Richard Trieu [Sat, 22 Jun 2019 00:32:19 +0000 (00:32 +0000)]
[ODRHash] Skip some typedef types.
In some cases, a typedef only strips aways a keyword for a type, keeping the
same name as the root record type. This causes some confusion when the type
is defined in one modules but only forward declared in another. Skipping the
typedef and going straight to the record will avoid this issue.
typedef struct S {} S;
S* s; // S is TypedefType here
Erich Keane [Fri, 21 Jun 2019 22:29:32 +0000 (22:29 +0000)]
Ensure Target Features always_inline error happens in C++ cases.
A handful of C++ cases as reported in PR42352 didn't actually give an
error when always_inlining with a different target feature list. This
resulted in broken IR.
Richard Smith [Fri, 21 Jun 2019 20:46:22 +0000 (20:46 +0000)]
PR42301: Abort cleanly if we encounter a huge source file rather than
crashing.
Ideally we wouldn't care about the size of a file so long as it fits in
memory, but in practice we have lots of hardocded assumptions that
unsigned can be used to index files, string literals, and so on.
Alexey Bataev [Fri, 21 Jun 2019 17:28:41 +0000 (17:28 +0000)]
[OPENMP]Fix PR42068: Vla type is not captured.
If the variably modified type is declared outside of the captured region
and then used in the cast expression along with array subscript
expression, the type is not captured and it leads to the compiler crash.
Leonard Chan [Fri, 21 Jun 2019 16:03:06 +0000 (16:03 +0000)]
[clang][NewPM] Add -fno-experimental-new-pass-manager to tests
As per the discussion on D58375, we disable test that have optimizations under
the new PM. This patch adds -fno-experimental-new-pass-manager to RUNS that:
- Already run with optimizations (-O1 or higher) that were missed in D58375.
- Explicitly test new PM behavior along side some new PM RUNS, but are missing
this flag if new PM is enabled by default.
- Specify -O without the number. Based on getOptimizationLevel(), it seems the
default is 2, and the IR appears to be the same when changed to -O2, so
update the test to explicitly say -O2 and provide -fno-experimental-new-pass-manager`.
Aaron Ballman [Fri, 21 Jun 2019 14:37:39 +0000 (14:37 +0000)]
Add an automated note to files produced by gen_ast_dump_json_test.py.
This also details what filters, if any, were used to generate the test output. Updates all the current JSON testing files to include the automated note.
[cmake] Add llvm-dwarfdump to clang test dependencies
Commit r363496 ("[Clang] Harmonize Split DWARF options with llc",
2019-06-15) introduced the use of llvm-dwarfdump in the clang tests,
so ensure the clang tests are dependent on llvm-dwarfdump.
Sam Elliott [Fri, 21 Jun 2019 10:03:31 +0000 (10:03 +0000)]
[RISC-V] Add -msave-restore and -mno-save-restore to clang driver
Summary:
The GCC RISC-V toolchain accepts `-msave-restore` and `-mno-save-restore`
to control whether libcalls are used for saving and restoring the stack within
prologues and epilogues.
Clang currently errors if someone passes -msave-restore or -mno-save-restore.
This means that people need to change build configurations to use clang. This
patch adds these flags, so that clang invocations can now match gcc.
As the RISC-V backend does not currently have a `save-restore` target feature,
we emit a warning if someone requests `-msave-restore`. LLVM does not error if
we pass the (unimplemented) target features `+save-restore` or `-save-restore`.
[clang] Small improvments after Adding APValue to ConstantExpr
Summary:
this patch has multiple small improvements related to the APValue in ConstantExpr.
changes:
- APValue in ConstantExpr are now cleaned up using ASTContext::addDestruction instead of there own system.
- ConstantExprBits Stores the ValueKind of the result beaing stored.
- VerifyIntegerConstantExpression now stores the evaluated value in ConstantExpr.
- the Constant Evaluator uses the stored value of ConstantExpr when available.
Jordan Rupprecht [Thu, 20 Jun 2019 22:35:52 +0000 (22:35 +0000)]
[CodeGen][test] Use FileCheck variable matchers for better test support
Summary: Depending on how clang is built, it may discard the IR names and use names like `%2` instead of `%result.ptr`, causing tests that rely on the IR name to fail. Using FileCheck matchers makes the test work regardless of how clang is built.
This test passes with both `-fno-discard-value-names` and `-fdiscard-value-names` to make sure it passes regardless of the build mode.
Clang :: Driver/cl-response-file.c currently FAILs on Solaris:
Command Output (stderr):
--
/vol/llvm/src/clang/dist/test/Driver/cl-response-file.c:10:11: error: CHECK: expected string not found in input
// CHECK: "-I" "{{.*}}\\Inputs\\cl-response-file\\" "-D" "FOO=2"
^
Looking at the generated response file reveals that this is no surprise:
/I/vol/llvm/src/clang/dist/test/Driver\Inputs
with no newline at the end. The echo command used to create it boils down to
echo 'a\cb'
However, one cannot expect \c to be emitted literally: e.g. bash's builtin
echo has
\c suppress further output
I've tried various combinations of builtin echo, /usr/bin/echo, GNU echo if
different, the same for printf, and the backslash unescaped and quoted
(a\cb and a\\cb). The only combination that worked reliably on Solaris,
Linux, and macOS was
printf 'a\\cb'
so this is what this patch uses. Tested on amd64-pc-solaris2.11 and
x86_64-pc-linux-gnu.
Richard Smith [Thu, 20 Jun 2019 19:49:13 +0000 (19:49 +0000)]
Fix crash and rejects-valid when a later template parameter or default
template argument contains a backreference to a dependently-typed
earlier parameter.
In a case like:
template<typename T, T A, decltype(A) = A> struct X {};
template<typename U> auto Y = X<U, 0>();
we previously treated both references to `A` in the third parameter as
being of type `int` when checking the template-id in `Y`. That`s wrong;
the type of `A` in these contexts is the dependent type `U`.
When we encounter a non-type template argument that we can't convert to
the parameter type because of type-dependence, we now insert a dependent
conversion node so that the SubstNonTypeTemplateParmExpr for the
template argument will have the parameter's type rather than whatever
type the argument had.
Leonard Chan [Thu, 20 Jun 2019 19:44:51 +0000 (19:44 +0000)]
[clang][NewPM] Do not eliminate available_externally durng `-O2 -flto` runs
This fixes CodeGen/available-externally-suppress.c when the new pass manager is
turned on by default. available_externally was not emitted during -O2 -flto
runs when it should still be retained for link time inlining purposes. This can
be fixed by checking that we aren't LTOPrelinking when adding the
EliminateAvailableExternallyPass.
Leonard Chan [Thu, 20 Jun 2019 19:35:25 +0000 (19:35 +0000)]
[clang][NewPM] Move EntryExitInstrumenterPass to the start of the pipeline
This fixes CodeGen/x86_64-instrument-functions.c when running under the new
pass manager. The pass should go before any other pass to prevent
`__cyg_profile_func_enter/exit()` from not being emitted by inlined functions.
Craig Topper [Thu, 20 Jun 2019 18:24:29 +0000 (18:24 +0000)]
[X86] Make _mm_mask_cvtps_ph, _mm_maskz_cvtps_ph, _mm256_mask_cvtps_ph, and _mm256_maskz_cvtps_ph aliases for their corresponding cvt_roundps_ph intrinsic.
These intrinsics should always take an immediate for the rounding mode.
The base instruction comes from before EVEX embdedded rounding. The
user should always provide the immediate rather than us assuming
CUR_DIRECTION.
Make the 512-bit versions also explicit aliases instead of copy
pasting the code.
[OpenMP] Add support for handling declare target to clause when unified memory is required
Summary:
This patch adds support for the handling of the variables under the declare target to clause.
The variables in this case are handled like link variables are. A pointer is created on the host and then mapped to the device. The runtime will then copy the address of the host variable in the device pointer.
Puyan Lotfi [Thu, 20 Jun 2019 16:59:48 +0000 (16:59 +0000)]
[clang-ifs] Clang Interface Stubs, first version (second landing attempt).
This change reverts r363649; effectively re-landing r363626. At this point
clang::Index::CodegenNameGeneratorImpl has been refactored into
clang::AST::ASTNameGenerator. This makes it so that the previous circular link
dependency no longer exists, fixing the previous share lib
(-DBUILD_SHARED_LIBS=ON) build issue which was the reason for r363649.
Clang interface stubs (previously referred to as clang-ifsos) is a new frontend
action in clang that allows the generation of stub files that contain mangled
name info that can be used to produce a stub library. These stub libraries can
be useful for breaking up build dependencies and controlling access to a
library's internal symbols. Generation of these stubs can be invoked by:
Notice that -fvisibility (along with use of visibility attributes) can be used
to control what symbols get generated. Currently the interface format is
experimental but there are a wide range of possibilities here.
Currently clang-ifs produces .ifs files that can be thought of as analogous to
object (.o) files, but just for the mangled symbol info. In a subsequent patch
I intend to add support for merging the .ifs files into one .ifs/.ifso file
that can be the input to something like llvm-elfabi to produce something like a
.so file or .dll (but without any of the code, just symbols).
[Sema] Diagnose addr space mismatch while constructing objects
If we construct an object in some arbitrary non-default addr space
it should fail unless either:
- There is an implicit conversion from the address space to default
/generic address space.
- There is a matching ctor qualified with an address space that is
either exactly matching or convertible to the address space of an
object.
Xing Xue [Thu, 20 Jun 2019 15:36:32 +0000 (15:36 +0000)]
AIX system headers need stdint.h and inttypes.h to be re-enterable
Summary:
AIX system headers need stdint.h and inttypes.h to be re-enterable when macro _STD_TYPES_T is defined so that limit macro definitions such as UINT32_MAX can be found. This patch attempts to allow that on AIX.
Puyan Lotfi [Thu, 20 Jun 2019 06:01:06 +0000 (06:01 +0000)]
[clang][AST] Refactoring ASTNameGenerator to use pimpl pattern (NFC).
The original pimpl pattern used between CodegenNameGenerator and
CodegenNameGeneratorImpl did a good job of hiding DataLayout making it so that
users of CodegenNameGenerator did not need to link with llvm core. This is an
NFC change to neatly wrap ASTNameGenerator in a pimpl.
Artem Dergachev [Wed, 19 Jun 2019 23:33:59 +0000 (23:33 +0000)]
[analyzer] exploded-graph-rewriter: Implement a --diff mode.
In this mode the tool would avoid duplicating the contents of the
program state on every node, replacing them with a diff-like dump
of changes that happened on that node.
This is useful because most of the time we only interested in whether
the effect of the statement was modeled correctly. A diffed graph would
also be much faster to load and navigate, being much smaller than
the original graph.
The diffs are computed "semantically" as opposed to plain text diffs.
I.e., the diff algorithm is hand-crafted separately for every state trait,
taking the underlying data structures into account. This is especially nice
for Environment because textual diffs would have been terrible.
On the other hand, it requires some boilerplate to implement.
Artem Dergachev [Wed, 19 Jun 2019 23:33:51 +0000 (23:33 +0000)]
[analyzer] Fix JSON dumps for store clusters.
Include a unique pointer so that it was possible to figure out if it's
the same cluster in different program states. This allows comparing
dumps of different states against each other.
Artem Dergachev [Wed, 19 Jun 2019 23:33:48 +0000 (23:33 +0000)]
[analyzer] Fix JSON dumps for location contexts.
Location context ID is a property of the location context, not of an item
within it. It's useful to know the id even when there are no items
in the context, eg. for the purposes of figuring out how did contents
of the Environment for the same location context changed across states.
Artem Dergachev [Wed, 19 Jun 2019 23:33:42 +0000 (23:33 +0000)]
[analyzer] NFC: Change evalCall() to provide a CallEvent.
This changes the checker callback signature to use the modern, easy to
use interface. Additionally, this unblocks future work on allowing
checkers to implement evalCall() for calls that don't correspond to any
call-expression or require additional information that's only available
as part of the CallEvent, such as C++ constructors and destructors.
Artem Dergachev [Wed, 19 Jun 2019 23:33:39 +0000 (23:33 +0000)]
[analyzer] DeadStores: Add a crude suppression files generated by DriverKit IIG.
IIG is a replacement for MIG in DriverKit: IIG is autogenerating C++ code.
Suppress dead store warnings on such code, as the tool seems to be producing
them regularly, and the users of IIG are not in position to address these
warnings, as they don't control the autogenerated code. IIG-generated code
is identified by looking at the comments at the top of the file.
Artem Dergachev [Wed, 19 Jun 2019 23:33:34 +0000 (23:33 +0000)]
[analyzer] RetainCount: Add support for OSRequiredCast().
It's a new API for custom RTTI in Apple IOKit/DriverKit framework that is
similar to OSDynamicCast() that's already supported, but crashes instead of
returning null (and therefore causing UB when the cast fails unexpectedly).
Kind of like cast_or_null<> as opposed to dyn_cast_or_null<> in LLVM's RTTI.
Historically, RetainCountChecker was responsible for modeling OSDynamicCast.
This is simply an extension of the same functionality.
Puyan Lotfi [Wed, 19 Jun 2019 20:51:35 +0000 (20:51 +0000)]
[clang][AST] ASTNameGenerator: A refactoring of CodegenNameGeneratorImpl (NFC).
This is a NFC refactor move of CodegenNameGeneratorImpl from clang::Index to
clang:AST (and rename to ASTNameGenerator). The purpose is to make the
highlevel mangling code more reusable inside of clang (say in places like clang
FrontendAction). This does not affect anything in CodegenNameGenerator, except
that CodegenNameGenerator will now use ASTNameGenerator (in AST).
[clang] Adapt ASTMatcher to explicit(bool) specifier
Summary:
Changes:
- add an ast matcher for deductiong guide.
- allow isExplicit matcher for deductiong guide.
- add hasExplicitSpecifier matcher which give access to the expression of the explicit specifier if present.
Leonard Chan [Wed, 19 Jun 2019 17:41:30 +0000 (17:41 +0000)]
[clang][NewPM] Fixing remaining -O0 tests that are broken under new PM
- CodeGen/flatten.c will fail under new PM becausec the new PM AlwaysInliner
seems to intentionally inline functions but not call sites marked with
alwaysinline (D23299)
- Tests that check remarks happen to check them for the inliner which is not
turned on at O0. These tests just check that remarks work, but we can make
separate tests for the new PM with -O1 so we can turn on the inliner and
check the remarks with minimal changes.
Alex Lorenz [Wed, 19 Jun 2019 17:07:36 +0000 (17:07 +0000)]
Unify DependencyFileGenerator class and DependencyCollector interface (NFCI)
Make DependencyFileGenerator a DependencyCollector as it was intended when
DependencyCollector was introduced. The missing PPCallbacks overrides are added to
the DependencyCollector as well.
This change will allow clang-scan-deps to access the produced dependencies without
writing them out to .d files to disk, so that it will be able collate them and
report them to the user.
Summary:
This patch applies a change similar to rC363069, but for SARIF files.
The `%diff_sarif` lit substitution invokes `diff` with a non-portable
`-I` option. The intended effect can be achieved by normalizing the
inputs to `diff` beforehand. Such normalization can be done with
`grep -Ev`, which is also used by other tests.
Additionally, this patch updates the SARIF output to have a newline at
the end of the file. This makes it so that the SARIF file qualifies as a
POSIX text file, which increases the consumability of the generated file
in relation to various tools.
Aaron Ballman [Wed, 19 Jun 2019 15:24:06 +0000 (15:24 +0000)]
Change the way we output templates for JSON AST dumping and dump information about template arguments.
Previously, we attempted to write out template parameters and specializations to their own array, but due to the architecture of the ASTNodeTraverser, this meant that other nodes were not being written out. This now follows the same behavior as the regular AST dumper and puts all the (correct) information into the "inner" array. When we correct the AST node traverser itself, we can revisit splitting this information into separate arrays again.
[OpenMP] Strengthen regression tests for task allocation under nowait depend clauses NFC
Summary:
This patch strengthens the tests introduced in D63009 by:
- adding new test for default device ID.
- modifying existing tests to pass device ID local variable to the task allocation function.
[OpenCL] Split type and macro definitions into opencl-c-base.h
Using the -fdeclare-opencl-builtins option will require a way to
predefine types and macros such as `int4`, `CLK_GLOBAL_MEM_FENCE`,
etc. Move these out of opencl-c.h into opencl-c-base.h such that the
latter can be shared by -fdeclare-opencl-builtins and
-finclude-default-header.
This changes the behaviour of -finclude-default-header when
-fdeclare-opencl-builtins is specified: instead of including the full
header, it will include the header with only the base definitions.
Hans Wennborg [Wed, 19 Jun 2019 11:34:08 +0000 (11:34 +0000)]
Revert r363116 "[X86] [ABI] Fix i386 ABI "__m64" type bug"
This introduced MMX instructions in code that wasn't previously using
them, breaking programs using 64-bit vectors and x87 floating-point in
the same application. See discussion on the code review for more
details.
> According to System V i386 ABI: the __m64 type paramater and return
> value are passed by MMX registers. But current implementation treats
> __m64 as i64 which results in parameter passing by stack and returning
> by EDX and EAX.
>
> This patch fixes the bug (https://bugs.llvm.org/show_bug.cgi?id=41029)
> for Linux and NetBSD.
>
> Patch by Wei Xiao (wxiao3)
>
> Differential Revision: https://reviews.llvm.org/D59744
As discussed in the review for D62952, this patch pre-normalizes the
reference expected output sarif files by removing lines containing
fields for which we expect differences that should be ignored.
Aaron Puchert [Tue, 18 Jun 2019 22:57:08 +0000 (22:57 +0000)]
Suggestions to fix -Wmissing-{prototypes,variable-declarations}
Summary:
I've found that most often the proper way to fix this warning is to add
`static`, because if the code otherwise compiles and links, the function
or variable is apparently not needed outside of the TU.
We can't provide a fix-it hint for variable declarations, because
multiple VarDecls can share the same type, and if we put static in front
of that, we affect all declared variables, some of which might have
previous declarations.
We also provide no fix-it hint for the rare case of an `extern` function
definition, because that would require removing `extern` and I have no
idea how to get the source location of the storage class specifier from
a FunctionDecl. I believe this information is only available earlier in
the AST construction from DeclSpec::getStorageClassSpecLoc(), but we
don't have that here.
Aaron Puchert [Tue, 18 Jun 2019 22:52:39 +0000 (22:52 +0000)]
Show note for -Wmissing-prototypes for functions with parameters
Summary:
There was a search for non-prototype declarations for the function, but
we only showed the results for zero-parameter functions. Now we show the
note for functions with parameters as well, but we omit the fix-it hint
suggesting to add `void`.
Alexey Bataev [Tue, 18 Jun 2019 20:29:06 +0000 (20:29 +0000)]
[OPENMP]Use host's mangling for 128 bit float types on the device.
Device have to use the same mangling as the host for 128bit float types. Otherwise, the codegen for the device is unable to find the parent function when it tries to generate the outlined function for the target region and it leads to incorrect compilation and crash at the runtime.
Alexey Bataev [Tue, 18 Jun 2019 19:04:27 +0000 (19:04 +0000)]
[OPENMP][NVPTX]Correct codegen for 128 bit long double.
If the host uses 128 bit long doubles, the compiler should generate correct code for NVPTX devices. If the return type has 128 bit long doubles, in LLVM IR this type must be coerced to int array instead.
Alexey Bataev [Tue, 18 Jun 2019 18:39:26 +0000 (18:39 +0000)]
[OPENMP]Use host's long double when compiling the code for device.
The device code must use the same long double type as the host.
Otherwise the code cannot be linked and executed properly. Patch adds
only basic support and checks for supporting of the host long double
double on the device.
projects / clang / log