|
| 异构加速绝对平面检测 |
| Heterogeneous acceleration absolute flatness testing |
| 投稿时间：2023-12-19 |
| DOI：10.3969/j.issn.1005-5630.202312190136 |
| 中文关键词: 绝对平面检测 异构并行计算 Zernike多项式拟合 |
| 英文关键词:absolute flatness testing heterogeneous parallel computing Zernike polynomial fitting |
| 基金项目:国家重点研发计划(2022YFF0607704) |
|
| 摘要点击次数: 197 |
| 全文下载次数: 271 |
| 中文摘要: |
| 光学干涉检测领域的不断发展要求检测仪器具备更高的横向分辨率。高分辨率意味着处理时间变长，测试效率变低。为提高测试效率，提出了一种利用 CPU/GPU 异构计算并行加速的Zernike多项式绝对平面检测方法。该方法使用CPU进行流程控制，利用GPU多核优势将检测平面中的元素离散并行求解，并在Zernike 系数求解中使用混合精度，在峰谷值和均方根值求解中使用线程束原语指令进一步优化性能。结果显示，使用RTX3070– Laptop，在512×512、1024×1024、2 048×2 048和4096×4096分辨率的光学平面检测中，该方法整体处理速度比CPU处理速度分别提高了近47、56、58和70倍。 |
| 英文摘要: |
| The growing field of optical interferometry requires instruments with higher lateral resolution. High-resolution implies long processing time, which significantly affects testing efficiency. In order to improve the testing efficiency, a Zernike polynomial absolute flatness testing utilizing parallel acceleration of CPU/GPU heterogeneous computing was proposed, which used CPU for process to control and GPU multi-core advantage to discretize the elements in the flatness for parallel solving. Performance was further optimized useing mixed-precision in Zernike coefficient solving and peak-to-valley (PV) and root mean square (RMS) were solved with warp-level primitive instructions. Using 512×512, 1024×1024, 2 048×2 048, and 4096 ×4096 optically flat surfaces on the RTX3070–Laptop hardware, the speed increased by 47, 56, 58, and 70 times respectively in the overall process compared to the CPU version. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
