cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

matze_de
Journeyman III

async_work_group_copy issue

multiple async_work_group_copy problem

Hello every one,

i have a problem with  async_work_group_copy if i use it twice in a kernel it somehow does not work. 

The Task is to create a MxN Matrix where every field is calculated with 2 x 45 x float3 (the 48 in copy is because of alignment (without using local memory it wirks fine).

The fields are calculated independently, i choose a 2-dim 16x16 ndrange.

So i thought i could copy the 2x16x48(45) float 3 to local memory because of multiple access.

My caching works if i only cache 16x48(45)

 

 

 

0 Likes
3 Replies
matze_de
Journeyman III

very sorry for posting this twice, my browser crashed during posting

0 Likes

It looks like the issues has not been explained completely. Can you provide your system configuration(CPU,GPU,SDK,DRIVER,OS) and a small  test case.

Thanks

0 Likes
genaganna
Journeyman III

Originally posted by: matze_de Hello every one,

 

i have a problem with  async_work_group_copy if i use it twice in a kernel it somehow does not work. 

 

The Task is to create a MxN Matrix where every field is calculated with 2 x 45 x float3 (the 48 in copy is because of alignment (without using local memory it wirks fine).

 

The fields are calculated independently, i choose a 2-dim 16x16 ndrange.

 

So i thought i could copy the 2x16x48(45) float 3 to local memory because of multiple access.

 

My caching works if i only cache 16x48(45)

 

 

Could you paste you kernel code here? It would be good if paste runtime code also.

0 Likes