vpgatherdd - I think that for newer CPUs it is faster than many loads + inserts, but if you are going to fault a lot, then it becomes slow. > The VGATHER instr

vpgatherdd - I think that for newer CPUs it is faster than many loads + inserts, but if you are going to fault a lot, then it becomes slow. > The VGATHER instructions are implemented as micro-coded flow. Latency is ~50 cycles. https://www.intel.com/content/www/us/en/content-details/814198/intel-64-and-ia-32-architectures-optimization-reference-manual-volume-1.html https://www.intel.com/content/www/us/en/content-details/8141...