Frequently Asked Questions
1. What can TornadoVM do?
TornadoVM accelerates parts of your Java applications on heterogeneous hardware devices such as multicore CPUs, GPUs, and FPGAs.
TornadoVM is currently being used to accelerate machine learning and deep learning applications, computer vision, physics simulations, financial applications, computational photography, natural language processing and signal processing.
2. Can I use TornadoVM in a commercial application?
Absolutely yes! TornadoVM employs many licenses as shown here, but its API is under CLASSPATH EXCEPTION, and hence it can be freely used in any application.
3. How can I use it?
List of compatible JDKs
TornadoVM can be currently executed with the following configurations:
TornadoVM with GraalVM (JDK 21): see the installation guide: Installation for GraalVM for JDK 21.0.1 on Linux and OSx.
TornadoVM with JDK21 (e.g. OpenJDK 21, Red Hat Mandrel 21, Amazon Corretto 21, Azul Zulu JDK 21): see the installation guide: TornadoVM for JDK 21 on Linux and OSx.
To run TornadoVM on Windows 10/11 OS, install TornadoVM with GraalVM. More information here: TornadoVM for Windows 10/11 using GraalVM.
ARM Mali GPUs and Linux
To run TornadoVM on ARM Mali, install TornadoVM with GraalVM and JDK 11. More information here: TornadoVM on ARM Mali GPUs.
4. Which programming languages does TornadoVM support?
Here you can find examples of how to use TornadoVM with GraalVM Polyglot.
5. Is TornadoVM a Domain Specific Language (DSL)?
No, TornadoVM is not a DSL. It compiles a subset of Java code to OpenCL C, NVIDIA PTX, and SPIR-V binary.
The TornadoVM API only provides two Java annotations (
@Reduce) plus an APIs to:
a) Create and define task-graphs (groups of Java methods to be accelerated by TornadoVM), and the data needed to execute those task-graphs.
b) Define execution plans.
6. Does it support the whole Java Language?
No, TornadoVM supports a subset of the Java programming language. A list of unsupported features along with the reasoning behind it can be found here: Unsupported Java features.
7. Can TornadoVM degrade the performance of my application?
No, TornadoVM can only increase the performance of your application because it can dynamically change the execution of a program at runtime onto another device. If a particular code segment cannot be accelerated, then execution falls back to the host JVM which will execute your code on the CPU as it would normally do.
Also with the Dynamic Reconfiguration, TornadoVM discovers the fastest possible device for a particular code segment completely transparently to the user.
8. Dynamic Reconfiguration? What is this?
It is a novel feature of TornadoVM, in which the user selects a metric on which the system decides how to map a specific computation on a particular device. Further details and instructions on how to enable this feature can be found here:
Dynamic reconfiguration: https://dl.acm.org/doi/10.1145/3313808.3313819.
9. Does TornadoVM support only OpenCL devices?
No. Currently, TornadoVM supports three compiler backends and therefore, it is able to generate OpenCL, PTX, and SPIR-V code depending on the hardware configuration.
10. Why is it called a VM?
The VM name is used because TornadoVM implements its own set of bytecodes for handling heterogeneous execution. These bytecodes are used for handling JIT compilation, device exploration, data management and live task-migration for heterogeneous devices (multi-core CPUs, GPUs, and FPGAs). We sometimes refer to a VM inside a VM (nested VM). The main VM is the Java Virtual Machine, and TornadoVM sits on top of that.
You can find more information here: https://dl.acm.org/doi/10.1145/3313808.3313819.
11. How does it interact with OpenJDK?
TornadoVM makes use of the Java Virtual Machine Common Interface (JVMCI) that is included from Java 9 to compile Java bytecode to OpenCL C/PTX/SPIR-V at runtime. As a JVMCI implementation, TornadoVM uses Graal (it extends the Graal IR and includes new backends for OpenCL C, PTX and SPIR-V code generation).
12. How do I know which parts of my application are suitable for acceleration?
Workloads with for-loops that do not have dependencies between iterations are very good candidates to offload on accelerators. Examples of this pattern are NBody computation, Black-scholes, DFT, KMeans, etc.
Besides, matrix-type applications are good candidates, such as matrix-multiplication widely used in machine and deep learning.
13. How can I contribute to TornadoVM?
TornadoVM is an open-source project, and, as such, we welcome contributions from all levels.
Solve issues reported on the GitHub page.
Work on New Proposals: We welcome new proposals and ideas. To work on a new proposal, use the discussion page on GitHub. Alternatively, you can open a shared document (e.g., a shared Google doc) where we can discuss and analyse your proposal.
Here you can find more information about how to contribute, code conventions, and tasks.
14. Does TornadoVM support calls to standard Java libraries?
Partially yes. TornadoVM currently supports calls to the Math library. However, invocations that imply I/O are not supported. Note that this restriction also applies to low-level parallel programming models such as OpenCL, SYCL, oneAPI and CUDA.
15. Do I need a GPU to run TornadoVM?
No. TornadoVM can also run on multi-core CPUs and/or FPGAs. What TornadoVM needs is a compatible driver/runtime installed in the machine. For example, to enable TornadoVM getting access to an Intel CPU, developers can use the Intel OpenCL runtime (e.g., from the Intel oneAPI base Toolkit).
To enable TornadoVM accessing FPGAs, developers can use the Intel and AMD OpenCL implementations for the Intel and Xilinx FPGAs, respectively.