[racket] performance of iteration through a vector

Or, use typed racket.

Robby

On Sat, Aug 30, 2014 at 8:18 PM, Matthew Butterick <mb at mbtype.com> wrote:
> Usually you'll get the best performance from `for` loops by using the in-*
> iterators.
> In this case, you can use `in-vector` [1], which on my machine is faster:
>
> (let ((t (current-inexact-milliseconds))
>       (sum 0))
>   (for ((j (in-range 1000000)))
>     (for ((v (in-vector vec)))
>       (set! sum (+ sum v))))
>   (displayln (/ (- (current-inexact-milliseconds) t) 1000.0)))
>
> I assume this is a synthetic example, but you can also improve performance
> by using appropriate versions of `for`. This is faster than the last one:
>
> (let ((t (current-inexact-milliseconds)))
>   (for*/fold ([sum 0])([j (in-range 1000000)][v (in-vector vec)])
>     (+ sum v))
>   (displayln (/ (- (current-inexact-milliseconds) t) 1000.0)))
>
> And this is faster still:
>
> (let ((t (current-inexact-milliseconds)))
>   (for*/sum ([j (in-range 1000000)][v (in-vector vec)])
>     v)
>   (displayln (/ (- (current-inexact-milliseconds) t) 1000.0)))
>
>
> [1]
> http://docs.racket-lang.org/reference/sequences.html?q=in-vector#%28def._%28%28lib._racket%2Fprivate%2Fbase..rkt%29._in-vector%29%29
>
>
> On Aug 30, 2014, at 3:54 PM, Dmitry Pavlov <dpavlov at ipa.nw.ru> wrote:
>
> Hello,
>
> Consider the following program:
>
> (define n 100)
> (define vec (for/vector ((i (in-range n))) i))
>
> (let ((t (current-inexact-milliseconds))
>       (sum 0))
>   (for ((j (in-range 1000000)))
>     (for ((i (in-range n)))
>       (set! sum (+ sum (vector-ref vec i)))))
>   (displayln (/ (- (current-inexact-milliseconds) t) 1000.0)))
>
> (let ((t (current-inexact-milliseconds))
>       (sum 0))
>   (for ((j (in-range 1000000)))
>     (for ((v vec))
>       (set! sum (+ sum v))))
>   (displayln (/ (- (current-inexact-milliseconds) t) 1000.0)))
>
>
> On my system (64-bit linux, Racket 6.1.0.2), it gives the following result:
>
> 1.016682861328125
> 6.3261611328125
>
>
> So we can make a conclusion that (for ((v vec)) ...) is
> 6x slower than (for ((i (in-range n))) ... (vector-ref vec i) ...)
> Is it normal? Would you advise to use the explicit (vector-ref)
> when performance matters?
>
> Best regards
>
> Dmitry
>
> ____________________
>  Racket Users list:
>  http://lists.racket-lang.org/users
>
>
>
>
> On Aug 30, 2014, at 3:54 PM, Dmitry Pavlov <dpavlov at ipa.nw.ru> wrote:
>
> Hello,
>
> Consider the following program:
>
> (define n 100)
> (define vec (for/vector ((i (in-range n))) i))
>
> (let ((t (current-inexact-milliseconds))
>       (sum 0))
>   (for ((j (in-range 1000000)))
>     (for ((i (in-range n)))
>       (set! sum (+ sum (vector-ref vec i)))))
>   (displayln (/ (- (current-inexact-milliseconds) t) 1000.0)))
>
> (let ((t (current-inexact-milliseconds))
>       (sum 0))
>   (for ((j (in-range 1000000)))
>     (for ((v vec))
>       (set! sum (+ sum v))))
>   (displayln (/ (- (current-inexact-milliseconds) t) 1000.0)))
>
>
> On my system (64-bit linux, Racket 6.1.0.2), it gives the following result:
>
> 1.016682861328125
> 6.3261611328125
>
>
> So we can make a conclusion that (for ((v vec)) ...) is
> 6x slower than (for ((i (in-range n))) ... (vector-ref vec i) ...)
> Is it normal? Would you advise to use the explicit (vector-ref)
> when performance matters?
>
> Best regards
>
> Dmitry
>
> ____________________
>  Racket Users list:
>  http://lists.racket-lang.org/users
>
>
>
> ____________________
>   Racket Users list:
>   http://lists.racket-lang.org/users
>