The drain function was used only before the API gained the ability to
wait on the completion of a graph. This is the correct way to "drain"
the task executor of work.
Signed-off-by: Jeremy Ong <jcong@amazon.com>
* EnvironmentVariableHolder: reduce the size of template instantiation.
Move almost all destruction logic to EnvironmentVariableHolderBase::UnregisterAndDestroy.
Specialized templates have DestructDispatchNoLock instead that can either destroy the held value,
or the holder itself.
UnregisterAndDestroy has been moved to the cpp file.
All of these changes reduce the profile build time and size on linux
Here, the size of bin/profile goes down by ~200MB.
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
* Requested changes/fixups.
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
* Use scoped_lock to simplify mutex management.
Updated comments.
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
* Hopefully a fix for env variables released at a wrong time
Conditional was using incorrect variable
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
* Comment fixup
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
* Missing negation in conditional
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
* Cleanup the internal logic in UnregisterAndDestroy
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
Inner scope numEnvironmentVars was shadowing the outer scope, and prevented env variable memory from being freed.
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
In case an asset platform setting is missing it was supposed to be set to a 'pc' value.
Instead it was set to an empty string.
Signed-off-by: nemerle <96597+nemerle@users.noreply.github.com>
"If constexpr" branches are evaluated at template instantiation time,
but static assertions receiving false are triggered even earlier.
Signed-off-by: Jeremy Ong <jcong@amazon.com>
Also, came up with more useful benchmarks that actually measure the
enqueue/dequeue operations for various simple workflows. For retained
graphs, time-of-flight from submission to execution is ~1us per job,
indicating job granularity should be >20us for retained jobs. For
dynamic jobs, where we need to pay the cost of allocation, a granularity
of ~100+ us may be advised.
Signed-off-by: Jeremy Ong <jcong@amazon.com>
A few 'typedefs' replaced by 'using's
This shouldn't have any functional changes at all, just c++17 modernization
It's a part 6 of a split #2847
Signed-off-by: Nemerle <nemerle5+git@gmail.com>
A few 'typedefs' replaced by 'using's
This shouldn't have any functional changes at all, just c++17 modernization
It's a part 4 of a split #2847
Signed-off-by: Nemerle <nemerle5+git@gmail.com>
A few 'typedefs' replaced by 'using's
This shouldn't have any functional changes at all, just c++17 modernization
It's a part 3 of a split #2847
Signed-off-by: Nemerle <nemerle5+git@gmail.com>
A few 'typedefs' replaced by 'using's
This shouldn't have any functional changes at all, just c++17 modernization
It's a part 2 of a split #2847
Signed-off-by: Nemerle <nemerle5+git@gmail.com>
A few 'typedefs' replaced by 'using's
This shouldn't have any functional changes at all, just c++17 modernization
It's a part 1 of a split #2847
Signed-off-by: Nemerle <nemerle5+git@gmail.com>
Nemerle
hultonha
igarri
Jeremy Ong
Esteban Papp
pappeste
Artur K
Yuriy Toporovskyy
yuriy0