The Curious Engineer

All blogs / Performance / optimization

Wojciech Muła --- website

0x80.pl

SIMD, AVX, AVX2, AVX512, optimization, algorithms, data structures

Performance / optimizationAlgorithms / data structuresCPU / microarchitecture

’10’15’20’25
Posts
174 from 29 Apr 2008 to 8 Jan 2026
Posting
now and then (1 in the last year)
Kind
personal blog
Feed
http://0x80.pl/feed.xml

2026 1

  1. Rust is perfectly imperfect

2025 8

  1. Change case of UTF-32-encoded strings
  2. LoongArch64 subjective higlights
  3. SIMD binary heap operations
  4. AVX512: printing u64 as binary
  5. Drawing trees
  6. Building full-text search in Javascript
  7. SIMD parallel bits deposit/extract
  8. Dividing unsigned 16-bit numbers

2024 3

  1. Dividing unsigned 8-bit numbers
  2. Myriad sequences of RISC-V code
  3. RISC-V Vector Extension overview

2023 13

  1. Simple suggestions using popcount
  2. AVX-512 conflict detection without resolving conflicts
  3. Modern perfect hashing for strings
  4. SIMD-ized faster parse of IPv4 addresses
  5. SWAR find any byte from set
  6. AVX512: finding first byte in lanes
  7. Finding lowest common ancestor of two nodes
  8. Faster fractional exponents
  9. Converting binary fraction to ratio
  10. AVX512: count trailing zeros
  11. AVX512: check if value belongs to a set
  12. AVX512: generating constants
  13. AVX512: histogram of sixteen nibbles

2022 5

  1. Faster hack
  2. Fast parsing HTTP verbs
  3. DDoS-ed by a service runs on AWS
  4. AVX512VBMI2 and packed varuint format
  5. Parsing hex numbers with validation

2021 7

  1. Bit test and reset vs compilers
  2. Conversion uint32 into decimal without division nor multiplication
  3. How to check if any word is zero
  4. Autovectorization status in MSVC in 2021
  5. Counting byte in byte stream with AVX512BW instructions
  6. How to detect if all bytes in SIMD register are the same?
  7. Autovectorization status in GCC & Clang in 2021

2020 2

  1. Use AVX512 to calculate binomial coefficient
  2. Use AVX512 Galois field affine transformation for bit shuffling

2019 9

  1. AVX512 8-bit positional population count procedure
  2. SIMDization of switch statements
  3. Malloc internal memory fragmentation footprint
  4. Auto-vectorization status in GCC, Clang, ICC and MSVC
  5. SIMDized counting byte in byte stream
  6. std::function and overloaded functions
  7. pyahocorasick stabilisation story
  8. C++ --- how to read a file into a string
  9. AVX512VBMI --- remove spaces from text

2018 20

  1. Python --- file modification time perils
  2. SIMDized sum of all bytes in the array --- part 2: signed bytes
  3. How many uops are there?
  4. A short report from code::dive 2018
  5. Speeding up multiple vector operations using SIMD
  6. SIMD --- why you shouldn't use static vector constants
  7. SIMDized sum of all bytes in the array
  8. SIMDized check which bytes are in a set
  9. Finding index of the minimum value using SIMD instructions
  10. AVX512 mask registers support in compilers
  11. AVX512 implementation of JPEG zigzag transformation
  12. Be careful with directory_iterator
  13. Parsing series of integers with SIMD
  14. Accidental recursion
  15. Is sorted using SIMD instructions
  16. When lock does not lock --- C++ story
  17. An awful part of C++17
  18. Intersection of ordered sets
  19. SSE/AVX: absolute value of difference of unsigned integers
  20. Is power of two --- BMI1 version

2017 2

  1. A short report from code::dive 2017
  2. ARM Neon and Base64 encoding & decoding

2016 17

  1. AVX512 --- first bit set in a large array
  2. SWAR check if all chars are digits
  3. Population count using XOP instructions
  4. SIMD-friendly algorithms for substring searching
  5. What does AVX512 conflict detection do?
  6. Detecting bit patterns with series of zeros followed by ones
  7. Byte-wise alignr in AVX512F
  8. GNU std::string::find is very slow
  9. Sorting an AVX512 register
  10. AVX512F base64 coding and decoding
  11. SIMD bit mask
  12. Building a bitmask
  13. Base64 encoding & decoding using AVX512BW instructions
  14. Implementing byte-wise lookup table with PSHUFB
  15. Base64 decoding with SIMD instructions
  16. Base64 encoding with SIMD instructions
  17. Speeding up letter case conversion

2015 15

  1. Fast conversion of floating-point values to string
  2. Base64 encoding --- implementation study
  3. Benefits from the obsession
  4. Implicit conversion --- the enemy
  5. Another C++ nasty feature
  6. Short report from code::dive 2015
  7. Boolean function for the rescue
  8. Tricky mistake
  9. Speeding up bit-parallel population count
  10. SIMD-ized searching in unique constant dictionary
  11. SIMD: detecting a bit pattern
  12. Compiler warnings are your future errors
  13. AVX512: ternary functions evaluation
  14. SSE/AVX2: Generating mask where n leading (trailing) bytes are set
  15. Not everything in AVX2 is 256-bit

2014 25

  1. Using SSE to convert from hexadecimal ASCII to number
  2. Parsing decimal numbers --- part 2: SSE
  3. Parsing decimal numbers --- part 1: SWAR
  4. Using PEXT to convert from hexadecimal ASCII to number
  5. Using PEXT to convert from binary ASCII to number
  6. Conversion numbers to octal representation
  7. Determining if an integer is a power of 2 --- part 2
  8. Conditionally fill word (for limited set of input values)
  9. Small win over compiler
  10. Interpolation search revisited
  11. Software emulation of PDEP
  12. Conversion numbers to hexadecimal representation
  13. Conversion numbers to binary representation
  14. C++ bitset vs array
  15. Quick and dirty ad-hoc git hosting
  16. Is const-correctness paranoia good?
  17. Scalar version of SSE move mask instruction
  18. SIMD-friendly Rabin-Karp modification
  19. C++ standard inaccuracy
  20. Integer log 10 of an unsigned integer --- SIMD version
  21. Mask for zero/non-zero bytes
  22. GCC --- asm goto
  23. Slow-paths in GNU libc strstr
  24. Penalties of errors in SSE floating point calculations
  25. x86 - ISA where 80% of instructions are unimportant

2013 14

  1. I accidentally created an infinite loop
  2. Calculate floor value without FPU/SSE instruction
  3. Convert float to int without FPU/SSE
  4. fopen a directory
  5. x86 extensions are useless
  6. Problems with PDO for PostgreSQL on 32-bit machines
  7. Encoding array of unsigned integers
  8. FBSTP --- the most complex instruction in x86 ISA
  9. Short story about PostgreSQL SUM function
  10. PostgreSQL --- faster reads from static tables
  11. PostgreSQL: printf in PL/pgSQL
  12. SSE: trie lookup speedup
  13. Detecting intersection of convex polygons in 2D
  14. PHP quirk

2012 2

  1. Average of two unsigned integers
  2. Speeding up LIKE '%text%' queries (at least in PostgeSQL)

2011 7

  1. SSE: conversion integers to decimal representation
  2. Traversing DAGs
  3. DAWG as dictionary? Yes!
  4. Python: C extensions --- sequence-like object
  5. Efficient trie representation
  6. Python: test if object is iterable
  7. Traversing tree without stack

2010 9

  1. Branchless set mask if value greater or how to print hex values
  2. Speedup reversing table of bytes
  3. Determining if an integer is a power of 2
  4. Brenchless conditional exchange
  5. STL: map with string as key --- access speedup
  6. Fill word with selected bit
  7. Branchless signum
  8. Transpose bits in byte using SIMD instructions
  9. PostgreSQL: get selected rows with given order

2008 15

  1. Join locate databases
  2. SSE4.1: PHMINPOSUW --- insertion sort
  3. SSSE3: PMADDUBSW and image crossfading
  4. SSE: conversion uint32 to float
  5. Floating point tricks
  6. RDTSC on Core2
  7. PABSQ --- absolute value of two singed 64-bit numbers
  8. GCC asm constraints
  9. SSSE3/SSE4: alpha blending --- operator over
  10. SSE4: grater/less or equal relations for unsigned bytes/words
  11. 16bpp/15bpp to 32bpp pixel conversions --- different methods
  12. SSE: modify 32bpp images with lookup tables
  13. SSE4 string search --- modification of Karp-Rabin algorithm
  14. SSSE3: fast popcount
  15. SSSE3: printing hex values