This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Remove all CSE skip blocks code

Hi Steven,

On Tue, 21 Nov 2006, Steven Bosscher wrote:
> 	* cse.c (enum taken): Remove PATH_AROUND.
> 	(addr_affects_sp_p, invalidate_skipped_set,
> 	invalidate_skipped_block): Remove.
> 	(cse_end_of_basic_block): Remove skip_blocks and related code.
> 	(cse_main): Don't test for flag_cse_skip_blocks.
> 	Update cse_end_of_basic_block call.
> 	(cse_basic_block): Likewise.  Remove PATH_AROUND case.  Remove
> 	code to lengthen the path if a jump was simplified.

This is OK for mainline, thanks.  I think we can defer a decision on
the -fcse-skip-blocks for a little while, until we see how things
progress with the CSE rewrite.  We might end up with a "pseudos live
at start" implementation that provides functionality equivalent to
path following.  In which case, it might make sense to retain the
historical option names.  Alternatively, future stage 1 changes may
deprecate other CSE/GCSE command line options, and it may make sense
to decide upon their fate at the same time.  As long as we have a
consistent strategy (and documentation) going into stage 2, I'm happy
to let things diverge temporarily in the meantime.

I'm not sure I agree with Richard's proposal of deprecating this command
line option for 4.2, where this option still does something useful.  Even
if it will go away.  I think we'll probably keep it as a no-op with a
deprecated warning for 4.3, and then remove it for 4.4 (as with Zdenek's
loop changes).


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]