M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A system for fast, exible, and high-level dynamic code generation. In ACM SIGPLAN '97 Conference on Programming Language Design and Implementation (PLDI), pages 109-121, Las Vegas, Nevada, May 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....these systems provide optimization opportunities not available with static compilation. Some systems include DyC and Tempo [59] based on C) and Fabius [58] based on ML) Calder et al. s work [26] as well as Consel and Noel s work [37] which takes a partial evaluation approach. The tcc system [67] provides a mechanism to specify and compose arbitrary expressions and statements at runtime. However, none of these systems were fully automatic; they all relied 100 on programmer directives to identify regions of code to be optimized. Calpa [62, 61] is a tool that automates the process of ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation (PLDI), pages 109--121, Las Vegas, Nevada, 15--18 June 1997.

....code generation systems. Tempo is a partial evaluator that can perform both compiletime and run time specialization based on programmer hints and simple program analysis [43, 114] C is an extension of ANSI C that allows the programmer to compose arbitrary fragments of dynamically generated code [118]. C includes two dynamic compilers the first is DCG [66] which is relatively heavyweight, and the second is VCODE [65] which is much faster and more lightweight. DyC is a dynamiccompilation system that uses programmer annotations to specify the variables and code on which dynamic ....

....the worst case i.e. everything escapes outright. 55 field. Compiling dynamically allows us to take advantage of actual run time information to selectively specialize with respect to a data value. Specialization with respect to values can allow order of magnitude increases in performance [118, 74]. However, specialization has a significant cost in time and memory, and therefore we must be careful what we attempt to specialize and on what values. We implemented an extension to BC2IR that allows us to analyze a method and make a prediction as to how much better we could do if we knew that a ....

[Article contains additional citation context not shown here]

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In Proceedings of the ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI-97), volume 32, 5 of ACM SIGPLAN Notices, pages 109--121, New York, June 15--18 1997. ACM Press.

....cannot be applied without heuristics that reduces the number of combinations. 5.1.2 Dynamic Code Generation Using a dynamic code generation technique, the specialization process could be accelerated so that meta level objects can be dynamically specialized at run time. Dynamic code generation[21, 27, 28, 61, 62, 99, 115] is a technique to construct a specializer for each target program. The construct specializer receives an input for the target program, and directly generates specialized code in machine instructions. By preparing compiled machine instructions at the construction time of the specializer, the code ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In Proceedings of Conference on Programming Language Design and Implementation (PLDI'97)

....of invariant inputs are typically based on partial evaluation [9] in which the program is specialized for the invariant fixed inputs. These program specialization optimizations are applied to entire regions of code that operate on the fixed inputs. Temp [4] DyC [7] Data Specialization [10] tcc [16], code specialization using value profiles [15] are all software techniques for exploiting coarse grained reuse. All known techniques for exploiting coarse grained reuse can be categorized as either a memoization or a specialization technique. Memoization is a technique based on looking up ....

....as either a memoization or a specialization technique. Memoization is a technique based on looking up previous results in a reuse table [2] Specialization involves optimizing the program by hardcoding the values produced by a reusable piece of code. Although various specialization techniques [4, 7, 15,16] have been studied, available opportunities for specialization in general purpose programs has not been studied before. The goal of this paper is to investigate reuse using a trace based measurement of its properties. We perform our study in the context of an online, dynamic optimizer. In ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A System for Fast, Flexible, and High-level Dynamic Code Generation. In Proceedings of the ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI-97), volume 32, 5 of ACM SIGPLAN Notices, pages 109--121, New York, June 15--18 1997. ACM Press.

....of invariant inputs are typically based on partial evaluation [9] in which the program is specialized for the invariant fixed inputs. These program specialization optimizations are applied to entire regions of code that operate on the fixed inputs. Temp [4] DyC [7] Data Specialization [10] tcc [16], code specialization using value profiles [15] are all software techniques for exploiting coarse grained reuse. All known techniques for exploiting coarse grained reuse can be categorized as either a memoization or a specialization technique. Memoization is a technique based on looking up ....

....as either a memoization or a specialization technique. Memoization is a technique based on looking up previous results in a reuse table [2] Specialization involves optimizing the program by hardcoding the values produced by a reusable piece of code. Although various specialization techniques [4, 7, 15,16] have 1 been studied, available opportunities for specialization in general purpose programs have not been studied before. The goal of this paper is to investigate reuse using a trace based measurement of program properties. We perform our study in the context of an online, dynamic optimizer. In ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A System for Fast, Flexible, and High-level Dynamic Code Generation. In Proceedings of the ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI-97), volume 32, 5 of ACM SIGPLAN Notices, pages 109--121, New York, June 15--18 1997. ACM Press.

....in the same register or memory location [23, 27, 31, 38, 41] In the compiler domain, it has been known for some time that value locality can be used to speed up programs by exploiting the fixed invariant inputs. Partial evaluation [28] data specialization [30] DyC [24] Tempo [12] C [37], and code specialization using value profiles [36] are different software techniques for exploiting value locality. Collectively, we refer to these as value reuse techniques because the input output behavior involving the invariant values can be repeatedly reused as opposed to being recomputed ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A System for Fast, Flexible, and High-level Dynamic Code Generation. In Proceedings of the ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI-97), volume 32, 5 of ACM SIGPLAN Notices, pages 109--121, New York, June 15-- 18 1997. ACM Press.

....bytecodes to a three address code, perform simple optimizations and register allocation, and then generate target machine code. A number of previous systems have utilized more specialized forms of dynamic compilation to selectively optimize program hot spots by exploiting run time constants [11, 3, 19, 14]. In general, these systems emphasize extremely fast dynamic compilation, often performing extensive off line precomputations to avoid constructing any explicit representation of the program fragment being compiled at dynamic compile time. Implementing a Java virtual machine and its related ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In SIGPLAN '97 Conference on Programming Language Design and Implementation, pages 109--121, June 1997.

....also investigated in out oforder execution techniques in dynamic binary translators. Other forms of systems that use dynamic translation to process non native code are Just In Time (JIT) compilers for Java [11,12] Dynamic compilation is generally found in dynamic compilers such as DyC[9] and tcc[10]. Systems such as these either require the programmer to annotate special parts of the program for dynamic compilation at run time, or the program must be written in a specialised language. In dynamic translation, the optimisation process is transparent to the user. Profiling the program s ....

Massimiliano Poletto, Dawson R. Engler and M. Frans Kaashoek. tcc: A System for Fast, Flexible, and High-level Dynamic Code Generation. In PLDI '97. Proceedings of the 1997 ACM SIGPLAN conference on Programming language design and implementation, pages 109-121, Las Vegas, NV, June 1997.

....virtual machines also include limited forms of online feedbackdirected optimization (e.g. inlining in Self 93) but do not develop general mechanisms for adaptive online feedbackdirected optimization. A number of research projects have explored more aggressive forms of dynamic compilation [35, 9, 12, 13, 8, 27, 28, 38, 37, 19, 39], using runtime information to tailor the executable to its current environment. Most of these systems were not fully automatic, and so far, few of these techniques have appeared in mainstream JVMs. However, these systems have demonstrated that online feedback directed optimizations can yield ....

....not statically determinable, these systems provide optimization opportunities not available with static compilation. Some systems include DyC [8, 27, 28] Tempo [38] based on C) Fabius [37] based on ML) and Consel and Noel s work [19] which takes a partial evaluation approach. The tcc system [39] provides a mechanism to specify and compose arbitrary expressions and statements at runtime. The main disadvantage of these techniques is that they rely on programmer directives to identify the regions of code to be optimized. Other work [17, 29] has explored offline profile directed compilation ....

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In Proceedings of the ACM SIGPLAN'97 Conference on Programming Language Design and Implementation (PLDI), pages 109--121, Las Vegas, Nevada, 15--18 June 1997. SIGPLAN Notices 32(5), May 1997.

....decisions. When a basic block counter reaches a threshold, the basic block is reoptimized and moved to the next optimization state, where more aggressive optimizations are performed. Another area of research considers performing runtime optimizations that exploit invariant runtime values [6, 18,26, 25, 15, 27]. The main disadvantage of these techniques is that they rely on programmer directives to identify the regions of code to be optimized. There are also nonadaptive systems that perform compilation optimization at runtime to avoid the cost of interpretation. This includes early work such as the ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In Proceedings of the ACM SIGPLAN'97 Conference on Programming Language Design and Implementation (PLDI), pages 109--121, Las Vegas, Nevada, 15--18 June 1997. SIGPLAN Notices 32(5), May 1997.

....determinable, these systems provide optimization opportunities not available with static compilation. Some systems include DyC [6, 16] and Tempo [24] based on C) and 10 Fabius [23] based on ML) as well as Consel and Noel s work [13] which takes a partial evaluation approach. The tcc system [25] provides a mechanism to specify and compose arbitrary expressions and statements at runtime. The main disadvantage of these techniques is that they rely on programmer directives to identify the regions of code to be optimized. There are also nonadaptive systems that perform ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A system for fast, flexible, and highlevel dynamic code generation. In Proceedings of the ACM SIGPLAN'97 Conference on Programming Language Design and Implementation (PLDI), pages 109--121, Las Vegas, Nevada, 15--18 June 1997. SIGPLAN Notices 32(5), May 1997.

....profiling [17] Le[8] investigates out of order execution techniques in dynamic binary translators, though their results are based on an interpreter based implementation. Many of the optimization techniques used in dynamic translators have been derived from dynamic compilers such as SELF[9] and tcc[10]. Runtime optimizations in such compilers can provide 0.9x 2x the performance of statically compiled programs. Such techniques have also been used in Just in time (JIT) compilers for Java. JITs from Sun[11] Intel[12] and others dynamically generate native machine code at runtime. To date, ....

....This information is currently not used by UQBT; only by UQDBT. UQBT relies on costly analysis to abstract higher level information, without depending on very low level details of the underlying hardware. 0000 0010 1101 1000 addl ebx, eax r[1] r[0] r[1] v[100] v[98] v[99] r[9] r[10] r[9] 1001 0010 0000 0010 0100 0000 0000 1010 add o1, o2, o1 Figure 4. Pentium to SPARC example 3. RESEARCH PROBLEMS Unlike other dynamic binary translators that are written with a fixed set of source and destination machines in mind, UQDBT is designed to handle a wide range of CISC and ....

Massimiliano Poletto, Dawson R. Engler and M. Frans Kaashoek. tcc: A System for Fast, Flexible, and High-level Dynamic Code Generation. In PLDI '97. Proceedings of the 1997 ACM SIGPLAN conference on Programming language design and implementation, pages 109-121, Las Vegas, NV, June 1997.

....dynamically typed. The result is a system with the convenience and functionality of an interpreter, but the speed of a compiled language [14, 4] C is an extension to C that includes syntax for dynamic code generation. Its compiler tcc generates code optimized with respect to run time constants [28]. Consel and Noel demonstrate a general system for efficient run time specialization in C [6] The Synthesis kernel and its successor Synthetix perform run time specialization of operating system services for efficiency. Specialization occurs when the system knows that certain variables in a ....

M. Poletto, D. Engler, and M. Kaashoek. tcc: a system for fast, flexible, and high-level dynamic code generation. In Proceedings of the 1997 ACM SIGPLAN Conference on Programming Language Design and Implementation, pages 109--121, 1997.

....as in other compiler optimizations; however, this ideal is beyond the current state ofthe art for general purpose programs. Instead, current dynamic compilation systems rely on some form of programmer direction to indicate where dynamic compilation should be applied. C [Engler et al. 96, Poletto et al. 97] and its predecessor dcg [Engler Proebsting 94] take a procedural approach to user direction, requiring the user to write programs that explicitly manipulate, compose, and compile program fragments at run time. These systems offer great flexibility and control to the programmer, but at the cost ....

....and DyC offers capabilities not available in C. For instance, C cannot (multi way) unroll loops with dynamic exit tests, because jumps to labels in other tick expressions are not permitted. C recently added limited support for automatic single way loop unrolling within a tick expression [Poletto et al. 97] Also, tick expressions cannot contain other tick expressions, so nested and overlapping dynamic regions cannot be supported. Both of these weaknesses would appear to prevent C from handling the simple interpreter example in Figure 1. C can support run time compiled functions with a dynamically ....

[Article contains additional citation context not shown here]

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A System for Fast, Flexible, and High-level Dynamic Code Generation. SIGPLAN Notices, page To Appear, June 1997. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design 20 and Implementation.

....code at run time. This makes it possible to perform optimizations which depend on information not available until run time, for example, the structure of a sparse matrix or the number of processors in a parallel application. Examples of such systems which generate native code are C (Tick C) [11, 26], Fabius [22] TempoC [3] and DyC [17] Code generation speeds as high as 6 cycles per generated instruction have been achieved. Runtime code modification can also be achieved 8 using dynamic reflection facilities available in languages such as Smalltalk and CLOS. Interfacing The concept of ....

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In PLDI'97, 1996.

....code at run time. This makes it possible to perform optimizations which depend on information not available until runtime, for example, the structure of a sparse matrix or the number of processors in a parallel application. Examples of such systems which generate native code are C (Tick C) [12, 26], and Fabius [23] Speeds as high as 6 cycles per generated instruction have been achieved. Recently, this technology has been extended to C [15] 4.5 Partial Evaluation Code generation is an essential part of active libraries. Over the past two decades, researchers in the field of Partial ....

M. Poletto, D. R. Engler, and M. F. Kaashoek, tcc: A system for fast, flexible, and high-level dynamic code generation, in PLDI'97, 1996.

.... 13] which specializes programs at sourceto source level, has been shown to be useful to optimize various programs, such as interpreters[17 19] scientific applications[5] and graphical applications[11] Recently, run time specialization (RTS) 1 techniques have been actively studied[6 9, 14, 15, 21, 22]. Those techniques e#ciently specialize programs at run time (1) by constructing a dedicated specializer for each target program at compile time, and (2) by directly manipulating native machine code without using source code at specialization time. The improved specialization speed enables us ....

....operations including interpretation and compilation are moved from the online processes (illustrated as shadowed steps in Figure 1) in RTS. Previous studies showed that RTS systems execute a small amount of instructions at online per generated machine instruction in the specialized program[7 9, 14, 15, 21]. Templates can be constructed by slicing o# appropriate fragments from an output of traditional compilers. Figure 2 shows the templates for power in Intel x86 mnemonic, which are sliced o# from the output of a C compiler. In addition to the dynamic expressions in both conditional branches ( 2) ....

[Article contains additional citation context not shown here]

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In PLDI'97 [2], pages 109--121.

....of ML into native MIPS code [19] Some issues like register allocation are decided at compile time whereas instruction selection is performed at run time. The Tick C compiler generates code at run time from a C program where computations are explicitly staged using Lisp like backquote notations [9, 27]. The DyC system compiles partially annotated C programs. Like Tempo, it produces templates which are compiled by the DEC Alpha compiler [3, 14] Unlike Tempo, it performs additional optimizations that can exploit template instantiation values and template assembly. Data are not yet available to ....

Massimiliano Poletto, Dawson R. Engler, and M. Frans Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In PLDI'97 [26], pages 109--121.

....and does not perform live range splitting. Priority based register allocation is effective in allocating registers to the most important variables in a function and can easily take into account call costs [18] An alternative linear time approach that does not consider priorities is described in [23]. Our algorithm is as follows: For each variable u, ordered by priority, the allocator performs a backwards depthfirst search through the flow graph, starting the search at all basic blocks in which u is used (representing a possible end of the live range) and terminating the search at a basic ....

M. Poletto, D.R. Engler and M.F. Kaashoek. tcc: A System for Fast, Flexible, and High-Level Dynamic Code Generation. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation, pages 109-121. ACM, June 1997.

....beyond the current state of the art for general purpose programs. Instead, current dynamic compilation systems rely on some form of programmer direction to indicate where dynamic compilation would most profitably be applied. Some previous dynamic compilation systems, such as C [Engler et al. 96, Poletto et al. 97] and its predecessor dcg [Engler Proebsting 94] take a procedural approach to user direction, requiring the user to write programs that explicitly manipulate, compose, and compile program fragments at run time. This kind of system offers great flexibility and control to the programmer, at the ....

....and our system offers capabilities not available in C. For instance, C cannot (multi way) unroll loops with dynamic exit tests because jumps to labels in other tick expressions are not permitted. C recently added limited support for automatic single way loop unrolling within a tick expression [Poletto et al. 97] Also, tick expressions cannot contain other tick expressions, so nested (not to mention overlapping) dynamic regions cannot be supported. Both of these weaknesses would appear to prevent C from handling the simple interpreter example in Figure 1. C can support run time compiled functions with ....

[Article contains additional citation context not shown here]

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A System for Fast, Flexible, and High-level Dynamic Code Generation. SIGPLAN Notices, page To Appear, June 1997. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation.

....is in a high level form. Yet, this property checking can be a fast process (e.g. syntactic) if the language has been appropriately defined. Run time specialization can be seen as an automatic approach to dynamic code generation. Poletto et al. propose a manual approach to dynamic code generation [11]. To do so, they extend the C programming language with constructs and types which allow the programmer to directly express the computations which construct program fragments dynamically. The advantage of this strategy is to allow the programmer to have tight control over the program fragments to ....

M. Poletto, D. Engler, and M. F. Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In ACM SIGPLAN Conference on Programming Language Design and Implementation, pages 109--121, Las Vegas, Nevada, June 1997. ACM Press.

....at run time. In turn, g will generate a function h each time it is called. Nested codegen s are thus used to generate code that generates code. The first version of Tempo did not support code that generates code (though it has recently been extended to do so) and some other systems, such as C [21, 22], also prohibit it. We decided to permit it in Cyclone, because it adds little complication to our type system or implementation. Nested codegen s are not generated automatically in Cyclone, because of the version of Tempo that we use, but the programmer can always write them explicitly. A final ....

....of using an unmodified, existing compiler limits the optimizations that it can perform. ML box, Meta ML, and C are all systems that add explicit code generation constructs to existing languages. MLbox and Meta ML are type safe dialects of ML [16, 26, 23] while C is an unsafe dialect of C [21, 22]. All three systems have features for combining code fragments that go beyond what we provide in Cyclone. For example, in C it is possible to generate functions that have n arguments, where n is a value computed at run time; this is not possible in Cyclone, ML box, or Meta ML. On the other hand, ....

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In Proceedings of the ACM SIGPLAN '97 Conference on Programming Language Design and Implementation (PLDI), pages 109--121, Las Vegas, Nevada, 15--18 June 1997. SIGPLAN Notices 32(5), May 1997.

....They all enable a high degree of control while remaining relatively easy to use. In general, however, as the amount of control increases, the ease of use decreases. At one end of this range is the C language, which is discussed in detail in Chapter 2 and has also been described elsewhere [23, 56, 57]. C is an extension of ANSI C that enables imperative dynamic code generation: the programmer can write fragments of C that denote dynamic code and then explicitly manipulate them to create large and complex dynamic code objects. C gives a vast amount of control over the compilation process and ....

.... register allocation algorithms that have been considered in the literature [35, 42, 31, 53] which in turn take their inspiration from an optimal off line replacement algorithm that was studied for virtual memory [5] Since the original description of the linear scan algorithm in Poletto et al. [56], Traub et al. have proposed a more complex linear scan algorithm, which they call second chance binpacking [73] This algorithm is an evolution and refinement of binpacking, a technique used for several years in the DEC GEM optimizing compiler [9] At a high level, the binpacking schemes are ....

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A system for fast, flexible, and high-level dynamic code generation. In Proceedings of the ACM SIGPLAN Conference on Programming Language Design and Implementation, pages 109--121, Las Vegas, NV, June 1997. ACM.

No context found.

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A system for fast, exible, and high-level dynamic code generation. In ACM SIGPLAN '97 Conference on Programming Language Design and Implementation (PLDI), pages 109-121, Las Vegas, Nevada, May 1997.

No context found.

M. Poletto, D. R. Engler, and M. F. Kaashoek. tcc: A System for Fast, Flexible, and High-level Dynamic Code Generation. In Proceedings of the ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI-97), volume 32, 5 of ACM SIGPLAN Notices, pages 109--121, New York, June 15--18 1997. ACM Press.

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