Conversation

Simon Brown

@sjb3d

·

May 30, 2013

Would be nice to have a #CUDA API to explicitly clone a module to get a new copy of constant memory but shared code.

1

Allan MacKinnon

@pixelio

·

May 30, 2013

Replying to

@sjb3d

@sjb3d

This is one approach: https://devtalk.nvidia.com/default/topic/540182/cuda-programming-and-performance/how-is-__constant__-memory-with-respect-to-cuda-streams-/post/3785705/#3785705… And another: http://pixel.io/blog/2013/5/9/kernel-arguments-vs-__constant__-variables.html…

1

2

Simon Brown

@sjb3d

Replying to

@pixelio

@pixelio

Thanks yep using kernel args but code cleaner with module level. Could repeat load modules but no idea if driver will share code!

1:50 PM · May 30, 2013·Twitter Web Client