[R] Enable OpenMP with AppleClang in XGBoost R package

[hcho3]

Follow-up to #5232. This PR uses Autoconf to detect the availability of OpenMP library (libomp) on Mac OSX. This is done by compiling the following program with command ${CC} -o conftest conftest.c /usr/local/lib/libomp.dylib -Xclang -fopenmp:

/* conftest.c */
#include <omp.h>

int main(void) {
  return (omp_get_max_threads() <= 1);
    // returns 0 when there's only one thread
    // returns 1 when there's more than one thread
}

Implication: Now Mac users can simply run install.packages('xgboost') and enjoy the performance of multiple threads, as long as they first install libomp package from Homebrew.

I tested this change as follows:

• I ran R CMD INSTALL . on my Macbook Pro. A sample XGBoost training job runs with all 4 threads:

library(xgboost)
data(agaricus.train, package='xgboost')
data(agaricus.test, package='xgboost')
train <- agaricus.train
test <- agaricus.test
bst <- xgboost(data = train$data, label = train$label, max.depth = 1, eta = 1,
               nthread = 4, nrounds = 1000, objective = "binary:logistic", verbose = 1,
               eval.metric = 'auc')

• Then I uninstalled libomp Homebrew and ran R CMD INSTALL . again. A prominent warning shows up, as I intended:

checking whether OpenMP will work in a package... no
*****************************************************************************************
WARNING: OpenMP is unavailable on this Mac OSX system. Training speed may be suboptimal.
         To use all CPU cores for training jobs, you should install OpenMP by running

             brew install libomp
*****************************************************************************************

Aside from the warning, XGBoost compiles successfully. The sample job ran without a problem, albeit this time with a single thread.

Note to reviewer: you should only review configure.ac. The configure script is generated from configure.ac when autoconf is run.

[trivialfis]

Awesome! Love it when everything is working out of box.