اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدAnatomy Of High-Performance Deep Learning Convolutions On SIMD Architectures
80
0
0.0
(
0
)
نشر من قبل
Evangelos Georganas
تاريخ النشر
2018
مجال البحث
الهندسة المعلوماتية
والبحث باللغة
English
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
Convolution layers are prevalent in many classes of deep neural networks, including Convolutional Neural Networks (CNNs) which provide state-of-the-art results for tasks like image recognition, neural machine translation and speech recognition. The computationally expensive nature of a convolution operation has led to the proliferation of implementations including matrix-matrix multiplication formulation, and direct convolution primarily targeting GPUs. In this paper, we introduce direct convolution kernels for x86 architectures, in particular for Xeon and XeonPhi systems, which are implemented via a dynamic compilation approach. Our JIT-based implementation shows close to theoretical peak performance, depending on the setting and the CPU architecture at hand. We additionally demonstrate how these JIT-optimized kernels can be integrated into a lightweight multi-node graph execution model. This illustrates that single- and multi-node runs yield high efficiencies and high image-throughputs when executing state-of-the-art image recognition tasks on CPUs.
قيم البحث
اقرأ أيضاً
The trend towards highly parallel multi-processing is ubiquitous in all modern computer architectures, ranging from handheld devices to large-scale HPC systems; yet many applications are struggling to fully utilise the multiple levels of parallelism
exposed in modern high-performance platforms. In order to realise the full potential of recent hardware advances, a mixed-mode between shared-memory programming techniques and inter-node message passing can be adopted which provides high-levels of parallelism with minimal overheads. For scientific applications this entails that not only the simulation code itself, but the whole software stack needs to evolve. In this paper, we evaluate the mixed-mode performance of PETSc, a widely used scientific library for the scalable solution of partial differential equations. We describe the addition of OpenMP threaded functionality to the library, focusing on sparse matrix-vector multiplication. We highlight key challenges in achieving good parallel performance, such as explicit communication overlap using task-based parallelism, and show how to further improve performance by explicitly load balancing threads within MPI processes. Using a set of matrices extracted from Fluidity, a CFD application code which uses the library as its linear solver engine, we then benchmark the parallel performance of mixed-mode PETSc across multiple nodes on several modern HPC architectures. We evaluate the parallel scalability on Uniform Memory Access (UMA) systems, such as the Fujitsu PRIMEHPC FX10 and IBM BlueGene/Q, as well as a Non-Uniform Memory Access (NUMA) Cray XE6 platform. A detailed comparison is performed which highlights the characteristics of each particular architecture, before demonstrating efficient strong scalability of sparse matrix-vector multiplication with significant speedups over the pure-MPI mode.
The SIMT execution model is commonly used for general GPU development. CUDA and OpenCL developers write scalar code that is implicitly parallelized by compiler and hardware. On Intel GPUs, however, this abstraction has profound performance implicatio
ns as the underlying ISA is SIMD and important hardware capabilities cannot be fully utilized. To close this performance gap we introduce C-For-Metal (CM), an explicit SIMD programming framework designed to deliver close-to-the-metal performance on Intel GPUs. The CM programming language and its vector/matrix types provide an intuitive interface to exploit the underlying hardware features, allowing fine-grained register management, SIMD size control and cross-lane data sharing. Experimental results show that CM applications from different domains outperform the best-known SIMT-based OpenCL implementations, achieving up to 2.7x speedup on the latest Intel GPU.
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 lo
nger 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 Facebooks DLRM benchmark. By enabling it to run on latest CPU hardware and software tailored for HPC, we are able to achieve more than two-orders of magnitude improvement in performance (110x) on a single socket compared to the reference CPU implementation, and high scaling efficiency up to 64 sockets, while fitting ultra-large datasets. This paper discusses the optimization techniques for the various operators in DLRM and which component of the systems are stressed by these different operators. The presented techniques are applicable to a broader set of DL workloads that pose the same scaling challenges/characteristics as DLRM.
We present details of our implementation of the Wuppertal adaptive algebraic multigrid code DD-$alpha$AMG on SIMD architectures, with particular emphasis on the Intel Xeon Phi processor (KNC) used in QPACE 2. As a smoother, the algorithm uses a domai
n-decomposition-based solver code previously developed for the KNC in Regensburg. We optimized the remaining parts of the multigrid code and conclude that it is a very good target for SIMD architectures. Some of the remaining bottlenecks can be eliminated by vectorizing over multiple test vectors in the setup, which is discussed in the contribution of Daniel Richtmann.
Todays high-performance computing (HPC) applications are producing vast volumes of data, which are challenging to store and transfer efficiently during the execution, such that data compression is becoming a critical technique to mitigate the storage
burden and data movement cost. Huffman coding is arguably the most efficient Entropy coding algorithm in information theory, such that it could be found as a fundamental step in many modern compression algorithms such as DEFLATE. On the other hand, todays HPC applications are more and more relying on the accelerators such as GPU on supercomputers, while Huffman encoding suffers from low throughput on GPUs, resulting in a significant bottleneck in the entire data processing. In this paper, we propose and implement an efficient Huffman encoding approach based on modern GPU architectures, which addresses two key challenges: (1) how to parallelize the entire Huffman encoding algorithm, including codebook construction, and (2) how to fully utilize the high memory-bandwidth feature of modern GPU architectures. The detailed contribution is four-fold. (1) We develop an efficient parallel codebook construction on GPUs that scales effectively with the number of input symbols. (2) We propose a novel reduction based encoding scheme that can efficiently merge the codewords on GPUs. (3) We optimize the overall GPU performance by leveraging the state-of-the-art CUDA APIs such as Cooperative Groups. (4) We evaluate our Huffman encoder thoroughly using six real-world application datasets on two advanced GPUs and compare with our implemented multi-threaded Huffman encoder. Experiments show that our solution can improve the encoding throughput by up to 5.0X and 6.8X on NVIDIA RTX 5000 and V100, respectively, over the state-of-the-art GPU Huffman encoder, and by up to 3.3X over the multi-thread encoder on two 28-core Xeon Platinum 8280 CPUs.
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
