tag:blogger.com,1999:blog-9054609445298212106.post5413401372085479267..comments2018-10-19T00:54:01.687+10:30Comments on A Hacker's Craic: oooh fastNotZedhttp://www.blogger.com/profile/09469760565180198154noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-9054609445298212106.post-54754191872002169752012-11-22T20:06:49.009+10:302012-11-22T20:06:49.009+10:30There's a version of it in here:
http://code...There's a version of it in here: <br /><br />http://code.google.com/p/socles/source/browse/trunk/socles/src/au/notzed/socle/ransac/ransac_homography.cl<br /><br />Parallelising it is fairly straightforward (if it fits within a workgroup) as most operations are independently across rows.<br /><br />NotZedhttps://www.blogger.com/profile/09469760565180198154noreply@blogger.comtag:blogger.com,1999:blog-9054609445298212106.post-53360164662396363942012-11-22T03:07:49.162+10:302012-11-22T03:07:49.162+10:30I need to perform SVD for a small matrix on GPU an...I need to perform SVD for a small matrix on GPU and wondered if your code is open source? It is for part of a university project and not really sure where to begin<br />giltechsolutions@gmail.comRichardGilmorehttps://www.blogger.com/profile/08364926942190123962noreply@blogger.comtag:blogger.com,1999:blog-9054609445298212106.post-27586823739753038162012-08-15T01:07:19.147+09:302012-08-15T01:07:19.147+09:30I'm only using it for very small matrices - e....I'm only using it for very small matrices - e.g. 9x8, used for solving 3x3 perspective matrices.<br /><br />So i'm just using 9 threads per matrix and doing 7 solutions per workgroup and all cross-thread communication is within LDS.<br /><br />For solving single instances of much bigger matrices it would require multiple kernel invocations and more complex logic - I haven't solved NotZedhttps://www.blogger.com/profile/09469760565180198154noreply@blogger.comtag:blogger.com,1999:blog-9054609445298212106.post-65211956632693150742012-08-13T12:13:00.308+09:302012-08-13T12:13:00.308+09:30Can you tell some more about how to parallel SVD i...Can you tell some more about how to parallel SVD into OpenCL? SVD require some global iteration to generate the final result, each iteration will require last times result as input, this will require global sync. While OpenCL do not support global sync, can you tell how to resolve this problem?Anonymoushttps://www.blogger.com/profile/07354876019308178340noreply@blogger.com