Awesome

WebAssembly in R using wasmer

The goal of wasmr is to run WebAssembly code from R using the wasmer runtime.

This is a early version with probably bugs, missing features and not the best code quality. But it already works and I am happy to hear feedback. The goal is that this package evolves into something stable.

The package is mainly written in C++ using the C API of the wasmer runtime which is written in rust. You therefore need a rust compiler to install the package.

Installation

remotes::install_github("dirkschumacher/wasmr")

Also make sure to have a very recent rust compiler. The best way to install the rust toolchain is to use rustup.

Example

library(wasmr)

f <- system.file("examples/sum.wasm", package = "wasmr")
instance <- instantiate(f)
instance$exports$sum(10, 20)
#> [1] 30

f <- system.file("examples/hello.wasm", package = "wasmr")
instance <- instantiate(f)
memory_pointer <- instance$exports$hello()
hi <- instance$memory$get_memory_view(memory_pointer)
rawToChar(hi$get(1:11))
#> [1] "Hello world"

f <- system.file("examples/fib.wasm", package = "wasmr")
instance <- instantiate(f)
instance$exports$fib(20)
#> [1] 6765

fib <- function(n) {
  if (n < 2) return(n)
  fib(n - 1) + fib(n - 2)
}

microbenchmark::microbenchmark(
  instance$exports$fib(20),
  fib(20)
)
#> Unit: microseconds
#>                      expr      min        lq       mean    median
#>  instance$exports$fib(20)   71.143   88.4745   158.3407   139.814
#>                   fib(20) 8037.195 9556.2085 13024.0362 10415.965
#>         uq       max neval
#>    150.472  1019.363   100
#>  14568.746 48506.287   100

Memory

f <- system.file("examples/fib.wasm", package = "wasmr")
instance <- instantiate(f)
instance$memory$get_memory_length()
#> [1] 2

# grow the number of pages
instance$memory$grow(1)
instance$memory$get_memory_length()
#> [1] 3

f <- system.file("examples/hello.wasm", package = "wasmr")
instance <- instantiate(f)
memory_pointer <- instance$exports$hello()
memory <- instance$memory$get_memory_view(memory_pointer)
memory$get(1:11)
#>  [1] 48 65 6c 6c 6f 20 77 6f 72 6c 64
rawToChar(memory$get(1:11))
#> [1] "Hello world"

# you can also write to the internal memory
memory$set(5:6, charToRaw("o_"))
rawToChar(memory$get(1:11))
#> [1] "Hello_world"

Imports

set.seed(42)
imports <- list(
  env = list(
    add = typed_function(
      function(a, b) {
        # use R's RNG to add a random number to the result
        a + b * as.integer(runif(1) * 100)
      },
      param_types = c("I32", "I32"),
      return_type = "I32"
    )
  )
)
f <- system.file("examples/sum_import.wasm", package = "wasmr")
instance <- instantiate(f, imports)
# sum: add(a, b) + 42
instance$exports$sum(1, 5)
#> [1] 498

Caveats and limitations

• It is not feature complete and does not support everything that wasm supports
• Imported functions can only have integer parameters
• No globals
• There is hardly any documentation except for the examples
• I32/I64 are mapped to IntegerVector and F32/F64 to NumericVector. Currently no way to differentiate.
• WIP

Inspiration and References

• wasmer Python - haven’t tried it but looked at the API/examples
• wasmer - especially the C api and the tests give some good examples.
• @jeroen ’s rust template

Contribute

While this is work in progress, the best way to contribute is to test the package and write/comment on issues. Before sending a PR it would great to post an issue first to discuss.

License

MIT just like wasmer.