Applying Code Transform Model to Newly Generated Program for Improving Execution Performance

<table class="table-group" id="tab31"><tr><td><table class="table"><tr><td class="thead-hr" colspan="4"><hr/></td></tr><tr class="thead"><td class="align_left">Time or percentage</td><td class="align_center">Java-codetool</td><td class="align_center">CodeGeneration</td><td class="align_center">Proposed approach with GPT-2</td></tr><tr><td class="thead-hr" colspan="4"><hr/></td></tr><tr><td class="align_left">Average time of generating a single instruction (s)</td><td class="align_center">0.17</td><td class="align_center">0.15</td><td class="align_center">0.01</td></tr><tr><td class="align_left">Reduced the percentage of code lines (%)</td><td class="align_center">0%</td><td class="align_center">0%</td><td class="align_center">24.34%</td></tr><tr><td class="align_left">Conformity of program execution results (%)</td><td class="align_center">—</td><td class="align_center">—</td><td class="align_center">97.60%</td></tr><tr class="table-tr"><td colspan="4"><hr class="tbody-hr"/></td></tr></table></td></tr><tr class="table-fn"><td><div>The symbol “−” stands for the information that is not available so far.<br/></div></td></tr></table>

<div>Performance evaluation of code generation model.</div>

Scientific Programming

tab31

Table 31

Table 31: Applying Code Transform Model to Newly Generated Program for Improving Execution Performance 