M. Guney, Kazushige Goto, Timothy B. Costa, S. Knepper, Louise Huot, A. Mitrano, S. Story
{"title":"Optimizing Matrix Multiplication on Intel® Xeon Phi TH x200 Architecture","authors":"M. Guney, Kazushige Goto, Timothy B. Costa, S. Knepper, Louise Huot, A. Mitrano, S. Story","doi":"10.1109/ARITH.2017.19","DOIUrl":"https://doi.org/10.1109/ARITH.2017.19","url":null,"abstract":"Matrix multiplication is ubiquitous in scientific computing. From computational science to machine learning, a large and diverse set of applications rely on the performance of general matrix-matrix multiplication (GEMM) subroutines. The Intel® Math Kernel Library(R) provides highly optimized GEMM subroutines that take full advantage of the available parallelism and vectorization in both Intel® Xeon® and Intel® Xeon Phi(TM) processors. In this paper we discuss the optimization of GEMM subroutines for the Intel® Xeon Phi(TM) x200 (code-named Knights Landing).","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134026905","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Algorithms and Arithmetic: Choose Wisely","authors":"G. Constantinides","doi":"10.1109/ARITH.2017.17","DOIUrl":"https://doi.org/10.1109/ARITH.2017.17","url":null,"abstract":"I will introduce a semi-formalism to allow us to conceptually reason about the differences between customised arithmetic design, as one might see in FPGAbased compute, and general purpose arithmetic, as one might find in microprocessor design. This framework will, I hope, expose to the reader the reason that we should be thinking carefully about appropriate data representations when designing custom hardware for compute, as well as clearly showing the link between these decisions and algorithmic ones. I will then provide a concrete example from the literature on matrix computation where some careful algorithmic tweaking results in the ability to use fixed-point arithmetic and, hence, far higher performance than would otherwise be achieved.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125433483","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Round-off Error Analysis of Explicit One-Step Numerical Integration Methods","authors":"S. Boldo, Florian Faissole, Alexandre Chapoutot","doi":"10.1109/ARITH.2017.22","DOIUrl":"https://doi.org/10.1109/ARITH.2017.22","url":null,"abstract":"Ordinary differential equations are ubiquitous in scientific computing. Solving exactly these equations is usually not possible, except for special cases, hence the use of numerical schemes to get a discretized solution. We are interested in such numerical integration methods, for instance Euler's method or the Runge-Kutta methods. As they are implemented using floating-point arithmetic, round-off errors occur. In order to guarantee their accuracy, we aim at providing bounds on the round-off errors of explicit one-step numerical integration methods. Our methodology is to apply a fine-grained analysis to these numerical algorithms. Our originality is that our floating-point analysis takes advantage of the linear stability of the scheme, a mathematical property that vouches the scheme is well-behaved.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127812612","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Normalizing or Not Normalizing? An Open Question for Floating-Point Arithmetic in Embedded Systems","authors":"S. González-Navarro, J. Hormigo","doi":"10.1109/ARITH.2017.16","DOIUrl":"https://doi.org/10.1109/ARITH.2017.16","url":null,"abstract":"Emerging embedded applications lack of a specific standard when they require floating-point arithmetic. In this situation they use the IEEE-754 standard or ad hoc variations of it. However, this standard was not designed for this purpose. This paper aims to open a debate to define a new extension of the standard to cover embedded applications. In this work, we only focus on the impact of not performing normalization. We show how eliminating the condition of normalized numbers, implementation costs can be dramatically reduced, at the expense of a moderate loss of accuracy. Several architectures to implement addition and multiplication for non-normalized numbers are proposed and analyzed. We show that a combined architecture (adder-multiplier) can halve the area and power consumption of its counterpart IEEE-754 architecture. This saving comes at the cost of reducing an average of about 10 dBs the Signal-to-Noise Ratio for the tested algorithms. We think these results should encourage researchers to perform further investigation in this issue.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"73 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124599826","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"High-Precision Anchored Accumulators for Reproducible Floating-Point Summation","authors":"D. Lutz, C. Hinds","doi":"10.1109/ARITH.2017.20","DOIUrl":"https://doi.org/10.1109/ARITH.2017.20","url":null,"abstract":"This paper introduces a new datatype that allows reproducible accumulation of floating-point (FP) numbers in a programmer-selectable range. The new datatype has a larger significand and a smaller range than existing FP formats and has much better arithmetic and computational properties. In particular, it is associative, parallelizable, reproducible and correct. For the modest ranges that will accommodate most problems, it is also much faster: 3 to 12 times faster on a single 256-bit SIMD implementation. The paper also describes a new instruction and associated datapath that support the proposed datatype, and discusses how a recently published software algorithm for reproducible FP summation could be implemented using the proposed approach.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122260938","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"The Classical Relative Error Bounds for Computing Sqrt(a^2 + b^2) and c / sqrt(a^2 + b^2) in Binary Floating-Point Arithmetic are Asymptotically Optimal","authors":"C. Jeannerod, J. Muller, A. Plet","doi":"10.1109/ARITH.2017.40","DOIUrl":"https://doi.org/10.1109/ARITH.2017.40","url":null,"abstract":"We study the accuracy of classical algorithms for evaluating expressions of the form $sqrt{a^2+b^2}$ and $c/sqrt{a^2+b^2}$ in radix-$2$, precision-$p$ floating-point arithmetic, assuming that the elementary arithmetic operations $pm$, $times$, $/$, $surd$ are rounded to nearest, and assuming an unbounded exponent range. Classical analyses show that the relative error is bounded by $2u + mathcal{O} (u^2)$ for $sqrt{a^2+b^2}$, and by $3u + mathcal{O} (u^2)$ for $c/sqrt{a^2+b^2}$, where $u = 2^{-p}$ is the unit roundoff. Recently, it was observed that for $sqrt{a^2+b^2}$ the $mathcal{O}(u^2)$ term is in fact not needed [1]. We show here that it is not needed either for $c/sqrt{a^2+b^2}$. Furthermore, we show that these error bounds are asymptotically optimal. Finally, we show that both the bounds and their asymptotic optimality remain valid when an FMA instruction is used to evaluate $a^2+b^2$.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117281262","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jack Koenig, David Biancolin, J. Bachrach, K. Asanović
{"title":"A Hardware Accelerator for Computing an Exact Dot Product","authors":"Jack Koenig, David Biancolin, J. Bachrach, K. Asanović","doi":"10.1109/ARITH.2017.38","DOIUrl":"https://doi.org/10.1109/ARITH.2017.38","url":null,"abstract":"We study the implementation of a hardware accelerator that computes a dot product of IEEE-754 floating-point numbers exactly. The accelerator uses a wide (640 or 4288 bits for single or double-precision respectively) fixed-point representation into which intermediate floating-point products are accumulated. We designed the accelerator as a generator in Chisel, which can synthesize various configurations of the accelerator that make different area-performance trade-offs.We integrated eight different configurations into an SoC comprised of RISC-V in-order scalar core, split L1 instruction and data caches, and unified L2 cache. In a TSMC 45 nm technology, the accelerator area ranges from 0.05 mm2 to 0.32 mm2, and all configurations could be clocked at frequencies in excess of 900MHz. The accelerator successfully saturates the SoC's memory system, achieving the same per-element efficiency (1 cycle-per-element) as Intel MKL running on an x86 machine with a similar cache configuration.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"57 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127538237","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A Sum Error Detection Scheme for Decimal Arithmetic","authors":"Álvaro Vázquez, E. Antelo","doi":"10.1109/ARITH.2017.34","DOIUrl":"https://doi.org/10.1109/ARITH.2017.34","url":null,"abstract":"Users of financial and e-commerce services demand a high degree of reliability and at the same time an increasing demand of speed of processing. On the other hand, soft errors are becoming more significant due to the higher densities and reduced CMOS integration technology sizes. Among the basic arithmetic operations, addition/subtraction is the most demanded. Although in the past, binary implementations were only considered, today decimal implementations are becoming important. In this context, we introduce a modular design for fast error checking of binary and decimal (BCD) addition/subtraction operations that avoids the whole replication of the arithmetic units. Unlike other error checkers based on parity prediction or residue checking, this is a separable design that lies completely off of the critical path of the protected adder without incurring in important penalties in area or performance.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126192679","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Modified Fused Multiply and Add for Exact Low Precision Product Accumulation","authors":"Nicolas Brunie","doi":"10.1109/ARITH.2017.29","DOIUrl":"https://doi.org/10.1109/ARITH.2017.29","url":null,"abstract":"The implementation of the Fused Multiply and Add (FMA) operation has been extensively studied in the literature on standard and large precisions. We suggest re- visiting those studies for 16-bit precision. We introduce a variation of the Mixed precision FMA targeted for applications processing low precision inputs (such as machine learning). We also introduce several versions of a fixed point based floating- point FMA which performs an exact accumulation of binary16 numbers. We study the implementation and area footprint of those operators in comparison with standard FMAs.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122821923","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"QRD for Parallel Arithmetic Structures","authors":"M. Langhammer","doi":"10.1109/ARITH.2017.26","DOIUrl":"https://doi.org/10.1109/ARITH.2017.26","url":null,"abstract":"We present a new organization of the QR decomposition (QRD), which is optimized for implementation on parallel arithmetic structures, such as found in current FPGAs. Data dependencies are hidden in the pipeline depths of the datapaths, allowing implementations to approach 100% sustained to peak throughput. The algorithm presented here is based on the Modified Gram-Schmidt (MGS) method, and is designed for floating point (FP) arithmetic, with a combination of separate dot product and multiply-add datapaths. In this short paper, we concentrate on the description of the algorithm and architecture, rather than the implementation, of the QRD.","PeriodicalId":279178,"journal":{"name":"2017 IEEE 24th Symposium on Computer Arithmetic (ARITH)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114126444","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}