This is the mail archive of the gcc@gcc.gnu.org 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]

Deprecating define_function_unit?


Hello,

Last month there were 160 define_function_units in all of GCC,
spread out over 11 different backends.  Today, there are only
76 left, in only 3 different backends:

arc/arc.md:2
mips/mips.md:62
sh/sh.md:12

The SH one is for SH5media; apparently a DFA pipeline description
for this CPU already exists outside of the FSF tree.

The MIPS ones are a bit harder, but if I understand correctly, most
of these MIPSen are not very important (they're old), or people are
working on writing new pipeline descriptions for them.

The ARC one, well...  I don't know anything about ARC, I can't find
any info about it.  It uses the 7th operand of define_function_unit
(the conflict litst), which makes it a bit more difficult to translate
to a new pipeline description.  It can be done if someone has some
documentation about that CPU.
Perhaps this is a good moment for the new ARC maintainers to step up
and help out ;-)  (it's only two define_function_units anyway!)

Anyway, I would like to have define_function_unit deprecated, perhaps
even removed before GCC 3.5 depending on how things work out for SH
and MIPS.  The benefit is ~1800 lines of code removed, mostly in the
gen* files and in haifa-sched.c (plus many many lines of sched_reorg
target hooks, but those are already gone).
With stage2 and stage3 ahead of us, I suppose it shouldn't be too much
of a problem for SH and MIPS to remove their old pipeline descriptions,
but obviously it's not _that_ urgent.

So.  Thoughts?  Can I add a note to the documentation (and in rtl.def)
that define_function_unit is deprecated?

Gr.
Steven



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