Take Advantage of All the MIPS - SYCL & C++
summary
SYCL claims you can write one C++ source that targets CPUs, GPUs, FPGAs and DSPs and the examples actually back that up: frontends and runtimes like DPC++, ComputeCpp, hipSYCL, triSYCL compile to backends including CUDA PTX Level Zero and OpenCL, tools used include compiler-explorer and NVIDIA Nsight for inspection, and measured results include an automated CUDA→SYCL n‑body port running 8.8 ms vs 10.19 ms on RTX 3060 while kernel tuning (inlining, memory coalescing, USM vs buffers) changes performance dramatically.
AI-written summary · about
chapters (10)
- 0:00Intro
- 15:24Migration
- 18:12Backend
- 21:22Results
- 23:43Complete Application
- 31:56Buffer Synchronization
- 35:43Hello World Example
- 39:56Inline Function
- 46:26SQL Implementation
- 52:36MD Span
related talks
24mOpen Acceleration for RISC-V – Portability, Performance and Partners...
01h 07mHeterogeneous Modern C++ with SYCL
37mC++ for multi-accelerator, multi-vendor systems
56m“Towards Heterogeneous Programming in C++"
01h 09mC++20 on Xilinx FPGA with SYCL for Vitis
01h 32mHeterogeneous Programming in C++ today
20mThe Journey to Enabling RVV Solutions with SYCL and the OneAPI Ecosystem
58mGenerating OpenCL/CUDA source code from C++ expressions in VexCL