Add parsing, sema analysis and serialization/deserialization for 'declare reduction' construct.
User-defined reductions are defined as
#pragma omp declare reduction( reduction-identifier : typename-list : combiner ) [initializer ( initializer-expr )]
These custom reductions may be used in 'reduction' clauses of OpenMP constructs. The combiner specifies how partial results can be combined into a single value. The
combiner can use the special variable identifiers omp_in and omp_out that are of the type of the variables being reduced with this reduction-identifier. Each of them will
denote one of the values to be combined before executing the combiner. It is assumed that the special omp_out identifier will refer to the storage that holds the resulting
combined value after executing the combiner.
As the initializer-expr value of a user-defined reduction is not known a priori the initializer-clause can be used to specify one. Then the contents of the initializer-clause
will be used as the initializer for private copies of reduction list items where the omp_priv identifier will refer to the storage to be initialized. The special identifier
omp_orig can also appear in the initializer-clause and it will refer to the storage of the original variable to be reduced.
Differential Revision: http://reviews.llvm.org/D11182
llvm-svn: 262582
This patch changes cc1 option for PGO profile use from
-fprofile-instr-use=<path> to -fprofile-instrument-use-path=<path>.
-fprofile-instr-use=<path> is now a driver only option.
In addition to decouple the cc1 option from the driver level option, this patch
also enables IR level profile use. cc1 option handling now reads the profile
header and sets CodeGenOpt ProfileUse (valid values are {None, Clang, LLVM}
-- this is a common enum for -fprofile-instrument={}, for the profile
instrumentation), and invoke the pipeline to enable the respective PGO use pass.
Reviewers: silvas, davidxl
Differential Revision: http://reviews.llvm.org/D17737
llvm-svn: 262515
This is like r262493, but for pragma detect_mismatch instead of pragma comment.
The two pragmas have similar behavior, so use the same approach for both.
llvm-svn: 262506
... and register them with CUDA runtime.
This is needed for commonly used cudaMemcpy*() APIs that use address of
host-side shadow to access their counterparts on device side.
Fixes PR26340
Differential Revision: http://reviews.llvm.org/D17779
llvm-svn: 262498
`#pragma comment` was handled by Sema calling a function on ASTConsumer, and
CodeGen then implementing this function and writing things to its output.
Instead, introduce a PragmaCommentDecl AST node and hang one off the
TranslationUnitDecl for every `#pragma comment` line, and then use the regular
serialization machinery. (Since PragmaCommentDecl has codegen relevance, it's
eagerly deserialized.)
http://reviews.llvm.org/D17799
llvm-svn: 262493
This patch introduces the -fwhole-program-vtables flag, which enables the
whole-program vtable optimization feature (D16795) in Clang.
Differential Revision: http://reviews.llvm.org/D16821
llvm-svn: 261767
Summary:
Different devices may in some cases require different code generation schemes in order to implement OpenMP. This is required not only for performance reasons, but also because it may not be possible to have the current (default) implementation working for these devices. E.g. GPU's cannot implement the same scheme a target such as powerpc or x86b would use, in the sense that it does not have the ability to fork threads, instead all the threads are always executing and need to be managed by the implementation.
This patch proposes a reorganization of the code in the OpenMP code generation to pave the way to have specialized implementation of OpenMP support. More than a "real" patch this is more a request for comments in order to understand if what is proposed is acceptable or if there are better/easier ways to do it.
In this patch part of the common OpenMP codegen infrastructure is moved to a new file under a new namespace (CGOpenMPCommon) so it can be shared between the default implementation and the specialized one. When CGOpenMPRuntime is created, an attempt to select a specialized implementation is done.
In the patch a specialization for nvptx targets is done which currently checks if the target is an OpenMP device and trap if it is not.
Let me know comments suggestions you may have.
Reviewers: hfinkel, carlo.bertolli, arpith-jacob, kkwli0, ABataev
Subscribers: Hahnfeld, cfe-commits, fraggamuffin, caomhin, jholewinski
Differential Revision: http://reviews.llvm.org/D16784
llvm-svn: 259977
Avoid crashing when printing diagnostics for vtable-related CFI
errors. In diagnostic mode, the frontend does an additional check of
the vtable pointer against the set of all known vtable addresses and
lets the runtime handler know if it is safe to inspect the vtable.
http://reviews.llvm.org/D16823
llvm-svn: 259716
In general CUDA does not allow dynamic initialization of
global device-side variables. One exception is that CUDA allows
records with empty constructors as described in section E2.2.1 of
CUDA 7.5 Programming guide.
This patch applies initializer checks for all device-side variables.
Empty constructors are accepted, but no code is generated for them.
Differential Revision: http://reviews.llvm.org/D15305
llvm-svn: 259592
Member pointers in the MS ABI are tricky for a variety of reasons.
The size of a member pointer is indeterminate until the program reaches
a point where the representation is required to be known. However,
*pointers* to member pointers may exist without knowing the pointee
type's representation. In these cases, we synthesize an opaque LLVM
type for the pointee type.
However, we can be in a situation where the underlying member pointer's
representation became known mid-way through the program. To account for
this, we attempted to manicure CodeGen's type-cache so that we can
replace the opaque member pointer type with the real deal while leaving
the pointer types unperturbed. This, unfortunately, is a problematic
approach to take as we will violate CodeGen's invariants.
These violations are mostly harmless but let's do the right thing
instead: invalidate the type-cache if a member pointer's LLVM
representation changes.
This fixes PR26313.
llvm-svn: 258839
* Runtime diagnostic data for cfi-icall changed to match the rest of
cfi checks
* Layout of all CFI diagnostic data changed to put Kind at the
beginning. There is no ABI stability promise yet.
* Call cfi_slowpath_diag instead of cfi_slowpath when needed.
* Emit __cfi_check_fail function, which dispatches a CFI check
faliure according to trap/recover settings of the current module.
* A tiny driver change to match the way the new handlers are done in
compiler-rt.
llvm-svn: 258745
Summary:
These aliases are done to support inline asm, but there's nothing we can
do: NVPTX doesn't support aliases.
Reviewers: tra
Subscribers: cfe-commits, jhen, echristo
Differential Revision: http://reviews.llvm.org/D16501
llvm-svn: 258734
This is part of a new statistics gathering feature for the sanitizers.
See clang/docs/SanitizerStats.rst for further info and docs.
Differential Revision: http://reviews.llvm.org/D16175
llvm-svn: 257971
Proper diagnostic and resolution of mangled names' conflicts in variables.
When there is a declaration and a definition using the same name but different
types, we emit what is in the definition. When there are two conflicting
definitions, we issue an error.
Differential Revision: http://reviews.llvm.org/D15686
llvm-svn: 257754
Clang got itself into the situation where we mangled the same
constructor twice with two different constructor types. After one of
the constructors were utilized, the tag used for one of the types
changed from class to struct because a class template became complete.
This resulted in one of the constructor types varying from the other
constructor.
Instead, force "base" constructor types to "complete" if the ABI doesn't
have constructor variants. This will ensure that GlobalDecls for both
variants will get the same mangled name.
This fixes PR26029.
llvm-svn: 257205
This patch attempts to fix the regressions identified when the patch was committed initially.
Thanks to Michael Liao for identifying the fix in the offloading metadata generation
related with side effects in evaluation of function arguments.
llvm-svn: 256933
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.
Reviewers: ABataev, hfinkel, tra, rjmccall, sfantao
Subscribers: FBrygidyn, piotr.rak, Hahnfeld, cfe-commits
Differential Revision: http://reviews.llvm.org/D12614
llvm-svn: 256842
Alexey Bataev
Rong Xu
Nico Weber
Artem Belevich
Peter Collingbourne
Samuel Antao
Renato Golin
Evgeniy Stepanov
Benjamin Kramer
Eric Christopher
David Majnemer
Justin Lebar
Andrey Bokhanko
Chad Rosier