Akira Hatanaka [Sat, 9 Dec 2017 00:02:37 +0000 (00:02 +0000)]
[CodeGen][X86] Fix handling of __fp16 vectors.
This commit fixes a bug in IRGen where it generates completely broken
code for __fp16 vectors on X86. For example when the following code is
compiled:
half4 hv0, hv1, hv2; // these are vectors of __fp16.
void foo221() {
hv0 = hv1 + hv2;
}
clang generates the following IR, in which two i16 vectors are added:
@hv1 = common global <4 x i16> zeroinitializer, align 8
@hv2 = common global <4 x i16> zeroinitializer, align 8
@hv0 = common global <4 x i16> zeroinitializer, align 8
define void @foo221() {
%0 = load <4 x i16>, <4 x i16>* @hv1, align 8
%1 = load <4 x i16>, <4 x i16>* @hv2, align 8
%add = add <4 x i16> %0, %1
store <4 x i16> %add, <4 x i16>* @hv0, align 8
ret void
}
To fix the bug, this commit uses the code committed in r314056, which
modified clang to promote and truncate __fp16 vectors to and from float
vectors in the AST. It also fixes another IRGen bug where a short value
is assigned to an __fp16 variable without any integer-to-floating-point
conversion, as shown in the following example:
__fp16 a;
short b;
void foo1() {
a = b;
}
@b = common global i16 0, align 2
@a = common global i16 0, align 2
Richard Smith [Fri, 8 Dec 2017 22:57:11 +0000 (22:57 +0000)]
Unify implementation of our two different flavours of -Wtautological-compare,
and fold together into a single function.
In so doing, fix a handful of remaining bugs where we would report false
positives or false negatives if we promote a signed value to an unsigned type
for the comparison.
This re-commits r320122 and r320124, minus two changes:
* Comparisons between a constant and a non-constant expression of enumeration
type never warn, not even if the constant is out of range. We should be
warning about the creation of such a constant, not about its use.
* We do not use more precise bit-widths for comparisons against bit-fields.
The more precise diagnostics probably are the right thing, but we should
consider moving them under their own warning flag.
Other than the refactoring, this patch should only change the behavior for the
buggy cases (where the warnings didn't take into account that promotion from
signed to unsigned can leave a range of inaccessible values in the middle of
the promoted type).
Vedant Kumar [Fri, 8 Dec 2017 19:51:42 +0000 (19:51 +0000)]
[ubsan] array-bounds: Ignore params with constant size
This is a follow-up to r320128. Eli pointed out that there is some gray
area in the language standard about whether the constant size is exact,
or a lower bound.
Hans Wennborg [Fri, 8 Dec 2017 16:54:08 +0000 (16:54 +0000)]
Revert "Unify implementation of our two different flavours of -Wtautological-compare."
> Unify implementation of our two different flavours of -Wtautological-compare.
>
> In so doing, fix a handful of remaining bugs where we would report false
> positives or false negatives if we promote a signed value to an unsigned type
> for the comparison.
This caused a new warning in Chromium:
../../base/trace_event/trace_log.cc:1545:29: error: comparison of constant 64
with expression of type 'unsigned int' is always true
[-Werror,-Wtautological-constant-out-of-range-compare]
DCHECK(handle.event_index < TraceBufferChunk::kTraceBufferChunkSize);
~~~~~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The 'unsigned int' is really a 6-bit bitfield, which is why it's always
less than 64.
I thought we didn't use to warn (with out-of-range-compare) when comparing
against the boundaries of a type?
In stdbool.h, define bool, false, true only in gnu++98
GCC has meanwhile corrected that with the similar
<https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=216679> "C++11
explicitly forbids macros for bool, true and false."
Hans Wennborg [Fri, 8 Dec 2017 05:19:12 +0000 (05:19 +0000)]
Revert r320124 "Fold together the in-range and out-of-range portions of -Wtautological-compare."
This broke Chromium:
../../base/trace_event/trace_log.cc:1545:29: error: comparison of constant 64
with expression of type 'unsigned int' is always true
[-Werror,-Wtautological-constant-out-of-range-compare]
DCHECK(handle.event_index < TraceBufferChunk::kTraceBufferChunkSize);
~~~~~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The 'unsigned int' is really a 6-bit bitfield, which is why it's always
less than 63.
Did this use to fall under the "in-range" case before? I thought we
didn't use to warn when comparing against the boundaries of a type.
Richard Smith [Fri, 8 Dec 2017 00:45:25 +0000 (00:45 +0000)]
Unify implementation of our two different flavours of -Wtautological-compare.
In so doing, fix a handful of remaining bugs where we would report false
positives or false negatives if we promote a signed value to an unsigned type
for the comparison.
Aaron Ballman [Thu, 7 Dec 2017 21:46:26 +0000 (21:46 +0000)]
Add new language mode flags for C17.
This adds -std=c17, -std=gnu17, and -std=iso9899:2017 as language mode flags for C17 and updates the value of __STDC_VERSION__ to the value based on the C17 FDIS. Given that this ballot cannot succeed until 2018, it is expected that we (and GCC) will add c18 flags as aliases once the ballot passes.
Aaron Ballman [Thu, 7 Dec 2017 21:37:49 +0000 (21:37 +0000)]
Add support for the __has_c_attribute builtin preprocessor macro.
This behaves similar to the __has_cpp_attribute builtin macro in that it allows users to detect whether an attribute is supported with the [[]] spelling syntax, which can be enabled in C with -fdouble-square-bracket-attributes.
George Rokos [Thu, 7 Dec 2017 20:27:31 +0000 (20:27 +0000)]
[OpenMP] NVPTX: Set default/minimum compute capability to sm_35
The current implementation of the nvptx runtime (to be upstreamed shortly) uses the atomicMax operation on 64-bit integers.
This is only supported in compute capabilities 3.5 and later. I've changed the clang default to sm_35.
Alexey Bataev [Thu, 7 Dec 2017 19:49:28 +0000 (19:49 +0000)]
[OPENMP] Do not capture private variables in the target regions.
Private variables are completely redefined in the outlined regions, so
we don't need to capture them. Patch adds this behavior to the
target-based regions.
Sjoerd Meijer [Thu, 7 Dec 2017 09:54:39 +0000 (09:54 +0000)]
[ARM] ACLE parallel arithmetic and DSP style multiplications
This is a follow up of r302131, in which we forgot to add SemaChecking
tests. Adding these tests revealed two problems which have been fixed:
- added missing intrinsic __qdbl,
- properly range checking ssat16 and usat16.
Zhihao Yuan [Thu, 7 Dec 2017 07:03:15 +0000 (07:03 +0000)]
Allow conditions to be decomposed with structured bindings
Summary:
This feature was discussed but not yet proposed. It allows a structured binding to appear as a //condition//
if (auto [ok, val] = f(...))
So the user can save an extra //condition// if the statement can test the value to-be-decomposed instead. Formally, it makes the value of the underlying object of the structured binding declaration also the value of a //condition// that is an initialized declaration.
Considering its logicality which is entirely evident from its trivial implementation, I think it might be acceptable to land it as an extension for now before I write the paper.
Yaxun Liu [Thu, 7 Dec 2017 01:39:52 +0000 (01:39 +0000)]
CodeGen: Fix invalid bitcasts for memcpy
CreateCoercedLoad/CreateCoercedStore assumes pointer argument of
memcpy is in addr space 0, which is not correct and causes invalid
bitcasts for triple amdgcn---amdgiz.
Richard Smith [Wed, 6 Dec 2017 19:23:19 +0000 (19:23 +0000)]
Delete special-case "out-of-range" handling for bools, and just use the normal
codepath plus the new "minimum / maximum value of type" diagnostic to get the
same effect.
Move the warning for an in-range but tautological comparison of a constant (0
or 1) against a bool out of -Wtautological-constant-out-of-range-compare into
the more-appropriate -Wtautological-constant-compare.
Jonas Hahnfeld [Wed, 6 Dec 2017 19:15:28 +0000 (19:15 +0000)]
Fix PR35542: Correct adjusting of private reduction variable
The adjustment is calculated with CreatePtrDiff() which returns
the difference in (base) elements. This is passed to CreateGEP()
so make sure that the GEP base has the correct pointer type:
It needs to be a pointer to the base type, not a pointer to a
constant sized array.
Taewook Oh [Wed, 6 Dec 2017 17:00:53 +0000 (17:00 +0000)]
Stringizing raw string literals containing newline
Summary: This patch implements 4.3 of http://open-std.org/jtc1/sc22/wg21/docs/papers/2014/n4220.pdf. If a raw string contains a newline character, replace each newline character with the \n escape code. Without this patch, included test case (macro_raw_string.cpp) results compilation failure.
[OpenCL] Fix layering violation by getOpenCLTypeAddrSpace
Commit 7ac28eb0a5 / r310911 ("[OpenCL] Allow targets to select address
space per type", 2017-08-15) made Basic depend on AST, introducing a
circular dependency. Break this dependency by adding the
OpenCLTypeKind enum in Basic and map from AST types to this enum in
ASTContext.
Erik Verbruggen [Wed, 6 Dec 2017 09:02:52 +0000 (09:02 +0000)]
[libclang] Add function to get the buffer for a file
This can be used by clients in conjunction with an offset returned by
e.g. clang_getFileLocation. Now those clients do not need to also
open/read the file.
Shoaib Meenai [Tue, 5 Dec 2017 21:49:56 +0000 (21:49 +0000)]
[CMake] Use PRIVATE in target_link_libraries for executables
We currently use target_link_libraries without an explicit scope
specifier (INTERFACE, PRIVATE or PUBLIC) when linking executables.
Dependencies added in this way apply to both the target and its
dependencies, i.e. they become part of the executable's link interface
and are transitive.
Transitive dependencies generally don't make sense for executables,
since you wouldn't normally be linking against an executable. This also
causes issues for generating install export files when using
LLVM_DISTRIBUTION_COMPONENTS. For example, clang has a lot of LLVM
library dependencies, which are currently added as interface
dependencies. If clang is in the distribution components but the LLVM
libraries it depends on aren't (which is a perfectly legitimate use case
if the LLVM libraries are being built static and there are therefore no
run-time dependencies on them), CMake will complain about the LLVM
libraries not being in export set when attempting to generate the
install export file for clang. This is reasonable behavior on CMake's
part, and the right thing is for LLVM's build system to explicitly use
PRIVATE dependencies for executables.
Unfortunately, CMake doesn't allow you to mix and match the keyword and
non-keyword target_link_libraries signatures for a single target; i.e.,
if a single call to target_link_libraries for a particular target uses
one of the INTERFACE, PRIVATE, or PUBLIC keywords, all other calls must
also be updated to use those keywords. This means we must do this change
in a single shot. I also fully expect to have missed some instances; I
tested by enabling all the projects in the monorepo (except dragonegg),
and configuring both with and without shared libraries, on both Darwin
and Linux, but I'm planning to rely on the buildbots for other
configurations (since it should be pretty easy to fix those).
Even after this change, we still have a lot of target_link_libraries
calls that don't specify a scope keyword, mostly for shared libraries.
I'm thinking about addressing those in a follow-up, but that's a
separate change IMO.
[analyzer] do not crash on cases where an array subscript is an rvalue
Array subscript is almost always an lvalue, except for a few cases where
it is not, such as a subscript into an Objective-C property, or a
return from the function.
This commit prevents crashing in such cases.
Artem Dergachev [Tue, 5 Dec 2017 17:14:39 +0000 (17:14 +0000)]
[analyzer] Mark heap-based symbolic regions in debug dumps.
They are now printed as HeapSymRegion{$x} in order to discriminate between that
and regular SymRegion{$x}, which are two different regions, having different
parent reginos (memory spaces) - HeapSpaceRegion and UnknownSpaceRegion
respectively.
Alex Lorenz [Tue, 5 Dec 2017 02:30:43 +0000 (02:30 +0000)]
[libclang] Store unsaved file hashes when recording parsing invocations
Storing the contents of unsaved files is too expensive.
Instead a hash is stored with a record invocation. When a reproducer is
generated, Clang will compare the stored hashes to the new hashes to determine
if the contents of a file has changed. This way we'll know when a reproducer was
generated for a different source to the one that triggered the original crash.
Hans Wennborg [Mon, 4 Dec 2017 22:28:45 +0000 (22:28 +0000)]
Correctly handle line table entries without filenames during AST serialization
The current code would hit an assert in ASTWriter when trying to write
out the filename for a line table entry that didn't have any. Fix this
by allowing the -1 sentinel value to round-trip through serialization.
Erich Keane [Mon, 4 Dec 2017 21:58:43 +0000 (21:58 +0000)]
Add _Float128 as alias to __float128 to enable compilations on Fedora27/glibc2-26
Fedora27 is using a new version of glibc that refers to the _Float128 type. This
patch adds that name as an alias to float128. I also added some predefined macro
values for the digits, mantissa, epilon, etc (FloatMacros). For the test case, I
copied an existing float128 test. This functionality needs work long term, but
it should be sufficient to tread water for a while. At Intel we have test
servers running our LLVM compiler with various open source workloads, the server
has been upgraded to Fedora27 so many workloads are failing due to _Float128.
Alex Lorenz [Mon, 4 Dec 2017 21:56:36 +0000 (21:56 +0000)]
[libclang] Record parsing invocation to a temporary file when requested
by client
This patch extends libclang by allowing it to record parsing operations to a
temporary JSON file. The file is deleted after parsing succeeds. When a crash
happens during parsing, the file is preserved and the client will be able to use
it to generate a reproducer for the crash.
These files are not emitted by default, and the client has to specify the
invocation emission path first.
Alexey Bataev [Mon, 4 Dec 2017 21:30:42 +0000 (21:30 +0000)]
[OPENMP] Fix PR35486: crash when collapsing loops with dependent iteration spaces.
Though it is incorrect from point of view of OpenMP standard to have
dependent iteration space in OpenMP loops, compiler should not crash.
Patch fixes this problem.
Two copies of getSymLERange in RangeConstraintManager are virtually
identical, which is clearly bad.
This patch uses lambdas to call one from another (assuming that we would
like to avoid getting ranges from the state when necessary).
Aaron Ballman [Mon, 4 Dec 2017 20:27:34 +0000 (20:27 +0000)]
Now that C++17 is official (https://www.iso.org/standard/68564.html), start changing the C++1z terminology over to C++17. NFC intended, these are all mechanical changes.
design document for a hardware-assisted memory safety (HWAMS) tool, similar to AddressSanitizer
Summary:
preliminary design document for a hardware-assisted memory safety (HWAMS) tool, similar to AddressSanitizer
The name TaggedAddressSanitizer and the rest of the document, are early draft, suggestions are welcome.
Manuel Klimek [Mon, 4 Dec 2017 08:53:16 +0000 (08:53 +0000)]
Fix bug where we wouldn't break columns over the limit.
Before, we would not break:
int a = foo(/* trailing */);
when the end of /* trailing */ was exactly the column limit; the reason
is that block comments can have an unbreakable tail length - in this case
2, for the trailing ");"; we would unconditionally account that when
calculating the column state at the end of the token, but not correctly
add it into the remaining column length before, as we do for string
literals.
The fix is to correctly account the trailing unbreakable sequence length
into our formatting decisions for block comments. Line comments cannot
have a trailing unbreakable sequence, so no change is needed for them.
Devin Coughlin [Mon, 4 Dec 2017 04:46:47 +0000 (04:46 +0000)]
[analyzer] Don't treat lambda-captures float constexprs as undefined
RegionStore has special logic to evaluate captured constexpr variables.
However, if the constexpr initializer cannot be evaluated as an integer, the
value is treated as undefined. This leads to false positives when, for example,
a constexpr float is captured by a lambda.
To fix this, treat a constexpr capture that cannot be evaluated as unknown
rather than undefined.
Richard Smith [Sat, 2 Dec 2017 02:48:42 +0000 (02:48 +0000)]
PR35456: Track definedness of variable template specializations separately from
whether they have an initializer.
We cannot distinguish between a declaration of a variable template
specialization and a definition of one that lacks an initializer without this,
and would previously mistake the latter for the former.
Sanjay Patel [Fri, 1 Dec 2017 23:15:52 +0000 (23:15 +0000)]
[CodeGen] convert math libcalls/builtins to equivalent LLVM intrinsics
There are 20 LLVM math intrinsics that correspond to mathlib calls according to the LangRef:
http://llvm.org/docs/LangRef.html#standard-c-library-intrinsics
We were only converting 3 mathlib calls (sqrt, fma, pow) and 12 builtin calls (ceil, copysign,
fabs, floor, fma, fmax, fmin, nearbyint, pow, rint, round, trunc) to their intrinsic-equivalents.
This patch pulls the transforms together and handles all 20 cases. The switch is guarded by a
check for const-ness to make sure we're not doing the transform if errno could possibly be set by
the libcall or builtin.
Richard Smith [Fri, 1 Dec 2017 21:24:36 +0000 (21:24 +0000)]
[c++17] When deducing the type of a non-type template parameter from the type
of its argument, perform function-to-pointer and array-to-pointer decay on the
parameter type first.
Otherwise deduction will fail, as the type of the argument will be decayed.