Project Summary
BLUESTONE: PROGRAM TRANSLATION AND SYNTHESIS FOR EXTREMELY
HETEROGENEOUS ARCHITECTURES
Jeffrey S. Vetter (Principal Investigator); Oak Ridge National Laboratory Franz Franchetti 
(Co-Investigator); Carnegie Mellon University Michael D. Franusich (Co-Investigator); SpiralGen 
Inc.

Goal.    The overarching goal of this Bluestone project is to enable new levels of performance 
portability of high-performance computing (HPC) and artificial intelligence (AI) applications on 
extremely heterogeneous (EH) architectures.  Given the current trends toward architectural 
diversification, any software developers that strive for performance portability of their software 
will need improved techniques, including program synthesis, that hide the increasing complexity of 
emerging EH architectures.  There are two key challenges for performance portability: (1) improved 
code generation and synthesis for EH devices on single node and across multiple EH architectures, 
and (2) effective run time scheduling of computation while accounting for device heterogeneity, 
locality, and data orchestration.
Approach.    Our project,  named Bluestone,  will take the approach of improving both the code 
genera- tion and run time systems by employing program synthesis techniques, including AI, 
constraint solving, SMT/SAT, and symbolic methods. To achieve these goals, our Bluestone project 
will integrate and build on several established components:  (a) the LLVM/MLIR compiler ecosystem; 
(b) the SnowWhite high-level reasoning engine; and (c) the IRIS heterogeneous run time system. We 
will initially target HPC and AI ap- plications on three relevant platforms of interest to DOE: 
Summit (IBM and NVIDIA), Frontier (AMD), and Aurora (Intel). Concurrently, we will focus on 
emerging architectures including FPGAs, systems on a chip (SoCs) (e.g., Qualcomm Snapdragon), and 
quantum devices via QASM. We will initially target popular lan- guages and programming models in 
HPC, such as a C, C++, SYCL, OpenCL, and Python (NumPy/SciPy), then move to support other languages 
and frameworks—such as Flang/Fortran, PyTorch, TensorFlow, and Julia—as we gain experience with 
them. We will initially target proxy applications, moving on to respective, important DOE 
applications later in the project.
Research objectives.    Our objectives are to investigate these goals with the following tasks: (a) 
integrate the LLVM/MLIR compiler with SnowWhite with to facilitate advanced optimization of compute 
kernels offloaded to heterogeneous devices like GPUs; (b) integrate SnowWhite with the IRIS run 
time system to allow advanced dynamic scheduling and efficient data movement; (c) develop support 
for creating analyti- cal performance models that can, in turn, be ingested by SnowWhite to speed 
up its optimization process;
(d) investigate approaches to Machine Learning (ML)-enabled transcoders for heterogeneous 
programming languages like OpenCL; (e) add support for quantum computing via QASM across Bluestone 
components to allow transparent access to QASM supported devices; and, (f) actively evaluate our 
ideas on proxy appli- cations, math kernels, and, later, DOE applications.
Team.    Our team comprises experts in the areas of programming systems and program synthesis from 
Oak Ridge National Laboratory (ORNL), Carnegie Mellon University, and SpiralGen Inc.  We actively 
develop and have substantial experience with relevant software stacks, emerging test beds, HPC 
architectures, and a variety of HPC and AI applications.  We work closely with US Department of 
Energy applications teams, among others, to use their applications as drivers and optimize their 
applications for these new architectures. We believe that these close collaborations will be 
necessary in the early stages of the emergence of these new architectures.   For the emerging 
architectures,  we have access to ORNL’s Experimental Computing Laboratory (https://excl.ornl.gov), 
which has several field-programmable gate arrays, systems on a chip, and other heterogeneous 
compute engines.