分段约束的超字并行向量发掘路径优化算法

超字并行(SLP)是一种针对基本块的向量并行发掘方法,结合循环展开可以发掘更多的并行性,但同时也会产生过多的发掘路径。针对上述问题,提出了一种分段约束的SLP发掘路径优化算法;采用分段的冗余删除方法,来保证冗余删除后段的同构性。采用段间的SLP发掘,来约束发掘路径;最后进行pack调整来处理访存重叠的情况。实验结果表明,该方法有效增强了SLP向量化功能,对于测试程序,向量化的平均加速比接近2。

Vector exploring path optimization algorithm of superword level parallelism with subsection constraints

Superword Level Parallelism (SLP) is a vector parallelism exploration approach for basic block. With loop unrolling, more parallel possibility can be explored. Simultaneously too much exploring paths are brought in. In order to solve the above problem, an optimized SLP method with subsection constraints was proposed. Redundant elimination on segmentation was used to obtain homogeneous segments. Inter-section exploring method based on SLP was used to restrain exploring paths and reduce the complexity of the algorithm. And finally pack adjustment was used to deal with the situation of overlap on memory access. The experimental results show that the vectorization capability of SLP is enhanced; for the test serial program, the average speedup of vectorized version is close to 2.