This is a proof-of-concept profiler for R based on Scalene.
The implementation is written into GNU R 4.3.3.
To use it it we need to build it from source code in the r directory:
./configuration
make
Our profiler does not support profiling bytecode. Therefore, we need to disable JIT.
That can be done either through environment variable R_ENABLE_JIT=0 or in code by compiler::enableJIT(0).
To enable our profiler we must use the R_SCALENE environment variable and set it to a filename where the profiling output is stored.
To ensure that we can obtain lines of code from srcref run the desired program by source("filename", keep.source = TRUE).
R_ENABLE_JIT=0 R_SCALENE=output.txt ./r/bin/R
The profiled code must be wrapped in the ..my_profile... function.
To start profiling we run the function.
If we pass names of functions to ..my_profile... call, then we profile the lines of those functions as well.
For an example of usage look at the following code.
facto <- function (n) {
if (n == 0) {
return(1)
} else {
return(n * facto(n - 1))
}
}
facti <- function (n) {
result <- 1
for (i in 1:n) {
result <- i * result
}
return (result)
}
..my_profile.. <- function (...) {
lapply(1:10000, function(i) {
lapply(1:50, function(i) {
facto(i)
facti(i)
})
})
}
..my_profile..()
We can run the profiler for example by R_ENABLE_JIT=0 R_SCALENE=output.txt ./r/bin/R where we run source("filename", keep.source = TRUE) with the file to be profiled.