Hadi Esmaeilzadeh
Adrian Sampson
Doug Burger
Abstract
As improvements in per-transistor speed and energy efficiency diminish, radical departures from conventional approaches are needed to continue improvements in the performance and energy efficiency of general-purpose processors. One such departure is approximate computing, where error in computation is acceptable and the traditional robust digital abstraction of near-perfect accuracy is relaxed. Conventional techniques in energy-efficient computing navigate a design space defined by the two dimensions of performance and energy, and traditionally trade one for the other. General-purpose approximate computing explores a third dimension---error---and trades the accuracy of computation for gains in both energy and performance. Techniques to harvest large savings from small errors have proven elusive. This paper describes a new approach that uses machine learning-based transformations to accelerate approximation-tolerant programs. The core idea is to train a learning model how an approximable region of code---code that can produce imprecise but acceptable results---behaves and replace the original code region with an efficient computation of the learned model. We use neural networks to learn code behavior and approximate it. We describe the Parrot algorithmic transformation, which leverages a simple programmer annotation ("approximable") to transform a code region from a von Neumann model to a neural model. After the learning phase, the compiler replaces the original code with an invocation of a low-power accelerator called a neural processing unit (NPU). The NPU is tightly coupled to the processor pipeline to permit profitable acceleration even when small regions of code are transformed. Offloading approximable code regions to NPUs is faster and more energy efficient than executing the original code. For a set of diverse applications, NPU acceleration provides whole-application speedup of 2.3× and energy savings of 3.0× on average with average quality loss of at most 9.6%. NPUs form a new class of accelerators and show that significant gains in both performance and efficiency are achievable when the traditional abstraction of near-perfect accuracy is relaxed in general-purpose computing.
- Baek, W. and Chilimbi, T.M. Green: A framework for supporting energy-conscious programming using controlled approximation. In PLDI (2010). Google Scholar
Digital Library
- Chakrapani, L.N., Akgul, B.E.S., Cheemalavagu, S., Korkmaz, P., Palem, K.V., and Seshasayee, B. Ultra-efficient (embedded) SOC architectures based on probabilistic CMOS (PCMOS) technology. In DATE (2006). Google ScholarDigital Library
- Chen, T., Chen, Y., Duranton, M., Guo, Q., Hashmi, A., Lipasti, M., Nere, A., Qiu, S., Sebag, M., Temam, O., and Bench, N.N. On the broad potential application scope of hardware neural network accelerators. In IISWC (2012). Google ScholarDigital Library
- de Kruijf, M., Nomura, S., and Sankaralingam, K. Relax: An architectural framework for software recovery of hardware faults. In ISCA (2010). Google ScholarDigital Library
- Dennard, R.H., Gaensslen, F.H., Rideout, V.L., Bassous, E., and LeBlanc, A.R. Design of ion-implanted MOSFET's with very small physical dimensions. IEEE J. Solid-State Circ. 9 (Oct. 1974), 256--268.Google ScholarCross Ref
- Esmaeilzadeh, H., Saeedi, P., Araabi, B.N., Lucas, C., and Fakhraie, S.M. Neural network stream processing core (NnSP) for embedded systems. In ISCAS (2006).Google ScholarCross Ref
- Esmaeilzadeh, H., Blem, E., St. Amant, R., Sankaralingam, K., and Burger, D. Power challenges may end the multicore era. Commun. ACM 56, 2 (Feb. 2013), 93--102. Google ScholarDigital Library
- Esmaeilzadeh, H. et al. Architecture support for disciplined approximate programming. In ASPLOS (2012). Google ScholarDigital Library
- Esmaeilzadeh, H., Sampson, A., Ceze, L., and Burger, D. Neural acceleration for general-purpose approximate programs. In MICRO (2012). Google ScholarDigital Library
- Galal, S. and Horowitz, M. Energy-efficient floating-point unit design. IEEE Trans. Comput. 60, 7 (2011) 913--922. Google ScholarDigital Library
- Govindaraju, V., Ho, C.H., and Sankaralingam, K. Dynamically specialized datapaths for energy efficient computing. In HPCA (2011). Google ScholarDigital Library
- Gupta, S., Feng, S., Ansari, A., Mahlke, S., and August, D. Bundled execution of recurring traces for energy-efficient general purpose processing. In MICRO (2011). Google ScholarDigital Library
- Guzhva, A., Dolenko, S., and Persiantsev, I. Multifold acceleration of neural network computations using GPU. In ICANN (2009). Google ScholarDigital Library
- Hameed, R., Qadeer, W., Wachs, M., Azizi, O., Solomatnikov, A., Lee, B.C., Richardson, S., Kozyrakis, C., and Horowitz, M. Understanding sources of inefficiency in general-purpose chips. In ISCA (2010). Google ScholarDigital Library
- Hardavellas, N., Ferdman, M., Falsafi, B., and Ailamaki, A. Toward dark silicon in servers. IEEE Micro 31, 4 (July-Aug. 2011), 6--15. Google ScholarDigital Library
- Hashmi, A., Berry, H., Temam, O., and Lipasti, M. Automatic abstraction and fault tolerance in cortical microarchitectures. In ISCA (2011). Google ScholarDigital Library
- Joubert, A., Belhadj, B., Temam, O., and Héliot, R. Hardware spiking neurons design: Analog or digital? In IJCNN (2012).Google Scholar
- Li, S., Ahn, J.H., Strong, R.D., Brockman, J.B., Tullsen, D.M., and Jouppi, N.P. McPAT: An integrated power, area, and timing modeling framework for multicore and manycore architectures. In MICRO (2009). Google ScholarDigital Library
- Liu, S., Pattabiraman, K., Moscibroda, T., and Zorn, B.G. Flikker: Saving refresh-power in mobile devices through critical data partitioning. In ASPLOS (2011). Google ScholarDigital Library
- Muralimanohar, N., Balasubramonian, R., and Jouppi, N. Optimizing NUCA organizations and wiring alternatives for large caches with CACTI 6.0. In MICRO (2007). Google ScholarDigital Library
- Narayanan, S., Sartori, J., Kumar, R., and Jones, D.L. Scalable stochastic processors. In DATE (2010). Google ScholarDigital Library
- Patel, A., Afram, F., Chen, S., and Ghose, K. MARSx86: A full system simulator for x86 CPUs. In DAC (2011). Google ScholarDigital Library
- Putnam, A., Bennett, D., Dellinger, E., Mason, J., Sundararajan, P., and Eggers, S. CHiMPS: A high-level compilation flow for hybrid CPU-FPGA architectures. In FPGA (2008). Google ScholarDigital Library
- Razdan, R. and Smith, M.D. A high-performance microarchitecture with hardware-programmable functional units. In MICRO (1994). Google ScholarDigital Library
- Rumelhart, D.E., Hinton, G.E., and Williams, R.J. Learning internal representations by error propagation. In Parallel Distributed Processing: Explorations in the Microstructure of Cognition. D.E. Rumelhart, J.L. McClelland, and PDP Research Group, eds. Volume 1. MIT Press, 1986, 318--362. Google ScholarDigital Library
- Sampson, A., Dietl, W., Fortuna, E., Gnanapragasam, D., Ceze, L., and Grossman, D. EnerJ: Approximate data types for safe and general low-power computation. In PLDI (2011). Google ScholarDigital Library
- Sidiroglou-Douskos, S., Misailovic, S., Hoffmann, H., and Rinard, M. Managing performance vs. accuracy trade-offs with loop perforation. In FSE (2011). Google ScholarDigital Library
- Temam, O. A defect-tolerant accelerator for emerging high-performance applications. In ISCA (2012). Google ScholarDigital Library
- Venkatesh, G., Sampson, J., Goulding, N., Garcia, S., Bryksin, V., Lugo-Martinez, J., Swanson, S., and Taylor, M.B. Conservation cores: Reducing the energy of mature computations. In ASPLOS (2010). Google ScholarDigital Library
- Zhu, J. and Sutton, P. FPGA implementations of neural networks: A survey of a decade of progress. In FPL (2003).Google ScholarCross Ref
Index Terms
- Neural acceleration for general-purpose approximate programs
Recommendations
Neural Acceleration for General-Purpose Approximate Programs
This work proposes an approximate algorithmic transformation and a new class of accelerators, called neural processing units (NPUs). NPUs leverage the approximate algorithmic transformation that converts regions of code from a Von Neumann model to a ...
Neural Acceleration for General-Purpose Approximate Programs
MICRO-45: Proceedings of the 2012 45th Annual IEEE/ACM International Symposium on MicroarchitectureThis paper describes a learning-based approach to the acceleration of approximate programs. We describe the \emph{Parrot transformation}, a program transformation that selects and trains a neural network to mimic a region of imperative code. After the ...
A performance study of general-purpose applications on graphics processors using CUDA
Graphics processors (GPUs) provide a vast number of simple, data-parallel, deeply multithreaded cores and high memory bandwidths. GPU architectures are becoming increasingly programmable, offering the potential for dramatic speedups for a variety of ...
Comments