29.01.2013 Views

Tutorial CUDA

Tutorial CUDA

Tutorial CUDA

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Execution Configuration: Constraints<br />

Maximum # of threads per block: 512<br />

# of active threads limited by resources:<br />

© NVIDIA Corporation 2008<br />

# of registers per multiprocessor (register pressure)<br />

Amount of shared memory per multiprocessor<br />

Use –maxrregcount=N flag to NVCC<br />

N = desired maximum registers / kernel<br />

At some point “spilling” into LMEM may occur<br />

Reduces performance – LMEM is slow<br />

Check .cubin file for LMEM usage

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!