|Re: [eigen] sse asin implementation|
[ Thread Index |
| More lists.tuxfamily.org/eigen Archives
- To: eigen@xxxxxxxxxxxxxxxxxxx
- Subject: Re: [eigen] sse asin implementation
- From: Benoit Jacob <jacob.benoit.1@xxxxxxxxx>
- Date: Tue, 7 Apr 2009 14:00:36 +0200
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:received:in-reply-to:references :date:message-id:subject:from:to:content-type :content-transfer-encoding; bh=yywY+OqnLEqOokg4roC4Kt2PtX2tl3yIiI+4LHN0KaQ=; b=iNitcVlzYoLd+cNW39pd2M/NjUwg4L3GuNJanocGMLsZFpkwT91Y7dhlmEc3/luGKA q1qLoiXNT0hPOhO/tnncySiBGgtB64X/Oxg+UyYmYDLLh1qwrhikP2MHMrKBLu+/5yfP qiRdK8ah+gILRewOevErDt1tclDIqrn47V524=
- Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type:content-transfer-encoding; b=hFPbToM1l8MbkHPTOtsEHMnxzEDVMh1teMPhpjraDXsa3TXBc/wG3xz2rvAhKTI4Pa mSDhxwdIPZp5d1ljC+Xe5KPXUcLjdaB5frutu+j4QFWMXsFBA2T8Rpf/0X9+2ehgCRax 7fFUvg0xJpMh+5C4poyl3L2d9FH4uI4VvhThU=
I think your idea of the unsupported/MoreVectorization module is a good one.
Anyway, since you mentioned asin in the todo in the first place,
please decide where to put Rohit's code, I'm OK either way.
About --ffast-math, I'm not competent. As soon as one uses SSE, one is
no longer IEEE754 compliant anyway. So pushing the conservative logic
to the extreme, one would disable vectorization unless -ffast-math is
enabled. So it's really a matter of measuring how much imprecise our
fast functions are. If it's just 1 bit of imprecision, then it doesn't
really matter. But if in certain cases the imprecision becomes worse,
then yes restrict that to when -ffast-math is set.
2009/4/7 Rohit Garg <rpg.314@xxxxxxxxx>:
> So will it be commited?
> On Thu, Apr 2, 2009 at 12:10 PM, Rohit Garg <rpg.314@xxxxxxxxx> wrote:
>>> it is relevant. Since I'm fine with both behavior I need people
>>> express their opinion to make that change or not. So I count one vote
>>> to "disable by default unless -ffast-math is set".
>> I vote for "disable by default unless -ffast-math is set" as well.
>> Rohit Garg
>> Senior Undergraduate
>> Department of Physics
>> Indian Institute of Technology
> Rohit Garg
> Senior Undergraduate
> Department of Physics
> Indian Institute of Technology