When building on linux and OPENGL_GLX OFF compilation fails with the
following error:
-----------------------------------------------------------------------
../../state_trackers/oxr/libst_oxr.a(oxr_session_gfx_gl.c.o): In function `oxr_session_populate_gl_xlib':
oxr_session_gfx_gl.c:(.text+0x5c): undefined reference to `xrt_gfx_provider_create_gl_xlib'
collect2: error: ld returned 1 exit status
src/xrt/targets/openxr/CMakeFiles/openxr_monado.dir/build.make:126: recipe for target 'src/xrt/targets/openxr/libopenxr_monado.so' failed
make[2]: *** [src/xrt/targets/openxr/libopenxr_monado.so] Error 1
CMakeFiles/Makefile2:2490: recipe for target 'src/xrt/targets/openxr/CMakeFiles/openxr_monado.dir/all' failed
make[1]: *** [src/xrt/targets/openxr/CMakeFiles/openxr_monado.dir/all] Error 2
Makefile:145: recipe for target 'all' failed
make: *** [all] Error 2
-----------------------------------------------------------------------
Apparently the gl_xlib backend really depends on GLX specifically, so
fix the issue by reflecting that in the conditionally compiled blocks.
This matches the OpenXR usage: the array is the plural of the element type,
and the count is the singular element type plus "count" (usually CountOutput
because of the two-call idiom)
Includes fixes to other code to match API changes.
We now have a cmake-format config file.
We no longer use list variables for sources, instead using
target_sources when we need to add, in accordance with current
best practice. (This makes it a lot easier to edit too.) There's no more
include_directories(), add_definitions(), or other gently-deprecated
directory-scoped commands, nor any CMake scripts that include
a parent directory reference (named targets instead)
Use a similar "hardcoded" idea as in p_factory_ensure_frameserver.
This fix usage of SLAM sources in other contexts like calibration, at the
cost of requiring a device to call create_tracked_slam at least once.
(again, similar to how psmv/psvr/hand tracking work already)
qwerty is auto probed, making HMDs that are not auto probed always take precedence.
When setting QWERTY_ENABLE=1 the intent is usually to exclusively use qwerty.
Therefore we default to disabling all other drivers when this variable is set.
To make the old behavior of adding qwerty devices with lower priority than actual
devices, the variable QWERTY_COMBINE=1 is introduced.
The hardcoded value 32 was actually wrong and caused a warning
../src/xrt/state_trackers/oxr/oxr_input.c:668:9: warning: 'oxr_binding_find_bindings_from_key' accessing 256 bytes in a region of size 128 [-Wstringop-overflow=]
668 | oxr_binding_find_bindings_from_key(log, profile, act->act_key, binding_points, &num);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
It produces a warning
../src/xrt/state_trackers/gui/gui_scene_debug.c:141:9: warning: ‘igInputFloat3’ accessing 12 bytes in a region of size 4 [-Wstringop-overflow=]
141 | igInputFloat3(name, (float *) &value.x, "%+f", ImGuiInputTextFlags_ReadOnly);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The STOPPING state has two possible follow up paths:
STOPPING -> IDLE -> READY
STOPPING -> IDLE -> EXITING
After EXITING, the application must call xrDestroySession; there is no meaningful
session state after EXITING.
To go to the READY state again, the application should first create a new session.
Applications that are lazy and drain the entire event queue and only handle
the last encountered state would be affected by "skipping" the EXITING state.
XR_KHR_vulkan_enable2:
physicalDevice VkPhysicalDevice must match the device specified by xrGetVulkanGraphicsDevice2KHR
XR_KHR_vulkan_enable:
physicalDevice VkPhysicalDevice must match the device specified by xrGetVulkanGraphicsDeviceKHR
XR_KHR_vulkan_enable:
Add a trivial check that xrGetVulkanGraphicsDeviceKHR is called before xrCreateSession.
(our cached suggested device will be XR_NULL_HANDLE if it has not been called).
The XR_KHR_vulkan_enable2 code path already contains this check.
It was necessary to add a list of xdevs to oxr_sdl2_hack_start and to
populate such list from its callees.
That includes sdl2_program.gui_program->xdevs which was not being filled
for the monado-service target.
The problem:
* xrCreateVulkanDeviceKHR is passed a VkPhysicalDevice, but not a VkInstance.
* xrGetVulkanGraphicsDevice2KHR is passed a VkInstance and returns a VkPhysicalDevice
that is a child of that instance.
* xrCreateVulkanDeviceKHR must verify that the xrGetVulkanGraphicsDevice2KHR
has been called and that the passed VkPhysicalDevice matches the one returned
by xrGetVulkanGraphicsDevice2KHR.
We have to consider:
* xrCreateVulkanDeviceKHR has to work on the "correct" VkInstance, which the passed
VkPhysicalDevice is a child of.
The reqirement
> If the vulkanPhysicalDevice parameter does not match the output of
> xrGetVulkanGraphicsDeviceKHR, then the runtime must return XR_ERROR_HANDLE_INVALID.
is not 100% clear whether calling xrCreateVulkanInstance multiple times is allowed
and how a second call to xrGetVulkanGraphicsDevice2KHR with a dfferent VkInstance
should be handled.
For this implementation xrCreateVulkanDeviceKHR will only consider the most recent call
to xrGetVulkanGraphicsDevice2KHR and the VkInstance that was used for this call.
This enforces at least that VkPhysicalDevice is a child of the cached VkInstance when
xrCreateVulkanDeviceKHR is called, because using a different VkPhysicalDevice would be
an error.
The generated files should be in auxiliary/bindings/*.{c,h}. For this to work
meson.build has to be in the bindings/ subdir:
https://github.com/mesonbuild/meson/issues/2320
Move CMakeLists.txt there too for some consistency.
Also fixes the previous include hack.
Move json profile generator to auxiliary/bindings and make generated_bindings static library.
aux/bindings: Add hardware type
aux/bindings: Add various steamvr specific things to bindings
Paths like "/input/menu/click" are used in SteamVR input profiles.
Subaction paths are the /user/X/Y part of the full path describing the input source/device.
"Subaction paths are a mechanism that enables applications to use the same action name and handle on multiple devices. Applications can query action state using subaction paths that differentiate data coming from each device."
Subpaths, are the input or output specific part of the full path, e.g.
"Each input source path must match the following pattern: …/input/<identifier>[_<location>][/<component>]"
hand_origin is confusing because it implies it is the origin of the coordinate system the hand is in.
It actually is the hand pose in the "global" coordinate system.
This commit is a band aid until a more proper room setup is implemented.
It allows moving the tracking offset for the device roles head, left and right by a fixed value.
A y tracking offset OXR_TRACKING_ORIGIN_OFFSET_Y=1.0 would tell monado that the current tracking
origin is 1 meter above the desired tracking origin, e.g. when the headset was calibrated to a
(0,0,0) position while sitting on table 1 meter above the floor.
This environment variable affects STAGE space, but not LOCAL space.
Rename XRT_FEATURE_OPENXR_LAYER_EQUIRECT_LEGACY to XRT_FEATURE_OPENXR_LAYER_EQUIRECT1.
Use correct define in verify_equirect1_layer function.
Rename equirect to equirect1.
meson: enable equirect1 by default.
Monado now calculates OpenHMD viewport dimensions and offsets correctly, we can pass those through.
The rotation of the rendered texture is performed with the distortion function.
v2: Use rotation matrix multiplication for distortion rotation
v3:
targets: Add Monado-SteamVR driver target
st/ovrd: Add OpenVR driver header
build: Factor out sdl hack into lib_sdl2_hack and update steamvr build
build: Revert lib_sdl2_refactor
steamvr: Emulate Index Controller by default
steamvr: Use oxr_handle_destroy instead of exposing oxr_instance_destroy
steamvr: don't use oxr internals
steamvr: communicate 3dof tracking to steamvr
steamvr: use util functions for device assignment and tracking origin setup
steamvr: Install plugin to <prefix>/share/steamvr-monado
steamvr: Use thread for updating poses every 1ms
Makes a big difference for the Index @144Hz on the vive driver.
Still somewhat choppy on survive driver - prediction should solve it.
Main-author: Christoph Haag <christoph.haag@collabora.com>
Co-author: Jakob Bornecrantz <jakob@collabora.com>
The compositor now moves immediately to visible/focused when polling.
The state tracker will generate relevant openxr state changes once the session is synchronized.
Properly working alternative to e03ee48dce (reverted in e7643de8db)
The application synchronizes its frame loop by "by calling xrWaitFrame,
xrBeginFrame and xrEndFrame in a loop."
Applications can discard frames by not calling xrEndFrame. If initial
frames are discarded, we should not consider the frame loop synchronized.
Previously a sequence like
xrBeginFrame, xrWaitFrame, xrBeginFrame, xrWaitFrame, xrPollEvent, xrBeginFrame
failed because xrPollEvent the compositor emitted transitions to visible
and focused but they were not emitted in the state tracker, because the
oxr session had not internally tranisitioned to the synchronized state.
Without this fix, compilation failed on Musl due to ssize_t being an
unknown type
In file included from ../src/xrt/state_trackers/prober/p_prober.c:15:
../src/xrt/state_trackers/prober/p_prober.h:67:2: error: unknown type name 'ssize_t'
67 | ssize_t interface;
| ^~~~~~~
