William Schottstaedt. Personal communication. Aug., 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....12.5] Thus B 0 2 P and B 1 2 P. For every i, the set B 1 is nonempty, so B 0 [ B 1 is infinite. Therefore B 0 or B 1 is infinite. Thus B 0 is an infinite polynomial time subset of A, or B 1 is an infinite polynomial time subset of A. Gasarch has found a simpler proof of our result [Gas87]. We present his proof. Proof: We take two cases. Case 1: For infinitely many values of n, when we run Algorithm A on input (0 the algorithm tells us A (0 ) or A (0 ) In this case, let B v = f0 j Algorithm A on input (0 ) or (0 ) yields A (0 ) vg; for v = 0; 1. ....

William I. Gasarch, 1987. Personal communication.

....in FQ k (n; B) partial recursive functions. Therefore, by Theorem 4.1.2(ii) there is a set B in Q(1; h) such that h is in FQ k (n; B) 4.2 The Nonspeedup Theorem Corollary 3.5.11 stated that F 2 = 2 FQ(1; B) for any oracle B. We generalize that to a result proved independently by Gasarch [Gas86]: If A is any nonrecursive set, then two parallel queries to A cannot be answered by making only one query to any set B. In addition, we show that 2 parallel queries to a nonrecursive set A cannot be answered by making only n serial queries to any set B. This is the strongest possible result, ....

William I. Gasarch, 1986. Personal communication.

....time. The single value problem for 1 pointers in Base 2ptr is PSPACE complete. Landi s dissertation shows that intra procedural pointer alias analysis is PSPACE complete if at least four levels of indirection are permitted [8] his proof can be adapted to require only two levels of indirection [10]. Landi s conclusion is that the diculty with pointer alias analysis is caused by multiple levels of indirection. This is obviously a valid conclusion, but does not get to the heart of the matter: what is the fundamental di erence between single level and multi level pointers that causes the ....

W. Landi, personal communication, June 1998.

.... Twobit and Larceny together have performance that is competitive with both Standard ML of New Jersey and commercial Common Lisp and Scheme systems [5] I have also received anecdotal evidence that Larceny is competitive with a recent Java system from Sun Microsystems on at least one GC benchmark [7]. Twobit incorporates a number of optimizations that limit heap allocation largely to that allocation performed explicitly by the user program. In particular, neither continuation frames nor environment structures are allocated on the heap unless the program requests it by capturing the ....

William Clinger. Personal communication, August 2000.

....The results in Sections 4. 1 and 4 were inspired by an open problems session led by Manfred Warmuth at our weekly Machine Learning Reading Group meeting, where he proposed the basic idea of using an approximate halving algorithm based on approximate and probabilistic 5 We note that William Chen [7] has recently extended the projective geometry argument to obtain a lower bound of Omega Gamma n p m lg k) for m n lg k. 40 counting, and also suggested the problem of learning a total order on n elements. We thank Tom Leighton for helping to improve the lower bound of Theorem 9. We also ....

William Chen. Personal communication.

....feeling of extra inertia when the head is moved. However, this would be a problem with any system of this type and as the ADL 1 s creator Bill Chernoff has pointed out, using any other commercial system such as the Fake Space Labs BOOM 1 for similar head movements would be extremely difficult [Cher93]. At US 1500 the ADL 1 represents low cost head tracking with very low latency. A device similar in concept to the ADL 1 is the Immersion Probe (Immersion Human Interface Corp. California USA) Rather than a head band, this smaller device features a stylus which can be held in the hand and used ....

Chernoff, William. Personal communication, June 1993.

....circuit breakers which contain electronics to detect arcing (so called arc fault detection technology) and which trip on the occurrence of arcing. One such has been developed by ETA Technologies in Germany (which amongst other things provides circuit breakers for European military aircraft) [14, 11]. Other detectors have been developed by the company Square D for home use [15] and Eaton Corp has patented arc fault detection technology suitable for aviation [10] The issues of faulty and deteriorated wiring have attracted the attention of US Congressional committees. 3.2.2 Summary of the ....

William Sells. Personal communication, September 21, 2000.

....time. The single value problem for 1 pointers in Base 2ptr is PSPACE complete. Landi s dissertation shows that intra procedural pointer alias analysis is PSPACE complete if at least four levels of indirection are permitted [7] his proof can be adapted to require only two levels of indirection [9]. Landi s conclusion is that the diculty with pointer alias analysis is caused by multiple levels of indirection. This is obviously a valid conclusion, but does not get to the heart of the matter: what is the fundamental di erence between single level and multilevel pointers that causes the ....

W. Landi, personal communication, June 1998.

....amount of time, one would not necessarily expect the error for the best possible model to be any better than for other models known to be worse. The behavior of chaotic systems can only be predicted locally in time, so an error function which tries to do otherwise is not likely to be a good one[39]. 2.4.3 Phase Fields Since we have restricted ourselves to autonomous systems, which do not depend on time, we can avoid the problems of the naive method by using an error function based on the notion of a phase portrait or phase space diagram. We have implemented the method for systems with two ....

....there is no mutation) and the algorithm will have converged. This picture of the GA also allows us to consider the entropy of the system as it evolves. One way of measuring entropy would be to impose a grid on the search space in order to break it up into a finite number of l dimensional boxes[39]. If the initial bounding box were a hypercube with sides of length b, and the grid cells were hypercubes with sides of length g, 6 there would of course be c = b=g) l grid cells. If we index the cells with the variable k, then we can define the concentration in each cell thus: p k = n k =n; ....

William K. Wootters. Personal Communication.

....methods were published prior to Recursive Contour Meshing in 1996 (Weigle and Banks 1996) Recursive Contour Meshing is the technique presented in the following chapter. Lorensen has developed an unpublished extension to Marching Cubes allowing the contouring of surfaces in four dimensional spaces (Lorensen 1996). Lorensen s technique will be included in the Visualization ToolKit (Schroeder, Martin, and Lorensen 1996) an object oriented visualization development package encompassing many of the techniques mentioned in this chapter as well as techniques for other types of 18 visualization. Lorensen s ....

....Meshing is the technique presented in the following chapter. Lorensen has developed an unpublished extension to Marching Cubes allowing the contouring of surfaces in four dimensional spaces (Lorensen 1996) Lorensen s technique will be included in the Visualization ToolKit (Schroeder, Martin, and Lorensen 1996), an object oriented visualization development package encompassing many of the techniques mentioned in this chapter as well as techniques for other types of 18 visualization. Lorensen s extension uses a new, procedurally generated table with 2 16 entries (there are 16 vertices in a 4 cell) It ....

Lorensen, William. 1996. Personal communication.

....results. 1 As described in the next chapter, predictor variables typically combine in a non linear fashion. Most predictor variables are clearly not independent of at least some others. 94 5. 1 What Level of Abstraction ( Because all previous related work in confidence annotation [12] [6] [51] 20] 5] 41] 8] 10] has been done at the word level, and because it is very simple to implement, the baseline metric is defined and tested on a word by word basis. All previous related work modeled only one error class, incorrect. I extend the technique by tagging words as either correct ....

William Byrne. Personal Communication. unpublished, 1996.

....informa Interprocedural Pointer Alias Analysis Delta 53 tion. Among the metrics they report is the number of thru deref assigns, which correspond to the write metrics reported in Table II. However, since their results included compiler introduced temporaries in their thru deref count [Landi 1997], a direct comparison is not possible. Stocks et al. 1997] use the same metric without including temporaries. Using the flow sensitive context sensitive analysis of Landi and Ryder [1992] the average number of objects ranges from 1.0 to 2.0 on the benchmarks we have in common (allroots, lex315, ....

....counted in the same manner in all cases. For example, in [Landi and Ryder 1992] invisible objects aliased to a formal parameter are represented as a single object, and statements such as cfree(TP) located in allroots are treated as modifying the structure freed, and thus as a pointer dereference [Landi 1997], where our analysis does not. On the three programs (allroots, lex315, simulator) in which our analysis reports the same, or close to the same, number of writes as thru derefs, our precision is identical to that reported in [Stocks et al. 1997] The relative precision of the flow insensitive ....

Landi, W. 1997. Personal communication.

....of over 73 and reduces analysis time by an average of 280 [13] 4 Indirect calls can result in several potential target functions. argc and argv. This function also initialized the iob array, used for standard I O. The added function is similar to the one added by Ruf [29, 30] and Landi et al. [19, 17]. Explicit and implicit initializations of global variables are automatically modeled as assignment statements in the dummy main function. Array initializations are expanded into an assignment for each array component. 4. CLIENT ANALYSES This section summarizes the client analyses used in this ....

W. Landi. Personal communication, Oct. 1997.

....pointer alias algorithm with limited context sensitive information. Among the metrics they report is the number of thru deref assigns, which corresponds to the write metric reported in Figure 14. However, since their results included compiler introduced temporaries in their thru deref count [Landi 1997], a direct comparison is not possible. Stocks et al. 1998] and Landi et al. 1998] use the same metric without including temporaries. Using the flow sensitive context sensitive analysis of Landi and Ryder [1992] the average number of objects ranges from 1.0 to 2.0 on the benchmarks we have in ....

....dereferences may not be counted in the same manner in all cases. For example, statements such as cfree(TP) located ACM Transactions on Programming Languages, Vol. 21, No. 4. 36 Delta Michael Hind et al. in allroots are treated as modifying the structure freed, and thus as a pointer dereference [Landi 1997], where our analysis does not. On the three programs (allroots, lex315, simulator) in which our analysis reports the same, or close to the same, number of writes as thru derefs, our precision is either identical or close to that reported in Stocks et al. 1998] The relative precision of the ....

Landi, W. 1997. Personal communication.

....to the main function, a dummy main function is added, which then calls the benchmark s main function, thus simulating the effects of argc and argv. This function also initializes the iob array, used for standard I O. The added function is similar to the one added by Ruf [35, 37] and Landi et al. [27, 25]. Initializations of global variables are automatically modeled as assignment statements in the dummy main function. 4 Results This section provides empirical evidence of the efficiency and precision for the six algorithms discussed in Section 2. The results were collected on a 333MHz IBM ....

....pointer alias algorithm with limited context sensitive information. Among the metrics they report is the number of thru deref assigns, which corresponds to the write metrics reported in Fig. 6. However, since their results include compiler introduced temporaries in their thru deref count [25], a direct comparison is not possible. Stocks et al. 47] use the same metric without including temporaries for the flow sensitive context sensitive analysis of Landi and Ryder [26] They report the average number of objects ranges from 1.0 to 2.0 on the nine common benchmarks. On these ....

[Article contains additional citation context not shown here]

William Landi. Personal communication, October 1997.

....number of aliased expressions per node can grow quite large: the make.c program serves as a worst case in [Lan92] with 675 may alias relations per node on average, and one node containing some 2000 relations. Research is underway to determine how relations translate into effects on storage names [Lan93] but currently the impact of so many alias relations on SSA form cannot be quantified. Our experiments show that the amount of alias information represented in SSA form can far exceed the amount of such information necessary for solving a given optimization problem: our algorithm typically ....

William A. Landi. personal communication,

....Z = a X) Y The values of X and Y may vary widely with scaling factor a ensuring that overflow and underflow do not occur. If instead the expression is evaluated as so correctly rounded conversion was not required over the entire range of floating point values due to implementation difficulties [59]. 8 The 80 bit double extended format is widely implemented in hardware. Another extant IEEE double extended format, 128 bit quad, has instruction level support in some architectures. However, on existing processors implementing such architectures, the actual quad operation trap to software. 9 ....

....can result in a normal number. 8 Z = a (X Y) exceptions designed to be avoided may occur. Alternatively, a X) may be an exact calculation, with the only rounding error introduced by the second multiplication; using a different evaluation order can increase the calculation s rounding error [59]. For predictable floating point behavior, a language must respect parentheses and not treat computations that are mathematically equivalent on infinitely precise real numbers as computationally equivalent on limited precision floating point numbers. For example, while multiplication of real ....

[Article contains additional citation context not shown here]

William Kahan, personal communication.

.... impossible to independently delete any of the individual objects, or the row (if the row were deleted, all the objects embedded in it would also be deleted; an embedded object cannot be deleted without deleting the row containing it, which in this case would also delete the other embedded objects) [KCS94]. These sorts of anomalies suggest that additional work will be necessary to fully integrate objects into SQL in a way that seems natural to users of other object oriented programming languages and database systems. Work in this area is currently in progress. For example, Sha94] is a proposal ....

William Kelley, Amelia Carlson, Phil Shaw, personal communication, August 1994.

....number of aliased expressions per node can grow quite large: the make.c program serves as a worst case in [Lan92] with 675 may alias relations per node on average, and one node containing some 2000 relations. Research is underway to determine how relations translate into effects on storage names [Lan93] but currently the impact of so many alias relations on SSA form cannot be quantified. Our experiments show that the amount of alias information represented in SSA form can far exceed the amount of such information necessary for solving a given optimization problem: our algorithm typically stops ....

William A. Landi. personal communication, 1993.

....exponent bits of the floating point argument. If scalb is defined as a bit manipulation operation, the result should not depend on the dynamic rounding mode and round to nearest should always be used. While the authors of the standard intend for scalb to act as an arithmetic, not bitwise operation [16], any ambiguity could be remedied by replacing the incongruous specification of scalb with a slightly longer specification that stated the desired behavior. Since the standard does not discuss the maximum allowed scaling range, the overflow behavior of scalb is also not fully specified. Depending ....

....3.5. scalb public static float scalb(double value, int scale exponent) throws OverflowExeption, UnderflowException, InexactException admits none yields overflow, underflow, inexact The intention of including scalb as an IEEE recommended function was to provide a quick way to scale entire arrays [16]. One technique to avoid floating point exceptions when calculating on an array is to first scan the array to determine a scaling factor that prevents floating point exceptions, scale the array elements, and finally perform the calculation in question. In this application of scalb, overflow does ....

[Article contains additional citation context not shown here]

William Kahan, personal communication.

....Thus B 0 2 P and B 1 2 P. For every i, the set B i 0 [ B i 1 is nonempty, so B 0 [ B 1 is infinite. Therefore B 0 or B 1 is infinite. Thus B 0 is an infinite polynomial time subset of A, or B 1 is an infinite polynomial time subset of A. Gasarch has found a simpler proof of our result [Gas87]. We present his proof. Proof: We take two cases. Case 1: For infinitely many values of n, when we run Algorithm A on input (0 n ; 0 n 1 ) the algorithm tells us A (0 n ) or A (0 n 1 ) In this case, let B v = f0 n j Algorithm A on input (0 n Gamma1 ; 0 n ) or (0 n ; 0 n 1 ....

William I. Gasarch, 1987. Personal communication.

....functions. Therefore, by Theorem 4.1.2(ii) there is a set B in Q(1; h) such that h is in FQ k (n; B) CHAPTER 4. NONRECURSIVE ORACLES 64 4.2 The Nonspeedup Theorem Corollary 3.5.11 stated that F K 2 = 2 FQ(1; B) for any oracle B. We generalize that to a result proved independently by Gasarch [Gas86]: If A is any nonrecursive set, then two parallel queries to A cannot be answered by making only one query to any set B. In addition, we show that 2 n parallel queries to a nonrecursive set A cannot be answered by making only n serial queries to any set B. This is the strongest possible result, ....

William I. Gasarch, 1986. Personal communication.

....contaminated water supplies, or through sneezing, the dispersal of the parasites can be quantified by the distance of dispersal from the infected host. The question can now be asked, what is the effect of the distance of dispersal on the evolution of virulence in parasites The Hypothesis Hamilton (1992) suggested that we should expect parasites that disperse over short distances to evolve to be less virulent than parasites that disperse over long distances. This follows from the continuum from vertical to horizontal transmission. Parasites that reproduce locally will be likely to share their ....

William D. Hamilton, 1992. Personal communication. Zoology Dept., South Park Rd., Oxford, OX1 3PS, UK.

No context found.

William Schottstaedt. Personal communication. Aug., 1991.

No context found.

William I Gasarch, 1996. Personal communication.

First 50 documents

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