Search for: All records

Recent trends in high-performance computing show an increase in the adoption of performance portable frameworks such as Kokkos and interpreted languages such as Python. PyKokkos follows these trends and enables programmers to write performance-portable kernels in Python which greatly increases productivity. One issue that programmers still face is how to organize parallel code, as splitting code into separate kernels simplifies testing and debugging but may result in suboptimal performance. To enable programmers to organize kernels in any way they prefer while ensuring good performance, we present PyFuser, a program analysis framework for automatic fusion of performance portable PyKokkos kernels. PyFuser dynamically traces kernel calls and lazily fuses them once the result is requested by the application. PyFuser generates fused kernels that execute faster due to better reuse of data, improved compiler optimizations, and reduced kernel launch overhead, while not requiring any changes to existing PyKokkos code. We also introduce automated code transformations that further optimize the fused kernels generated by PyFuser. Our experiments show that on average PyFuser achieves speedups compared to unfused kernels of 3.8x on NVIDIA and AMD GPUs, as well as Intel and AMD CPUs. 

We present a design and implementation of an in-memory object graph store, dubbed εStore. Our key innovation is a storage model - epsilon store - that equates an object on the heap to a node in a graph store. Thus any object on the heap (without changes) can be a part of one, or multiple, graph stores, and vice versa, any node in a graph store can be accessed like any other object on the heap. Specifically, each node in a graph is an object (i.e., instance of a class), and its properties and its edges are the primitive and reference fields declared in its class, respectively. Necessary classes, which are instantiated to represent nodes, are created dynamically by εStore. εStore uses a subset of the Cypher query language to query the graph store. By design, the result of any query is a table (ResultSet) of references to objects on the heap, which users can manipulate the same way as any other object on the heap in their programs. Moreover, a developer can include (transitively) an arbitrary object to become a part of a graph store. Finally, εStore introduces compile-time rewriting of Cypher queries into imperative code to improve the runtime performance. εStore can be used for a number of tasks including implementing methods for complex in-memory structures, writing complex assertions, or a stripped down version of a graph database that can conveniently be used during testing. We implement εStore in Java and show its application using the aforementioned tasks. 

We present LeJit, a template-based framework for testing Java just-in-time (JIT) compilers. Like recent template-based frameworks, LeJit executes a template---a program with holes to be filled---to generate concrete programs given as inputs to Java JIT compilers. LeJit automatically generates template programs from existing Java code by converting expressions to holes, as well as generating necessary glue code (i.e., code that generates instances of non-primitive types) to make generated templates executable. We have successfully used LeJit to test a range of popular Java JIT compilers, revealing five bugs in HotSpot, nine bugs in OpenJ9, and one bug in GraalVM. All of these bugs have been confirmed by Oracle and IBM developers, and 11 of these bugs were previously unknown, including two CVEs (Common Vulnerabilities and Exposures). Our comparison with several existing approaches shows that LeJit is complementary to them and is a powerful technique for ensuring Java JIT compiler correctness.