Autodesk, Inc.

Dominic Laflamme

Ian F Stewart

Jérôme Couture-Gagnon

Ronald Beirouti

Maxime Beaudry

Eric Cabot

Peter Zion

It is desirable for a fragment shader to have access to non-interpolated values for each vertex of the primitive in which the fragment is located. For example, a fragment shader may use the distortion of the primitive with respect to an original state of the primitive as part of the function the fragment shader performs. Due to the specification of fragment shaders and vertex shaders, fragments shaders receive only interpolated values, and thus cannot receive non-interpolated values of, for example, one solution to this problem would be to modify the processing engine for the shader language, and the shader specifications themselves, so that a fragment shader can receive non-interpolated values from the vertices of the primitive on which the fragment is located. Desirable values to receive would be at least the vertex coordinates. Another solution is to specify and use varyings in a manner that pass data to a fragment shader that permit the fragment shader to reconstruct the non-interpolated values. One way to achieve this is to a. allocate varyings and assign them indices, b. assign indices to the vertices and c. have each a shader contribute only to those varyings having the same index as the vertex being processed, and otherwise contribute a null value, such as 0, to the varyings with other indices. In this manner, when the interpolated value for the indexed varying is received by the fragment shader, the indexed varying contains the contribution of only one vertex, scaled by an interpolation parameter. Another indexed varying can be used to pass the interpolation parameter, allowing the original value for the vertex to be computed by the fragment shader.

Exchanging data between vertex shaders and fragment shaders on a graphics processing unit

An operator graph representing three-dimensional animation can be analyzed to identify subgraphs of the operator graph in which operators are not required to operate in a serialized manner. Such a condition may arise, for example, when two operators are not dependent on each other for data. This condition may arise when the operators are operating on different elements in a scene. Such operators may be evaluated in parallel. To identify these operators, a dependency graph is created. The dependency graph indicates which operators have inputs that are dependent on outputs provided by other operators. Using this graph, operators that are independent of each other can be readily identified. These operators can be evaluated in parallel. In an interactive editing system for three-dimensional animation or other rich media, such an analysis of an operator graph would occur when changes are made to the animation. The creation and use of the dependency graph enables faster processing of a scene while changes to the scene are being made.

Efficient processing of operator graphs representing three-dimensional character animation

Growing community of inventors

Montreal, Canada

Dominic Laflamme