| 赵皓宇,王重熙,宋鹏皓,章隆兵.基于无裁剪图形流水线的三维图形处理器[J].高技术通讯(中文),2024,34(7):681~691 |
| 基于无裁剪图形流水线的三维图形处理器 |
| A 3D graphics processor with clip-free graphics pipeline |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2024. 07. 002 |
| 中文关键词: 三维图形处理器； 图形流水线； 裁剪； 向量内积 |
| 英文关键词: 3D graphics processor, graphics pipeline, clipping, vector inner product |
| 基金项目: |
| 作者 | 单位 | | 赵皓宇 | （处理器芯片全国重点实验室（中国科学院计算技术研究所）北京 100190）
（中国科学院大学北京 100049） | | 王重熙 | | | 宋鹏皓 | | | 章隆兵 | |
|
| 摘要点击次数: 323 |
| 全文下载次数: 275 |
| 中文摘要: |
| 传统的三维图形处理器通过裁剪操作获取三角形的可见区域。然而，裁剪操作的延迟长且硬件开销高，大量的裁剪操作会降低图形处理器的性能。本文设计了一款基于OpenGL ES 2.0标准的三维图形处理器芯片，采用了统一渲染架构。该图形处理器采用高效的无裁剪图形流水线结构，消除了裁剪所带来的硬件开销和性能损耗。此外，本文为该图形处理器设计了一个符合IEEE-754标准的三维向量内积（DP3）计算单元，用于固定功能流水线，以提高图形处理器的性能，并消除图形渲染过程中浮点乘加操作的误差，增强了图形处理器的图形渲染鲁棒性。该三维图形处理器每秒能够处理500M个顶点和8G个纹素，功耗为1000mW，采用了28nm工艺，面积为7.92mm2。实现结果表明，与之前的工作相比，本文设计的图形处理器的性能-功耗比提高了27.8%。 |
| 英文摘要: |
| Traditional 3D graphics processors rely on clipping operations to determine the visible area of triangles. However, clipping operations introduce latency, high hardware overhead, and can negatively impact performance when dealing with a large number of clipping operations. This paper presents the design of a 3D graphics processor with a unified rendering architecture based on the OpenGL ES 2.0 specification. This proposed graphics processor incorporates an efficient clip-free graphics pipeline architecture, effectively eliminating the hardware overhead and performance loss associated with clipping. Furthermore, an IEEE-754 compliant 3D vector inner product unit is introduced into the fixed function pipeline of the proposed processor, enhancing performance and eliminating errors in floating-point multiplication and addition operations during graphics rendering, improving the robustness of the graphics rendering. The implemented 3D graphics processor achieves a processing capacity of 500M vertices and 8G texels per second, consuming 1000mW of power. Utilizing a 28nm process with an area of 7.92mm2, the proposed implementation demonstrates a 27.8% improvement in the performance-to-power ratio compared with previous work. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
