Recent SteamVR versions have bugs with it, especially after triggering a
recenter operation.
In SteamVR, recentering fires referenceSpaceChangePending for the LOCAL
space, then the STAGE space, then the LOCAL space again, all with
different changeTimes. No poseInPreviousSpace is given.
Recreating the main reference space whenever this event is received
leads to strange, inconsistent issues. Sometimes the local/stage spaces
end up on top of each other, other times one or both will be way up in
the air (putting the headset at negative y coordinates).
This bug is even present when recentering in the compositor, so it's not
an issue with lovr. Cautiously disabling the local-floor emulation on
SteamVR runtimes and just always using the STAGE space until things are
sorted out.
The "vec3 is 4 floats" thing was consistently confusing to people. It's
reverted everywhere except for Curve.
maf now has full sets of methods for vec2/vec3/vec4, for consistency.
Vector bindings now use luax_readvec* helper functions for the
number/vector variants, and use maf for most functionality, which cleans
things up a lot.
Some compile fixes and a rename from gpu_wgpu to gpu_web, since wgpu
refers to a specific implementation of WebGPU and I'm really bad at
typing it for some reason.
- Adds Pass:setViewCull to enable/disable frustum culling.
- Renames Pass:setCullMode to Pass:setFaceCull (with backcompat).
Some stuff currently missing:
- Text is not culled, but should be.
- VR view frusta are not merged yet.
It's important that the bits for the vector type occupy the least
significant bits, so that vectors can be distinguished from pointer
lightuserdata.
When the vector pool was expanded, this broke, causing e.g. Blob
pointers to exhibit undefined behavior when trying to use them as
vectors.
tbh I still don't understand the union/bitfield memory layout.
Quest added a thing where they emulate grip pose when hand tracking is
active. This is actually pretty cool, and maybe LÖVR should do it too
on other runtimes, but it messed up the Quest hand mesh animation, for
some complicated reasons:
- Previously, getPose('hand/*') was returning the wrist pose because
LÖVR fell back to hand tracking data when the controller wasn't
tracked.
- Because of this, coupled with the fact that hand/controller models are
expected to be drawn at the hand pose, hand meshes were animated such
that the root node was located at the wrist pose.
- When Oculus added grip pose emulation for hand tracking, it caused a
discrepancy:
- Hand meshes were still being animated relative to their wrist pose
- getPose was now returning grip-style poses
- This resulted in hand meshes being off by approximately 90 degrees.
The fix is to locate skeletal joints relative to the grip pose when
animating Oculus hand meshes, and to place the origin/wrist at its real
pose instead of assuming it's the origin.
Origin type used to be a query-able property of the VR system that
indicated whether the tracking was roomscale or seated-scale.
The t.headset.offset config value could be used to design an
origin-agnostic experience, which by default shifted content up 1.7
meters when tracking was seated-scale. That way, stuff rendered at
y=1.7m was always at "eye level". It worked pretty well.
It's getting replaced with a t.headset.seated flag.
- If seated is false (the default), the origin of the coordinate space
will be on the floor, enabling the y=1.7m eye level paradigm. If
tracking is not roomscale, a floor offset of 1.7m will be emulated.
- If seated is true, the origin of the coordinate space will be y=0
at eye level (where the headset was when the app started). This is
the case on both roomscale and seated-scale tracking.
So basically 'seated' is an opt-in preference for where the app wants
its vertical origin to be.
One advantage of this is that it's possible to consistently get a y=0
eye level coordinate space, which was not possible before. This makes
it easier to design simpler experiences that only need to render a
floating UI and don't want to render a full environment or deal with
offsetting everything relative to a 'floor'. This also makes it easier
to implement hybrid VR+flatscreen experiences, because the camera is at
y=0 when the headset module is disabled.
The opt-in nature of the flag, coupled with the fact that it is
consistent across all types of tracking and hardware, is hopefully a
more useful design.
You can do lovr.headset.getPose('floor') to get the offset of the stage
relative to the local origin if you want to draw something at the center
of the play area.
Also lovr.headset.isTracked('floor') basically tells you if it's roomscale.
If the very first graphics-related thing done in a frame is drawing a
model, the reanimation logic would skip because a new frame hasn't
started yet. lovrModelAnimateVertices needs to unconditionally start a
new frame. (Previously, a new frame was guaranteed to be started
because all passes were temporary, but this is no longer the case).
- rm Pass:getTallyCount. It's unclear if this reports the current tally
count, or the number of tallies in the last submit. lovr was even
getting this confused internally (fixed).
- rm tally index argument from Pass:beginTally and Pass:finishTally.
The tally index is now an autoincremented value managed internally,
and both :beginTally/:finishTally return it. If someone wants to use
their own indices, a lookup table can be used to do the mapping.
- lovr.headset.getPassthrough returns current passthrough mode
- lovr.headset.setPassthrough sets the passthrough mode
- nil --> uses the default passthrough mode for the headset
- bool --> false = opaque, true = one of the transparent modes
- string --> explicit PassthroughMode
- lovr.headset.getPassthroughModes returns a table of supported modes
Creates a lightweight copy of a Model, for situations where a single
model needs to be rendered with multiple poses in a single frame, which
is currently not possible.
Enables automatic CPU/GPU timing for all passes. Defaults to true
when graphics debugging is active, but can be enabled/disabled manually.
When active, Pass:getStats will return submitTime and gpuTime table
keys, respectively indicating CPU time the Pass took to record and the
time the Pass took to run on the GPU. These have a delay of a few
frames.
This doesn't include a way to get "global" timing info for a submit.
This information would be useful because it doesn't require lovrrs to
sum all the timing info for all the passes and it would include other
work like transfers, synchronization, and CPU waits. However, this is
more challenging to implement under the current architecture and will be
deferred to later. Even if this were added, these per-pass timings will
remain useful.
- Make sure to reset barriers for compute/canvas resources too
- Delay stream ending so OpenXR layout transitions actually go in an
active command buffer.
If you switch to/from a compute shader and the other shader is either
nil or a graphics shader, clear bindings.
Maybe if you switch to/from nil the bindings shouldn't be cleared, but
this is a bit more complicated to implement and it's not clear that
there's any reason not to treat nil shaders as graphics shaders.
Previously, if you wanted to run compute operations that depend on the
results of prior compute operations, you had to put these in 2 different
passes, because logically all of the compute calls in a pass run "at the
same time" (or we're at least giving the GPU the freedom to do that).
Having to set up an entirely new pass just to synchronize 2 :compute
calls is pretty cumbersome, and incurs extra overhead. It would be
possible to change things so *every* :compute call waits for previous
computes to finish, but this would destroy GPU parallelism.
The Pass:barrier method lets compute calls within a pass synchronize
with each other, without requiring multiple passes. Adding a barrier
basically means "hey, wait for all the :compute calls before the barrier
to finish before running future :computes".
This lets things remain highly parallel but allows them to be easily
synchronized when needed.
Pass stores draw commands rather than sending them to Vulkan
immediately.
The main motivation is to allow more flexibility in the Lua API. Passes
are now regular objects, aren't invalidated whenever submit is called,
and can cache their draws across multiple frames. Draws can also be
internally culled, sorted, and batched.
Some API methods (tallies) are missing, and there are still some bugs to
fix, notably with background color.
View count is well-defined to be 2 with the current view configuration,
and people should be able to rely on getViewCount even before the views
are tracked. It returns the number of views in the view configuration,
not the number of views with valid data.
- If timestamp is zero (before .update is called), return empty data
instead of erroring.
- Check for valid position/orientation separately, and return empty data
for anything that's invalid. Previously both position/orientation
were used if either was valid, which returns undefined results.
- Add Buffer:newReadback
- Add Buffer:getData
- Buffer:getPointer works with permanent buffers
- Buffer:setData works with permanent buffers
- Buffer:clear works with permanent buffers
- Add Texture:newReadback
- Add Texture:getPixels
- Add Texture:setPixels
- Add Texture:clear
- Add Texture:generateMipmaps
- Buffer readbacks can now return tables in addition to Blobs using Readback:getData
Tally is coming back soon with an improved API, it's temporarily removed
since it made the transfer rework a bit easier.
Note that synchronous readbacks (Buffer:getData, Texture:getPixels)
internally call lovr.graphics.submit, so they invalidate existing Pass
objects. This will be improved soon.
If the number of skinned vertices in a Model doesn't fit in a single
dispatch (~2M vertices on 32-sized subgroups or ~500K vertices on
8-sized subgroups), split it into multiple dispatches.
- Pass:mesh accepts tables for vertices/indices
- Add Pass:setVertexFormat to set format used for table-based meshes
- Pass:send accepts tables for buffers
- Pass:send supports arbitrarily nested structs/arrays for push constants
- Buffer formats support arbitrarily nested structs/arrays
- Zero-length buffers are valid and represent structs
- Fields can have names using 'name'
- Field types can be tables of other fields (structs)
- Fields can have 'length' key
- newBuffer syntax has been changed to put format first (old version
still works)
- Buffers can be created from shader variables, avoiding need to declare
matching format.
- Pass:clear/Pass:read use byte offsets instead of indices
- Pass:copy uses byte offsets when copying a Buffer to a Buffer
- Deprecate lovr.graphics.getBuffer (tables can be used instead)
- core/spv just returns the type of image variables instead of trying to
validate them.
- When Shader is loading resources, it will reject combined image
samplers, uniform/texel buffers, and input attachments, with better
error messages that include the binding number of the invalid resource.
There were certain cases where an unrenderable texture view could be
created from a renderable parent texture (e.g. it has multiple mipmap
levels). This would leave renderView as NULL, which would cause a
crash.
If the target index is missing, the state will apply to all targets.
Fixes undefined behavior when setting color state in a pass with
multiple color attachments.
They are copied by value. Upon popping, they are pushed as temporary
vectors. Matrices are allocated on the heap, everything else is stored
in the Variant itself.
- Allow multisampled render pass to have a single-sample depth attachment
- Add a new depthResolve feature, indicating whether it's supported
- Fix stencil load/save
- Minor changes to render pass caching
- Currently the depth resolve is done using the first sample. A future
improvement would be to expose/use the min/max/average resolve modes.
- Put channel into thread module file.
- Make thread internals private.
- Handle more thread bookkeeping in thread module instead of Lua API.
- Fix a few race conditions/leaks nobody was probably ever going to hit.
Zip archives weren't enumerating in the root directory when they were
mounted with a non-empty mountpoint. Additionally, zips mounted at the
root directory weren't listing files properly. This fixes both by
normalizing the mountpoint prefix (it had a prepended slash when it was
empty, which messed up hashing), and ensuring there is a "root node" in
the tree with an empty string.
Although the name is unfortunate, this allows access to lovr.headset
when no window is opened or when the graphics module is disabled. This
requires the XR_MND_headless extension to be supported by the runtime.
Only when a readback is read back before a pass is created.
Should really change gpu to know if the frame has started yet and adjust
the tick index accordingly.
Some Android header defines DEPTH, which clashes with a symbol in the
OpenXR driver. This change just stops using Android headers in there
and declares more granular private functions. It also removes a few
unused private os functions.
ModelData:getTriangles currently adds a fresh set of vertices for every
mesh in a node. This is technically correct, but it wastes space when 2
nodes reference the same set of vertices with different index buffers,
which is pretty common when a node has multiple materials. It also
breaks ODE, who doesn't like it when vertices outnumber indices too
much.
- Add helper functions for creating shapes to avoid duplication between
newShape and newShapeCollider.
- Add lovr.physics.newMeshShape and lovr.physics.newTerrainShape
- Register TerrainShape so it has all the base Shape methods
- Smooth out a few TerrainShape warnings
Fixes easily-encounterable GPU OOM on discrete cards.
Currently when mapping CPU-accessible GPU memory, there are only two
types of memory: write and read.
The "write" allocations try to use the special 256MB pinned memory
region, with the thought that since this memory is usually for vertices,
uniforms, etc. it should be fast.
However, this memory is also used for staging buffers for buffers and
textures, which can easily exceed the 256MB (or 246MB on NV) limit upon
creating a handful of large textures.
To fix this, we're going to separate WRITE mappings into STREAM and
STAGING. STREAM will act like the old CPU_WRITE mapping type and use
the same memory type. STAGING will use plain host-visible memory and
avoid hogging the precious 256MB memory region.
STAGING also uses a different allocation strategy. Instead of creating
a big buffer with a zone for each tick, it's a more traditional linear
allocator that allocates in 4MB chunks and condemns the chunk if it ever
fills up. This is a better fit for staging buffer lifetimes since there's
usually a bunch of them at startup and then a small/sporadic amount
afterwards. The buffer doesn't need to double in size, and it doesn't
need to be kept around after the transfers are issued. The memory
really is single-use and won't roll over from frame to frame like the
other scratchpads.
The "system" button on Valve Index controller may not be exposed to
applications through OpenXR. Oculus runtime throws error when binding
for that button is attempted.
The animation compute shader was not specializing the workgroup size
properly, so it was only working on GPUs with a subgroup size of 32.
The Quest 1 has a subgroup size of 32 and the Quest 2 has a subgroup
size of 64, so this resulted in hand models breaking on Quest 2 only!
A null-char is valid part of Lua string. When such a string is sent
through the channel, its length should be stored as well to be able to
correctly reconstruct it on the other thread.
The bug was triggered with this code:
s1 = 'a \0 b'
print(#s1) -- 5
ch:push(s1)
s2 = ch:pop()
print(#s2) -- 2