Giuseppe Attardi and Tito Flagella. Customising Object Allocation. In ECOOP'94, volume 821, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....It assumes no knowledge of register or stack area layouts, but it does assume that all pointers in heapallocated data can be found accurately thanks to the registration of all internal roots [Bar90] by the developer. Other experiments around garbage collector customization were carried out in [AF94] and [AFI95] Their Customizable Memory Management (CMM) allows users to customize object management by specifying at each object allocation which policy to adopt for its storage, and by providing the optimal traversal routines for each type. The major interest of the latter kind of manual ....

Giuseppe Attardi and Tito Flagella. Customising Object Allocation. In ECOOP'94, volume 821, 1994.

....machines. This is the reason why the prototype parallel version for the Sun Sparc multiprocessor architecture gives no significant speedup [16] Therefore, it is planned to develop a parallel mostly copying algorithm. The mostly copying garbage collection developed by Bartlett [4, 5] see also [1]) is a variant of the classical stop and copy algorithm, where all accessible objects on the heap are copied into a new space. The advantages of this strategy are that it results in memory compaction and its running time is proportional to the amount of accessible memory. However, at the beginning ....

....about 1 2 log p multiplications for an average p. Therefore we will focus on the squaring operation. The following MuPAD function computes the square of a polynomial a with any kind of coefficients using a parallel for loop. square : proc(a) local d,i,x,n; begin d: degree(a) x: op(a,[2,1]) poly(plus( for n from 0 to 2 d parallel INRIA 7 n seq. 1 pr. 2 pr. 4 pr. 8 pr. 16 pr. 32 pr. 500 60s 124s 121s 91s 73s 49s 34s 1000 672s 851s 534s 445s 297s 191s 141s Figure 2: Time needed for squaring a polynomial of degree n with coefficients of n bits. private i,s; s: if n mod 2=0 ....

Attardi, G., and Flagella, T. Customising object allocation. In ECOOP '94 -- Object-Oriented Programming (Berlin, July 1994), M. Tokoro and R. Pareschi, Eds., vol. 821 of Lecture Notes in Computer Science, Springer-Verlag, pp. 299--319.

....machines. This is the reason why the prototype parallel version for the Sun Sparc multiprocessor architecture gives no significant speedup [16] Therefore, it is planned to develop a parallel mostly copying algorithm. The mostly copying garbage collection developed by Bartlett [4, 5] see also [1]) is a variant of the classical stop and copy algorithm, where all accessible objects on the heap are copied into a new space. The advantages of this strategy are that it results in memory compaction and its running time is proportional to the amount of accessible memory. However, at the beginning ....

....about 1 2 log p multiplications for an average p. Therefore we will focus on the squaring operation. The following MuPAD function computes the square of a polynomial a with any kind of coefficients using a parallel for loop. square : proc(a) local d,i,x,n; begin d: degree(a) x: op(a,[2,1]) poly(plus( for n from 0 to 2 d parallel private i,s; s: if n mod 2=0 then coeff(a,n 2)2 else 0 endif) for i from max(0,n d) to min(d,iquo(n 1,2) do s: s 2 coeff(a,i) coeff(a,n i) endfor; s xn endfor) op(a,2. 3) endproc: Each parallel task computes one term snx n of a 2 ; the parallel ....

Attardi, G., and Flagella, T. Customising object allocation. In ECOOP '94 -- Object-Oriented Programming (Berlin, July 1994), M. Tokoro and R. Pareschi, Eds., vol. 821 of Lecture Notes in Computer Science, Springer-Verlag, pp. 299--319.

....are stored on a given page) BiBoP obviates the need for a header word per object but it also makes it difficult to associate state with each object, and to allocate quickly. To the best of our knowledge since Bartlett s original work there has only been one other mostly copying collector: CMM [4, 5, 6] (for C ) CMM improves on Bartlett s algorithm in that it associates a bitmap with each page which allows it to mark individual objects, thus achieving less false retention. However, CMM did not make any fundamental changes and therefore still requires two passes over the live data. Some data ....

....each page which allows it to mark individual objects, thus achieving less false retention. However, CMM did not make any fundamental changes and therefore still requires two passes over the live data. Some data comparing CMM to Bartlett s collector and to BWC in the context of C is published in [4]. 4 Comparison of Overall Costs for ML To compare MCC to BWC, we added a C back end to the TIL compiler [30, 29] and targeted it to use both collectors. We chose TIL because we were familiar with it, and because we were able to generate fairly natural C code. In particular, the generated code ....

G. Attardi and T. Flagella. Customising object allocation. In M. Tokoro and R. Pareschi, editors, Object Oriented Programming, Proceedings of the 8th ECOOP, Lecture Notes in Computer Science 821, pages 320--343. Berlin: Springer-Verlag, 1994.

....arrays of Item can be created normally, for instance: Item anArrayOfItems = new Item[20] The constructor for class Item with no argument will be called for each Item in the array. The operator [ can be used to access such arrays, e. g: anArrayOfItems[i] print( Item anItem = anArrayOfItems[3]; 8 Heap Classes To manage a heap one normally has to maintain the set of roots for the objects in the heap, manage the pages where objects are allocated, and implement the memory allocation and recovery primitives. A suitable encapsulation for these functionalities is provided by the Heap ....

....to specific applications or to perform debugging and profiling. Customisation of the allocator is achieved by supplying specialised functions to get raw memory and to manage allocation. Of course Vo s allocator does not perform automatic memory reclamation. The basic design of CMM was presented in [2, 3]: the major differences with the present work are in a new version of the collector algorithm to limit the cases of recursion, statistics on its benefits, support for arrays of CmmObjects, support for virtual base classes and references in classes, the extension of the framework with the ....

G. Attardi and T. Flagella, `Customising object allocation', in M. Tokoro and R. Pareschi (eds.) Object-Oriented Programming, Proceedings of the 8th ECOOP, Lecture Notes in Computer Science 821. Berlin:Springer-Verlag, 320--343 (1994).

No context found.

Attardi, G., Flagella, T. (1994b). Customising object allocation. in M. Tokoro and R. Pareschi (eds.) Object-Oriented Programming, Proceedings of the 8th ECOOP, Lecture Notes in Computer Science 821. Berlin:Springer-Verlag, 320--343.

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