Skip to main content

Resultados relacionados

Resultados totales (Incluyendo duplicados): 32
Encontrada(s) 1 página(s)

UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

OpenMP taskloop dependences

  • Maroñas Bravo, Marcos
  • Teruel García, Xavier
  • Beltran, Vincenç
Exascale systems will contain multicore/manycore processors with high core count in each node. Therefore, using a model that relaxes the synchronization, such as data-flow, is crucial to adequately exploit the potential of the hardware. The flexibility of the data-flow execution model relies on the dynamic management of data-dependences among tasks. The OpenMP standard already provides a construct, known as taskloop, that distributes the loop iteration space into several tasks, but this construct does not support the use of the depend clause yet. In this paper we propose the use of the induction variable to define data dependences in tasks created by the taskloop construct. By using the induction variable, each task will contain its own dependences based on the partition of work they received. We also aim to demonstrate that using taskloop with dependences provides an enhancement in terms of programmability with respect to using stand-alone tasks to parallelize a loop. Our implementation does not introduce any significant overhead on the taskloop implementation and, in certain cases, it outperforms the stand-alone task version., This research has received funding from the European Union’s Horizon 2020/EuroHPC research and innovation programme under grant agreement No 955606 (DEEP-SEA); and the support of the Spanish Ministry of Science and Innovation (Computacion de Altas Prestaciones VIII: PID2019-107255GB)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

Communication-aware sparse patterns for the factorized approximate inverse preconditioner

  • Laut Turón, Sergi
  • Casas, Marc|||0000-0003-4564-2093
  • Borrell Pol, Ricard
The Conjugate Gradient (CG) method is an iterative solver targeting linear systems of equations Ax=b where A is a symmetric and positive definite matrix. CG convergence properties improve when preconditioning is applied to reduce the condition number of matrix A. While many different options can be found in the literature, the Factorized Sparse Approximate Inverse (FSAI) preconditioner constitutes a highly parallel option based on approximating A-1. This paper proposes the Communication-aware Factorized Sparse Approximate Inverse preconditioner (FSAIE-Comm), a method to generate extensions of the FSAI sparse pattern that are not only cache friendly, but also avoid increasing communication costs in distributed memory systems. We also propose a filtering strategy to reduce inter-process imbalance. We evaluate FSAIE-Comm on a heterogeneous set of 39 matrices achieving an average solution time decrease of 17.98%, 26.44% and 16.74% on three different architectures, respectively, Intel Skylake, Fujitsu A64FX and AMD Zen 2 with respect to FSAI. In addition, we consider a set of 8 large matrices running on up to 32,768 CPU cores, and we achieve an average solution time decrease of 12.59%., Marc Casas is supported by Grant RYC-2017-23269 funded by MCIN/AEI/ 10.13039/501100011033 and by “ESF Investing in your future”. This work has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 955606. This work has been supported by the Computación de Altas Prestaciones VIII (BSC-HPC8) project. It has also been partially supported by the EXCELLERAT project funded by the European Commission’s ICT activity of the H2020 Programme under grant agreement number: 823691 and by the Spanish Ministry of Science and Innovation (Nucleate, Project PID2020-117001GB-I00)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

FASE: A fast, accurate and seamless emulator for custom numerical formats

  • Osorio Ríos, John Haiber
  • Armejach Sanosa, Adrià|||0000-0003-2869-668X
  • Petit, Eric
  • Henry, Greg
  • Casas, Marc|||0000-0003-4564-2093
Deep Neural Networks (DNNs) have become ubiquitous in a wide range of application domains. Despite their success, training DNNs is an expensive task that has motivated the use of reduced numerical precision formats to improve performance and reduce power consumption. Emulation techniques are a good fit to understand the properties of new numerical formats on a particular workload. However, current SoA techniques are not able to perform these tasks quickly and accurately on a wide variety of workloads.We propose FASE, a Fast, Accurate, and Seamless Emulator that leverages dynamic binary translation to enable emulation of custom numerical formats. FASE is fast: allowing emulation of large unmodified workloads; accurate: emulating at the instruction operand level; and seamless: as it does not require any code modifications and works on any application or DNN framework without any language, compiler, or source code access restrictions., Marc Casas has been partially supported by the Grant RYC2017-23269 funded by MCIN/AEI/ 10.13039/501100011033 and by “ESF Investing in your future”. Adria Armejach is a Serra Hunter Fellow and has been partially supported by the Grant IJCI-2017-33945 funded by MCIN/AEI/ 10.13039/501100011033. John Osorio has been partially supported by the Grant PRE2019- 090406 funded by MCIN/AEI/ 10.13039/501100011033 and by “ESF Investing in your future”. This work has been partially supported by Intel under the BSC-Intel collaboration and European Union’s Horizon 2020 research and innovation programme under grant agreement No 955606 - DEEP-SEA EU project., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

A generator of numerically-tailored and high-throughput accelerators for batched GEMMs

  • Ledoux Pardo, Luis Eduardo
  • Casas, Marc|||0000-0003-4564-2093
We propose a hardware generator of GEMM accelerators. Our generator produces vendor-agnostic HDL describing highly customizable systolic arrays guided by accuracy and energy efficiency goals. The generated arrays have three main novel aspects. First, the accelerators handle a large variety of computer number formats using intermediate representations based on our Sign Scale Significand (S3) format. Second, the processing elements perform all intermediate dot-product arithmetic operations required by the GEMM kernel without any intermediate rounding, which makes it possible to deliver better energy efficiency than state-of-the-art approaches while offering more accuracy and reproducible results. Third, our accelerators feature the Half-Speed Sink Down (HSSD) mechanism, which maximizes the overlap of host-accelerator data transfers with GEMM computations.We evaluate our automatically generated designs in a cutting-edge setup composed of a POWER9 host, CAPI (Coherent Accelerator Processor Interface) link, and a Virtex Ultrascale Plus FPGA. Arrays can operate at the speed of the link and saturate it to reach a 13GB/s throughput. Our fine-grain customization approach allows to cover a wide range of accuracy versus efficiency scenarios and can reach 0.65GOps/s/W while producing 1024 accurate bits or 148.7GOps/s/W with 6 accurate bits. Our configurations achieve up to 1613GOps/s system performance and power efficiencies of up to 240GOps/s/W for the FPGA. This automatic generator is the first being able to produce such a variety of designs. We improve the single-precision energy efficiency of state-of-the-art FPGA GEMM accelerators by 1.86×., This work has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 955606 Marc Casas is supported by Grant RYC-2017-23269 funded by MCIN/AEI/ 10.13039/501100011033 and by “ESF Investing in your future”, Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Artículo científico (JournalArticle)

O(n) key–value sort with active compute memory

  • Esmaili Dokht, Pouya
  • Guiot Cusido, Miquel
  • Radojkovic, Petar
  • Martorell Bofill, Xavier|||0000-0002-0417-3430
  • Ayguadé Parra, Eduard|||0000-0002-5146-103X
  • Labarta Mancho, Jesús José|||0000-0002-7489-4727
  • Adlard, Jason
  • Amato, Paolo
  • Sforzin, Marco
We propose the Active Compute Memory (ACM), a near-memory-processing architecture capable of performing key–value sort directly in the DRAM. In the ACM architecture, sort is merely the writing of data into memory with one addressing protocol (perspective) and reading it back with different perspective. The first perspective is conventional, based on the data address; the second perspective is the sorted order. The ACM requires additional tables to store the meta-data and moderate control logic enhancements that can be implemented directly in the DRAM silicon. By these modest enhancements to DRAM, ACM exploits the parallelism inherently available in the row buffer to enable sort with O ( n ) complexity. This leads to an order of magnitude improvement in ACM performance and energy compared to conventional O ( n log n ) CPU-centric sort algorithms. The ACM also shows superior performance compared to other near-memory sort accelerators. This is because the ACM processing is done near the row buffer and it exploits much lower memory access latency, higher bandwidth and wider parallel processing. The sort operation covered in this paper is just an example of an address management operation that can be efficiently implemented directly in the DRAM silicon. We release as an open source the simulation infrastructure for the ACM performance and energy modeling. We would encourage the community to use it, adapt it to other PIM proposals, and share their own evaluations., This work was funded by the Collaboration Agreement between Micron Technology, Inc. and BSC. The work was also supported by the Spanish Government, under the contracts PID2019-107255GB-C21 and CEX2021-001148-S funded by MCIN/AEI/ 10.13039/501100011033. The work also received funding from the Department of Research and Universities of the Government of Catalonia to the AccMem Research Group (Code: 2021 SGR 00807) and project DEEP-SEA funded by European Union’s Horizon 2020 research and innovation programme , under Grant Agreement no.955606 and the Spanish Government, under the contract PCI2021-121958 funded by MCIN/AEI/ 10.13039/501100011033 and “European Union NextGenerationEU/PRTR”., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Artículo científico (JournalArticle)

Optimizing iterative data-flow scientific applications using directed cyclic graphs

  • Álvarez Robert, David
  • Beltran Querol, Vicenç
Data-flow programming models have become a popular choice for writing parallel applications as an alternative to traditional work-sharing parallelism. They are better suited to write applications with irregular parallelism that can present load imbalance. However, these programming models suffer from overheads related to task creation, scheduling and dependency management, limiting performance and scalability when tasks become too small. At the same time, many HPC applications implement iterative methods or multi-step simulations that create the same directed acyclic graphs of tasks on each iteration. By giving application programmers a way to express that a specific loop is creating the same task pattern on each iteration, we can create a single task directed acyclic graph (DAG) once and transform it into a cyclic graph. This cyclic graph is then reused for successive iterations, minimizing task creation and dependency management overhead. This paper presents the taskiter, a new construct we propose for the OmpSs-2 and OpenMP programming models, allowing the use of directed cyclic task graphs (DCTG) to minimize runtime overheads. Moreover, we present a simple immediate successor locality-aware heuristic that minimizes task scheduling overhead by bypassing the runtime task scheduler. We evaluate the implementation of the taskiter and the immediate successor heuristic in 8 iterative benchmarks. Using small task granularities, we obtain a geometric mean speedup of 2.56x over the reference OmpSs-2 implementation, and a 3.77x and 5.2x speedup over the LLVM and GCC OpenMP runtimes, respectively., This work was supported in part by the European Union’s Horizon 2020/EuroHPC Research and Innovation Programme (DEEP-SEA) under Grant 955606; in part by the Spanish State Research Agency—Ministry of Science and Innovation, Generalitat de Catalunya, under Project PCI2021121958 and Project 2021-SGR-01007; in part by the Spanish Ministry of Science and Technology under Contract PID2019-107255GB; and in part by Severo Ochoa under Grant CEX2021-001148-S/MCIN/AEI/10.13039/501100011033., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Artículo científico (JournalArticle)

A BF16 FMA is all you need for DNN training

  • Osorio Ríos, John Haiber
  • Armejach Sanosa, Adrià|||0000-0003-2869-668X
  • Petit, Eric
  • Henry, Greg
  • Casas, Marc|||0000-0003-4564-2093
Fused Multiply-Add (FMA) functional units constitute a fundamental hardware component to train Deep Neural Networks (DNNs). Its silicon area grows quadratically with the mantissa bit count of the computer number format, which has motivated the adoption of the BrainFloat16 format (BF16). BF16 features 1 sign, 8 exponent and 7 explicit mantissa bits. Some approaches to train DNNs achieve significant performance benefits by using the BF16 format. However, these approaches must combine BF16 with the standard IEEE 754 Floating-Point 32-bit (FP32) format to achieve state-of-the-art training accuracy, which limits the impact of adopting BF16. This article proposes the first approach able to train complex DNNs entirely using the BF16 format. We propose a new class of FMA operators, FMAbf16 n m, that entirely rely on BF16 FMA hardware instructions and deliver the same accuracy as FP32. FMAbf16 n m operators achieve performance improvements within the 1.28- 1.35X range on ResNet101 with respect to FP32. FMAbf16 n m enables training complex DNNs on simple low-end hardware devices without requiring expensive FP32 FMA functional units., Marc Casas was partially supported by the under Grant RYC-2017-23269 funded by MCIN/AEI/10.13039/501100011033, and by ESF Investing in your future. Adrià Armejach is a Serra Hunter Fellow and has been partially supported by the under Grant IJCI-2017-33945 funded by MCIN/AEI/10.13039/501100011033. John Osorio has been partially supported by the under Grant PRE2019-090406 funded by MCIN/AEI/10.13039/501100011033 and by ESF Investing in your future. This work has been partially supported by Intel under the BSC-Intel collaboration and European Union Horizon 2020 research and innovation programme under Grant 955606 - DEEP-SEA EU project., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

Transparent load balancing of MPI programs using OmpSs-2@Cluster and DLB

  • Aguilar Mena, Jimmy|||0000-0001-6802-2247
  • Ali, Omar Shaaban Ibrahim
  • López Herrero, Víctor
  • Garcia Casulla, Marta
  • Carpenter, Paul Matthew
  • Ayguadé Parra, Eduard|||0000-0002-5146-103X
  • Labarta Mancho, Jesús José|||0000-0002-7489-4727
Load imbalance is a long-standing source of inefficiency in high performance computing. The situation has only got worse as applications and systems increase in complexity, e.g., adaptive mesh refinement, DVFS, memory hierarchies, power and thermal management, and manufacturing processes. Load balancing is often implemented in the application, but it obscures application logic and may need extensive code refactoring. This paper presents an automated and transparent dynamic load balancing approach for MPI applications with OmpSs-2 tasks, which relieves applications from this burden. Only local and trivial changes are required to the application. Our approach exploits the ability of OmpSs-2@Cluster to offload tasks for execution on other nodes, and it reallocates compute resources among ranks using the Dynamic Load Balancing~(DLB) library. It employs LeWI to react to fine-grained load imbalances and DROM to address coarse-grained load imbalances by reserving cores on other nodes that can be reclaimed on demand. We use an expander graph to limit the amount of point-to-point communication and state. The results show 46% reduction in time-to-solution for micro-scale solid mechanics on 32 nodes and a 20% reduction beyond DLB for $n$-body on 16 nodes, when one node is running slow. A synthetic benchmark shows that performance is within 10% of optimal for an imbalance of up to 2.0 on 8 nodes. All software is released open source., This research has received funding from the European Union’s Horizon 2020/EuroHPC research and innovation programme under grant agreement No 955606 (DEEP-SEA) and 754337 (EuroEXA). It is supported by the Spanish State Research Agency - Ministry of Science and Innovation (contract PID2019-107255GB and Ramon y Cajal fellowship RYC2018-025628-I) and by the Generalitat de Catalunya (2017-SGR-1414)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

Efficient data redistribution for malleable applications

  • Martínez Alvarez, Iker
  • Aliaga, José I
  • Castrillo, Maribel
  • Iserte, Sergio
Process malleability can be defined as the ability of a distributed MPI parallel job to change the number of processes on–the–fly without stopping its execution, reallocating the compute resources originally assigned to the job, and without storing application data to disk. MPI malleability consists of four stages: resource reallocation, process management, data redistribution and execution resuming. Among them, data redistribution is the most time-consuming and determines the reconfiguration time. In this paper, we compare different implementations of this stage using point-to-point and collective MPI operations, and discuss the impact of overlapping computation-communication. We then combine these strategies with different methods to expand/shrink jobs, using a synthetic application to emulate MPI-based codes and their malleable counterparts, in order to evaluate the effect of different malleability methods in parallel distributed applications. The results show that the use of asynchronous techniques speeds up execution by 1.14 and 1.21, depending on the network used., This work has been funded by project PID2020-113656RB-C21 supported by MCIN/AEI/10.13039/501100011033. Researcher I. MartínÁlvarez was supported by the predoctoral fellowship ACIF/2021/260 from Valencian Region Government and European Social Funds. The European DEEP-SEA project has received funding from the European Commission’s EuroHPC-JU program under grant agreement No 955606, the JU receives support from the European Union’s Horizon 2020 research and innovation programme and by the Spanish Ministry under Grant PCI2021-121958 funded by MCIN / AEI /10.13039 / 501100011033 and by the “European Union NextGenerationEU/PRTR”., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Artículo científico (JournalArticle)

Proteo: a framework for the generation and evaluation of malleable MPI applications

  • Martín Álvarez, Iker
  • Aliaga, José I.
  • Castillo, Maribel
  • Iserte, Sergio
Applying malleability to HPC systems can increase their productivity without degrading or even improving the performance of running applications. This paper presents Proteo, a configurable framework that allows to design benchmarks to study the effect of malleability on a system, and also incorporates malleability into a real application. Proteo consists of two modules: SAM allows to emulate the computational behavior of iterative scientific MPI applications, and MaM is able to reconfigure a job during execution, adjusting the number of processes, redistributing data, and resuming execution. An in-depth study of all the possibilities shows that Proteo is able to behave like a real malleable or non-malleable application in the range [0.85, 1.15]. Furthermore, the different methods defined in MaM for process management and data redistribution are analyzed, concluding that asynchronous malleability, where reconfiguration and application execution overlap, results in a 1.15 speedup., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. This work has been funded by project PID2020-113656RB-C21 supported by MCIN/AEI/10.13039/501100011033. Researcher I. Martín-Álvarez was supported by the predoctoral fellowship ACIF/2021/260 from Valencian Region Government and European Social Funds. The European DEEP-SEA project has received funding from the European Commission’s EuroHPC-JU program under grant agreement No. 955606; the JU receives support from the European Union’s Horizon 2020 research and innovation programme and by the Spanish Ministry under Grant PCI2021-121958 funded by MCIN / AEI / 10.13039 / 501100011033 and by the “European Union NextGenerationEU/PRTR.”, Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

Combining dynamic concurrency throttling with voltage and frequency scaling on task-based programming models

  • Navarro Muñoz, Antoni
  • Lorenzon, Arthur F.
  • Ayguadé Parra, Eduard|||0000-0002-5146-103X
  • Beltran Querol, Vicenç
Being on the verge of exascale performance has shifted the prioritization of performance in applications to the inclusion of power-performance efficiency as a primary objective in the High Performance Computing (HPC) community. Simultaneously, this has surfaced hardware and software efforts that employ techniques such as dynamic voltage and frequency scaling (DVFS) for core and uncore units or dynamic concurrency throttling (DCT) to exploit hardware resources efficiently, by saving energy while maintaining performance. These techniques are complementary, so they can be used together. However, employing them is not a straightforward task, as they have to be adjusted based on the workload, and it is even more complex to combine them properly. Thus, these techniques should be applied transparently by a runtime system, without relying on application developers. In this paper, we extend a task-based runtime system with an infrastructure that categorizes workloads based on their computational profile – memory-bounded, compute-bounded, or balanced. This categorization is done in an on-line manner and with a negligible overhead. With this additional information, we enhance the CPU-manager and scheduler of OmpSs-2, a task-based parallel programming model, to automatically combine DVFS and DCT techniques based on workloads. Moreover, we show that our heuristics transparently improve energy efficiency on average by 15% with no significant performance loss and either equal or surpass the energy efficiency of the best static configuration available., This research has received funding from the European Union’s Horizon 2020/EuroHPC research and innovation programme under grant agreement N.955606 (DEEP-SEA), and is supported by the Spanish State Research Agency - Ministry of Science and Innovation (contract PID2019-107255GB), and by the Generalitat de Catalunya (2017-SGR-1414). This work was also supported by Project HPC- EUROPA3, with the support of the EC Research Innovation Action under the H2020 Programme., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

An OpenMP free agent threads implementation

  • López, Victor
  • Criado Ledesma, Joel
  • Peñacoba Veigas, Raúl
  • Ferrer Ibañez, Roger
  • Teruel García, Xavier
  • Garcia Gasulla, Marta|||0000-0003-3682-9905
In this paper, we introduce a design and implementation of the free agent threads for OpenMP. These threads increase the malleability of the OpenMP programming model, offering resource managers and runtime systems flexibility to manage threads and resources efficiently. We demonstrate how free agent threads can address load imbalances problems at the OpenMP level and at an MPI level or higher. We use two mini-apps extracted from two real HPC applications and representative of real-world codes to demonstrate this. We conclude that more malleability in thread management is necessary, and free agents can be regarded as a practical starting point to increase malleability in thread management., This work has been done as part of the European Processor Initiative project. The European Processor Initiative (EPI) (FPA: 800928) has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement EPI-SGA1: 826647. It has also received funding from the European Union's Horizon 2020/EuroHPC research and innovation programme under grant agreement No 955606 (DEEP-SEA); and the support of the Spanish Ministry of Science and Innovation (Computacion de Altas Prestaciones VIII: PID2019-107255GB)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Artículo científico (JournalArticle)

Automated generation of high-performance computational fluid dynamics codes

  • Macià Sorrosal, Sandra
  • Martínez Ferrer, Pedro J.
  • Ayguadé Parra, Eduard|||0000-0002-5146-103X
  • Beltran Querol, Vicenç
Domain-Specific Languages (DSLs) improve programmers productivity by decoupling problem descriptions from algorithmic implementations. However, DSLs for High-Performance Computing (HPC) have two additional critical requirements: performance and scalability. This paper presents the automated process of generating, from abstract mathematical specifications of Computational Fluid Dynamics (CFD) problems, optimised parallel codes that perform and scale as manually optimised ones. We consciously combine within Saiph, a DSL for solving CFD problems, low-level optimisations and parallelisation strategies, enabling high-performance single-core executions which effectively scale to multi-core and distributed environments. Our results demonstrate how high-level DSLs can offer competitive performance by transparently leveraging state-of-the-art HPC techniques., This research has received funding from the European Union’s Horizon 2020/EuroHPC research and innovation programme under grant agreement N. 955606 (DEEP-SEA), and is supported by the Spanish State Research Agency - Ministry of Science and Innovation (contract PID2019-107255 GB), and by the Generalitat de Catalunya (2017-SGR1414). This work is also supported by the Ministry of Economy of Spain through Severo Ochoa Center of Excellence Program (SEV-2015-0493)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

Page size aware cache prefetching

  • Vavouliotis, Georgios
  • Chacon, Gino
  • Álvarez Martí, Lluc|||0000-0003-0506-8867
  • Gratz, Paul V.
  • Jiménez, Daniel A.
  • Casas, Marc|||0000-0003-4564-2093
The increase in working set sizes of contemporary applications outpaces the growth in cache sizes, resulting in frequent main memory accesses that deteriorate system per- formance due to the disparity between processor and memory speeds. Prefetching data blocks into the cache hierarchy ahead of demand accesses has proven successful at attenuating this bottleneck. However, spatial cache prefetchers operating in the physical address space leave significant performance on the table by limiting their pattern detection within 4KB physical page boundaries when modern systems use page sizes larger than 4KB to mitigate the address translation overheads. This paper exploits the high usage of large pages in modern systems to increase the effectiveness of spatial cache prefetch- ing. We design and propose the Page-size Propagation Module (PPM), a µarchitectural scheme that propagates the page size information to the lower-level cache prefetchers, enabling safe prefetching beyond 4KB physical page boundaries when the accessed blocks reside in large pages, at the cost of augmenting the first-level caches’ Miss Status Holding Register (MSHR) entries with one additional bit. PPM is compatible with any cache prefetcher without implying design modifications. We capitalize on PPM’s benefits by designing a module that consists of two page size aware prefetchers that inherently use different page sizes to drive prefetching. The composite module uses adaptive logic to dynamically enable the most appropriate page size aware prefetcher. Finally, we show that the proposed designs are transparent to which cache prefetcher is used. We apply the proposed page size exploitation techniques to four state-of-the-art spatial cache prefetchers. Our evalua- tion shows that our proposals improve single-core geomean performance by up to 8.1% (2.1% at minimum) over the original implementation of the considered prefetchers, across 80 memory-intensive workloads. In multi-core contexts, we report geomean speedups up to 7.7% across different cache prefetchers and core configurations., This work is supported by the Spanish Ministry of Science and Technology through the PID2019-107255GB project, the Generalitat de Catalunya (contract 2017-SGR-1414), the European Union Horizon 2020 research and innovation program under grant agreement No 955606 (DEEP-SEA EU project), the National Science Foundation through grants CNS-1938064 and CCF-1912617, and the Semiconductor Research Corporation project GRC 2936.001. Georgios Vavouliotis has been supported by the Spanish Ministry of Economy, Industry, and Competitiveness and the European Social Fund under the FPI fellowship No. PRE2018-087046. Marc Casas has been partially supported by the Grant RYC2017-23269 funded by MCIN/AEI/10.13039/501100011033 and ESF ‘Investing in your future’., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Contribución de congreso (ConferenceOutput)

ecoHMEM: Improving object placement methodology for hybrid memory systems in HPC

  • Jordà Peroliu, Marc
  • Rai, Siddharth
  • Ayguadé Parra, Eduard|||0000-0002-5146-103X
  • Labarta Mancho, Jesús José|||0000-0002-7489-4727
  • Peña Monferrer, Antonio José
Recent byte-addressable persistent memory (PMEM) technology offers capacities comparable to storage devices and access times much closer to DRAMs than other non-volatile memory technology. To palliate the large gap with DRAM performance, DRAM and PMEM are usually combined. Users have the choice to either manage the placement to different memory spaces by software or leverage the DRAM as a cache for the virtual address space of the PMEM. We present novel methodology for automatic object-level placement, including efficient runtime object matching and bandwidth-aware placement. Our experiments leveraging Intel® Optane™ Persistent Memory show from matching to greatly improved performance with respect to state-of-the-art software and hardware solutions, attaining over 2x runtime improvement in miniapplications and over 6% in OpenFOAM, a complex production application., This paper received funding from the Intel-BSC Exascale Laboratory SoW 5.1, the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No. 749516, the EPEEC project from the European Union’s Horizon 2020 research and innovation program under grant agreement No 801051, the DEEP-SEA project from the European Commission’s EuroHPC program under grant agreement 955606, and the Ministerio de Ciencia e Innovacion—Agencia Estatal de Investigación (PID2019-107255GB-C21/AEI/10.13039/501100011033)., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publication Artículo científico (JournalArticle)

Role-shifting threads: Increasing OpenMP malleability to address load imbalance at MPI and OpenMP

  • Criado Ledesma, Joel
  • López Herrero, Víctor
  • Vinyals Ylla Català, Joan
  • Ramirez Miranda, Guillem
  • Teruel García, Xavier
  • Garcia Gasulla, Marta|||0000-0003-3682-9905
This paper presents the evolution of the free agent threads for OpenMP to the new role-shifting threads model and their integration with the Dynamic Load Balancing (DLB) library. We demonstrate how free agent threads can improve resource utilization in OpenMP applications with load imbalance in their nested parallel regions. We also demonstrate how DLB efficiently manages the malleability exposed by the role-shifting threads to address load imbalance issues. We use three real-world scientific applications, one of them to demonstrate that free agents alone can improve the OpenMP model without external tools, and two other MPI+OpenMP applications, one of them with a coupling case, to illustrate the potential of the free agent threads’ malleability with an external resource manager to increase the efficiency of the system. In addition, we demonstrate that the new implementation is more usable than the former one, letting the runtime system automatically make decisions that were made by the programmer previously. All software is released open-source., This work has received funding from the DEEP Projects, at the European Commission’s FP7, H2020, and EuroHPC Programmes, under Grant Agreements 287530, 610476, 754304, and 955606. The PCI2021-121958 financed by the Spanish State Research Agency - Ministry of Science and Innovation. And it also has the support of the Spanish Ministry of Science and Innovation (Computacion de Altas Prestaciones VIII: PID2019-107255GB)., Peer Reviewed
Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

Role-shifting threads: Increasing OpenMP malleability to address load imbalance at MPI and OpenMP

  • Criado Ledesma, Joel
  • López Herrero, Víctor
  • Vinyals Ylla Català, Joan
  • Ramirez Miranda, Guillem
  • Teruel García, Xavier
  • Garcia Gasulla, Marta
This paper presents the evolution of the free agent threads for OpenMP to the new role-shifting threads model and their integration with the Dynamic Load Balancing (DLB) library. We demonstrate how free agent threads can improve resource utilization in OpenMP applications with load imbalance in their nested parallel regions. We also demonstrate how DLB efficiently manages the malleability exposed by the role-shifting threads to address load imbalance issues. We use three real-world scientific applications, one of them to demonstrate that free agents alone can improve the OpenMP model without external tools, and two other MPI+OpenMP applications, one of them with a coupling case, to illustrate the potential of the free agent threads’ malleability with an external resource manager to increase the efficiency of the system. In addition, we demonstrate that the new implementation is more usable than the former one, letting the runtime system automatically make decisions that were made by the programmer previously. All software is released open-source., This work has received funding from the DEEP Projects, at the European Commission’s FP7, H2020, and EuroHPC Programmes, under Grant Agreements 287530, 610476, 754304, and 955606. The PCI2021-121958 financed by the Spanish State Research Agency - Ministry of Science and Innovation. And it also has the support of the Spanish Ministry of Science and Innovation (Computacion de Altas Prestaciones VIII: PID2019-107255GB)., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

Page size aware cache prefetching

  • Vavouliotis, Georgios
  • Chacon, Gino
  • Álvarez Martí, Lluc
  • Gratz, Paul V.
  • Jiménez, Daniel A.
  • Casas, Marc
The increase in working set sizes of contemporary applications outpaces the growth in cache sizes, resulting in frequent main memory accesses that deteriorate system per- formance due to the disparity between processor and memory speeds. Prefetching data blocks into the cache hierarchy ahead of demand accesses has proven successful at attenuating this bottleneck. However, spatial cache prefetchers operating in the physical address space leave significant performance on the table by limiting their pattern detection within 4KB physical page boundaries when modern systems use page sizes larger than 4KB to mitigate the address translation overheads. This paper exploits the high usage of large pages in modern systems to increase the effectiveness of spatial cache prefetch- ing. We design and propose the Page-size Propagation Module (PPM), a µarchitectural scheme that propagates the page size information to the lower-level cache prefetchers, enabling safe prefetching beyond 4KB physical page boundaries when the accessed blocks reside in large pages, at the cost of augmenting the first-level caches’ Miss Status Holding Register (MSHR) entries with one additional bit. PPM is compatible with any cache prefetcher without implying design modifications. We capitalize on PPM’s benefits by designing a module that consists of two page size aware prefetchers that inherently use different page sizes to drive prefetching. The composite module uses adaptive logic to dynamically enable the most appropriate page size aware prefetcher. Finally, we show that the proposed designs are transparent to which cache prefetcher is used. We apply the proposed page size exploitation techniques to four state-of-the-art spatial cache prefetchers. Our evalua- tion shows that our proposals improve single-core geomean performance by up to 8.1% (2.1% at minimum) over the original implementation of the considered prefetchers, across 80 memory-intensive workloads. In multi-core contexts, we report geomean speedups up to 7.7% across different cache prefetchers and core configurations., This work is supported by the Spanish Ministry of Science and Technology through the PID2019-107255GB project, the Generalitat de Catalunya (contract 2017-SGR-1414), the European Union Horizon 2020 research and innovation program under grant agreement No 955606 (DEEP-SEA EU project), the National Science Foundation through grants CNS-1938064 and CCF-1912617, and the Semiconductor Research Corporation project GRC 2936.001. Georgios Vavouliotis has been supported by the Spanish Ministry of Economy, Industry, and Competitiveness and the European Social Fund under the FPI fellowship No. PRE2018-087046. Marc Casas has been partially supported by the Grant RYC2017-23269 funded by MCIN/AEI/10.13039/501100011033 and ESF ‘Investing in your future’., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

Automated generation of high-performance computational fluid dynamics codes

  • Macià Sorrosal, Sandra
  • Martínez Ferrer, Pedro J.
  • Ayguadé Parra, Eduard
  • Beltran Querol, Vicenç
Domain-Specific Languages (DSLs) improve programmers productivity by decoupling problem descriptions from algorithmic implementations. However, DSLs for High-Performance Computing (HPC) have two additional critical requirements: performance and scalability. This paper presents the automated process of generating, from abstract mathematical specifications of Computational Fluid Dynamics (CFD) problems, optimised parallel codes that perform and scale as manually optimised ones. We consciously combine within Saiph, a DSL for solving CFD problems, low-level optimisations and parallelisation strategies, enabling high-performance single-core executions which effectively scale to multi-core and distributed environments. Our results demonstrate how high-level DSLs can offer competitive performance by transparently leveraging state-of-the-art HPC techniques., This research has received funding from the European Union’s Horizon 2020/EuroHPC research and innovation programme under grant agreement N. 955606 (DEEP-SEA), and is supported by the Spanish State Research Agency - Ministry of Science and Innovation (contract PID2019-107255 GB), and by the Generalitat de Catalunya (2017-SGR1414). This work is also supported by the Ministry of Economy of Spain through Severo Ochoa Center of Excellence Program (SEV-2015-0493)., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

Efficient data redistribution for malleable applications

  • Martínez Alvarez, Iker
  • Aliaga, José I
  • Castrillo, Maribel
  • Iserte, Sergio
Process malleability can be defined as the ability of a distributed MPI parallel job to change the number of processes on–the–fly without stopping its execution, reallocating the compute resources originally assigned to the job, and without storing application data to disk. MPI malleability consists of four stages: resource reallocation, process management, data redistribution and execution resuming. Among them, data redistribution is the most time-consuming and determines the reconfiguration time. In this paper, we compare different implementations of this stage using point-to-point and collective MPI operations, and discuss the impact of overlapping computation-communication. We then combine these strategies with different methods to expand/shrink jobs, using a synthetic application to emulate MPI-based codes and their malleable counterparts, in order to evaluate the effect of different malleability methods in parallel distributed applications. The results show that the use of asynchronous techniques speeds up execution by 1.14 and 1.21, depending on the network used., This work has been funded by project PID2020-113656RB-C21 supported by MCIN/AEI/10.13039/501100011033. Researcher I. MartínÁlvarez was supported by the predoctoral fellowship ACIF/2021/260 from Valencian Region Government and European Social Funds. The European DEEP-SEA project has received funding from the European Commission’s EuroHPC-JU program under grant agreement No 955606, the JU receives support from the European Union’s Horizon 2020 research and innovation programme and by the Spanish Ministry under Grant PCI2021-121958 funded by MCIN / AEI /10.13039 / 501100011033 and by the “European Union NextGenerationEU/PRTR”., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

OpenMP taskloop dependences

  • Maroñas Bravo, Marcos
  • Teruel García, Xavier
  • Beltran, Vincenç
Exascale systems will contain multicore/manycore processors with high core count in each node. Therefore, using a model that relaxes the synchronization, such as data-flow, is crucial to adequately exploit the potential of the hardware. The flexibility of the data-flow execution model relies on the dynamic management of data-dependences among tasks. The OpenMP standard already provides a construct, known as taskloop, that distributes the loop iteration space into several tasks, but this construct does not support the use of the depend clause yet. In this paper we propose the use of the induction variable to define data dependences in tasks created by the taskloop construct. By using the induction variable, each task will contain its own dependences based on the partition of work they received. We also aim to demonstrate that using taskloop with dependences provides an enhancement in terms of programmability with respect to using stand-alone tasks to parallelize a loop. Our implementation does not introduce any significant overhead on the taskloop implementation and, in certain cases, it outperforms the stand-alone task version., This research has received funding from the European Union’s Horizon 2020/EuroHPC research and innovation programme under grant agreement No 955606 (DEEP-SEA); and the support of the Spanish Ministry of Science and Innovation (Computacion de Altas Prestaciones VIII: PID2019-107255GB)., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

O(n) key–value sort with active compute memory

  • Esmaili Dokht, Pouya
  • Guiot Cusido, Miquel
  • Radojkovic, Petar
  • Martorell Bofill, Xavier
  • Ayguadé Parra, Eduard
  • Labarta Mancho, Jesús José
  • Adlard, Jason
  • Amato, Paolo
  • Sforzin, Marco
We propose the Active Compute Memory (ACM), a near-memory-processing architecture capable of performing key–value sort directly in the DRAM. In the ACM architecture, sort is merely the writing of data into memory with one addressing protocol (perspective) and reading it back with different perspective. The first perspective is conventional, based on the data address; the second perspective is the sorted order. The ACM requires additional tables to store the meta-data and moderate control logic enhancements that can be implemented directly in the DRAM silicon. By these modest enhancements to DRAM, ACM exploits the parallelism inherently available in the row buffer to enable sort with O ( n ) complexity. This leads to an order of magnitude improvement in ACM performance and energy compared to conventional O ( n log n ) CPU-centric sort algorithms. The ACM also shows superior performance compared to other near-memory sort accelerators. This is because the ACM processing is done near the row buffer and it exploits much lower memory access latency, higher bandwidth and wider parallel processing. The sort operation covered in this paper is just an example of an address management operation that can be efficiently implemented directly in the DRAM silicon. We release as an open source the simulation infrastructure for the ACM performance and energy modeling. We would encourage the community to use it, adapt it to other PIM proposals, and share their own evaluations., This work was funded by the Collaboration Agreement between Micron Technology, Inc. and BSC. The work was also supported by the Spanish Government, under the contracts PID2019-107255GB-C21 and CEX2021-001148-S funded by MCIN/AEI/ 10.13039/501100011033. The work also received funding from the Department of Research and Universities of the Government of Catalonia to the AccMem Research Group (Code: 2021 SGR 00807) and project DEEP-SEA funded by European Union’s Horizon 2020 research and innovation programme , under Grant Agreement no.955606 and the Spanish Government, under the contract PCI2021-121958 funded by MCIN/AEI/ 10.13039/501100011033 and “European Union NextGenerationEU/PRTR”., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

Proteo: a framework for the generation and evaluation of malleable MPI applications

  • Martín Álvarez, Iker
  • Aliaga, José I.
  • Castillo, Maribel
  • Iserte, Sergio
Applying malleability to HPC systems can increase their productivity without degrading or even improving the performance of running applications. This paper presents Proteo, a configurable framework that allows to design benchmarks to study the effect of malleability on a system, and also incorporates malleability into a real application. Proteo consists of two modules: SAM allows to emulate the computational behavior of iterative scientific MPI applications, and MaM is able to reconfigure a job during execution, adjusting the number of processes, redistributing data, and resuming execution. An in-depth study of all the possibilities shows that Proteo is able to behave like a real malleable or non-malleable application in the range [0.85, 1.15]. Furthermore, the different methods defined in MaM for process management and data redistribution are analyzed, concluding that asynchronous malleability, where reconfiguration and application execution overlap, results in a 1.15 speedup., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. This work has been funded by project PID2020-113656RB-C21 supported by MCIN/AEI/10.13039/501100011033. Researcher I. Martín-Álvarez was supported by the predoctoral fellowship ACIF/2021/260 from Valencian Region Government and European Social Funds. The European DEEP-SEA project has received funding from the European Commission’s EuroHPC-JU program under grant agreement No. 955606; the JU receives support from the European Union’s Horizon 2020 research and innovation programme and by the Spanish Ministry under Grant PCI2021-121958 funded by MCIN / AEI / 10.13039 / 501100011033 and by the “European Union NextGenerationEU/PRTR.”, Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

Combining dynamic concurrency throttling with voltage and frequency scaling on task-based programming models

  • Navarro Muñoz, Antoni
  • Lorenzon, Arthur F.
  • Ayguadé Parra, Eduard
  • Beltran Querol, Vicenç
Being on the verge of exascale performance has shifted the prioritization of performance in applications to the inclusion of power-performance efficiency as a primary objective in the High Performance Computing (HPC) community. Simultaneously, this has surfaced hardware and software efforts that employ techniques such as dynamic voltage and frequency scaling (DVFS) for core and uncore units or dynamic concurrency throttling (DCT) to exploit hardware resources efficiently, by saving energy while maintaining performance. These techniques are complementary, so they can be used together. However, employing them is not a straightforward task, as they have to be adjusted based on the workload, and it is even more complex to combine them properly. Thus, these techniques should be applied transparently by a runtime system, without relying on application developers. In this paper, we extend a task-based runtime system with an infrastructure that categorizes workloads based on their computational profile – memory-bounded, compute-bounded, or balanced. This categorization is done in an on-line manner and with a negligible overhead. With this additional information, we enhance the CPU-manager and scheduler of OmpSs-2, a task-based parallel programming model, to automatically combine DVFS and DCT techniques based on workloads. Moreover, we show that our heuristics transparently improve energy efficiency on average by 15% with no significant performance loss and either equal or surpass the energy efficiency of the best static configuration available., This research has received funding from the European Union’s Horizon 2020/EuroHPC research and innovation programme under grant agreement N.955606 (DEEP-SEA), and is supported by the Spanish State Research Agency - Ministry of Science and Innovation (contract PID2019-107255GB), and by the Generalitat de Catalunya (2017-SGR-1414). This work was also supported by Project HPC- EUROPA3, with the support of the EC Research Innovation Action under the H2020 Programme., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

Communication-aware sparse patterns for the factorized approximate inverse preconditioner

  • Laut Turón, Sergi
  • Casas, Marc
  • Borrell Pol, Ricard
The Conjugate Gradient (CG) method is an iterative solver targeting linear systems of equations Ax=b where A is a symmetric and positive definite matrix. CG convergence properties improve when preconditioning is applied to reduce the condition number of matrix A. While many different options can be found in the literature, the Factorized Sparse Approximate Inverse (FSAI) preconditioner constitutes a highly parallel option based on approximating A-1. This paper proposes the Communication-aware Factorized Sparse Approximate Inverse preconditioner (FSAIE-Comm), a method to generate extensions of the FSAI sparse pattern that are not only cache friendly, but also avoid increasing communication costs in distributed memory systems. We also propose a filtering strategy to reduce inter-process imbalance. We evaluate FSAIE-Comm on a heterogeneous set of 39 matrices achieving an average solution time decrease of 17.98%, 26.44% and 16.74% on three different architectures, respectively, Intel Skylake, Fujitsu A64FX and AMD Zen 2 with respect to FSAI. In addition, we consider a set of 8 large matrices running on up to 32,768 CPU cores, and we achieve an average solution time decrease of 12.59%., Marc Casas is supported by Grant RYC-2017-23269 funded by MCIN/AEI/ 10.13039/501100011033 and by “ESF Investing in your future”. This work has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 955606. This work has been supported by the Computación de Altas Prestaciones VIII (BSC-HPC8) project. It has also been partially supported by the EXCELLERAT project funded by the European Commission’s ICT activity of the H2020 Programme under grant agreement number: 823691 and by the Spanish Ministry of Science and Innovation (Nucleate, Project PID2020-117001GB-I00)., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

ecoHMEM: Improving object placement methodology for hybrid memory systems in HPC

  • Jordà Peroliu, Marc
  • Rai, Siddharth
  • Ayguadé Parra, Eduard
  • Labarta Mancho, Jesús José
  • Peña Monferrer, Antonio José
Recent byte-addressable persistent memory (PMEM) technology offers capacities comparable to storage devices and access times much closer to DRAMs than other non-volatile memory technology. To palliate the large gap with DRAM performance, DRAM and PMEM are usually combined. Users have the choice to either manage the placement to different memory spaces by software or leverage the DRAM as a cache for the virtual address space of the PMEM. We present novel methodology for automatic object-level placement, including efficient runtime object matching and bandwidth-aware placement. Our experiments leveraging Intel® Optane™ Persistent Memory show from matching to greatly improved performance with respect to state-of-the-art software and hardware solutions, attaining over 2x runtime improvement in miniapplications and over 6% in OpenFOAM, a complex production application., This paper received funding from the Intel-BSC Exascale Laboratory SoW 5.1, the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No. 749516, the EPEEC project from the European Union’s Horizon 2020 research and innovation program under grant agreement No 801051, the DEEP-SEA project from the European Commission’s EuroHPC program under grant agreement 955606, and the Ministerio de Ciencia e Innovacion—Agencia Estatal de Investigación (PID2019-107255GB-C21/AEI/10.13039/501100011033)., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

A generator of numerically-tailored and high-throughput accelerators for batched GEMMs

  • Ledoux Pardo, Luis Eduardo
  • Casas, Marc
We propose a hardware generator of GEMM accelerators. Our generator produces vendor-agnostic HDL describing highly customizable systolic arrays guided by accuracy and energy efficiency goals. The generated arrays have three main novel aspects. First, the accelerators handle a large variety of computer number formats using intermediate representations based on our Sign Scale Significand (S3) format. Second, the processing elements perform all intermediate dot-product arithmetic operations required by the GEMM kernel without any intermediate rounding, which makes it possible to deliver better energy efficiency than state-of-the-art approaches while offering more accuracy and reproducible results. Third, our accelerators feature the Half-Speed Sink Down (HSSD) mechanism, which maximizes the overlap of host-accelerator data transfers with GEMM computations.We evaluate our automatically generated designs in a cutting-edge setup composed of a POWER9 host, CAPI (Coherent Accelerator Processor Interface) link, and a Virtex Ultrascale Plus FPGA. Arrays can operate at the speed of the link and saturate it to reach a 13GB/s throughput. Our fine-grain customization approach allows to cover a wide range of accuracy versus efficiency scenarios and can reach 0.65GOps/s/W while producing 1024 accurate bits or 148.7GOps/s/W with 6 accurate bits. Our configurations achieve up to 1613GOps/s system performance and power efficiencies of up to 240GOps/s/W for the FPGA. This automatic generator is the first being able to produce such a variety of designs. We improve the single-precision energy efficiency of state-of-the-art FPGA GEMM accelerators by 1.86×., This work has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 955606 Marc Casas is supported by Grant RYC-2017-23269 funded by MCIN/AEI/ 10.13039/501100011033 and by “ESF Investing in your future”, Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

A BF16 FMA is all you need for DNN training

  • Osorio Ríos, John Haiber
  • Armejach Sanosa, Adrià
  • Petit, Eric
  • Henry, Greg
  • Casas, Marc
Fused Multiply-Add (FMA) functional units constitute a fundamental hardware component to train Deep Neural Networks (DNNs). Its silicon area grows quadratically with the mantissa bit count of the computer number format, which has motivated the adoption of the BrainFloat16 format (BF16). BF16 features 1 sign, 8 exponent and 7 explicit mantissa bits. Some approaches to train DNNs achieve significant performance benefits by using the BF16 format. However, these approaches must combine BF16 with the standard IEEE 754 Floating-Point 32-bit (FP32) format to achieve state-of-the-art training accuracy, which limits the impact of adopting BF16. This article proposes the first approach able to train complex DNNs entirely using the BF16 format. We propose a new class of FMA operators, FMAbf16 n m, that entirely rely on BF16 FMA hardware instructions and deliver the same accuracy as FP32. FMAbf16 n m operators achieve performance improvements within the 1.28- 1.35X range on ResNet101 with respect to FP32. FMAbf16 n m enables training complex DNNs on simple low-end hardware devices without requiring expensive FP32 FMA functional units., Marc Casas was partially supported by the under Grant RYC-2017-23269 funded by MCIN/AEI/10.13039/501100011033, and by ESF Investing in your future. Adrià Armejach is a Serra Hunter Fellow and has been partially supported by the under Grant IJCI-2017-33945 funded by MCIN/AEI/10.13039/501100011033. John Osorio has been partially supported by the under Grant PRE2019-090406 funded by MCIN/AEI/10.13039/501100011033 and by ESF Investing in your future. This work has been partially supported by Intel under the BSC-Intel collaboration and European Union Horizon 2020 research and innovation programme under Grant 955606 - DEEP-SEA EU project., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Artículo científico (JournalArticle)

Optimizing iterative data-flow scientific applications using directed cyclic graphs

  • Álvarez Robert, David
  • Beltran Querol, Vicenç
Data-flow programming models have become a popular choice for writing parallel applications as an alternative to traditional work-sharing parallelism. They are better suited to write applications with irregular parallelism that can present load imbalance. However, these programming models suffer from overheads related to task creation, scheduling and dependency management, limiting performance and scalability when tasks become too small. At the same time, many HPC applications implement iterative methods or multi-step simulations that create the same directed acyclic graphs of tasks on each iteration. By giving application programmers a way to express that a specific loop is creating the same task pattern on each iteration, we can create a single task directed acyclic graph (DAG) once and transform it into a cyclic graph. This cyclic graph is then reused for successive iterations, minimizing task creation and dependency management overhead. This paper presents the taskiter, a new construct we propose for the OmpSs-2 and OpenMP programming models, allowing the use of directed cyclic task graphs (DCTG) to minimize runtime overheads. Moreover, we present a simple immediate successor locality-aware heuristic that minimizes task scheduling overhead by bypassing the runtime task scheduler. We evaluate the implementation of the taskiter and the immediate successor heuristic in 8 iterative benchmarks. Using small task granularities, we obtain a geometric mean speedup of 2.56x over the reference OmpSs-2 implementation, and a 3.77x and 5.2x speedup over the LLVM and GCC OpenMP runtimes, respectively., This work was supported in part by the European Union’s Horizon 2020/EuroHPC Research and Innovation Programme (DEEP-SEA) under Grant 955606; in part by the Spanish State Research Agency—Ministry of Science and Innovation, Generalitat de Catalunya, under Project PCI2021121958 and Project 2021-SGR-01007; in part by the Spanish Ministry of Science and Technology under Contract PID2019-107255GB; and in part by Severo Ochoa under Grant CEX2021-001148-S/MCIN/AEI/10.13039/501100011033., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

Transparent load balancing of MPI programs using OmpSs-2@Cluster and DLB

  • Aguilar Mena, Jimmy
  • Ali, Omar Shaaban Ibrahim
  • López Herrero, Víctor
  • Garcia Casulla, Marta
  • Carpenter, Paul Matthew
  • Ayguadé Parra, Eduard
  • Labarta Mancho, Jesús José
Load imbalance is a long-standing source of inefficiency in high performance computing. The situation has only got worse as applications and systems increase in complexity, e.g., adaptive mesh refinement, DVFS, memory hierarchies, power and thermal management, and manufacturing processes. Load balancing is often implemented in the application, but it obscures application logic and may need extensive code refactoring. This paper presents an automated and transparent dynamic load balancing approach for MPI applications with OmpSs-2 tasks, which relieves applications from this burden. Only local and trivial changes are required to the application. Our approach exploits the ability of OmpSs-2@Cluster to offload tasks for execution on other nodes, and it reallocates compute resources among ranks using the Dynamic Load Balancing~(DLB) library. It employs LeWI to react to fine-grained load imbalances and DROM to address coarse-grained load imbalances by reserving cores on other nodes that can be reclaimed on demand. We use an expander graph to limit the amount of point-to-point communication and state. The results show 46% reduction in time-to-solution for micro-scale solid mechanics on 32 nodes and a 20% reduction beyond DLB for $n$-body on 16 nodes, when one node is running slow. A synthetic benchmark shows that performance is within 10% of optimal for an imbalance of up to 2.0 on 8 nodes. All software is released open source., This research has received funding from the European Union’s Horizon 2020/EuroHPC research and innovation programme under grant agreement No 955606 (DEEP-SEA) and 754337 (EuroEXA). It is supported by the Spanish State Research Agency - Ministry of Science and Innovation (contract PID2019-107255GB and Ramon y Cajal fellowship RYC2018-025628-I) and by the Generalitat de Catalunya (2017-SGR-1414)., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

FASE: A fast, accurate and seamless emulator for custom numerical formats

  • Osorio Ríos, John Haiber
  • Armejach Sanosa, Adrià
  • Petit, Eric
  • Henry, Greg
  • Casas, Marc
Deep Neural Networks (DNNs) have become ubiquitous in a wide range of application domains. Despite their success, training DNNs is an expensive task that has motivated the use of reduced numerical precision formats to improve performance and reduce power consumption. Emulation techniques are a good fit to understand the properties of new numerical formats on a particular workload. However, current SoA techniques are not able to perform these tasks quickly and accurately on a wide variety of workloads.We propose FASE, a Fast, Accurate, and Seamless Emulator that leverages dynamic binary translation to enable emulation of custom numerical formats. FASE is fast: allowing emulation of large unmodified workloads; accurate: emulating at the instruction operand level; and seamless: as it does not require any code modifications and works on any application or DNN framework without any language, compiler, or source code access restrictions., Marc Casas has been partially supported by the Grant RYC2017-23269 funded by MCIN/AEI/ 10.13039/501100011033 and by “ESF Investing in your future”. Adria Armejach is a Serra Hunter Fellow and has been partially supported by the Grant IJCI-2017-33945 funded by MCIN/AEI/ 10.13039/501100011033. John Osorio has been partially supported by the Grant PRE2019- 090406 funded by MCIN/AEI/ 10.13039/501100011033 and by “ESF Investing in your future”. This work has been partially supported by Intel under the BSC-Intel collaboration and European Union’s Horizon 2020 research and innovation programme under grant agreement No 955606 - DEEP-SEA EU project., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publication Contribución de congreso (ConferenceOutput)

An OpenMP free agent threads implementation

  • López, Victor
  • Criado Ledesma, Joel
  • Peñacoba Veigas, Raúl
  • Ferrer Ibañez, Roger
  • Teruel García, Xavier
  • Garcia Gasulla, Marta
In this paper, we introduce a design and implementation of the free agent threads for OpenMP. These threads increase the malleability of the OpenMP programming model, offering resource managers and runtime systems flexibility to manage threads and resources efficiently. We demonstrate how free agent threads can address load imbalances problems at the OpenMP level and at an MPI level or higher. We use two mini-apps extracted from two real HPC applications and representative of real-world codes to demonstrate this. We conclude that more malleability in thread management is necessary, and free agents can be regarded as a practical starting point to increase malleability in thread management., This work has been done as part of the European Processor Initiative project. The European Processor Initiative (EPI) (FPA: 800928) has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement EPI-SGA1: 826647. It has also received funding from the European Union's Horizon 2020/EuroHPC research and innovation programme under grant agreement No 955606 (DEEP-SEA); and the support of the Spanish Ministry of Science and Innovation (Computacion de Altas Prestaciones VIII: PID2019-107255GB)., Peer Reviewed, Postprint (author's final draft)