WASM SIMD Data

This repository contains code and data which can be used to generate a JSON file containing information about the WASM SIMD proposal. For an example of the output, see this gist. Note that the current output contains many more instructions.

The data/ directory contains descriptions of the API, including opcodes, instruction names and parameters, as well as data about the corresponding higher-level APIs in both wasm_simd128.h and WAV.

The lowering/ directory then uses that information to generate more data on which instructions are expected to be used on various architectures to implement the WASM SIMD instructions, as well as performance data (number of uOps and latency). Note that this data should not currently be considered accurate; the lowering/ directory has more information, but basically many of the functions have not yet been optimized and engines such as V8 are likely to contain much better implementations.

For information on how to generate the lowering data, please see the README in the lowering/ directory.

Once you have generated the lowering data, you can generate the final output using scripts/generate.py. If you provide an argument to that scrip it will be used as an output file for a minified (no extra whitespace) JSON dump. If you do not provide an argument a non-minified JSON dump of the same data intended to be (relatively speaking) human-readable will be sent to stdout.

Please note that the format of this data should not be considered stable (yet?).

License

The license for the data has not yet been determined. Currently, copy some information from the WebAssembly 128-bit packed SIMD Extension proposal, which means it would be subject to the W3C license, but I'm leaning towards re-writing those sections anyways. Not only would this allow us to use a more permissive license (perhaps even a public domain dedication), it would allow more flexibility in how the content is organized.

To generate the lowering data, the lowering/generate-makefile.py script is first used to generate a Makefile which contains rules to allow generating the rest of the lowering data in parallel. To use it, just redirect the output to Makefile; for example, ./generate-makefile.py > Makefile.