GPJSON is a GPU-based JSON data processing system which can be used to execute JSONPath queries on newline-delimited JSON datasets. It has support for a subset of JSONPath queries and can be used from any GraalVM language.

GPJSON uses Gradle as a build system. There is no need to install Gradle, the Gradle wrapper can be used to build GPJSON.

The only requirements for running GPJSON are GraalVM and CUDA (and a supported GPU). It has been tested on GraalVM CE 21.0.0.2 Java 8, downloadable here. It has only been tested on Linux using CUDA 11.2, but should work on other OS when CUDARuntime.java is adapted.

To build the project, execute:

./gradlew build

This will generate two files in gpjson/build/libs:

• gpjson-0.1-SNAPSHOT.jar: Contains the compiled code of GPJSON, without any dependencies. This will not run without including GPJSON's dependencies on the classpath.
• gpjson-0.1-SNAPSHOT-all.jar: Contains the compiled code of GPJSON, including all dependencies. This will run without any other JAR files required on the classpath.

To use GPJSON in GraalVM, copy the gpjson-0.1-SNAPSHOT-all.jar file to $GRAALVM_HOME/jre/languages/gpjson. This can also be done automatically by setting the graalVMDirectory project property. For example, if you have set the GRAALVM_HOME environment variable, use:

./gradlew copyToGraalVM -PgraalVMDirectory=$GRAALVM_HOME

Create a file gpjson.js:

const gpjson = Polyglot.eval('gpjson', 'jsonpath');
const result = gpjson.query('dataset.json', '$.user.lang');

Then, run:

$GRAALVM_HOME/bin/node --polyglot --jvm gpjson.js

GPJSON is licensed under the BSD 3-Clause license available here.

Parts of the source code are from grCUDA and are licensed under the BSD 3-Clause license available here.

GPJSON also depends on Truffle APIs licensed under the Universal Permissive License (UPL), Version 1.0 (https://opensource.org/licenses/UPL).