A series index for LLM attention kernels and GPU primitives: fused softmax, online softmax, FlashAttention, PagedAttention kernels, Triton/CUDA, and memory-access optimization.