B.3. conv2d

conv2d -- 2D convolution
   -1 -- float, support=full
   -2 -- float, support=same
   -3 -- float, support=min

 Parameters:
  -p:rows     ROWS -- rows in input matrix
                      (default 16)
  -p:m        M    -- rows in coefficient matrix
                      (default 3)
  -p:n        N    -- columns in coefficient matrix
                      (default 3)
  -p:mn       MN   -- if not zero, set both M and N
                      to MN (default 0)
  -start      N    -- starting problem size 2^N
                      (default 4, that is 16 points)
  -loop_start N    -- initial number of calibration
                      loops (default 5000)