ToDo: Eliminate non-coalesced memory operations for CUDA/OpenCL/Stream cores

http403 / pyrit

Automatically exported from code.google.com/p/pyrit

0 stars 0 forks source link

ToDo: Eliminate non-coalesced memory operations for CUDA/OpenCL/Stream cores #66

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago

Open ToDo nobody is currently working on:

The CUDA/OpenCL/Stream cores make heavy use of non-coalesced memory
operations. This causes pipeline stalls due to memory latency which we
currently need to drown a in huge amount of threadblocks.

Starting with the CUDA-core, we need to rewrite the memory access patterns
to reduce the performance penalty due to memory latency.

Original issue reported on code.google.com by lukas.l...@gmail.com on 4 Nov 2009 at 7:44

GoogleCodeExporter commented 9 years ago

Original comment by lukas.l...@gmail.com on 4 Nov 2009 at 7:44

GoogleCodeExporter commented 9 years ago

Original comment by lukas.l...@gmail.com on 4 Nov 2009 at 7:48

Added labels: Type-Enhancement
Removed labels: Type-Defect

GoogleCodeExporter commented 9 years ago

here is something, see draft... data passed to kernel should be coalesced, and 
16byte aligned
what about?

Original comment by masterzorag on 10 Feb 2011 at 11:30

Attachments:

draft