java-stream

java-stream

This is an implementation of BabelStream in Java 8 which contains the following implementations:

• jdk-plain - Single threaded for
• jdk-stream - Threaded implementation using JDK8's parallel stream API
• tornadovm - A TornadoVM implementation for PTX/OpenCL
• aparapi - A Aparapi implementation for OpenCL

Build & Run

Prerequisites

• JDK >= 8

To run the benchmark, first create a binary:

> cd java-stream
> ./mvnw clean package

The binary will be located at ./target/java-stream.jar. Run it with:

> java -version                                                                                                    ✔  11.0.11+9 ☕  tom@soraws-uk  05:03:20 
openjdk version "11.0.11" 2021-04-20
OpenJDK Runtime Environment GraalVM CE 21.1.0 (build 11.0.11+8-jvmci-21.1-b05)
OpenJDK 64-Bit Server VM GraalVM CE 21.1.0 (build 11.0.11+8-jvmci-21.1-b05, mixed mode)
> java -jar target/java-stream.jar --help

For best results, benchmark with the following JVM flags:

-XX:-UseOnStackReplacement     # disable OSR, not useful for this benchmark as we are measuring peak performance  
-XX:-TieredCompilation         # disable C1, go straight to C2 
-XX:ReservedCodeCacheSize=512m # don't flush compiled code out of cache at any point 

Worked example:

> java -XX:-UseOnStackReplacement -XX:-TieredCompilation -XX:ReservedCodeCacheSize=512m -jar target/java-stream.jar
BabelStream
Version: 3.4
Implementation: jdk-stream; (Java 11.0.11;Red Hat, Inc.; home=/usr/lib/jvm/java-11-openjdk-11.0.11.0.9-4.fc33.x86_64)
Running all 100 times
Precision: double
Array size: 268.4 MB (=0.3 GB)
Total size: 805.3 MB (=0.8 GB)
Function    MBytes/sec  Min (sec)   Max         Average     
Copy        17145.538   0.03131     0.04779     0.03413     
Mul         16759.092   0.03203     0.04752     0.03579     
Add         19431.954   0.04144     0.05866     0.04503     
Triad       19763.970   0.04075     0.05388     0.04510     
Dot         26646.894   0.02015     0.03013     0.02259 

If your OpenCL/CUDA installation is not at the default location, TornadoVM and Aparapi may fail to detect your devices. In those cases, you may specify the library directly, for example:

> LD_PRELOAD=/opt/rocm-4.0.0/opencl/lib/libOpenCL.so.1.2 java -jar target/java-stream.jar ...

Instructions for TornadoVM

The TornadoVM implementation requires you to run the binary with a patched JVM. Follow the official instructions or use the following simplified instructions:

Prerequisites

• CMake >= 3.6
• GCC or clang/LLVM (GCC >= 5.5)
• Python >= 2.7
• Maven >= 3.6.3
• OpenCL headers >= 1.2 and/or CUDA SDK >= 9.0

First, get a copy of the TornadoVM source:

> cd
> git clone https://github.com/beehive-lab/TornadoVM tornadovm

Take note of the required GraalVM version in tornadovm/assembly/src/docs/10_INSTALL_WITH_GRAALVM.md. We'll use 21.1.0 in this example. Now, obtain a copy of GraalVM and make sure the version matches the one required by TornadoVM:

> wget https://github.com/graalvm/graalvm-ce-builds/releases/download/vm-21.1.0/graalvm-ce-java11-linux-amd64-21.1.0.tar.gz
> tar -xf graalvm-ce-java11-linux-amd64-21.1.0.tar.gz

Next, create ~/tornadovm/etc/sources.env and populate the file with the following:

#!/bin/bash
export JAVA_HOME=<path to GraalVM 21.1.0 jdk>
export PATH=$PWD/bin/bin:$PATH
export TORNADO_SDK=$PWD/bin/sdk
export CMAKE_ROOT=/usr          # path to CMake binary

Proceed to compile TornadoVM:

> cd ~/tornadovm
> . etc/sources.env
> make graal-jdk-11-plus BACKEND={ptx,opencl}

To test your build, source the environment file:

> source ~/tornadovm/etc/sources.env
> LD_PRELOAD=/opt/rocm-4.0.0/opencl/lib/libOpenCL.so.1.2 tornado --devices
Number of Tornado drivers: 1
Total number of OpenCL devices  : 3
Tornado device=0:0
        AMD Accelerated Parallel Processing -- gfx1012
                Global Memory Size: 4.0 GB
                Local Memory Size: 64.0 KB
                Workgroup Dimensions: 3
                Max WorkGroup Configuration: [1024, 1024, 1024]
                Device OpenCL C version: OpenCL C 2.0

Tornado device=0:1
        Portable Computing Language -- pthread-AMD Ryzen 9 3900X 12-Core Processor
                Global Memory Size: 60.7 GB
                Local Memory Size: 8.0 MB
                Workgroup Dimensions: 3
                Max WorkGroup Configuration: [4096, 4096, 4096]
                Device OpenCL C version: OpenCL C 1.2 pocl

Tornado device=0:2
        NVIDIA CUDA -- NVIDIA GeForce GT 710
                Global Memory Size: 981.3 MB
                Local Memory Size: 48.0 KB
                Workgroup Dimensions: 3
                Max WorkGroup Configuration: [1024, 1024, 64]
                Device OpenCL C version: OpenCL C 1.2

You can now use TornadoVM to run java-stream:

> tornado -jar ~/java-stream/target/java-stream.jar --impl tornadovm --arraysize 65536                              1 ✘  11.0.11+9 ☕  tom@soraws-uk  05:31:34 
BabelStream
Version: 3.4
Implementation: tornadovm; (Java 11.0.11;GraalVM Community; home=~/graalvm-ce-java11-21.1.0)
Running all 100 times
Precision: double
Array size: 0.5 MB (=0.0 GB)
Total size: 1.6 MB (=0.0 GB)
Using TornadoVM device:
 - Name     : NVIDIA GeForce GT 710 CL_DEVICE_TYPE_GPU (available)
 - Id       : opencl-0-0
 - Platform : NVIDIA CUDA
 - Backend  : OpenCL
Function    MBytes/sec  Min (sec)   Max         Average     
Copy        8791.100    0.00012     0.00079     0.00015     
Mul         8774.107    0.00012     0.00061     0.00014     
Add         9903.313    0.00016     0.00030     0.00018     
Triad       9861.031    0.00016     0.00030     0.00018     
Dot         2799.465    0.00037     0.00056     0.00041