My last patch did break runops_slow_core (first instruction got traced)
- fixed.
And finally I found an hopefully sane way, to implement the idea from
#17495:
- stacktop is initially set in test_main
- if running the CGgoto core and it's cleared in interpreter
- in cg_core stacktop get's set again beyond the opfunc jump table.
This speeds up life.pasm from 325 to 406 on my 800 Mhz Athlon.
This should finally need a test, if the compiler supports nested
functions (if no nested funtions are available, a save way to put
lo_var_ptr beyond the jump table in core_ops_cg.c (aka ops2cg.pl)).
leo