CSC4700-GPU Programming, the C++ way

This lecture introduces GPU programming using C++ and the Kokkos library. The primary goal is to demonstrate how to write portable C++ code that can run efficiently on both CPUs and GPUs, addressing the challenges of diverse HPC architectures and programming models. The lecture covers Kokkos's API, parallel patterns (algorithms), views for data management, and integration with HPX for asynchronous operations. [00:00:00] - Introduction and Announcements [00:01:04] - Review of Previous Lecture: GPUs [00:03:02] - GPU Programming Model Differences [00:03:21] - Bridge Between CPU and GPU Programming [00:04:05] - Overview of Programming Models and Challenges [00:07:05] - Introduction to Kokkos Library [00:09:37] - Kokkos Users, Capabilities, and Backends [00:11:39] - Kokkos Kernels and Patterns [00:12:20] - Parallelization Example: Sequential and Parallel [00:16:04] - Passing Work to Kokkos: Function Objects and Lambdas [00:22:40] - Importance of Lambda Captures in GPUs [00:24:58] - Parallel Reduction Patterns in Kokkos [00:31:39] - Naming and Debugging Kokkos Kernels [00:32:52] - Summary So Far & Transition [00:33:45] - Kokkos Views: Motivation and Usage [00:36:04] - Kokkos Views: Dimensions and Syntax [00:40:48] - View Allocation and Memory Management [00:43:45] - Questions and Limitations About Views [00:44:28] - Execution Spaces in Kokkos [00:48:12] - Memory Spaces and Data Placement in Kokkos [00:51:39] - Host, Device, and Mirrored Views [00:59:07] - Data Layouts in Kokkos [01:01:09] - Integrating Kokkos with HPX [01:03:04] - Asynchronous Execution and HPX Futures [01:08:16] - Kokkos Executives and Execution Policies in HPX [01:12:09] - Final Example and Multi-dimensional Iteration [01:12:54] - Conclusion and Summary