Kareem Omar

CRC. Fastest CRC32 for x86, Intel and AMD, + comprehensive derivation and discussion of various approaches

RGB2Y. Fastest CPU (AVX/SSE) RGB to grayscale: 2-4x faster than OpenCV. For image processing/computer vision.

KFAST. Implementation of FAST feature detector for computer vision (Rosten 2006) using AVX2 to outperform canonical implementation by up to 600%.

SortingNetworks. Fastest CPU SIMD (SSE4) sorting networks for small integer arrays (2-6 elements), also optimal amd64 assembly and notes on getting compilers to generate optimal sorting networks.

FastArrayOps. Extremely fast x86 / AVX2 assembly implementations of common operations for linear arrays: checking whether array contains element, finding index of element, finding min/max element, finding index of min/max element.

KORAL. Novel extreme-performance CPU-GPU cooperative feature detector-descriptor for computer vision.

LATCH. Fastest CPU implementation of the LATCH 512-bit binary feature descriptor; fully scale- and rotation-invariant

CLATCH. Insanely fast CUDA LATCH: fully scale- and rotation-invariant 512-bit binary descriptor for computer vision

FastDivide. Divide 64-bit integers faster than hardware. Or precompute for a given denom and quickly divide repeatedly.

CUDAKfNN. Fastest CUDA SIFT or other 128-float vector matcher for computer vision

KPS. Infrastructure for simultaneous orbital and attitude propagation, with attitude-based real-time analytical aerodynamics simulation

KLERP. Fastest CPU (AVX2) Bilinear and Nearest-Neighbor Interpolation: 25-100% faster than OpenCV. For computer vision / image processing.

KNES. Complete, lightweight NES emulator in C++, speedcoded in 3 days.

CUDAK2NN. Insanely fast CUDA 2NN 512-bit binary descriptor matcher for computer vision

CUDARGB2Y. Fastest CUDA RGB to grayscale: 5-30x faster than OpenCV. For image processing/computer vision.

KfNN. Fastest CPU (AVX/SSE) SIFT or other 128-float vector matcher for computer vision

CUDALERP. Fast CUDA (GPU) Bilinear and Nearest-Neighbor Interpolation at high accuracy - uint8_t data

CUDAHammingMean. Fastest GPU implementation of a brute-force Hamming-weight matrix sum/mean for 512-bit binary descriptors.

FastIntegerSqrt. Fastest implementations of 32-bit and 64-bit integer square roots for x86-64

BoxBlur. Fastest CPU (AVX/SSE) Horizontal Box Blur for image processing and computer vision

CUDAFLERP. Fast CUDA (GPU) Bilinear and Nearest-Neighbor Interpolation at high accuracy - float32 data

popcount. Fastest possible x86 implementation of popcount/population count/Hamming weight/counting set bits

K2NN. Fast bruteforce and Multi-Index Hash (MIH) accelerated 2NN matchers for 512-bit binary descriptors for computer vision

FastThreadPool. Fast lock-free thread pool

BitOps. Basic, efficient, header-only bit ops and bit array primitives for modern x86. Tests provided.

ULATCH. Fastest CPU implementation of the LATCH 512-bit binary feature descriptor for computer vision (upright)

EllipticCurveFactorization. Fast, single-file, MIT-licensed large integer factorization using ECM combined with other techniques.

FeatureAngle. Extremely fast SSE gradient (angle of rotation) computation of grayscale features in an image, for image processing and computer vision.

UCLATCH. Insanely fast CUDA LATCH 512-bit binary descriptor for computer vision (upright)

ModularSqrt. Fast modular square root of primes and prime powers, including 2. Interface uses GMP bigints.

PrimeSieve. Super fast, dynamically expanding prime sieve for primality queries, forward or backward iteration

ArduinoPhysics. Realtime 2D physics and collision detection on an Arduino with 60 fps output to a Sharp memory LCD.

MemoryOrder. Demos of 3 ways even the strong memory model of x86 can exhibit architectural memory reordering, leading to bugs

UnsignedIntegralToFloatingPoint. Notes on fast standards-compliant conversion of U32/U64 to and from float/double, which compilers do not get right.

MATLAB-KDrag. Orbital and attitude propagator with B-dot and *dynamic* aerodynamic drag simulation, including torque computation for aero-stabilized bodies.

CUDAKfNN_packed. Fastest CUDA SIFT or other 128-float *packed as uint8_t* vector matcher for computer vision

Factorization-Primality. Extremely fast, single-file factorization and primality testing for 32-bit and 64-bit integers on x86.

PyCruiseControl. Modified divorced PID controller applied to car cruise control and accompanying physics simulation and visualizations

SolveModularQuadratic. Generate all solutions to a modular quadratic equation. Supports any modulus. Interface uses GMP bigints.

CudaBoids. Numerical simulation of flocking behavior using CUDA and OpenGL

U128. Fast unsigned 128-bit integer class for MSVC since it doesn't natively support __uint128_t yet

Leftpack. Fast AVX2 leftpack/compress implementations (keep and contiguously pack a subset of elements)

Sudoku. Fast sudoku solver with detection of no solution/single solution/multiple solutions/invalid initial board

KFAST_OpenMVG. Custom version of KFAST for integration into OpenMVG

FastDivide128. Getting __udivti3 or __umodti3 errors? Just want faster division/modulo for 128-bit ints on Clang? Look no further.

MATLABCruiseControl. Modified divorced PID controller applied to car cruise control and accompanying physics simulation and visualizations - MATLAB port

SMC-Demo. Minimal demo of self-modifying code on Windows. Still doable, still useful.

Bheap. Lightweight binary heap that greatly outperforms std::priority_queue and other commonly available heap implementations

KHALF. Optimized special-case bilinear interpolation, halving the width and not changing the height, for computer vision dual-frame display.

smart_tm. a smart, leap-second- and leap-day-aware, fast, 64-bit-capable replacement for the ctime 'tm' struct

SingleLinePythonSudoku. Single-line Python Sudoku solver

Boids_SDL. Numerical simulation of flocking behavior using pure CPU and SDL.

Schematic. Basic toy Lisp interpreter in a few hundred lines of C++.

SimpleOrbitalPropagator. Super super simple MATLAB orbital propagator in just a few lines. Experiment with orbits!