[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [f-cpu] stats on insn pairing



Sorry I was in hurry. I did data flow analysis on gcc
internal repesetation after last scheduling and computed:
1) avg fanout, how many insn uses result of single insn
2) fanout histogram
3) register value producer-consumer pairs
4) same as 3 but distinguishing insns with and without
   immediate values

Notes:
store-store link is because of post-increment dependency
const_int-plus is adding of const in 2 insns because of
        too big value
move-move is value copying (creating bigger fanout)

hope it is more clear,
devik

On Fri, 21 Feb 2003, Michael Riepe wrote:

> On Fri, Feb 21, 2003 at 04:37:03PM +0100, devik wrote:
>
> > while discussing with nico I wanted to know read insn-insn
> > pairing and value fanout. I computed stats for gcc source
> > and linux kernel one.
> > They are attached ...
>
> Does `pair' mean that the second insn consumes a value produced by the
> first insn?  Or does it just follow immediately in the instruction
> stream?
>
> --
>  Michael "Tired" Riepe <Michael.Riepe@stud.uni-hannover.de>
>  "All I wanna do is have a little fun before I die"
> *************************************************************
> To unsubscribe, send an e-mail to majordomo@seul.org with
> unsubscribe f-cpu       in the body. http://f-cpu.seul.org/
>


*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/