Re: [RFD] MIPS/gas: Optimisation cannot be set to 0

[Thiemo Seufer <ths at networkno dot de>]

Maciej W. Rozycki wrote:
> Hello,
> 
>  Browsing through gas code I have noticed this snippet:
> 
> /* Whether we are optimizing.  The default value of 2 means to remove
>    unneeded NOPs and swap branch instructions when possible.  A value
>    of 1 means to not swap branches.  A value of 0 means to always
>    insert NOPs.  */
> static int mips_optimize = 2;
> 
> and all the three values are checked against throughout.  However, it is 
> not possible to set the variable to 0 -- the only piece of code doing 
> initialisation is this:
> 
>     case 'O':
>       if (arg && arg[0] == '0')
> 	mips_optimize = 1;
>       else
> 	mips_optimize = 2;
>       break;
> 
> in md_parse_option().  It looks like it has been like this since our CVS 
> tree was born.  There is this ChangeLog entry which may be of relevance 
> though:
> 
> Wed Feb 12 14:36:29 1997  Ian Lance Taylor  <ian@cygnus.com>
> 
> 	* config/tc-mips.c (md_parse_option): When debugging, set
> 	mips_optimize to 1, not 0.
> 
>  Anyway, the outcome is we have dead code and unclear semantics.  The most 
> obvious solution is mapping the argument of -O directly to the value of 
> mips_optimize.  The drawback is changed semantics.  Another possibility is 
> getting rid of what currently mips_optimize == 1 means entirely.  And 
> there may be more reasonable options available.
> 
>  I have no strong preference, but I think gas should have a mode where its 
> input is assembled intact.  And swapping branches may produce surprising 
> interactions with debugging information.  I am therefore in favour to the 
> first proposal above, but I would like to hear from the others.

I also prefer the first option, given that gcc carefully avoids to pass
-O0 in most cases:

#ifndef SUBTARGET_ASM_OPTIMIZING_SPEC
#define SUBTARGET_ASM_OPTIMIZING_SPEC "\
%{noasmopt:-O0} \
%{!noasmopt:%{O:-O2} %{O1:-O2} %{O2:-O2} %{O3:-O3}}"
#endif

The noasmopt option is undocumented.


Thiemo