We don't need to use the predetermined data-sharing attributes for the
loop counters if the user explicitly specified correct data-sharing
attributes for such variables.
llvm-svn: 352543
According to the report, better to keep the original strict compare
operation as the loop condition with unsigned loop counters to make the
loop countable. This allows further loop transformations.
llvm-svn: 352526
With commit r351627, LLVM gained the ability to apply (existing) IPO
optimizations on indirections through callbacks, or transitive calls.
The general idea is that we use an abstraction to hide the middle man
and represent the callback call in the context of the initial caller.
It is described in more detail in the commit message of the LLVM patch
r351627, the llvm::AbstractCallSite class description, and the
language reference section on callback-metadata.
This commit enables clang to emit !callback metadata that is
understood by LLVM. It does so in three different cases:
1) For known broker functions declarations that are directly
generated, e.g., __kmpc_fork_call for the OpenMP pragma parallel.
2) For known broker functions that are identified by their name and
source location through the builtin detection, e.g.,
pthread_create from the POSIX thread API.
3) For user annotated functions that carry the "callback(callee, ...)"
attribute. The attribute has to include the name, or index, of
the callback callee and how the passed arguments can be
identified (as many as the callback callee has). See the callback
attribute documentation for detailed information.
Differential Revision: https://reviews.llvm.org/D55483
llvm-svn: 351629
Summary: Change the strategy for computing loop index variables after collapsing a loop nest via the collapse clause by replacing the expensive remainder operation with multiplications and additions.
Reviewers: ABataev, caomhin
Reviewed By: ABataev
Subscribers: guansong, arphaman, cfe-commits
Differential Revision: https://reviews.llvm.org/D56413
llvm-svn: 350759
Summary: Introduce a compiler flag for cases when the user knows that the collapsed loop counter can be safely represented using at most 32 bits. This will prevent the emission of expensive mathematical operations (such as the div operation) on the iteration variable using 64 bits where 32 bit operations are sufficient.
Reviewers: ABataev, caomhin
Reviewed By: ABataev
Subscribers: hfinkel, kkwli0, guansong, cfe-commits
Differential Revision: https://reviews.llvm.org/D55928
llvm-svn: 350758
Summary:
There is a minor issue in how the implicit data-sharings for nested tasks are computed.
For the following example:
```
int x;
#pragma omp task shared(x)
#pragma omp task
x++;
```
We compute an implicit data-sharing of shared for `x` in the second task although I think that it should be firstprivate. Below you can find the part of the OpenMP spec that covers this example:
- // In a task generating construct, if no default clause is present, a variable for which the data-sharing attribute is not determined by the rules above and that in the enclosing context is determined to be shared by all implicit tasks bound to the current team is shared.//
- //In a task generating construct, if no default clause is present, a variable for which the data-sharing attribute is not determined by the rules above is firstprivate.//
Since each implicit-task has its own copy of `x`, we shouldn't apply the first rule.
Reviewers: ABataev
Reviewed By: ABataev
Subscribers: cfe-commits, rogfer01
Tags: #openmp
Differential Revision: https://reviews.llvm.org/D56430
llvm-svn: 350734
Each we create the target regions with the teams distribute inner
region, we can better estimate number of the teams required to execute
the target region. Function __kmpc_push_target_tripcount() is used for
purpose, which accepts device_id and the number of the iterations,
performed by the associated loop.
llvm-svn: 350571
As discussed in D56113, this patch refactors the implementation of the
const restriction for linear to reuse a function introduced by D56113.
A side effect is that, if a variable has mutable members, this
diagnostic is now skipped, and the diagnostic for the variable not
being an integer or pointer is reported instead.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D56299
llvm-svn: 350441
As discussed in D56113, this patch refactors the implementation of the
const restriction for reductions to reuse a function introduced by
D56113. A side effect is that diagnostics sometimes now say
"variable" instead of "list item" when a list item is a variable.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D56298
llvm-svn: 350440
The following appears in OpenMP 3.1 sec. 2.9.1.1 as a predetermined
data-sharing attribute:
> Variables with const-qualified type having no mutable member are
> shared.
It does not appear in OpenmP 4.0, 4.5, or 5.0. This patch removes the
implementation of that attribute when the requested OpenMP version is
greater than 3.1.
One effect of that removal is that `default(none)` affects const
variables without mutable members.
Also, without this patch, if a const variable without mutable members
was explicitly lastprivate or private, it was an error because it was
predetermined shared. Now, clang instead complains that it's const
without mutable fields, which is a more intelligible diagnostic. That
should be fine for all of the above versions because they all have
something like the following, which is quoted from OpenMP 5.0
sec. 2.19.3:
> A variable that is privatized must not have a const-qualified type
> unless it is of class type with a mutable member. This restriction does
> not apply to the firstprivate clause.
reduction and linear clauses already have separate checks for const
variables. Future patches will merge the implementations.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D56113
llvm-svn: 350439
nvvm_barrier0.
Use runtime functions instead of the direct call to the nvvm intrinsics.
It allows to prevent some dangerous LLVM optimizations, that breaks the
code for the NVPTX target.
llvm-svn: 350328
For constants with the predefined data-sharing clauses we may had
troubles with the target combined directives. It may cause compiler
crash in some corner cases.
llvm-svn: 350127
Not all %itanium_abi_triple values support TLS. Makes
OpenMP/declare_reduction_codegen.cpp, OpenMP/parallel_copyin_codegen.cpp for
%itanium_abi_triples without TLS support.
Alternatively we could pass -fnoopenmp-use-tls and tweak some of the CHECK
lines, but this seems a bit simpler.
Fixes PR40156.
Differential Revision: https://reviews.llvm.org/D56086
llvm-svn: 350067
Instead of generating llvm.mem.parallel_loop_access metadata, generate
llvm.access.group on instructions and llvm.loop.parallel_accesses on
loops. There is one access group per generated loop.
This is clang part of D52116/r349725.
Differential Revision: https://reviews.llvm.org/D52117
llvm-svn: 349823
Need to mark the loop as started when the initialization statement is
found. It is required to prevent possible incorrect loop iteraton
variable detection during template instantiation and fix the compiler
crash during the codegen.
llvm-svn: 349657
Without this patch, clang doesn't complain that X needs explicit data
sharing attributes in the following:
```
#pragma omp target teams default(none)
{
#pragma omp parallel num_threads(X)
;
}
```
However, clang does produce that complaint after the braces are
removed. With this patch, clang complains in both cases.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D55861
llvm-svn: 349635
A map clause with the close map-type-modifier is a hint to
prefer that the variables are mapped using a copy into faster
memory.
Patch by Ahsan Saghir (saghir)
Differential Revision: https://reviews.llvm.org/D55719
llvm-svn: 349551
buffer.
Seems to me, nvlink has a bug with the proper support of the weakly
linked symbols. It does not allow to define several shared memory buffer
with the different sizes even with the weak linkage. Instead we always
use 128 bytes buffer to prevent nvlink from the error message emission.
llvm-svn: 349540
Inlined runtime with the current implementation of the interwarp copy
function leads to the undefined behavior because of the not quite
correct implementation of the barriers. Start using generic
__kmpc_barier function instead of the custom made barriers.
llvm-svn: 349192
If the array section is based on pointer and this sections is mapped in
target region + then it is used in the inner parallel region, it also
must be globalized as the pointer itself is passed by value, not by
reference.
llvm-svn: 348492
Critical regions in NVPTX are the constructs, which, generally speaking,
are not supported by the NVPTX target. Instead we're using special
technique to handle the critical regions. Currently they are supported
only within the loop and all the threads in the loop must execute the
same critical region.
Inside of this special regions the regions still must be emitted as
critical, to avoid possible data races between the teams +
synchronization must use __kmpc_barrier functions.
llvm-svn: 348272
__kmpc_barrier runtime functions must be marked as convergent to prevent
some dangerous optimizations. Also, for NVPTX target all barriers must
be emitted as simple barriers.
llvm-svn: 348271
This moves everything primarily testing the functionality of -ast-dump and -ast-print into their own directory, rather than leaving the tests spread around the testing directory.
llvm-svn: 348017
Summary: This patch adds a new runtime for the SPMD deinit kernel function which replaces the previous function. The new function takes as argument the flag which signals whether the runtime is required or not. This enables the compiler to optimize out the part of the deinit function which are not needed.
Reviewers: ABataev, caomhin
Reviewed By: ABataev
Subscribers: jholewinski, guansong, cfe-commits
Differential Revision: https://reviews.llvm.org/D54970
llvm-svn: 347915
Fixed emission of the target regions found in the virtual functions.
Previously we may end up with the situation when those regions could be
skipped.
llvm-svn: 347793
modes.
If the region is inside target|teams|distribute region, we can emit the
locations with the correct info for execution mode and runtime mode.
Patch adds this ability to the NVPTX codegen to help the optimizer to
produce better code.
llvm-svn: 347583
For the NVPTX target default locations should be emitted as constants +
additional info must be emitted in the reserved_2 field of the ident_t
structure. The 1st bit controls the execution mode and the 2nd bit
controls use of the lightweight runtime. The combination of the bits for
Non-SPMD mode + lightweight runtime represents special undefined mode,
used outside of the target regions for orphaned directives or functions.
Should allow and additional optimization inside of the target regions.
llvm-svn: 347425
Loop-control variables with the default data-sharing attributes should
not be captured in the OpenMP region as they are private by default.
Also, default attributes should be emitted for such variables in the
inner OpenMP regions for the correct data sharing during codegen.
llvm-svn: 347409
forms of random access iterator
In OpenMP 4.5, only 4 relational operators are supported: <, <=, >,
and >=. This work is to enable support for relational operator
!= (not-equal) as one of the canonical forms.
Patch by Anh Tuyen Tran
Differential Revision: https://reviews.llvm.org/D54441
llvm-svn: 347405
This patch adjusts a test not to depend on deprecated FileCheck
behavior that permits overlapping matches within a block of CHECK-DAG
directives. Thus, this patch also removes uses of FileCheck's
-allow-deprecated-dag-overlap command-line option.
There were two issues in this test:
1. There were sets of patterns for store instructions in which a
pattern X could match a superset of a pattern Y. While X appeared
before Y, Y's intended match appeared before X's intended match. The
result was that X matched Y's intended match. Under the old
overlapping behavior, Y also matched Y's intended match. Under the
new non-overlapping behavior, Y had nothing left to match. This patch
fixes this by gathering these sets in one place and putting the most
specific patterns (Y) before the more general patterns (X).
2. The CHECK-DAG patterns involving the variables CBPADDR3 and
CBPADDR4 were the same, but there was only one match in the text, so
CBPADDR4 patterns had nothing to match under the new non-overlapping
behavior. Moreover, a preceding related series of directives had
variables (SADDR0, BPADDR0, etc.) numbered only 0 through 4, but this
series had variables numbered 0 through 5. Assuming CBPADDR4's
directives were not intended, this patch removes them.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D54765
llvm-svn: 347351
This patch adjusts a test not to depend on deprecated FileCheck
behavior that permits overlapping matches within a block of CHECK-DAG
directives. Thus, this patch also removes uses of FileCheck's
-allow-deprecated-dag-overlap command-line option.
Specifically, the FileCheck variables DBG_LOC_START, DBG_LOC_END, and
DBG_LOC_CANCEL were all set to the same value. As a result, three
TERM_DEBUG-DAG patterns, one for each variable, all matched the same
text under the old overlapping behavior. Under the new
non-overlapping behavior, that's not permitted. This patch's solution
is to replace these variables with one variable and replace these
patterns with one pattern.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D54764
llvm-svn: 347350
If lambda is used inside of the OpenMP region and captures `this`, we
should recapture it in the OpenMP region also. But we should do this
only if the OpenMP region is used in the context of the same class, just
like the lambda.
llvm-svn: 347096
reductions.
Fixed previously committed code for the reduction support in
teams/parallel constructs taking into account new design of the NVPTX
support in the compiler. Teams reduction are not fully functional yet,
it is going to be fixed in the following patches.
llvm-svn: 347081
If the statements between target|teams|distribute directives does not
require execution in master thread, like constant expressions, null
statements, simple declarations, etc., such construct can be xecuted in
SPMD mode.
llvm-svn: 346551
target|teams|distribute variables.
If the total size of the variables, declared in target|teams|distribute
regions, is less than the maximal size of shared memory available, the
buffer is allocated in the shared memory.
llvm-svn: 346507
The base pointer for the lambda mapping must point to the lambda capture
placement and pointer must point to the captured variable itself. Patch
fixes this problem.
llvm-svn: 346408
Fixed lookup for the target regions in unused virtual functions + fixed
processing of the global variables not marked as declare target but
emitted during debug info emission.
llvm-svn: 346343
Coalesced memory access requires use of the new function
`__kmpc_data_sharing_coalesced_push_stack` instead of the
`__kmpc_data_sharing_push_stack`.
llvm-svn: 345991
The previously used combination `PTR_AND_OBJ | PRIVATE` could be used for mapping of some data in Fortran. Changed it to `PTR_AND_OBJ | LITERAL`.
llvm-svn: 345982
target/teams/distribute regions.
Target/teams/distribute regions exist for all the time the kernel is
executed. Thus, if the variable is declared in their context and then
escape it, we can allocate global memory statically instead of
allocating it dynamically.
Patch captures all the globalized variables in target/teams/distribute
contexts, merges them into the records, one per each target region.
Those records are then joined into the union, one per compilation unit
(to save the global memory). Those units are organized into
2 x dimensional arrays, where the first dimension is
the number of blocks per SM and the second one is the number of SMs.
Runtime functions manage this global memory space between the executing
teams.
llvm-svn: 345978
Alexey Bataev