[RFC] Quad-float support, round 4

[Jerry DeLisle]

On 09/14/2010 02:41 PM, Tim Prince wrote:
> On 9/14/2010 8:41 AM, Steve Kargl wrote:
>> gfc ifc
>> sqrtq 200 160 (ie, __float128 is XXX times slower than double)
>> sinq 100 6
>> cosq 150 10
>> asinhq 120 10
>> erfcq 150 6
>>
>> on the same hardware. Note, gfc = gfortran, and ifc = ifort.
>>
>> The question then becomes what are the algorithms that Intel uses
>> to compute the last four functions.
> As far as I know, the algorithms in the Intel quad precision math
> library are closely held, although you might learn something by objdump
> or similar tools. I would be concerned that doing this myself would run
> into conflicts with my employer, as my gcc paperwork will not be
> reconfirmed by them.

Read this:

http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4857011

for some insight into the problems one can get into.

If older implementations of FPU have issues with useble range, then we will have 
to do range reduction for certain.

Jerry