That was my intuition, since in order to inspect memory, all debug symbols need to be there. Apparently it builds then in debug mode too (which isn’t strictly necessary, you can enable symbols with -ggdb3 with -Os optimization, too…).
A classic that I have encountered here are delay loops which Keil can’t optimize away but GCC does, resulting in no delay. Have a look at the lower half of the post Build GD32 project with Platformio - #19 by maxgerhardt. The other gotcha with variables which are modified in ISRs (or other threads concurrently) but are not marked as volatile is the other big gotcha that will hurt when GCC optimizes the code.
Other than that it’s of course hard to tell without the code, but you seem to be on the case.