performance_optimization_tips.md

• Code Optimization - mostly C++

  • Intel 64 and IA-32 Architectures Optimization Reference Manual
  • Знакомьтесь, loop fracking
  • Приемы использования масочных регистров в AVX512 коде
  • LLVM для исследователей
  • Оптимизация циклов: нужны блоки
  • «Ра-а-авняйсь, смирно!». Выравниваем данные
  • Городские легенды о медленных вызовах виртуальных функций
  • Основные проблемы влияющие на производительность вычислительного ядра и приложения и методы их решения компилятором
  • Как правильно скопировать массив и при чем тут SFINAE
  • Linux Performance

• g++ autovectorization tips and issues - SSE,AVX

  • GCC auto-vectorization has no effect on runtime, even when supposedly “profitable”
  • Vectorize Source-Code
  • How to Write Fast Numerical Code
  • SSE multiplication of 4 32-bit integers
  • Vectorization in gcc
  • C++ - Getting started with SSE
  • FastC++: Coding Cpp Efficiently

• Is Parallel Programming Hard, And, If So, What Can You Do About It?

  • Paul E. McKenney

• Performance Monitor

  • What are perf cache events meaning?
  • perfmon2
  • perf wiki
    • also
    • Where Do I get Source package for Perf tool
  • Linux Performance Monitoring, any way to monitor per-thread?
    • An easy-to-use interface to the Linux perf_event API.
  • oprofile
    • tl;dr

    operf [ options ] [ --system-wide | --pid=<PID> | [ command [ args ] ] ]
    A typical usage might look like this:
    operf ./my_test_program my_arg

• MMU - TLB - huge pages

  • Virtual Memory
  • Which CPUs support 1GB pages?
  • How to use Intel Westmere 1GB pages on Linux?
    • Some benchmarks to evaluate NUMA remote memory overhead
  • HugeTLB - Large Page Support in the Linux Kernel
  • Hugepages
  • Huge pages part 1 (Introduction)
    • libhugetlbfs
    • Huge pages part 1 (Introduction)
    • Huge pages part 5: A deeper look at TLBs and costs
    • Memory part 7: Memory performance tools
      • Enabling Large Page Support
      • Using mmap and madvise for huge pages
      • mmap failed when trying to map huge page (1GB)
      • How do I allocate a DMA buffer backed by 1GB HugePages in a linux kernel module?
  • How Bad Can 1GB Pages Be?
  • Multiple Page Size Support in the Linux Kernel
  • mmap/munmap tips
    • munmap()--Remove Memory Mapping
  • Writing an OS in Rust - Page Tables

• Cache issues

  • Determine if cache is write-back or write-through
  • What causes a L3 cache miss in CPU?
  • What is “cache-friendly” code?
  • Cache Optimizing
  • Caches and multithreading
  • Optimizing for loops

• Other Memory issues

  • stream memory benchmarks

  • Poor memcpy Performance on Linux - AVX code for memcpy

  • various memcpy tips

  • fast memcpy

  • new technologies from Intel : Cache Monitoring Technology (CMT), Memory Bandwidth Monitoring (MBM), Cache Allocation Technology (CAT) and Code and Data Prioritization (CDP) Technology

    • Cache Monitoring Technology (CMT), Memory Bandwidth Monitoring (MBM), Cache Allocation Technology (CAT) and Code and Data Prioritization (CDP) Technology
    • User space software for Intel(R) Resource Director Technology http://www.intel.com/content/www/us/en/architecture-and-technology/resource-director-technology.html
    • processors supporting CAT Cache Allocation Technology
    • A few experiments with the Cache Allocation Technology
    • CAT white paper

• Timers

  • Precise Linux Timing - What Determines the Resolution of clock_gettime()?
  • Measure time in Linux - time vs clock vs getrusage vs clock_gettime vs gettimeofday vs timespec_get?

• Videos from The 3rd annual JuliaCon 2016 (MIT)

  • The 3rd annual JuliaCon 2016 (MIT) | Introduction to Writing High Performance Julia | Arch D. Robison

• Useful pieces of information

  • Difference between Mutex, Semaphore & Spin Locks
  • Does endless While loop take up CPU resources?

• Assembler

  • Compiler Explorer C++
  • Whirlwind Tour of ARM Assembly

• Scientific Methodology and Performance Evaluation

  • Scientific Methodology and Performance Evaluation for Computer Scientists
  • analytical GPU performance model based on Little’s law, that expresses the kernel execution time in terms of latency bound, throughput bound, and achieved occupancy
  • Understanding Latency Hiding on GPUs PhD Thesis by Vasily Volkov , EECS BerkeleyUnderstanding Latency Hiding on GPUs PhD Thesis by Vasily Volkov , EECS Berkeley

• Binary Instrumentation

  • review of binary instrumentation methods: Fuzzing binary-only programs with AFL++

  • Intel GTPin

    • Binary Instrumentation Technology GTPin
    • GTPin - A Dynamic Binary Instrumentation Framework
    • Binary Instrumentation with Graphics Technology Pin (GT Pin)

  • DynamoRIO

    • DynamoRIO
    • DRK: DynamoRIO as a Linux Kernel Module
    • dynamorio-sys: A crate using bindgen to automatically generate Rust bindings to DynamoRIO
    • dynamorio-rs crate provides safe Rust bindings to the DynamoRIO dynamic binary instrumentation framework