Skip to Content.
Sympa Menu

coq-club - RE: [Coq-Club] How does Int31 "underlying mechanism for hardware-efficient computations" work?

coq-club AT inria.fr

Subject: The Coq mailing list

List archive

RE: [Coq-Club] How does Int31 "underlying mechanism for hardware-efficient computations" work?


Chronological Thread 
  • From: "Soegtrop, Michael" <michael.soegtrop AT intel.com>
  • To: "coq-club AT inria.fr" <coq-club AT inria.fr>
  • Subject: RE: [Coq-Club] How does Int31 "underlying mechanism for hardware-efficient computations" work?
  • Date: Wed, 29 Jun 2016 09:45:23 +0000
  • Accept-language: de-DE, en-US
  • Authentication-results: mail3-smtp-sop.national.inria.fr; spf=None smtp.pra=michael.soegtrop AT intel.com; spf=Pass smtp.mailfrom=michael.soegtrop AT intel.com; spf=None smtp.helo=postmaster AT mga09.intel.com
  • Ironport-phdr: 9a23:ydTuFh3Z/ZVZBHEQsmDT+DRfVm0co7zxezQtwd8ZsegQL/ad9pjvdHbS+e9qxAeQG96Ks7Qd1qGO7OjJYi8p2d65qncMcZhBBVcuqP49uEgeOvODElDxN/XwbiY3T4xoXV5h+GynYwAOQJ6tL2PbrnD61zMOABK3bVMzfbWtXNePxJzon8mJuLTrKz1SgzS8Zb4gZD6Xli728vcsvI15N6wqwQHIqHYbM85fxGdvOE7B102kvpT4r9Zf9HEasPU4ssVETK/SfqIiTLUeAi5sezQ+49Suvh3eRyOO4GEdWyMYiEwbLRLC6USwZZD8vTfgsfI5kAybNs3/QLR+EWCn7qxrQRLszjwAOjEl6mbPos12kK9f5hmmokoskMbvfIiJOa8mLevmdtQASD8ZUw==

Dear Arnaud,

 

thanks a lot for the detailed explanation – the “Register” process doesn’t seem to be documented in the reference manual, so this is very welcome.

 

I would say the trust base increase is not that much of a problem, as long as this can be switched off easily. I don’t see how this would be done in the current implementation of Int32 (other than copying the library and removing the register keyword). I would use such features mostly for interactive development and use the plain Coq implementation in overnight check runs, where speed doesn’t matter that much.

 

Btw.: has it been tried to use e.g. hex digit representations in Coq? This thesis analyses the speed up of multi-bit digit arithmetic in HOL:

 

FORMAL COMPUTATIONS AND METHODS by Alexey Solovyev, Pittsburgh, 2007

 

http://d-scholarship.pitt.edu/16721/7/SolovyevThesis_(v._1.02).pdf

 

For 5 decimal digit numbers the speed up is about 2.5, from 10 decimal digits on one gets a speed increase of about 4 times for 4 bit digits.

 

Best regards,

 

Michael

Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928




Archive powered by MHonArc 2.6.18.

Top of Page