tag:blogger.com,1999:blog-5683237053150646122.post4626161150251054530..comments2014-10-03T06:40:26.920+04:00Comments on What your mother never told you about graphics development: View frustum culling optimization – Representation mattersArseny Kapoulkinehttp://www.blogger.com/profile/18310595345818946666noreply@blogger.comBlogger6125tag:blogger.com,1999:blog-5683237053150646122.post-72549485490688872112010-09-23T22:34:58.235+04:002010-09-23T22:34:58.235+04:00Well since most of the objects are going to need t...Well since most of the objects are going to need their world view projection matrices calculated anyway, then there's little loss, otherwise you have to solve all the planes for the frustum in world space for each object. I think a matrix calculation is probably cheaper than a world space frustum cull test.TheMipmapperhttp://graphics.comnoreply@blogger.comtag:blogger.com,1999:blog-5683237053150646122.post-24960945065763387972009-12-13T23:06:38.368+03:002009-12-13T23:06:38.368+03:00Since you're likely to have many many objects ...Since you're likely to have many many objects with bounding boxes in world space, and only 6 frustum planes in projection space, wouldn't it be faster in the long run to just bring the planes to world space and to the test there?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-5683237053150646122.post-20110639020839877202009-05-05T18:30:00.000+04:002009-05-05T18:30:00.000+04:00I use the p/n-vertex approach on the wii and perfo...I use the p/n-vertex approach on the wii and perform the test in ~145 cycles (with paired-single instructions)<br /><br />This is great if you use occluding planes as well since you can use the almost exact algorithm.<br /><br />For an AABB, getting the p/n vertex is 2 fsel instructions (not counting the load instructions) on the wii, maybe 1 on spu.xzenoxhttp://www.blogger.com/profile/03415735783122993959noreply@blogger.comtag:blogger.com,1999:blog-5683237053150646122.post-75157430776817775952009-04-01T00:52:00.000+04:002009-04-01T00:52:00.000+04:00As for trig, I wrote that in the post.As for "no n...As for trig, I wrote that in the post.<BR/>As for "no need to transform", no, you are wrong. Adding matrix columns is just another way to do transformation of (1, 0, 0, 1)-like vectors.<BR/>Although there is a typo in the post - you need inversed transposed inverse view projection matrix (which is of course equal to just transposed). Thanks for leading me to it.Arseny Kapoulkinehttp://www.blogger.com/profile/18310595345818946666noreply@blogger.comtag:blogger.com,1999:blog-5683237053150646122.post-31446223346312463122009-04-01T00:41:00.000+04:002009-04-01T00:41:00.000+04:00You can extract the planes directly from the proje...You can extract the planes directly from the projection matrix, there is no need for trig and there is no need to transform the clip space planes by the inverse projection.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-5683237053150646122.post-21926858734869863192009-03-19T11:34:00.000+03:002009-03-19T11:34:00.000+03:00Great series of posts - please keep them coming!I ...Great series of posts - please keep them coming!<BR/><BR/>I am quite enjoying them.Anonymousnoreply@blogger.com