TY  - GEN
T1  - Assembly program performance analysis metrics instructions performed and program latency exemplified on loop unroll
A1  - Cempron, Jonathan Paul
LA  - English
YR  - 2018
UL  - https://tuklas.up.edu.ph/Record/UP-1685594773862388556
AB  - Software program optimization for improved execution speed can be achieved through modifying the program. Said program is usually written in High-Level languages then later translated into Low-Level languages, a language specific to the processor used. A larger coverage of optimization can be achieved through optimizing in Low-Level Language ? rather than in the High-Level language ? because all High-Level languages are eventually translated to Low-Level. One method that has been used in the past is Loop Unrolling, which is done by transforming iterative looping blocks into longer sequential code blocks. This method of optimization increases code length but reduces branching instructions and the latencies introduced by said instructions. However, measuring the performance difference between the original code against the loop unroll optimized code cannot be exposed using current static performance metrics, which rely on IC. Alternative metrics ? Instructions Performed and Instruction Latency ? are proposed for examining the effectivity of optimization due to the limitations in traditional metrics based on IC. As an extension of loop unrolling, its specific explanation in this paper is discussed as a pre-processor for autovectorization. The specific methods of vectorization, however, will not be a part of this paper?s scope.
CN  - ARTICLE-2741
KW  - Computer science.
KW  - Assembly programming.
KW  - Instruction set architecture.
KW  - Loop unroll.
KW  - Metric.
KW  - Optimization.
KW  - Vectorization.
ER  -