Parallelizing the Slant Stack Transform with CUDA

In geophysics, the slant stack transform is a method used to align signals from different sensors. We focus on the use of the transform within passive refraction microtremor (ReMi) surveys, in order to produce high resolution slowness-frequency plots for use as samples in a machine learning model. Running on a single central processing unit (CPU) thread, this process takes approximately 45 min, 99.5% of which consists of the slant stack transform. In order to reduce the time taken to perform the transform, we use NVIDIA CUDA programming model. Using the same CPU, augmented with a GeForce RTX 2080 Ti we were able to reduce this time down to as little as 0.5 s.