This repository contains standalone HLSL matrix multiplication shaders and native D3D12 linear algebra samples.
This repository demonstrates different approaches based on feature availability:
- Standalone shaders: three GEMM implementations using groupshared memory, wave-scope matrix operations, and threadgroup-scope matrix operations.
- sin-network: a complete native D3D12 sample that
approximates
sin(x)with a 16-neuron network and thread-scope matrix-vector multiplication. Its README contains build instructions and a walkthrough of the C++ host, HLSL shader, resource layout, and dispatch flow. - xor-network: a complete native D3D12 sample that
classifies the four XOR inputs as one batch with a
2 -> 4 -> 1network, compares FP16 threadgroup matrix multiplication with a wave-tiled GEMM implementation, and verifies both output matrices agree.