Using shared memory should be with caution. As the size of shared memory is typically small, using shared memory could limit the amount of blocks that a GPU core could accommodate, which reduces GPU's ability of hiding memory latency, therefore slow the program.
Something I learned from asst2.
Using shared memory should be with caution. As the size of shared memory is typically small, using shared memory could limit the amount of blocks that a GPU core could accommodate, which reduces GPU's ability of hiding memory latency, therefore slow the program.