You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Given that CUDA now supports shuffle instructions with 64-bit values, would it be possible to replace the 32-bit address formats with full 64 bit pointers within each slab to eliminate the translation cost?
The text was updated successfully, but these errors were encountered:
Given that CUDA now supports shuffle instructions with 64-bit values, would it be possible to replace the 32-bit address formats with full 64 bit pointers within each slab to eliminate the translation cost?
The text was updated successfully, but these errors were encountered: