(Internal change: 1663209)

(Internal change: 1663368)

(Internal change: 1663496)

…rprimPath

separately. Also, the instanceContext is now returned as a vector of paths instead of
a vector of UsdPrims, since Hd/SceneDelegate can't know about USD.

In the case where the last instance root itself is the rprim, the returned rprimPath will
be empty. The last path in instanceContext will contain the rprim path.

Updated unit test to exercise the modifications to the API.

(Internal change: 1663831)

(Internal change: 1663896)

(Internal change: 1663953)

…prim::Sync so it would first try to get the shader from the delegate and then try to use that shader. This change surfaced an issue down in the UsdImaging where it would not know how to resolve the shader when the instance was a transform (for instance).

This change fix the instanceAdapter so it will first try to use any shader in the instance and then use what ever was configured in the master.

(Internal change: 1664329)

adding minimal GL window utilitiy for unit tests.
(not implemeneted yet on macos)

(Internal change: 1664362)

Node-culling is a crucial optimization in Pcp composition that
is a win against the natural tendency for scene structure to
accumulate down namespace.

The basic idea is that, as we walk down namespace, we can prune
off branches of the composition structure as soon as we know
they contribute no further opinions.

However, the rules for which nodes we can cull are subtle due
to certain features that require preserving empty structure.
But it turns out we can be more aggressive with variant culling
than we had been.

(Internal change: 1664515)

(Internal change: 1664537)

remove Qt dependency from Hd and Hdx.
unit tests use GarchGLDebugWindow instead.

(Internal change: 1664580)

Replace TF_FOR_ALL's with C++11 range-for (limited to clips specific files,
stage and few others need this treatment, but its outside the scope of this change, methinks.)
Change the TimeMapping obj from a std::pair<..> typedef to a struct with named members, update
all usage points. Condense some time mapping code in clip.cpp. Put large debug notice into function instead
of rewriting it in 5 or so places. Lastly, this adds two sentinel
values, Usd_ClipTimesBegin/End which are aliases for std::numeric_limits<double>::max()(-/+)
which are authored silently at the end of the clipTimes range.

(Internal change: 1664596)

remove a stray endif(), inadvertently added at the previous change

(Internal change: 1664655)

The Hydra mesh Rprim sets up a copy computation to combine vertex buffers that have different formats.  When were running the computation, we discovered that the source of the copy didn't have a GPU resource allocated to it.  The reason why their was no GPU resource is because the points array was empty, so because of that the packed smooth normals was producing no output and the memory manager was determing the size was 0, so didn't provide a resource for it.

This all happens in a deferred way, so we don't know at the time of adding the computation that the result is going to be empty, therefore the copy computation is seeing these empty unallocated buffers at execution time.

Changes:
- Reordered the copy computation execution function, so that the empty check comes before the is resource allocated check.

(Internal change: 1664714)

implements GarchGLDebugWindow on osx (still quite minimal)

(Internal change: 1664718)

(Internal change: 1664738)

one step forward to remove GL fixed pipeline functions.
(for OSX port)

(Internal change: 1664742)

Changes:
- Added blendWireframeColor to HdMeshReprDesc.
- This flag, then passed to the Mesh Shader Key which uses it to choose the appropriate shader fragment.
- Modified the fullback value for GetWireframeColor to 0,0,0,0.
- Renamed shader fragment MeshWire.Fragment.EdgeOnly to MeshWire.Fragment.EdgeOnlyBlendColor.
- Added new shader fragment MeshWire.Fragment.EdgeOnlyNoBlend, which removes the lerp.
- Updated testHdCodeGen and testHdDrawBatching to support new shader key parameters.
- Added new subtests to testHdCodeGen for edgeonly fragments.
- Updated baselines for testHdCodeGen for new default Wireframe color.

(Internal change: 1664785)

use GarchGLDebugWindow instead of QGLWidget for the unit tests.

(Internal change: 1664839)

a workaround for some modifier key symbols are inconsistent
between SDK versions.

also fixing typo of BOOL/bool causing a warning.

(Internal change: 1664853)

commit e212084 was too hasty to remove fixed GL matrix stack,
which is still used for axis/bbox drawing.
Partially revert the change so that Hydra drawing is taking the new
path with other stuffs working.

(Internal change: 1665023)

(Internal change: 1665049)

(Internal change: 1665057)

…e don't crash when binding lights and the shadow pass hadn't been run due to some other error.

(Internal change: 1665181)

glfq is the last Qt C++ dependency in pxr. In pxr, it's being used for enabling debug context
in usdview. it's useful, but we think it doesn't worth to ask Qt dependency just for it,
and we have another way to debug GL errors such as API trace and GL debugger.

For GL debugging in usdview, we'll revisit to add that functionality in another way later on.

(Internal change: 1665204)

…re that has not been correctly inserted. Also, this change makes it so hd/surfaceshader does not insert textures in the registry, this is only done by hd/texture.

[Hd] Surface shader now does not insert textures in the registry, it only checks if they are there. There is also better error handling for missing textures in SurfaceShaders. In order to support this changes it has been added a functionality in the instanceRegistry to check if an instance is there or not without registering it.

(Internal change: 1665282)

It's failing because currently OPENSUBDIV_VERSION = v3_1_0_RC2
and that is parsed as version 2. fixing to see the first 3 numbers only.

Fixes PixarAnimationStudios#82

(Internal change: 1665286)

In the case of single layers, this seems restrictive, but in the general
case, it protects users from overwriting a thin root layer with a composed
stage.

(Internal change: 1665421)

…uld contain redundant commas and spaces. A good example of this issue is in the documentation string of certain properties(eg. proxyPrim).

(Internal change: 1665541)

…ons (create textures...) if they are needed

[Hd] First check if we need to do anything before we do any operation on the texture

(Internal change: 1665615)

…ing output.

(Internal change: 1665639)

…/USD into dev_win_ip_g

…to dev_win_ip_g

Bringing dev_wip_ip_g up to date with dev

…ed it exported for usdImagingGL.

This change provides Hydra support for reloading shader programs for Surface and Geometry shaders.  It doesn't currently support Lighting, Render Pass and Culling shaders.  The engine marks all prims as dirty to trigger an re-evaluation of the shader state, marks all surface shaders as dirty to cause those prims to be synced.  It also removes all the geometric shader instances from the GeometricShaderRegistry.  Finally it causes the Fallback surface shader to be reloaded.
Reloading Renderpass shader is now partially supported as we will correctly compile and re-link the shader if the contents of the file changes, but I'm not currently reloading the glslfxFile.

Changes:
- Added ReloadAllShaders that dirties state so it gets reevaluated and removes cache entries from the resource registry.
- Added ability to remove all entries from the HdInstanceRegistry.
- The Render Index can now reload the fallback surface shader, but creating an new one and swapping it with the old one.
- Renderpass shader now uses the hash of the contents of the file rather than a hash of the filename.

(Internal change: 1665781)

(Internal change: 1665957)

(Internal change: 1665958)

(Internal change: 1665959)

(Internal change: 1665960)

(Internal change: 1665961)

(Internal change: 1665962)

(Internal change: 1665963)

(Internal change: 1665964)

(Internal change: 1665965)

(Internal change: 1665966)

(Internal change: 1665967)

(Internal change: 1665968)

(Internal change: 1665969)

(Internal change: 1665970)

(Internal change: 1665971)

(Internal change: 1666059)

(Internal change: 1666060)

(Internal change: 1666067)

…sdin

(Internal change: 1666142)

(Internal change: 1666189)

(Internal change: 1666308)

(Internal change: 1666328)

set HD_ENABLE_BINDLESS_BUFFER to false by default.
this needs nvidia driver 367.44 or later to work correctly (nvidia bug 1723330)

the caveat of this workaround is, with falling back to SSBO we can only
have up to level 4 of instancing or so
(depending how much instance primvars we need).

(Internal change: 1666334)

…. For instance, if the texture wants to use a specific wrap mode. Also, we now setup a default anisotropic filter and a default border color.

(Internal change: 1666496)

no longer require it.

(Internal change: 1666501)

(Internal change: 1666513)

Calling realpath too frequently can be a performance killer for network
filers. This saves many stat calls per plugin discovered (one stat call
for every element of the path at least, more if there are actually
symlinks in the path).

(Internal change: 1666553)

(Internal change: 1666566)

QGLWidget.renderText requires GL fixed pipeline, thus it doesn't
work on OSX with core profile enabled (needed for hydra).
Instead we draw HUD contents to QImage using QPainter API,
and copy it back to QGLWidget as a texture. There are two
GLSL implementations to blit the textures back to QGLWidget,
for GL 3.0 or higher (includes Core profile OSX) and for GL 2.1
(compatibility profile OSX).

This change reduces ~3000 GL classic API calls for each frame.
To even reduce the overhead of texture transfer, split the HUD into
4 regions (TopLeft, TopRight, BottomLeft, BottomRight).

Also removed unused menu (VBO stats display).

(Internal change: 1666598)

checkling GL version using GLEW_VERSION can be problematic when
glewExperimental is set to detect extensions for core profile context.
This fix instead takes a standard way of GL_VERSION and
GL_SHADING_LANGUAGE_VERSION parsing to determine the GL version.

(Internal change: 1666745)

(Internal change: 1666835)

(Internal change: 1666850)

(Internal change: 1666851)

Add another way of passing lighting parameters (GlfSimpleLight,
GlfSimpleMaterial) in usdview. we no longer need to push the parameters
in GL fixed pipeline state and suck them back. RefEngine takes
those parameters and calls corresponding GL calls for compatibility.

SetLightingStateFromOpenGL is still being used in the third party
plugins and dress group. They should be updated in the same way
then that function can be removed.

(Internal change: 1666888)

(Internal change: 1667080)

(Internal change: 1667139)

(Internal change: 1667174)

It was giving folks problems and shouldn't be required anyway.

…ake variable PXR_LIB_PREFIX

By default shared libraries will have the prefix 'lib'  (NOTE: this does not apply to made PYTHON libraries)
To change the prefix to a custom prefix or to remove it use the cmake variable PXR_LIB_PREFIX like this

PXR_LIB_PREFIX:STRING="prefix_string_"

If you wish to remove the prefix from the libraries set the variable with empty string like this

PXR_LIB_PREFIX:STRING=""

(Internal change: 1667206)

(Internal change: 1667216)

(Internal change: 1667224)

UsdStage::GetLoadSet asserted that all prims in its load
set will always exist on the stage. This was incorrect; it's
actually very easy to get into such a state -- for example,
by deactivating the ancestor of a loaded prim.

This change removes the assertion and will simply return
the previously-loaded path in this case. This accurately
reflects what payloads are loaded in the underlying
PcpCache.

This restores the behavior of GetLoadSet in this case
to what it was prior to the introduction of the TF_VERIFY.
This behavior is confusing, though. A good follow-up
might be to unload prims whenever we deactivate them,
but that's a larger change.

(Internal change: 1667228)

(Internal change: 1667281)

(Internal change: 1667284)

(Internal change: 1667308)

…m attributes are identity

In particular, we don't want to be authoring shear XformOps unnecessarily.

(Internal change: 1667366)

(Internal change: 1667506)

As Pcp indexes scene description it discovers composition arcs
representing dependencies ("deps") on other scene description.
Pcp must track deps to manage data lifetime, to process later
changes to scene data, to invalidate caches, and to support
features that must reason about the dependency graph structure,
such as namespace editing.

Each dependency points to a "site" representing all the SdfSpecs
("specs") that reside at a path in a layer stack.
Historically our old composition system tracked deps in terms
of spec stacks, not sites.  We retained that in Pcp, while
suspecting that tracking specs is a mismatch for the task,
because of the principle that composition arcs connect sites,
not specs, and the specs can always be derived from sites.

Further experience confirmed there were some issues.
The usd project found that spec tracking was too expensive
in memory terms, so it turns them off entirely, but at the
cost of not supporting all dependency features correctly.
The other issue is complexity: there are cases that
spec deps are insufficient to handle, so Pcp_Dependencies
was actually a hybrid that also tracked sites, and would
use one representation or the other in various cases.
This made the code tricky to follow, and required additional
concepts to emerge in the API like "fallback ancestor".

The new system tracks the sites used by every prim index,
and stores a reverse mapping from sites back to index paths.

details

* PcpDependencyType is now a full-fledged categorization
  of dependency types.  PcpClassifyNodeDependency()
  can compute the dep type of a given node.

* Rather than maintaining a self-contained external bimap,
  Pcp_Dependencies now takes advantage of lock-step update
  with PcpPrimIndex entries in PcpCache so it can store
  only the inverse map and use indexes for the forward map.
  This cuts the runtime memory usage in half.

* The above change means we can simplify the update process
  and no longer need an async "flush" pass.

* PcpPrimIndex no longer builds a side list of deps as it goes.
  Instead, we rely on the node graph to fully represent the
  actual dependency structure.

  To support this, we now introduce variant nodes for
  variant selections even when the variant specs at the
  target do not yet exist.  This lets us track the dependency
  purely as a node, and is consistent with other kinds of arcs.

  To handle "spooky dependencies" we now propagate implied
  relocate nodes up to the root -- formally representing
  what was previously implicit structure, and allowing
  "spooky" cases to be handled by the normal code paths.

  This obviates the "shouldContributeDependencies" node flag.
  "fallbackAncestor" is also no longer a necessary API concept,
  because we always have ancestral site deps to use.

* PcpLayerStackSite now holds a ref rather than weakptr to
  its layer stack.  This is consistent with its purpose
  in representing a dependency on a layer stack (not a
  weak reference!); allows us to avoid having a duplicate
  side-structure for tracking the actual references
  already described by the sites; and shrinks the size
  of a pcp graph node (48 B -> 40 B).  (It doesn't make
  any difference to the # of atomic ref count ops we
  do since weakptrs refcount their remnant; now we
  refcount the layerstack instead.)

results

* less code, fixes correctness bugs
* less runtime memory -- ~5-7x less Pcp dependency memory;
  one production set usd goes from 933MB to 883MB (total)
* depenency tracking overhead now scales with dependency
  arc complexity of the scene, not the prim count

bugs: 135974, 137333, 96529

(Internal change: 1667509)

(Internal change: 1667564)

Eliminates all glBegin/glEnd and other fixed pipeline calls.

(Internal change: 1667610)

(Internal change: 1667633)

(Internal change: 1667647)

…minary work for the introduction

of configurable C++ namespaces into the pxr codebase. For this system to work, we don't want to have
explicit global qualifications. This is broken up into per lib changes, so many will follow this doing the same thing.

(Internal change: 1667717)

(Internal change: 1667718)

(Internal change: 1667720)

(Internal change: 1667721)

(Internal change: 1667722)

(Internal change: 1667724)

(Internal change: 1667725)