Release Tasks

Current Work-in-progress:

• CreateArray factory functions
• Fix CreateKernel so that ComputePipeline is saved, CommandBuffer is ephemeral and not reusable
• Fix # threads / # workgroups to allow span in x, y, and z directions instead of being 1D
  • allow for flexibility in specifying # threads in each direction so it doesn't assume it's 1:1 with output dimensions
• allow for more flexible output signatures (multiple buffers? 1 or more params?)
• physics demo
• fix projection in rendering demo

P0

• User testing and feedback
• Separate template project repo (based on examples/hello-world but without the relative directory pointers)
• Tutorial program - update
• Updated slide presentation

P1

• Implement srush GPU puzzles
• Implement transformer block
• Benchmarking suite
• Optimize Kernels to perform competitively
• Wrap current test code in a testing framework
• Implement Sibhoem matmul series in WGSL, optimized matmul
• Raymarching terminal shader demo

P2

• Implement an end-to-end architecture - gemma and/or gpt2

Done

• README
• Decouple logging from global spdlog instance (soln for now: we've embedded a mini logging library)
• Multi-shader kernels in a single command buffer
• Naming convention?
• Refactor Shape to not be dynamically allocated
• Builds work for all examples
• Update deprecated SetDeviceLostCallback --> SetDeviceLostCallbackInfo