search

MiguelMonteiro / permutohedral_lattice

Permutohedral Lattice C++/CUDA implementation + TensorFlow Op (CPU/GPU)
83 stars 18 forks source link

Possibly incorrect number of blocks/blockSize? #23

Closed achirkin closed 4 years ago

achirkin commented 4 years ago

I've been working on some adaptation of your code and noticed something inconsistent:

https://github.com/MiguelMonteiro/permutohedral_lattice/blob/fe04895526dd39e5ea01c35194b7199c04f9e680/src/PermutohedralLatticeGPU.cuh#L476

Shouldn't here be blocks.y = 1;? blockSize.y seems to be equal to one already. In my understanding, the result is that the same computation is run pd + 1 times. Though it's not entirely clear if this is harmful.

MiguelMonteiro commented 4 years ago

Sorry it's been awhile I don't know how to answer your question.