This package brings the geiger C++14 micro-benchmarking library to R.
Note that using it requires a C++14 compiler. It may go to CRAN once R 3.4.0 is released as this will bring support for C++14 compilation to R.
It can make use of the PAPI interface via e.g. libpapi-dev on Debian or Ubuntu. The
default is to build without PAPI unless enabled. We plan to make this an automatic
detection via configure.
In its first release, RcppGeiger contains two of the examples from geiger:
The fist example contains two simple lambda function with an single operation each. We
replaced the call to std::rand() with one to R's RNG:
// [[Rcpp::export]]
void simple() {
geiger::init();
// A benchmark suite that does only time measurement
geiger::suite<> s;
s.add("R::runif",
[]() {
//std::rand();
R::runif(0, RAND_MAX);
});
s.add("vector push_back",
[]() {
std::vector<int> v;
v.push_back(1000);
});
// Redirection of each test result to the "console" printer
s.set_printer<geiger::printer::console<>>();
// Run all benchmarks
s.run();
}
Running this from R shows the following outpu on my laptop:
R> simple()
Test Iterations ns / iteration
----------------------------------------------------
R::runif 109,254,138 9
vector push_back 18,143,368 55
R> The second example compares a linear walk through a vector with a random walk. We once
again replace std::rand with R::runif. The example show how calls to the
geiger::suite instance can be chained.
static const int size = 1024 * 1024 * 16;
static const int batch = 64;
static const int mask = size - 1;
static const int prime = 7919;
static int sum = 0;
auto linear_walk() {
std::vector<char> v(size, 'a');
return [v = std::move(v)]() {
static std::size_t pos = /*std::rand()*/ static_cast<int>(R::runif(0, RAND_MAX)) & mask;
for (int i = 0; i < batch; ++i) {
sum += v[pos];
pos = (pos + 1) & mask;
}
};
}
auto random_walk() {
std::vector<char> v(size, 'a');
return [v = std::move(v)]() {
static std::size_t pos = /*std::rand()*/ static_cast<int>(R::runif(0, RAND_MAX)) & mask;
for (int i = 0; i < batch; ++i) {
sum += v[pos];
pos = (pos * prime) & mask;
}
};
}
//' @rdname simple
// [[Rcpp::export]]
void walk() {
geiger::init();
#ifdef USE_PAPI
geiger::suite<geiger::cache_profiler> s;
#else
geiger::suite<> s;
#endif
s.add("linear walk", linear_walk())
.add("random walk", random_walk())
.set_printer<geiger::printer::console<>>()
.run(size / batch);
}
It produces the following output on the laptop:
R> walk()
Test Iterations ns / iteration
-----------------------------------------------
linear walk 262,144 57
random walk 262,144 800
R> On a server with both PAPI support and an appropriate cpu, we see
R> walk()
Test Iterations ns / iteration L1 DCM L2 DCM L3 TCM
--------------------------------------------------------------------------------------
linear walk 262,144 78 1.00 0.03 0.03
random walk 262,144 322 63.94 63.48 25.55
R> The package is not yet on CRAN and must be installed from the repository. For now cloning and installing locally may be easiest.
Of course
R> remotes::install_github("eddelbuettel/rcppgeiger")works as well.
The package is pretty new and raw. Expect changes.
There are several TODO items (such as collecting data and bringing it back to R) which should be listed as well.
Dirk Eddelbuettel
GPL (>= 2)