SPIRV-Interpreter

An interpreter for SPIR-V shaders/kernels, which aims to support the full SPIR-V specification. Outputs results of running the program for given inputs.

Use -h or --help on the spirv-run executable to see command line arguments and flags.

Features

• Support for vertex, fragment, compute, hull, and raytracing shaders
• Specify inputs and print outputs in YAML or JSON
• Textures may be read from and written to .png files
• Generate template files for expected inputs
• Check against expected results
• Verbose trace and interactive program execution
• 58 test examples, and counting

Limitations

Validation

For the sake of runtime speed and code clarity, the interpreter expects syntactically correct SPIR-V inputs. It is not, nor does it intend to be, a SPIR-V validator. If that is what you need, use spirv-val in SPIRV-Tools. If the input SPIR-V file is invalid, the behavior of the interpretation is undefined. This is unlikely to be a problem since most SPIR-V is auto-generated.

Image Processing

The SPIR-V specification does not explicitly define how some image operations should be done. This presents some challenges to the interpreter which aims to be a correct reference.

Many vendors use anisotropic filtering, but these algorithms have not been open sourced. Therefore, the interpreter uses n-linear interpolation instead, which should be easy to deconstruct and/or extend as needed.

Ray Tracing

Similarly to image processing, the SPIR-V specification does not define all characteristics of ray tracing functions and structures. A best guess approximation is used for the interpreter, which relies on information common to all implementations.

Building

The source can be built using a compiler which supports C++20 with modules (such as clang++ version ≥ 16 or g++ version ≥ 14). The source should be platform independent, but little to no testing has been done on Windows or Mac.

Here are a few build commands to run from the repository's root directory:

git submodule update --init --recursive
cmake -B build -G Ninja
ninja -C build all

assuming you have cmake version ≥ 3.28 and ninja version ≥ 1.11 installed.

After building, you should find the spirv-run executable at build/src/spirv-run.

Testing

The project has two complementary approaches to testing:

1. Integration tests on SPIR-V shaders in examples, commonly launched through test/example-runner.py.
2. Unit tests via Catch2 in the test directory

Note

You do not need to have Catch2 to build or run the spirv-run executable or the integration tests.

If you wish to run the unit tests, follow the steps to build, also including -DBUILD_TESTING=ON as an argument to cmake. After building, you should find a tests executable at build/test/tests.

Contributing

Contributions via merge requests are welcome! Contributions should:

• Be provided under the same license as the project (MPL 2.0).
• Follow the coding style. We recommend using clang-format (see settings at .clang-format).
• Be well-documented and have test cases (in examples and/or test directories).
• Not break other code. Include the results of running test/example-runner.py and tests in requests as proof.

License

The interpreter's novel source is licensed with the MPL v2.0. Full license terms are in LICENSE.md.

© SPIRV-Interpreter @ https://github.com/mmoult/SPIRV-Interpreter
This Source Code Form is subject to the terms of the Mozilla Public
License, v. 2.0. If a copy of the MPL was not distributed with this
file, You can obtain one at https://mozilla.org/MPL/2.0/.

There are a few code dependencies. You may review each individually (the last two are handled via git's submodule, so the links may not work until they are initialized) for the complete terms:

• SPIR-V: Custom Permissive License
• GLSL: Custom Permissive License
• stb: dual-licensed public domain or MIT
• glm: MIT License
• Catch2: Boost Software License

Examples in examples may use their own licenses. Each has a LICENSE file with the necessary documentation. Examples are only for testing and demonstrating behavior- they are not bundled in the spirv-run executable or any other project build (and as such, the interpreter is not a "derivative work" in the legal sense).