At Fri, 6 Nov 2020 20:14:51 +0100, Dominik Pantůček wrote:
> I assume that CS' unsafe-flvector-set! is actually pretty safe when it
> comes to flonum-convertible numbers. It might be a bit faster because it
> lacks the contract, but definitely it is not a "low level" variant.
Hm, right --- it's implemented by `bytevector-ieee-double-native-set!`,
which is defined to convert its last argument to a flonum (so the
conversion applies even in unsafe mode). I hadn't noticed that before,
so thanks for pointing it out!
I will investigate faster option. A primitive without conversion could
make the safe `flvector-set!` slightly faster, too, by avoiding a
redundant check.
> Also given the fact that unsafe-fl+ and others give no performance
> advantage
I think `unsafe-fl+` can still be faster. For example, the second loop
below is the unsafe one, and it's 10-20% faster. I constructed this
program carefully, though, to avoid flonum inference and to avoid
allocating flonum results (which swamps the cost of checks).
The performance guide is pretty conservative about suggesting
performance improvements via unsafe operations, already, but
suggestions are welcome.
Mathew
----------------------------------------
#lang racket/base
(require racket/unsafe/ops
racket/flonum)
(define N 100000000)
(define (f x)
(fl< (fl+ x x) 0.0))
(define (u-f x)
(fl< (unsafe-fl+ x x) 0.0))
(set! f f)
(set! u-f u-f)
(collect-garbage)
(time
(for/fold ([v #f]) ([i (in-range N)])
(f 1.0)))
(collect-garbage)
(time
(for/fold ([v #f]) ([i (in-range N)])
(u-f 1.0)))
--
You received this message because you are subscribed to the Google Groups
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to racket-users+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/racket-users/20201106124546.1c8%40sirmail.smtps.cs.utah.edu.
