From: Andrew MacLeod <amacleod@redhat.com>
To: Aldy Hernandez <aldyh@redhat.com>,
Richard Biener <richard.guenther@gmail.com>
Cc: GCC patches <gcc-patches@gcc.gnu.org>
Subject: Re: [COMMITTED] Convert nonzero mask in irange to wide_int.
Date: Tue, 4 Oct 2022 09:27:20 -0400 [thread overview]
Message-ID: <55c5aebd-6b51-982b-3dc7-73513c727f58@redhat.com> (raw)
In-Reply-To: <CAGm3qMW+8NjcFu99pkp5w2y9gdDHHfZ4Vfx6KEVQ4PdqkH8b3g@mail.gmail.com>
On 10/4/22 08:13, Aldy Hernandez via Gcc-patches wrote:
> On Tue, Oct 4, 2022, 13:28 Aldy Hernandez <aldyh@redhat.com> wrote:
>
>> On Tue, Oct 4, 2022 at 9:55 AM Richard Biener
>> <richard.guenther@gmail.com> wrote:
>>>
>>> Am 04.10.2022 um 09:36 schrieb Aldy Hernandez via Gcc-patches <
>> gcc-patches@gcc.gnu.org>:
>>>> The reason the nonzero mask was kept in a tree was basically inertia,
>>>> as everything in irange is a tree. However, there's no need to keep
>>>> it in a tree, as the conversions to and from wide ints are very
>>>> annoying. That, plus special casing NULL masks to be -1 is prone
>>>> to error.
>>>>
>>>> I have not only rewritten all the uses to assume a wide int, but
>>>> have corrected a few places where we weren't propagating the masks, or
>>>> rather pessimizing them to -1. This will become more important in
>>>> upcoming patches where we make better use of the masks.
>>>>
>>>> Performance testing shows a trivial improvement in VRP, as things like
>>>> irange::contains_p() are tied to a tree. Ughh, can't wait for trees in
>>>> iranges to go away.
>>> You want trailing wide int storage though. A wide_int is quite large.
>> Absolutely, this is only for short term storage. Any time we need
>> long term storage, say global ranges in SSA_NAME_RANGE_INFO, we go
>> through vrange_storage which will stream things in a more memory
>> efficient manner. For irange, vrange_storage will stream all the
>> sub-ranges, including the nonzero bitmask which is the first entry in
>> such storage, as trailing_wide_ints.
>>
>> See irange_storage_slot to see how it lives in GC memory.
>>
> That being said, the ranger's internal cache uses iranges, albeit with a
> squished down number of subranges (the minimum amount to represent the
> range). So each cache entry will now be bigger by the difference between
> one tree and one wide int.
>
> I wonder if we should change the cache to use vrange_storage. If not now,
> then when we convert all the subranges to wide ints.
>
> Of course, the memory pressure of the cache is not nearly as problematic as
> SSA_NAME_RANGE_INFO. The cache only stores names it cares about.
Rangers cache can be a memory bottleneck in pathological cases..
Certainly not as bad as it use to be, but I'm sure it can still be
problematic. Its suppose to be a memory efficient representation
because of that. The cache can have an entry for any live ssa-name
(which means all of them at some point in the IL) multiplied by a factor
involving the number of dominator blocks and outgoing edges ranges are
calculated on. So while SSA_NAME_RANGE_INFO is a linear thing, the
cache lies somewhere between a logarithmic and exponential factor based
on the CFG size.
if you are growing the common cases of 1 to 2 endpoints to more than
double in size (and most of the time not be needed), that would not be
very appealing :-P If we have any wide-ints, they would need to be a
memory efficient version. The Cache uses an irange_allocator, which is
suppose to provide a memory efficient objects.. hence why it trims the
number of ranges down to only what is needed. It seems like a trailing
wide-Int might be in order based on that..
Andrew
PS. which will be more problematic if you eventually introduce a
known_ones wide_int. I thought the mask tracking was/could be
something simple like HOST_WIDE_INT.. then you only tracks masks in
types up to the size of a HOST_WIDE_INT. then storage and masking is
all trivial without going thru a wide_int. Is that not so/possible?
> Aldy
>
next prev parent reply other threads:[~2022-10-04 13:27 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-10-04 7:35 Aldy Hernandez
2022-10-04 7:55 ` Richard Biener
2022-10-04 11:28 ` Aldy Hernandez
2022-10-04 12:13 ` Aldy Hernandez
2022-10-04 13:27 ` Andrew MacLeod [this message]
2022-10-04 14:30 ` Aldy Hernandez
2022-10-04 14:34 ` Richard Biener
2022-10-04 15:14 ` Aldy Hernandez
2022-10-04 15:42 ` Andrew MacLeod
2022-10-05 10:14 ` Aldy Hernandez
2022-10-07 9:23 ` Aldy Hernandez
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=55c5aebd-6b51-982b-3dc7-73513c727f58@redhat.com \
--to=amacleod@redhat.com \
--cc=aldyh@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=richard.guenther@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).