0003526: wish: control of the floating point rounding mode - MantisBT

Mantis Bug Tracker

View Issue Details [ Jump to Notes ]

[ Issue History ] [ Print ]

ID

Project

Category

View Status

Date Submitted

Last Update

0003526

OCaml

OCaml general

public

2005-03-10 08:09

2007-12-02 09:43

Reporter

administrator

Assigned To

Priority

normal

Severity

feature

Reproducibility

always

Status

acknowledged

Resolution

open

Platform

OS

OS Version

Product Version

Target Version

Fixed in Version

Summary

0003526: wish: control of the floating point rounding mode

Description

Full_Name: Sebastian Egner
Version: Ocaml
OS: Linux/FreeBDS/Windows
Submission from: dslam164-147-59-81.dyndsl.versatel.nl (81.59.147.164)

For implementing efficient interval arithmetics it is important to have control
of the rounding mode of floating point operations. While this could be done with
calls to a C-function, the overhead would be significant---and floating point
ops would become dependent on global state. Implementing the entire interval
arithmetics as an external module in C, and linking against it, is also not a
very attractive option as the Ocalm compiler cannot fully be deployed for
optimization.

Tags

No tags attached.

Relationships

duplicate of

acknowledged

Wish: Controle du mode d'arrondi

Notes
(0004343) Christophe Troestler (reporter) 2007-11-22 19:37	Merge with 0001743

(0004356) thelema (reporter) 2007-11-26 23:09	Could the semantics of FP math in OCaml be further standardized so that each arch just needs some code to set the right flags and everyone gets the same behavior? This feature sounds like an excuse to build more incompatibilities between different architectures.

(0004374) xleroy (administrator) 2007-12-02 09:43	> Could the semantics of FP math in OCaml be further standardized > so that each arch just needs some code to set the right flags > and everyone gets the same behavior? This is exactly what we aim for, but actually all modern platforms that we know of set the FPU to the mode we want (no exceptions, round to nearest, etc). The last know platform that needed additional code to get this mode was FreeBSD pre 2002. So, basically, if you have examples of modern platforms where the default floating-point mode isn't the "standard" one, please report it and we'll see what we can do.

Issue History
Date Modified	Username	Field	Change
2005-11-18 10:13	administrator	New Issue
2007-11-22 19:37	Christophe Troestler	Note Added: 0004343
2007-11-22 19:39	Christophe Troestler	Note Added: 0004345
2007-11-22 19:39	Christophe Troestler	Note Deleted: 0004345
2007-11-22 23:27	doligez	Relationship added	duplicate of 0001743
2007-11-26 23:09	thelema	Note Added: 0004356
2007-12-02 09:43	xleroy	Note Added: 0004374
2007-12-02 09:43	xleroy	Description Updated

Copyright © 2000 - 2011 MantisBT Group