Tugbars Heptaskin
GitHub

Tugbars Heptaskin

@Tugbars

View GitHub Profile

About

Embedded Systems Engineer

Location Denmark
Company LINAK
Followers 111
Following 7
Public Repositories 35
Expertise

Skills & Technologies

C Cuda C++
GitHub Work

Projects & Repositories

Recent public projects and repositories from this profile.

VectorFFT

C

A FFT library in C, emitted by a Ocaml DAG FFT compiler. Supports both interleaved and split layouts, In-place & out-of-place C2C/R2C/C2R, 2D, and DCT/DST/DHT, with a calibrated wisdom planner. Beats Intel MKL on C2C and on multithreaded real/2D transforms. Pure C, zero dependencies.

⭐ 5 Forks: 1

Flash-Attention-PTX-CUDA

Cuda

Hand-written PTX flash attention kernel hitting 202 TFLOPS FP16 (86% of theoretical peak) on RTX 5080: beats or matches vLLM's FlashAttention-2 on most shapes without WGMMA, TMA, or datacenter hardware.

⭐ 12 Forks: 2

ICEEMDAN-MKL

C++

High-performance ICEEMDAN implementation using Intel MKL. Header-only C++17, OpenMP parallelized, ~11ms @ 2048 samples. Cubic/Akima splines, multiple processing modes (Standard/Finance/Scientific).

⭐ 3 Forks: 0

Savitzky-Golay-Filter

C

High-performance Savitzky-Golay filter in C: batch, streaming, and 2D image processing. Embedded-friendly with coefficient export for MCUs. MATLAB-validated.

⭐ 30 Forks: 4

Adaptive-Savitzky-Golay-Filter

C

A Savitzky-Golay implementation that dynamically applies the most optimal window size and polynomial order based on the shape and noise level of the dataset.

⭐ 10 Forks: 1

RLS-Driven-Peak-Analyzer

C

Microcontroller-friendly C implementation of Recursive Least Squares (RLS) for running gradient analysis, including linear, cubic, and quadratic trends, optimized for real-time trend and peak detection

⭐ 5 Forks: 0