Here's something interesting. Since i'm just sitting at a computer writing i plugged in my phone and set up an automated testing script and tried it out to compare various versions and compiler flags. Being able to test what improves the performance, and by how much, is going to be important going forward.

Here are the results in percentage full speed (three one-minute tests for each data point, averaged unless one was an outlier):
The perceived speedup going from 0.135 to 0.137 is indeed a reality (about 10% improvement), which is good news. I attributed it to the switch to C++ compiler above, but there's really no way of knowing.

As a reference, gauntlet was running at between 60-65% with version 0.135 before PR1.1 came along.

Definitely got a ways to go with these.

FYI i also ran with the -nosound option, but the results were not significantly different, as expected. I made a small change to the startup script already, so combined with the slight bump in speed from using arm-specific compiler flags it seemed worth an update. The autobuilder is chewing on it right now. Edit: and it's ready.

Edit: I forgot to mention that i dropped back to -o2 level to see if that would solve the problem with galaga88. No such luck. I've added a specific todo/roadmap list to the first post to keep track of what's up next.