Assembler works.

C works in simulator.

Full speed on target does not work, unless memory init is excluded.

So what's the problem? Did your assembler version have a memory initialization loop like STARTUP.A51?

Is your code depending upon memory locations being cleared to 00h? Or perhaps NOT being cleared (some non-0 random value that makes it work)?

Does your target have a watchdog timer that's timing out during the memory init?

I'm out of ideas.
Rgs,
Steve