Optimizing ML with run-time code generation

LeePeter,; LeoneMark,

doi:10.1145/249069.231407

Cited by 29 publications

(15 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some contemporary run-time code generation systems are Fabius, Tempo, and DyC. Fabius performs run-time code generation for programs written in a functional subset of ML [9]. Tempo supports both compile-time partial evaluation and run-time specialization of C programs [4].…”

Section: Relatedmentioning

confidence: 99%

Facile

Schnarr¹,

Hill

Larus

2001

SIGPLAN Not.

View full text Add to dashboard Cite

Architectural simulators are essential tools for computer architecture and systems research and development. Simulators, however, are becoming frustratingly slow, because they must now model increasingly complex micro-architectures running realistic workloads. Previously, we developed a technique called fast-forwarding, which applied partial evaluation and memoization to improve the performance of detailed architectural simulations by as much as an order of magnitude [14].While writing a detailed processor simulator is difficult, implementing fast-forwarding is even more complex. This paper describes Facile, a domain-specific language for writing detailed, accurate micro-architecture simulators. Architectural descriptions written in Facile can be compiled, using partial evaluation techniques, into fast-forwarding simulators that achieve significant performance improvements with far less programmer effort. Facile and its compiler make this performance-enhancing technique accessible to computer architects.

show abstract

Section: Relatedmentioning

confidence: 99%

Facile

Schnarr¹,

Hill

Larus

2001

SIGPLAN Not.

View full text Add to dashboard Cite

show abstract

“…Recently, they have extended their compiler, FABIUS, to accept a functional subset of ML [32]. They achieve low code generation costs using some of the techniques we independently derived for tee.…”

Section: Related Workmentioning

confidence: 99%

“…In such systems, the static compiler emits templates that at run time are filled with appropriate values, optimized in relatively simple but effective ways, and emitted directly. In practice, however, only Leone and Lee [32] appear to generate code more quickly than tee. Furthermore, improving performance in this way has a usability cost: no existing template-based system gives the programmer as much freedom and expressive power as 'C.…”

Section: Related Workmentioning

confidence: 99%

“…Since many optimizations on dynamic code can be profitably performed only at run time, improvements in code quality are generally obtained at the expense of greater code generation time. One way to generate good code with low overhead is to structure the system so that most optimizations can be performed statically [ 1,11,321, but in all existing systems this ability comes at the expense of language flexibility and expressiveness (furthermore, in practice, only Leone and Lee [32] appear to generate code more quickly than tee). 'C, by contrast, allows the user to dynamically compose arbitrary pieces of code, which makes static analysis of the dynamic code problematic.…”

Section: Architecturementioning

confidence: 99%

See 1 more Smart Citation

tcc

1997

View full text Add to dashboard Cite

tee is a compiler that provides efficient and high-level access to dynamic code generation. It implements the 'C ("Tick-C") programming language, an extension of ANSI C that supports dynamic code generation [ 151. 'C gives power and flexibility in specifying dynamically generated code: whereas most other systems use annotations to denote run-time invariants. 'C allows the programmer to specify and compose arbitrary expressions and statements at run time. This degree of control is needed to efficiently implement some of the most important applications of dynamic code generation, such as "just in time" compilers [ 171 and efficient simulators [ 10, 48, 461.The paper focuses on the techniques that allow tee to provide 'C's flexibility and expressiveness without sacrificing run-timecode generation efficiency. These techniques include fast register allocation, efficient creation and composition of dynamic code specifications, and link-time analysis to reduce the size of dynamic code generators. tee also implements two different dynamic code generation strategies, designed to address the tradeoff of dynamic compilation speed versus generated code quality. To characterize the effects of dynamic compilation, we present performance measurements for eleven programs compiled using tee. On these applications, we measured performance improvements of up to one order of magnitude.To encourage further experimentation and use of dynamic code generation, we are making the tee compiler available in the public domain. This is, to our knowledge, the first high-level dynamic compilation system to be made available.

show abstract

“…ory behavior of each program run to produce optimizations tailored to that run. Specialization, or partial evaluation, is a related, more aggressive strategy by which hot portions of code are heavily optimized by "hard-coding" frequently occurring values, and other values that depend on them, directly into the instruction stream; when these values turn up again, the optimized code is invoked [2,8,13,14,17,21,22,26,28,31,35,36,43,44,45]. For certain classes of programs, such as interpreters, raytracers, and database query executors, in which a few popular values consistently dictate execution behavior, employing this technique can result in marked speedups, up to 5x in some cases [23].…”

Section: Introductionmentioning

confidence: 99%

Runtime specialization with optimistic heap analysis

Shankar

Sastry

Bodík

et al. 2005

Proceedings of the 20th Annual ACM SIGPLAN Conference on Object-Oriented Programming, Systems, Languages, and Applications

View full text Add to dashboard Cite

We describe a highly practical program specializer for Java programs. The specializer is powerful, because it specializes optimistically, using (potentially transient) constants in the heap; it is precise, because it specializes using data structures that are only partially invariant; it is deployable, because it is hidden in a JIT compiler and does not require any user annotations or offline preprocessing; it is simple, because it uses existing JIT compiler ingredients; and it is fast, because it specializes programs in under 1s.These properties are the result of (1) a new algorithm for selecting specializable code fragments, based on a notion of influence; (2) a precise store profile for identifying constant heap locations; and (3) an efficient invalidation mechanism for monitoring optimistic assumptions about heap constants. Our implementation of the specializer in the Jikes RVM has low overhead, selects specialization points that would be chosen manually, and produces speedups ranging from a factor of 1.2 to 6.4, comparable with annotationguided specializers.

show abstract

Optimizing ML with run-time code generation

Cited by 29 publications

References 22 publications

Facile

Facile

tcc

Runtime specialization with optimistic heap analysis

Contact Info

Product

Resources

About