Decoupling -O from -fvia-C

[Simon Marlow]

Wolfgang Thaller wrote:
> I think it is again time to discuss whether -O should still imply - fvia-C.
> 
> I have recently committed a few patches that allow the NCG to compile  
> the RTS. Position-independent code works in the NCG on Darwin/x86 and  
> (almost untested) on Linux/x86, but will never work in for -fvia-C  
> compilation. So now there are situations where -fasm is required, and  
> there are fewer and fewer situations where -fvia-C is required.
> 
> The Makefile in compiler/ adds -O to the HC_OPTS for some files,  which 
> overrides any global -fasm, so compiling a libHSghc_dyn.dylib  (or .so) 
> requires editing the Makefile.
> 
> The only argument for having -O imply -fvia-C that I am aware of is  
> that -fvia-C code is slightly faster than -fasm on x86. This was not  
> even the case on PowerPC, last time I checked.

Yes, this has been at the back of my mind for a while.  We should change this. 
Please go forth and commit...

Cheers,
	Simon