|
From: | John W. Eaton |
Subject: | Re: floating point precision control |
Date: | Tue, 12 Jul 2016 15:28:03 -0400 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Icedove/38.5.0 |
On 07/10/2016 03:48 PM, Mike Miller wrote:
jwe, all,
We still have a couple possibly-related bugs, that the same tests that were failing on Windows with a 32-bit Octave also fail with a 64-bit Octave, and the FP control word setting doesn't seem to affect it at all in this case (bugs #48364 and #48365).
Isn't the FP control word and extended precision specific to x86 hardware? Did you look at this mailing list thread? I've only started to read it. http://lkml.iu.edu/hypermail/linux/kernel/0103.0/0453.htmlMaybe we should just try to do the best we can with 64-bit IEEE arithmetic and not try to use extended precision since it is impossible to always use 80-bit precision everywhere (intermediate values may be stored in 64-bit memory locations, for example) and a mixture of 64- and 80-bit arithmetic seems like it may be asking for unpredictable results?
jwe
[Prev in Thread] | Current Thread | [Next in Thread] |