yume-pdq

Optimized kernels from hashing to vectorized or Vulkan shader-based exact matching for near real-time high-throughput high-recall PDQ-based image screening, including a hand-vectorized PDQ variant that prioritizes low latency, high throughput with high statistical compatibility, with options of using AVX2 intrinsics, portable-simd, or AVX512 intrinsics, with no-std and LLVM SafeStack+CFI hardening support.