Blender Git Commits

Merge branch 'master' into object_nodes

Simple node for concatenating 2 meshes into one.

Ignore nodes that don't have the 'compile' method, to allow frames.

Added back automatic type conversion.

Now automatic type conversion for sockets happens in the python code.

Fix for refcounting in the mesh value node.

This node was overwriting the entire stack pointer (incl. the refcount)
but should only store the DM itself.

Make sure invalid node connections are rejected at the basic level.

The internal bvm nodes require strict type matching. Any implicit
automatic type conversions should be added on the level of DNA->bvm
conversion. This keeps the bvm nodes simple, and also allows for
semantics on the DNA node level that could not be properly implemented
in bvm.

Improved python API for constructing the internal node graph.

Socket connections can now be handled consistently through single py objects,
so the compilation code becomes more concise and readable.

Fix for dangling root sockets when skipping 'pass' nodes.

This can happen when proxies are created, and should always create
a 'value' node to act as a clean root terminator.

Improved RNA API for bvm node graphs.

Now there are actual collections of inputs/outputs to perform proper checks
on valid names/indices and types.

Sanity check for array count <= 0, otherwise CDDM keeps spewing warnings about empty mesh.

Added an "Iteration" node as a general-purpose input for expressions.

This is used by the Array modifier node to give a meaningful per-instance
input to calculate a transform from.

Use the jump address in the array modifier to evaluate a transform function per copy.

Codegen: for function inputs store the entry point in the instructions.

Generate instructions for a subgraph in a separate codegen function.

This will be useful for separately generating kernel functions, which
are to be stored as own code blocks so kernels can call them repeatedly
as needed.

Fix for API changes in pynodes.

Fix for stack index assignment to graph outputs.

Use a simple value node for graph outputs that are only connected to pass nodes.

Changing graph output mechanism to use proxy nodes.

Distinguish 'function' and 'kernel' node types.

Only kernel nodes will be allowed to have function inputs. This is
necessary because any kernel node will define the boundaries of
node subtrees used for function inputs.

Node inputs now have a 'value type' to specify how input connections are treated.

VALUE_CONSTANT inputs can not have connections, their values are known at compile time.
bvm stores these values directly in the instruction list.

VALUE_VARIABLE inputs store values at evaluation time. bvm stores a stack index for them
which is used to load and store from/to the evaluation stack.

VALUE_FUNCTION inputs are passed to kernels to be called internally when needed.
bvm stores an instruction index and allows a jump to this instruction (TODO).