Re: [ping #4][patch] Fix PR80929: Realistic PARALLEL cost in seq_cost.

[Segher Boessenkool <segher@kernel.crashing.org>]

On Wed, Jul 12, 2017 at 03:30:00PM +0200, Georg-Johann Lay wrote:
> On 12.07.2017 14:11, Segher Boessenkool wrote:
> >On Tue, Jul 11, 2017 at 10:47:27AM +0200, Georg-Johann Lay wrote:
> >>This small addition improves costs of PARALLELs in
> >>rtlanal.c:seq_cost().  Up to now, these costs are
> >>assumed to be 1 which gives gross inexact costs for,
> >>e.g. divmod which is represented as PARALLEL.
> >
> >insn_rtx_cost returns 0 ("unknown") for such a PARALLEL, so your
> >current patch does not change this at all?
> 
> Huh?  It returns the costs of 1st SET in a PARALLEL (provided it
> has one), no?  Or even costs for come compares.

No, it returns 0 if there is more than one normal SET (or more than
one compare).

> >>+      else if (INSN_P (seq)
> >>+	       && PARALLEL == GET_CODE (PATTERN (seq)))
> >
> >Yoda conditions have we should not.
> 
> hmm, I didn't find something like PARALLEL_P (rtx).
> Is comparing rtx_codes deprecated now?

I meant it should be written

  else if (INSN_P (seq) && GET_CODE (PATTERN (seq)) == PARALLEL)

i.e. constant on the right.

> >This whole thing could be something like
> >
> >   if (INSN_P (seq))
> >     {
> >       int t = insn_rtx_cost (PATTERN (seq), speed);
> 
> This will behave differently.

Yes, I know, I even said so.

> >(Why do you need a check for INSN_P here?  Why does it increment the
> 
> >cost for non-insns?  So many questions).
> 
> Again, you'll have to ask the original author for reasoning.

Since you want to change the code, to make it better, I was hoping
you would dig in a bit.  To make it better, not just different :-/


Segher