ScalaCL lets you run Scala code on GPUs through OpenCL (BSD-licensed).
WORK IN PROGRESS (see ScalaCL if you want something that works, albeit only on Scala 2.9.x).
See slides from ScalaCL + Reified talk @ scala.io 2013.
Features of the new design (v3, rewritten from scratch again!):
- Much better asynchronicity support (now requires OpenCL 1.1), and much better performance in general
- Support for captures of constants and OpenCL arrays
- Support for lazy clones for fast zipping
- Kernels are now fully specialized on static types and generated at compile-time (allows much faster startup and caching at runtime)
- ScalaCL Collections no longer fit in regular Scala Collections, to avoid silent data transfers / conversions when using unaccelerated methods (syntax stays the same, though)
- No more CLRange: expecting compiler to do its job
TODO
- Finish Scalaxy/Reified integration (started under CLFunc / CLFuncUtils)
- Add more tests: DataIO, CodeConversion, scheduling, uniqueness / caching of kernels
- Implement more DataIO[T], support case classes as tuples
- Catch up with compiler plugin:
- Auto-vectorization
- 1D works
- Add 2D
- add filters
- Import Scalaxy streams, make them work with scala.reflection.api.Universe
- Auto-vectorization
- Plug some v2 runtime code back (filtered array compaction, reduceSymmetric, parallel sums...)
- Benchmarks!
- Wanna help? Ping the NativeLibs4Java mailing-list!
Usage
scalaVersion := "2.11.4"
libraryDependencies += "com.nativelibs4java" %% "scalacl" % "0.3-SNAPSHOT"
// Avoid sbt-related macro classpath issues.
fork := true
// Scalaxy/Reified snapshots are published on the Sonatype repository.
resolvers += Resolver.sonatypeRepo("snapshots")Examples
The following example currently works:
import scalacl._
case class Matrix(data: CLArray[Float],
rows: Int,
columns: Int)
(implicit context: Context)
{
def this(rows: Int, columns: Int)
(implicit context: Context) =
this(new CLArray[Float](rows * columns), rows, columns)
def this(n: Int)
(implicit context: Context) =
this(n, n)
def putProduct(a: Matrix, b: Matrix): Unit = {
assert(a.columns == b.rows)
assert(a.rows == rows)
assert(b.columns == columns)
kernel {
// This block will either be converted to an OpenCL kernel or cause compilation error
for (i <- 0 until rows;
j <- 0 until columns) {
// c(i, j) = sum(k, a(i, k) * b(k, j))
data(i * columns + j) = (
for (k <- 0 until a.columns) yield
a.data(i * a.columns + k) * b.data(k * b.columns + j)
).sum
}
}
}
def putSum(a: Matrix, b: Matrix): Unit = {
assert(a.columns == b.columns && a.columns == columns)
assert(a.rows == b.rows && a.rows == rows)
kernel {
for (i <- 0 until rows; j <- 0 until columns) {
val offset = i * columns + j
data(offset) = a.data(offset) + b.data(offset)
}
}
}
}
implicit val context = Context.best
val n = 10
val a = new Matrix(n)
val b = new Matrix(n)
val out = new Matrix(n)
out.putProduct(a, b)
println(out.data)