The resultant of two vectors, calculated by adding respective components of each vector is called vector sum.

main():
1. Allocate space for 3 vectors A, X, Y (of length 2000000).
2. Define vectors X and Y (A = X + Y will be computed).
3. Calculate expected value for varifying results.
4. Run vector sum on with various approaches.
5. Free vectors A, X, Y.

$ nvcc -std=c++17 -Xcompiler -O3 main.cu
$ ./a.out

# CPU vector-sum ...
# Execution time: 0.0 ms
#
# GPU vector-sum, simple ...
# Execution time: 6.2 ms
#
# GPU vector-sum, multiple ...
# Execution time: 6.1 ms
#
# GPU vector-sum, chunked with stream ...
# Execution time: 7.5 ms
#
# GPU vector-sum, chunked with streams ...
# Execution time: 7.6 ms
#
# GPU vector-sum, chunked with interleaved streams ...
# Execution time: 6.7 ms

See main.cu for code.

References

• CUDA by Example :: Jason Sanders, Edward Kandrot