文章目录
- 题目：HLS-Based Optimization and Design Space Exploration for Applications with Variable Loop Bounds
- 时间：2018
- 会议：ICCAD
- 研究机构：UCLA Jason Cong
1 缩写 & 引用
2 abstract & introduction & previous work
如果循环的边界是变量决定的话，HLS无法进行很好的优化，因为循环边界不能在编译时决定，这里为了解决，主要需要两点：
- source to source HLS代码转换
- cycle and resource估计模型
- 具体研究了floating-point variable-loop reduction和prefix sum pattern
our work is more focused on optimizing these innermost loops by exploiting fine-grain parallelism and pipelining, accurately estimating resource sharing among these serial loops, and efficiently allocating non-sharable resource for overall latency minimization
3 HLS code transformation for variable-bound loop
变量循环边界一共有三种情况：完全并行、reduction、prefix sum
3.1 loop pipelining and loop unrolling based on the maximum loop bound
现行情况下，如果循环边界是变量，HLS通常只能pipeline，如果想要unroll，就只能按最大的边界展开，这就会导致PE利用率低
3.2 partial unrolling with pipelining
上述的代码重写之后就是下面的代码