I should add that this could happen without the register width
difference, just from order of operations.
David's insistence that the codec output must be binary-identical across
modifications is a poor substute for a test suite. Unfortunately he's
probably the only person who could write it, so keep looking at
optimizer flags, please.
Thanks
Bruce
On 06/06/2013 08:27 AM, Bruce Perens wrote:
> Richard,
>
> There's a longer optimization list at
> http://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
> Not just the floating point specific flags, but pretty much any
> instruction optimization can cause a difference.
> The problem is that the floating point registers are 80 bits wide and
> the memory storage for double precision floating point is 64 bits.
> So, you can have changes to the order in which operations are done that
> are "mathematically equivalent" but cause small differences in values
> because the 80-bit to 64-bit conversion is done at different points.
>
> Thanks
>
> Bruce
>
> ------------------------------------------------------------------------------
> How ServiceNow helps IT people transform IT departments:
> 1. A cloud service to automate IT design, transition and operations
> 2. Dashboards that offer high-level views of enterprise services
> 3. A single system of record for all IT processes
> http://p.sf.net/sfu/servicenow-d2d-j
> _______________________________________________
> Freetel-codec2 mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
------------------------------------------------------------------------------
How ServiceNow helps IT people transform IT departments:
1. A cloud service to automate IT design, transition and operations
2. Dashboards that offer high-level views of enterprise services
3. A single system of record for all IT processes
http://p.sf.net/sfu/servicenow-d2d-j
_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
