From: Uros Bizjak <ubizjak@gmail.com>
To: Florian Weimer <fw@deneb.enyo.de>
Cc: libc-alpha@sourceware.org, Richard Henderson <rth@twiddle.net>
Subject: Re: [PATCH, alpha]: Add earlyclobber to sqrtt/sqrtf insns.
Date: Fri, 14 Apr 2017 13:02:00 -0000 [thread overview]
Message-ID: <CAFULd4b3daAPoddnBC1=bPf3zPGMqq6NCZ-nc4SmnSEqJ95PSg@mail.gmail.com> (raw)
In-Reply-To: <877f2nm9nk.fsf@mid.deneb.enyo.de>
On Fri, Apr 14, 2017 at 2:54 PM, Florian Weimer <fw@deneb.enyo.de> wrote:
>>>> Add earlyclobber to sqrtt/sqrtf insns.
>>>>
>>>> When using software completions, we have to prevent assembler to match
>>>> input and output operands of sqrtt/sqrtf insn. Add earlyclobber to
>>>> output operand to avoid unwanted operand matching.
>>>
>>> Is this so that the trap handler can recover the original input?
>>>
>>> In this case, please add this to the commit message for those of us
>>> who are not familiar with the Alpha architecture.
>>
>> I don't know the low-level details, perhaps Richard can explain these
>> requirements.
>>
>> The sqrtt simply returns zero for subnormal operand with software
>> completions when input and output reg are matched.
>
> Did you see that in an emulator, or on actual hardware?
I have seen this on actual hardware, AlphaServer ES40, EV68AL Tsunami.
Without the patch, there are many instances of problematic insns,
simple grep sqrt[ts] of libm dump shows
~ $ grep sqrt[ts] aaa
6e94: 6b bd eb 53 sqrtt/sud $f11,$f11
6ef0: 70 bd f0 53 sqrtt/sud $f16,$f16
cb8c: 60 bd e0 53 sqrtt/sud $f0,$f0
cc14: 60 bd e0 53 sqrtt/sud $f0,$f0
d28c: 70 bd e2 53 sqrtt/sud $f2,$f16
d350: 70 bd e2 53 sqrtt/sud $f2,$f16
d588: 62 bd e2 53 sqrtt/sud $f2,$f2
d60c: 62 bd e2 53 sqrtt/sud $f2,$f2
dad0: 70 bd e2 53 sqrtt/sud $f2,$f16
db64: 70 bd e2 53 sqrtt/sud $f2,$f16
ddbc: 62 bd e2 53 sqrtt/sud $f2,$f2
de24: 62 bd e2 53 sqrtt/sud $f2,$f2
e028: 62 bd e2 53 sqrtt/sud $f2,$f2
e628: 71 bd f1 53 sqrtt/sud $f17,$f17
12290: 6a bd ea 53 sqrtt/sud $f10,$f10
133ec: 6b bd eb 53 sqrtt/sud $f11,$f11
13460: 6b bd eb 53 sqrtt/sud $f11,$f11
1dff4: 60 bd e0 53 sqrtt/sud $f0,$f0
23de4: 62 bd e2 53 sqrtt/sud $f2,$f2
23f00: 62 bd e2 53 sqrtt/sud $f2,$f2
24000: 70 bd f0 53 sqrtt/sud $f16,$f16
24040: 64 bd ea 53 sqrtt/sud $f10,$f4
24084: 64 bd ea 53 sqrtt/sud $f10,$f4
24124: 6a bd ea 53 sqrtt/sud $f10,$f10
2414c: 70 bd e0 53 sqrtt/sud $f0,$f16
24168: 61 bd e1 53 sqrtt/sud $f1,$f1
25cac: 63 bd ea 53 sqrtt/sud $f10,$f3
25d94: 71 bd e5 53 sqrtt/sud $f5,$f17
25eb4: 65 bd e5 53 sqrtt/sud $f5,$f5
25f24: 63 bd e3 53 sqrtt/sud $f3,$f3
25f84: 6b bd eb 53 sqrtt/sud $f11,$f11
25f98: 6a bd ea 53 sqrtt/sud $f10,$f10
25fa8: 6b bd eb 53 sqrtt/sud $f11,$f11
2607c: 6c bd ec 53 sqrtt/sud $f12,$f12
26094: 6a bd ea 53 sqrtt/sud $f10,$f10
2613c: 6b bd eb 53 sqrtt/sud $f11,$f11
26150: 63 bd eb 53 sqrtt/sud $f11,$f3
2995c: 6b b9 ea 53 sqrts/sud $f10,$f11
29ad0: 6b b9 ea 53 sqrts/sud $f10,$f11
29c00: 70 b9 f0 53 sqrts/sud $f16,$f16
29c60: 6b b9 eb 53 sqrts/sud $f11,$f11
29d50: 6d b9 ea 53 sqrts/sud $f10,$f13
29e18: 6b b9 ea 53 sqrts/sud $f10,$f11
2aa00: 6a bd ea 53 sqrtt/sud $f10,$f10
2aebc: 62 b9 e2 53 sqrts/sud $f2,$f2
2af6c: 62 b9 e2 53 sqrts/sud $f2,$f2
2b1a8: 62 b9 e2 53 sqrts/sud $f2,$f2
2b230: 62 b9 e2 53 sqrts/sud $f2,$f2
2b630: 70 b9 e2 53 sqrts/sud $f2,$f16
2b6c8: 70 b9 e2 53 sqrts/sud $f2,$f16
2b940: 62 b9 e2 53 sqrts/sud $f2,$f2
2b9b8: 62 b9 e2 53 sqrts/sud $f2,$f2
2d29c: 60 b9 f0 53 sqrts/sud $f16,$f0
2e528: 6a b9 ea 53 sqrts/sud $f10,$f10
2fb68: 6b b9 eb 53 sqrts/sud $f11,$f11
2fbfc: 6b b9 eb 53 sqrts/sud $f11,$f11
33948: 60 b9 e0 53 sqrts/sud $f0,$f0
37c30: 63 b9 e3 53 sqrts/sud $f3,$f3
37d64: 63 b9 e3 53 sqrts/sud $f3,$f3
37e64: 70 b9 f0 53 sqrts/sud $f16,$f16
37eac: 62 b9 ea 53 sqrts/sud $f10,$f2
37ed8: 62 b9 ea 53 sqrts/sud $f10,$f2
37f88: 6a b9 ea 53 sqrts/sud $f10,$f10
37fa8: 70 b9 e0 53 sqrts/sud $f0,$f16
37fb8: 61 b9 e1 53 sqrts/sud $f1,$f1
3909c: 65 b9 ea 53 sqrts/sud $f10,$f5
39184: 65 b9 e7 53 sqrts/sud $f7,$f5
39294: 67 b9 e7 53 sqrts/sud $f7,$f7
39310: 65 b9 e5 53 sqrts/sud $f5,$f5
39370: 6b b9 eb 53 sqrts/sud $f11,$f11
39384: 65 b9 e5 53 sqrts/sud $f5,$f5
39390: 6a b9 ea 53 sqrts/sud $f10,$f10
39458: 6c b9 ec 53 sqrts/sud $f12,$f12
39470: 6a b9 ea 53 sqrts/sud $f10,$f10
39514: 6a b9 ea 53 sqrts/sud $f10,$f10
39528: 65 b9 ea 53 sqrts/sud $f10,$f5
7429c: 6a bd f0 53 sqrtt/sud $f16,$f10
742f8: 6a bd f0 53 sqrtt/sud $f16,$f10
Uros.
next prev parent reply other threads:[~2017-04-14 13:02 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-04-14 12:30 Uros Bizjak
2017-04-14 12:38 ` Florian Weimer
2017-04-14 12:47 ` Uros Bizjak
2017-04-14 12:54 ` Florian Weimer
2017-04-14 13:02 ` Uros Bizjak [this message]
2017-04-14 13:55 ` Richard Henderson
2017-04-14 18:01 ` Uros Bizjak
2017-04-26 13:44 ` Richard Henderson
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='CAFULd4b3daAPoddnBC1=bPf3zPGMqq6NCZ-nc4SmnSEqJ95PSg@mail.gmail.com' \
--to=ubizjak@gmail.com \
--cc=fw@deneb.enyo.de \
--cc=libc-alpha@sourceware.org \
--cc=rth@twiddle.net \
/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).