SCAMP's performance drops considerably when using multithreading when the output matrix is large. This is because it makes a copy of the the output matrix for each thread.
We can fix this by only allocating what the tile needs and transferring only those values.
SCAMP's performance drops considerably when using multithreading when the output matrix is large. This is because it makes a copy of the the output matrix for each thread.
We can fix this by only allocating what the tile needs and transferring only those values.