From: Andrew Pinski <pinskia@gmail.com>
To: Kugan Vivekanandarajah <kugan.vivekanandarajah@linaro.org>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [RFC] Tree Loop Unroller Pass
Date: Tue, 20 Feb 2018 19:53:00 -0000 [thread overview]
Message-ID: <CA+=Sn1mJn7dkknvkCun0KCjEhdzRM7ynY5UKLJFDSTg1B91s=Q@mail.gmail.com> (raw)
In-Reply-To: <CAELXzTNzMks8yg-vTc=m9txFQBXWp-UJt--9XnBxWS8sp43_pw@mail.gmail.com>
On Mon, Feb 12, 2018 at 3:55 PM, Kugan Vivekanandarajah
<kugan.vivekanandarajah@linaro.org> wrote:
> Implements tree loop unroller using the infrastructure provided.
>
> gcc/ChangeLog:
>
> 2018-02-12 Kugan Vivekanandarajah <kuganv@linaro.org>
>
> * Makefile.in (OBJS): Add tree-ssa-loop-unroll.o.
> * common.opt (ftree-loop-unroll): New option.
> * passes.def: Add pass_tree_loop_uroll
> * timevar.def (TV_TREE_LOOP_UNROLL): Add.
> * tree-pass.h (make_pass_tree_loop_unroll): Declare.
> * tree-ssa-loop-unroll.c: New file.
I think it was decided to name new gimple passes as gimple-* rather
than tree-ssa-*.
The option should most likely just take over -funrol-loops, etc.
Shouldn't:
+ if (targetm.hw_max_mem_read_streams
+ && (max_load_streams = targetm.hw_max_mem_read_streams ()) > 0)
+ {
+ load_streams = count_mem_load_streams (loop, max_load_streams);
+ if (load_streams > 0)
+ {
+ signed t = 1 << (floor_log2 (max_load_streams / load_streams));
+ if (t < n_unroll)
+ n_unroll = t;
+ }
+ }
this be a target hook instead of doing inline here. It seems too
target specific notion of what a stream is. Even the notion of a
stream here seems micro-arch specific and not generic enough.
A few more comments about the pass.
You don't check can_duplicate_loop_p on the loop.
You use optimize_function_for_size_p on the whole function instead of
checking if the loop is cold (by checking optimize_loop_for_size_p).
In my version of gimple loop unrolling, I had to add
lpt_decision.already_unrolled and mark the loop as already unrolled so
the rtl loop unroller would not happen again.
-fopt-report does not report when the unrolling has happened unlike
the RTL version.
Thanks,
Andrew
next prev parent reply other threads:[~2018-02-20 19:53 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-02-12 23:56 Kugan Vivekanandarajah
2018-02-20 19:53 ` Andrew Pinski [this message]
-- strict thread matches above, loose matches on Subject: below --
2018-02-13 13:05 [RFC] Tree loop unroller pass Wilco Dijkstra
2018-02-15 22:31 ` Kugan Vivekanandarajah
2018-02-16 11:56 ` Richard Biener
2018-02-16 14:22 ` Wilco Dijkstra
2018-02-16 15:00 ` Richard Biener
2018-02-16 15:32 ` Wilco Dijkstra
2018-02-20 15:46 ` Michael Matz
2018-02-20 0:17 ` Kugan Vivekanandarajah
2018-02-12 23:52 Kugan Vivekanandarajah
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='CA+=Sn1mJn7dkknvkCun0KCjEhdzRM7ynY5UKLJFDSTg1B91s=Q@mail.gmail.com' \
--to=pinskia@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=kugan.vivekanandarajah@linaro.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).