Skip to content
A yEnc encoder using vectorized ARM64 assembly
Assembly
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
README.md Initial commit Nov 14, 2019
encoder.s

README.md

yEnc-aarch64

This is a yEnc encoder I built to learn ARM64 assembly programming on Linux.

I've documented the code excessively. If you're looking to learn assembly or low level systems programming, I think reading the source isn't a bad place to start.

Also, check out why using yEnc encoding is a bad idea.

Features

  • Pure assembly with no dependencies
  • Optimized (vectorized) implementation using SIMD instructions
  • Low memory requirements

Build

On a 64-bit ARM host, build a static ELF executable with: gcc -nostdlib -static encoder.s.

Performance

You'll need access to an aarch64 machine with hardware performance counters exposed. The bare metal Cavium ThunderX servers from Packet.net worked well for me.

Install perf and, ideally, taskset. On Ubuntu, that's sudo apt install linux-tools-common linux-tools-$(uname -r).

Profile with taskset -c 2 perf stat ./a.out input output.

 Performance counter stats for './a.out /dev/shm/random_1M /dev/shm/output':
 
          4.919380      task-clock (msec)         #    0.914 CPUs utilized
                 0      context-switches          #    0.000 K/sec
                 0      cpu-migrations            #    0.000 K/sec
               269      page-faults               #    0.055 M/sec
         9,829,420      cycles                    #    1.998 GHz
         3,743,103      instructions              #    0.38  insn per cycle
           376,389      branches                  #   76.511 M/sec
           117,801      branch-misses             #   31.30% of all branches

       0.005381709 seconds time elapsed
You can’t perform that action at this time.