B. Calder, D. Grunwald, and B. Zorn. Quantifying behavioural differences between C and C ++ programs. Journal of Programming Languages, 2:313-- 351, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....would eliminate pipeline stalls, as the target of the call is unambiguously known. A related empirical study of behavioral differences between C and C programs reveals that C functions are usually small (in terms of lines of code) and C programs contain more calls than C programs [CGZ95]. The higher calling frequency usually degrades analysis performance because of the inherent difficulty of interprocedural analysis contrasted with that of traditional intraprocedural analysis. Therefore, these findings suggest that inlining a uniquely resolved function may be a crucial ....

B. Calder, D. Grunwald, and B. Zorn. Quantifying behavioural differences between C and C ++ programs. Journal of Programming Languages, 2:313-- 351, 1995.

....empirical results with modular points to analysis are encouraging; however, this is a proof of concept implementation and there is scope for optimization. Table 1 contains some characteristics of the thirteen C programs we have analyzed. These are some of the benchmarks used in [PR96, BS96, CGZ95] 11 The columns lines, ICFG nodes, methods, virtual calls, SCC s and Max SCC respectively show the number of lines of code, ICFG nodes, methods, dynamically dispatched call sites, nodes in SCCDAG and methods in the maximum sized SCC for each program. Table 2 contains the timings using a ....

B. Calder, D. Grunwald, and B. Zorn. Quantifying behavioural differences between C and C ++ programs. Journal of Programming Languages, 2:313--351, 1995.

....subset of C . We have completed the implementation of the algorithm for single level types and the implementation for the general types is in progress. Table 1 contains data about call graph decomposition of 9 C programs obtained from Hemant Pande [PR96] David Bacon [BS96] and Ben Zorn [CGZ95] We used hierarchy analysis for constructing the initial call graphs. Table 1 shows that the call graphs of these programs have a large number of components and the cycles in these call graphs are of small size. This is encouraging evidence for effectiveness of analysis using abstract values for ....

B. Calder, D. Grunwald, and B. Zorn. Quantifying behavioural differences between C and C ++ programs. Journal of Programming Languages, 2:313-- 351, 1995.

....the execution stream (e.g. trace caches) need to pay more attention to indirect calls than would be suggested by the SPEC95 benchmark suite. The increased number of indirect calls in desktop applications may imply that most of the desktop applications are written in an object oriented style. Calder et al. 94] observed that on average, C programs tend to have more indirect calls because C programs tend to use dynamic dispatch (i.e. virtual function calls) to take the place of Application Branches Calls CB (T) DB IB DC IC DLL acrord32 89.3 (58) 9.9 0.7 82.3 10.7 7.0 netscape 91.2 ....

....target for the entire duration of a program run. conditional logic (i.e. if then else or switch statements) For our benchmark programs, netscape, photoshp, and most of powerpnt are written in C , while acrord32 is written in object oriented style C. Winword is written mostly in C. However, Calder et al. 94] also found that C programs tend to have smaller function sizes and longer basic block sizes. None of the desktop applications have particularly longer basic blocks and only powerpnt has significantly smaller functions. To determine the impact of the larger executable size on the application ....

Calder, B., Grunwald, D., and Zorn, B. Quantifying behavioural differences between C and C++ programs. Journal of Programming Languages, 2(4):313--351, December 1994.

....be be unwise to inline x, because then f 100 would be evaluated twice instead of once. Informally, we say that a transformation is W safe if it guarantees not to duplicate work. 3 This difference may soon decrease as the increased use of object oriented languages leads to finer gained procedures (Calder, Grunwald Zorn [1994]) In the case of WHNFs, the trade off is simply between code size and the benefit of inlining. Atoms and constructor applications are easy: they always small enough to inline. Recall that constructor applications must have atomic arguments. Functions, in contrast, can be large, so the effect ....

B Calder, D Grunwald & B Zorn [1994], "Quantifying behavioural differences between C and C++ programs," Journal of Programming Languages 2(4), Dec 1994, .

....are constrained to be atomic. Dead code elimination discards let bindings that are no longer used; this usually occurs when all occurrences of a variable have been inlined. 2 This difference may soon decrease as the increased use of object oriented languages leads to finer gained procedures (Calder, Grunwald Zorn [1994]) Beta reduction replaces ( x E) A by E[A x] An analogous transformation deals with type applications. Beta reduction is particularly simple in our setting. Since the argument A is bound to be atomic, there is no risk of duplicating a redex, and we can simply replace x by A throughout E. ....

B Calder, D Grunwald & B Zorn [Dec 1994], "Quantifying behavioural differences between C and C++ programs," Journal of Programming Languages 2, 313--351.

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