linux-sparse.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Linus Torvalds <torvalds@linux-foundation.org>
To: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
Cc: Dibyendu Majumdar <mobile@majumdar.org.uk>,
	Linux-Sparse <linux-sparse@vger.kernel.org>
Subject: Re: Potential incorrect simplification
Date: Tue, 28 Mar 2017 14:57:01 -0700	[thread overview]
Message-ID: <CA+55aFyWtn3_o+D5f1V9NvwHSDBFHi57tv4vOjJAffhCyqL7mA@mail.gmail.com> (raw)
In-Reply-To: <20170328202746.bmamqhxitmpvfsad@macpro.local>

On Tue, Mar 28, 2017 at 1:27 PM, Luc Van Oostenryck
<luc.vanoostenryck@gmail.com> wrote:
> On Tue, Mar 28, 2017 at 11:02:32AM -0700, Linus Torvalds wrote:
>>
>> The single-store shortcut is actually pretty important, because it is
>> what turns a *lot* of local variables from stack memory things into
>> pseudos. That's partly because the sparse optimizations aren't all
>> that smart.
>
> Everytime I look at this part of the code, I think the same:
>   why turn all vars into stack memory to have to turn them later
>   into pseudos?

Oh, you absolutely *cannot* turn them into pseudo's directly.

A pseudo is a "register", and in SSA format. Local variables are not
registers, and do not honor SSA.

So local variables are very much *not* pseudos at any time. Local
variables can have their address taken, local variables can be
assigned multiple times, local variables can have complex types, none
of which is pseudo-like behavior.

Now, some *very* limited cases of local variables end up being
trivially very similar to pseudos, and those simple cases get turned
into pseudos. But that only happens for the simple cases - when their
address is not taken or used, and when they only have a single
assignment to them.

But local variables do not start out as pseudos, exactly because the
pseudo case is a very very very limited case of the full local
variable case.

                        Linus

  reply	other threads:[~2017-03-28 21:57 UTC|newest]

Thread overview: 59+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-03-28 12:40 Potential incorrect simplification Dibyendu Majumdar
2017-03-28 13:34 ` Luc Van Oostenryck
2017-03-28 13:58   ` Dibyendu Majumdar
2017-03-28 14:11     ` Luc Van Oostenryck
2017-03-28 14:19       ` Dibyendu Majumdar
2017-03-28 16:20         ` Linus Torvalds
2017-03-28 17:00           ` Luc Van Oostenryck
2017-03-28 18:02             ` Linus Torvalds
2017-03-28 20:27               ` Luc Van Oostenryck
2017-03-28 21:57                 ` Linus Torvalds [this message]
2017-03-28 22:28                   ` Luc Van Oostenryck
2017-03-28 22:22           ` Dibyendu Majumdar
2017-08-06 12:46           ` Christopher Li
2017-08-06 14:00             ` Luc Van Oostenryck
2017-08-06 14:24               ` Christopher Li
2017-08-06 14:54                 ` Christopher Li
2017-08-06 15:07                 ` Luc Van Oostenryck
2017-08-06 15:51                   ` Christopher Li
2017-08-06 16:51                     ` Luc Van Oostenryck
2017-08-06 18:35                       ` Christopher Li
2017-08-06 19:51                         ` Dibyendu Majumdar
2017-08-06 20:08                           ` Luc Van Oostenryck
2017-08-06 19:52                         ` Luc Van Oostenryck
2017-08-06 23:34                           ` Christopher Li
2017-08-07  0:31                             ` Luc Van Oostenryck
2017-08-07  0:38                               ` Christopher Li
2017-08-06 15:52                 ` Dibyendu Majumdar
2017-08-06 16:56                   ` Luc Van Oostenryck
2017-08-06 17:04                     ` Dibyendu Majumdar
2017-08-06 17:45                       ` Luc Van Oostenryck
2017-08-06 17:58                         ` Dibyendu Majumdar
2017-08-06 18:15                           ` Luc Van Oostenryck
2017-08-06 18:18                             ` Dibyendu Majumdar
2017-08-06 18:31                               ` Luc Van Oostenryck
2017-08-07 19:11                   ` [PATCH v2 0/8] fix loading of partially defined bitfield Luc Van Oostenryck
2017-08-07 19:11                     ` [PATCH v2 1/8] Remove single-store shortcut Luc Van Oostenryck
2017-08-07 21:42                       ` Linus Torvalds
2017-08-10  0:29                         ` Christopher Li
2017-08-10  0:41                           ` Luc Van Oostenryck
2017-08-10  0:53                             ` Christopher Li
2017-08-10 11:01                       ` Christopher Li
2017-08-10 12:26                         ` Luc Van Oostenryck
2017-08-10 13:25                           ` Christopher Li
2017-08-07 19:11                     ` [PATCH v2 2/8] new helper: def_opcode() Luc Van Oostenryck
2017-08-07 19:12                     ` [PATCH v2 3/8] reuse nbr_pseudo_users() Luc Van Oostenryck
2017-08-07 19:12                     ` [PATCH v2 4/8] change the masking when loading bitfields Luc Van Oostenryck
2017-08-07 19:12                     ` [PATCH v2 5/8] simplify ((A & M') | B ) & M when M' & M == 0 Luc Van Oostenryck
2017-08-07 19:12                     ` [PATCH v2 6/8] transform (A & M) >> S to (A >> S) & (M >> S) Luc Van Oostenryck
2017-08-08  0:22                       ` Christopher Li
2017-08-08  0:29                         ` Luc Van Oostenryck
2017-08-08  1:48                           ` Christopher Li
2017-08-08  1:00                         ` Linus Torvalds
2017-08-08  1:38                           ` Luc Van Oostenryck
2017-08-08  1:50                           ` Christopher Li
2017-08-07 19:12                     ` [PATCH v2 7/8] transform (A << S) >> S into A & (-1 " Luc Van Oostenryck
2017-08-07 21:54                       ` Linus Torvalds
2017-08-07 22:08                         ` Luc Van Oostenryck
2017-08-07 22:27                           ` Luc Van Oostenryck
2017-08-07 19:12                     ` [PATCH v2 8/8] fix: cast of OP_AND only valid if it's an OP_CAST Luc Van Oostenryck

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+55aFyWtn3_o+D5f1V9NvwHSDBFHi57tv4vOjJAffhCyqL7mA@mail.gmail.com \
    --to=torvalds@linux-foundation.org \
    --cc=linux-sparse@vger.kernel.org \
    --cc=luc.vanoostenryck@gmail.com \
    --cc=mobile@majumdar.org.uk \
    /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 NNTP newsgroup(s).