search

jroose / clpp

Automatically exported from code.google.com/p/clpp
0 stars 1 forks source link

What's the difference between "default scan" and "GPU scan"? #6

Open GoogleCodeExporter opened 8 years ago

GoogleCodeExporter commented 8 years ago 
I am new to clpp. Just want to know what is the difference between these two 
scans?

[I did not find a better place to put this question]

Original issue reported on code.google.com by rongguod...@gmail.com on 21 Jun 2012 at 9:48

GoogleCodeExporter commented 8 years ago 
[deleted comment]
GoogleCodeExporter commented 8 years ago 
For details of the algorithm behind each of these 2 implementations, read the 
references cited in each of the 2 individual .cl files. 

"GPU scan" is much simpler and performance much better (at least on my GPU 
hardware platform), using a single invokation of a kernel instead of multilple 
GPU kernels that "default scan" does, and it also does a good job optimizing to 
"warp sizes", or SIMT size, of the underlying GPU hardware to eliminate GPU 
thread synchronization.

Original comment by Edward.K...@gmail.com on 1 May 2013 at 10:27