From: "Lehua Ding" <lehua.ding@rivai.ai>
To: "Jeff Law" <jeffreyalaw@gmail.com>,
gcc-patches <gcc-patches@gcc.gnu.org>
Cc: "juzhe.zhong" <juzhe.zhong@rivai.ai>,
"rdapp.gcc" <rdapp.gcc@gmail.com>,
"kito.cheng" <kito.cheng@gmail.com>, palmer <palmer@rivosinc.com>
Subject: Re: [PATCH] RISC-V: Fix error combine of pred_mov pattern
Date: Sat, 12 Aug 2023 00:30:03 +0800 [thread overview]
Message-ID: <D667ADA3F89EED6D+tencent_D1FA83EE7A6E261CF9A89E7BF785B3D38E05@qq.com> (raw)
In-Reply-To: <680c9180-6db6-0dd7-d93a-aa12cdfbbbac@gmail.com>
[-- Attachment #1: Type: text/plain, Size: 4465 bytes --]
> But combine doesn't run at -O0. So something is inconsistent. I
> certainly believe we need to avoid the mem->mem case, but that's
> independent of combine and affects all optimization levels.
This is an new bug when running all tests after fixing the combine bug.
I understand that maybe I should send a separate patch to fix the problem.
Maybe this problem was exposed after I changed the pattern. I will continue to track it.
> I think we can simplify to just
> !(MEM_P (operands[0]) && MEM_P (operands[1])
> I would have expected those to be handled by the constraints rather than
> the pattern's condition.
Yeh, the condition of the V2 becomes much simpler after split.
------------------ Original ------------------
From: "Jeff Law" <gcc-patches@gcc.gnu.org>;
Date: Fri, Aug 11, 2023 11:57 PM
To: "Lehua Ding"<lehua.ding@rivai.ai>;"gcc-patches"<gcc-patches@gcc.gnu.org>;
Cc: "juzhe.zhong"<juzhe.zhong@rivai.ai>;"rdapp.gcc"<rdapp.gcc@gmail.com>;"kito.cheng"<kito.cheng@gmail.com>;"palmer"<palmer@rivosinc.com>;
Subject: Re: [PATCH] RISC-V: Fix error combine of pred_mov pattern
On 8/8/23 21:54, Lehua Ding wrote:
> Hi Jeff,
>
> > The pattern's operand 0 explicitly allows MEMs as do the constraints.
> > So forcing the operand into a register just seems like it's papering
> > over the real problem.
>
> The added of force_reg code is address the problem preduced after
> address the error combine.
> The more restrict condtion of the pattern forbidden mem->mem pattern
> which will
> produced in -O0. I think the implementation forgot to do this force_reg
> operation before
> when doing the intrinis expansion The reason this problem isn't exposed
> before is because
> the reload pass will converts mem->mem to mem->reg; reg->mem based on
> the constraint.
So if the core issue if mem->mem, that is a common thing to avoid.
Basically in the expander you use a force_reg and then have a test like
!(MEM_P (op0) && MEM_P (op1)) in the define_insn's condition.
But the v1 had a much more complex condition. It looks like that got
cleaned up in the v2. So I'll need to look at that one more closely.
>
> > This comment doesn't make sense in conjuction with your earlier details.
> > In particular combine doesn't run at -O0, so your earlier comment that
> > combine creates the problem seems inconsistent with the comment above.
>
> As the above says, the code addresses the problem which produced
> after addressing the combine problem.
But combine doesn't run at -O0. So something is inconsistent. I
certainly believe we need to avoid the mem->mem case, but that's
independent of combine and affects all optimization levels.
>
> > Umm, wow. I haven't thought deeply about this, but the complexity of
> > that insn condition is a huge red flag that our operand predicates
> > aren't correct for this pattern.
>
> This condition is large because the vsetvl info need (compare to scalar
> mov or *mov<mode>_whole pattern),
> but I think this condition is enough clear to understand. Let me explain
> briefly.
>
> (register_operand (operands[0], <MODE>mode) && MEM_P (operands[3]))
> || (MEM_P (operands[0]) && register_operand(operands[3], <MODE>mode))
>
> This two conditons mean allow mem->reg and reg->mem pattern.
I think we can simplify to just
!(MEM_P (operands[0]) && MEM_P (operands[1])
>
> (register_operand (operands[0], <MODE>mode) &&
> satisfies_constraint_Wc1 (operands[1]))
>
> This condition mean the mask must be all trues for reg->reg_or_imm
> pattern since> reg->reg insn doen't support mask operand.
I would have expected those to be handled by the constraints rather than
the pattern's condition.
Jeff
next prev parent reply other threads:[~2023-08-11 16:30 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-08-09 3:54 Lehua Ding
2023-08-10 12:29 ` Lehua Ding
2023-08-11 15:57 ` Jeff Law
2023-08-11 16:30 ` Lehua Ding [this message]
2023-08-11 16:40 ` Lehua Ding
2023-08-28 21:34 ` Jeff Law
2023-08-18 10:30 ` Lehua Ding
-- strict thread matches above, loose matches on Subject: below --
2024-02-20 4:21 Alexandre Oliva
2024-02-23 7:39 ` Jeff Law
2023-08-08 11:57 Lehua Ding
2023-08-08 16:10 ` Jeff Law
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=D667ADA3F89EED6D+tencent_D1FA83EE7A6E261CF9A89E7BF785B3D38E05@qq.com \
--to=lehua.ding@rivai.ai \
--cc=gcc-patches@gcc.gnu.org \
--cc=jeffreyalaw@gmail.com \
--cc=juzhe.zhong@rivai.ai \
--cc=kito.cheng@gmail.com \
--cc=palmer@rivosinc.com \
--cc=rdapp.gcc@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).