Enable Slang render tests in pytest framework and streamline slang-rhi integration

[ppenenko]

Summary

This issue outlines the roadmap for enabling Slang render validation within our pytest-based rendering framework (contrib/tests) in metashade/MaterialX.

The primary objective is to achieve cross-backend rendering parity (GLSL vs. Slang) to prove that Metashade generates functional Slang shaders that render identically to GLSL baselines on actual GPU hardware.

---

Background & Current State

1. Shader Generation (MaterialXGenSlang):

   • The Slang code generator is fully supported and enabled by default (MATERIALX_BUILD_GEN_SLANG=ON).
   • Python bindings (PyMaterialXGenSlang) and JS bindings exist.
   • Syntax and node implementation tests exist in Catch under [genslang].

2. Slang Renderer (MaterialXRenderSlang):

   • Implements ShaderRenderer using slang-rhi (supporting D3D12 on Windows, Vulkan on Linux, and Metal on macOS).
   • MaterialXTest contains a [renderslang] test case in source/MaterialXTest/MaterialXRenderSlang/RenderSlang.cpp.
   • Currently requires manually supplying -DMATERIALX_SLANG_RHI_SOURCE_DIR at CMake configure time.
   • Omitted from upstream CI because slang-rhi is not pre-packaged on runner images.

3. Python Render Bindings Missing:

   • While PyMaterialXRenderGlsl, PyMaterialXRenderMsl, and PyMaterialXRenderOsl exist, there is currently no PyMaterialXRenderSlang module.
   • As a result, Python-based test harnesses (including our pytest framework) cannot currently instantiate a native Slang renderer.

---

Implementation Roadmap

Phase 1: Zero-Friction slang-rhi Integration (Upstream Candidate)

• CMake FetchContent with local fallback:
  • Introduce MATERIALX_BUILD_RENDER_SLANG option.
  • If MATERIALX_SLANG_RHI_SOURCE_DIR is not provided, automatically fetch and build a pinned version of shader-slang/slang-rhi via CMake FetchContent.
  • Retain full support for MATERIALX_SLANG_RHI_SOURCE_DIR so developers hacking on / debugging local slang-rhi checkouts can point to their local directory.
• Developer Experience:
  • Add a windows-slang configure/build preset in CMakePresets.json.
  • Add VS Code task and debug configuration for running MaterialXTest [renderslang].
• Upstream Benefit: Makes MaterialXTest [renderslang] completely reproducible on any developer machine and CI runner with a single flag.

Phase 2: Python Bindings for Slang Renderer (PyMaterialXRenderSlang) (Upstream Candidate)

• Add source/PyMaterialX/PyMaterialXRenderSlang/PySlangRenderer.cpp (~30 lines of pybind11 code matching PyGlslRenderer.cpp / PyMslRenderer.mm).
• Bind SlangRenderer::create, initialize, createProgram, �alidateInputs,
  ender,
  enderTextureSpace, captureImage, and getProgram.
• Configure the PyMaterialXRenderSlang CMake module.

Phase 3: pytest Framework Integration (metashade/MaterialX)

• Shader Generator Registry:
  • Register mx_gen_slang.SlangShaderGenerator.create() under target genslang in mtlxutils/mxshadergen.py.
• Slang Render Environment:
  • Implement SlangRenderer in mtlxutils/mxrenderer.py wrapping PyMaterialXRenderSlang.
  • Add --renderer slang CLI option and fixtures in contrib/tests/conftest.py.
• Cross-Backend Validation:
  • Run contrib/tests/test_render.py with Slang backend and compare rendered outputs directly against GLSL references using perceptual FLIP gating (--flip-threshold 0.05).
  • Add VS Code launch configuration and task: pytest: Metashade Render Slang.

Phase 4: CI & Automation

• Add a GitHub Actions CI matrix job on Windows / Linux runners to execute both MaterialXTest [renderslang] and pytest contrib/tests/test_render.py --renderer slang.
• Include Slang runs in scheduled builds to catch compiler and driver regressions early.