Port lawnjelly's dynamic BVH implementation from 3.x to be used in
both 2D and 3D broadphases.
Removed alternative broadphase implementations which are not meant to be
used anymore since they are much slower.
Includes changes in Rect2, Vector2, Vector3 that help with the template
implementation of the dynamic BVH by uniformizing the interface between
2D and 3D math.
Co-authored-by: lawnjelly <lawnjelly@gmail.com>
Use ThreadWorkPool to process physics step tasks in multiple threads. Collisions are all processed in parallel and solving impulses is
processed in parallel for rigid body islands.
Additional changes:
- Proper islands for soft bodies linked to active bodies
- All moving areas are on separate islands (can be parallelized)
- Fix inconsistencies with body islands (Kinematic bodies could link
bodies together or not depending on the processing order)
- Completely prevent static bodies to be active (it could cause islands
to be wrongly created and cause dangerous multi-threading operations as
well as inconsistencies in created islands)
- Apply impulses only on dynamic bodies to avoid unsafe multi-threaded
operations (static bodies can be on multiple islands)
- Removed inverted iterations when populating body islands, it's now
faster in regular order (maybe after fixing inconsistencies)
Several optimizations in the way solver islands are processed in both
2D and 3D physics:
- Use LocalVector instead of linked list to avoid cache misses (with
persistent storage based on worst case scenario)
- Remove pairs when setup fails (no valid contact) to avoid unnecessary
solving of non-colliding rigid bodies just to return immediately
In 3D, collision is disabled between kinematic/static bodies when
contacts are generated only to report them.
In 2D, this case was already fixed but the code is cleaned to make it
easier to follow.
- Based on C++11's `atomic`
- Reworked `SafeRefCount` (based on the rewrite by @hpvb)
- Replaced free atomic functions by the new `SafeNumeric<T>`
- Replaced wrong cases of `volatile bool` by the new `SafeFlag`
- Platform-specific implementations no longer needed
Co-authored-by: Hein-Pieter van Braam-Stewart <hp@tmm.cx>
This change makes test_body_motion more reliable when the kinematic body
recovers from being stuck.
- When recovery occurs, the rest information is generated, in order to
make sure collision results from test_move, move_and_collide and
move_and_slide are consistent and return a collision in case of overlap.
- The new calculation for recovery vector makes sure the recovery is
never more than the overlap depth between shapes.
This can help with cases where the kinematic body overlaps with several
shapes.
Recovery is made iteratively, without forcing a full overlap at each
step. This helps with getting proper rest information when recovery
occurs.
- One Way Collision:
When attempting motion, contact direction is checked against motion
before skipping in order to solve cases where kinematic bodies can sink
into one-way collision shapes.
Rest info now sets max contact depth in order to properly handle one-way
collision.
- Low speed motion is now handled in the rest info, by never setting
min_allowed_depth lower than motion length.
Separation is always applied with full margin, otherwise contact is lost
when low speed motion occurs right after higher speed motion.
- Similar changes are applied to 3D in order to make 2D and 3D
consistent.
-Rendering server now uses a split RID allocate/initialize internally, this allows generating RIDs immediately but initialization to happen later on the proper thread (as rendering APIs generally requiere to call on the right thread).
-RenderingServerWrapMT is no more, multithreading is done in RenderingServerDefault.
-Some functions like texture or mesh creation, when renderer supports it, can register and return immediately (so no waiting for server API to flush, and saving staging and command buffer memory).
-3D physics server changed to be made multithread friendly.
-Added PhysicsServer3DWrapMT to use 3D physics server from multiple threads.
-Disablet Bullet (too much effort to make multithread friendly, this needs to be fixed eventually).
This change allows collide_shape, intersect_shape, cast_motion and
rest_info in both 2D and 3D to ignore disabled shapes and make them
consistent with the physics simulation.
In some other cases, _cull_aabb_for_body is used and filters shapes out
internally, but whenever a physics query uses the broadphase directly
without calling _cull_aabb_for_body, disabled shapes can be returned
and need to be filtered out.
- Based on C++11's `thread` and `thread_local`
- No more need to allocate-deallocate or check for null
- No pointer anymore, just a member variable
- Platform-specific implementations no longer needed (except for the few cases of non-portable functions)
- Simpler for `NO_THREADS`
- Thread ids are now the same across platforms (main is 1; others follow)
For RigidBodies, uses the collision normal determined by relative motion
to determine whether or not a one-way collision has occurred.
For KinematicBodies, performs additional checks to ensure a one-way
collision has occurred, and averages the recovery step over all collision
shapes.
Co-authored-by: Sergej Gureev <sergej.gureev@relex.fi>
Happy new year to the wonderful Godot community!
2020 has been a tough year for most of us personally, but a good year for
Godot development nonetheless with a huge amount of work done towards Godot
4.0 and great improvements backported to the long-lived 3.2 branch.
We've had close to 400 contributors to engine code this year, authoring near
7,000 commit! (And that's only for the `master` branch and for the engine code,
there's a lot more when counting docs, demos and other first-party repos.)
Here's to a great year 2021 for all Godot users 🎆
- Fixes Godot physics failing when the cast Shape is inside of, or
already colliding with another Shape.
- Fixes Bullet physics failing when there is no motion.
- Ensures Godot and Bullet physics behave the same.
- Updates the documentation to exclude the caveats for the failures and
differences.
The thread model option for physics (2D) and rendering (single-unsafe,
single-safe, multithread), was causing crashes/locks when set as
multithreaded and exported for a platform that does not support threads
(namely HTML5).
This commit ensures that when threads support is not available, that
option is ignored, and the equivalent of "single-unsafe" is always used
instead.
I couldn't find a tool that enforces it, so I went the manual route:
```
find -name "thirdparty" -prune \
-o -name "*.cpp" -o -name "*.h" -o -name "*.m" -o -name "*.mm" \
-o -name "*.glsl" > files
perl -0777 -pi -e 's/\n}\n([^#])/\n}\n\n\1/g' $(cat files)
misc/scripts/fix_style.sh -c
```
This adds a newline after all `}` on the first column, unless they
are followed by `#` (typically `#endif`). This leads to having lots
of places with two lines between function/class definitions, but
clang-format then fixes it as we enforce max one line of separation.
This doesn't fix potential occurrences of function definitions which
are indented (e.g. for a helper class defined in a .cpp), but it's
better than nothing. Also can't be made to run easily on CI/hooks so
we'll have to be careful with new code.
Part of #33027.
Which means that reduz' beloved style which we all became used to
will now be changed automatically to remove the first empty line.
This makes us lean closer to 1TBS (the one true brace style) instead
of hybridating it with some Allman-inspired spacing.
There's still the case of braces around single-statement blocks that
needs to be addressed (but clang-format can't help with that, but
clang-tidy may if we agree about it).
Part of #33027.
Part of #33027, also discussed in #29848.
Enforcing the use of brackets even on single line statements would be
preferred, but `clang-format` doesn't have this functionality yet.
Configured for a max line length of 120 characters.
psf/black is very opinionated and purposely doesn't leave much room for
configuration. The output is mostly OK so that should be fine for us,
but some things worth noting:
- Manually wrapped strings will be reflowed, so by using a line length
of 120 for the sake of preserving readability for our long command
calls, it also means that some manually wrapped strings are back on
the same line and should be manually merged again.
- Code generators using string concatenation extensively look awful,
since black puts each operand on a single line. We need to refactor
these generators to use more pythonic string formatting, for which
many options are available (`%`, `format` or f-strings).
- CI checks and a pre-commit hook will be added to ensure that future
buildsystem changes are well-formatted.
EngineDebugger is the new interface to access the debugger.
It tries to be as agnostic as possible on the data that various
subsystems can expose.
It allows 2 types of interactions:
- Profilers:
A subsystem can register a profiler, assigning it a unique name.
That name can be used to activate the profiler or add data to it.
The registered profiler can be composed of up to 3 functions:
- Toggle: called when the profiler is activated/deactivated.
- Add: called whenever data is added to the debugger
(via `EngineDebugger::profiler_add_frame_data`)
- Tick: called every frame (during idle), receives frame times.
- Captures: (Only relevant in remote debugger for now)
A subsystem can register a capture, assigning it a unique name.
When receiving a message, the remote debugger will check if it starts
with `[prefix]:` and call the associated capture with name `prefix`.
Port MultiplayerAPI, Servers, Scripts, Visual, Performance to the new
profiler system.
Port SceneDebugger and RemoteDebugger to the new capture system.
The LocalDebugger also uses the new profiler system for scripts
profiling.
- Removed platform-specific implementations.
- Now all semaphores are in-object, unless they need to be conditionally created.
- Similarly to `Mutex`, provided a dummy implementation for when `NO_THREADS` is defined.
- Similarly to `Mutex`, methods are made `const` for easy use in such contexts.
- Language bindings updated: `wait()` and `post()` are now `void`.
- Language bindings updated: `try_wait()` added.
Bonus:
- Rewritten the `#ifdef` in `mutex.h` to meet the code style.
Main:
- It's now implemented thanks to `<mutex>`. No more platform-specific implementations.
- `BinaryMutex` (non-recursive) is added, as an alternative for special cases.
- Doesn't need allocation/deallocation anymore. It can live in the stack and be part of other classes.
- Because of that, it's methods are now `const` and the inner mutex is `mutable` so it can be easily used in `const` contexts.
- A no-op implementation is provided if `NO_THREADS` is defined. No more need to add `#ifdef NO_THREADS` just for this.
- `MutexLock` now takes a reference. At this point the cases of null `Mutex`es are rare. If you ever need that, just don't use `MutexLock`.
- Thread-safe utilities are therefore simpler now.
Misc.:
- `ScopedMutexLock` is dropped and replaced by `MutexLock`, because they were pretty much the same.
- Every case of lock, do-something, unlock is replaced by `MutexLock` (complex cases where it's not straightfoward are kept as as explicit lock and unlock).
- `ShaderRD` contained an `std::mutex`, which has been replaced by `Mutex`.
- Renames PackedIntArray to PackedInt32Array.
- Renames PackedFloatArray to PackedFloat32Array.
- Adds PackedInt64Array and PackedFloat64Array.
- Renames Variant::REAL to Variant::FLOAT for consistency.
Packed arrays are for storing large amount of data and creating stuff like
meshes, buffers. textures, etc. Forcing them to be 64 is a huge waste of
memory. That said, many users requested the ability to have 64 bits packed
arrays for their games, so this is just an optional added type.
For Variant, the float datatype is always 64 bits, and exposed as `float`.
We still have `real_t` which is the datatype that can change from 32 to 64
bits depending on a compile flag (not entirely working right now, but that's
the idea). It affects math related datatypes and code only.
Neither Variant nor PackedArray make use of real_t, which is only intended
for math precision, so the term is removed from there to keep only float.
Happy new year to the wonderful Godot community!
We're starting a new decade with a well-established, non-profit, free
and open source game engine, and tons of further improvements in the
pipeline from hundreds of contributors.
Godot will keep getting better, and we're looking forward to all the
games that the community will keep developing and releasing with it.
For clarity, assign-to-release idiom for PoolVector::Read/Write
replaced with a function call.
Existing uses replaced (or removed if already handled by scope)
