Systemtap global variable can be translated to constant immediate value(or just a macro) if no one does write access to it(or, const suffix like as global XXX:const).
(In reply to comment #0) > Systemtap global variable can be translated to constant immediate value(or just > a macro) if no one does write access to it This should be straightforward. We already know when locking is unneeded for just such globals: bug #2341.
(In reply to comment #1) > This should be straightforward. We already know when locking is unneeded for just > such globals: bug #2341. Sure, this request is more advanced one. We expect some formula including immediate values are compounded to simpler formula. For example, --- global TICKS:const=100 global HZ:const=1000 function ticktohz:long (tick) { return tick * HZ / TICKS; # this will be translated to tick * 10 } ---
> Sure, this request is more advanced one. We expect some formula including > immediate values are compounded to simpler formula. For example, > --- > global TICKS:const=100 > global HZ:const=1000 > > function ticktohz:long (tick) { > return tick * HZ / TICKS; # this will be translated to tick * 10 > } Reusing the same analysis already done for lock elision, it should not be necessary to designate anything with ":const" for this treatment.
Note that this conflicts with bug #5642 and existing staprun GLOBAL=VALUE functionality, since globals can be reinitialized later via staprun. That means we couldn't literally substitute them at translate time.
(In reply to comment #4) > Note that this conflicts with bug #5642 and existing staprun GLOBAL=VALUE > functionality, since globals can be reinitialized later via staprun. > That means we couldn't literally substitute them at translate time. Hmm, even though, as I mentioned in comment #1, const suffix or something like that can solve that conflict. (or, const XXXXX=NNNN instead of global XXXXX:const=NNNNN)
this may now be implemented via macros @define const_global_foo %( 1 %) probe begin { println(@const_global_foo) }