Scalar processors

Almost all applications programming in chemistry, and structural chemistry in particular, is performed in the FORTRAN language and the molecular mechanics calculations for which this hardware/software design exercise was undertaken is no exception. There are two problems which must be solved in order to build a FORTRAN microcomputer system with good scalar and array computational performance and these are firstly, the design of an efficient scalar processor with a transparent FORTRAN interface to the hardware and secondly, the design of an efficient AFPP,... 

Performance of the (MC)SOR algorithm on a range of computers. A single processor CPU times (in seconds) required to perform 100 (MC)SOR iterations on a 294 x 298 grid are given with the corresponding MFLOPS rates in parentheses. Note that the MCSOR scheme should not be used for calculations performed on a single scalar processor. 

We note, that also [HIKB96] work with a logic of uninterpreted functions, but use the techniques of reducing property verification to verifying the property for finite-instantiations of the model, which in general (and in particular for the application to super-scalar processors discussed in this paper) is not complete. 

Mesh Number of vertices Number of cells CPU-time ratio (Cray Y-MP) algorithm section III.D/ algorithm section IV.B.l CPU-time ratio for scalar processor... 

One advantage of the bond fluctuation method is that it works well on a variety of platforms, including vector computers, massively parallel computers, and super scalar processors. For a dense polymer melt in three dimensions, the vectorized code gives ca. 1.7 x 10 attempted moves per second for one processor on the Cray YMP for a volume fractions of (f) = 0.41. On massively parallel computers, the program has only been run to date in such a way that each processor ran one independent system. While the method should be parallelizable across multiple processors, this has not been done yet. 

Mathematical models require computation to secure concrete predictions. Successes in relatively simple cases spurs interest in more complex situations. Somewhat specialized computer hardware and software have emerged in response to these demands. Examples are the high-end processors with vector architecture, such as the Cray series, the CDC Cyber 205, and the recently announced IBM 3090 with vector attachment. When a computation can effectively utilize vector architecture, such machines will out-perform even the most powerful conventional scalar machine by a substantial margin. Such performance has given rise to the term supercomputer. ... 

Perhaps the first question to be considered in contemplating the use of a vector processor in Quantum Chemistry (QC) is just how much advantage is obtainable with the minimum amount of effort i.e. by simply implementing software from a scalar machine with little or no modification. The answer to this question is readily obtainable by benchmarking the machine against some standard on a variety of widely used QC packages. Such an exercise would shed... 

With the advent of vector processors over the last ten years, the vector computer has become the most efficient and in some instances the only affordable way to solve certain computational problems. One such computer, the Texas Instruments Advanced Scientific Computer (ASC), has been used extensively at the Naval Research Laboratory to model atmospheric and combustion processes, dynamics of laser implosions, and other plasma physics problems. Furthermore, vectorization is achieved in these programs using standard Fortran. This paper will describe some of the hardware and software differences which distinguish the ASC from the more conventional scalar computer and review some of the fundamental principles behind vector program design. 

This article reviews some of the progress made in using parallel processor systems to study macromolecules. After an initial introduction to the key concepts required to understand parallelisation, the main part of the article focuses on molecular dynamics. It is shown that simple replicated data methods can be used to carry out molecular dynamics effectively, without the need for major changes from the approach used in scalar codes. Domain decomposition methods are then introduced as a path toward reducing inter-processor communication costs further to produce truly scalable simulation algorithms. Finally, some of the methods available for carrying out parallel Monte Carlo simulations are discussed. 

One chain can then be accepted based on its Rosenbluth weight. This strategy of multiple chain growth has been tried on a scalar machine, and has been shown to improve sampling efficiency within configuration bias MC [37]. It should also translate to parallel machines because a reasonable amount of computational time is required to generate a chain (or multiple chains) on each processor, relative to the inter-processor communication required at the end of chain regrowth. 

The time required for the 4-index transformation depends on the number of 2-electron FSGO integrals, and on the number of molecular orbitals M utilized in the Cl excitation process. Generally, M has been chosen to be between 20 to 50 molecular orbitals in most previous studies. It should be noted that the time for a 4-index transformation may be reduced by at least an order of magnitude using a vector processor, relative to the time required for scalar execution with the same machine. 

The single-processor vector machine will have only one of the vector processors depicted, and the system may even have its scalar floating-point capability shared with the vector processor. The early vector processors indeed possessed only one VPU, while present-day models can house up to 64 feeding on the same shared memory. It may be noted that the VPU in Fig. 1 does not show a cache. The majority of vector processors do not employ a cache anymore. In many cases the vector unit cannot take advantage of it and the execution speed may even be unfavorably affected because of frequent cache overflow. 

One class of mainframe now available is the vector processor, which is optimised for operations on consecutive elements of arrays, which, in FORTRAN, often means DO loops. Might not these processors provide a speeding up of the DO loops The answer is no, because each DO loop only involves perhaps one addition to the abundances. But nevertheless the speed of these processors is often such that they are to be preferred to normal, scalar machines. 

---