Thinking Machines Corporation. Paris Reference Manual. Cambridge, MA, Feb. 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Thus, the CM 2 has configurations ranging between 8K and 64K, inclusive, bitserial processors. In this analysis, we view each Sprint node as a single 32 bit processing node. The CM 2 programming model of C uses a canonical data layout similar to that on the CM 5 described in Section 2.2. 1 ([39], 38] 44] 4] The only difference on the CM 2 and CM 5 is that the Sprint node, or major grid, portion of each data element address is in reflected binary gray code, insuring that nearest neighbor communications are at most one hop away in the hypercube interconnection network. We now ....

Thinking Machines Corporation, Cambridge, MA. Paris Reference Manual, Version 6.0 edition, June 1991.

....a shortest possible clique tree. 2.3. The Connection Machine. 2.3.1. Architecture. The Connection Machine (model CM 2) is a local memory, SIMD parallel computer. The description we present here corresponds to the machine architecture presented by the assembly language instruction set Paris [34]. We programmed the CM in lisp, which is compiled into Paris. A full sized CM has 2 = 65,536 processors, each of which can directly access 65,536 bits of memory. Since this work was done, larger memories have become available. The processors are connected by a communication network called ....

Thinking Machines Corporation, Cambridge, Massachusetts, Paris reference manual, version 5.0.

....University of Illinois, Urbana. Cedar FORTRAN permitted the expression of both SIMD and MIMD parallelism. 2. 2 Distributed Memory SIMD Languages The CM 2 machine [41] from Thinking Machines was a distributed memory array processor and its assembly language, called Paris (Parallel Instruction Set) [40], had FORTRAN, C and Lisp interfaces that permit programmers to write high level language programs with Paris commands embedded in them. The resulting languages were called FORTRAN Paris, C Paris and Lisp Paris, and they are examples of distributed memory SIMD languages. The programming model of ....

Thinking Machines Corp., Cambridge, MA 02139. Paris Reference Manual, 1991.

....and the portable MPI implementation. For all four implementations we discuss the particular solutions chosen to the CVL porting decisions outlined in Section 2, and their impact on development time and final performance. 3. 1 CM 2 CVL The CM 2 implementation of CVL is written in C and Paris [19], a parallel instruction set for the CM 2 s SIMD processing array. The main alternative was CM Fortran [21] but it could not be used because at the time it did not have the ability to alias arrays (for example, storing an array of floating point numbers where an array of integers used to be) and ....

Thinking Machines Corporation, Cambridge, MA. Paris Reference Manual, February 1991. Version 6.0.

....Thus, the CM 2 has configurations ranging between 8K and 64K, inclusive, bitserial processors. In this analysis, we view each Sprint node as a single 32 bit processing node. The CM 2 programming model of C uses a canonical data layout similar to that on the CM 5 described in Section 2.2. 1 ([39], 38] 44] 4] The only difference on the CM 2 and CM 5 is that the Sprint node, or major grid, portion of each data element address is in reflected binary gray code, insuring that nearest neighbor communications are at most one hop away in the hypercube interconnection network. We now ....

Thinking Machines Corporation, Cambridge, MA. Paris Reference Manual, Version 6.0 edition, June 1991.

....prefix operation on each segment, starting from zero on each segment boundary. The segmented versions of the instructions are critical for the implementation of nested parallelism (see Section 3. 1) The notion of segments also appears in some of the library routines of the Connection Machines CM 2 [57] and CM 5 [60] and has been adopted in the PREFIX and SUFFIX operations of High Performance Fortran [33] The internal VCODE representation of the segment descriptor is machine dependent: our serial implementation uses a sequence of the lengths of each segment, while our implementations on the ....

....only uses one processor, we have implemented prototypes of multiprocessor vector operations [15] and we are working on a complete multiprocessor library using the Cray microtasking facilities [25] 3.5. 2 CM 2 CVL The implementation of CVL for the CM 2 is built on top of the Paris instruction set [57]. As a machinespecific language, Paris has several nice features for implementing CVL: a clean interface to C, a model of the CM 2 as a collection of virtual processors, and direct support for scan, reduce, and other vector operations. It also has three significant drawbacks, however. First, it ....

[Article contains additional citation context not shown here]

Thinking Machines Corporation, Cambridge, MA. Paris Reference Manual, February 1991.

....the portable MPI implementation. For each implementation, we discuss the particular solutions chosen to the CVL porting decisions outlined in Section 2, and their impact on development time and final performance. 3. 1 CM 2 CVL The implementation of CVL for the TMC CM 2 is written in C and Paris [18], a parallel instruction set for the CM 2 s SIMD processing array. CM Fortran [20] was not used because at the time it did not have the ability to alias subsections of arrays (for example, storing a vector of integers into the middle of where a vector of floating point numbers used to be) and ....

Thinking Machines Corporation, Cambridge, MA. Paris Reference Manual, February 1991. Version 6.0.

....parallel prefix operation on each segment, starting from zero on each segment boundary. The segmented versions of the instructions are critical for the implementation of nested parallelism. The notion of segments also appears in some of the library routines of the Connection Machines CM 2 and CM 5 [43, 42] and has been adopted in the PREFIX and SUFFIX operations of High Performance Fortran. NESL compiler: The Nesl compiler translates Nesl code into Vcode. The most important compilation step is the use of a technique called flattening nested parallelism [13, 7] This process converts nested ....

....on each step. In our CM Fortran implementation, as these elements get packed to the bottom of an array, they become more imbalanced across the processors. Although 1 On the CM 2 there is an additional important source of inefficiency: CM 2 CVL is built on top of the Paris instruction set [43]. Although working with Paris has many advantages, it forces use of the older fieldwise representation of data, instead of the more efficient slicewise representation generated by the CM Fortran compiler. 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 Time per element (usec) 8K 32K 128K 512K 2M 8M ....

Thinking Machines Corporation, Cambridge, MA. Paris Reference Manual, Feb. 1991.

....by ones in the match matrix, shown in Figure 5, produce the maximum match with the minimum cost for the corresponding weight matrix. 11 22 4 76 32 95 2 8 100 55 6 333 1 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0 25 0 88 7 Weight Matrix Match Matrix Figure 5: Maximum match with the minimum cost A C Paris [11] implementation of the algorithm was adapted for the case of Single Instruction Multiple Data (SIMD) machines. 2 A critical component of the algorithm is the sorting routine. The strategy of this routine is to sort columns by the row position of matches, where the first row equals 0, the second ....

Thinking Machines Corp., Cambridge, Massachusetts, Paris Reference Manual, version 6.0 ed., Feb. 1991.

....UC constructs into simpler ones; a few rules have been implemented to perform simple optimizations like loop fusion. The third phase of the compiler is the architecture specific code generation phase, where the intermediate language program is translated to a target language, such as: C Paris[32] or C[25] for synchronous architectures like the CM 2) ANSI C (for the sequential implementation) or Maisie[3] a message passing language, or PVM[5] a message passing library (for implementation on asynchronous architectures and workstation networks) For the CM 2 implementation, the UC ....

Thinking Machines Corporation, Cambridge, MA. Paris Reference Manual, 6.0 edition, Febuary 1991.

No context found.

Thinking Machines Corporation. Paris Reference Manual. Cambridge, MA, Feb. 1991.

No context found.

Thinking Machines Corporation, Cambridge, Massachusetts. Paris Reference Manual, February 1989.

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