Efficient Hammersley distribution on the GPU

Started by
1 comment, last by B_old 10 years, 4 months ago

I'm looking for an efficient way to get hammersley distribution points on the GPU. I found this page with some GLSL code that relies on bit-shifts. Is this efficient on the GPU? Are you aware of more efficient implementation? Is it better to create a look-up-table?

Advertisement

I'm looking for an efficient way to get hammersley distribution points on the GPU. I found this page with some GLSL code that relies on bit-shifts. Is this efficient on the GPU? Are you aware of more efficient implementation? Is it better to create a look-up-table?

On Nvidia hardware I think bitwise shifts are probably going to be the most expensive integer operation, and integer operations are already slightly more expensive than floating point. For instance, a GTX 760 is going to be able to handle a theoretical 188 billion integer shifts per second, compared to a theoretical 2.25 trillion floating point operations per second.

OK, that's good to know.

This topic is closed to new replies.

Advertisement