M. Annaratone, E. Arnould, T. Gross, H. T. Kung, and M. Lam. "The Warp Computer: Architecture, Implementation, and Performance". IEEE Trans. Comput., 36(12):1523--1538, 1987.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... Goodyear MPP [50, 470] used by NASA to process satellite images, the IBM GF11 [54, 342] dedicated for a whole year to a single computation involving the masses of eight fundamental particles, in an effort to validate the predictions of quantum chromo dynamics [93] and the Warp systolic array [24], which is used for the real time signal processing required in robotic navigation and vision. The use of dedicated machines is not a privilege of large government agencies and philanthropists in search of basic truths. Businesses as diverse as oil companies, car manufacturers, and wall street ....

....hardware scheduling 4.1 [92] Two level scheduler from Washington dynamic partitioning which may change at runtime to reflect load 3.4 [633] Victor from IBM Research variable partitioning by users 3.3 [523] VORX on HPC variable partitioning by users 3. 3 [303] Warp systolic array dedicated [24] Xylem on Cedar at Illinois gang scheduling on PE clusters 5.3 [187, 188] Table 7: finished. ffl There is no overhead for context switching, except that for redistributing the PEs when the load changes. The second level of scheduling within the application is assumed to require less ....

M. Annaratone, E. Arnould, T. Gross, H. T. Kung, M. Lam, O. Menzilcioglu, and J. A. Webb, "The Warp computer: architecture, implementation, and performance". IEEE Trans. Comput. C-36(12), pp. 1523--1538, Dec 1987.

....connection is critical to achieve good performance from the coprocessor. Thus, the objective of minimizing the number of accesses over the limited bandwidth connection is considered in the mapping process. This is in contrast to other approaches of building general purpose systolic computers [24, 25, 26, 27, 7]. Thus, Chapter 5 discusses design methods under constraints of fixed bandwidth and area, and objectives of yield (clock frequency) or speedup, and number of accesses. The mapping process incorporates the General Parameter Method (Chapters 2 and 4) to map partitioned dependence graphs of the given ....

....can be extended to processor arrays of arbitrary dimensions. We choose to study linear arrays because they are easier to build and program than arrays of higher dimension. Hence, several linear arrays have been implemented for specific applications as well as for general purpose computing [34, 35, 24, 25]. The organization of this chapter is as follows. Section 1.2 describes the model of algorithms targeted in this thesis, followed by a discussion of previous and related work in Section 2.1. Section 2.2 presents the definitions of parameters, followed by the constraint equations for valid systolic ....

[Article contains additional citation context not shown here]

M. Annaratone, E. Arnould, T. Gross, H. T. Kung, M. Lam, O. Menzilcioglu, and J. A. Webb, "The Warp computer: Architecture, implementation and performance," IEEE Transactions on Computers, vol. C-36, pp. 1523--1538, Dec. 1987.

....on a mesh topology such as the Intel Paragon, the Cray T3D, T3E, etc. The most famous example of such an architecture is the iWarp [27] a circuit designed by Intel Corporation in 1988 in cooperation with Carnegie Mellon University (CMU) Its architecture is derived from the original Warp project [28] started in 1984 at CMU. The iWarp chip was designed as a building block for developing powerful and programmable parallel systems for high speed signal, image and scientific computing. An iWarp system is simply an assembly of iWarp cells (iWarp chip and memory) connected by means of a dedicated ....

M. Annaratone et al., "The Warp computer: architecture, implementation, and performance," IEEE Transactions on Computers, vol. C-36(12), pp. 1523--38, 1987.

....that meets those requirements, and provide algorithms for that hardware [29] In turn, algorithmic development inspires new types of hardware support, bringing to light new computational possibilities, influencing the way we think about machine vision. Similar research efforts can be found in [32, 21, 20, 26, 1] and others, with more or less emphasis on either the architecture or the vision end of the research. A common thread in these studies is that low level vision involves more than the window based operations that dominated earlier research. What makes our work unique is that our programming model ....

M. Annaratone, E. Arnould, T. Gross, H.T. Kung, M.Lam, O. Menzilcioglu, J.A. Webb (1987): "The Warp Computer: Architecture, Implementation, and Performance," IEEE Trans. on Computers, C-36 (12).

....ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 8, NO. 8, AUGUST 1997 J: PRODUCTION TPDS 2 INPROD 100936 100936 1.DOC regularpaper97.dot KSM 19,968 04 21 97 8:36 AM 2 17 There have been numerous efforts to develop generalpurpose systolic computers in the past ten years. These include Warp and iWarp [1], Matrix 1 [8] SLAPP [6] medium grain architecture for image and signal processing [27] VATA [30] pseudosystolic linear array [22] 24] and a host of others. However, many of these designs have powerful processors with large local memories, and highbandwidth data interconnect between ....

....more processing units to the system in a modular fashion. The architectures considered in our comparison are: Systolic General Purpose Processors (SGPP) These are programmable general purpose PAs that have been built for a class of applications. Examples include iWarp (in systolic mode) [1], SLAPP [6] Matrix 1 [8] and medium grain image processing architectures [27] Partitioned Systolic Arrays (PSA) These include research efforts aimed at designing fixed size systolic arrays for solving large problems [21] 19] 15] 2] 25] 29] 17] 3] 5] 32] Systolic Arrays ....

M. Annaratone, E. Arnould, T. Gross, H.T. Kung, M. Lam, O. Menzilcioglu, and J.A. Webb, "The Warp Computer: Architecture, Implementation and Performance," IEEE Trans. Computers, vol. 36, no. 12, pp. 1,523-1,538, Dec. 1987.

.... queues for loads to communicate with computation instructions, and processor to memory queues for computation instructions to communicate with stores [2, 17] Likewise, systolic arrays use systolic queues that allow some computation instructions to communicate with others directly [3, 7]. If one adheres to a load store architecture in which all explicit communication is through the ISA visible registers (as we expect most superscalar implementations to be) then the auxiliary storage must take on the role of implicitly forwarding a register instance directly from the producer to ....

M. Annaratone et al, "The Warp Computer: Architecture, Implementation and Performance," IEEE Transactions on Computers, vol. C-36, pp. 1523-1538, December 1987.

....suitable for medium to high level vision algorithms requiring for more complex and nonuniform processing. IUA (Image Understanding Architecture) 13] and NETRA [14] are constructed as a hierarchical structure. The CMU warp processor is a systolic array machine built for image understanding tasks [15]. The adaptive parallel computer vision system (APVIS) approach is to design a hybrid system that can be performed adaptively for different types of parallelism. The APVIS can perform adaptively for different levels of processing steps in vision tasks as a cost effective system. Especially, two ....

M.Annaratone et al., "The Warp computer: Architecture, implementation, and performance," IEEE Trans. Computer, Dec. 1987.

....Symmetry. Delta Systolic arrays [41] which are 1 or 2 dimensional arrays of simple processors (cells) connected to their nearest neighbours. The cells on the edge of the array are usually connected to a general purpose machine which acts as a controller. Examples are the Warp and iWarp machines [2, 6] and several machines described in [53] Variations on the idea of systolic arrays are wavefront arrays [45] and instruction systolic arrays [58] The categories listed above are not mutually exclusive. For example, there is an overlap between vector processors and shared memory multiprocessors. ....

M. Annaratone, E. Arnould, T. Gross, H. T. Kung, M. Lam, O. Menzilcioglu and J. A. Webb, "The Warp computer: architecture, implementation and performance ", IEEE Transactions on Computers, C-36 (1987), 1523-1538.

....in a single clock cycle, but may take several cycles to complete. VLIW architectures make extensive use of compiler techniques to detect parallelism and package them into long instruction words. Examples of VLIW architectures include Trace family machines [53, 52] the Cydra 5 [108] and iWARP [7]. 2.5 Issues and Challenges In 1983, Arvind and Iannucci set forth two fundamental issues in multiprocessing. They are [10] memory latency: Most von Neumann processors are likely to idle during long memory references, and such references are unavoidable in parallel machines. cost of ....

Marco Annaratone, Emmanuel Arnould, Thomas Gross, H. T. Kung, Monica Lam, Onat Menzilicioglu, and Jon A. Webb, "The Warp computer: Architecture, implementation and performance," IEEE Transactions on Computers, 36(12):1523--1538, December 1987.

....reduced. This is important because processor meshes can be implemented much more efficiently, and with much higher interprocessor bandwidth, than binary tree arrays. Adapt has been implemented on Unix serial architectures using the serial implementation method, on the Carnegie Mellon Warp machine [2] in two separate implementations, one using row partitioning and the other using column partitioning, and on the Carnegie Mellon Nectar computer using column partitioning. We now discuss these implementations. 4.1 Serial implementation The serial implementation of Adapt is extremely ....

Annaratone, M., Arnould, E., Gross, T., Kung, H. T., Lam, M., Menzilcioglu, O. and Webb, J. A. "The Warp Computer: Architecture, Implementation and Performance". IEEE Transactions on Computers C-36, 12 (December 1987), 1523-1538.

No context found.

M. Annaratone, E. Arnould, T. Gross, H. T. Kung, and M. Lam. "The Warp Computer: Architecture, Implementation, and Performance". IEEE Trans. Comput., 36(12):1523--1538, 1987.

No context found.

Marco Annaratone, Emmanuel Arnould, Thomas Gross, H. T. Kung, Monica Lam, Onat Menzilcioglu, and Jon Webb. "The Warp Computer: Architecture, Implementation, and Performance". IEEE Transactions on Computers 36, 12 (December 1987), 1523-1538.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC