reduce interrupt frequency in ray_shade

[brodieG]

I'm working on a response to Wolf Vollprecht's Next Journal article, and I noticed when running benchmarks that ray_trace runs substantially slower than the Julia version. I didn't think this should be the case, so I dug into it and noticed that Rcpp::checkUserInterrupt() is called in the innermost for loop. This is a somewhat expensive operation and can be just as effective if we check every few thousand loops.

Making the change increases performance by 5x (see below). Presumably we could make a similar change to the multi core version, and to other places were interrupts are used, but I haven't checked that. I figure you might have opinions about how to do this, so I didn't want to spend too much time changing things before getting a sense for what you want.

I did also make a change to NEWS, but only because I've been asked to that in pull requests into other projects. Feel free to edit as you see fit (obviously).

Before changes:

library(rayshader)
elev <- seq(-90, 90, length=25)
microbenchmark::microbenchmark(times=10,
  ray_shade(volcano, elev, 45, lambert=FALSE, progbar=FALSE)
)
## Unit: milliseconds
##                                                            expr      min
##  ray_shade(volcano, elev, 45, lambert = FALSE, progbar = FALSE) 134.2335
##        lq     mean   median       uq      max neval
##  135.0545 162.9438 143.9956 177.8761 242.3845    10

After changes:

library(rayshader)
elev <- seq(-90, 90, length=25)
microbenchmark::microbenchmark(times=10,
  ray_shade(volcano, elev, 45, lambert=FALSE, progbar=FALSE)
)
## Unit: milliseconds
##                                                            expr     min
##  ray_shade(volcano, elev, 45, lambert = FALSE, progbar = FALSE) 25.9512
##        lq     mean   median       uq      max neval
##  26.52456 28.63295 27.44805 27.87292 41.60245    10

[tylermorganwall]

Thanks for the bug report! I just pushed an update that moves the check to the outer loop.

[tylermorganwall]

I didn't pull your request (moving the call to the inner loop was all that was required), but I did include you on the NEWS for the update.

[brodieG]

No problem. I started off with that change first, but then wanted something that you could use for both this and the multicore version which doesn't loop through rows.