- Skip generating profile data on `__global__` function in the host
compilation. It's a host-side stub function only and don't have
profile instrumentation generated on the real function body. The extra
profile data results in the malformed instrumentation profile data.
- Skip generating region mapping on functions in the wrong-side, i.e.,
+ For the device compilation, skip host-only functions; and,
+ For the host compilation, skip device-only functions (including
`__global__` functions.)
- As the device-side profiling is not ready yet, only host-side profile
code generation is checked.
Differential Revision: https://reviews.llvm.org/D85276
On the frontend side, this patch recovers AIX static init implementation to
use the linkage type and function names Clang chooses for sinit related function.
On the backend side, this patch sets correct linkage and function names on aliases
created for sinit/sterm functions.
Differential Revision: https://reviews.llvm.org/D84534
Add address space to indirect abi info and use it for kernels.
Previously, indirect arguments assumed assumed a stack passed object
in the alloca address space using byval. A stack pointer is unsuitable
for kernel arguments, which are passed in a separate, constant buffer
with a different address space.
Start using the new byref for aggregate kernel arguments. Previously
these were emitted as raw struct arguments, and turned into loads in
the backend. These will lower identically, although with byref you now
have the option of applying an explicit alignment. In the future, a
reasonable implementation would use byref for all kernel arguments
(this would be a practical problem at the moment due to losing things
like noalias on pointer arguments).
This is mostly to avoid fighting the optimizer's treatment of
aggregate load/store. SROA and instcombine both turn aggregate loads
and stores into a long sequence of element loads and stores, rather
than the optimizable memcpy I would expect in this situation. Now an
explicit memcpy will be introduced up-front which is better understood
and helps eliminate the alloca in more situations.
This skips using byref in the case where HIP kernel pointer arguments
in structs are promoted to global pointers. At minimum an additional
patch is needed to allow coercion with indirect arguments. This also
skips using it for OpenCL due to the current workaround used to
support kernels calling kernels. Distinct function bodies would need
to be generated up front instead of emitting an illegal call.
Summary:
Introduced OMPChildren class to handle all associated clauses, statement
and child expressions/statements. It allows to represent some directives
more correctly (like flush, depobj etc. with pseudo clauses, ordered
depend directives, which are standalone, and target data directives).
Also, it will make easier to avoid using of CapturedStmt in directives,
if required (atomic, tile etc. directives).
Also, it simplifies serialization/deserialization of the
executable/declarative directives.
Reduces number of allocation operations for mapper declarations.
Reviewers: jdoerfert
Subscribers: yaxunl, guansong, jfb, cfe-commits, sstefan1, aaron.ballman, caomhin
Tags: #clang
Differential Revision: https://reviews.llvm.org/D83261
The second argument of getNaturalAlignIndirect() was `bool ByRef`, but
the implementation was just delegating to getIndirect() with `ByRef`
passed unchanged to `bool ByVal` parameter of getIndirect().
Fix a couple of /*ByRef=*/ comments as well.
Reviewed By: rjmccall
Differential Revision: https://reviews.llvm.org/D85113
The _ExtInt(1) in getTypeForMem was hitting the bool logic for expanding
to an 8 bit value. The result was an assert, or store i1 %0, i8* %2, align 1
since the parameter IS an i1. This patch changes the 'forMem' test to
exclude ext-int from the bool test.
Without this patch, the following example fails but shouldn't
according to OpenMP TR8:
```
#pragma omp target enter data map(alloc:i)
#pragma omp target data map(present, alloc: i)
{
#pragma omp target exit data map(delete:i)
} // fails presence check here
```
OpenMP TR8 sec. 2.22.7.1 "map Clause", p. 321, L23-26 states:
> If the map clause appears on a target, target data, target enter
> data or target exit data construct with a present map-type-modifier
> then on entry to the region if the corresponding list item does not
> appear in the device data environment an error occurs and the
> program terminates.
There is no corresponding statement about the exit from a region.
Thus, the `present` modifier should:
1. Check for presence upon entry into any region, including a `target
exit data` region. This behavior is already implemented correctly.
2. Should not check for presence upon exit from any region, including
a `target` or `target data` region. Without this patch, this
behavior is not implemented correctly, breaking the above example.
In the case of `target data`, this patch fixes the latter behavior by
removing the `present` modifier from the map types Clang generates for
the runtime call at the end of the region.
In the case of `target`, we have not found a valid OpenMP program for
which such a fix would matter. It appears that, if a program can
guarantee that data is present at the beginning of a `target` region
so that there's no error there, that data is also guaranteed to be
present at the end. This patch adds a comment to the runtime to
document this case.
Reviewed By: grokos, RaviNarayanaswamy, ABataev
Differential Revision: https://reviews.llvm.org/D84422
This patch simplified IR generation for __builtin_btf_type_id().
For __builtin_btf_type_id(obj, flag), previously IR builtin
looks like
if (obj is a lvalue)
llvm.bpf.btf.type.id(obj.ptr, 1, flag) !type
else
llvm.bpf.btf.type.id(obj, 0, flag) !type
The purpose of the 2nd argument is to differentiate
__builtin_btf_type_id(obj, flag) where obj is a lvalue
vs.
__builtin_btf_type_id(obj.ptr, flag)
Note that obj or obj.ptr is never used by the backend
and the `obj` argument is only used to derive the type.
This code sequence is subject to potential llvm CSE when
- obj is the same .e.g., nullptr
- flag is the same
- metadata type is different, e.g., typedef of struct "s"
and strust "s".
In the above, we don't want CSE since their metadata is different.
This patch change IR builtin to
llvm.bpf.btf.type.id(seq_num, flag) !type
and seq_num is always increasing. This will prevent potential
llvm CSE.
Also report an error if the type name is empty for
remote relocation since remote relocation needs non-empty
type name to do relocation against vmlinux.
Differential Revision: https://reviews.llvm.org/D85174
This patch added the following additional compile-once
run-everywhere (CO-RE) relocations:
- existence/size of typedef, struct/union or enum type
- enum value and enum value existence
These additional relocations will make CO-RE bpf programs more
adaptive for potential kernel internal data structure changes.
For existence/size relocations, the following two code patterns
are supported:
1. uint32_t __builtin_preserve_type_info(*(<type> *)0, flag);
2. <type> var;
uint32_t __builtin_preserve_field_info(var, flag);
flag = 0 for existence relocation and flag = 1 for size relocation.
For enum value existence and enum value relocations, the following code
pattern is supported:
uint64_t __builtin_preserve_enum_value(*(<enum_type> *)<enum_value>,
flag);
flag = 0 means existence relocation and flag = 1 for enum value.
relocation. In the above <enum_type> can be an enum type or
a typedef to enum type. The <enum_value> needs to be an enumerator
value from the same enum type. The return type is uint64_t to
permit potential 64bit enumerator values.
Differential Revision: https://reviews.llvm.org/D83242
In order to follow NEC Aurora SX VE ABI correctly, change to sign/zero
extend integer arguments and return values smaller than 64 bits in clang.
Also update regression test.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D85071
Specified in https://github.com/WebAssembly/simd/pull/237, these
instructions load the first vector lane from memory and zero the other
lanes. Since these instructions are not officially part of the SIMD
proposal, they are only available on an opt-in basis via LLVM
intrinsics and clang builtin functions. If these instructions are
merged to the proposal, this implementation will change so that the
instructions will be generated from normal IR. At that point the
intrinsics and builtin functions would be removed.
This PR also changes the opcodes for the experimental f32x4.qfm{a,s}
instructions because their opcodes conflicted with those of the
v128.load{32,64}_zero instructions. The new opcodes were chosen to
match those used in V8.
Differential Revision: https://reviews.llvm.org/D84820
notail on x86-64
This is needed because the epilogue code inserted before tail calls on
x86-64 breaks the handshake between the caller and callee.
Calls to objc_retainAutoreleasedReturnValue used to have the same
problem, which was fixed in https://reviews.llvm.org/D59656.
rdar://problem/66029552
Differential Revision: https://reviews.llvm.org/D84540
Provides AMDGCN and NVPTX specific specialization of getGPUWarpSize,
getGPUThreadID, and getGPUNumThreads methods. Adds tests for AMDGCN
codegen for these methods in generic and simd modes. Also changes the
precondition in InitTempAlloca to be slightly more permissive. Useful for
AMDGCN OpenMP codegen where allocas are created with a cast to an
address space.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D84260
fptosi/fptoui have similar, but not identical, semantics. In
particular, the behavior on overflow is different.
Fixes https://bugs.llvm.org/show_bug.cgi?id=46844 for 64-bit. (The
corresponding patch for 32-bit is more involved because the equivalent
intrinsics don't exist, as far as I can tell.)
Differential Revision: https://reviews.llvm.org/D84703
We previously used a non-aggregate RValue to represent the passed value,
which violated the assumptions of call arg lowering in some cases, in
particular on 32-bit Windows, where we'd end up producing an FCA store
with TBAA metadata, that the IR verifier would reject.
When we use the OpenMPIRBuilder for the parallel region we need to also
use it to get the thread ID (among other things) in the body. This is
because CGOpenMPRuntime::getThreadID() and
CGOpenMPRuntime::emitUpdateLocation implicitly assumes that if they are
called from within a parallel region there is a certain structure to the
code and certain members of the OMPRegionInfo are initialized. It might
make sense to initialize them even if we use the OpenMPIRBuilder but we
would preferably get rid of such state instead.
Bug reported by Anchu Rajendran Sudhakumari.
Depends on D82470.
Reviewed By: anchu-rajendran
Differential Revision: https://reviews.llvm.org/D82822
We need to keep track of the alloca insertion point (which we already
communicate via the callback to the user) as we place allocas as well.
Reviewed By: fghanim, SouraVX
Differential Revision: https://reviews.llvm.org/D82470
Need to map the base pointer for all directives, not only target
data-based ones.
The base pointer is mapped for array sections, array subscript, array
shaping and other array-like constructs with the base pointer. Also,
codegen for use_device_ptr clause was modified to correctly handle
mapping combination of array like constructs + use_device_ptr clause.
The data for use_device_ptr clause is emitted as the last records in the
data mapping array.
Reviewed By: ye-luo
Differential Revision: https://reviews.llvm.org/D84767
Need to map the base pointer for all directives, not only target
data-based ones.
The base pointer is mapped for array sections, array subscript, array
shaping and other array-like constructs with the base pointer. Also,
codegen for use_device_ptr clause was modified to correctly handle
mapping combination of array like constructs + use_device_ptr clause.
The data for use_device_ptr clause is emitted as the last records in the
data mapping array.
It applies only for global pointers.
Differential Revision: https://reviews.llvm.org/D84767
Previously ctor homing was omitting debug info for classes if they
have both trival and nontrivial constructors, but we should only omit debug
info if the class doesn't have any trivial constructors.
retained types list.
bug: https://bugs.llvm.org/show_bug.cgi?id=46537
Differential Revision: https://reviews.llvm.org/D84870
This patch implements Clang front end support for the OpenMP TR8
`present` motion modifier for `omp target update` directives. The
next patch in this series implements OpenMP runtime support.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D84711
This patch introduces 2 new address spaces in OpenCL: global_device and global_host
which are a subset of a global address space, so the address space scheme will be
looking like:
```
generic->global->host
->device
->private
->local
constant
```
Justification: USM allocations may be associated with both host and device memory. We
want to give users a way to tell the compiler the allocation type of a USM pointer for
optimization purposes. (Link to the Unified Shared Memory extension:
https://github.com/intel/llvm/blob/sycl/sycl/doc/extensions/USM/cl_intel_unified_shared_memory.asciidoc)
Before this patch USM pointer could be only in opencl_global
address space, hence a device backend can't tell if a particular pointer
points to host or device memory. On FPGAs at least we can generate more
efficient hardware code if the user tells us where the pointer can point -
being able to distinguish between these types of pointers at compile time
allows us to instantiate simpler load-store units to perform memory
transactions.
Patch by Dmitry Sidorov.
Reviewed By: Anastasia
Differential Revision: https://reviews.llvm.org/D82174
Instead, pattern match extends of extract_subvectors to generate
widening operations. Since extract_subvector is not a legal node, this
is implemented via a custom combine that recognizes extract_subvector
nodes before they are legalized. The combine produces custom ISD nodes
that are later pattern matched directly, just like the intrinsic was.
Also removes the clang builtins for these operations since the
instructions can now be generated from portable code sequences.
Differential Revision: https://reviews.llvm.org/D84556
This patch implements Clang front end support for the OpenMP TR8
`present` motion modifier for `omp target update` directives. The
next patch in this series implements OpenMP runtime support.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D84711
types.
We previously did not treat a function type as dependent if it had a
parameter pack with a non-dependent type -- such a function type depends
on the arity of the pack so is dependent even though none of the
parameter types is dependent. In order to properly handle this, we now
treat pack expansion types as always being dependent types (depending on
at least the pack arity), and always canonically being pack expansion
types, even in the unusual case when the pattern is not a dependent
type. This does mean that we can have canonical types that are pack
expansions that contain no unexpanded packs, which is unfortunate but
not inaccurate.
We also previously did not treat a typedef type as
instantiation-dependent if its canonical type was not
instantiation-dependent. That's wrong because instantiation-dependence
is a property of the type sugar, not of the type; an
instantiation-dependent type can have a non-instantiation-dependent
canonical type.
This patch implements Clang front end support for the OpenMP TR8
`present` map type modifier. The next patch in this series implements
OpenMP runtime support.
This patch does not attempt to implement TR8 sec. 2.22.7.1 "map
Clause", p. 319, L14-16:
> If a map clause with a present map-type-modifier is present in a map
> clause, then the effect of the clause is ordered before all other
> map clauses that do not have the present modifier.
Compare to L10-11, which Clang does not appear to implement yet:
> For a given construct, the effect of a map clause with the to, from,
> or tofrom map-type is ordered before the effect of a map clause with
> the alloc, release, or delete map-type.
This patch also does not implement the `present` implicit-behavior for
`defaultmap` or the `present` motion-modifier for `target update`.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D83061
Pass LowerMatrixIntrinsics wasn't running yet running under the new pass
manager, and this adds LowerMatrixIntrinsics to the pipeline (to the
same place as where it is running in the old PM).
Differential Revision: https://reviews.llvm.org/D84180
Reapply 49e5f603d4
which had been reverted in c94332919b.
Originally reverted because I hadn't updated it in quite a while when I
got around to committing it, so there were a bunch of missing changes to
new code since I'd written the patch.
Reviewers: aaron.ballman
Differential Revision: https://reviews.llvm.org/D76646
Sometimes we also want to avoid merging inline assembly. This patch add
the nomerge function attribute to inline assembly.
Reviewed By: zequanwu
Differential Revision: https://reviews.llvm.org/D84225
Summary:
Need to avoid an optimization for base pointer mapping for target data
directives.
Reviewers: jdoerfert, ye-luo
Subscribers: yaxunl, guansong, cfe-commits, sstefan1, caomhin
Tags: #clang
Differential Revision: https://reviews.llvm.org/D84182
OptNoneInstrumentation is part of StandardInstrumentations. It skips
functions (or loops) that are marked optnone.
The feature of skipping optional passes for optnone functions under NPM
is gated on a -enable-npm-optnone flag. Currently it is by default
false. That is because we still need to mark all required passes to be
required. Otherwise optnone functions will start having incorrect
semantics. After that is done in following changes, we can remove the
flag and always enable this.
Reviewed By: ychen
Differential Revision: https://reviews.llvm.org/D83519
Use common interface for accessing target specific GPU grid values in NVPTX
OpenMP codegen as proposed in https://reviews.llvm.org/D80917
Originally authored by Greg Rodgers (@gregrodgers).
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D83492
The `intrinsics_gen` target exists in the CMake exports since r309389
(see LLVMConfig.cmake.in), hence projects can depend on `intrinsics_gen`
even it they are built separately from LLVM.
Reviewed By: MaskRay, JDevlieghere
Differential Revision: https://reviews.llvm.org/D83454
On AIX, the semantic of global_dtors contains __sterm functions associated with C++
cleanup actions and user-declared __attribute__((destructor)) functions. We should
never merely register __sterm with atexit(), so currently
-fregister_global_dtors_with_atexit does not work well on AIX: It would cause
finalization actions to not occur when unloading shared libraries. We need to figure
out a way to handle that when we start supporting user-declared
__attribute__((destructor)) functions.
Currently we report_fatal_error on this option temporarily.
Differential Revision: https://reviews.llvm.org/D83974
Refactors CGOpenMPRuntimeNVPTX as CGOpenMPRuntimeGPU to make it a
generalization for OpenMP GPU Codegen. Target specific specialized
methods for NVPTX are defined in class CGOpenMPRuntimeNVPTX. This
paves the way for a clean and maintainable extension to more GPU
targets for OpenMP Codegen.
For original author (git blame) list of CGOpenMPRuntimeGPU code,
look in history of CGOpenMPRuntimeNVPTX.cpp and .h, after this commit.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D83723
Michael Liao