*For me both "-cl-opt-disable" and "-op-disable" work.
But those options do not disable kernel optimizations
*Access to memory with volatile qualifier is optimized out by compiler ( only first access is generated to IL/ISA , following reads from the same address are removed ).
Sample code
global volatile flot4* v;
a1 = v[0]; <- this read is generated
a2 = v[0]; <- optimized out
a3 = v[0]; <- optimized out
*It is not that the threads in OpenCL are tiled, but blocked linear.
Linear is sequential along a single line, blocked linear is sequential within a block and each block is sequential. In tiled mode it usually follows a hierarchical Z pattern.
Saturday, 7 November 2009
Subscribe to:
Post Comments (Atom)
0 comments:
Post a Comment