This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| PGI (NVIDIA) | |
|---|---|
| Name | PGI (NVIDIA) |
| Type | Subsidiary |
| Industry | Software |
| Founded | 1980s |
| Fate | Acquired |
| Headquarters | California |
| Products | Compilers, tools |
| Parent | NVIDIA |
PGI (NVIDIA) is a compiler and tools brand within NVIDIA Corporation focused on high-performance compilers for heterogeneous computing. Originating from the Portland Group, the effort provided Fortran, C, and C++ compilers aimed at accelerating code on multi-core CPUs and GPU accelerators. The technology intersects with software ecosystems developed by vendors and institutions across scientific computing, supercomputing, and enterprise domains.
PGI traces its origins to the Portland Group, a company established to deliver optimized compilers for high-performance computing customers including laboratories and universities. The Portland Group engaged with organizations such as Lawrence Livermore National Laboratory, Los Alamos National Laboratory, Oak Ridge National Laboratory, Argonne National Laboratory, and collaborations with vendors like Cray Research, IBM, Intel Corporation, Hewlett-Packard, and SGI. Over time PGI's technology adapted to changes in parallel computing exemplified by projects such as MPI, OpenMP, CUDA, and accelerator deployments at centers like National Center for Supercomputing Applications and NERSC. In the 2010s PGI was acquired by a major GPU vendor, integrating into a corporate strategy alongside products from NVIDIA, linking to initiatives such as CUDA Toolkit, Tesla (GPU), NVIDIA DGX, and collaborations with research projects at institutions like MIT, Stanford University, University of California, Berkeley, and Lawrence Berkeley National Laboratory.
PGI provided a suite of compilers and tools historically branded under names like PGI Compilers and PGI Accelerator. Key offerings included Fortran, C, and C++ compilers supporting host and device code generation, performance libraries, and debugging aids. Related technologies and tools interfaced with ecosystems from GNU Project toolchains, LLVM Project, Intel Math Kernel Library, OpenBLAS, cuBLAS, MKL integrations, and debugging/profiling tools such as NVIDIA Nsight, gdb, Valgrind, Allinea Forge, and TotalView. The product line also connected with build and package systems used by projects like CMake, Make (software), Autotools, and high-performance application frameworks used by groups at NASA, European Organization for Nuclear Research, CERN, and Max Planck Society.
PGI compilers targeted languages and standards relevant to HPC workflows including Fortran 77, Fortran 90, Fortran 95, standards used by communities at institutions such as Argonne National Laboratory and DOE labs, as well as C99, C11, and C++03/C++11 families. Support extended to parallel programming models and standards like OpenMP, OpenACC, and NVIDIA’s proprietary CUDA. PGI also supported interoperability with standards-driven libraries and interfaces adopted by consortia such as ISO/IEC committees and research centers including European Centre for Medium-Range Weather Forecasts and Met Office.
PGI compilers implemented optimization passes designed to target both multicore CPUs and GPU accelerators, leveraging code generation strategies analogous to those in LLVM and vendor compilers from Intel Corporation and Cray Research. Techniques included loop transformations, vectorization for instruction sets like AVX, SSE, and NEON, aggressive inlining, interprocedural optimization, and auto-parallelization tuned for systems like Blue Gene and clusters using InfiniBand interconnects. For GPU offload, PGI supported directive-driven approaches enabling mappings from compute kernels to devices, akin to approaches in projects such as OpenACC and compiler efforts from PGI’s contemporaries. Performance engineering workflows employed profiling and tracing tools from NVIDIA Nsight Systems, TAU Performance System, HPCToolkit, and monitoring stacks used at supercomputing centers like Oak Ridge National Laboratory.
Under NVIDIA stewardship, PGI technologies were integrated with the broader software stack including CUDA Toolkit, NVIDIA TensorRT, cuDNN, and systems such as NVIDIA DGX Station and NVIDIA DGX A100. Integration efforts aimed to facilitate execution on accelerator families like NVIDIA Ampere architecture, NVIDIA Tesla, and subsequent GPU lines, while interoperating with orchestration platforms from Kubernetes, container ecosystems like Docker, and cloud services from providers such as Amazon Web Services, Microsoft Azure, and Google Cloud Platform used by research groups at Stanford University and University of Cambridge.
Historically PGI compilers were offered under commercial licensing models with options for academic and enterprise support, aligning with procurement practices at national labs and universities including Lawrence Livermore National Laboratory and University of Illinois Urbana-Champaign. Post-acquisition licensing and redistribution were harmonized with NVIDIA’s policies, impacting deployment on supercomputers such as Summit and Perlmutter and affecting CI/CD workflows used by projects hosted on platforms like GitHub and GitLab. Some components and pathways adopted open-source collaborations seen in efforts by the LLVM Project and community-driven toolchains.
PGI’s compilers and tools were widely cited in scientific publications and technical reports from institutions such as Argonne National Laboratory, Oak Ridge National Laboratory, Lawrence Livermore National Laboratory, European Centre for Medium-Range Weather Forecasts, and universities including MIT, Princeton University, and University of California, Berkeley. The technology influenced adoption of directive-based GPU programming and contributed to performance-portability discussions alongside initiatives like Kokkos, RAJA, OpenACC, and OpenMP. Industry reception compared PGI offerings to compilers from Intel Corporation, GNU Project, and Cray Research, with practitioners reporting gains in productivity and acceleration on heterogeneous platforms deployed across national laboratories, research institutes, and commercial R&D centers.
Category:Compilers