R. Cohn and P. G. Lowney, Hot Cold Optimization of Large Windows/NT Applications, MICRO-29, pp. 80-89, Paris, France, December 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....at a time. Therefore, the compiler does not have the opportunity to optimize the full program as a whole, even when performing sophisticated inter procedural transformations. A second reason for the interest in binary optimization has been the emergence of pro le directed compilation techniques [26, 9, 12, 20]. The compiler can use to great advantage the pro ling information. However, the same problem of separate compilation plagues the use of pro le feedback: large projects will be forced to re build every le to take advantage of the pro ling information. Furthermore, the pro le guided ....

....body of work on interprocedural data ow analyses design to analyze only part, but not all, of a program (see, for example, 3, 6, 16, 29] although only some of them use pro le information to guide their decisions. This pro le information is, however, widely used when performing optimizations [26, 9, 12, 20]. On the other hand, while speculation has been commonly used in the compiler world for optimizing programs [21, 17, 22, 20, 27] as far as we know, this is the rst attempt to introduce unsafe speculations into a data ow analysis algorithm. Finally, di erent alternatives for collecting pro le ....

R. Cohn and P. G. Lowney. Hot cold optimization of large Windows/NT applications. In Proceedings of the 29th Annual International Symposium on Microarchitecture, pages 80-89, Paris, France, Dec. 2-4, 1996. ACM Press.

....body of work on interprocedural dataflow analyses design to analyze only part, but not all, of a program (see, for example, 3, 5, 13, 26] although only some of them use profile information to guide their decisions. This profile information is, however, widely used when performing optimizations [24, 7, 10, 18]. On the other hand, while speculation has been commonly used in the compiler world for optimizing programs [19, 15, 20, 18, 25] as far as we know, this is the first attempt to introduce unsafe speculations into a dataflow analysis algorithm. 6. Summary and future directions greatly from ....

R. Cohn and P. G. Lowney. Hot cold optimization of large Windows/NT applications. In Proceedings of the 29th Annual International Symposium on Microarchitecture, pages 80--89, Paris, France, Dec. 2--4, 1996. ACM Press.

....orthogonal to this approach, and can improve the benefits of whole procedure code placement. These techniques involve rearranging the code inside procedures, with the goal of allowing the instruction fetching unit to work more efficiently. One simple and effective example is hot cold splitting [10,27], also known as fluff removal . The cold part (or fluff ) of a procedure consists of those basic blocks whose execution frequency is below some threshold, based on profile data. Hot cold splitting moves the cold parts of a procedure out of line and thus compacts the hot parts of the procedure. ....

....It is even better, if slightly harder to implement, to place the cold parts from all procedures in some contiguous area of the executable. This reduces the effective footprint of the procedures, which makes it easier for a procedure placement algorithm to avoid conflicts. Cohn and Lowney [10] address not only the cache effect of hot cold splitting but also the other optimizations it enables. Removing the cold code from the path makes it possible to optimize the hot code by removing partially dead code, eliminating unnecessary preserved registers and stack adjusts, and performing ....

R. Cohn and G. Lowney, "Hot Cold Optimization of Large Windows NT Applications. " Proc. of the 29th Annual International Symposium on Microarchitecture, IEEE, p. 80, December 1996.

....to some degree, runtime code modifiers. Both Dyninst [6] and Vulcan [32] can insert code into running processes. It is based on dynamic instrumentation technology [18] developed as part of the Paradyn Parallel Performance Tools project [26] Other related fields include link time optimization [28, 10] and lowoverhead profiling [2, 14] 0.5 1 1.5 2 2.5 Fib2 fib ref Bubble Matrix2 Matrix Sieve2 Sieve loop const. prop. dead code dead code const. prop call return all opt.s Figure 6. Contributions of Separate Optimizations, Normalized to DynamoRIO with Logical Tracing 10 Future Work ....

R. Cohn and P. G. Lowney. Hot cold optimization of large Windows/NT applications. In 29th Annual International Symposium on Microarchitecture (MICRO '96), Dec. 1996.

....is based on dynamic instrumentation technology [20] developed as part of the Paradyn Parallel Performance Tools project [29] Because these tools modify the original code by inserting trampolines, extensive modification of the code is unwieldy. Other related fields include link time optimization [30, 10] and low overhead profiling [2, 15] 7 Conclusions This paper presents a flexible yet efficient infrastructure for the development of adaptive dynamic optimizations. Dynamic optimization has great potential to solve the problems of static compilation of modern, dynamic software. However, there ....

R. Cohn and P. G. Lowney. Hot cold optimization of large Windows/NT applications. In 29th Annual International Symposium on Microarchitecture (MICRO '96), Dec. 1996.

....mispredictions is comparable to the number of conditional branch (direction) mispredictions. Detailed statistics are given in section 3.3. Due to the popularity of object oriented programming and dynamically linked libraries, the importance of register indirect jump can only become greater [7,11,29] than what is reported here using the SPEC 2000 integer programs. All but eon (written in C ) are written in C and all are statically linked command line programs. 2. Dynamic Code Straightening 2.1 Dynamic software trace cache framework Our dynamic software trace cache (DSTC) framework follows ....

Robert Cohn and P. Geoffrey Lowney, "Hot Cold Optimization of Large Windows/NT Applications," Proc. the 29 Int. Symp. Microarchitecture, Dec 1996.

....allocates registers for procedures using a depth first traversal on the call graph of the program. Chow [1] presents a one pass inter procedural register allocation scheme by using bottom up processing of the procedures call graph as in [18] Several post link tools such as ALTO [13] and SPIKE [2, 3] also try to attack this problera Similar to [1] they distinguish between rarely (cold) and frequently (hot) executed paths within each function and try to avoid execution of instructions in the hot paths at the expense of the colder ones. SPIKE applies the Hot Cold Optimization (HCO) 3] on a ....

....SPIKE [2, 3] also try to attack this problera Similar to [1] they distinguish between rarely (cold) and frequently (hot) executed paths within each function and try to avoid execution of instructions in the hot paths at the expense of the colder ones. SPIKE applies the Hot Cold Optimization (HCO) [3] on a duplicated copy of the hot path. To accomplish this, they also make use of data and control flow analyses. Others [4, 8, 9, 10, 16] handle the same problem in similar ways using control flow, data flow and call graphs representations of the program, which are needed in order to obtain ....

R. Cohn and P. G. Lowney, "Hot Cold Optimization of Large Windows/NT Applications", Proceedings of the 29th Annual IEEE/ACM International Symposium on Microarchitecture, pp. 80 - 89, Dec. 1996.

....the semantics of the program, ATOM requires access to relocation information for the text segment. For this purpose, the linker provided by Digital retains a special type of compact relocations [4] in the executable image. More recent works from Digital such as Hot Cold Optimization (HCO) [6] and Spike [7] implement profile based optimizations on executable files under the Alpha Windows NT system. The task of PostMorph is similar to the first phase of an OM transformation. The goal of PostMorph is to facilitate late program optimizations similar to those performed in OM, HCO and ....

Robert Cohn and P. Geoffrey Lowney, "Hot Cold Optimization of Large Windows NT Applications," in Proceedings of the 29th Annual International Symposium on Microarchitecture, IEEE, December 1996, pages 80-89.

....[Wal92] OM [SW92] also performed optimizations on programs after compilation. These tools were designed for UNIX systems and required either an intermediate form representation of the program or an executable that includes relocation information. More recent systems for Windows NT include Spike [CL96, Goo97] and Etch [RVL97] Both implement profile based optimizations, Etch for the Intel x86 architecture and Spike for Digital Alpha based systems. The Digital Continuous Profiling Infrastructure (DCPI) developed at the Digital Systems Research Center, supports continuous monitoring of the entire ....

....interactive benchmarks. The ideal test for our system would be a suite of large, interactive benchmarks in a format that permits reoptimization. Unfortunately, source or intermediate form representations for such benchmarks are not commonly available. Recent developments in late code optimization [CL96, Rub96, RVL97] offer some promise in addressing these issues. 6. Conclusions This paper describes a system for continuous low overhead profile collection designed to meet the unique requirements of automatic optimization. We achieve these low overheads through statistical sampling of all system activity and ....

R. Cohn and G. Lowney, "Hot Cold Optimization of Large Windows NT Applications." In Proceedings of the 29th Annual International Symposium on Microarchitecture, IEEE, pages 80-89, December 1996.

....on another. Dynamic compilation employs numerous techniques relevant to software dynamic optimization. Dynamic compilation is used both for interpreted languages as just in time compilation [2, 17, 10, 14] and for compiled languages [13, 8] Other related elds include link time optimization [18, 7], which shares with dynamic optimization the fact that it optimizes binary code, and lowoverhead pro ling [1, 11] which is crucial in a dynamic optimization system for quickly identifying important regions of code. 7 Conclusions We have described a dynamic optimization framework for the Windows ....

Robert Cohn and P. Georey Lowney. Hot cold optimization of large Windows/NT applications. In 29th Annual International Symposium on Microarchitecture (MICRO '96), December 1996.

No context found.

R. Cohn and P. G. Lowney, Hot Cold Optimization of Large Windows/NT Applications, MICRO-29, pp. 80-89, Paris, France, December 1996.

No context found.

R. Cohn and P. G. Lowney. Hot cold optimization of large Windows/NT applications. In Proceedings of the 29th Annual International Symposium on Microarchitecture, pages 80--89, Paris, France, Dec. 2--4, 1996. ACM Press.

No context found.

R. Cohn and P. G. Lowney, "Hot Cold Optimization of Large Windows/NT Applications", Proc. MICRO29, Dec. 1996.

No context found.

R. Cohn and P. G. Lowney. Hot cold optimization of large Windows/NT applications. In Proc. 29th Annual International Symposium on Microarchitecture, pages 80--89, December 1996.

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