Currently, we use an `UnrefQueue` which supposedly aims to transfer destruction costs to another thread. The implications of this unusual pattern can not be well understood because some allocators might free resources more efficiently if they are freed by the same thread that allocated them. In addition, `UnrefQueue` complicates the validation of thread safety in our engine. Lastly, our current usage of `UnrefQueue` triggers `ref()`, `unref()` atomic operations as objects are passed into the queue. These operations could be more expensive than the actual destruction.
With this PR we thus remove `UnrefQueue`. We can expect these changes to have a minor impact at most because we free most resources elsewhere in `ResourceSystem::updateCache`.
This variable is only used in ShadowsBin constructor and it's initialized each
time before constructor call so required value can be just passed into
ShadowsBin ctor.
Make ShadowsBin default constructor private because it is required by osg even
it's not actually called.
A component to load ESM content files with limited support for record types and
selection which of them to load. Supported record types are:
ACTI, CELL, CONT, DOOR, GMST, LAND, STAT.
Currently, we disable a paging root node that we only need in exterior cells by setting its node mask to 0 when transitioning into an interior cell. Node masks are not ideal for this usage case because Node::getBound is unaware of masks. With this PR we just detach the unused node from the scene graph. _shadowedScene->getBound() in the MWShadowTechnique should return a much better value in interior cells with these changes.
We currently build a large map of a BSAFile's contents unused by Open MW. We already map archive contents in VFS. With this PR we remove the map from BSAFile and reimplement its only current use in BSATool.
We currently use a version guard to adapt to a change in the number of parameters supplied to osg::TriangleFunctor's operator() template functor. The differing parameter is unused in our code. Crucially, operator() is not an override, so we can just add a default value for the differing parameter. Such a default allows us to apply identical code to both versions of the library without regressing functionality.
This PR proposes a simple change to `RigGeometry` `dirtyGLObjects` logic.
1. We will avoid dirtying unmodified arrays.
2. We can drop an osg version guard since `Drawable::dirtyGLObjects` is not nearly as harmful as `Geometry::dirtyGLObjects`.
3. We will avoid crashes in an as yet unfinished future PR concerning `Array` sharing improvements.
To reduce amount of computations on the caller side and restrict possible
values.
* verts can't be non-int because it's a number of things.
* worldsize is initially defined as int by ESM::Land::REAL_SIZE.
* Put function to calculate heightfied shift into components to be able to
reuse by other binaries.
osg::NodeVisitor is designed to recursively call virtual apply signatures until we find an implemented signature. Encountered nodes that we do not explicitely handle will trigger additional virtual function calls. With this PR we avoid these additional virtual function calls in the particularly costly ComputeLightSpaceBounds by adding some explicit signatures.
Currently, we always traverse the scene graph an additional time with a ComputeLightSpaceBounds visitor during shadow casting. ComputeLightSpaceBounds is only useful when the shadow casting mask allows us to shrink the bounds of the rendered scene, so we guard its traversal with a check against getCastsShadowTraversalMask. In practice, this guard never works because we build the traversal mask inclusively.
With this PR we limit the getCastsShadowTraversalMask check to relevant masks. This new check allows us to skip a superfluous ComputeLightSpaceBounds traversal with most settings.
Currently, Open MW's basic file access speed is limited by a peculiar layer of virtualisation in BSAFile's interface. This PR removes such virtualisation by properly separating BSAFile from CompressedBSAFile in low level contexts.
Tests which no longer work are commented out.
Some of these don't work because they're effectively testing for the
presence of bugs in the old implementation.
Others don't work because we're no longer accidentally disabling the
boost::program_options feature where it generates an error if only part
of a token gets consumed.
These will be fixed by later commits.
We skip this during node path iterations. this is not a node we are interested in.
We avoid allocating a new mGeomToSkelMatrix per frame and avoid a ref_ptr associated with its update.
We speed up a search for the Skeleton node by adding a continue; condition prior to an expensive dynamic_cast.
We can expect marginally improved loading times with this PR. Drawable, Transform and Node counts in stats panels are expected to remain unchanged - this PR does not add new scene graph optimisations, it just increases the speed with which we apply existing ones.
1. We add explicit `NodeVisitor::apply` overrides for commonly encountered node types to avoid additional virtual function calls per node associated with the default `apply` implementation.
2. We skip pushing `StateSet`s when `_mergeAlphaBlending` is enabled or the `StateSet` contains no relevant state.
3. We add a specialised variant of `CollectLowestTransformsVisitor::addTransform` accepting `MatrixTransform` to avoid matrix copies and multiplications.
With this PR we optimise a function that is called quite often when loading new cells.
We remove avoidable dynamic_casts.
We remove an unused pair.second element.
We convert a map to an unordered_map because its ordering is irrelevant in this case.
We avoid adding the root Skeleton node to the bones' node path.
Currently, we create a new ComputeLightSpaceBounds visitor per frame. Within this visitor, we require excessive memory allocations, mainly a new osg::RefMatrix per encountered Transform node.
With this PR we reuse a single ComputeLightSpaceBounds visitor across frames and enable the createOrReuseMatrix functionality to avoid allocating new matrices every frame. osgUtil::CullVisitor internally uses the same approach.
Currently, we run culling tests against all lights in the scene during LightListCallback::pushLightState. We can avoid most of these tests by removing off-screen lights at an earlier stage. We should benchmark the cumulative time spent within LightListCallback::pushLightState before and after this PR.
With this PR we test out osg's shader define system for a somewhat harmless feature. As we can see, our code becomes more concise and efficient in this case. Most importantly, we no longer create unneeded vertex shader objects.
Firstly, this PR reintroduces commit "Recreate a special case for IntersectionVisitor on QuadTreeWorld" we forgot to reapply while reverting a revert commit. Secondly, in cases we still need to build a view for an intersection visitor, we now use the available `osgUtil::IntersectionVisitor::getReferenceEyePoint` instead of falling back to the origin position that was previously causing long rebuild times.
`handle_stateset` is not needed because `UpdateMatrixTransform` is a `NodeCallback` only allowed to be set on a `Node`. `Geode` and `Drawable` do not need explicit logic because they are both derived from `Node`.
* avoids creating empty statesets on drawables
Currently, we attempt to skip creating state on drawable nodes when this state matches the default state. This attempt is incomplete because we still create an avoidable empty stateset in the default case.
* renderingmanager.cpp
* nifloader.cpp
* nifloader.cpp
* shadervisitor.cpp
* avoids creating empty statesets on drawables
Currently, we attempt to skip creating state on drawable nodes when this state matches the default state. This attempt is incomplete because we still create an avoidable empty stateset in the default case.
* renderingmanager.cpp
* nifloader.cpp
* avoids creating empty statesets for collada nodes
With this PR we avoid creating empty statesets for collada nodes which will be detrimental to osg's draw performance.
* scenemanager.cpp
We can take elsid's commit 605cb8d further by avoiding the terrain sync completely in most cases. Currently in changeCellGrid we wait for a new preloading task to ensure the getPagedRefnums for the new active cells have been filled in by object paging. This is usually not necessary because we have already completed a preload in the past containing these active cells. With this PR we remember what we preloaded and skip the terrain sync if it is not needed.
This reduces average time spent on in. 5 milliseconds as a base precision is
quite a lot considering that for 60 FPS frame time is 1000/16 = ~16.67 ms
when it's a cell loading frame and there is more important work to do rather
than sleeping.
Each binary depending on components library requires OSG plugins to be linked.
Duplicating dependecies for each binary does not give benefits and brings
problems when new binary is added.
Sample use case: users are using the Morrowind Code Patch feature that allows modders to enable this flag to differentiate editor-made potions from player crafted potions for tooltips.
CID 332936 (#1 of 1): Result is not floating-point (UNINTENDED_INTEGER_DIVISION)
integer_division: Dividing integer expressions size and 2, and then converting
the integer quotient to type float. Any remainder, or fractional part of the
quotient, is ignored.
* Change job change type to remove when tile is outside allowed range.
* Swap try number and change type in job priority. To make sure remove jobs
always processed before any other.
Define components_qt static library with all qt dependent components that also
depends on other components. Link only cs, wizard and launcher with qt
dependent components.
Even when target is not reachable actor will try to run there either because
target navmesh polygon is selected within extended area or because partial path
is built to the closest possible polygon.
This saves cache capacity when a scene contains objects contantly transforming
by scripts and causing changes in navmesh. The probability to get cache hit for
such states is almost zero because even a constant change in a single float
value may give up to 2^32 different states.
OpenCS - Fix moved reference - Issues #3514 and #4752
See merge request OpenMW/openmw!1051
(cherry picked from commit 2bee171c7990522da33c2667f7d079fa35f4ede0)
36c30f7f Fix for Issue #3514 where moving a reference to another cell is not handled properly.
40327681 Update the changelog.
(cherry picked from commit 496b3aef88b8fd867dcdd23a6ca43144573b1b2f)
Stereo friendly water
(cherry picked from commit 0e22c55e48a7f965367d3d430c1bef5357b22748)
Option to disable per view mapping.
Include memory header
De-hardcode settings and buffers.
formatting error
Update water.cpp (whitespace)
Update water.cpp (more whitespace)
include render order
c array -> c++ array
To make sure RecastMesh objects are equal if built with the same data but in
different order. Will be used later when there will be more than one place
building RecasMesh objects.
To reduce cache size and make it more flexible.
Adding off mesh connections to the navmesh is the last step of navmesh
generation and it's very fast comparing to other steps (microseconds vs
milliseconds). Having less cache size makes get and set operations almost 2x
times faster that also have an order of microseconds. So in total there is
no performance impact.
This is less error prone approach than use of MovedCellRef fields.
Also make separate functions for skipping and reading moved cell refs to avoid
passing special flags logic and null pointers for unused arguments.
Instead of explicit work queue stop before any possibly used engine manager
is destructed. Based on an assumption that any engine manager can be destructed
independently from the work queue destruction. This model is already used in
CellPreloader that conflicts with explicit work queue stop.
After the work queue is requested to be stopped, any client waiting for a not
started work item to be done will wait forever because the work item is dropped
from the queue. Therefore either clients should not wait for own work items to
be completed in destructor or the work queue should not drop items before
clients are destructed. Other approaches are possible but are not considered
due to increasing complexity.
CellPreloader already tries to wait for all created work items to be done so
keep it that way and extend the model to AsyncScreenCaptureOperation and Scene.
Additionally abort all scheduled work items when owner is destructed. This
prevents a long exit when multiple screenshots are scheduled right before
exiting the game.
Factoring common code parts outside of a template
is apparently a good practise to reduce code duplication
(and the size of openmw by around 0.5%),
and should improve a bit the performances,
since the whole `std::to_string` * 2 + string concatenation
dance results in quite a lot of code, preventing inlining on my machine.