Stricter check for eliding for Shader Optimization

[bowald]

This is a follow up to the PR #2746.

Me and @rasmustilljander continued to look into issues related to upgrading our MaterialX 1.38 materials to MaterialX 1.39.
The PR #2746 solved an issue for custom nodes getting classified as CONSTANTS but without inputs. However, we discover some more appearances having shaders generated with syntax errors not fixed by last PR. This PR tightens the check to make sure only constant/dot nodes are elided.

Here is a "simplified Material" that fails shader code generation on main, where the custom nodes is considered for bypassing even if they have 2 outputs (i.e. not a constant node).

<?xml version="1.0"?>
<materialx version="1.39" xmlns:xi="http://www.w3.org/2001/XInclude">
  <nodedef name="ND_custom_a" node="custom_a">
    <input name="specular_roughness_raw_in" type="float" value="0.5" />
    <output name="specular_roughness_out" type="float" />
    <output name="specular_ior_out" type="float" />
  </nodedef>
  <nodegraph name="NG_custom_a" nodedef="ND_custom_a">
    <constant name="constant_0" type="float">
      <input name="value" type="float" interfacename="specular_roughness_raw_in"/>
    </constant>
    <output name="specular_roughness_out" type="float" nodename="constant_0" />
    <output name="specular_ior_out" type="float" nodename="constant_0" />
  </nodegraph>
  <nodedef name="ND_custom_b" node="custom_b">
    <output name="specular_roughness_raw_out" type="float" />
    <output name="specular_anisotropy_out" type="float" />
    <input name="normal_raw_in" type="vector3" value="0.5, 0.5, 1.0" />
  </nodedef>
  <nodegraph name="custom_b" nodedef="ND_custom_b">
    <output name="specular_roughness_raw_out" type="float" nodename="constant0_ST0054" />
    <output name="specular_anisotropy_out" type="float" nodename="spec_anisotropy_ST0054" />
    <constant name="spec_anisotropy_ST0054" type="float">
      <input name="value" type="float" value="0.37" />
    </constant>
    <constant name="spec_rotation_ST0054" type="float">
      <input name="value" type="float" value="0.37" />
    </constant>
    <constant name="constant0_ST0054" type="float">
      <input name="value" type="float" value="0.37" />
    </constant>
  </nodegraph>
  <nodegraph name="NG_APPEARANCE">
    <output name="specular_roughness" type="float" nodename="custom_a_inst01" output="specular_roughness_out" />
    <output name="specular_anisotropy" type="float" nodename="custom_b_inst01" output="specular_anisotropy_out" />
    <custom_a name="custom_a_inst01" type="multioutput">
      <input name="specular_roughness_raw_in" type="float" nodename="custom_b_inst01" output="specular_roughness_raw_out" />
    </custom_a>
    <custom_b name="custom_b_inst01" type="multioutput">
    </custom_b>
  </nodegraph>
  <standard_surface name="SR_APPEARANCE" type="surfaceshader">
    <input name="base_color" type="color3" value="0.73, 0.68, 0.57" />
    <input name="specular_roughness" type="float" nodegraph="NG_APPEARANCE" output="specular_roughness" />
    <input name="specular_anisotropy" type="float" nodegraph="NG_APPEARANCE" output="specular_anisotropy" />
  </standard_surface>
  <surfacematerial name="APPEARANCE" type="material">
    <input name="surfaceshader" type="surfaceshader" nodename="SR_APPEARANCE" />
  </surfacematerial>
</materialx>

This is the content of the nodegraph NG_APPEARANCE, where both custom_a_inst01, custom_b_inst01 are considered constant and on main they will be elided even though that is incorrect.

[jstone-lucasfilm]

This looks like a good improvement, thanks @bowald, and I had just a few minor notes for clarity.

[jstone-lucasfilm]

Let's just inline these two tests, rather than creating an intermediate constant, as I believe that will read more clearly.

[bowald]

Fixed

[jstone-lucasfilm]

Perhaps "Constant node doesn't follow expected interface, cannot elide"?

[bowald]

hehe, I need to include a spell checker in my IDE. Thanks for addressing this.

[jstone-lucasfilm]

Perhaps "Dot note doesn't follow expected interface, cannot elide", matching the comment above?

[bowald]

Fixed

[jstone-lucasfilm]

Thanks for the fix, @bowald!