From: Richard Sandiford <rsandifo@linux.vnet.ibm.com>
To: Paulo Matos <pmatos@broadcom.com>
Cc: "gcc\@gcc.gnu.org" <gcc@gcc.gnu.org>
Subject: Re: Mode change for bswap pattern expansion
Date: Mon, 27 Jan 2014 17:00:00 -0000 [thread overview]
Message-ID: <87vbx5s735.fsf@sandifor-thinkpad.stglab.manchester.uk.ibm.com> (raw)
In-Reply-To: <19EB96622A777C4AB91610E763265F463F15EE@SJEXCHMB14.corp.ad.broadcom.com> (Paulo Matos's message of "Mon, 27 Jan 2014 16:24:04 +0000")
Paulo Matos <pmatos@broadcom.com> writes:
>> -----Original Message-----
>> From: Richard Sandiford [mailto:rdsandiford@googlemail.com]
>> Sent: 27 January 2014 16:06
>> To: Paulo Matos
>> Cc: gcc@gcc.gnu.org
>> Subject: Re: Mode change for bswap pattern expansion
>>
>> Paulo Matos <pmatos@broadcom.com> writes:
>> > On a vector processor we can do a bswapsi with two instructions, by first
>> rotating half-words (16 bits) by 8 and then rotating full words by 16.
>> > However, this means expanding:
>> > (set (match_operand:SI 0 "register_operand" "")
>> > (bswap:SI (match_operand:SI 1 "register_operand" "")))
>> >
>> > to:
>> > (set (match_dup:V2HI 0)
>> > (rotate:V2HI (match_dup:V2HI 1)
>> > (const_int 8)))
>> > (set (match_dup:SI 0)
>> > (rotate:SI (match_dup:SI 0)
>> > (const_int 16)))
>> >
>> > This is obviously not correct, because match_dup cannot set the mode. The point
>> I am trying to make is that I can't find a good way to deal with the mode
>> changes. I don't think GCC is too happy if I change the modes of the
>> same operand
>> from one instruction to the other right? The only other way is to emit
>> paradoxical subregs. So something along these lines:
>> > (set (subreg:V2HI (match_dup 0) 0)
>> > (rotate:V2HI (subreg:V2HI (match_dup 1) 0)
>> > (const_int 8)))
>> > (set (match_dup 0)
>> > (rotate:SI (match_dup 0)
>> > (const_int 16)))
>>
>> It's usually better not to hard-code the subregs in the pattern.
>> Instead you could use C code to create the subregs, e.g.:
>>
>> [(set (match_dup 3)
>> (rotate:V2HI (match_dup 2)
>> (const_int 8)))
>> (set (match_dup 0)
>> (rotate:SI (match_dup 4)
>> (const_int 16)))]
>> ""
>> {
>> operands[2] = gen_lowpart (V2HImode, operands[1]);
>> operands[3] = gen_reg_rtx (V2HImode);
>> operands[4] = gen_lowpart (SImode, operands[3]);
>> }
>>
>> so that any hard regs are correctly handled. Or it might be easier to code
>> it using emit_insn (gen_* (...))s instead.
>>
>> BTW, paradoxical subregs are where the outer mode is strictly larger
>> than the inner mode.
>>
>
> That's right. My mis-understanding.
>
>> MIPS uses essentially the same sequence, except that it has a special
>> instruction to do the first rotate (WSBH), rather than it being an instance
>> of a general vector rotate. For MIPS we just model it as an unspec SImode
>> operation. Maybe that would be easier here too.
>>
>
> I will look at how MIPS is doing it.
>
> However, the unspec SI has severe performance penalties on my port since
> it is able to issue more that one instruction per cycle, therefore
> having each instruction separately allows the scheduler to issue each of
> the bswapsi parts into different slots with other instructions.
Sorry, I meant we use an unspec for the first ("V2HI") rotate.
I.e. rather than:
(set (subreg:V2HI (match_dup 2) 0)
(rotate:V2HI (subreg:V2HI (match_dup 1) 0)
(const_int 8)))
(set (match_dup 0)
(rotate:SI (match_dup 2)
(const_int 16)))
we have:
(set (match_dup 2) (unspec:SI [(match_dup 1)] UNSPEC_FOO))
(set (match_dup 0)
(rotate:SI (match_dup 2)
(const_int 16)))
In your case the define_insn for the UNSPEC_FOO pattern would have the
same attributes as a V2HI rotate, so it should get scheduled in the same way.
Thanks,
Richard
next prev parent reply other threads:[~2014-01-27 16:49 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-01-27 11:28 Paulo Matos
2014-01-27 16:24 ` Richard Sandiford
2014-01-27 16:28 ` Paulo Matos
2014-01-27 17:00 ` Richard Sandiford [this message]
2014-01-27 17:57 ` Paulo Matos
2014-01-27 19:29 ` pinskia
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=87vbx5s735.fsf@sandifor-thinkpad.stglab.manchester.uk.ibm.com \
--to=rsandifo@linux.vnet.ibm.com \
--cc=gcc@gcc.gnu.org \
--cc=pmatos@broadcom.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).