Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[SYSTEMML-769] [WIP] Support for automatic detection of native BLAS a…
…nd GPU backend 1. Support for automatic detection of BLAS (MKL and OpenBLAS) and GPU backend. 2. Added native matmult and conv2d functions. In case if native library is not available, we fallback to Java implementation. 3. This will allow us to explore distributed GPU solution.
- Loading branch information
Niketan Pansare
committed
Jan 6, 2017
1 parent
7a30925
commit e2d9a16
Showing
29 changed files
with
558 additions
and
173 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,137 @@ | ||
| --- | ||
| layout: global | ||
| title: Using systemml-accelerator | ||
| description: Using systemml-accelerator | ||
| --- | ||
| <!-- | ||
| {% comment %} | ||
| Licensed to the Apache Software Foundation (ASF) under one or more | ||
| contributor license agreements. See the NOTICE file distributed with | ||
| this work for additional information regarding copyright ownership. | ||
| The ASF licenses this file to you under the Apache License, Version 2.0 | ||
| (the "License"); you may not use this file except in compliance with | ||
| the License. You may obtain a copy of the License at | ||
| http://www.apache.org/licenses/LICENSE-2.0 | ||
| Unless required by applicable law or agreed to in writing, software | ||
| distributed under the License is distributed on an "AS IS" BASIS, | ||
| WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| See the License for the specific language governing permissions and | ||
| limitations under the License. | ||
| {% endcomment %} | ||
| --> | ||
|
|
||
| * This will become a table of contents (this text will be scraped). | ||
| {:toc} | ||
|
|
||
| <br/> | ||
|
|
||
| ## Introduction | ||
|
|
||
| The [systemml-accelerator](https://github.com/niketanpansare/systemml-accelerator) packages system-dependent libraries | ||
| to simplify deployment. It can be used to allow SystemML to use native BLAS as well as use hardware accelerators (such as Nvidia's GPU). | ||
|
|
||
| If you are [installing SystemML using pip](https://apache.github.io/incubator-systemml/beginners-guide-python#install-systemml), | ||
| no additional action is required. If you intend to use SystemML in any other way, you have to ensure that `systemml-accelerator.jar` | ||
| is available. | ||
|
|
||
| ## Using native BLAS | ||
|
|
||
| By default, SystemML implements all its matrix operations in Java. This simplifies deployment especially in a distributed environment. | ||
| However, in some cases (such as deep learning), the user might want to use native BLAS rather than SystemML's internal Java library. | ||
| The current version supports only 64-bit JVM and Intel MKL (recommended) and OpenBLAS. For any other setup, we fallback to SystemML's internal Java library. | ||
|
|
||
| ### Steps for installing Intel MKL | ||
|
|
||
| Download and install the [community version of Intel MKL](https://software.intel.com/sites/campaigns/nest/). | ||
| Intel requires you to first register your email address and then sends the download link to your email address with license key. | ||
|
|
||
| <div class="codetabs"> | ||
| <div data-lang="Linux" markdown="1"> | ||
| ```bash | ||
| 1. Extract the downloaded .tgz file and execute install.sh. | ||
| ``` | ||
| </div> | ||
| <div data-lang="Windows" markdown="1"> | ||
| ```bash | ||
| 1. Execute the downloaded .exe file and follow the guided setup. | ||
| ``` | ||
| </div> | ||
| </div> | ||
|
|
||
| ### Steps for installing OpenBLAS | ||
|
|
||
| <div class="codetabs"> | ||
| <div data-lang="Linux" markdown="1"> | ||
| ```bash | ||
| 1. Install OpenBLAS via yum/apt-get | ||
| # Fedora, Centos | ||
| sudo yum install openblas | ||
| # Ubuntu | ||
| sudo apt-get install openblas | ||
|
|
||
| 2. If OpenBLAS not getting picked up, double-check if you are using 64-bit Java and if `libopenblas` is available. | ||
| ldconfig -p | grep libopenblas | ||
| # You may have to add explicit link using following command: | ||
| sudo ln -s /lib64/libopenblas.so.0 /usr/lib64/libopenblas.so | ||
| ``` | ||
| </div> | ||
| <div data-lang="Windows" markdown="1"> | ||
| ```bash | ||
| Download the pre-built binaries or install from the source. | ||
| ``` | ||
| </div> | ||
| </div> | ||
| Links: | ||
| 1. [Pre-built OpenBLAS binaries](https://sourceforge.net/projects/openblas/), | ||
| 2. [OpenBLAS source](https://github.com/xianyi/OpenBLAS). | ||
| By default, SystemML searches for Intel MKL and then OpenBLAS to select the underlying BLAS. | ||
| If both are not found, we fallback to SystemML's internal Java library. | ||
| If you want to explicitly select the underlying BLAS or disable native BLAS, please set | ||
| the environment variable `SYSTEMML_BLAS` to `mkl/openblas/none`. | ||
| ## Using GPU | ||
| SystemML requires that CUDA 8.0 and CuDNN 5.1 is installed on the machine to exploit GPU. | ||
| If these libraries are not installed, we fall back to non-GPU plan. | ||
| Like native BLAS, to exploit GPU we require that `systemml-accelerator.jar` is available. | ||
| If you want to explicitly disable using GPU, please set the environment variable `SYSTEMML_GPU` to `none` (default: `cuda`). | ||
| To test if BLAS and/or GPU is enabled, please follow the below steps: | ||
| <div class="codetabs"> | ||
| <div data-lang="PySpark" markdown="1"> | ||
| ```bash | ||
| from systemml import random | ||
| m1 = random.uniform(size=(1000,1000)) | ||
| m2 = random.uniform(size=(1000,1000)) | ||
| m3 = m1.dot(m2).toNumPy() | ||
| ``` | ||
| </div> | ||
| <div data-lang="Scala" markdown="1"> | ||
| ```bash | ||
| SYSTEMML_HOME=`python -c 'import imp; import os; print imp.find_module("systemml")[1]'` | ||
| SYSTEMML_JAR=`ls $SYSTEMML_HOME/systemml-java/systemml*incubating*.jar` | ||
| ACCELERATOR_JAR=`ls $SYSTEMML_HOME/systemml-java/systemml-accelerator.jar` | ||
| $SPARK_HOME/bin/spark-shell --jars $SYSTEMML_JAR,$ACCELERATOR_JAR | ||
| scala> import org.apache.sysml.api.mlcontext._ | ||
| scala> import org.apache.sysml.api.mlcontext.ScriptFactory._ | ||
| scala> val ml = new MLContext(sc) | ||
| scala> val script = dml("X = matrix(0.1, rows=1000, cols=1000); Y = matrix(0.2, rows=1000, cols=1000); Z = X %*% Y; print(sum(Z))") | ||
| scala> ml.execute(script) | ||
| ``` | ||
| </div> | ||
| </div> | ||
| The above script should output either . | ||
| ```bash | ||
| accelerator.BLASHelper: Found BLAS: (mkl/openblas) | ||
| or | ||
| accelerator.LibraryLoader: Unable to load (MKL/OpenBLAS) | ||
| ``` | ||
| Note: if `systemml-accelerator.jar` is not included via `--jars` (for spark-shell or spark-submit), then we fall back to SystemML's internal Java library. |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,27 @@ | ||
| #------------------------------------------------------------- | ||
| # | ||
| # Licensed to the Apache Software Foundation (ASF) under one | ||
| # or more contributor license agreements. See the NOTICE file | ||
| # distributed with this work for additional information | ||
| # regarding copyright ownership. The ASF licenses this file | ||
| # to you under the Apache License, Version 2.0 (the | ||
| # "License"); you may not use this file except in compliance | ||
| # with the License. You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, | ||
| # software distributed under the License is distributed on an | ||
| # "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY | ||
| # KIND, either express or implied. See the License for the | ||
| # specific language governing permissions and limitations | ||
| # under the License. | ||
| # | ||
| #------------------------------------------------------------- | ||
| X = matrix(0.1, rows=$1, cols=$2) | ||
| Y = matrix(0.2, rows=$2, cols=$3) | ||
| Z = matrix(0, rows=$1, cols=$3) | ||
| for(i in 1:$4) { | ||
| Z = Z + X %*% Y | ||
| } | ||
| print(as.scalar(Z[1,1])) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,67 @@ | ||
| #!/bin/bash | ||
| #------------------------------------------------------------- | ||
| # | ||
| # Licensed to the Apache Software Foundation (ASF) under one | ||
| # or more contributor license agreements. See the NOTICE file | ||
| # distributed with this work for additional information | ||
| # regarding copyright ownership. The ASF licenses this file | ||
| # to you under the Apache License, Version 2.0 (the | ||
| # "License"); you may not use this file except in compliance | ||
| # with the License. You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, | ||
| # software distributed under the License is distributed on an | ||
| # "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY | ||
| # KIND, either express or implied. See the License for the | ||
| # specific language governing permissions and limitations | ||
| # under the License. | ||
| # | ||
| #------------------------------------------------------------- | ||
|
|
||
| #------------------------------------------------------------- | ||
| export JAVA_HOME=... # 64 bit JVM | ||
| export SPARK_HOME=... | ||
|
|
||
| CONF="--master local[*] --executor-memory 5g" | ||
| invoke_systemml() { | ||
| iter=$4 | ||
| setup=$5 | ||
| echo "Testing "$setup" with "$iter" iterations and using setup ["$1", "$2"] %*% ["$2", "$3"]" | ||
| tstart=$(date +%s.%N) | ||
| echo $JAVA_OPTS $OUTPUT_SYSTEMML_STATS | ||
| $SPARK_HOME/bin/spark-submit $CONF --class org.apache.sysml.api.DMLScript $6 SystemML.jar -f matmult.dml -stats -args $1 $2 $3 $4 | ||
| ttime=$(echo "$(date +%s.%N) - $tstart" | bc) | ||
| echo $setup","$iter","$1","$2","$3","$ttime >> time.txt | ||
| } | ||
|
|
||
|
|
||
| rm time.txt | ||
| iter=1000 | ||
| export SYSTEMML_GPU=none | ||
| echo "-------------------------" | ||
| for i in 1 10 100 1000 2000 5000 10000 | ||
| do | ||
| for j in 1 10 100 1000 2000 5000 10000 | ||
| do | ||
| for k in 1 10 100 1000 2000 5000 10000 | ||
| do | ||
| # Intel MKL | ||
| export SYSTEMML_BLAS=mkl | ||
| invoke_systemml $i $j $k $iter IntelMKL "--jars ./systemml-accelerator.jar" | ||
|
|
||
| # OpenBLAS | ||
| export SYSTEMML_BLAS=openblas | ||
| invoke_systemml $i $j $k $iter OpenBLAS "--jars ./systemml-accelerator.jar" | ||
|
|
||
| # Java | ||
| invoke_systemml $i $j $k $iter Java "" | ||
|
|
||
| # GPU | ||
| export SYSTEMML_GPU=cuda | ||
| invoke_systemml $i $j $k $iter GPU "--jars ./systemml-accelerator.jar" | ||
| export SYSTEMML_GPU=none | ||
| done | ||
| done | ||
| done |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.