Optimization, my favorite Nim topic!
Runtime of your version for me: 0.766 s
res.add board[d[i]]
This line resizes the `res` seq, instead you should allocate it to have the
correct size directly. Like this:
proc do_something(board: seq[char]): seq[char] =
var res = newSeq[char](board.len * Directions.len)
for i, p in board:
for j, d in Directions:
res[i * Directions.len + j] = board[d[i]]
return res
New runtime: 0.073 s
With the new devel branch of Nim you can also use `newSeqOfCap` if you don't
want to manually calculate the index:
proc do_something(board: seq[char]): seq[char] =
var res = newSeqOfCap[char](board.len * Directions.len)
for i, p in board:
for d in Directions:
res.add board[d[i]]
return res
As a workaround you can also do `var res = newSeq[char](board.len *
Directions.len); res.setLen(0)` to achieve the same effect.
res = do_something(board)
You create a new `res` seq for every call to do_something. Instead you could
reuse a single data structure and pass it to `do_something`. But that's
probably not what you want to benchmark then.
Full code with some cleanup (use implicit result variable, no `$` for echo,
arrays instead of seqs, more consts):
# Nim example to test execution speed against a Python version
import times, os
const
board = ['0', 'p', '.', 'p', 'P', '.', 'p', 'P', '.', 'p', '.', 'p', 'P',
'.', 'p']
NE = [8, 2, 13, 4, 5, 7, 8, 3, 9, 12, 7, 8, 3, 9, 12]
Directions = [NE, NE, NE, NE] # for short: in real application the array
elements differ
proc do_something(board: openarray[char]): seq[char] =
result = newSeqOfCap[char](board.len * Directions.len)
for i in 0 .. board.high:
for d in Directions:
result.add board[d[i]]
let t0 = cpuTime()
var res: seq[char]
const count = 100_000
for i in 1..count:
res = do_something(board)
let t1 = cpuTime()
echo "***** Time elapsed for Nim: ", t1 - t0, " Counts: ", count
echo res.repr
New runtime: 0.065 s
For comparison, Python runtime: 0.783 s
