Parallel Algorithms, Ranges and oneDPL for Hardware Accelerators

Explore a conference talk from Meeting C++ 2024 that delves into parallel algorithms, ranges, and Intel's oneDPL implementation. Learn about proposals for parallelization (P2300, P2500, P3179) and their impact on hardware accelerators like GPUs and FPGAs. Discover how senders and receivers facilitate async task scheduling across threadpools and accelerator interfaces, while understanding the implementation of user-defined execution policies for specific accelerators. Examine the adaptation of parallel algorithms, execution contexts, and policies for accelerator backends like oneDPL and Thrust, including platform-specific execution policies that extend std::execution principles. Gain insights into ranges parallel algorithms with execution policies, focusing on computation call fusion and overhead reduction. See practical demonstrations using oneDPL to solve computationally intensive problems on GPUs and other accelerators, with examples showcasing the implementation of these principles in real-world scenarios.