<html><body><p><font size="2">--------------------</font><br><font size="2">Wed, March 25, 2020:</font><br><font size="2">--------------------</font><br><br><font size="2">- Input Dependencies in DDG</font><br><font size="2"> - Input edges are not necessary for semantic correctness but are</font><br><font size="2"> useful for developing profitability heuristics</font><br><font size="2"> - They can cause cycles in the graph or impose ordering</font><br><font size="2"> constraints unnecessarily</font><br><font size="2"> - After considering various alternatives decided to explore</font><br><font size="2"> the following approach:</font><br><font size="2"> - Do not establish input dependence edges, but provide the pair</font><br><font size="2"> of input dependent nodes in a side table in the DDG so that </font><br><font size="2"> interested consumers can look them up.</font><br><br><font size="2">- Unroll-and-Jam (</font><a href="https://reviews.llvm.org/D76132"><font size="2">https://reviews.llvm.org/D76132</font></a><font size="2">)</font><br><font size="2"> - Legality checks using dependence info still waiting for review.</font><br><font size="2"> - It may be useful to call a meeting among active participants</font><br><font size="2"> in the code review to discuss issues and reach agreement.</font><br><br><font size="2">- Dependence Analysis (</font><a href="https://reviews.llvm.org/D75628"><font size="2">https://reviews.llvm.org/D75628</font></a><font size="2">)</font><br><font size="2"> - Agreed to contact Phillip to seek more advice on potential</font><br><font size="2"> improvements to SCEV and/or dependence analysis.</font><br><br><font size="2">Bardia Mahjour<br>Compiler Optimizations<br>IBM Toronto Software Lab</font><BR>
</body></html>