On Wed, 27 Jan 2021, Jakub Jelinek wrote:
> On Wed, Jan 27, 2021 at 03:40:38PM +0100, Richard Biener wrote:
> > The following avoids repeatedly turning VALUE RTXen into
> > sth useful and re-applying a constant offset through get_addr
> > via DSE check_mem_read_rtx. Instead perform this once for
> > all stores to be visited in check_mem_read_rtx. This avoids
> > allocating 1.6GB of garbage PLUS RTXen on the PR80960
> > testcase, fixing the memory usage regression from old GCC.
> >
> > Bootstrap and regtest running on x86_64-unknown-linux-gnu, OK?
> >
> > Thanks,
> > Richard.
> >
> > 2021-01-27 Richard Biener <rguent...@suse.de>
> >
> > PR rtl-optimization/80960
> > * dse.c (check_mem_read_rtx): Call get_addr on the
> > offsetted address.
> > ---
> > gcc/dse.c | 5 +++++
> > 1 file changed, 5 insertions(+)
> >
> > diff --git a/gcc/dse.c b/gcc/dse.c
> > index c88587e7d94..da0df54a2dd 100644
> > --- a/gcc/dse.c
> > +++ b/gcc/dse.c
> > @@ -2219,6 +2219,11 @@ check_mem_read_rtx (rtx *loc, bb_info_t bb_info)
> > }
> > if (maybe_ne (offset, 0))
> > mem_addr = plus_constant (get_address_mode (mem), mem_addr, offset);
> > + /* Avoid passing VALUE RTXen as mem_addr to canon_true_dependence
> > + which will over and over re-create proper RTL and re-apply the
> > + offset above. See PR80960 where we almost allocate 1.6GB of PLUS
> > + RTXen that way. */
> > + mem_addr = get_addr (mem_addr);
> >
> > if (group_id >= 0)
> > {
>
> Does that result in any changes on how much does DSE optimize?
> I mean, if you do 2 bootstraps/regtests, one with this patch and another one
> without it, and at the end of rest_of_handle_dse dump
> locally_deleted, globally_deleted
> for each CU/function, do you get the same counts except perhaps for dse.c?
I can check but all immediate first uses of mem_addr are in
true_dependece_1 which does x_addr = get_addr (x_addr); as the
first thing on it. So the concern would be that
get_addr (get_addr (x_addr)) != get_addr (x_addr) which I think
shouldn't be the case (no semantic difference at least).
I'll see to do the test tomorrow.
Thanks,
Richard.
