gem5 bootcamp 2022: GPGPU model

In this section, we go over the AMD GPU model. Slides: https://ucdavis365-my.sharepoint.com/:p:/g/personal/jlowepower_ucdavis_edu/EVrhvLYF4olFncllsswZyNIBE0v1DfQbhvvJbx5OIIw1wg?e=WtPA60&action=embedview More info: https://gem5bootcamp.github.io/gem5-bootcamp-env/ For comments, use the gem5-bootcamp channel in the gem5 slack. Invitation here: https://join.slack.com/t/gem5-workspace/shared_invite/zt-1aal963w6-_cqn0u8QgBh3GeeSi2Ja7g gem5 bootcamp 2022: GPGPU model Timecode 00:00 Intro 03:54 Disclaimers 05:42 Contributors 10:44 Graphics Processing Units (GPU) 13:56 Flynn's Taxonomy 17:34 Execution Model Comparison 18:39 GPU's & Memory 25:21 Programming GPUs 29:17 How do we do efficient Memory Access? 38:43 SIMT Unit - a GPU Pipeline 41:39 Need for Specialized Memories 47:24 What a GPU app Running in gem5 requires 51:18 Creating Portable gem5 Resources 59:47 APU vs dGPU 01:01:27 Key GPU Code Locations 01:03:58 How does GPU Kernel Actually Run 01:09:49 How does an Instruction Run Through GPU 01:16:53 Running GPU Programs in gem5 01:33:01 Comparing Results 01:37:26 Comparing Simple and Dynamic Register Allocation 01:40:25 Running Multi-Kernel GPU Applications 01:44:08 Adding m5ops Steps 02:04:43 Impact of m5ops