HotakaYagi
commented
5 years ago 残り： "That causes taking much time for computation"
p1: "and excessively long computation times often result"
p1: "four double precisions of data are"
Figures: Plots are too small, especially font size. Figure 4 similarly contains pseudo-code which is very small and almost impossible to read if printed on paper, and the illustrations are overlapping in places.
The discussion of Roofline performance limits in Section 3 might benefit from drawing the corresponding Roofline diagram(s) as illustration.
The paper is difficult to read. Language and typesetting makes the text hard to parse, and in addition, the content and disposition of the article makes it hard to follow. Section 4 contains experimental results and is over half of the content, with many nested subheadings. The content should be clearly split up into theory and approach, with results clearly presented afterwards.
The importance of high-precision FP arithmetic could be motivated better, e.g. by providing example applications (with or without associated performance benchmark).
Several statements could be better supported by references, such as the discussion of loop nest structures for mat-vec and mat-mat multiplication.
Unclear: What is the number of cores used in the experiments?
Writing issues:
abstract: "synchlonization" "That causes taking much time for computation"
p1: "and excessively long computation times often result"
p1: "four double precisions of data are"
p1: "thread level parallelism" -> thread-level
p2: "qusai quadruple" -> quasi
p2: "but more costed" -> more expensive
p3: "but it enables to process using multicore" -> multi-core processing
p3: bytess
p3: "For experiment environment" -> add "the"
There are similar spelling and grammar issues in the rest of the paper. A language consultant should be involved as part of the revision.
Citations: Please put a blank space before citations.
Figures: Plots are too small, especially font size. Figure 4 similarly contains pseudo-code which is very small and almost impossible to read if printed on paper, and the illustrations are overlapping in places.
The discussion of Roofline performance limits in Section 3 might benefit from drawing the corresponding Roofline diagram(s) as illustration.