From: Richard Sandiford <richard.sandiford@arm.com>
To: Jeff Law via Gcc-patches <gcc-patches@gcc.gnu.org>
Subject: Re: [PATCH 6/6] Extend SLP permutation optimisations
Date: Tue, 30 Aug 2022 15:50:44 +0100 [thread overview]
Message-ID: <mpt8rn5eqez.fsf@arm.com> (raw)
Message-ID: <20220830145044.YKQYbHtRgqGG1oNJuU7OAMbSMrrgJTGhoTuiWQFcXIk@z> (raw)
In-Reply-To: <1106b505-65b0-4681-c53b-fe7579490e92@gmail.com> (Jeff Law via Gcc-patches's message of "Fri, 26 Aug 2022 10:26:13 -0600")
Jeff Law via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> On 8/25/2022 7:07 AM, Richard Sandiford via Gcc-patches wrote:
>> Currently SLP tries to force permute operations "down" the graph
>> from loads in the hope of reducing the total number of permutations
>> needed or (in the best case) removing the need for the permutations
>> entirely. This patch tries to extend it as follows:
>>
>> - Allow loads to take a different permutation from the one they
>> started with, rather than choosing between "original permutation"
>> and "no permutation".
>>
>> - Allow changes in both directions, if the target supports the
>> reverse permutation.
>>
>> - Treat the placement of permutations as a two-way dataflow problem:
>> after propagating information from leaves to roots (as now), propagate
>> information back up the graph.
>>
>> - Take execution frequency into account when optimising for speed,
>> so that (for example) permutations inside loops have a higher
>> cost than permutations outside loops.
>>
>> - Try to reduce the total number of permutations when optimising for
>> size, even if that increases the number of permutations on a given
>> execution path.
>>
>> See the big block comment above vect_optimize_slp_pass for
>> a detailed description.
>>
>> The original motivation for doing this was to add a framework that would
>> allow other layout differences in future. The two main ones are:
>>
>> - Make it easier to represent predicated operations, including
>> predicated operations with gaps. E.g.:
>>
>> a[0] += 1;
>> a[1] += 1;
>> a[3] += 1;
>>
>> could be a single load/add/store for SVE. We could handle this
>> by representing a layout such as { 0, 1, _, 2 } or { 0, 1, _, 3 }
>> (depending on what's being counted). We might need to move
>> elements between lanes at various points, like with permutes.
>>
>> (This would first mean adding support for stores with gaps.)
>>
>> - Make it easier to switch between an even/odd and unpermuted layout
>> when switching between wide and narrow elements. E.g. if a widening
>> operation produces an even vector and an odd vector, we should try
>> to keep operations on the wide elements in that order rather than
>> force them to be permuted back "in order".
> Very cool. Richi and I discussed this a bit a year or so ago --
> basically noting that bi-directional movement is really the way to go
> and that the worst thing to do is push a permute down into the *middle*
> of a computation chain since that will tend to break FMA generation.
> Moving to the loads or stores or to another permute point ought to be
> the goal.
Hmm, I hadn't thought specifically about the case of permutes
ending up in the middle of a fusable operation. The series doesn't
address that directly. If we have:
permute(a) * permute(b) + c
then the tendency will still be to convert that into:
permute(a * b) + c
Damn. Another case to think about ;-)
I've pushed the series for now though (thanks to Richi for the reviews).
Richard
next prev parent reply other threads:[~2022-08-30 14:50 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-08-25 13:04 [PATCH 0/6] Optimise placement of SLP permutations Richard Sandiford
2022-08-25 13:05 ` [PATCH 1/6] Split code out of vectorizable_slp_permutation Richard Sandiford
2022-08-25 13:05 ` [PATCH 2/6] Split code out of vect_transform_slp_perm_load Richard Sandiford
2022-08-25 13:05 ` [PATCH 3/6] Make graphds_scc pass the node order back to callers Richard Sandiford
2022-08-25 13:06 ` [PATCH 4/6] Rearrange unbounded_hashmap_traits Richard Sandiford
2022-08-25 13:06 ` [PATCH 5/6] Add base hash traits for vectors Richard Sandiford
2022-08-25 13:07 ` [PATCH 6/6] Extend SLP permutation optimisations Richard Sandiford
2022-08-26 16:26 ` Jeff Law
2022-08-30 14:50 ` Richard Sandiford [this message]
2022-08-30 14:50 ` Richard Sandiford
2022-08-31 14:38 ` Jeff Law
2022-08-26 9:25 ` [PATCH 0/6] Optimise placement of SLP permutations Richard Biener
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=mpt8rn5eqez.fsf@arm.com \
--to=richard.sandiford@arm.com \
--cc=gcc-patches@gcc.gnu.org \
/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).