasteful SmallDenseMap of
SmallVector<.., 16> (16!!!!) objects to a simple SmallVector of pairs.
This no longer de-duplicates the common function pointers used during
deallocation, but this doesn't really seem worth the complexity and
overhead of managing the map-of-vectors. Notably, there is no reason to
assume that functions have the 4-byte alignment that DenseMap relies on,
and indeed this prevents checking the alignment of the DenseMap keys
because we can't even meaningfully query the alignment of functions
using our existing alignment tools.
Generally, function pointers don't seem like a great idea for keys in
a DenseMap. =]
I chatted with Richard Smith about this a bit as well and have written
down a FIXME because this *does* waste some memory and in general seems
a very clumsy memory management strategy. He would like to see a more
fundamental fix eventually here that tries to build a better pattern.
Chandler Carruth [Wed, 30 Dec 2015 02:51:00 +0000 (02:51 +0000)]
[ptr-traits] Stop using two bogus types as stand-ins for flags
indicating the nature of the default argument in a ParmVarDecl.
Instead, this adds a proper enum stored exclusively in the ParmVarDecl
bits (which we have plenty of) for this. This even allows us to track
a previously unrepresented state in Clang when we parse a function
declaration with a default argument on a parameter but we cannot even
form an invalid expression node (for example, it is an invalid token).
Now, we can model this state in the AST at least, and potentially
improve recovery in this area in the future.
I've also cleaned up the functions managing both variable initializer
expressions and parameter default argument expresssions as much as
possible. I've left some comments about further improvements based on
a discussion with Richard Smith. Lots of credit to him for walking me
through exactly which of the *many* tradeoffs here he felt was the best
fit.
Should be NFC for now. I've tried my best to preserve existing behavior.
This is part of a series of patches to allow LLVM to check for complete
pointee types when computing its pointer traits. This is absolutely
necessary to get correct (or reproducible) results for things like how
many low bits are guaranteed to be zero.