Hardware Performance Counter Dataset for GPU Kernel Auto-Tuning: Four HPC Kernels on NVIDIA RTX A4000

Exhaustive hardware performance counter dataset for four GPU compute kernels: GEMM, 2D Convolution, Hotspot, and Dedispersion. Each kernel's full tuning parameter search space was explored on an NVIDIA RTX A4000 Ada GPU (collected on the DAS-6). Collection performed using Kernel Tuner, and source code of the kernels available at AutoTuningAssociation/benchmark_hub. The dataset was created as part of a bachelor thesis investigating whether hardware performance counters can serve as transferable guidance signals in GPU kernel auto-tuning. Each configuration includes runtime measurements and approximately 300 hardware performance counters covering the memory hierarchy (L1, L2, DRAM), warp scheduling, and execution pipelines.Both the raw JSON cache files and the more filesize optimized Parquet format files are provided.