[RFC] Kill gen_sequence

Joern Rennecke amylaar@onetel.net.uk
Sat Jun 8 18:13:00 GMT 2002

David S. Miller wrote:
>    From: Joern Rennecke <amylaar@onetel.net.uk>
>    Date: Sun, 9 Jun 2002 01:55:55 +0100 (BST)
>    I agree that we don't need SEQUENCE rtl.  But I disagree with throwing
>    away information that we need again straight ahead, i.e. a pointer to
>    the last insn.
> 9 times out of ten the information is still there, via
> get_last_insn().  And like I said the other cases can be handled
> too.

The information isn't still there.  end_sequence restores last_insn
to what it was before the matching call to start_sequence.

> We both agree that the information needs to be passed around,
> I'm saying using throw away RTL is not the way we should do this.
> Your response was to use some kind of cache, and I said that
> fixed limitations in sequence depth is not acceptable.

No, using a cache doesn't imply a limit in sequence depth.  But that is
moot, since there is a simpler way:
having end_sequence return the struct sequence_stack.

> Can we just install my change and move on to the next thing?

Your patch is removing useful functionality, and to put it back we'd
have to patch largely the same places again.

> We are arguing in circles at this point and it is at least mildly
> frustrating. 

So could you please take notice of my proposal to use the struct
sequence_stack to pass the information about the sequence.

More information about the Gcc-patches mailing list