Pierre Alexandre Tremblay

Am I deluded (on that respect as well I should say) to believe it is not possible to use conditional statements in jit.expr?

jit.expr @expr "(in[0].p[0] > 0.5) ? (2) : (1)"

jit.expr @expr "if (in[0].p[0] > 0.5) then (2) else (1)"

conditionals-in-jit-expr

thanks for the help, but that was only a theoretical problem. I need a proper conditional in my case that I can post (or that you can see on the ftm list if you happen to follow it)

I just checked out your post on the FTM archives and your initial way of doing it (using multiply to select based on conditionals) was what I was going to suggest.

I'd guess jit.expr probably does not support conditional branching on individual elements - this is somewhat against the notion of processing blocks / vectors of data....

What exactly is you ultimate goal other than to make your code faster? I'd guess that you may need to move in c or java to do this and just roll your own external.

Hmm - just looked again and it may be possible to do this more neatly, but I don't totally understand it yet..

- I can't see where you figure out when you've reached the target - how does this work?

- how is the increment value calculated?
- I can't see where you figure out when you've reached the target - how does this work?

OK - scratch all of those questions - it's all become patently clear to me.

So - I think this can be faster / neater.

1 - don't use jit.expr - use jit.op - this saves you recalculating values all over the place.

2 - use a better (faster) linear interpolation when selecting between two values like so:

that's one multiply rather than two, and hence faster.

3 - I think if you store the direction of travel you can make a more efficient target reached test (as I presume that equal to counts) assuming you can do an XOR test

That's more efficient than what you have now but you'll need another comparison for the direction calculation - still 2 comparisons is better than 4

Neat patch by the way. Maybe you could post it here so that this willl make sense to anyone else reading this thread?

t = target
d = 1 if moving up, 0 if down
c = current value

Have you seen Norbert's version in FTM? It does smell like yours ;-) I'll port his implementation (and try yours) later today and will post both... actually, I am tempted to post it here with a special request to Andrew to look at it as he is top class ;-)

Have you seen Norbert's version in FTM?  It does smell like yours ;-)  I'll port his implementation (and try yours) later today and will post both... actually, I am tempted to post it here with a special request to Andrew to look at it as he is top class ;-)

conditionals in jit.expr