Summary:
We use this `dyn_ptr` argument in Clang/OpenMP to handle the
`KernelLaunchEnvironment`. This is a per-kernel argument used to share
some information. Currenetly, it's prepended to the argument list and we
generate storage for it in the runtime.
This is bad for a few reasons:
1. It changes the ABI by shifting user arguments
2. It cannot be trivially be left uninitialized if unused
3. The runtime must allocate its own memory for it
This PR changes it to be appended instead. Additionally, space for this
is always emitted. This means the OMPIRBuilder itself will provide the
storage, we simply need to populate it in the runtime if it is used.
This means that if it's unused we don't always pay the cost and it's
easier for non-OpenMP users to ignore it.
Backward compatibility is maintained by auto-upgrading the kernel
arguments. In `libomptarget` we completely allocate a new buffer to
store this in the new format. The plugins still need to respect the old
ABI of the called device object, so we simply rotate it if it's the old
version.
Make CreatePtrDiff() emit the pattern `ptrtoaddr(p1)-ptrtoaddr(p2)`.
This makes a few changes:
* The return type is now the address type instead of hardcoded to i64.
I've adjusted callers to deal with this where they didn't already.
* Don't use `ConstantExpr::getSizeOf()` and instead get the actual size
from DataLayout. These sizeof expressions will be removed as part of the
ptradd migration.
* Add a convenience overload without the element type, for the case
where you want a pure pointer difference.
I also adjusted some OpenMP code to consistently use zext for sizes, as
I had issues updating the test coverage otherwise (as we ended up
randomly picking zext or sext depending on the exact code path).
This adds support for using `ATTACH` map-type for proper
pointer-attachment when mapping list-items that have base-pointers.
For example, for the following:
```c
int *p;
#pragma omp target enter data map(p[1:10])
```
The following maps are now emitted by clang:
```
(A)
&p[0], &p[1], 10 * sizeof(p[1]), TO | FROM
&p, &p[1], sizeof(p), ATTACH
```
Previously, the two possible maps emitted by clang were:
```
(B)
&p[0], &p[1], 10 * sizeof(p[1]), TO | FROM
(C)
&p, &p[1], 10 * sizeof(p[1]), TO | FROM | PTR_AND_OBJ
````
(B) does not perform any pointer attachment, while (C) also maps the
pointer p, both of which are incorrect.
-----
With this change, we are using ATTACH-style maps, like `(A)`, for cases
where the expression has a base-pointer. For example:
```cpp
int *p, **pp;
S *ps, **pps;
... map(p[0])
... map(p[10:20])
... map(*p)
... map(([20])p)
... map(ps->a)
... map(pps->p->a)
... map(pp[0][0])
... map(*(pp + 10)[0])
```
#### Grouping of maps based on attach base-pointers
We also group mapping of clauses with the same base decl in the order of
the increasing complexity of their base-pointers, e.g. for something
like:
```
S **spp;
map(spp[0][0], spp[0][0].a), // attach-ptr: spp[0]
map(spp[0]), // attach-ptr: spp
map(spp), // attach-ptr: N/A
```
We first map `spp`, then `spp[0]` then `spp[0][0]` and `spp[0][0].a`.
This allows us to also group "struct" allocation based on their attach
pointers. This resolves the issues of us always mapping everything from
the beginning of the symbol `spp`. Each group is mapped independently,
and at the same level, like `spp[0][0]` and its member `spp[0][0].a`, we
still get map them together as part of the same contiguous struct
`spp[0][0]`. This resolves issue #141042.
#### use_device_ptr/addr fixes
The handling of `use_device_ptr/addr` was updated to use the attach-ptr
information, and works for many cases that were failing before. It has
to be done as part of this series because otherwise, the switch from
ptr_to_obj to attach-style mapping would have caused regressions in
existing use_device_ptr/addr tests.
#### Handling of attach-pointers that are members of implicitly mapped
structs:
* When a struct member-pointer, like `p` below, is a base-pointer in a
`map` clause on a target construct (like `map(p[0:1])`, and the base of
that struct is either the `this` pointer (implicitly or explicitly), or
a struct that is implicitly mapped on that construct, we add an implicit
`map(p)` so that we don't implicitly map the full struct.
```c
struct S { int *p;
void f1() {
#pragma omp target map(p[0:1]) // Implicitly map this->p, to ensure
// that the implicit map of `this[:]` does
// not map the full struct
printf("%p %p\n", &p, p);
}
```
#### Scope for improvement:
* We may be able to compute attach-ptr expr while collecting
component-lists in Sema.
* But we cache the computation results already, and `findAttachPtrExpr`
is fairly simple, and fast.
* There may be a better way to implement semantic expr comparison.
#### Needs future work:
* Attach-style maps not yet emitted for declare mappers.
* Mapping of class member references: We are still using PTR_AND_OBJ
maps for them. We will likely need to change that to handle
`ref_ptr/ref_ptee`, and `attach` map-type-modifier on them.
* Implicit capturing of "this" needs to map the full `this[0:1]` unless
there is an explicit map on one of the members, or a map with a member
as its base-pointer.
* Implicit map added for capturing a class member pointer needs to also
add a zero-length-array-section map.
* `use_device_addr` on array-sections-on-pointers need further
improvements (documented using FIXMEs)
#### Why a large PR
While it's unfortunate that this PR has gotten large and difficult to
review, the issue is that all the functional changes have to be made
together, to prevent regressions from partially implemented changes.
For example, the changes to capturing were previously done separately
(#145454), but they would still cause stability issues in absence of
full attach-mapping. And attach-mapping needs those changes to be able
to launch kernels.
We extracted the utilities and functions, like those for finding
attach-ptrs, or comparing exprs, out as a separate NFC PR that doesn't
call those functions, just adds them (#155625). Maybe the change that
adds a new error message for use_device_addr on array-sections with
non-var base-pointers could have been extracted out too (but that would
have had to be a follow-up change in that case, and we would get
comp-fails with this PR when the erroneous case was not
caught/diagnosed).
---------
Co-authored-by: Alex Duran <alejandro.duran@intel.com>
Add nuw attribute to inbounds GEPs where the expression used to form the
GEP is an addition of unsigned indices.
Relands #105496, which was reverted because it exposed a miscompilation
arising from #98608. This is now fixed by #106512.
Generate nuw GEPs for struct member accesses, as inbounds + non-negative
implies nuw.
Regression tests are updated using update scripts where possible, and by
find + replace where not.
This patch migrates the CGOpenMPRuntimeGPU::emitReduction and related functions to the OpenMPIRBUilder. In future patches MLIR OpenMP translation would be making use of these functions.
Co-authored-by: Jan Leyonberg <jan.leyonberg@amd.com>
A kernel implicit parameter (dyn_ptr) was introduced some time back.
This patch increments the kernel args version for a compiler supporting
dyn_ptr. The version will be used by the runtime to determine whether
the implicit parameter is generated by the compiler. The versioning is
required to support use cases where code generated by an older compiler
is linked with a newer runtime.
If approved, this patch should be backported to release 18.
Summary:
Currently, OpenMP handles the `omp requires` clause by emitting a global
constructor into the runtime for every translation unit that requires
it. However, this is not a great solution because it prevents us from
having a defined order in which the runtime is accessed and used.
This patch changes the approach to no longer use global constructors,
but to instead group the flag with the other offloading entires that we
already handle. This has the effect of still registering each flag per
requires TU, but now we have a single constructor that handles
everything.
This function removes support for the old `__tgt_register_requires` and
replaces it with a warning message. We just had a recent release, and
the OpenMP policy for the past four releases since we switched to LLVM
is that we do not provide strict backwards compatibility between major
LLVM releases now that the library is versioned. This means that a user
will need to recompile if they have an old binary that relied on
`register_requires` having the old behavior. It is important that we
actively deprecate this, as otherwise it would not solve the problem of
having no defined init and shutdown order for `libomptarget`. The
problem of `libomptarget` not having a define init and shutdown order
cascades into a lot of other issues so I have a strong incentive to be
rid of it.
It is worth noting that the current `__tgt_offload_entry` only has space
for a 32-bit integer here. I am planning to overhaul these at some point
as well.
Before we tracked the size of the teams reduction buffer in order to
allocate it at runtime per kernel launch. This patch splits the number
into two parts, the size of the reduction data (=all reduction
variables) and the (maximal) length of the buffer. This will allow us to
allocate less if we need less, e.g., if we have less teams than the
maximal length. It also allows us to move code from clangs codegen into
the runtime as we now know how large the reduction data is.
The alignment did likely not help much but increases the memory
requirement. Note that half of the affected accesses are all performed
by a single thread in each block. The reads are by consecutive threads
in a single block.
The KernelEnvironment is for compile time information about a kernel. It
allows the compiler to feed information to the runtime. The
KernelLaunchEnvironment is for dynamic information *per* kernel launch.
It allows the rutime to feed information to the kernel that is not
shared with other invocations of the kernel. The first use case is to
replace the globals that synchronize teams reductions with per-launch
versions. This allows concurrent teams reductions. More uses cases will
follow, e.g., per launch memory pools.
Fixes: https://github.com/llvm/llvm-project/issues/70249
We used to have two separate implementations to derive the number of
threads used in a target region. This lead us to sometimes miss out on
user provided thread bounds (num_threads, or thread_limit) when we
looked for "constant default values". If we might miss out on the
presence of those bounds, we cannot set the thread_limit statically
since the runtime will try to honor user input rather than cap it at the
"preferred default". This patch replaces the secondary implementation
with the primary in a mode that will not emit code but just look for the
presence, and potentially upper bounds, of thread limiting clauses.
The runtime test would not pass without this rewrite as we missed some
clauses, set the static limit on the device to the preferred value, but
then violated that value at runtime.
Fixes: https://github.com/llvm/llvm-project/issues/64845
Differential Revision: https://reviews.llvm.org/D158381
This allows use with non-0 address space stacks. llvm_ptr_ty should
never be used. This could use some more percolation up through mlir,
but this is enough to fix existing tests.
https://reviews.llvm.org/D156666
This patch introduces per kernel environment. Previously, flags such as execution mode are set through global variables with name like `__kernel_name_exec_mode`. They are accessible on the host by reading the corresponding global variable, but not from the device. Besides, some assumptions, such as no nested parallelism, are not per kernel basis, preventing us applying per kernel optimization in the device runtime.
This is a combination and refinement of patch series D116908, D116909, and D116910.
Depend on D155886.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D142569
This patch introduces per kernel environment. Previously, flags such as execution mode are set through global variables with name like `__kernel_name_exec_mode`. They are accessible on the host by reading the corresponding global variable, but not from the device. Besides, some assumptions, such as no nested parallelism, are not per kernel basis, preventing us applying per kernel optimization in the device runtime.
This is a combination and refinement of patch series D116908, D116909, and D116910.
Depend on D155886.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D142569
This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over
their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes
`IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to
`-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed
to `omp.is_target_device`. Getters and setters of all these renamed properties
are also updated accordingly. Many unit tests have been updated to use the new
names, but an alias for the `-fopenmp-is-device` option is created so that
external programs do not stop working after the name change.
`IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only
valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the
`-fopenmp-is-target-device` compiler frontend option, which is only added to
the OpenMP device invocation for offloading-enabled programs.
Differential Revision: https://reviews.llvm.org/D154591
The itanium ABI for certain platforms requires a minimum alignments for
member function pointers to reserve certain bits for distinguishing
virtual and non-virtual functions.
Our implementation of this however depends on the alignment of the
function involved, which may however not reflect the true alignment of
function pointers on certain targets for which the alignment is
independent of the function (e.g. AIX). Worse, the 2-byte alignment
we use may be less than the ABI minimum for the target, and in the case
we are using explicit sections will result in invalid codegen.
This patch attempts to correct this situation by considering the target
alignment of function pointers as part of making the decision about
whether we need to adjust the function alignment to conform to the ABI.
Targets which do not provide the function ptr alignment information
will return a value of 1 when queried and will conservatively retain
the old alignment.
Differential Revision: https://reviews.llvm.org/D147184
We use this helper to make several internal global variables during
codegen. currently we do not specify any alignment which allows the
alignment to be set incorrectly after some changes in how alignment was
handled. This patch explicitly aligns these variables to the natural
alignment as specified by the data layout
Fixes https://github.com/llvm/llvm-project/issues/62668
Reviewed By: tianshilei1992, gchatelet
Differential Revision: https://reviews.llvm.org/D150461
This reverts commit 35cfadfbe2decd9633560b3046fa6c17523b2fa9.
It makes a couple of buildbots unhappy because of the following test failures:
- `Transforms/OpenMP/add_attributes.ll'`
- `mapping/declare_mapper_target_data.cpp` on AMDGPU
This patch introduces per kernel environment. Previously, flags such as execution mode are set through global variables with name like `__kernel_name_exec_mode`. They are accessible on the host by reading the corresponding global variable, but not from the device. Besides, some assumptions, such as no nested parallelism, are not per kernel basis, preventing us applying per kernel optimization in the device runtime.
This is a combination and refinement of patch series D116908, D116909, and D116910.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D142569
This patch prefixes omp outlined helpers and reduction funcs
with the original function's name.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D140722
This patch attempts to prefix omp outlined helpers and reduction funcs
with the original function's name.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D140722
If an inlined kernel is called in a loop, the launch point alloca would
lead to increasing stack usage every time the kernel is invoked. This
could make the application run out of stack space and crash. This problem
is fixed by using the alloca insertion point while creating the alloca instruction.
Fixes https://github.com/llvm/llvm-project/issues/60602
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D145820
We already created a versioned `__tgt_kernel_arguments` struct but it
was only briefly used and its content was passed in isolation anyway.
This makes it hard to add more information in the future. With this
patch we fully embrace the struct as means to pass information from the
compiler to the plugin as part of a kernel launch.
The patch also extends and renames the struct, bumping the version
number to 2. Version 1 entries are auto-upgraded. This is in preparation
for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP
lowering, etc.
The `__tgt_target_kernel_nowait` interface was deprecated as it was
unused. Once we actually implement support for something like that, we
can add an appropriate API.
Note: Only plugins with the `launch_kernel` interface are now supported.
That means that a new clang won't be able to use an old runtime.
An old clang can still use the new runtime since the libomptarget
interface did not change.
Differential Revision: https://reviews.llvm.org/D141232
Parallel regions are outlined as functions with capture variables explicitly generated as distinct parameters in the function's argument list. That complicates the fork_call interface in the OpenMP runtime: (1) the fork_call is variadic since there is a variable number of arguments to forward to the outlined function, (2) wrapping/unwrapping arguments happens in the OpenMP runtime, which is sub-optimal, has been a source of ABI bugs, and has a hardcoded limit (16) in the number of arguments, (3) forwarded arguments must cast to pointer types, which complicates debugging. This patch avoids those issues by aggregating captured arguments in a struct to pass to the fork_call.
Reviewed By: jdoerfert, jhuber6, ABataev
Differential Revision: https://reviews.llvm.org/D102107
This is successor for D125291. This revision would try to use
@llvm.threadlocal.address in clang to access TLS variable. The reason
why the OpenMP tests contains a lot of change is that they uses
utils/update_cc_test_checks.py to update their tests.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D129833
Previously we added the `push_target_tripcount` function to send the
loop tripcount to the device runtime so we knew how to configure the
teams / threads for execute the loop for a teams distribute construct.
This was implemented as a separate function mostly to avoid changing the
interface for backwards compatbility. Now that we've changed it anyway
and the new interface can take an arbitrary number of arguments via the
struct without changing the ABI, we can move this to the new interface.
This will simplify the runtime by removing unnecessary state between
calls.
Depends on D128550
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D128816
This patch changes the code we generate to enter a target region on the
device. This is in-line with the new definition in the runtime that was
added previously. Additionally we implement this in the OpenMPIRBuilder
so that this code can be shared with Flang in the future.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D128550
Currently when using `atomic update` with floating-point variables, if
the operation is add or sub, `cmpxchg`, instead of `atomicrmw` is emitted, as
shown in [1]. In fact, about three years ago, llvm-svn: 351850 added the
support for FP operations. This patch adds the support in OpenMP as well.
[1] https://godbolt.org/z/M7b4ba9na
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D124724
This adds -no-opaque-pointers to clang tests whose output will
change when opaque pointers are enabled by default. This is
intended to be part of the migration approach described in
https://discourse.llvm.org/t/enabling-opaque-pointers-by-default/61322/9.
The patch has been produced by replacing %clang_cc1 with
%clang_cc1 -no-opaque-pointers for tests that fail with opaque
pointers enabled. Worth noting that this doesn't cover all tests,
there's a remaining ~40 tests not using %clang_cc1 that will need
a followup change.
Differential Revision: https://reviews.llvm.org/D123115
The EmitLoadOfPointer() call already specified the right pointer
type, but it did not match the Address we're loading from, so we
need to insert a bitcast first.
Rather than using a dummy void pointer type, we should specify the
correct private type and perform the bitcast beforehand rather than
afterwards. This way, the Address will have correct alignment
information.
Rather than specifying a dummy type in EmitLoadOfPointer() and
then casting it to the correct one, we should instead specify the
correct type and cast beforehand. Otherwise the computed alignment
will be incorrect.
