Obviously, the ippsDiv_32f_A24() function has the best accuracy of functions ippsDiv_32f_A11(), ippsDiv_32f_A21() or ippsDiv_32f_A24(), but where in this mix is ippsDiv_32f() ?
… is ippsDiv_32f_A24() more accurate that ippsDiv_32f() ?
For IPP 7.0.7, 9.0.2 and the latest IPP libraries ?
>>… is ippsDiv_32f_A24() more accurate that ippsDiv_32f() ?
yes, that's correct.
ippsDiv_32f() produces the same accuracy like A21 functions do with the fixed accuracy.
We should probably add this info into IPP documentation ( Developer Reference) in the next versions and keep this thread informed.
Regarding functions with fixed accuracy, it worth to remind "The choice of accuracy levels should be based on practical experience and identified application demands. Available options are specified in the function name suffix and include A11 , A21 , or A24 for the single-precision, and A26 , A50 , or A53 for the double-precision data format. Flavors A11 , A21 , A26 , and A50 provide approximately 3, 6, 8, and 15 exact decimal digits, respectively. For flavors A24 and A53 , the maximum guaranteed error is within 1 ulp and in most cases does not exceed 0.55 ulp."
here is the link to this IPP Developer Reference page: https://software.intel.com/content/www/us/en/develop/documentation/ipp-dev-reference/top/volume-1-si...
As we have not heard back from you, we are considering that your issue has been resolved and we have answered all your queries. So we will no longer respond to this thread. If you require additional assistance from Intel, please start a new thread. Any further interaction in this thread will be considered community only
Have a Good day!