Tim Jones / MathWorks Inc.

GPU轉碼器(CPU Coder)可以將MATLAB的深度學習、嵌入式視覺和無人自主系統的程式碼轉成優化過的CUDA程式碼。 本演講將讓您了解在桌機或嵌入式NVIDIA GPU硬體上使用GPU轉碼器自動產生GPU程式碼：

• 從MATLAB自動產生優化的CUDA程式碼，用於深度學習、嵌入式視覺和無人自主系統
• 將生成的代碼整合成原始碼和靜態或是動態的函式庫
• 所生成的CUDA，可攜到NVIDIA的 GPU之間
• 所生成的程式碼是優化過的NVIDIA CUDA函式庫，包括cuDNN，cuSolver和cuBLAS
• 可將手寫CUDA程式碼併入到MATLAB演算法中並產生程式碼
• 可將演算法在GPU上如NVIDIA Tesla®和NVIDIATegra®上進行原型化
• 在MATLAB中使用生成的CUDA程式碼來加速MATLAB程式碼的密集運算

GPU Coder generates optimized CUDA code from MATLAB code for deep learning, embedded vision, and autonomous systems. See how you can use GPU Coder to automatically generate and execute GPU Code on your desktop or embedded NVIDIA GPU. In this presentation, we will explore the following:

• Generate optimized CUDA code from MATLAB code for deep learning, embedded vision, and autonomous systems
• Integrate generated code as source code and static or dynamic libraries
• Generated CUDA is portable across NVIDIA GPUs
• Generated code calls optimized NVIDIA CUDA libraries, including cuDNN, cuSolver, and cuBLAS
• Incorporate handwritten CUDA code into MATLAB algorithms and generated code
• Prototype algorithms on GPUs such as the NVIDIA Tesla® and NVIDIA Tegra®
• Use generated CUDA code within MATLAB to accelerate computationally intensive portions of your MATLAB code