tag:blogger.com,1999:blog-9218555604707336107.post4639787122026371054..comments2017-11-25T06:57:16.866-08:00Comments on Baum Dev Blog: Fast Kahan summation implementation in AssemblerBaumBloggerhttp://www.blogger.com/profile/12668078702880913731noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-9218555604707336107.post-1333004207907727102010-11-12T14:18:02.141-08:002010-11-12T14:18:02.141-08:00Sure, why not? It won't always be about optimi...Sure, why not? It won't always be about optimization but yes - Computer Science is what I do :-)BaumBloggerhttps://www.blogger.com/profile/12668078702880913731noreply@blogger.comtag:blogger.com,1999:blog-9218555604707336107.post-12234389527182949702010-11-12T13:51:05.179-08:002010-11-12T13:51:05.179-08:00Awesome.Awesome.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-9218555604707336107.post-21580643503563649712010-11-12T12:02:10.265-08:002010-11-12T12:02:10.265-08:00This is an amazing post. Will you post more compu...This is an amazing post. Will you post more computer science here? I am extremely interested in these kinds of optimizations, thank you.Maethalionhttps://www.blogger.com/profile/03677599930153154043noreply@blogger.comtag:blogger.com,1999:blog-9218555604707336107.post-49541244764871618152010-11-04T15:19:12.082-07:002010-11-04T15:19:12.082-07:00Generally that depends on the distribution of the ...Generally that depends on the distribution of the numbers that you use. In the ideal case, the error might cancel but this is not the worst case.<br /><br />A float has just below 7 digits precision. So in float it is (2 + 1.19E-7) = 2. This may introduce an error of 127 * 1.19E-7 = 1.5E-5 which may seem low (0.0005%). In IIR filters for example, these results are taken for the following calculations and the errors may propagate.BaumBloggerhttps://www.blogger.com/profile/12668078702880913731noreply@blogger.comtag:blogger.com,1999:blog-9218555604707336107.post-87988794595372214712010-11-02T00:07:49.389-07:002010-11-02T00:07:49.389-07:00Out of interest, what sort of error would you be l...Out of interest, what sort of error would you be looking at with, say, adding 128 floats ranging from -2f to +2f using naive summing?Anonymousnoreply@blogger.com