|
FZGPUModules 2.0
GPU-accelerated modular compression pipelines
|
#include <warp_op_params.h>
cuSZp2 / SZp: linear-ABS quant + 1-D Lorenzo, block reset every 32*epl elements. epl (= block_size/32) lets the policy address element 32*m+lane of block b for epl > 1; the delta chain runs across the whole block.