Introduce ExportedComparatorTypeDesc/ComparatorLibDesc with inline
sanity checks, and add a core comparator shlib exporting three stub types.
Co-authored-by: Cursor <cursoragent@cursor.com>
We added a new centralized OpenCL Compute manager. This can later
be extended to support CUDA, SyCL, etc. SMO can be configured at
build time to choose which API it will use for compute.
Moreover, the ComputeMgr allows us to register buffers which are
available to all cl_contexts.
This symbol is defined as a static member object inside of a
boost detail header. When boost headers are used in a project
that uses Boost in both the main binary as well as dlopen()'d
shlibs, the top_ symbol gets duplicated and the metadata gets
partitioned.
We use the Boost shlib to unify both the main binary and the
shlibs to use the same memory address for top_.
This involves marking the templated object call_stack::top_ as
"extern" and then declaring to Boost that we intend to use the
shlibs.
This determines the maximum rate at which stimbuffs will be refreshed
with data from their device.
A device may refresh less frequently than this, but not more
frequently. The goal here is to give us control over the max
rate at which a device produces data.
We added a timestamp to each Lockvoker so that we can detect when
a lockvoker has been in a qutex for "too long", where "too long"
is defined arbitrarily as 500ms.
Next we're going to change the way we create callbacks to enable
us to more explicitly access the sh_ptr<AsyncContin> via
the callback object.
We now detect that a deadlock is likely when
CONFIG_DEBUG_QUTEX_DEADLOCK_TIMEOUT_MS has elapsed. This is the
preliminary work required to do a backtrace through the call
stack and figure out if a deadlock has really occured.
To do this, we'd have to go through the async call chain and
search for a previous caller which acquired the same qutex as
the one that first failed during this Lockvoker LockSet acquisition
attempt.
This makes the initialization sequence much cleaner and conceptually
well encapsulated.
We also now dynamically allocate the Mind objects. They're allocated
dynamically by Mrntt inside of initializeReq. This means that we no
longer have to worry about jolting and cleaning up the running threads
of global mind object even when we never explicitly called
Mind.initializeReq.
Along with other conceptual improvements to our abstractions, this
patch also gets us to a real "end of program initialization" point
for the first time.
Updated Boost dependency to version 1.73.0 to address segfault issues with boost::asio in dynamic libraries. Refactored heartbeat socket management to use raw UDP sockets instead of boost::asio, improving compatibility and error handling during socket operations.
This commit significantly restructures the way we setup threading in
SMO. We now don't use the CRT main() thread at all. It's only used
as a mechanism to ensure that Marionette doesn't execute before
global constructors have been executed.
JOLTing:
This is a simple ASIO post()ed message that makes each thread setup
its thread-local data pointer to its own ComponentThread object,
and then enter its main ASIO run() loop to await commands from
Marionette.
Exception bubbling:
We now cleanly cause mind threads to report their exceptions
to marionette, so that marionette can cleanly shut the mind down
in an orderly fashion.
Thread Control messaging API:
A namespace of asynchronous messages to be post()ed to threads to
control them. It enables us to pause and resume threads. This will
be very useful for Marionette when we add the ability for it to
suspend Salmanoff's running mind, inject new goals, inspect current
state, etc; and then resume the mind's execution.
hayodea
latentprion