Akira Hatanaka [Wed, 18 Nov 2015 00:15:28 +0000 (00:15 +0000)]
Produce a better diagnostic for global register variables.
Currently, when there is a global register variable in a program that
is bound to an invalid register, clang/llvm prints an error message that
is not very user-friendly.
This commit improves the diagnostic and moves the check that used to be
in the backend to Sema. In addition, it makes changes to error out if
the size of the register doesn't match the declared variable size.
Richard Smith [Tue, 17 Nov 2015 23:32:01 +0000 (23:32 +0000)]
[modules] When a #include is mapped to a module import and appears somewhere
other than the top level, we issue an error. This breaks a fair amount of C++
code wrapping C libraries, where the C library is #included within a namespace
/ extern "C" combination, because the C library (probably) includes C++
standard library headers which may be within modules.
Without modules, this setup is harmless if (and *only* if) the corresponding
standard library module was already included outside the namespace, so
downgrade the error to a default-error extension in that case, so that it can
be selectively disabled for such misbehaving libraries.
Artem Belevich [Tue, 17 Nov 2015 22:28:55 +0000 (22:28 +0000)]
[CUDA] Make CUDA compilation usable by default.
Currently clang requires several additional command
line options in order to enable new features needed
during CUDA compilation. This patch makes these
options default.
* Automatically include cuda_runtime.h if we've found
a valid CUDA installation.
* Disable automatic CUDA header inclusion during unit tests.
* Added test case for command line construction.
* Enabled target overloads and relaxed call checks that are
needed in order to include CUDA headers.
* Added CUDA-7.5 installation path to the CUDA installation search list.
* Define __CUDA__ macro to indicate CUDA compilation.
Artem Belevich [Tue, 17 Nov 2015 22:28:52 +0000 (22:28 +0000)]
[CUDA] Added a wrapper header for inclusion of stock CUDA headers.
Header files that come with CUDA are assuming split host/device
compilation and are not usable by clang out of the box.
With a bit of preprocessor magic it's possible to twist them
into something clang can use.
This wrapper always includes CUDA headers exactly the same way during
host and device compilation passes and produces identical preprocessed
content during host and device side compilation for sm_35 GPUs. Device
compilation passes for older GPUs will see a smaller subset of device
functions supported by particular GPU.
The wrapper assumes specific contents of CUDA header files and works
only with CUDA 7.0 and 7.5.
Artem Belevich [Tue, 17 Nov 2015 22:28:50 +0000 (22:28 +0000)]
[CUDA] Detect and link with CUDA's libdevice bitcode library.
- added detection of libdevice bitcode file and API to find one appropriate for the GPU we're compiling for.
- pass additional cc1 options for linking with detected libdevice bitcode
- added -nocudalib to prevent automatic linking with libdevice
- added test cases to verify new functionality
Artem Belevich [Tue, 17 Nov 2015 22:28:46 +0000 (22:28 +0000)]
[CUDA] added include paths for both sides of CUDA compilation.
In order to compile a CUDA file clang must be able to find
include files for both both host and device.
This patch passes AuxToolchain to AddPreprocessingOptions and
uses it to add include paths for the opposite side of compilation.
We also must be able to find CUDA include files. If the driver
found CUDA installation, it adds appropriate include path
to CUDA headers. This can be disabled with '-nocudainc'.
- Added include paths for the opposite side of compilation.
- Added include paths to detected CUDA installation.
- Added -nocudainc to prevent adding CUDA include path.
- Added test cases to verify new functionality.
Artem Belevich [Tue, 17 Nov 2015 22:28:40 +0000 (22:28 +0000)]
[CUDA] use -aux-triple to pass target triple of opposite side of compilation
Clang needs to know target triple for both sides of compilation so that
preprocessor macros and target builtins from both sides are available.
This change augments Compilation class to carry information about
toolchains used during different CUDA compilation passes and refactors
BuildActions to use it when it constructs CUDA jobs.
Removed DeviceTriple from CudaHostAction/CudaDeviceAction as it's no
longer needed.
Tim Northover [Tue, 17 Nov 2015 18:27:27 +0000 (18:27 +0000)]
ARM: fix mismatch between Clang and backend on exception type.
"-arch armv7k" is only normally used with a watchos target, but Clang doesn't
stop you from giving a "-miphoneos-version-min" option too, converting the
triple to "thumbv7k-apple-ios". In this case the backend will decide to use
SjLj exceptions, so Clang needs to agree so it can create the correct
predefines.
Fortunately, there's a handy function to make the decision for us now.
Manuel Klimek [Tue, 17 Nov 2015 15:40:10 +0000 (15:40 +0000)]
Revert "Make FP_CONTRACT ON the default."
This reverts commit r253269.
This leads to assert / segfault triggering on the following reduced example:
float foo(float U, float base, float cell) { return (U = 2 * base) - cell; }
Richard Smith [Tue, 17 Nov 2015 03:02:41 +0000 (03:02 +0000)]
[modules] Fix some more cases where we used to reject a conflict between two
declarations that are not simultaneously visible, and where at least one of
them has internal/no linkage.
Eric Christopher [Mon, 16 Nov 2015 18:29:59 +0000 (18:29 +0000)]
When producing error messages for always_inline functions with the
target attribute, don't include "negative" subtarget features in the
list of required features. Builtins are positive by default so don't
need this change, but we pull the default list of features from the
command line and so need to make sure that we only include features
that are turned on for code generation in our error.
Oliver Stannard [Mon, 16 Nov 2015 14:58:50 +0000 (14:58 +0000)]
[ARM,AArch64] Fix __rev16l and __rev16ll intrinsics
These two intrinsics are defined in arm_acle.h.
__rev16l needs to rotate by 16 bits, bit it was actually rotating by 2 bits.
For AArch64, where long is 64 bits, this would still be wrong.
__rev16ll was incorrect, it reversed the bytes in each 32-bit word, rather than
each 16-bit halfword. The correct implementation is to apply __rev16 to the top
and bottom words of the 64-bit value.
For AArch32 targets, these get compiled down to the hardware rev16 instruction
at -O1 and above. For AArch64 targets, the 64-bit ones get compiled to two
32-bit rev16 instructions, because there is not currently a pattern for the
64-bit rev16 instruction.
Daniel Jasper [Mon, 16 Nov 2015 12:38:56 +0000 (12:38 +0000)]
clang-format: Enable #include sorting by default.
This has seen quite some usage and I am not aware of any issues. Also
add a style option to enable/disable include sorting. The existing
command line flag can from now on be used to override whatever is set
in the style.
Keno Fischer [Mon, 16 Nov 2015 09:04:13 +0000 (09:04 +0000)]
[CGDebugInfo] Set the size and align for reference types
In r253186, I changed the DIBuilder API to now take size and align
for reference types as well. This was done in preparation for upcoming
changes to the Verifier that will validate that sizes match between
DI types and IR values that are declared as having those types.
This updates clang to actually pass the information through.
Devin Coughlin [Sun, 15 Nov 2015 17:48:22 +0000 (17:48 +0000)]
[analyzer] Handle calling ObjC super method from inside C++ lambda.
When calling a ObjC method on super from inside a C++ lambda, look at the
captures to find "self". This mirrors how the analyzer handles calling super in
an ObjC block and fixes an assertion failure.
Craig Topper [Sun, 15 Nov 2015 03:32:11 +0000 (03:32 +0000)]
Fix a layering oddity by passing Sema to DeclSpec::Finish instead of DiagnosticsEngine and Preprocessor. Everything the preprocessor was being used for can be acquired from Sema.
Devin Coughlin [Sun, 15 Nov 2015 03:07:17 +0000 (03:07 +0000)]
[analyzer] Refer to capture field to determine if capture is reference.
The analyzer incorrectly treats captures as references if either the original
captured variable is a reference or the variable is captured by reference.
This causes the analyzer to crash when capturing a reference type by copy
(PR24914). Fix this by refering solely to the capture field to determine when a
DeclRefExpr for a lambda capture should be treated as a reference type.
David Majnemer [Sun, 15 Nov 2015 03:04:34 +0000 (03:04 +0000)]
[Sema] Don't crash trying to diagnose abs called on a pointer type
Clang tries to figure out if a call to abs is suspicious by looking
through implicit casts to look at the underlying, implicitly converted
type.
Interestingly, C has implicit conversions from pointer-ish types like
function to less exciting types like int. This trips up our 'abs'
checker because it doesn't know which variant of 'abs' is appropriate.
Instead, diagnose 'abs' called on function types upfront. This sort of
thing is highly suspicious and is likely indicative of a missing
pointer dereference/function call/array index operation.
Eric Christopher [Sat, 14 Nov 2015 02:38:37 +0000 (02:38 +0000)]
Add support for the always_inline + target feature diagnostic to print
out the first missing target feature that's required and reword
the diagnostic accordingly.
David Blaikie [Sat, 14 Nov 2015 01:10:38 +0000 (01:10 +0000)]
Make some tests LLVM-optimization agnostic and remove some others that were beyond value/repair
Several of these tests (the two deleted, and the one removal edit) were
relying on the optimizer to collapse things to test some frontend
feature. The tests were really old and features seemed amply covered by
other parts of the test suite, so I just removed them.
If anyone thinks they're valuable enough to keep/fix, we can play around
with that, for sure.
Juergen Ributzka [Fri, 13 Nov 2015 19:08:07 +0000 (19:08 +0000)]
Fix auto-link for text-based dynamic library SDKs.
When linking against text-based dynamic library SDKs the library name of a
framework has now more than one possible filename extensions. This fix tests for
both possible extensions (none, and .tbd).
Richard Smith [Fri, 13 Nov 2015 05:14:45 +0000 (05:14 +0000)]
[modules] When a declaration has non-trivial visibility, check whether it's
actually hidden before we check its linkage. This avoids computing the linkage
"too early" for an anonymous struct with a typedef name for linkage.
Richard Smith [Fri, 13 Nov 2015 03:52:13 +0000 (03:52 +0000)]
[modules] Follow the C++ standard's rule for linkage of enumerators: they have
the linkage of the enumeration. For enumerators of unnamed enumerations, extend
the -Wmodules-ambiguous-internal-linkage extension to allow selecting an
arbitrary enumerator (but only if they all have the same value, otherwise it's
ambiguous).
Richard Smith [Thu, 12 Nov 2015 22:19:45 +0000 (22:19 +0000)]
[modules] Simplify and generalize the existing rule for finding hidden
declarations in redeclaration lookup. A declaration is now visible to
lookup if:
* It is visible (not in a module, or in an imported module), or
* We're doing redeclaration lookup and it's externally-visible, or
* We're doing typo correction and looking for unimported decls.
We now support multiple modules having different internal-linkage or no-linkage
definitions of the same name for all entities, not just for functions,
variables, and some typedefs. As previously, if multiple such entities are
visible, any attempt to use them will result in an ambiguity error.
This patch fixes the linkage calculation for a number of entities where we
previously didn't need to get it right (using-declarations, namespace aliases,
and so on). It also classifies enumerators as always having no linkage, which
is a slight deviation from the C++ standard's definition, but not an observable
change outside modules (this change is being discussed on the -core reflector
currently).
This also removes the prior special case for tag lookup, which made some cases
of this work, but also led to bizarre, bogus "must use 'struct' to refer to type
'Foo' in this scope" diagnostics in C++.
Richard Smith [Thu, 12 Nov 2015 22:04:34 +0000 (22:04 +0000)]
DR407: Rationalize how we handle tags being hidden by typedefs. Even with
DR407, the C++ standard doesn't really say how this should work. Here's what we
do (which is consistent with DR407 as far as I can tell):
* When performing name lookup for an elaborated-type-specifier, a tag
declaration hides a typedef declaration that names the same type.
* When performing any other kind of lookup, a typedef declaration hides
a tag declaration that names the same type.
In any other case where lookup finds both a typedef and a tag (that is, when
they name different types), the lookup will be ambiguous. If lookup finds a
tag and a typedef that name the same type, and finds anything else, the lookup
will always be ambiguous (even if the other entity would hide the tag, it does
not also hide the typedef).
Richard Smith [Thu, 12 Nov 2015 21:55:58 +0000 (21:55 +0000)]
Revert r240335.
This failed to solve the problem it was aimed at, and introduced just as many
issues as it resolved. Realistically, we need to deal with the possibility that
multiple modules might define different internal linkage symbols with the same
name, and this isn't a problem unless two such symbols are simultaneously
visible.
The case where two modules define equivalent internal linkage symbols is
handled by r252063: if lookup finds multiple sufficiently-similar entities from
different modules, we just pick one of them as an extension (but we keep them
separate).
James Y Knight [Thu, 12 Nov 2015 18:37:29 +0000 (18:37 +0000)]
Correct atomic libcall support for __atomic_*_fetch builtins.
In r244063, I had caused these builtins to call the same-named library
functions, __atomic_*_fetch_SIZE. However, this was incorrect: while
those functions are in fact supported by GCC's libatomic, they're not
documented by the spec (and gcc doesn't ever call them).
Instead, you're /supposed/ to call the __atomic_fetch_* builtins and
then redo the operation inline to return the final value.
James Molloy [Thu, 12 Nov 2015 15:36:04 +0000 (15:36 +0000)]
[C++] Add the "norecurse" attribute to main() if in C++ mode
The C++ spec (3.6.1.3) says "The function `main` shall not be used within a program". This implies that it cannot recurse, so add the norecurse attribute to help the midend out a bit.
Re-recommit: Add support for the new mips-mti-linux toolchain.
Last time, this caused two Windows buildbots and a single ARM buildbot to fail.
I XFAIL'd the failing test on win32,win64 machines in order to see if the ARM
buildbot complains again.
projects / clang / log