Peephole optimizers are typically constructed using human-written pattern matching rules, an approach that requires expertise and time, as well as being less than systematic at exploiting all opportunities for optimization. We explore fully automatic construction of peephole optimizers using brute force superoptimization. While the optimizations discovered by our automatic system may be less general than human-written counterparts, our approach has the potential to automatically learn a database of thousands to millions of optimizations, in contrast to the hundreds found in current peephole optimizers. We show experimentally that our optimizer is able to exploit performance opportunities not found by existing compilers; in particular, we show speedups from 1.7 to a factor of 10 on some compute intensive kernels over a conventional optimizing compiler.

We report studies on an L-asparaginase from Pyrococcus furiosus, cloned and expressed in Escherichia coli and purified to homogeneity. Protein stability and enzyme kinetic parameters were determined. The enzyme was found to be thermostable, natively dimeric, and glutaminase-free, with optimum activity at pH 9.0. It showed a K(m) of 12 mM and a substrate inhibition profile above 20 mM L-asparagine. Urea could not induce unfolding and enzyme inactivation; however, with guanidine hydrochloride (GdnCl) a two-state unfolding pattern was observed. Reduced activity and an altered near-UV-CD signal for protein at low GdnCl concentration (1 M) suggested tertiary structural changes at the enzyme active site. A homology three-dimensional model was developed and the structural information was combined with activity and stability data to give functional clues about the asparaginase.

No abstract

Covalent linkers bridging the domains of multidomain proteins are considered to be crucial for assembly and function. In this report, an exception in which the linker of a two-domain dimeric L-asparaginase from Pyrococcus furiosus (PfA) was found to be dispensable is presented. Domains of this enzyme assembled without the linker into a conjoined tetrameric form that exhibited higher activity than the parent enzyme. The global shape and quaternary structure of the conjoined PfA were also similar to the wild-type PfA, as observed by their solution scattering profiles and X-ray crystallographic data. Comparison of the crystal structures of substrate-bound and unbound enzymes revealed an altogether new active-site composition and mechanism of action. Thus, conjoined PfA is presented as a unique enzyme obtained through noncovalent, linker-less assembly of constituent domains that is stable enough to function efficiently at elevated temperatures.

Peephole optimizers are typically constructed using human-written pattern matching rules, an approach that requires expertise and time, as well as being less than systematic at exploiting all opportunities for optimization. We explore fully automatic construction of peephole optimizers using brute force superoptimization. While the optimizations discovered by our automatic system may be less general than human-written counterparts, our approach has the potential to automatically learn a database of thousands to millions of optimizations, in contrast to the hundreds found in current peephole optimizers. We show experimentally that our optimizer is able to exploit performance opportunities not found by existing compilers; in particular, we show speedups from 1.7 to a factor of 10 on some compute intensive kernels over a conventional optimizing compiler.