Previously, `olDestroyQueue` would not actually destroy the queue,
instead leaving it for the device to clean up when it was destroyed.
Now, the queue is either released immediately if it is complete or put
into a list of "pending" queues if it is not. Whenever we create a new
queue, we check this list to see if any are now completed. If there are
any we release their resources and use them instead of pulling from
the pool.
This prevents long running programs that create and drop many queues
without syncing them from leaking memory all over the place.
This is equivalent to `cuOccupancyMaxPotentialBlockSize`. It is
currently
only implemented on Cuda; AMDGPU and Host return unsupported.
---------
Co-authored-by: Callum Fare <callum@codeplay.com>
The purpose of this fence is to ensure that any `dataSubmit`s inserted
into a queue before a `dataFence` finish before finish before any
`dataSubmit`s
inserted after it begin.
This is a no-op for most queues, since they are in-order, and by design
any operations inserted into them occur in order.
But the interface is supposed to be functional for out-of-order queues.
The addition of the interface means that any operations that rely on
such ordering (like ATTACH map-type support in #149036) can invoke it,
without worrying about whether the underlying queue is in-order or
out-of-order.
Once a plugin supports out-of-order queues, the plugin can implement
this function, without requiring any change at the libomptarget level.
---------
Co-authored-by: Alex Duran <alejandro.duran@intel.com>
This sprinkles a few mutexes around the plugin interface so that the
olLaunchKernel CTS test now passes when ran on multiple threads.
Part of this also involved changing the interface for device synchronise
so that it can optionally not free the underlying queue (which
introduced a race condition in liboffload).
Add a device function to check if a device queue is empty. If liboffload
tries to create an event for an empty queue, we create an "empty" event
that is already complete.
This allows `olCreateEvent`, `olSyncEvent` and `olWaitEvent` to run
quickly for empty queues.
Enables AMD data center class GPUs to use memory manager memory pooling
up to 3GB allocation by default, up from the "1 << 13" threshold that
all plugin-nextgen devices use.
The following patch introduces a new interop interface implementation
with the following characteristics:
* It supports the new 6.0 prefer_type specification
* It supports both explicit objects (from interop constructs) and
implicit objects (from variant calls).
* Implements a per-thread reuse mechanism for implicit objects to reduce
overheads.
* It provides a plugin interface that allows selecting the supported
interop types, and managing all the backend related interop operations
(init, sync, ...).
* It enables cooperation with the OpenMP runtime to allow progress on
OpenMP synchronizations.
* It cleanups some vendor/fr_id mismatchs from the current query
routines.
* It supports extension to define interop callbacks for library cleanup.
When `unloadBinary` is called, any entries in the JITEngine's cache
for that binary will be cleared. This fixes a nasty issue with
liboffload program handles. If two handles happen to have had the same
address (after one was free'd, for example), the cache would be hit and
return the wrong program.
Summary:
We rely on these flags to do things in the runtime and print the
contents of binaries correctly. CUDA updated their ABI encoding recently
and we didn't handle that. it's a new ABI entirely so we just select on
it when it shows up.
Fixes: https://github.com/llvm/llvm-project/issues/148703
The `GlobalTy` helper has been extended to make both the Size and Ptr be
optional. Now `getGlobalMetadataFromDevice`/`Image` is able to write the
size of the global to the struct, instead of just verifying it.
This is a generated file which contains a macro for all Device Info
keys. This is visible to the plugin interface so that it can use the
definitions in a future patch.
The `unloadBinaryImpl` method on the host plugin is now implemented
properly (rather than just being a stub). When an image is unloaded,
it is deallocated and the library associated with it is closed.
GenericKernelTy has a pointer to the name that was used to create it.
However, the name passed in as an argument may not outlive the kernel.
Instead, GenericKernelTy now contains a std::string, and copies the
name into there.
This allows removal of a specific Image from a Device, rather than
requiring all image data to outlive the device they were created for.
This is required for `ol_program_handle_t`s, which now specify the
lifetime of the buffer used to create the program.
Previously, if a binary failed to load due to failures when jit
compiling, the function would return success with nullptr. Now it
returns a new plugin error, `COMPILE_FAILURE`.
Rather than being "stringly typed", store values as a std::variant that
can hold various types. This means that liboffload doesn't have to do
any string parsing for integer/bool device info keys.
Previously, device info was returned as a queue with each element having
a "Level" field indicating its nesting level. This replaces this queue
with a more traditional tree-like structure.
This should not result in a change to the output of
`llvm-offload-device-info`.
This pull request fixes coverage mapping on GPU targets.
- It adds an address space cast to the coverage mapping generation pass.
- It reads the profiled function names from the ELF directly. Reading it
from public globals was causing issues in cases where multiple
device-code object files are linked together.
Previously we decided to check in files that we generate with tablegen.
The justification at the time was that it helped reviewers unfamiliar
with `offload-tblgen` see the actual changes to the headers in PRs.
After trying it for a while, it's ended up causing some headaches and is
also not how tablegen is used elsewhere in LLVM.
This changes our use of tablegen to be more conventional. Where
possible, files are still clang-formatted, but this is no longer a hard
requirement. Because `OffloadErrcodes.inc` is shared with libomptarget
it now gets generated in a more appropriate place.
Summary:
We try to clamp these to ones known to work, but we should probably just
optimistically accept these. I'd prefer to update the flag check, but
since NVIDIA refuses to publish their ELF format it's too much effort to
reverse engineer.
Fixes: https://github.com/llvm/llvm-project/issues/138532
[Offload] Use new error code handling mechanism
This removes the old ErrorCode-less error method and requires
every user to provide a concrete error code. All calls have been
updated.
In addition, for consistency with error messages elsewhere in LLVM, all
messages have been made to start lower case.
A new ErrorCode enumeration is present in PluginInterface which can
be used when returning an llvm::Error from offload and PluginInterface
functions.
This enum must be kept up to sync with liboffload's ol_errc_t enum, so
both are automatically generated from liboffload's enum definition.
Some error codes have also been shuffled around to allow for future
work. Note that this patch only adds the machinery; actual error codes
will be added in a future patch.
~~Depends on #137339 , please ignore first commit of this MR.~~ This has
been merged.
Summary:
We treated the missing kernel environment as a unique mode, but it was
kind of this random bool that was doing the same thing and it explicitly
expects the kernel environment to be zero. It broke after the previous
change since it used to default to SPMD and didn't handle zero in any of
the other cases despite being used. This fixes that and queries for it
without needing to consume an error.
This pull request is the third part of an ongoing effort to extends PGO
instrumentation to GPU device code and depends on
https://github.com/llvm/llvm-project/pull/93365. This PR makes the
following changes:
- Allows PGO flags to be supplied to GPU targets
- Pulls version global from device
- Modifies `__llvm_write_custom_profile` and `lprofWriteDataImpl` to
allow the PGO version to be overridden
Summary:
This patch moves the RPC server handling to be a header only utility
stored in the `shared/` directory. This is intended to be shared within
LLVM for the loaders and `offload/` handling.
Generally, this makes it easier to share code without weird
cross-project binaries being plucked out of the build system. It also
allows us to soon move the loader interface out of the `libc` project so
that we don't need to bootstrap those and can build them in LLVM.
The module currently stores the target triple as a string. This means
that any code that wants to actually use the triple first has to
instantiate a Triple, which is somewhat expensive. The change in #121652
caused a moderate compile-time regression due to this. While it would be
easy enough to work around, I think that architecturally, it makes more
sense to store the parsed Triple in the module, so that it can always be
directly queried.
For this change, I've opted not to add any magic conversions between
std::string and Triple for backwards-compatibilty purses, and instead
write out needed Triple()s or str()s explicitly. This is because I think
a decent number of them should be changed to work on Triple as well, to
avoid unnecessary conversions back and forth.
The only interesting part in this patch is that the default triple is
Triple("") instead of Triple() to preserve existing behavior. The former
defaults to using the ELF object format instead of unknown object
format. We should fix that as well.
This pull request is the second part of an ongoing effort to extends PGO
instrumentation to GPU device code and depends on #76587. This PR makes
the following changes:
- Introduces `__llvm_write_custom_profile` to PGO compiler-rt library.
This is an external function that can be used to write profiles with
custom data to target-specific files.
- Adds `__llvm_write_custom_profile` as weak symbol to libomptarget so
that it can write the collected data to a profraw file.
- Adds `PGODump` debug flag and only displays dump when the
aforementioned flag is set
Summary:
If the user deallocates an RPC device this can sometimes fail if the RPC
server is still running. This will happen if the modification happens
while the server is still checking it. This patch adds a mutex to guard
modifications to it.
Summary:
This patch just changes the interface to make starting the thread
multiple times permissable since it will only be done the first time.
Note that this does not refcount it or anything, so it's onto the user
to make sure that they don't shut down the thread before everyone is
done using it. That is the case today because the shutDown portion is
run by a single thread in the destructor phase.
Another question is if we should make this thread truly global state,
because currently it will be private to each plugin instance, so if you
have an AMD and NVIDIA image there will be two, similarly if you have
those inside of a shared library.
Summary:
Pretty dumb mistake of me, forgot that this is run per-device and
per-plugin, which fell through the cracks with my testing because I have
two GPUs that use different plugins.
Summary:
Handling the RPC server requires running through list of jobs that the
device has requested to be done. Currently this is handled by the thread
that does the waiting for the kernel to finish. However, this is not
sound on NVIDIA architectures and only works for async launches in the
OpenMP model that uses helper threads.
However, we also don't want to have this thread doing work
unnnecessarily. For this reason we track the execution of kernels and
cause the thread to sleep via a condition variable (usually backed by
some kind of futex or other intelligent sleeping mechanism) so that the
thread will be idle while no kernels are running.
