From: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>
To: Kyrill Tkachov <kyrylo.tkachov@foss.arm.com>,GCC Patches
<gcc-patches@gcc.gnu.org>
Cc: Richard Biener <rguenther@suse.de>
Subject: Re: [PATCH][v3] GIMPLE store merging pass
Date: Wed, 07 Sep 2016 20:44:00 -0000 [thread overview]
Message-ID: <B678985F-4595-4EA2-A69B-C939FECA5AC9@gmail.com> (raw)
In-Reply-To: <57CEDD67.6010801@foss.arm.com>
On September 6, 2016 5:14:47 PM GMT+02:00, Kyrill Tkachov <kyrylo.tkachov@foss.arm.com> wrote:
>Hi all,
s/contigous/contiguous/
s/ where where/ where/
+struct merged_store_group
+{
+ HOST_WIDE_INT start;
+ HOST_WIDE_INT width;
+ unsigned char *val;
+ unsigned int align;
+ auto_vec<struct store_immediate_info *> stores;
+ /* We record the first and last original statements in the sequence because
+ because we'll need their vuse/vdef and replacement position. */
+ gimple *last_stmt;
s/ because because/ because/
Why aren't these two HWIs unsigned, likewise in store_immediate_info and in most other spots in the patch?
+ fprintf (dump_file, "Afer writing ");
s/Afer /After/
/access if prohibitively slow/s/ if /is /
I'd get rid of successful_p in imm_store_chain_info::output_merged_stores.
+unsigned int
+pass_store_merging::execute (function *fun)
+{
+ basic_block bb;
+ hash_set<gimple *> orig_stmts;
+
+ FOR_EACH_BB_FN (bb, fun)
+ {
+ gimple_stmt_iterator gsi;
+ HOST_WIDE_INT num_statements = 0;
+ /* Record the original statements so that we can keep track of
+ statements emitted in this pass and not re-process new
+ statements. */
+ for (gsi = gsi_after_labels (bb); !gsi_end_p (gsi); gsi_next (&gsi))
+ {
+ gimple_set_visited (gsi_stmt (gsi), false);
+ num_statements++;
+ }
+
+ if (num_statements < 2)
+ continue;
What about debug statements? ISTM you should skip those.
(Isn't visited reset before entry of a pass?)
Maybe I missed the bikeshedding about the name but I'd have used -fmerge-stores instead.
Thanks,
>
>The v3 of this patch addresses feedback I received on the version
>posted at [1].
>The merged store buffer is now represented as a char array that we
>splat values onto with
>native_encode_expr and native_interpret_expr. This allows us to merge
>anything that native_encode_expr
>accepts, including floating point values and short vectors. So this
>version extends the functionality
>of the previous one in that it handles floating point values as well.
>
>The first phase of the algorithm that detects the contiguous stores is
>also slightly refactored according
>to feedback to read more fluently.
>
>Richi, I experimented with merging up to MOVE_MAX bytes rather than
>word size but I got worse results on aarch64.
>MOVE_MAX there is 16 (because it has load/store register pair
>instructions) but the 128-bit immediates that we ended
>synthesising were too complex. Perhaps the TImode immediate store RTL
>expansions could be improved, but for now
>I've left the maximum merge size to be BITS_PER_WORD.
>
>I've disabled the pass for PDP-endian targets as the merging code
>proved to be quite fiddly to get right for different
>endiannesses and I didn't feel comfortable writing logic for
>BYTES_BIG_ENDIAN != WORDS_BIG_ENDIAN targets without serious
>testing capabilities. I hope that's ok (I note the bswap pass also
>doesn't try to do anything on such targets).
>
>Tested on arm, aarch64, x86_64 and on big-endian arm and aarch64.
>
>How does this version look?
>Thanks,
>Kyrill
>
>[1] https://gcc.gnu.org/ml/gcc-patches/2016-08/msg01512.html
>
>2016-09-06 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
>
> PR middle-end/22141
> * Makefile.in (OBJS): Add gimple-ssa-store-merging.o.
> * common.opt (fstore-merging): New Optimization option.
> * opts.c (default_options_table): Add entry for
> OPT_ftree_store_merging.
> * params.def (PARAM_STORE_MERGING_ALLOW_UNALIGNED): Define.
> * passes.def: Insert pass_tree_store_merging.
> * tree-pass.h (make_pass_store_merging): Declare extern
> prototype.
> * gimple-ssa-store-merging.c: New file.
> * doc/invoke.texi (Optimization Options): Document
> -fstore-merging.
>
>2016-09-06 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
> Jakub Jelinek <jakub@redhat.com>
>
> PR middle-end/22141
> * gcc.c-torture/execute/pr22141-1.c: New test.
> * gcc.c-torture/execute/pr22141-2.c: Likewise.
> * gcc.target/aarch64/ldp_stp_1.c: Adjust for -fstore-merging.
> * gcc.target/aarch64/ldp_stp_4.c: Likewise.
> * gcc.dg/store_merging_1.c: New test.
> * gcc.dg/store_merging_2.c: Likewise.
> * gcc.dg/store_merging_3.c: Likewise.
> * gcc.dg/store_merging_4.c: Likewise.
> * gcc.dg/store_merging_5.c: Likewise.
> * gcc.dg/store_merging_6.c: Likewise.
> * gcc.target/i386/pr22141.c: Likewise.
> * gcc.target/i386/pr34012.c: Add -fno-store-merging to dg-options.
next prev parent reply other threads:[~2016-09-07 20:41 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-09-06 15:16 Kyrill Tkachov
2016-09-06 15:33 ` Jakub Jelinek
2016-09-06 16:21 ` Kyrill Tkachov
2016-09-06 16:34 ` Jakub Jelinek
2016-09-06 16:38 ` Kyrill Tkachov
2016-09-07 9:11 ` Richard Biener
2016-09-07 12:43 ` Jeff Law
2016-09-07 13:32 ` Bernd Schmidt
2016-09-07 20:47 ` Jakub Jelinek
2016-09-07 20:44 ` Bernhard Reutner-Fischer [this message]
2016-09-08 8:54 ` Kyrill Tkachov
2016-09-08 15:47 ` Bernhard Reutner-Fischer
2016-09-13 9:47 ` Kyrill Tkachov
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=B678985F-4595-4EA2-A69B-C939FECA5AC9@gmail.com \
--to=rep.dot.nop@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=kyrylo.tkachov@foss.arm.com \
--cc=rguenther@suse.de \
/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).