Several CUDA function names have the word Symbol in them.
e.g. cudaMemcpyToSymbol() , cudaMemcpyFromSymbol(), cudaGetSymbolAddress() etc.
The explanation of what these functions do is clear enough in the Programming guide. What I
dont get is the choice of the word “symbol” in naming these functions. What does it mean?
They use the word “symbol” because it is possible for these functions to take the name of a device variable as a string. For instance:
Presumably, this strange syntax allows some use case that I have yet to think of 🙂