https://community.intel.com/t5/Intel-Integrated-Performance/beginners-questions/m-p/922512#M15851 <DIV>1. that is not too optimistic. please use ipp 2d fft in ippi lib on P4, ipp is exactly faster than nsp. not clear why do you need forming a "column" row. just use the 2d transform</DIV> <DIV>2. you can compare yourself, but for example MKL 1D DFT calls IPP inside</DIV> <DIV></DIV> Sat, 18 Dec 2004 01:50:29 GMT borix 2004-12-18T01:50:29Z https://community.intel.com/t5/Intel-Integrated-Performance/beginners-questions/m-p/922511#M15850 <DIV> <DIV>Hi</DIV> <DIV>I'm drowning in IPP and MKL documentation</DIV> <DIV>I have been using nsp up till now but now consider changing to IPP or MKL.</DIV> <DIV>It would be very helpful for me if I can get some suggestions on porting.</DIV> <DIV>I'll then would have a starting point.</DIV> <DIV>Myapplication works with lots of large (512*512 up tp 8k*8k) images and fft's,but only one in memory at a time.</DIV> <DIV>I have used the strategy of allocating each row for it self, both image and fft.</DIV> <DIV>That means that I have to build coulmn vectors myself for the vertical FFTwhich slows things down I'll guess.</DIV> <DIV> <DIV>To use a "Intel" memor allocation would make bigger demand on contimuous free space, wouldnt it?</DIV></DIV> <DIV>Up to now, I have used this type ofcall:</DIV> <DIV>nspcrFft( PFloat(floatRArray<Y>),<BR /> PFloat(floatIArray<Y>),<BR /> Bits,<BR /> NSP_Forw OR NSP_NoBitRev OR NSP_DoFastMMX);</Y></Y></DIV> <DIV></DIV> <DIV>Can someone please comment on the validit of my heap concerns and offer</DIV> <DIV>a suggestion to replacement function (just as a starting point for me)</DIV> <DIV>Also</DIV> <DIV>1: is it to optimistic to hope for a doubled performance on a P4 ?</DIV> <DIV>2: how do I choose between IPP and MKL ?</DIV> <DIV>3: I will extend to 3d soon with objects up to 1k*1k*64, does that have any bearing on the choise above?</DIV></DIV> Fri, 17 Dec 2004 20:24:50 GMT https://community.intel.com/t5/Intel-Integrated-Performance/beginners-questions/m-p/922511#M15850 uabpsab 2004-12-17T20:24:50Z https://community.intel.com/t5/Intel-Integrated-Performance/beginners-questions/m-p/922512#M15851 <DIV>1. that is not too optimistic. please use ipp 2d fft in ippi lib on P4, ipp is exactly faster than nsp. not clear why do you need forming a "column" row. just use the 2d transform</DIV> <DIV>2. you can compare yourself, but for example MKL 1D DFT calls IPP inside</DIV> <DIV></DIV> Sat, 18 Dec 2004 01:50:29 GMT https://community.intel.com/t5/Intel-Integrated-Performance/beginners-questions/m-p/922512#M15851 borix 2004-12-18T01:50:29Z