During the past decade, Deep Learning (DL) algorithms, programming systems and hardware have converged with the High Performance Computing (HPC) counterparts. Nevertheless, the programming methodology of DL and HPC systems is stagnant, relying on highly-optimized, yet platform-specific and inflexible vendor-optimized libraries. Such libraries provide close-to-peak performance on specific platforms, kernels and shapes thereof that vendors have dedicated optimizations efforts, while they underperform in the remaining use-cases, yielding non-portable codes with performance glass-jaws. This work introduces a framework to develop efficient, portable DL and HPC kernels for modern CPU architectures. We decompose the kernel development in two steps: 1) Expressing the computational core using Tensor Processing Primitives (TPPs): a compact, versatile set of 2D-tensor operators, 2) Expressing the logical loops around TPPs in a high-level, declarative fashion whereas the exact instantiation (ordering, tiling, parallelization) is determined via simple knobs. We demonstrate the efficacy of our approach using standalone kernels and end-to-end workloads that outperform state-of-the-art implementations on diverse CPU platforms.

During the past decade, novel Deep Learning (DL) algorithms, workloads and hardware have been developed to tackle a wide range of problems. Despite the advances in workload and hardware ecosystems, the programming methodology of DL systems is stagnant. DL workloads leverage either highly-optimized, yet platform-specific and inflexible kernels from DL libraries, or in the case of novel operators, reference implementations are built via DL framework primitives with underwhelming performance. This work introduces the Tensor Processing Primitives (TPP), a programming abstraction striving for efficient, portable implementation of DL workloads with high-productivity. TPPs define a compact, yet versatile set of 2D-tensor operators (or a virtual Tensor ISA), which subsequently can be utilized as building-blocks to construct complex operators on high-dimensional tensors. The TPP specification is platform-agnostic, thus code expressed via TPPs is portable, whereas the TPP implementation is highly-optimized and platform-specific. We demonstrate the efficacy and viability of our approach using standalone kernels and end-to-end DL & HPC workloads expressed entirely via TPPs that outperform state-of-the-art implementations on multiple platforms.

Full-batch training on Graph Neural Networks (GNN) to learn the structure of large graphs is a critical problem that needs to scale to hundreds of compute nodes to be feasible. It is challenging due to large memory capacity and bandwidth requirements on a single compute node and high communication volumes across multiple nodes. In this paper, we present DistGNN that optimizes the well-known Deep Graph Library (DGL) for full-batch training on CPU clusters via an efficient shared memory implementation, communication reduction using a minimum vertex-cut graph partitioning algorithm and communication avoidance using a family of delayed-update algorithms. Our results on four common GNN benchmark datasets: Reddit, OGB-Products, OGB-Papers and Proteins, show up to 3.7× speed-up using a single CPU socket and up to 97× speed-up using 128 CPU sockets, respectively, over baseline DGL implementations running on a single CPU socket.

Deep learning (DL) is one of the most prominent branches of machine learning. Due to the immense computational cost of DL workloads, industry and academia have developed DL libraries with highly-specialized kernels for each workload/architecture, leading to numerous, complex code-bases that strive for performance, yet they are hard to maintain and do not generalize. In this work, we introduce the batch-reduce GEMM kernel and show how the most popular DL algorithms can be formulated with this kernel as the basic building-block. Consequently, the DL library-development degenerates to mere (potentially automatic) tuning of loops around this sole optimized kernel. By exploiting our new kernel we implement Recurrent Neural Networks, Convolution Neural Networks and Multilayer Perceptron training and inference primitives in just 3K lines of high-level code. Our primitives outperform vendor-optimized libraries on multi-node CPU clusters, and we also provide proof-of-concept CNN kernels targeting GPUs. Finally, we demonstrate that the batch-reduce GEMM kernel within a tensor compiler yields high-performance CNN primitives, further amplifying the viability of our approach.

During the last two years, the goal of many researchers has been to squeeze the last bit of performance out of HPC system for AI tasks. Often this discussion is held in the context of how fast ResNet50 can be trained. Unfortunately, ResNet50 is no longer a representative workload in 2020. Thus, we focus on Recommender Systems which account for most of the AI cycles in cloud computing centers. More specifically, we focus on Facebook's DLRM benchmark. By enabling it to run on latest CPU hardware and software tailored for HPC, we are able to achieve up to two-orders of magnitude improvement in performance on a single socket compared to the reference CPU implementation, and high scaling efficiency up to 64 sockets, while fitting ultra-large datasets which cannot be held in single node's memory. Therefore, this paper discusses and analyzes novel optimization and parallelization techniques for the various operators in DLRM. Several optimizations (e.g. tensorcontraction accelerated MLPs, framework MPI progression, BFLOAT16 training with up to 1.8× speed-up) are general and transferable to many other deep learning topologies.

Google's neural machine translation (GNMT) is state-of-the-art recurrent neural network (RNN/LSTM) based language translation application. It is computationally more demanding than well-studied convolutional neural networks (CNNs). Also, in contrast to CNNs, RNNs heavily mix compute and memory bound layers which requires careful tuning on a latency machine to optimally use fast on-die memories for best single processor performance. Additionally, due to massive compute demand, it is essential to distribute the entire workload among several processors and even compute nodes. To the best of our knowledge, this is the first work which attempts to scale this application on an Intel CPU cluster. Our CPU-based GNMT optimization, the first of its kind, achieves this by the following steps: (i) we choose a monolithic long short-term memory (LSTM) cell implementation from LIBXSMM library (specifically tuned for CPUs) and integrate it into TensorFlow, (ii) we modify GNMT code to use fused time step LSTM op for the encoding stage, (iii) we combine Horovod and Intel MLSL scaling libraries for improved performance on multiple nodes, and (iv) we extend the bucketing logic for grouping similar length sentences together to multiple nodes for achieving load balance across multiple ranks. In summary, we demonstrate that due to these changes we are able to outperform Google's stock CPU-based GNMT implementation by ~2x on single node and potentially enable more than 25x speedup using 16 node CPU cluster.

We present a new approach for multithreaded communication and asynchronous progress in MPI applications, wherein we offload communication processing to a dedicated thread. The central premise is that given the rapidly increasing core counts on modern systems, the improvements in MPI performance arising from dedicating a thread to drive communication outweigh the small loss of resources for application computation, particularly when overlap of communication and computation can be exploited. Our approach allows application threads to make MPI calls concurrently, enqueuing these as communication tasks to be processed by a dedicated communication thread. This not only guarantees progress for such communication operations, but also reduces load imbalance. Our implementation additionally significantly reduces the overhead of mutual exclusion seen in existing implementations for applications using MPI_THREAD_MULTIPLE. Our technique requires no modification to the application, and we demonstrate significant performance improvement (up to 2X) for QCD, 1-D FFT and deep learning CNN applications.

Intel Xeon Phi coprocessor-based clusters offer high compute and memory performance for parallel workloads and also support direct network access. Many real world applications are significantly impacted by network characteristics and to maximize the performance of such applications on these clusters, it is particularly important to effectively saturate network bandwidth and/or hide communications latency. We demonstrate how to do so using techniques such as pipelined DMAs for data transfer, dynamic chunk sizing, and better asynchronous progress. We also show a method for, and the impact of avoiding serialization and maximizing parallelism during application communication phases. Additionally, we apply application optimizations focused on balancing computation and communication in order to hide communication latency and improve utilization of cores and of network bandwidth. We demonstrate the impact of our techniques on three well known and highly optimized HPC kernels running natively on the Intel Xeon Phi coprocessor. For the Wilson-Dslash operator from Lattice QCD, we characterize the improvements from each of our optimizations for communication performance, apply our method for maximizing concurrency during communication phases, and show an overall 48% improvement from our previously best published result. For HPL/LINPACK, we show 68.5% efficiency with 97 TFLOPs on 128 Intel Xeon Phi coprocessors, the first ever reported native HPL efficiency on a coprocessor-based supercomputer. For FFT, we show 10.8 TFLOPs using 1024 Intel Xeon Phi coprocessors on the TACC Stampede cluster, the highest reported performance on any Intel Architecture-based cluster and the first such result to be reported on a coprocessor-based supercomputer.

The gap between the cost of moving data and the cost of computing continues to grow, making it ever harder to design iterative solvers on extreme-scale architectures. This problem can be alleviated by alternative algorithms that reduce the amount of data movement. We investigate this in the context of Lattice Quantum Chromo dynamics and implement such an alternative solver algorithm, based on domain decomposition, on Intel®Xeon Phi co-processor (KNC) clusters. We demonstrate close-to-linear on-chip scaling to all 60 cores of the KNC. With a mix of single- and half-precision the domain-decomposition method sustains 400-500 Gflop/s per chip. Compared to an optimized KNC implementation of a standard solver [1], our full multi-node domain-decomposition solver strong-scales to more nodes and reduces the time-to-solution by a factor of 5.

A new sparse high performance conjugate gradient benchmark (HPCG) has been recently released to address challenges in the design of sparse linear solvers for the next generation extreme-scale computing systems. Key computation, data access, and communication pattern in HPCG represent building blocks commonly found in today's HPC applications. While it is a well known challenge to efficiently parallelize Gauss-Seidel smoother, the most time-consuming kernel in HPCG, our algorithmic and architecture-aware optimizations deliver 95% and 68% of the achievable bandwidth on Xeon and Xeon Phi, respectively. Based on available parallelism, our Xeon Phi shared-memory implementation of Gauss-Seidel smoother selectively applies block multi-color reordering. Combined with MPI parallelization, our implementation balances parallelism, data access locality, CG convergence rate, and communication overhead. Our implementation achieved 580 TFLOPS (82% parallelization efficiency) on Tianhe-2 system, ranking first on the most recent HPCG list in July 2014. In addition, we demonstrate that our optimizations not only benefit HPCG original dataset, which is based on structured 3D grid, but also a wide range of unstructured matrices.

Modern high-speed interconnection networks are designed with capabilities to support communication from multiple processor cores. The MPI endpoints extension has been proposed to ease process and thread count tradeoffs by enabling multithreaded MPI applications to efficiently drive independent network communication. In this work, we present the first implementation of the MPI endpoints interface and demonstrate the first applications running on this new interface. We use a novel library-based design that can be layered on top of any existing, production MPI implementation. Our approach uses proxy processes to isolate threads in an MPI job, eliminating threading overheads within the MPI library and allowing threads to achieve process-like communication performance. We evaluate the performance advantages of our implementation through several benchmarks and kernels. Performance results for the Lattice QCD Dslash kernel indicate that endpoints provides up to 2.9× improvement in communication performance and 1.87× overall performance improvement over a highly optimized hybrid MPI+OpenMP baseline on 128 processors.

The Non-Uniform Fast Fourier Transform (NUFFT) is a generalization of FFT to non-equidistant samples. It has many applications which vary from medical imaging to radio astronomy to the numerical solution of partial differential equations. Despite recent advances in speeding up NUFFT on various platforms, its practical applications are still limited, due to its high computational cost, which is significantly dominated by the convolution of a signal between a non-uniform and uniform grids. The computational cost of the NUFFT is particularly detrimental in cases which require fast reconstruction times, such as iterative 3D non-Cartesian MRI reconstruction. We propose novel and highly scalable parallel algorithm for performing NUFFT on x86-based multi-core CPUs. The high performance of our algorithm relies on good SIMD utilization and high parallel efficiency. On convolution, we demonstrate on average 90% SIMD efficiency using SSE, as well up to linear scalability using a quad-socket 40-core Intel(R) Xeon(R) E7-4870 Processors based system. As a result, on dual socket Intel(R) Xeon(R) X5670 based server, our NUFFT implementation is more than 4x faster compared to the best available NUFFT3D implementation, when run on the same hardware. On Intel(R) Xeon(R) E5-2670 processor based server, our NUFFT implementation is 1.5X faster than any published NUFFT implementation today. Such speed improvement opens new usages for NUFFT. For example, iterative multi channel reconstruction of a 240×240×240 image could execute in just over 3 minutes, which is on the same order as contemporary non-iterative (and thus less-accurate) 3D NUFFT-based MRI reconstructions.

Multigrid methods are widely used to accelerate the convergence of iterative solvers for linear systems used in a number of different application areas. In this paper, we explore optimization techniques for geometric multigrid on existing and emerging multicore systems including the Opteron-based Cray XE6, Intel® Xeon® E5-2670 and X5550 processor-based Infiniband clusters, as well as the new Intel® Xeon Phi coprocessor (Knights Corner). Our work examines a variety of novel techniques including communication-aggregation, threaded wavefront-based DRAM communication-avoiding, dynamic threading decisions, SIMDization, and fusion of operators. We quantify performance through each phase of the V-cycle for both single-node and distributed-memory experiments and provide detailed analysis for each class of optimization. Results show our optimizations yield significant speedups across a variety of subdomain sizes while simultaneously demonstrating the potential of multi- and manycore processors to dramatically accelerate single-node performance. However, our analysis also indicates that improvements in networks and communication will be essential to reap the potential of manycore processors in large-scale multigrid calculations.

Address re-mapping techniques in so-called active memory systems have been shown to dramatically increase the performance of applications with poor cache and/or communication behavior on shared memory multiprocessors. However, these systems require custom hardware in the memory controller for cache line assembly/disassembly, address translation between re-mapped and normal addresses, and coherence logic. In this paper we make the important observation that on a traditional flexible distributed shared memory (DSM) multiprocessor node, equipped with a coherence protocol thread context as in SMTp or a simple dedicated in-order protocol processing core as in a CMP, the address re-mapping techniques can be implemented in software running on the protocol thread or core without custom hardware in the memory controller while delivering high performance. We implement the active memory address re-mapping techniques of parallel reduction and matrix transpose (two popular kernels in scientific, multimedia, and data mining applications) on these systems, outline the novel coherence protocol extensions needed to make them run efficiently in software protocols, and evaluate these protocols on four different DSM multiprocessor architectures with multi-threaded and/or dual-core nodes. The proposed protocol extensions yield speedup of 1.45 for parallel reduction and 1.29 for matrix transpose on a 16-node DSM multiprocessor when compared to non-active memory baseline systems and achieve performance comparable to the existing active memory architectures that rely on custom hardware in the memory controller