| Commit message (Collapse) | Author | Age | Files | Lines |
| ... | |
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
operators where one type is a C complex type, and to emit both the
efficient and correct implementation for complex arithmetic according to
C11 Annex G using this extra information.
For both multiply and divide the old code was writing a long-hand
reduced version of the math without any of the special handling of inf
and NaN recommended by the standard here. Instead of putting more
complexity here, this change does what GCC does which is to emit
a libcall for the fully general case.
However, the old code also failed to do the proper minimization of the
set of operations when there was a mixed complex and real operation. In
those cases, C provides a spec for much more minimal operations that are
valid. Clang now emits the exact suggested operations. This change isn't
*just* about performance though, without minimizing these operations, we
again lose the correct handling of infinities and NaNs. It is critical
that this happen in the frontend based on assymetric type operands to
complex math operations.
The performance implications of this change aren't trivial either. I've
run a set of benchmarks in Eigen, an open source mathematics library
that makes heavy use of complex. While a few have slowed down due to the
libcall being introduce, most sped up and some by a huge amount: up to
100% and 140%.
In order to make all of this work, also match the algorithm in the
constant evaluator to the one in the runtime library. Currently it is
a broken port of the simplifications from C's Annex G to the long-hand
formulation of the algorithm.
Splitting this patch up is very hard because none of this works without
the AST change to preserve non-complex operands. Sorry for the enormous
change.
Follow-up changes will include support for sinking the libcalls onto
cold paths in common cases and fastmath improvements to allow more
aggressive backend folding.
Differential Revision: http://reviews.llvm.org/D5698
llvm-svn: 219557
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
It's possible to construct cases where the first field we are trying to
copy is in the middle of an IR field. In some complicated cases, we
would fail to use an appropriate offset inside the object. Earlier
builds of clang seemed to miscompile the code by copying an insufficient
number of bytes. Up until now, we would assert: the copying offset was
insufficiently aligned.
This fixes PR21232.
llvm-svn: 219524
|
| |
|
|
| |
llvm-svn: 219504
|
| |
|
|
|
|
| |
Also remove some unnecessary virtual keywords. NFC.
llvm-svn: 219497
|
| |
|
|
| |
llvm-svn: 219491
|
| |
|
|
|
|
| |
Moved CGOpenMPRegionInfo from CGOpenMPRuntime.h to CGOpenMPRuntime.cpp file and reworked the code for this change. Also added processing of ThreadID variable passed as an argument in outlined functions in parallel and task directives.
llvm-svn: 219490
|
| |
|
|
|
|
| |
This patch makes class OMPPrivateScope a common class for all private variables. Reworked processing of firstprivate variables (now it is based on OMPPrivateScope too).
llvm-svn: 219486
|
| |
|
|
|
|
| |
the try block.
llvm-svn: 219470
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Make it possible to pass NULL through variadic functions on 64-bit
Windows targets. The Visual C++ headers define NULL to 0, when they
should define it to 0LL on Win64 so that NULL is a pointer-sized
integer.
Fixes PR20949.
Reviewers: thakis, rsmith
Differential Revision: http://reviews.llvm.org/D5480
llvm-svn: 219456
|
| |
|
|
|
|
|
|
| |
Assertion failed: "Computed __func__ length differs from type!"
Reworked PredefinedExpr representation with internal StringLiteral field for function declaration.
Differential Revision: http://reviews.llvm.org/D5365
llvm-svn: 219393
|
| |
|
|
|
|
| |
Includes parsing and semantic analysis for 'omp teams' directive support from OpenMP 4.0. Adds additional analysis to 'omp target' directive with 'omp teams' directive.
llvm-svn: 219385
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
the current usage.
Summary:
The current code uses memset to re-initialize EHCleanupScope objects
with breaks the assumptions of the upcoming asan's intra-object-overflow checker.
If there is no DTOR, the new checker will refuse to work.
Test Plan: bootstrap with asan
Reviewers: rnk
Reviewed By: rnk
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D5656
llvm-svn: 219331
|
| |
|
|
|
|
| |
Revert this patch while I investigate some sanitizer failures off-line.
llvm-svn: 219307
|
| |
|
|
|
|
|
|
| |
This patch generates some helper variables that used as private copies of the corresponding original variables inside an OpenMP 'parallel' directive. These generated variables are initialized by copy using values of the original variables (with the copy constructor, if any). For arrays, initializator is generated for single element and in the codegen procedure this initial value is automatically propagated between all elements of the private copy.
In outlined function, references to original variables are replaced by the references to these private helper variables. At the end of the initialization of the private variables an implicit barier is generated by calling __kmpc_barrier(...) runtime function to be sure that all threads were initialized using original values of the variables.
Differential Revision: http://reviews.llvm.org/D5140
llvm-svn: 219306
|
| |
|
|
|
|
|
|
|
| |
Boostrapping LLVM+Clang+LLDB without threshold on object size for
lifetime markers insertion has shown there was no significant change
in compile time, so let the stack slot colorizer do its optimization
for all slots.
llvm-svn: 219303
|
| |
|
|
|
|
| |
Still troubles with OpenMP/parallel_firstprivate_codegen.cpp (now in ARM buildbots).
llvm-svn: 219298
|
| |
|
|
|
|
|
|
| |
This patch generates some helper variables that used as private copies of the corresponding original variables inside an OpenMP 'parallel' directive. These generated variables are initialized by copy using values of the original variables (with the copy constructor, if any). For arrays, initializator is generated for single element and in the codegen procedure this initial value is automatically propagated between all elements of the private copy.
In outlined function, references to original variables are replaced by the references to these private helper variables. At the end of the initialization of the private variables an implicit barier is generated by calling __kmpc_barrier(...) runtime function to be sure that all threads were initialized using original values of the variables.
Differential Revision: http://reviews.llvm.org/D5140
llvm-svn: 219297
|
| |
|
|
|
|
| |
To fix issues with test OpenMP/parallel_firstprivate_codegen.cpp
llvm-svn: 219296
|
| |
|
|
|
|
|
|
| |
This patch generates some helper variables that used as private copies of the corresponding original variables inside an OpenMP 'parallel' directive. These generated variables are initialized by copy using values of the original variables (with the copy constructor, if any). For arrays, initializator is generated for single element and in the codegen procedure this initial value is automatically propagated between all elements of the private copy.
In outlined function, references to original variables are replaced by the references to these private helper variables. At the end of the initialization of the private variables an implicit barier is generated by calling __kmpc_barrier(...) runtime function to be sure that all threads were initialized using original values of the variables.
Differential Revision: http://reviews.llvm.org/D5140
llvm-svn: 219295
|
| |
|
|
|
|
|
|
|
| |
semantic analysis for 'omp teams' directive support from OpenMP 4.0. Adds additional analysis to 'omp target' directive with 'omp teams' directive."
This reverts commit r219197 because it broke ARM self-hosting buildbots with
segmentation fault errors in many tests.
llvm-svn: 219289
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
Previously CodeGen assumed that static locals were emitted before they
could be accessed, which is true for automatic storage duration locals.
However, it is possible to have CodeGen emit a nested function that uses
a static local before emitting the function that defines the static
local, breaking that assumption.
Fix it by creating the static local upon access and ensuring that the
deferred function body gets emitted. We may not be able to emit the
initializer properly from outside the function body, so don't try.
Fixes PR18020. See also previous attempts to fix static locals in
PR6769 and PR7101.
Reviewers: rsmith
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D4787
llvm-svn: 219265
|
| |
|
|
| |
llvm-svn: 219258
|
| |
|
|
|
|
|
|
|
|
|
| |
We used to avoid these, but it looks like we did so just because we were
not handling dllexport alias correctly.
Dario Domizioli fixed that, so allow these aliases.
Based on a patch by Dario Domizioli!
llvm-svn: 219206
|
| |
|
|
|
|
| |
Includes parsing and semantic analysis for 'omp teams' directive support from OpenMP 4.0. Adds additional analysis to 'omp target' directive with 'omp teams' directive.
llvm-svn: 219197
|
| |
|
|
|
|
|
|
|
| |
No functional changes intended.
Renamed EmitOMPSimdLoop to EmitOMPInnerLoop, I plan to re-use
it to emit inner loop in the future patches for CodeGen of the
worksharing loop directives (omp for, omp for simd).
llvm-svn: 219195
|
| |
|
|
|
|
| |
operator as expected. NFC, should fix the MSVC build bots.
llvm-svn: 219116
|
| |
|
|
| |
llvm-svn: 219104
|
| |
|
|
|
|
|
|
| |
By leaving these members out of the member list, we avoid them being
emitted into type unit definitions - while still allowing the
definition/declaration to be injected into the compile unit as expected.
llvm-svn: 219101
|
| |
|
|
|
|
|
|
|
|
| |
list of class members
By leaving these members out of the member list, we avoid them being
emitted into type unit definitions - while still allowing the
definition/declaration to be injected into the compile unit as expected.
llvm-svn: 219100
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
This add support for the C++11 feature, thread_local global variables.
The ABI Clang implements is an improvement of the MSVC ABI. Sadly,
further improvements could be made but not without sacrificing ABI
compatibility.
The feature is implemented as follows:
- All thread_local initialization routines are pointed to from the
.CRT$XDU section.
- All non-weak thread_local variables have their initialization routines
call from a single function instead of getting their own .CRT$XDU
section entry. This is done to open up optimization opportunities to
the compiler.
- All weak thread_local variables have their own .CRT$XDU section entry.
This entry is in a COMDAT with the global variable it is initializing;
this ensures that we will initialize the global exactly once.
- Destructors are registered in the initialization function using
__tlregdtor.
Differential Revision: http://reviews.llvm.org/D5597
llvm-svn: 219074
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We already add the align parameter attribute for function parameters that have
the align_value attribute (or those with a typedef type having that attribute),
which is an important special case, but does not handle pointers with value
alignment assumptions that come into scope in any other way. To handle the
general case, emit an @llvm.assume-based alignment assumption whenever we load
the pointer-typed lvalue of an align_value-attributed variable (except for
function parameters, which we already deal with at entry).
I'll also note that this is more general than Intel's described support in:
https://software.intel.com/en-us/articles/data-alignment-to-assist-vectorization
which states that the compiler inserts __assume_aligned directives in response
to align_value-attributed variables only for function parameters and for the
initializers of local variables. I think that we can make the optimizer deal
with this more-general scheme (which could lead to a lot of calls to
@llvm.assume inside of loop bodies, for example), but if not, I'll rework this
to be less aggressive.
llvm-svn: 219052
|
| |
|
|
|
|
|
|
| |
single threaded systems
http://reviews.llvm.org/D4985
llvm-svn: 219027
|
| |
|
|
|
|
|
|
|
|
|
|
| |
When the aligned clause of an OpenMP simd pragma is not provided with an
explicit alignment, a target-dependent default must be used. This adds such a
default of PPC targets.
This will become slightly more complicated when BG/Q support is added (because
then it will depend on the type). For now, 16 is a correct value for all
systems, and covers Altivec and VSX vectors.
llvm-svn: 218994
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds support for the align_value attribute. This attribute is supported by
Intel's compiler (versions 14.0+), and several of my HPC users have requested
support in Clang. It specifies an alignment assumption on the values to which a
pointer points, and is used by numerical libraries to encourage efficient
generation of vector code.
Of course, we already have an aligned attribute that can specify enhanced
alignment for a type, so why is this additional attribute important? The
problem is that if you want to specify that an input array of T is, say,
64-byte aligned, you could try this:
typedef double aligned_double attribute((aligned(64)));
void foo(aligned_double *P) {
double x = P[0]; // This is fine.
double y = P[1]; // What alignment did those doubles have again?
}
the access here to P[1] causes problems. P was specified as a pointer to type
aligned_double, and any object of type aligned_double must be 64-byte aligned.
But if P[0] is 64-byte aligned, then P[1] cannot be, and this access causes
undefined behavior. Getting round this problem requires a lot of awkward
casting and hand-unrolling of loops, all of which is bad.
With the align_value attribute, we can accomplish what we'd like in a well
defined way:
typedef double *aligned_double_ptr attribute((align_value(64)));
void foo(aligned_double_ptr P) {
double x = P[0]; // This is fine.
double y = P[1]; // This is fine too.
}
This attribute does not create a new type (and so it not part of the type
system), and so will only "propagate" through templates, auto, etc. by
optimizer deduction after inlining. This seems consistent with Intel's
implementation (thanks to Alexey for confirming the various Intel-compiler
behaviors).
As a final note, I would have chosen to call this aligned_value, not
align_value, for better naming consistency with the aligned attribute, but I
think it would be more useful to users to adopt Intel's name.
llvm-svn: 218910
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Prior to GCC 4.4, __sync_fetch_and_nand was implemented as:
{ tmp = *ptr; *ptr = ~tmp & value; return tmp; }
but this was changed in GCC 4.4 to be:
{ tmp = *ptr; *ptr = ~(tmp & value); return tmp; }
in response to this change, support for sync_fetch_and_nand (and
sync_nand_and_fetch) was removed in r99522 in order to avoid miscompiling code
depending on the old semantics. However, at this point:
1. Many years have passed, and the amount of code relying on the old
semantics is likely smaller.
2. Through the work of many contributors, all LLVM backends have been updated
such that "atomicrmw nand" provides the newer GCC 4.4+ semantics (this process
was complete July of 2014 (added to the release notes in r212635).
3. The lack of this intrinsic is now a needless impediment to porting codes
from GCC to Clang (I've now seen several examples of this).
It is true, however, that we still set GNUC_MINOR to 2 (corresponding to GCC
4.2). To compensate for this, and to address the original concern regarding
code relying on the old semantics, I've added a warning that specifically
details the fact that the semantics have changed and that we provide the newer
semantics.
Fixes PR8842.
llvm-svn: 218905
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
Currently, with struct my_struct { int x; method_ptr y; };
a call to foo(my_struct s) may end up dropping the last 4 bytes
of the method pointer for x86_64 NaCl and x32.
When checking Has64BitPointers, also check if the method pointer
straddles an eightbyte boundary and classify Hi as well as Lo if needed.
Test Plan: test/CodeGenCXX/x86_64-arguments-nacl-x32.cpp
Reviewers: dschuff, pavel.v.chupin
Subscribers: jfb
Differential Revision: http://reviews.llvm.org/D5555
llvm-svn: 218889
|
| |
|
|
|
|
|
|
| |
Reapplying now that r218887 is in.
This reverts commit r218882, reapplying r218880.
llvm-svn: 218888
|
| |
|
|
|
|
|
|
| |
r218879 has been reverted for now, this needs to go to match.
This reverts commit r218880.
llvm-svn: 218882
|
| |
|
|
| |
llvm-svn: 218880
|
| |
|
|
|
|
|
| |
This will give more information to the optimizers so that they can reuse stack slots
and reduce stack usage.
llvm-svn: 218865
|
| |
|
|
|
|
|
| |
Update for corresponding LLVM API change for
`DIBuilder::createExpression()`.
llvm-svn: 218798
|
| |
|
|
|
|
|
|
|
|
| |
Complex address expressions are no longer part of DIVariable, but
rather an extra argument to the debug intrinsics.
http://reviews.llvm.org/D4919
rdar://problem/17994491
llvm-svn: 218788
|
| |
|
|
|
|
| |
"Update CGDebugInfo to the updated API in LLVM."
llvm-svn: 218781
|
| |
|
|
|
|
|
|
|
|
| |
Complex address expressions are no longer part of DIVariable, but
rather an extra argument to the debug intrinsics.
http://reviews.llvm.org/D4919
rdar://problem/17994491
llvm-svn: 218777
|
| |
|
|
|
|
| |
These were uncoveredby my yet undelivered patch.
llvm-svn: 218774
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This patch implements collapsing of the loops (in particular, in
presense of clause 'collapse'). It calculates number of iterations N
and expressions nesessary to calculate the nested loops counters
values based on new iteration variable (that goes from 0 to N-1)
in Sema. It also adds Codegen for 'omp simd', which uses
(and tests) this feature.
Differential Revision: http://reviews.llvm.org/D5184
llvm-svn: 218743
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When generating coverage regions, we were doing a linear search
through the existing regions in order to try to merge related ones.
Most of the time this would find what it was looking for in a small
number of steps and it wasn't a big deal, but in cases with many
regions and few mergeable ones this leads to an absurd compile time
regression.
This changes the coverage mapping logic to do a single sort and then
merge as we go, which is a bit simpler and about 100 times faster.
I've also added FIXMEs on a couple of behaviours that seem a little
suspect, while keeping them behaving as they were - I'll look into
these soon.
The test changes here are mostly tedious reorganization, because the
ordering of regions we output has become slightly (but not completely)
more consistent from the almost completely arbitrary ordering we got
before.
llvm-svn: 218738
|
| |
|
|
|
|
|
|
| |
This struct has some members that are accessed directly and others
that need accessors, but it's all just public. This is confusing, so
I've changed it to a class and made more members private.
llvm-svn: 218737
|
| |
|
|
| |
llvm-svn: 218697
|
| |
|
|
|
|
| |
Differential Revision: http://reviews.llvm.org/D5499
llvm-svn: 218660
|