search

kmammou / v-hacd

Automatically exported from code.google.com/p/v-hacd
BSD 3-Clause "New" or "Revised" License
1.37k stars 256 forks source link

New branch of V-HACD called 'performance-improvements' available #83

Open jratcliff63367 opened 4 years ago

jratcliff63367 commented 4 years ago

I have created a new branch of V-HACD with an important bug fix as well as a number of performance optimizations.

You can read more about it here.

https://codesuppository.blogspot.com/2020/04/just-short-post-today-to-cover-some.html

jratcliff63367 commented 4 years ago

This morning I did some performance measurements of the 'performance-improvements' branch to the master branch and found that with a high triangle count mesh and a very high voxel resolution (10 million voxels) the performance branch is 7.5 times faster.

There were many performance improvements made on this branch so I cannot with confidence say which optimization had the greatest benefit. But it's a substantial increase for sure.

submagr commented 4 years ago

Hi @jratcliff63367, I want to use the improved version but couldn't find a binary (OSX and Linux) for it. Can you please provide steps for getting the binary?

Thank you

jratcliff63367 commented 4 years ago

That's why I didn't merge this to master. You have to build it from source code. I'm not sure what process Khaled did to build his binaries in the past.

submagr commented 4 years ago

Thanks. I will try this.

By the way, Can you please share the binaries you used for testing here (as a zip) just in case, if the compilation doesn't work for me?

pmurph0305 commented 4 years ago

This branch appears to produce convex hulls with a number of vertices greater than the Parameter's m_maxNumVerticesPerCH in many cases. Is this expected?

Primarily occurs with a low number of convex hulls, so it's easiest to see with max convex hulls set to 1 and a low number of max vertices per convex hull as well.

jratcliff63367 commented 4 years ago

This branch appears to produce convex hulls with a number of vertices greater than the Parameter's m_maxNumVerticesPerCH in many cases. Is this expected?

That would be a bug. I will take a look at this.

pmurph0305 commented 4 years ago

This branch appears to produce convex hulls with a number of vertices greater than the Parameter's m_maxNumVerticesPerCH in many cases. Is this expected?

That would be a bug. I will take a look at this.

Hope you don't mind, but I looked into it and discovered that the order of simplifying and merging convex hulls was swapped in this branch and causes the issue. I submitted a pull request that fixes the issue. #89