lms-intrinsics: Intel Intrinsics for Lightweight Modular Staging Framework (LMS)

lms-intrinsics is a package that enables the use of SIMD x86 instructions in the Lightweight Modular Staging Framework (LMS). While most SIMD instruction are available as a low-level machine code, the lms-intrinsics package focuses on the C SIMD instrinsics, which are supported by most modern C compilers such as gcc, Intel Compiler, LLVM, etc, and provides the appropriate generation of a vectorized C code. Currently the following instruction sets are supported:

• MMX
• SSE / SSE2 / SSE3 / SSSE3 / SSE4.1 / SSE4.2
• AVX / AVX2
• FMA
• AVX-512
• KNC
• SVML

While the same set of instructions are also available on other x86 CPUs, the focus of lms-intrinsics is mainly towards Intel supported intrinsics, with the aim to include all listed intrinsics in the Intel Intrinsics Guide. Hopefully, this work will inspire the inclusion of other instruction sets and extensions such as NEON and others in LMS.

This implementation of lms-intrinsics is work done by Ivaylo Toskov as part of a master thesis project at the Department of Computer Science at ETH Zurich Switzerland, supervised by Alen Stojanov and Markus Püschel.

Usage - SBT

lms-intrinsics can be used though SBT directly, just include the following in your build.sbt file:

resolvers += Resolver.sonatypeRepo("snapshots")

// LMS dependency
libraryDependencies += "org.scala-lang.lms" % "lms-core_2.11" % "1.0.0-SNAPSHOT"

// support for unsigned primives in Scala
libraryDependencies += "ch.ethz.acl" %% "scala-unsigned" % "0.1-SNAPSHOT"

// the main lms-intrinsics package
libraryDependencies += "ch.ethz.acl" %% "lms-intrinsics" % "0.0.2-SNAPSHOT"

If you like to use the latest version of the lms-intrinsics GitHub repository, add project dependency in your Build.scala file:

.dependsOn(RootProject(uri("https://github.com/ivtoskov/lms-intrinsics")))

Note that the lms-intrinsics package contains the definition of more than 5000 Intel Intrinsics and other SIMD functions, therefore compilation time with Scala can be quite substantial.

Quick Start

A minimal example of how lms-intrinsics can be used, can be illustrated in 3 steps:

Import LMS, the support for unsigned primives, and import lms-intrinsics:

import scala.lms.common._
import ch.ethz.acl.passera.unsigned._
import ch.ethz.acl.intrinsics._

Then, create an IR that spports AVX and AVX2 instructons:

val AVX2_IR = new BaseExp
  with ...
  with IntrinsicsBase
  with AVX
  with AVX2 { }

Stage a simple function that takes two arrays of floats, a and b asarguments. Extracts the first 8 elements from the first array, adds them, and writes the result to the first 8 elements of array b:

import AVX2_IR._
def add_first_8elements(a: Rep[Array[Float]], b: Rep[Array[Float]]): Rep[Unit] = {
  val b_write = reflectMutableSym(b.asInstanceOf[Sym[Array[Float]]])
  val tmp1 = _mm256_loadu_ps(a, Const(0))
  val tmp2 = _mm256_add_ps(tmp1, tmp1)
  _mm256_storeu_ps(b_write, tmp2, Const(0))
}

That's it. A complete example reflecting these steps is available in the lms-intrinsics-example repository.

Contribution policy

Contributions via GitHub pull requests are gladly accepted from their original author.

License

This code is open source software licensed under the Apache 2.0 License. Feel free to use it accordingly.