https://community.intel.com/t5/Intel-Integrated-Performance/whole-number-integer-power/m-p/823497#M5003 For small whole number (integer) constant powers, is it better to iteratively multiply into a temporary buffer, or use Powx?<DIV></DIV><DIV>What is the tradeoff in performance vs. precision for each approach (recognizing Powx provides optional precisions)?</DIV><DIV></DIV><DIV>Thanks!</DIV> Thu, 12 May 2011 22:31:57 GMT jeffc111 2011-05-12T22:31:57Z https://community.intel.com/t5/Intel-Integrated-Performance/whole-number-integer-power/m-p/823497#M5003 For small whole number (integer) constant powers, is it better to iteratively multiply into a temporary buffer, or use Powx?<DIV></DIV><DIV>What is the tradeoff in performance vs. precision for each approach (recognizing Powx provides optional precisions)?</DIV><DIV></DIV><DIV>Thanks!</DIV> Thu, 12 May 2011 22:31:57 GMT https://community.intel.com/t5/Intel-Integrated-Performance/whole-number-integer-power/m-p/823497#M5003 jeffc111 2011-05-12T22:31:57Z https://community.intel.com/t5/Intel-Integrated-Performance/whole-number-integer-power/m-p/823498#M5004 <P>Hi, </P><P>I think simple iterative multiplication will have a better performance and good precision with small power N. You might just have a simple test with the following two ways, and see which one is better for your data: </P><P>ippsPowx_32f_A21( indata, N, ouputdata, length );</P><P>or: </P><P>for(int j=0;j<N></N></P><P>ippsMul_32f_I(outputdata, indata,length);</P><P>}</P><P>Thanks,<BR />Chao</P> Mon, 16 May 2011 07:44:32 GMT https://community.intel.com/t5/Intel-Integrated-Performance/whole-number-integer-power/m-p/823498#M5004 Chao_Y_Intel 2011-05-16T07:44:32Z