Quanquan Hu, Jian Ge, Luoxi Jin, Kevin Willis

Computational efficiency is a critical requirement for transit searches in modern large-scale photometric surveys. We present Graphics Processing Units Transit Least Squares (GTLS), a GPU-accelerated implementation of the Transit Least Squares algorithm designed to reduce the computational cost of periodic transit detection while preserving TLS-like sensitivity to transit-shaped signals. GTLS parallelizes the dominant steps of the TLS search, including phase folding, transit-duration evaluation, moving-window depth estimation, and chi-squared calculation. Using Kepler-like long-cadence light curves and synthetic Kepler-like time series, we benchmark GTLS against the reference CPU implementation of TLS and the GPU-based BLS implementation in cuvarbase. On an AMD Ryzen 9 7950X CPU and an NVIDIA RTX 4090 GPU, GTLS processes a 3000-day synthetic light curve in approximately 138 seconds, compared with 3289 seconds for TLS. With two RTX 4090 GPUs, the runtime is reduced to approximately 79 seconds. In recovery tests, GTLS achieves detection performance statistically consistent with TLS, with a precision of 9.3 percent and recall of 79.4 percent, compared with 9.4 percent and 81.1 percent for TLS. These results demonstrate that GTLS enables efficient TLS-style searches for large photometric data sets from Kepler, TESS, PLATO, ET, and future missions. The source code is publicly available.